Stacks and Moduli

working draft

Jarod Alper

jarod@[Link]

February 17, 2025

Stacks and Moduli © 2020 by Jarod Alper is licensed under CC BY-NC-ND 4.0.
To view a copy of this license, visit
[Link]

ii
Preface

These notes develop the foundations of moduli theory in algebraic geometry with the
goal of providing self-contained proofs in characteristic 0 of the following theorems:

Theorem A. The moduli stack Mg of stable curves of genus g ≥ 2 is a smooth,

proper, and irreducible Deligne–Mumford stack of dimension 3g − 3 which admits a
projective coarse moduli space M g .

Theorem B. The moduli stack Bun ss r,d (C) of semistable vector bundles of rank
r and degree d over a smooth, connected, and projective curve C of genus g is a
smooth, universally closed, and irreducible algebraic stack of dimension r2 (g − 1)
ps
which admits a projective good moduli space Mr,d (C).

Along the way we build the foundations of algebraic spaces and stacks, which
provide a convenient language to discuss and establish geometric properties of moduli
spaces. Introducing these foundations requires developing several themes at the
same time including:

– embracing the functorial perspective in algebraic geometry, where a scheme

X is determined by its relation to all other schemes, i.e., by its representable
functor Sch → Sets, T 7→ Mor(T, X);
– introducing stacks as a categorical gadget that encodes the symmetries of
objects, e.g., we will think of Mg as the assignment of a scheme T to the
groupoid of all families of smooth curves over T ;
– replacing the Zariski topology on a scheme with the étale topology: we will
introduce Grothendieck topologies as a generalization of topological spaces,
and we will systematically use descent theory for étale morphisms; and
– relying on several advanced topics not typically seen in a first algebraic geometry
course, e.g., properties of flat, étale, and smooth morphisms of schemes,
algebraic groups and their actions, deformation theory, and the birational
geometry of surfaces.

Choosing a linear order to present the foundations is no easy task. We mitigate this
challenge by relegating much of the background to appendices, while keeping the
main body of the notes focused on moduli theory with the above two theorems in
mind.
Prerequisites. We assume that you are familiar with schemes, e.g., at the level of
[Har77, Ch. 2–3] or [Vak17, Parts II-V], but not much more. On the other hand,
you will see many important concepts in algebraic geometry in action in the context
of moduli theory, e.g., the valuative criterion for properness and cohomology and
base change. This provides the opportunity to fill in gaps in your background or

iii
to reinforce your understanding of key concepts in scheme theory. In fact, every
section of [Har77] and [Vak17] will be relevant for at least some part of this text.
Sources. Moduli spaces have a rich history with its modern origins in Bernhard
Riemann’s work in the 19th century (see §0.1). Our approach to the development of
moduli theory is due to Alexander Grothendieck, David Mumford, Michael Artin,
János Kollàr, Johan de Jong, and many others. A primary source for many topics
in this book is the Stacks Project [SP]. The moduli space of stable curves had its
origins in [GIT] and [DM69] while the textbooks [HM98] and [ACG11] offer further
details. Similarly, the origins of the moduli space of semistable vector bundles can
be traced to [Mum63] and [Ses67], and the books [LP97] and [HL10] provide more
extensive treatments.
Our approach. We choose a route to the summits of Mt. Mgbar (Theorem A) and
Mt. BunC (Theorem B) with a gradual slope while minimizing technical pitches
and crevasses. Our path offers frequent views of the summits ahead as well as the
surrounding valleys and peaks. The appendices provide background training in the
technical gear necessary for our climb, and we take a warm-up ascent of Mts. Hilbert
and Quot (Theorems 1.1.2 and 1.1.3) to get practice with the equipment. While
strongly advised against in mountain climbing, you can also head directly for the
summits without any gear or training, waiting only until you fall into a crevasse
before acquiring the background knowledge.
Motivated by the ethos that mathematical understanding is often best obtained
by a thorough study of examples, this book illustrates concepts in moduli theory
though three main examples: Hilbert and Quot schemes (as moduli spaces of
objects with no automorphisms), the moduli stack Mg of stable curves (with finite
automorphisms), and the moduli stack Bun ss r,d (C) of semistable vector bundles (with
infinite automorphisms) on a fixed curve. Certainly, before you hope to study moduli
of higher dimensional varieties or moduli of Bridgeland semistable complexes on a
surface, you ought to have a solid understanding of these examples.
This book is long (sorry!) and you should not read it linearly from start to finish.
The hope is that each chapter and even each section can be read independently,
and that the reader can skip around as desired. Throughout the text, we attempt
to provide examples, counterexamples, and exercises for each new concept, always
starting with the most elementary cases. The exercises come in different flavors, e.g.,
working out additional examples, proving further properties, or filling in technical
details of a proof, but the purpose remains the same: to provide the reader a chance
to practice and play around with the new mathematical objects.
While the approach in the Stacks Project is to prove theorems in the utmost level
of generality, we often sacrifice the level of generality if it allows for a more intuitive
or geometric proof. Most notably, while we work over any field k (or base scheme S)
when possible, in both our proof of Stable Reduction (5.5.1) for curves and for the
construction of the moduli space of semistable vector bundles, we assume that the
field k is of characteristic 0. We sometimes add a noetherian hypothesis even if it
is not strictly necessary, e.g., in the Valuative Criterion for Properness (3.8.7) for
algebraic stacks, or we will establish a theorem only for Deligne–Mumford stacks
rather than for algebraic stacks with finite inertia, e.g., in the Keel–Mori theorem
(4.4.6).
Whenever possible, we give pointers to the Stacks Project and original sources
containing either more general results or alternative arguments. We hope that this
text can serve as a guide to the much more exhaustive development in the Stacks
Project. With the tools and techniques developed in this book, the reader should be

iv
empowered to study geometric properties of these moduli spaces as in [HM98] or
[HL10].
Acknowledgements. This book originated from an online graduate course Math
582C: Introduction to Stacks and Moduli taught by the author at the University
of Washington in the Winter Quarter of 2021. The lecture notes have expanded
over the years as I’ve had opportunities to teach similar graduate courses at the
University of Washington and several graduate minicourses— University of Es-
sex, Universidad Católica de Chile, Scuola Normale Superiore di Pisa, Universität
Duisburg-Essen, L’Institut de Mathématiques de Bourgogne, University of Illinois at
Chicago, Università degli Studi di Salerno, Johannes Gutenberg-Universität Mainz,
Pohang University of Science and Technology, and Universidad de los Andes. I
would like to thank the participants for their valuable feedback.
I would like to thank Ravi Vakil and Johan de Jong for introducing me to moduli
spaces and stacks, and for their encouragement to write this textbook. This book
would not have been possible without de Jong’s contributions in the Stacks Project
and the countless other arguments and examples that he has personally shared with
me. I would also like to thank my collaborators—especially Maksym Fedorchuk,
Jack Hall, Daniel Halpern-Leistner, Jochen Heinloth, Andrew Kresch, David Rydh,
and David Smyth—for sharing their knowledge and insights. I thank Dhyan Aranha,
Arend Bayer, Jim Bryan, Ben Church, Hannah Dell, Neeraj Deshmukh, Patricio
Gallardo, Davide Gori, Vu Hien Giovanni Inchiostro, Adam Kapilow, Rega Karim,
Myeonhu Kim, Aaron Landsman, Dion Leijnse, Jake Levinson, Max Lieblich, Ben-
jamin Lim, Siddarth Matthur, Andrés Ibáñez Núñez, Brian Nungent, Dragos Oprea,
Lachlan Potter, Denis Resin, Minseon Shin, Tuomas Tajakka, Alex Schefflin, Ian
Selvaggi, Reinier Sorgdrager, Dario Weissmann, Maxime Wybouw, Xucheng Zhang,
and many others for their comments and corrections.

v
Table of Contents

0 Introduction and motivation 1

0.1 A brief history of moduli . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.2 Moduli sets of curves, vector bundles, and triangles . . . . . . . . . . 13
0.3 The functorial worldview . . . . . . . . . . . . . . . . . . . . . . . . . 16
0.4 Moduli groupoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
0.5 Why the étale topology? . . . . . . . . . . . . . . . . . . . . . . . . . 32
0.6 Moduli stacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
0.7 Constructing projective moduli spaces . . . . . . . . . . . . . . . . . 41

1 Hilbert and Quot schemes 47

1.1 The Grassmannian, Hilbert, and Quot functors . . . . . . . . . . . . 47
1.2 Projectivity of the Grassmannian . . . . . . . . . . . . . . . . . . . . 49
1.3 Castelnuovo–Mumford regularity . . . . . . . . . . . . . . . . . . . . 54
1.4 Projectivity of Hilb and Quot . . . . . . . . . . . . . . . . . . . . . . 59
1.5 An invitation to the geometry of Hilbert schemes . . . . . . . . . . . 65

2 Sites, sheaves, and stacks 69

2.1 Descent theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.2 Grothendieck topologies and sites . . . . . . . . . . . . . . . . . . . . 80
2.3 Presheaves and sheaves . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.4 Prestacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.5 Stacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

3 Algebraic spaces and stacks 103

3.1 Definitions of algebraic spaces and stacks . . . . . . . . . . . . . . . 103
3.2 Representability of the diagonal . . . . . . . . . . . . . . . . . . . . . 115
3.3 First properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.4 Equivalence relations and groupoids . . . . . . . . . . . . . . . . . . 127
3.5 Dimension, tangent spaces, and residual gerbes . . . . . . . . . . . . 133
3.6 Characterization of Deligne–Mumford stacks . . . . . . . . . . . . . . 139
3.7 Smoothness and the Infinitesimal Lifting Criteron . . . . . . . . . . . 142
3.8 Properness and the Valuative Criterion . . . . . . . . . . . . . . . . . 146
3.9 Further examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

4 Geometry of Deligne–Mumford stacks 162

4.1 Quasi-coherent sheaves and cohomology . . . . . . . . . . . . . . . . 162
4.2 Quotients by finite groups . . . . . . . . . . . . . . . . . . . . . . . . 172
4.3 The local structure of Deligne–Mumford stacks . . . . . . . . . . . . 178
4.4 Coarse moduli spaces and the Keel–Mori Theorem . . . . . . . . . . 182
4.5 Algebraic spaces versus schemes . . . . . . . . . . . . . . . . . . . . . 192

vi
 4.6 Finite covers of Deligne–Mumford stacks . . . . . . . . . . . . . . . . 201

5 Moduli of stable curves 207

5.1 Review of smooth curves . . . . . . . . . . . . . . . . . . . . . . . . . 207
5.2 Nodal curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
5.3 Stable curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
5.4 The stack of all curves . . . . . . . . . . . . . . . . . . . . . . . . . . 233
5.5 Stable reduction and the properness of Mg,n . . . . . . . . . . . . . 239
5.6 Contraction, forgetful, and gluing morphisms . . . . . . . . . . . . . 250
5.7 Irreducibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
5.8 Projectivity following Mumford . . . . . . . . . . . . . . . . . . . . . 277
5.9 Projectivity following Kollár . . . . . . . . . . . . . . . . . . . . . . . 291
5.10 Glimpse of the geometry of M g . . . . . . . . . . . . . . . . . . . . . 301

6 Geometry of algebraic stacks 303

6.1 Quasi-coherent sheaves and cohomology theories . . . . . . . . . . . 303
6.2 Global quotient stacks and the resolution property . . . . . . . . . . 315
6.3 The fppf topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
6.4 Gerbes, rigidification, and Picard schemes . . . . . . . . . . . . . . . 321
6.5 Affine GIT and good moduli spaces . . . . . . . . . . . . . . . . . . . 335
6.6 Coherent Tannaka duality and coherent completeness . . . . . . . . . 351
6.7 Local structure of algebraic stacks . . . . . . . . . . . . . . . . . . . 360
6.8 Gm -actions and the Białynicki-Birula Stratification . . . . . . . . . . 372
6.9 Valuative criteria for algebraic stacks: Θ- and S-completeness . . . . 380
6.10 Existence of good moduli spaces . . . . . . . . . . . . . . . . . . . . 391

7 Geometric Invariant Theory 405

7.1 Good quotients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
7.2 Projective GIT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
7.3 The Cartan Decomposition and the Destabilization Theorem . . . . 411
7.4 The Hilbert–Mumford Criterion . . . . . . . . . . . . . . . . . . . . . 413
7.5 Examples in GIT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
7.6 Kempf’s Optimal Destabilization Theorem . . . . . . . . . . . . . . . 421
7.7 The HKKN stratification of the unstable locus . . . . . . . . . . . . 428

8 Moduli of semistable vector bundles 435

8.1 Semistable vector bundles . . . . . . . . . . . . . . . . . . . . . . . . 436
8.2 Moduli stack of semistable vector bundles . . . . . . . . . . . . . . . 447
8.3 Semistable reduction, moduli spaces, and stratifications . . . . . . . 456

Appendix A Morphisms of schemes 465

A.1 Morphisms locally of finite presentation . . . . . . . . . . . . . . . . 465
A.2 Flatness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466
A.3 Étale, smooth, and unramified morphisms . . . . . . . . . . . . . . . 473
A.4 Properness and the Valuative Criterion . . . . . . . . . . . . . . . . . 480
A.5 Dévissage and finiteness of cohomology . . . . . . . . . . . . . . . . . 483
A.6 Cohomology and Base Change . . . . . . . . . . . . . . . . . . . . . 485
A.7 Quasi-finite morphisms and Zariski’s Main Theorem . . . . . . . . . 494

vii
Appendix B Further topics in scheme theory 497
B.1 Algebraic groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
B.2 Birational geometry and positivity . . . . . . . . . . . . . . . . . . . 517
B.3 Limits of schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 527
B.4 Pushouts of schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . 530
B.5 Completions, henselizations, and Artin Approximation . . . . . . . . 532

Appendix C Deformation theory 539

C.1 First order deformations . . . . . . . . . . . . . . . . . . . . . . . . . 540
C.2 Higher-order deformations and obstructions . . . . . . . . . . . . . . 545
C.3 Cotangent complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . 552
C.4 Versal formal deformations and Rim–Schlessinger’s Criteria . . . . . 556
C.5 Effective formal deformations and Grothendieck’s Existence Theorem 563
C.6 Artin Algebraization . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
C.7 Artin’s Axioms for Algebraicity . . . . . . . . . . . . . . . . . . . . . 572

Bibliography 584

viii
Chapter 0

Introduction and motivation

Besides being a form of cartography, the theory of moduli spaces has the
wonderful feature of having many doors, many techniques by which this
theory can be developed. Of course, there is traditional algebraic geometry,
but there is also invariant theory, complex-analytic techniques such as
Teichmüller theory, global topological techniques, and purely characteristic
p methods such as counting objects over finite fields. This is another part
of its charm.

David Mumford [Mum04, preface]

Moduli spaces arise as solutions to one of the most fundamental problems in

mathematics:

Classification problem: Can we classify the isomorphism classes

of mathematical objects of a given type?

There are many types of objects that we may want to classify:

– subspaces V ⊆ Cn of dimension k;
– plane curves C ⊆ P2 of degree d;
– curves C of genus g together with a degree d morphism C → P1 ;
– vector bundles on a fixed projective variety X; and
– representations of a group, e.g., an absolute Galois group Gal(Qp /Qp ), the
fundamental group π1 (Σ) of a topological surface of genus g, or the path
algebra of quiver.

Our primary interest is in the following two examples, which will be used
throughout this book to illustrate the concepts of moduli.
(1) smooth (or more generally stable) projective curves of genus g, and
(2) vector bundles (or more specifically semistable vector bundles) on a fixed
smooth curve C.

A moduli space is a space whose points are in natural bijection

with isomorphism classes of the objects of a given type.

1
 The keyword above is ‘natural’, and it is probably not clear to you what this
could mean. Indeed, one of the main challenges in developing moduli theory is
formulating precisely what this means. After all, any two complex manifolds or
varieties of positive dimension are bijective as they both have the cardinality of the
continuum. We do not want our moduli space to be a disjoint of a points, with
one point for every isomorphism class. We need to specify a sense of proximity for
objects. Our approach for formalizing the notion of a ‘natural bijection’ will be to
introduce the notion of a family of objects and require that every family of objects
over a space S corresponds uniquely to a map from S to the moduli space; see §0.3.
The structure of the ‘space’ depends on the context: if we are classifying topo-
logical objects, we might hope for the structure of a topological space, while if we
are classifying analytic objects, we might hope for the structure of a manifold. In
this book, we are mainly focused on classifying algebraic structures, and we desire
a moduli space with the structure of an algebraic variety, and ideally a projective
variety.
Ubiquity of moduli. Once you start viewing spaces through the lens of moduli,
everything begins looking like a moduli space! This is true in a precise sense: every
space M is the moduli space of its points. It is of course more interesting when
there are equivalent descriptions with geometric meanings. Projective space P1 is
the set of points in P1 (not so interesting), or lines in the plane passing through the
origin (more interesting), or subschemes of A2 of length 2 supported at the origin
(also interesting), or isomorphism classes of stable elliptic curves (very interesting).
Fascinatingly, moduli spaces often afford equivalent viewpoints across different
mathematical fields, as is the case in our primary two examples:
(1) Mg is the moduli of smooth projective algebraic curves of genus g, or the
moduli of compact Riemann surfaces of genus g, or the moduli of complex
structures on a fixed compact oriented topological surface Σg of genus g up to
biholomorphisms, or the moduli of hyperbolic metrics on Σg up to isometries.
(2) Bun ss
r,d (C) is the moduli of semistable algebraic vector bundle on a fixed
curve C, or the moduli of holomorphic vector bundles on C with flat unitary
connection, or the moduli of irreducible unitary representation of π1 (C).
This leads to a rich interplay between algebraic, analytic, and topological approaches.
Discrete vs continuous moduli. Depending on the types of objects, the moduli
space could be discrete or continuous, or a combination of the two.
• The moduli space of line bundles on P1 is the discrete set Z: every line bundle
on P1 is isomorphic to O(n) for a unique integer n ∈ Z.
• The moduli space parameterizing quadric plane curves C ⊆ P2 is the connected
space P5 : a curve defined by a0 x2 + a1 xy + a2 xz + a3 y 2 + a4 yz + a5 z 2 is
uniquely determined by the point [a0 , . . . , a5 ] ∈ P5 , and as a plane curve varies
continuously (i.e., by varying the coefficients ai ), the corresponding point in
P5 does too.
• For smooth curves, the genus g is a discrete invariant. For fixed genus g, the
moduli space Mg of smooth curves is a variety; it is in fact an irreducible
quasi-projective variety, but we are now getting far ahead of ourselves.
` The
moduli space of all smooth curves can be viewed as the disjoint union g Mg .
• For vector bundles on a smooth curve C, the rank r and degree d are discrete
invariants while the moduli space Mr,d
ss
(C) of semistable bundles of rank r and
degree d is an irreducible projective variety.

2
Why study moduli spaces? Properties of moduli spaces can inform us about the
properties of the objects themselves. Many properties of objects are best formulated
in terms of moduli spaces. For instance, to express the condition that a general
genus 3 curve can be parameterized by an explicit coordinate system—namely a
general genus 3 curve is canonically embedded into P2 as a plane quartic and thus
parameterized by a point in the space P(Γ(P2 , O(4))) ∼
= P14 —we could say that the
moduli space M3 is unirational, i.e., there is a dominant rational map P14 99K M3 .
Do moduli spaces exist? Before we can begin to study the geometric properties
of a moduli space, we need to know that they exist. This is no easy task and is the
main goal of this book. We develop the foundations of moduli theory in order to
prove that there is a projective moduli space parameterizing stable curves of genus
g (Theorem A) and a projective moduli space parameterizing semistable vector
bundles of rank r and degree d on a fixed smooth curve (Theorem B).
It is an astonishing mathematical coincidence that moduli spaces of algebraic
objects often exist as algebraic varieties. Their existence is the starting point of
moduli theory. The beauty, allure, and elegance of moduli spaces is what first
attracted me and countless others to the subject.
Trichotomy of moduli. A recurring theme in moduli is the influence of automor-
phism groups on both the properties of a moduli space and the techniques used to
study its geometry. There is a trichotomy in moduli theory depending on the size of
the automorphism groups: (1) no automorphisms, (2) finite automorphisms, and
(3) infinite automorphisms. In (3), the moduli spaces are particularly well-behaved
when the closed points—sometimes referred to as polystable objects—of the moduli
stack have reductive automorphisms.

Automorphisms None Finite Reductive

at Closed Points

Type of space Scheme/algebraic Deligne–Mumford algebraic stack

space stack
Defining Zariski/étale locally étale locally an smooth locally an
property an affine scheme affine scheme affine scheme
Examples Pn , Gr(q, n), Hilb, Mg Bun ss
r,d (C)
Quot
Quotient action is free finite stabilizers reductive stabiliz-
stacks [X/G] ers at closed orbits
Existence of fine moduli space coarse moduli good moduli space
moduli space space

Table 0.0.1: Trichotomy of moduli.

Roadmap of this chapter. We motivate the approach of this text by gradually

adding more enriched structures to moduli sets of objects. We first introduce
families of objects and the functorial worldview in §0.3, and then develop the
groupoid perspective in §0.4. After motivating the étale topology in §0.5, we

3
combine these perspectives by introducing moduli stacks in §0.6. We then sketch
two main techniques to construct a projective moduli space in §0.7.

Sets Groupoids

Topological Functors or
Prestacks over Sch
spaces presheaves on Sch

Ringed spaces Sheaves on Schét Stacks over Schét

Algebraic spaces Deligne–Mumford Algebraic

Schemes
stacks stacks

Figure 0.0.2: Schematic diagram of algebro-geometric enrichments of sets and

groupoids.

0.1 A brief history of moduli

The spirit of Riemann will move future generations as it has moved us.

Lars Ahlfors [Ahl53, p. 53]

The historical development of moduli theory provides a first glimpse of many

themes in moduli.

0.1.1 Bernhard Riemann and the origins of Mg

Die 3p − 3 übrigen Verzweigungswerthe in jenen Systemen gle-

ichverzweigter µ werthiger Functionen können daher beliebige Werthe
annehmen; und es hängt also eine Klasse von Systemen gleichverzweigter
2p + 1 fach zusammenhängender Functionen und die zu ihr gehörende
Klasse algebraischer Gleichungen von 3p−3 stetig veränderlichen Grössen
ab, welche die Moduln dieser Klasse genannt werden sollen.
Translation: The remaining 3p − 3 branch points in these systems of
similarly branching µ-valued functions can therefore be assigned any given
values; and thus a class of systems of similarly branching functions with
connectivity 2p + 1, and the corresponding class of algebraic equations,
depends on 3p − 3 continuous variables, which we shall call the moduli of
the class.
Bernhard Riemann [Rie57, p. 33]

This is a remarkable sentence in a remarkable paper—Riemann both introduces

the concept of ‘moduli’ and computes that the ‘number of moduli’ of Mg is 3g − 3.
Riemann’s idea went something like this: instead of considering abstract smooth

4
curves, let us view curves as branched covers over P1 and consider the moduli space

• C is a smooth curve of genus g

 
1
Hurd,g = [C → P ] .
• C → P1 is a simply branched covering of degree d
(0.1.1)
Formally studied later by Hurwitz [Hur91], these moduli spaces—which are now re-
ferred to as Hurwitz spaces—play an essential role in the study of Mg and specifically
in the proofs of irreducibility (see §5.7).
A simply branched covering is a finite map of smooth curves where the ramification
indices are at most two and every fiber has at most one ramification point; see
also Definition 5.7.2. By Riemann–Hurwitz (5.7.4), every simply branched covering
C → P1 is branched over 2g + 2d − 2 distinct points of P1 . This gives a commutative
diagram
 / Hurd,g
Hurd (C) (0.1.2)

&
 { / Mg
z
[C] Sym2d+2g−2 P1

where

– the map Hurd,g → Sym2d+2g−2 P1 sends a covering [C → P1 ] to the 2g +

2d − 2 branched points; here SymN P1 = (P1 )N /SN is the space classifying N
unordered points,
– the map Hurd,g → Mg is defined by [C → P1 ] 7→ [C], and
– Hurd (C) is the preimage of [C] ∈ Mg under Hurd,g → Mg , i.e., Hurd (C)
classifies simply branched coverings C → P1 where C is fixed.

If d is sufficiently large, then for every general collection of 2d + 2g − 2 points of P1 ,

there exists a genus g curve C and a simply branched covering C → P1 branched
over precisely these points, and moreover there are at most finitely many such maps.
In other words, Hurd,g → Sym2d+2g−2 P1 has dense image and finite fibers; see §5.7.2
for precise details. Therefore,

dim Mg = dim Hurd,g − dim Hurd (C)

(0.1.3)
= 2d + 2g − 2 − dim Hurd (C).

To compute the dimension of Hurd (C), we observe that a simply branched covering
C → P1 is the data of a degree d line bundle L (i.e., an element of Picd (C)) and two
base point free sections such that the induced map to P1 is simply branched. Since
a general choice of two sections defines a simply branched covering (Lemma 5.7.16),
we can compute

dim Hurd (C) = dim Picd (C) + 2h0 (C, L) − 1, (0.1.4)

where we subtract one since scaling two sections defines the same map to P1 .
Riemann–Roch (5.1.5) tells us that h0 (C, L) = d + 1 − g. On the other hand
dim Picd = dim Pic0 = g; this can be seen using the exponential sequence: 0 → Z →
exp
OC −−→ OC ∗
→ 0 yields a long exact sequence
deg
∗
H1 (C, Z) → H1 (C, OC ) → H1 (C, OC ) −−→ H2 (C, Z), (0.1.5)
| {z } | {z } | {z } | {z }
Z2g Cg Pic(C) Z

5
and provides an identification Pic0 (C) ∼
= Cg /Z2g . Therefore, dim Pic0 (C) = g, and
using (0.1.4), we compute that dim Hurd (C) = g + 2(d + 1 − g) − 1 = 2d − g + 1.
Plugging this into (0.1.3) yields

dim Mg = (2d + 2g − 2) − (2d − g + 1) = 3g − 3.

Riemann in fact gave several other heuristic arguments computing the dimension of
Mg . See [GH78, pp. 255-257] or [Mir95, pp. 211-215] for further discussion on the
number of moduli of Mg , and see [AJP16] for historical background of Riemann’s
computations.

Riemann’s moduli problem: Does Mg exist as a complex

analytic space?

While Riemann’s argument can be made completely rigorous with today’s meth-
ods (as we do ourselves later in this text), there are foundational issues with
Riemann’s method—today we would say that Riemann computed the dimension of
a ‘local deformation space’. Most notably, Mg was not known to exist and it was
not clear what type of space Mg was supposed to be. Despite this, Riemann had
an instinctive grasp of its geometry—in fact in the same paper [Rie57], Riemann
introduced the word ‘Mannigfaltigkeit’ (or ‘manifold’) to describe its geometry. Man-
ifolds were not formally defined until much later in the 1940s following Teichmüller,
Chern, and Weil.

0.1.2 Moduli of curves of low genus

Ritengo probabile che la varietà sia razionale o quanto meno che sia
riferibile ad un’involuzione di gruppi di punti in uno spazio lineare...

Francesco Severi’s conjecture that Mg is unirational for all

g [Sev15, p. 881].

Genus 0. For n ≥ 3, the moduli space M0,n of smooth genus 0 curves with n
ordered distinct points can be described as

n−3
M0,n = P1 ∖ {0, 1, ∞} ∖ (all diagonals).

Indeed, given n ordered distinct points p1 , . . . , pn on P1 , there is a unique auto-

morphism g ∈ Aut(P1 ) ∼ = PGL2 taking (p1 , p2 , p3 ) to (0, 1, ∞). When n = 4, we
obtain that a bijection M0,4 = P1 ∖ {0, 1, ∞} given by the classical cross-ratio of
four points in P1 first discovered by Pappus of Alexandria [Ale86] in 300 AD; see
also Example 7.2.8.

Genus 1. The moduli of elliptic curves was known to Dedekind and Klein. Every
elliptic curve (E, p), i.e., a smooth genus 1 curve E with a marked point p ∈ E, can
be described as a plane cubic in Weierstrass form

Eλ = V (y 2 z − x(x − z)(x − λz)) ⊆ P2

for some λ ̸= 0, 1, where p = [0 : 1 : 0] ∈ Eλ . However, the choice of λ is not unique:

the values λ, 1/λ, 1 − λ, 1/(1 − λ), λ/(λ − 1), and (λ − 1)/λ determine isomorphic

6
elliptic curves. In other words, the map A1 ∖ {0, 1} → M1,1 given by λ 7→ [Eλ ] is a
6-to-1 surjective map. The j-invariant on the other hand

(λ2 − λ + 1)3
j = 28
λ2 (λ − 1)2
uniquely determines the isomorphism class of the curve and thus gives a bijection
M1,1 ∼
= A1 . For a modern treatment, see [Har77, §4].

Genus 2. Every smooth genus 2 curve C is hyperelliptic and can be written as a

double cover y 2 = (x − a1 ) · · · (x − a6 ) over P1 . This is a consequence of the sheaf of
differentials ΩC being a base point free line bundle of degree 2 with 2 global sections;
the induced map C → P1 is ramified over 6 points by Riemann–Hurwitz (5.7.4). We
obtain the description that

M2 = Γ(P1 , O(6)) ∖ ∆ / GL2 ,

where ∆ ⊆ Γ(P1 , O(6)) denotes the locus of binary sextics with a double root. After
a projective change of coordinates on P1 , we can arrange that the curve is ramified
over 0, 1, ∞ and 3 other points a4 , a5 , a6 ∈ P1 ∖ (0, 1, ∞). In this way, we obtain a
surjective map (P1 ∖ {0, 1, ∞})3 ∖ ∆ → M2 .
Invariant theory of binary sextics (see [Cle70]) provides an even sharper descrip-
tion: the ring of invariant polynomials, i.e., polynomials in the coefficients of a binary
sextic that are invariant under automorphisms of P1 , is generated by invariants
J2 , J4 , J6 , J8 , J10 , and J15 , whose degree is indicated by the subscript, with a
single relation J15 2
= G(J2 , J4 , J6 , J10 ) for a polynomial G. The invariant J10 is the
discriminant of a binary sextic, while J15 does not affect the scheme structure. This
yields that M2 is an open subset of weighted projective space

M2 = Proj C[J2 , J4 , J6 , J10 ] ∖ {J10 = 0},

which implies that M2 is an affine variety embedded into A8 via

J25 J23 J4 J22 J6 J2 J42 J2 J63 J45 J4 J6 J65

, , , , 2 , 2 , , 3 .
J10 J10 J10 J10 J10 J10 J10 J10

We can identify this coordinate ring with the invariant ring of the action of Z/5
on A3 where a generator ζ ∈ Z/5 acts via ζ · (x, y, z) = (ζx, ζ 2 y, ζ 3 z); the above
functions are identified with the invariants x5 , x3 y, x2 z, xy 2 , xz 3 , y 5 , yz, z 5 . This
yields the rather elegant global description

M2 = A3 /(Z/5).

This was studied classically by Bolza [Bol87] and more recently by Igusa [Igu60].

Genus 3. A non-hyperelliptic smooth genus 3 curve embeds as a quartic in P2 under

the canonical embedding. Letting ∆ ⊆ Γ(P2 , O(4)) be the locus of singular quartics,
we can describe the

open locus in M3 of non-hyperelliptic curves as the quotient
Γ(P2 , O(4)) ∖ ∆ / GL3 —this is the first time that our description only describes a
general curve. On the other hand, a hyperelliptic genus 3 curve is a double cover of
P1 ramified over 8 points, and we obtain a set-theoretic decomposition
a
M3 = Γ(P2 , O(4)) ∖ ∆ / GL3 Γ(P1 , O(8)) ∖ ∆ / GL2 ,



| {z } | {z }
dim=6 dim=5

7
suggesting that the locus of hyperelliptic curves is a divisor in M3 .

Genus 4. A non-hyperelliptic smooth curve C of genus 4 embeds into P3 under

its canonical embedding, and can be realized as the intersection C = Q ∩ S of a
quadric surface Q and a cubic surface S. This gives a rational map Γ(P3 , O(2)) ×
Γ(P3 , O(3)) 99K M4 whose image is the locus of non-hyperelliptic

curves; as above the
hyperelliptic locus can be parameterized by Γ(P1 , O(10)) ∖ ∆ / GL2 . Alternatively,
a general non-hyperelliptic smooth genus 4 curve can be realized as the normalization
of a plane quintic with precisely two nodes, or as a degree 3 cover of P1 branched
over 12 points.

Genera 5–10. Classically, curves of low genus were described either as plane curves
with prescribed singularities via the image of a map C → P2 , or as branched covers
C → P1 . For a general genus g curve C, the smallest degree d such that C is realized
as the normalization of a singular plane curve is d = ⌊ 2g+8
3 ⌋. If the plane curve has
at worst nodal singularities, then the number of nodes is δ := (d − 1)(d − 2)/2 − g.
Meanwhile, the minimum degree of a map C → P1 is ⌊ g+3 2 ⌋. See also [Mum75a,
p. 21].

(d+1)(d+2)
g d = min degree δ = # of nodes 2 − 3δ min degree # of branch pts
of im(C → P2 ) of C → P1
0 1 0 3 1 0
1 3 0 10 2 4
2 4 1 12 2 6
3 4 0 15 3 10
4 5 2 15 3 12
5 6 5 13 4 16
6 6 4 16 4 18
7 7 8 12 5 22
8 8 13 6 5 24
9 8 12 9 6 28
10 9 18 1 6 30
11 10 25 -9 7 34

Table 0.1.6: General curves of low genus.

In [Sev15] and [Sev21], Severi used such descriptions to show that Mg is unira-
tional for g ≤ 10. Like other mathematicians of his era, Severi did not precisely
formulate what it meant for Mg to be a moduli space.
What goes wrong for g ≥ 11? As the genus grows, it becomes more difficult
to describe a general genus g curve. To give an indication of the challenges for
g ≥ 11, let us try to describe a genus g curve as a degree d = ⌊ 2g+8 3 ⌋ planar curve
with δ = (d − 1)(d − 2)/2 − g nodes at prescribed points p1 , . . . , pδ ∈ P2 . If the
plane curve defined by f ∈ Γ(P2 , O(d)) has a node at each pi , then the equations
fx (pi ) = fy (pi ) = fz (pi ) = 0 imposes 3δ linear equations on Γ(P2 , O(d)). For such
nodal plane curves to exist, we would need
 
(d + 1)(d + 2) (d − 1)(d − 2)
dim Γ(P2 , O(d)) − 3δ = −3 − g > 0.
2 2

8
As illustrated by Table 0.1.6, g = 11 is the first case where this does not hold!
Severi’s conjecture. While Severi’s conjecture that Mg is unirational for all
g turned out to be false, it nevertheless motivated mathematicians for decades:
“Whether more Mg ’s, g ≥ 11, are unirational or not is a very interesting problem,
but one which looks very hard too, especially if g is quite large” [Mum75a, p. 37].
In the 1980s, Eisenbud, Harris, and Mumford disproved this conjecture by showing
that in some sense quite the opposite is true in large genus: Mg is of general type
for g ≥ 24 [HM82], [EH87].

Petri’s description of canonical curves.

[Petri’s approach] is unavoidably a bit messy, but just to be able to brag,

I think it is a good idea to be able to say ‘I have seen every curve once.’

David Mumford [Mum75a, p.17]

While most 19th and early 20th century mathematicians described curves as either
plane singular curves or as covers of P1 , Petri’s explicit description [Pet23] of
canonically embedded curves was an exception and is more reminiscent of modern
approaches. Building on M. Noether’s result [Noe80] that the canonical embedding
C ,→ Pg−1 of a non-hyperelliptic smooth curve C is projectively normal—that is,
φ : Sym∗ Γ(C, ΩC ) → d≥0 Γ(C, Ω⊗d C ) is surjective—and also building on work of
L
Enriques [Enr19] and Babbage [Bab39], Petri showed that the homogeneous ideal
I = ker φ is generated by quadrics unless C is a plane quintic (g = 6) or trigonal
(i.e., a triple covering of P1 ) in which case I is generated in degree 2 and 3. Petri’s
analysis was remarkably constructive leading to explicit equations in Pg−1 cutting
out C along with explicit syzygies among the equations. Petri’s work continues
to inspire research in the theory of moduli and syzygies. We will not cover this
perspective further in this text but we recommend [SD73], [Mum75a, pp. 17-21],
[AS78], [Gre82], [Gre84], and [ACGH85, §III.3].

0.1.3 Analytic approaches and the Teichmüller space

It does not of necessity follow that, if the work delights you with its grace,
the one who wrought it is worthy of your esteem.

Plutarch, The Parallel Lives

In the late 19th and early 20th century, Riemann surfaces were described as
quotients of the upper half plane by a discrete subgroup of PSL2 (R); such subgroups
are named Fuchsian groups after Fuchs [Fuc66]. Fricke and Klein classified Fuchsian
groups using the theory of automorphic functions in their 1300 page volumes [FK92],
[FK12]. They constructed what is now known as the Teichmüller space, showed that
it is a contractible space, and even exhibited complex structures. Torelli showed
that a Riemann surface can be constructed from its Jacobian [Tor13], and Siegel
constructed the moduli space Ag of abelian varieties of dimension g as an analytic
space [Sie35].
Oswald Teichmüller was the first to give a precise formulation of Riemann’s
moduli problem, to construct Mg as a complex analytic space, and to interpret 3g − 3
as its complex dimension [Tei40], [Tei44]. Teichmüller constructed the Teichmüller
space Tg parameterizing complex structures on a topological surface Σg of genus

9
g up to homeomorphism. The space Tg is homeomorphic to a ball in C3g−3 and
inherits an action of the mapping class group Γg of diffeomorphisms of Σg modulo
the subgroup of diffeomorphisms isotopic to the identity. This action is properly
discontinuous, and Mg is realized as the quotient Tg /Γg . Although largely forgotten
for nearly 20 years, Teichmüller theory was later greatly expanded by Ahlfors, Bers,
and Weil among others; see [Wei57], [Wei58], [AB60], [Ber60], and [Ahl61]. For
modern expository treatments, see [Ber72], [Hub06], and [FM12].
Teichmüller also introduced the notion of families of Riemann surfaces and showed
that the Teichmüller space satisfies a universal property. Grothendieck, in a series
of ten lectures at Cartan’s seminar [Gro61], developed a general theory of analytic
moduli spaces in the language of categories and functors, reformulated Teichmüller
theory in this setting, and showed that Tg represents a functor parameterizing
families of Riemann surfaces. This set the stage for Grothendieck’s later work on
algebraic moduli: “One can hope that we shall be able one day to eliminate analysis
completely from the theory of Teichmüller space, which should be purely geometric”
[Gro61, Lecture I].

0.1.4 The origins of algebraic moduli theory

As for Mg there is virtually no doubt that it can be provided with the
structure of an algebraic variety.

André Weil [Wei58, p. 383]

Boole, Cayley, Gordan, and Hilbert. The invariant theorists of the 19th
century were interested in classifying homogeneous polynomials of degree d in n
variables up to projective automorphisms, or in other words in the moduli space
Γ(Pn−1 , O(d))/ PGLn . They attempted to describe this moduli space by exhibiting
explicit invariant polynomials in the coefficients aI of a polynomial f = I aI xI .
P
The origins of invariant theory lie in work of George Boole [Boo41] and Arthur
Cayley [Cay45], and were further developed by Paul Gordan and David Hilbert along
with many others. Gordan exhibited explicit generators of the ring of invariants of
binary forms (n = 2) [Gor68] and Hilbert later proved that the ring of invariants
is finitely generated for any ring [Hil90], [Hil93]. We prove and discuss Hilbert’s
theorem in Corollary 6.5.8(3) and Remark 6.5.10.
Cayley constructed the moduli space of curves in P3 [Cay60], [Cay62], which is
now referred to as the Chow variety. His idea was to associate to a degree d curve
C ⊆ P3 the set of lines L ⊆ P3 meeting C non-trivially. This is a hypersurface of
degree d in Gr(1, P3 ), and the Chow variety is the closure of all hypersurfaces in
Γ(Gr(1, P3 ), O(d)) obtained this way. This construction was generalized later by
Chow and van der Waerdan [CW37] to subvarieties of Pn of arbitrary dimension;
see Section 1.4.5. Chow varieties will play a role in the construction of M g in §5.8.

Weil. In André Weil’s work on the Riemann hypothesis for curves over finite fields
[Wei48], he needed to construct the Jacobian of a curve parameterizing degree 0 line
bundles. At that point, varieties had only been considered as embedded in affine or
projective space, and in his foundational work [Wei62], Weil enlarged the category to
abstract varieties. This was enough to construct the Jacobian and give a proof—in
fact his second proof—of the Riemann hypothesis for curves. Later Weil and Chow
independently showed that the Jacobian was projective.

10
Baily. Walter Baily constructed the moduli space Ag of principally polarized
abelian varieties as a quasi-projective variety [Bai60a], [Bai60b], showed that Satake’s
topological compactification [Sat56] is algebraic [Bai58], and together with Borel
introduced what is now known as the Baily–Borel compactification [BB66]. Using the
period map Mg → Ag associating a curve to its Jacobian and Torelli’s theorem that
this map is injective, Baily concluded that Mg has the structure of a quasi-projective
variety. However, he did not prove that this provided a ‘natural’ structure of a
variety nor that it had any uniqueness properties, i.e., that Mg is a coarse moduli
space.

Thirdly, in order to call E the variety of moduli of Riemann surfaces

of genus n, one should be able to state that it is unique and in some
sense universal among normal parameter varieties of algebraic systems of
curves of genus n. Namely, given any normal algebraic system of curves
of genus n (by which we mean that the parameter variety is a normal
variety) there should exist a natural map of the parameter variety of the
nonsingular members of this system into E. — Baily [Bai60b, pp. 59-60]

Mumford credits Baily for the quasi-projectivity of Mg in [Mum75a, p. 98] just as

Gieseker does in his commentary in [Mum04].

Grothendieck. After Alexander Grothendieck’s formalization of analytic moduli

theory in [Gro61], he applied his functorial approach to algebraic geometry in his
‘FGA series’ [FGAI]–[FGAVI]: he introduced the Hilbert, Quot, and Picard functors,
showed that they were representable by projective schemes, developed descent theory,
and introduced the notions of prestacks and stacks. Grothendieck of course later
redeveloped the entire foundations of algebraic geometry by developing scheme
theory. His profound influence on algebraic geometry and more broadly mathematics
helped shape the future of moduli theory.
Although he did not publish on Mg , Grothendieck was nevertheless very much
interested in the existence of Mg as a quasi-projective variety and its connectedness in
any characteristic, as demonstrated from his written correspondence with Mumford
and others in the early 1960s [Mum10, §II]. Grothendieck was aware that the presence
of automorphisms obstructed the representability of the functor parameterizing
smooth families of curves. He rigidified the moduli problem by also parameterizing
a level n structure on a curve, i.e., a symplectic basis of H1 (C, Z/nZ). While
he could show that the functor of smooth curves with level structure n ≥ 3 was
representable by a scheme, he struggled to show that it was quasi-projective. The
idea was to construct Mg as a quotient of the rigidified moduli space by taking the
quotient by the finite group acting on the choice of level n structure. The lack of
quasi-projectivity impeded this approach as the quotient of a non-quasi-projective
variety by a finite group need not exist as a variety.

Mumford. Motivated by Riemann’s moduli space as well as by constructions of

Chow varieties, Picard varieties, and the moduli of abelian varieties in the early 20th
century, David Mumford made immense contributions to the foundations of moduli
theory, and was the first to systematically study their geometry. By integrating
Grothendieck’s formalism of scheme theory with 19th century invariant theory,
Mumford developed a theory of quotients in algebraic geometry now known as
Geometric Invariant Theory (or GIT ), which we develop in Chapter 7. Mumford
applied his theory of GIT to construct both Mg and Ag . His theory was originally
sketched in [Mum61] and fully worked out in the definitive text [GIT]. In fact,

11
Mumford gave two constructions [GIT, Thms. 5.11 and 7.13] of the coarse moduli
scheme Mg over Spec Z and moreover that Mg is quasi-projective over Spec Z[1/p]
for every prime p.1 The projectivity of M g over Spec Z was established later by
other methods [Knu83b, Mum77, Gie82], which were more directly applicable to
other moduli spaces.
Mumford also constructed a quasi-projective variety parameterizing stable vector
bundles on a fixed smooth curve [Mum63], and Seshadri then showed that the moduli
space of semistable vector bundles provides a projective compactification [Ses67].
In the seminal joint work [DM69], Deligne and Mumford introduced stable curves
and the compactification Mg of Mg , proved the Stable Reduction Theorem (5.5.1),
and were the first to introduce the notion of an algebraic stack—now referred to
as Deligne–Mumford stacks. Finally, they offered two proofs2 of the connectedness
of Mg in any characteristic; see §5.7. Unfortunately, Deligne and Mumford did not
prove any of their statements on stacks: “The proofs of the results of this section
will be given elsewhere” [DM69, p. 76]. The lack of rigorous foundations contributed
to the formidable reputation of algebraic stacks over the following decades. It wasn’t
until the 2000s that algebraic stacks entered into mainstream algebraic geometry.
In the 50+ years following Deligne and Mumford’s paper, most of their statements
now have full proofs in the literature. There are now excellent textbooks covering
algebraic stacks [LMB00], [Ols16], and Johan de Jong’s Stacks Project [SP] has
provided an unquestionably solid foundation.

Artin. The theory of algebraic spaces and stacks was developed by Michael Artin.
Similar to Weil’s enlargement of affine and projective varieties to abstract varieties,
enlarging the category of schemes to algebraic spaces allows us to construct the
quotient of finite group actions or more generally any étale equivalence relation.3
Knutson, a student of Artin, was the first to write down the theory of algebraic
spaces [Knu71].
In the two papers [Art69a] and [Art69b], Artin proved two crucial results in
moduli theory: Artin Approximation (B.5.18) and Artin Algebraization (C.6.8). In
his groundbreaking paper [Art74b], Artin introduced a broader concept of algebraic
stacks than Deligne and Mumford including stacks such as Bun r,d (C) with possibly
infinite automorphism groups. He also provided a local deformation-theoretic
characterization of algebraic stacks called Artin’s Axioms for Algebraicity (C.7.1
and C.7.4), which can be applied to verify that a given moduli stack is algebraic;
see §C.7.2.

For further historical background, we recommend [Mum75a], [Oor81], [Kle05],

[JP13], [AJP16], and [Kol21].

1 Interestingly, neither of Mumford’s constructions actually uses GIT, or at least what is often

considered as the ‘standard GIT machinery’ by verifying GIT stability using the Hilbert–Mumford
Criterion. One of Mumford’s constructions relies on the existence of the moduli space Ag of
principally polarized abelian varieties, and the other on ad hoc method using covariants.
2While both proofs used the compactification M and the Stable Reduction Theorem in a
g
fundamental way, the use of algebraic stacks was not essential.
3 Matsusaka also built a theory of Q-varieties by considering certain quotients of equivalence

relations [Mat64] but it was not as robust as the theory of algebraic spaces.

12
0.2 Moduli sets of curves, vector bundles, and tri-
angles
To understand stacks, you really must first understand the moduli space
of triangles.

attributed to Michael Artin

To define a moduli space as a set entails specifying two things:

(1) a class of certain types of objects, and
(2) an equivalence relation on objects.
Here is our first attempt at defining Mg :
Example 0.2.1 (Moduli set of smooth curves). The objects of the moduli set of
smooth curves, denoted as Mg , are smooth, connected, and projective curves of
genus g over C. Two curves are declared equivalent if they are isomorphic. There are
many variants obtained by parameterizing additional structures or choosing different
equivalence relations.
– We already saw the Hurwitz moduli set Hurd,g in (0.1.1) parameterizing
branched covers C → P1 of degree d.
– The moduli set Mg,n of n-pointed smooth genus g curves parameterizes the data
of a smooth curve C together with n ordered distinct points p1 , . . . , pn ∈ C; two
objects (C, pi ) ∼ (C ′ , p′i ) are equivalent if there is an isomorphism α : C → C ′
with α(pi ) = p′i .
– The moduli set Mg [n] of smooth genus g curves with level n structure parame-
terizes smooth, connected, and projective curves C of genus g over C together
with a basis (α1 , . . . , αg , β1 , . . . , βg ) of H1 (C, Z/nZ) such that the intersection
pairing is symplectic, while two objects (C, αi , βi ) ∼ (C ′ , αi′ , βi′ ) are declared
equivalent if there is an isomorphism C → C ′ taking αi and βi to αi′ and βi′ .
– For the moduli set whose objects are plane curves C ⊂ P2 , there are several
choices for equivalence relations C ∼ C ′ : (a) C and C ′ are equal as subschemes,
(b) C and C ′ are projectively equivalent (i.e., there is an automorphism of P2
taking C to C ′ ), or (c) C and C ′ are abstractly isomorphic.
Example 0.2.2 (Moduli set of vector bundles on a curve). The moduli set Bunr,d (C)
parameterizes vector bundles of rank r and degree d on a fixed smooth, connected, and
projective curve C; the equivalence relation here is isomorphism. The special case of
r = 1 yields the set Picd (C) parameterizing degree d line bundles on C. This is non-
canonically identified with with the abelian variety H1 (C, OC )/H1 (C, Z) = Cg /Z2g
via the exponential exact sequence (0.1.5).
A recurring theme in moduli is the presentation of moduli spaces as quotients of
group actions.
Example 0.2.3 (Moduli set of orbits). Given a group action of a group G on a set
X, we define the moduli set of orbits by taking the objects to be all elements x ∈ X
and by declaring x to be equivalent to x′ if they have the same orbit Gx = Gx′ . In
other words, the moduli set of orbits is the quotient set X/G.
Some examples to keep in mind are the Z/2-action on A1 via (−1) · x = −x
and the usual scaling action of Gm on An via t · (x1 , . . . , xn ) = (tx1 , . . . , txn ). The

13
quotient set (An ∖ 0)/Gm is identified with Pn−1 . The quotient An /Gm including
the origin—and particularly the case of A1 /Gm —shows up repeatedly in this text.
Another interesting example is the Gm -action on A2 given by t · (x, y) = (tx, t−1 y).

0.2.1 Toy example: moduli of triangles

Before diving deeper into Mg and Bunr,d (C), let us study the simple yet surprisingly
fruitful example of the moduli of triangles. These moduli spaces are easy to visualize
and are useful to illustrate various themes of stacks and moduli.
Example 0.2.4 (Labeled triangles). A labeled triangle is a triangle in R2 where the
vertices are labeled with ‘1’, ‘2’ and ‘3’, and the distances of the edges are denoted
as a, b, and c. We require that triangles have nonzero area or equivalently that their
vertices are not collinear.
2
b

a 3 Figure 0.2.5: To keep track of the

labeling, we color the edges.
c
1

We define the moduli set of labeled triangles M as the set of labeled triangles
where two triangles are said to be equivalent if they are the same triangle in R2 with
the same vertices and same labeling. By writing (x1 , y1 ), (x2 , y2 ), and (x3 , y3 ) as
the coordinates of the labeled vertices, we obtain a bijection
   
x2 − x1 x3 − x1
M∼ = (x1 , y1 , x2 , y2 , x3 , y3 ) | det ̸= 0 ⊆ R6 (0.2.6)
y2 − y1 y3 − y1

with the open subset of R6 whose complement is the codimension 1 closed subset
defined by the condition that the vectors (x2 , y2 ) − (x1 , y1 ) and (x3 , y3 ) − (x1 , y1 )
are linearly dependent.

y3

Figure 0.2.7: Picture of the slice

of the moduli space M where
(x1 , y1 ) = (0, 0) and (x2 , y2 ) =
x3 (1, 0). Triangles are described by
their third vertex (x3 , y3 ) with
y3 ̸= 0. We have drawn represen-
tative triangles for a handful of
points in the x3 y3 −plane.

Example 0.2.8 (Labeled triangles up to similarity). We define the moduli set of

labeled triangles up to similarity, denoted by M lab , by taking the same class of

14
objects as in the previous example—labeled triangles—but changing the equivalence
relation to label-preserving similarity.

similar not similar

Figure 0.2.9: The two triangles on the left are similar, but the third is not.

Every labeled triangle is similar to a unique labeled triangle with perimeter

a + b + c = 2. We have the description
 

 a+b+c=2 

0<a<b+c
 
M lab = (a, b, c) . (0.2.10)

 0<b<a+c 

0<c<a+b
 

By setting c = 2 − a − b, we may visualize M lab as the analytic open subset of R2

defined by pairs (a, b) satisfying 0 < a, b < 1 and a + b > 1.

Figure 0.2.11: M lab is the

shaded triangle. The magenta
lines represent the right trian-
gles defined by a2 + b2 = c2 ,
a2 + c2 = b2 , and b2 + c2 = a2 ,
the blue lines represent isosce-
les triangles defined by a = b,
b = c, and a = c, and the
green point is the unique equi-
lateral triangle defined by a =
b = c.

Example 0.2.12 (Unlabeled triangles up to similarity). We now turn to the moduli

of unlabeled triangles up to similarity, which reveals a new feature not seen in to
the two previous examples: symmetry!
We define the moduli set of unlabeled triangles up to similarity, denoted by
M unl , where the objects are unlabeled triangles in R2 and the equivalence relation
is similarity. We can describe an unlabeled triangle uniquely by the ordered tuple
(a, b, c) of increasing side lengths:
 
0<a≤b≤c<a+b
M unl = (a, b, c) . (0.2.13)
a+b+c=2

15
 Figure 0.2.14: Picture of M unl .

The isosceles triangles with a = b or b = c and the equilateral triangle with

a = b = c have symmetry groups of Z/2 and S3 , respectively. This is unfortunately
not encoded into our description M unl above. Note that we can identify M unl as
the quotient M lab /S3 under the natural action of S3 on the labelings, and that the
stabilizers of isosceles and equilateral triangles are precisely their symmetry groups
Z/2 and S3 . The action of S3 on the locus of triangles that are not isosceles nor
equilateral is free.

0.3 The functorial worldview

Mathematical objects are determined by—and understood by—the network
of relationships they enjoy with all the other objects of their species.

Barry Mazur [Maz08]

Defining a moduli functor requires specifying:

(1) families of objects,
(2) when two families of objects are equivalent, and
(3) how families pull back under morphisms.
In the algebraic category, this is packaged with a contravariant functor

F : Sch → Sets, S 7→ {families of objects over S}, (0.3.1)

where for a map f : S → T , F (f ) : F (T ) → F (S) gives the pullback map—sometimes

written simply as f ∗ —from a family over T to a family over S. (To be a functor,
pullback maps must commute with composition: F (g ◦ f ) = F (f ) ◦ F (g) for maps
f : S → T and g : T → U .)

0.3.1 Family matters

Families allow us to provide a precise formulation of a moduli space M . A family of
objects C over a space S defines a set-theoretic map

S → M, s 7→ [Cs ], (0.3.2)

16
where the fiber [Cs ] ∈ M is the pullback of C under the inclusion {s} ,→ S. In the
topological (resp., algebraic) category, we desire that the map S → M is continuous
(resp., algebraic). Ideally, there is a bijective correspondence between families over S
and morphisms S → M , or in other words that the space M represents the functor
F of (0.3.1). If this happens, we call M a fine moduli space, but we will see shortly
that this is often too much to hope for.

Defining moduli spaces via families has advantages:

– We can endow the moduli set M with enriched structures. To provide M a

topology, we declare a subset U ⊆ M to be open if for every family of objects C
over S, the locus {s ∈ S | [Cs ] ∈ U } is an open subset of S. A global function
f on M can be defined as the data of compatible global functions fX on S for
every family X over S.

– When M is a fine moduli space, the identity map id : M → M corresponds to

a family of objects U over the moduli space M . This is the universal family:
for any other family C over S, there is a unique morphism S → M (given
set-theoretically by (0.3.2)) such that the universal family U pulls back to C.

This is certainly a giant leap in abstraction! And it may seem that we just
made life more difficult: rather than introducing a space by specifying its points, its
topology, and possibly a complex or algebraic structure, we must specify an immense
amount of categorical data. In practice, however, it is usually quite straightforward
to define well-behaved notions of families. This change of perspective is enshrined in
Yoneda’s Lemma (0.3.12), which asserts that the functor of maps to a space uniquely
determines the space.

Example 0.3.3 (Families of labeled triangles). Revisiting the moduli of labeled

triangles up to similarity introduced in Example 0.2.8, we define a family of labeled
triangles over a topological space S as a tuple (T , σ1 , σ2 , σ3 ) where T → S is a
fiber bundle with three sections σi : S → T equipped with a continuous distance
function d : T ×S T → R≥0 . We require that for every point s ∈ S, the restriction
ds : Ts × Ts → R≥0 is a metric on the fiber Ts such that Ts isometric to a triangle
with vertices σi (s).

We say two families (T , (σi )) and (T ′ , (σi′ )) of labeled triangles over S ∈ Top are
similar if there is a homeomorphism f : T → T ′ over S compatible with the sections
(i.e., f ◦ σi = σi′ ) such that for each s ∈ S, the induced map Ts → Ts′ on fibers is a
similarity of triangles, i.e., an isometry after rescaling. Given a family T → S of
labeled triangles and a continuous map S ′ → S, the pullback family is defined as
the fiber product T ×S S ′ of sets together with the pullback sections σi′ : S ′ → T ′
and its inherited distance function.

17
Figure 0.3.4: A family of labeled triangles over a curve corresponds to an arc in the
moduli space.

We define the moduli functor of labeled triangles as

FM lab : Top → Sets, S 7→ {families (T → S, σi ) of labeled triangles}/(similarity).

Recall from (0.2.10) that the assignment of a triangle to its side lengths yields a
bijection between FM lab and
 

 a+b+c=2 

0<a<b+c
 
lab
M = (a, b, c) .

 0<b<a+c 

0<c<a+b
 

Since this extends to compatible isomorphisms FM lab (S) → Mor(S, M lab ) for every
topological space S, the topological space M lab represents the functor FM lab . Con-
sequently, there is a universal family Tuniv → M lab with sections σi : M lab → Tuniv .

Figure 0.3.5: The universal family U lab → M lab of labeled triangles up to similarity.

18
Example 0.3.6 (Families of unlabeled triangles). Revisiting Example 0.2.12, we
define a family of unlabeled triangles as a fiber bundle T → S equipped with a
continuous distance function d : T ×S T → R≥0 that restricts to a metric on every
fiber and such that every fiber is isometric to a triangle. Two families T → S and
T ′ → S are similar if there is a homeomorphism f : T → T ′ over S compatible with
the sections inducing similarities of triangles on fibers.
We define the functor

FM unl : Top → Sets, S 7→ {families of unlabeled triangles} /(similarity)

but we can already see complications arising from the presence of symmetries of our
objects—each equilateral triangle has symmetry group S3 while the isosceles triangles
have symmetry groups Z/2. This functor is not representable by Proposition 0.3.29as
there are non-trivial families of triangles T such that all fibers are similar triangles .
For instance, we construct a non-trivial family of triangles over S 1 by gluing two
trivial families via a symmetry of an equilateral triangle.

Figure 0.3.7: A trivial (left) and non-trivial (right) family of equilateral triangles.
Image taken from a video produced by Jonathan Wise: see [Link]
edu/~[Link]/visual/moduli/[Link].

0.3.2 Moduli functors of curves, vector bundles, and orbits

Defining a moduli functor F : Sch/C → Sets in the category of C-schemes entails

specifying for every C-scheme S a set F (S) of families of objects over S, and a
pullback map F (S) → F (S ′ ) for every morphism S ′ → S of C-schemes which are
compatible under composition.
To gain intuition for a moduli functor, it is always useful to plug in special test
schemes. For instance, by plugging in S = Spec C, we obtain the underlying moduli
set F (Spec C) of objects. By plugging in S = C[ϵ]/(ϵ2 ), we obtain a set of pairs
consisting of a C-point and a tangent vector, and plugging in a curve (or a DVR)
gives families of objects over the curve.

Example 0.3.8 (Moduli functor of smooth curves). A family of smooth curves of

genus g is a smooth, proper morphism C → S of schemes such that for every s ∈ S,
the fiber Cs is a connected (smooth proper) curve of genus g.

19
 Cs C
Ct

s S

t
Figure 0.3.9: A family of curves over a curve S.

The moduli functor of smooth curves of genus g is

FMg : Sch/C → Sets, S 7→ {families of smooth curves C → S of genus g} / ∼,

where two families C → S and C ′ → S are equivalent if there is an isomorphism

C → C ′ over S. If S ′ → S is a map of C-schemes and C → S is a family of curves,
the pullback is defined as the family C ×S S ′ → S ′ .
Example 0.3.10 (Moduli functor of vector bundles on a curve). Let C be a fixed
smooth, connected, and projective curve over C. A family of vector bundles of rank
r and degree d over a C-scheme S is a vector bundle E on C × S such that for every
s ∈ S, the restriction Es := E|Cκ(s) of E to Cκ(s) := C ×C κ(s) has rank r and degree
d. The moduli functor of vector bundles on C of rank r and degree d is

vector bundles E on C × S 
 
Sch/C → Sets S 7→ (isomorphism),
of rank r and degree d

If f : S ′ → S is a map of C-schemes and E is a vector bundle on C × S, the pullback

is defined as the vector bundle (id ×f )∗ E on C × S ′ .
We will see in Section 0.3.5 that these two functors are not representable, and
correspondingly that there is no fine moduli space.
Example 0.3.11 (Moduli functor of orbits). Consider the action of an algebraic
group G over C acting on a C-scheme X. For every C-scheme S, the abstract group
G(S) acts on the set X(S)—in fact, giving such actions functorial in S uniquely
specifies the group action (Exercise B.1.10). We can consider the functor

Sch/C → Sets S 7→ X(S)/G(S).

This is a very naive candidate for a moduli functor of a quotient, and very far from
being representable even for free actions (see Exercise 0.3.37). We will modify this
example in §0.6.5.
In some cases, you may know precisely which objects you want to parameterize,
but it may not be straightforward to introduce a notion of families. Or there may be

20
several candidate notions for a family of objects, which could translate to different
scheme structures on the same topological space. This happens for instance for the
moduli of higher dimensional varieties.

0.3.3 Yoneda’s lemma and representable functors

The Yoneda lemma is the hardest trivial thing in mathematics.

Dan Piponi

Following Grothendieck, we will study a scheme X by studying all maps to it!

This psychological trick is the heart of the functorial approach: a geometric object
is determined by its relationship to all other objects. This is made rigorous by the
Yoneda Lemma: for an object X of a category C, the contravariant functor
hX : C → Sets, S 7→ Mor(S, X)
recovers the object X itself.
Lemma 0.3.12 (Yoneda Lemma). Let C be a category and X be an object. For
every contravariant functor G : C → Sets, the map
Mor(hX , G) → G(X), α 7→ αX (idX )
is bijective and functorial with respect to both X and G, where the left-hand side
denotes the set of natural transformations hX → G and αX denotes the map
hX (X) = Mor(X, X) → G(X).
Exercise 0.3.13. (to be done at least once in your lifetime) Prove Yoneda’s lemma.
This requires spelling out precisely what ‘functorial with respect to both X and G’
means.
Caution 0.3.14. Throughout this book, we will consistently abuse notation by
conflating an element g ∈ G(X) and the corresponding morphism hX → G, which
we will often write simply as X → G.
Definition 0.3.15 (Representable functors and fine moduli spaces). We say that a
functor F : Sch → Sets is representable by a scheme if there exists a scheme X and
∼
an isomorphism of functors F → hX .
When F is a moduli functor representable by a scheme M , we say that M is a
fine moduli space.
By the Yoneda Lemma (0.3.12), if a functor is representable, then it is repre-
sentable by a unique scheme. One of our aims is to understand when a given moduli
functor F has a fine moduli space, i.e., is representable by a scheme.
Example 0.3.16 (Projective space as a functor). By [Har77, Thm. II.7.1], there is
a functorial bijection
L is a line bundle on S globally
 
Mor(S, PnZ ) ∼
= (L, s0 , . . . , sn ) / ∼,
generated by s0 , . . . , sn ∈ Γ(S, L)
where (L, si ) ∼ (L′ , s′i ) if there exists an isomorphism α : L → L′ such that si = α∗ s′i
for all i. In other words, the functor on the right is representable by the scheme PnZ .
The condition that the sections si are globally generated translates to the condition
that for every x ∈ S, at least one section si (x) ∈ L ⊗ κ(x) is nonzero, or equivalently
to the surjectivity of (s0 , . . . , sn ) : OS⊕n+1 → L.

21
Example 0.3.17 (The Grassmannian functor). As a set, the Grassmannian Gr(q, n)
parameterizes q-dimensional quotients of n-dimensional space.4 But what are families
of q-dimensional quotients over a scheme S? A naive guess might be quotients
p : OSn ↠ OSk but this has no chance to be representable (see Exercise 0.3.37). The
case of projective space suggests we define the Grassmannian functor as

Gr(q, n) : Sch → Sets

 
S 7→ OS⊕n ↠ Q Q is a vector bundle of rank q
 
/ ∼,

∼
where [p : OS⊕n ↠ Q ∼ [p′ : OS⊕n ↠ Q′ if there exists an isomorphism Ψ : Q → Q′
 

such that
OS⊕n
p
/Q

Ψ
p′
! 
Q′
commutes (i.e., p′ = Ψ ◦ p), or equivalently if ker(p) = ker(p′ ). Pullbacks are defined
in the natural manner.
We will later show that Gr(q, n) is representable by a scheme projective over
Z (Theorem 1.1.1). The proof of this result is a good illustration of the utility of
the functorial approach and a warmup for the representability of Hilb and Quot
(Theorems 1.1.2 and 1.1.3).
These exercises will give you some practice with the functorial approach.
Exercise 0.3.18 (Affine and Projective Space). Let S be a scheme and E be a
vector bundle on S.
(a) Show that the affine space A(E) := SpecS (Sym∗ E) represents the functor
assigning f : T → S to Γ(T, f ∗ E ∨ ), where E ∨ := H omOS (E, OS ). Note that
a k-point of A(E) is an element of the dual of E ⊗OS k.5 Observe also the
special case A(OS⊕n ) ∼
= AnS .
(b) Show that the projectivization P(E) := ProjS (Sym∗ E) of E represents the
functor
Sch/S → Sets
f
→ S) 7→ {quotients q : f ∗ E ↠ L where L is a line bundle on T }/ ∼
(T −

where [q : f ∗ E ↠ L] ∼ [q ′ : f ∗ E ↠ L′ ] if ker(q) = ker(q ′ ) (or equivalently

there is an isomorphism α : L → L′ with q ′ = α ◦ q). Note that E is naturally
identified with the pushforward of OP(E) (1) along P(E) → S, and that when
E is trivial, there is an identification P(OSn+1 ) ∼
= Pn .
Exercise 0.3.19. Provide functorial descriptions of:
(a) An ∖ 0;
(b) Spec k[x1 , . . . , xn ]/(f1 , . . . , fm );
(c) SpecS A where A is a quasi-coherent sheaf of algebras on a scheme S; and
4 Alternatively, the points could be considered as q-dimensional subspaces but in these notes, we

will follow Grothendieck’s convention using quotients.

5 This is consistent with [Har77, Exc. 5.18, Def. p.162], [EGA, II.4.1.1], and [SP, Tag 01OB],

but beware that some authors use the dual E ∨ instead of E in defining A(E) and P(E).

22
 (d) Proj R where R is a positively graded ring.

Exercise 0.3.20 (moderate, good practice). Let X be a scheme and let F1 and F2
be vector bundles on X. Show that the functor

H omOX (F1 , F2 ) : Sch/X → Sets, (f : T → X) 7→ HomOT (f ∗ F1 , f ∗ F2 )

is representable by A(H omOX (F1 , F2 )∨ ) → X.

Note the distinction between HomOX (F1 , F2 ) (group of OX -module homomorphisms),
H omOX (F1 , F2 ) (sheaf of OX -module homomorphisms), and H omOX (F1 , F2 ) (func-
tor or scheme parameterizing OX -module homomorphisms).

Exercise 0.3.21 (moderate). Let X be a scheme over a field k and let F1 and F2
be vector bundles on X. Show that the functor

Ext1OX (F1 , F2 ) : AffSch/k → Sets, T 7→ Ext1OX×T (p∗1 F1 , p∗1 F2 ),

from the category of affine k-schemes, is representable by A(Ext1OX (F1 , F2 )∨ ). See

also [LP97, §7.3] and Proposition 8.2.11.

Exercise 0.3.22 (moderate). Let S be a noetherian scheme and let E → F be

homomorphism of coherent sheaves on PnS with F flat over S. Show that the
subfunctor of S (or more precisely of hS = Mor(−, S)) defined by

Sch → Sets, T 7→ {morphisms T → S such that ET → FT is zero}

is representable by a closed subscheme of X.

Exercise 0.3.23 (Weil Restriction, hard). If S ′ → S is a morphism of schemes, the

Weil restriction of a morphism X ′ → S ′ is the functor

ReS ′ /S (X ′ ) : Sch/S → Sets, (T → S) 7→ X ′ (T ×S S ′ ).

(1) If k/k′ is a field extension of degree d, show that Rek′ /k A1 ∼

= Ad .
(2) Show that T := ReC/R (Gm,C ) is an algebraic group over R, which is a non-split
torus of rank 2, i.e., T ×R C ∼= G2m,C but T ≇ G2m,R .
(3) (hard) Assume that S ′ → S is finite and flat, and that for every s ∈ S, every
finite set of points of the fiber Xs′ is contained in an affine, then ReS ′ /S (X ′ ) is
representable. See also [BLR90, Thm. 7.4].

0.3.4 Universal families

Definition 0.3.24. If F : Sch → Sets is a moduli functor representable by a
∼
scheme M via an isomorphism α : F → hM of functors, then the universal family
of F is the object u ∈ F (M ) corresponding under α to the identity morphism
idM ∈ hM (M ) = Mor(M, M ).

Exercise 0.3.25 (easy but important). Let F : Sch → Sets be a functor repre-
sentable by a scheme M , and let u ∈ F (M ) be the universal family. Show that if
a ∈ F (T ) is an object over a scheme T corresponding to a map fa : T → M , then
the object a is the pullback of u under fa , i.e., a = F (fa )(u).

23
 Suspend your skepticism for a moment and suppose that there actually exists
a scheme Mg representing the moduli functor of smooth curves of genus g (Exam-
ple 0.3.8). Then corresponding to the identity map Mg → Mg is a family of genus g
curves Ug → Mg satisfying the following universal property: for every smooth family
of curves C → S over a scheme S, there is a unique map S → Mg and cartesian
diagram
C / Ug

□
 
S / Mg .

The map S → Mg sends a point s ∈ S to the curve [Cs ] ∈ Mg . While there does not
exist a scheme Mg representing the moduli functor, there is an algebraic stack Mg
parameterizing smooth curves which is equipped with a universal family; see §3.1.29.

C
Ug

[C] Mg

[D]

Figure 0.3.26: Visualization of a (non-existent) universal family over Mg .

Example 0.3.27. The universal family of the moduli functor of projective space
(Example 0.3.16) is the line bundle O(1) on Pn together with the sections x0 , . . . , xn ∈
Γ(Pn , O(1)).
Example 0.3.28 (Classifying spaces in algebraic topology). Let G be a topological
group and Toppara be the category of paracompact topological spaces where mor-
phisms are defined up to homotopy. It is a theorem in algebraic topology that the
functor
Toppara → Sets, S 7→ {principal G-bundles P → S}/(isomorphism),
is representable by a topological space, which we denote by BG and call the
classifying space. The universal family is usually denoted by EG → BG. For
example, the classifying space BC∗ is the infinite dimensional manifold CP∞ . In
algebraic geometry, however, the classifying stack BGm,C is an algebraic stack of
dimension −1.

24
0.3.5 Examples of non-representable moduli functors
If F : Sch/C → Sets is a moduli functor, then an object E ∈ F (C) with a non-trivial
automorphism can prevent the functor F from being representable. This is because
we may glue trivial families using the automorphism to construct a non-trivial family
E over a scheme S such that every fiber Es (i.e., the pullback of E along Spec C → S)
is isomorphic to E.

Proposition 0.3.29. Let F : Sch/C → Sets be a moduli functor. If there is a family

of objects E ∈ F (S) over a variety S such that
(a) the fibers Es are isomorphic for s ∈ S(C), and
(b) the family E is non-trivial, i.e., is not equal to the pullback of an object
E ∈ F (C) along the structure map S → Spec C,
then F is not representable.

Proof. Suppose by way of contradiction that F is representable by a scheme X. By

condition (a), the restriction E := Es is independent of s ∈ S(C) and defines a unique
x
point x ∈ X(C). As S is reduced, the map S → X factors as S → Spec C − → X. This
x
implies that the family E is the pullback under the constant map S → Spec C − → X,
i.e., E is a trivial family, which contradicts condition (b).

Exercise 0.3.30 (easy). Show that the moduli functor F : Sch/C → Sets assigning a
scheme S to the set of isomorphism classes of vector bundles on S is not representable.

Example 0.3.31 (Moduli of elliptic curves). An elliptic curve is a pair (E, p) where
E is a smooth, connected, and projective curve E of genus 1 and p ∈ E(C). A family
of elliptic curves over a scheme S is a pair (E → S, σ) where E → S is smooth proper
morphism with a section σ : S → E such that for every s ∈ S, the fiber (Es , σ(s)) is
an elliptic curve over the residue field κ(s). The moduli functor of elliptic curves is

FM1,1 : Sch → Sets

S 7→ {families (E → S, σ) of elliptic curves } / ∼,

where (E → S, σ) ∼ (E ′ → S, σ ′ ) if there is an S-isomorphism α : E → E ′ compatible

with the sections (i.e., σ ′ = α ◦ σ).

Exercise 0.3.32 (good practice). Consider the family of elliptic curves defined over
A1 ∖ 0 (with coordinate t) by

E := V (y 2 z − x3 + tz 3 )  / (A1 ∖ 0) × P2


A1 ∖ 0

with section σ : A1 ∖ 0 → E given by t 7→ [0, 1, 0]. Show that (E → A1 ∖ 0, σ) satisfies

(a) and (b) in Proposition 0.3.29, and conclude that FM1,1 is not representable.

Example 0.3.33 (Moduli functor of smooth curves). Let C be a curve with a

non-trivial automorphism α ∈ Aut(C), and let N be the nodal cubic curve, which
we can think of as P1 with the points 0 and ∞ glued together. We can construct
a family C → N by taking the trivial family π : C × P1 → P1 and gluing the fiber
π −1 (0) with π −1 (∞) via the automorphism α. To show that the moduli functor

25
Sch/C → Sets of smooth curves is not representable, it suffices to show that C → N
is non-trivial.

0 ∞

Figure 0.3.34: Family of curves over the nodal cubic obtaining by gluing the fibers
over 0 and ∞ of the trivial family over P1 via α. (It would be more illustrative to
draw a Möbius band as the family of curves over the nodal cubic.)

Exercise 0.3.35 (details). Show that C → N is a non-trivial family.

Exercise 0.3.36 (good practice). Show that the moduli functor of vector bundles
over a curve C is not representable.

0.3.6 Schemes are sheaves in the big Zariski topology

If F : Sch → Sets is representable by a scheme X, then F is necessarily a sheaf in
the big Zariski topology, that is, for every scheme S, the presheaf on the Zariski
topology of S, defined by assigning to an open subset U ⊆ S the set F (U ), is a
sheaf on the Zariski topology of S. This is a restatement that morphisms into the
fixed scheme X glue uniquely. The failure to be a sheaf therefore provides another
obstruction to the representability of a given moduli functor F .

Exercise 0.3.37 (good practice).

(a) Show that the following naive Grassmannian functor

F : Sch → Sets, S 7→ {quotients q : OSn ↠ OSk }/ ∼

is not representable.
(b) Under the usual scaling action of Gm on An+1 ∖ 0, show that the functor
S 7→ (An+1 ∖ 0)(S)/Gm (S) is not a sheaf.

The presence of non-trivial automorphisms often implies that a given moduli

functor is not a sheaf in the big Zariski topology.

Example 0.3.38. Consider the moduli functor FMg of smooth curves from Ex-
ample 0.3.8. Let {Si } be a Zariski open covering of a scheme S, and suppose that
∼
Ci → Si are families of smooth curves Ci → Si with isomorphisms αij : Ci |Sij → Cj |Sij
on the intersection Sij := Si ∩ Sj . The requirement that FMg be a sheaf (when re-
stricted to the Zariski topology on S) implies that the families Ci → Si glue uniquely
to a family of curves C → S. However, we have not required the isomorphisms αi to

26
be compatible on the triple intersection (i.e., αij |Sijk ◦ αjk |Sijk = αik |Sijk ), which is
necessary for gluing schemes [Har77, Exc. II.2.12]. For this reason, FMg fails to be a
sheaf.
Exercise 0.3.39. Show that the moduli functors of smooth curves and elliptic
curves are not sheaves by explicitly exhibiting a scheme S, an open cover {Si }, and
families of curves over Si that do not glue to a family over S.

0.3.7 The yoga of functors

Contravariant functors F : Sch → Sets form a category Fun(Sch, Sets) where mor-
phisms are natural transformations. This category has fiber products: given mor-
phisms α : F → G and β : G′ → G, we define

F ×G G′ : Sch → Sets
S 7→ {(a, b) ∈ F (S) × G′ (S) | αS (a) = βS (b) ∈ G(S)} .

Exercise 0.3.40 (easy). Show that F ×G G′ satisfies the universal property for
fiber products in Fun(Sch, Sets).
Definition 0.3.41.
(1) We say that a morphism F → G of contravariant functors is representable by
schemes if for every map S → G from a scheme S, the fiber product F ×G S
is representable by a scheme.
(2) We say that a morphism F → G is an open immersion or that a subfunctor
F ⊆ G is open if for every morphism S → G from a scheme S, F ×G S is
representable by an open subscheme of S.
(3) We say that a set of open subfunctors {Fi } of F is a Zariski open cover if for
every morphism S → F from a scheme S, {Fi ×F S} is a Zariski open cover of
S (and in particular each Fi is an open subfunctor of F ).
Each of these conditions can be checked on affine schemes.
These definitions give a recipe for checking that a given functor F is representable
by a scheme: find a Zariski open cover {Fi } where each Fi is representable.
Exercise 0.3.42 (good practice).
(a) Let F : Sch → Sets be a functor which is a sheaf in the big Zariski topology,
and let {Fi } be a Zariski open cover of F . Show that if each Fi is representable
by a scheme, then so is F .
(b) Show that a collection
` of open subfunctors {Fi } of F is a Zariski open cover if
and only if the map i Fi (k) → F (k) is surjective for each algebraically closed
field k.
(c) Given morphisms of schemes X → Y and Y ′ → Y , reprove the existence of the
fiber product X ×Y Y ′ in the category of schemes by exhibiting a Zariski open
cover {Fi } of X ×Y Y ′ where each Fi is representable by an affine scheme.
Exercise 0.3.43 (Functorial definition of a scheme). Show that a scheme can be
equivalently defined as a contravariant functor F : AffSch → Sets on the category
of affine schemes (or covariant functor on the category of rings) as follows. Let C
be a full subcategory of the category Fun(AffSch, Sets) of contravariant functors.
Extending Definitions 0.3.15 and 0.3.41, we define a functor F : AffSch → Sets to
be representable in C if there exists an object X ∈ C and a functorial equivalence

27
F (S) = Mor(S, X) for every S ∈ AffSch. We say that a map F → G of functors
from AffSch to Sets is representable by open immersions in C if for every morphism
Spec B → G, the fiber product F ×G Spec B is representable by an object X ∈ C
which is an open subscheme of Spec B. Finally, we say that a collection {Fi } of
subfunctors of F is a Zariski open C-cover if each Fi → F is representable by open
immersions in C and for each algebraically closed field k, the map i Fi (k) → F (k)
`
is surjective.
(a) Letting C = AffSch, show that a scheme with affine diagonal can be equivalently
defined as a functor F : AffSch → Sets such that there exists a Zariski open
C-cover {Fi } of F with each Fi representable in C.
(b) Letting C be the category of schemes with affine diagonal, show that a scheme
can be equivalently defined as a functor F : AffSch → Sets such that there
exists a Zariski open C-cover {Fi } with each Fi representable in C.
(c) Alternatively, show that a scheme can be defined as a suitable contravariant
functor on the category of quasi-affine schemes.

Replacing Zariski opens with étale morphisms in the above exercise leads to the
definition of an algebraic space (Definition 3.1.2).

Exercise 0.3.44 (hard). Let X → Y be a morphism of schemes each proper

over a scheme S. If X is flat over S, show that the subfunctor F ⊆ Mor(−, S),
∼
parameterizing maps T → S of schemes such that XT → YT is an isomorphism, is
representable by an open subfunctor.
Hint: If s ∈ S is a point such that Xs → Ys is an isomorphism, use the Fibral
Flatness Criterion (A.2.10) to show that X → Y is flat over s. Then reduce to the
case when X → Y is finite étale.

0.4 Moduli groupoids

La conclusion pratique à laquelle je suis arrivé dès maintenant, c’est
que chaque fois que en vertu de mes critères, une variété de modules
(ou plutôt, un schéma de modules) pour la classification des variations
(globales, ou infinitésimales) de certaines structures (variétés complètes
non singulières, fibrés vectoriels, etc.) ne peut exister, malgré de bonnes
hypothèses de platitude, propreté, et non singularité éventuellement, la
raison en est seulement l’existence d’automorphismes de la structure qui
empêche la technique de descente de marcher.

Alexander Grothendieck, letter to Serre, 1959 [CS01, p. 94]

We now change our perspective: rather than specifying when two objects are
identified, we specify how !
One of the most desirable properties of a moduli space is the existence of a
universal family (see §0.3.4), and the presence of automorphisms obstructs its
existence (see §0.3.5). Encoding automorphisms into our descriptions will allow us
to get around this problem. To define a moduli groupoid, we need to specify
(1) objects; and
(2) a set of equivalences (possibly empty) between any two objects.
Shortly we will combine the functorial worldview of the last section with this
groupoid perspective to define moduli stacks.

28
0.4.1 Groupoids
A convenient mathematical structure to encode objects and their identifications is a
groupoid.

Definition 0.4.1. A groupoid is a category C where every morphism is an isomor-

phism.

Two groupoids C1 and C2 are equivalent if there is an equivalence of categories

∼
F : C1 → C2 , i.e., there is a functor G : C2 → C1 such that F ◦ G and G ◦ F are
isomorphic to the identity functors, or equivalently F is fully faithful and essentially
surjective.

Example 0.4.2 (Sets are groupoids). If Σ is a set, the category CΣ , whose objects
are elements of Σ and whose morphisms consist of only the identity morphisms, is a
groupoid. We say that a groupoid C is equivalent to a set Σ if there is an equivalence
of categories C → CΣ .

Example 0.4.3 (Classifying groupoid). If G is a group, the classifying groupoid BG

of G is defined as the category with one object ⋆ such that Aut(⋆) = Mor(⋆, ⋆) = G.

Example 0.4.4. The category FB of finite sets where morphisms are bijections
is a groupoid. The isomorphism classes of FB are in bijection with N while
Aut({1, . . . , n}) = Sn is the permutation group.

Example 0.4.5 (Projective space). Projective space is identified with the moduli
groupoid of lines L ⊂ An+1 through the origin where the only morphisms are the iden-
tity maps. Alternatively, the objects are nonzero linear maps x = (x0 , . . . , xn ) : C →
Cn+1 and there is a unique morphism x → x′ if and only if im(x) = im(x′ ) ⊆ Cn+1
(i.e., there exists a λ ∈ C∗ such that x′ = λx).

0.4.2 Moduli groupoid of orbits

Example 0.4.6 (Moduli groupoid of orbits). Given an action of a group G on a set
X, we define the moduli groupoid of orbits [X/G]6 by taking the objects to be all
elements x ∈ X and by declaring Mor(x, x′ ) = {g ∈ G | x′ = gx}.

0
A1 A1

[A1 /(Z/2)] [A1 /Gm ]

Z/2 Gm {1}
Figure 0.4.7: Pictures of the scaling actions of Z/2 = {±1} and Gm on A1 over C with
the automorphism groups listed in blue. Note that [A1 /Gm ] has two isomorphism
classes of objects—0 and 1—corresponding to the two orbits—0 and A1 ∖ 0—such
that 0 ∈ {1} if the set A1 /Gm is endowed with the quotient topology.
6We use brackets to distinguish the groupoid quotient [X/G] from the set quotient X/G. Later

when G is an algebraic group and X is a scheme, [X/G] will denote the quotient stack (which
always exists) while X/G will denote the quotient space (if it exists).

29
Exercise 0.4.8 (easy). Show that the moduli groupoid of orbits [X/G] in Exam-
ple 0.4.6 is equivalent to a set if and only if the action of G on X is free.
Example 0.4.9. Consider the category C with two objects x1 and x2 such that
Mor(xi , xj ) = {±1} for i, j = 1, 2 where composition of morphisms is given by
multiplication. Then C is equivalent B(Z/2).
1 1 1
1
-1 -1 -1
-1

1 xi x
x1 x2 x
-1
Figure 0.4.10: An equivalence of groupoids.

Exercise 0.4.11 (easy). In Example 0.4.9, show that there is an equivalence

of categories inducing a bijection on objects between C and either [(Z/2)/(Z/4)]
or [(Z/2)/(Z/2 × Z/2)], where the action is given by surjections Z/4 → Z/2 or
Z/2 × Z/2 → Z/2.
Example 0.4.12 (Projective space as a quotient). The moduli groupoid of projective
space (Example 0.4.5) can also be described as the moduli groupoid of orbits
[(An+1 ∖ 0)/Gm ]. We can also consider the quotient groupoid [An+1 /Gm ], which
is equivalent to the groupoid whose objects are (possibly zero) linear maps x =
(x0 , . . . , xn ) : C → Cn+1 such that Mor(x, x′ ) = {t ∈ C∗ | x′i = txi for all i}. In this
way, Pn is a subgroupoid of [An+1 /Gm ].
Exercise 0.4.13 (easy, good practice). If a group G acts on a set X and x ∈ X is
a point with stabilizer Gx , show that there is a fully faithful functor BGx → [X/G].
If the action is transitive, show that it is an equivalence.
A morphisms of groupoids C1 → C2 is by definition a functor. The category
Mor(C1 , C2 ) has functors as objects and natural transformations as morphisms.
Exercise 0.4.14 (easy). If C1 and C2 are groupoids, show that Mor(C1 , C2 ) is a
groupoid.
Exercise 0.4.15 (moderate, good practice). If H and G are groups, show that
there is an equivalence
a
B C G (im ϕ)


Mor(BH, BG) =
ϕ∈Hom(H,G)/G

where Hom(H, G)/G denotes equivalence classes of homomorphisms H → G up to

conjugation by G, and C G (im ϕ) denotes the centralizer of im ϕ in G.
Exercise 0.4.16. Provide an example of group actions of H and G on sets X
and Y and a map [X/H] → [Y /G] of groupoids that does not arise from a group
homomorphism ϕ : H → G and a ϕ-equivariant map X → Y .

0.4.3 Examples of moduli groupoids

Example 0.4.17 (Moduli groupoid of smooth curves). The objects are smooth,
connected, and projective curves of genus g over C and for two curves C, C ′ , the set
of morphisms is defined as the set of isomorphisms
∼
Mor(C, C ′ ) = {isomorphisms α : C → C ′ }.

30
Example 0.4.18 (Moduli groupoid of vector bundles on a curve). The objects are
vector bundles E of rank r and degree d on a fixed curve C, and the morphisms are
isomorphisms of vector bundles.

Example 0.4.19 (Moduli groupoid of unlabeled triangles). Let us revisit the moduli
M unl of unlabeled triangles up to similarity from Example 0.2.12. Recall that we
have already introduced families of unlabeled triangles and shown that this functor
is not representable (Example 0.3.6).
We define the moduli groupoid of unlabeled triangles up to similarity, denoted
by Munl (note the calligraphic font), where the objects are unlabeled triangles and
the morphisms are similarities. For example, an isosceles triangle and an equilateral
triangle have automorphism groups Z/2 and S3 . We can draw essentially the same
picture as Figure 0.2.14 except we record the automorphisms.

b
1
b=
c
Z/
Figure 0.4.20: Picture of the 2
equilateral
moduli groupoid Munl with
a
+

non-trivial automorphism 2/3

b=

S3
groups labeled.
c

2
Z/
a2 + b2 = c2
b

1/2
=
a

1/2 2/3 a

There is a functor

Munl → M unl ,

from the moduli groupoid to the moduli set, which is an equivalence on isomorphism
classes of objects and collapses all morphisms to the identity. This is a first example
of a coarse moduli space.

Exercise 0.4.21. Recalling the description of the moduli set M lab of labeled
triangles up to similarity from (0.2.6), show that there is a natural action of S3
on the moduli set M lab of labeled triangles up to similarity and that there is an
identification Munl ∼
= [M lab /S3 ] of groupoids.

Exercise 0.4.22. Define a moduli groupoid of oriented triangles and investigate its
relation to the moduli groupoids of labeled/unlabeled triangles.

For a more detailed exposition of the moduli stack of triangles, see [Beh14].

31
0.5 Why the étale topology?
On peut dire qu’en passant de la topologie de Zariski à topologie étale,
“on a fait ce qu’il fallait” pour obtenir “le bon” H1 [...] pour un groupe de
coefficients constant fini G. C’est un fait remarquable, qui sera démontré
dans la suite de ce séminaire, que cela suffit également pour trouver les
“bons” Hi (X, G) pour tout groupe de coefficients de torsion (du moins si
G est premier aux caractéristiques résiduelles de X).

Alexander Grothendieck, [SGA4, VII.2.1]

Moduli stacks will be introduced in the next section by combining moduli functors
with groupoids: one needs to specify families of objects over every scheme S along
with identifications and pullbacks. For such data to define a stack, we will require
that objects and their morphisms glue in the étale topology! Apparently Grothendieck
coined the word ‘étale’ because étale morphisms reminded him of a calm sea at high
tide under a full moon, just as Victor Hugo wrote in Les travailleurs de la mer : “La
mer était étale, mais le reflux commençait à se faire sentir.”
Why is the Zariski topology not sufficient for our purposes? The short answer is
that there are not enough Zariski-open subsets and that étale morphisms serve as a
good replacement of analytic open subsets.

0.5.1 What is an étale morphism anyway?

I have sometimes been baffled when a student is intimidated by étale morphisms,
especially when she has already mastered conceptually more difficult notions of say
properness and flatness. One factor could be the fact that the definition is buried in
[Har77, Exercises III.10.3-6] and its importance is not highlighted there.

A1 x

A1 x2

Figure 0.5.1: Picture of an étale double cover of A1 ∖ 0.

The geometric picture to have in your mind is a covering space. There are several
ways in which we can formulate an étale morphism f : X → Y of schemes of finite
type over C:

– f is smooth of relative dimension 0 (i.e., f is flat and all fibers are smooth of
dimension 0);
– f is flat and unramified (i.e., for `
all y ∈ Y (C), the scheme-theoretic fiber Xy
is isomorphic to a disjoint union i Spec C of points);
– f is flat and ΩX/Y = 0;
– for all x ∈ X(C), the induced map O bX,x on completions is an
bY,f (x) → O
isomorphism; and

32
 – assuming in addition that X and Y are smooth: for all x ∈ X(C), the induced
map TX,x → TY,f (x) on tangent spaces is an isomorphism.
We say that f is étale at x ∈ X if there is an open neighborhood U of x such that f |U
is étale. See §A.3 for more background. These characterizations are all equivalent,
but by no means this should be clear to you—some of the proofs are quite involved.
Nevertheless, if you can take the equivalences on faith, it requires very little effort
to not only internalize the concept but to master its use.
Exercise 0.5.2 (good practice). Show that f : A1 → A1 , x 7→ x2 is étale over
A1 ∖ 0 but is not étale at the origin. Try to show this for as many of the above
characterizations as you can.

0.5.2 What can you see in the étale topology?

Working with the étale topology is like getting a pair of magnifying lenses for your
algebraic geometry glasses allowing you to see finer details that you already could
observe with your differential geometry glasses.
Example 0.5.3 (Reducibility of a node). Consider the plane nodal cubic C defined
by y 2 = x2 (x − 1) in the plane. While there is an analytic open neighborhood of
the node p = (0, 0) which is reducible, there is no such Zariski-open neighborhood.
However, taking a ‘square root’ of x − 1 yields a reducible étale neighborhood. More
specifically, define C ′ = Spec C[x, y, t]t /(y 2 − x3 + x2 , t2 − x + 1) and consider
C ′ → C, (x, y, t) 7→ (x, y)
Since y 2 − x3 + x2 = (y − xt)(y + xt), we see that C ′ is reducible. This is illustrated
by the following picture, which is also featured in [Har77, Exc. III.10.6].

Figure 0.5.4: After an étale cover, the nodal cubic becomes reducible.

Example 0.5.5 (Étale cohomology). Sheaf cohomology for the Zariski topology can
be extended to the étale topology leading to the extremely robust theory of étale

33
cohomology. For example, for a smooth projective curve C of genus g over C, the étale
cohomology H1 (Cét , Z/n) of the finite constant sheaf Z/n is isomorphic to (Z/n)2g
just like the ordinary cohomology groups, while the sheaf cohomology H1 (C, Z/n)
in the Zariski topology is 0. Finally, we would be remiss without mentioning the
spectacular application of étale cohomology to prove the Weil conjectures.

Example 0.5.6 (Étale fundamental group). Have you ever thought that there
is a similarity between the bijection in Galois theory between intermediate field
extensions and subgroups of the Galois group, and the bijection in algebraic topology
between covering spaces and subgroups of the fundamental group? Well, you are
in good company—Grothendieck also considered this and developed a beautiful
theory of the étale fundamental group which packages Galois groups and fundamental
groups in the same framework.

Example 0.5.7 (Quotients by free actions of finite groups). If G is a finite group

acting freely on a projective variety X, then there exists a quotient X/G as a projec-
tive variety. The essential reason for this is that every G-orbit (or in fact every finite
set of points) is contained in an affine variety U , which is the complement of some
hypersurface. Then the intersection V = g gU of the G-translates is a G-invariant
T

affine open containing Gx and V /G = Spec Γ(V, OV )G . These local quotients glue
to form a global quotient X/G. See Corollary 4.2.13 and Exercise 4.2.14.
However, if X is not projective, the quotient does not necessarily exist as a scheme.
As with most phenomena for smooth proper varieties that are not projective, a
counterexample can be constructed by using Hironaka’s examples of smooth, proper
3-folds; [Har77, App. B, Ex. 3.4.1]. There is a smooth, proper 3-fold with a free
action by G = Z/2 such that there is an orbit Gx not contained in any G-invariant
affine open. This shows that X/G cannot exist as a scheme; indeed, if it did, then
the image of x under the finite morphism X → X/G would be contained in some
affine and its inverse would be an affine open subset of X containing Gx. See [Knu71,
Ex. 1.3] or [Ols16, Ex. 5.3.2] for details.
Nevertheless, for a free action of a finite group G on a scheme X, then every
point x ∈ X has a G-equivariant étale neighborhood Ux → X where Ux is an affine
scheme, and the quotients Ux /G can be glued in the étale topology to construct X/G
as an algebraic space (Corollary 3.1.14). The upshot is that we can always take
quotients of free actions by finite groups. This is a very desirable feature given the
ubiquity of group actions in algebraic geometry, but it comes at the cost of enlarging
our category from schemes to algebraic spaces.

Example 0.5.8 (Artin Approximation). Artin Approximation (B.5.18) is a powerful

and extremely deep result, due to Michael Artin, which implies that most properties
which hold for the completion O bX,x of the local ring also in an étale neighborhood
of x. For instance, since the completion of the local ring at a nodal singularity is
reducible, Artin Approximation implies that there is a reducible étale neighborhood.

0.5.3 Working with the étale topology: descent theory

Another reason why the étale topology is so useful is that many properties of schemes
and their morphisms can be checked on étale covers. Almost every property that can
be checked on a Zariski-open cover {Ui } of scheme X can `also be checked on an étale
cover {Ui → X}; here each map Ui → X is étale and i Ui → X is surjective. In
fact, most properties can even be verified on a smooth or fppf cover. Descent theory
is developed in §2.1 and is used to prove just about everything about algebraic

34
spaces and stacks. Indeed, algebraic stacks (resp., Deligne–Mumford stacks) have
by definition a smooth (resp., étale) cover by affine schemes. Similar to how many
properties of schemes can be reduced to properties of affine schemes by taking
a Zariski open affine cover, properties of algebraic spaces and stacks can also be
established by reducing to affine schemes.

0.6 Moduli stacks

Was du ererbt von deinen Vätern hast, erwirb es, um es zu besitzen.

Johann Wolfgang von Goethe

As promised, we now synthesize moduli functors with the groupoid perspective.

To define a moduli stack, we need to specify:
(1) families of objects;
(2) how two families of objects are isomorphic; and
(3) how families pull back under morphisms.
Notice the difference from specifying a moduli functor is that rather than specifying
when two families are isomorphic, we specify how. In other words, we need to specify
an assignment
F : Sch → Groupoids, S 7→ FamS
taking a scheme S to a groupoid of families of objects over S. But what exactly do
we mean by this? Groupoids form a ‘2-category’ as they have objects (groupoids),
morphisms (functors between groupoids), and 2-morphisms (natural transformations
between functors). How can we precisely formulate such an assignment in down-
to-earth terms? Well, we certainly need pullback functors f ∗ : FamT → FamS for
f g
each morphism f : S → T . Given a composition S − →T − → U of schemes, we should
∼
also have an isomorphism of functors (i.e., a 2-morphism) µf,g : (f ∗ ◦ g ∗ ) → (g ◦ f )∗ .
f
Should the isomorphisms µf,g satisfy a compatibility condition under triples S −
→
g h
T −→U − → V ? Yes! This leads to the notion of a pseudo-functor but we will not
spell it out here; we encourage the reader to work it out, or to look it up in [Vis05,
Def. 3.10] or [SP, Tag 003N]. We take a slightly different approach using prestacks
which is technically more convenient. It is nevertheless useful to think of a prestack
as an assignment Sch → Groupoids.

0.6.1 Motivating the definition of a prestack

Instead of trying to define an assignment S 7→ FamS , we will build one massive
category X encoding all of the groupoids FamS which will live over the category
Sch of schemes. Loosely speaking, the objects of X will be a family a of objects
over a scheme S, i.e., a ∈ FamS , and a morphism a → b between a family a over S
and a family b over T will be the data of a morphism f : S → T together with an
∼
isomorphism a → f ∗ b of a and the pullback family of b.
A prestack over Sch is a category X together with a functor p : X → Sch, which
we visualize as
X a
α /b
_ _
p
  
Sch S
f
/T

35
where the lower case letters a, b are objects in X and the upper case letters S, T are
schemes. We say that a is over S and that α : a → b is over f : S → T . Moreover, we
need to require that certain natural axioms hold for p : X → Sch. Loosely speaking,
we require the existence and uniqueness of pullbacks: given a map S → T and
α
object b ∈ X over T , there should exist an arrow a −→ b over f satisfying a suitable
universal property; see Definition 2.4.1.
Given a scheme S, the fiber category X (S) is defined as the category of objects
over S whose morphisms are over the identity. If X is built from the groupoids
FamS as above, then X (S) = FamS .
Example 0.6.1 (Viewing a set-valued functor as a prestack). A moduli functor
F : Sch → Sets can be encoded as a moduli prestack as follows: we define the category
XF of pairs (S, a) where S is a scheme and a ∈ F (S). A map (S ′ , a) → (S, a) is
a map f : S ′ → S such that a′ = f ∗ a, where f ∗ is convenient shorthand for
F (f ) : F (S) → F (S ′ ). Observe that the fiber categories XF (S) are equivalent (even
equal) to the set F (S).
Example 0.6.2 (Moduli prestack of smooth curves). The moduli prestack of smooth
curves is the category Mg of families of smooth curves C → S together with the
functor p : Mg → Sch defined by (C → S) 7→ S. A morphism (C ′ → S ′ ) → (C → S)
in Mg is the data of maps α : C ′ → C and f : S ′ → S such that the diagram

C′
α /C

□
 
S′
f
/S

is cartesian. Note that in the fiber category Mg (C), an object is a smooth curve C
and the set of morphisms C → C is identified with the automorphism group Aut(C).
Example 0.6.3 (Moduli prestack of vector bundles). The moduli prestack of vector
bundles on a smooth curve C over C is the category Bun r,d (C) of pairs (E, S), where
S is a C-scheme and E is a vector bundle on CS = C ×C S such that for every
s ∈ S, the restriction Es to C ×C Spec κ(s) has rank r and degree d. The functor
p : Bun r,d (C) → Sch/C is defined by (E, S) 7→ S. A map (E ′ , S ′ ) → (E, S) consists
∼
of a map of schemes f : S ′ → S together with an isomorphism (id ×f )∗ E → E ′ .

0.6.2 Motivating the definition of a stack

A stack is to a prestack as a sheaf is to a presheaf. The concept could not be more
intuitive: we require that objects and morphisms glue uniquely.
Example 0.6.4 (Moduli stack of sheaves). Define the category X over Sch of
pairs (E, S) where E is a sheaf of abelian groups on a scheme S, and the functor
p : X → Sch is given by (E, S) 7→ S. A map (E ′ , S ′ ) → (E, S) in X is a map of
∼
schemes f : S ′ → S together with an isomorphism f −1 E → E ′ (or alternatively a
map E → f∗ E whose adjoint is an isomorphism).
′

You already know that morphisms of sheaves glue: let E and F be sheaves on
schemes S and T and let f : S → T be a map. If {Si } is a Zariski open cover of S,
then a map f : S → T is the same data as a collection of morphisms fi : Si → T such
∼
that fi |Sij = fj |Sij , where Sij = Si ∩ Sj . Similarly, an isomorphism f −1 E → E ′ is
the same data as isomorphisms (f E)|Si → E |Si that agree on the intersections
−1 ′

Sij [Har77, Exc. II.1.15]. Putting these together, a morphism α : (E, S) → (F, T )

36
is equivalent to morphisms αi : (E|Si , Si ) → (F, T ) such that αi |Sij = αj |Sij . You
also know how sheaves glue—it is more complicated than gluing morphisms as
sheaves have automorphisms, and given two sheaves, we prefer to say that they are
isomorphic rather than equal. If {Si } is a Zariski open cover of a scheme S, then
giving a sheaf E on S is equivalent to giving a sheaf Ei on Si and isomorphisms
ϕij : Ei |Sij → Ej |Sij satisfying the cocycle condition ϕik = ϕjk ◦ ϕij on the triple
intersection Sijk = Si ∩ Sj ∩ Sk [Har77, Exc. II.1.22].

In a similar way, we could have considered the stack of O-modules, quasi-coherent

sheaves, or vector bundles. Or for a scheme X, we could have stacks of sheaves,
O-modules, quasi-coherent sheaves, or vector bundles over X, where an object over
a scheme S is a sheaf on X × S.

The definition of a stack (Definition 2.5.1) simply axiomatizes these two natural
gluing concepts.

0.6.3 Motivating the definition of an algebraic stack

For a stack to be a geometric object, we need to specify that it is locally like a

scheme in a suitable sense. Without imposing such a condition would be like trying
to study the geometry of an arbitrary ringed space (X, OX ) or a non-representable
sheaf F : Sch → Sets. If we wish to utilize our algebraic geometry toolkit (e.g.,
coherent sheaves, commutative algebra, cohomology, ...) to study stacks in a similar
way that we study schemes, we must impose an algebraicity condition.

The conditions we impose are quite natural. In increasing generality, we define:

(1) A functor X : Sch → Sets is an algebraic space if it is a sheaf (i.e., objects glue
uniquely in the étale topology), and there is an étale cover {Ui → X} where
each Ui is an affine scheme.
(2) A stack X → Sch is Deligne–Mumford if there is an étale cover {Ui → X }
where each Ui is an affine scheme.
(3) A stack X → Sch is algebraic if there is a smooth cover {Ui → X } where each
Ui is an affine scheme.

Of course, we need to make precise the notions of étale and smooth covers. For
a first approximation, when we say that {Ui → X} is an étale cover, we require that
for every map T → X of functors where T is representable by a scheme, the fiber
product of`functors is representable by a scheme Ti , and moreover that Ti → T is
étale and Ti → T is surjective. Note that in (1), if we replace ‘étale’ with ‘Zariski’,
we would recover the notion of a scheme; see Exercise 0.3.42. It will take some time
to develop the foundations to make this completely rigorous; precise definitions are
postponed until §3.1.

37
 Algebro-geometric space Type of object Obtained by gluing

Schemes ringed space/ affine schemes in the

sheaf Zariski topology
Algebraic spaces sheaf affine schemes in the
étale topology
Deligne–Mumford stacks stack affine schemes in the
étale topology
Algebraic stacks stack affine schemes in the
smooth topology

Table 0.6.5: Schemes, algebraic spaces, Deligne–Mumford stacks, and algebraic

stacks are obtained by gluing affine schemes.

Why smooth covers? After all the fuss motivating the étale topology above,
you might be surprised to see that an algebraic stack is smooth locally a scheme.
For Deligne–Mumford stacks—which turn out to be precisely algebraic stacks with
finite automorphism groups—étale covers are sufficient. But for algebraic stacks like
Bun r,d (C) with infinite automorphism groups, we need smooth covers. For instance,
we would like to be able to form the quotient [Spec C/Gm ] (which we will call the
classifying stack BGm ) of the trivial action of Gm (or C∗ ) on a point, and this will
have no étale cover by a scheme.

0.6.4 Examples of moduli stacks

Given a stack encoding a moduli problem, constructing a smooth cover is a geometric
problem inherent to the moduli problem. It can often be solved by rigidifying the
moduli problem by parameterizing additional information. This concept is best
absorbed in examples.

Example 0.6.6 (Moduli stack of elliptic curves). An elliptic curve (E, p) is embedded
into P2 via OE (3p) such that E is defined by a Weierstrass equation

y 2 z = x(x − z)(x − λz)

for some λ =
̸ 0, 1 [Har77, Prop. IV.4.6]. Setting U = A1 ∖ {0, 1} with coordinate λ,
the family E ⊆ U × P2 of elliptic curves defined by this Weierstrass equation defines
a map U → M1,1 which is an étale cover.

Example 0.6.7 (Moduli stack of smooth curves). For a smooth curve C of genus g ≥
2, the line bundle Ω⊗3 ⊗3 ∼
C is very ample and defines an embedding C ,→ P(Γ(C, ΩC )) =
P 5g−6
. There is a Hilbert scheme Hilb (P
P 5g−6
) (see Theorem 1.1.2) parameterizing
closed subschemes of P5g−6 with the same Hilbert polynomial P (z) = (6z − 1)(g −
1) as C ⊆ P5g−6 , and there is a locally closed subscheme H ′ ⊆ HilbP (P5g−6 )
parameterizing smooth subschemes such that Ω⊗3 ∼
C = OC (1). The universal subscheme
over H ′ defines a map H ′ → Mg which is a smooth cover (see Theorem 3.1.17 for
details) and thus Mg is an algebraic stack. We will show that it is Deligne–Mumford
in Corollary 3.6.10.

38
Example 0.6.8 (Moduli stack of vector bundles). For every vector bundle E of
rank r and degree d on a smooth curve C with an ample line bundle OC (1), for m
sufficiently large, the twist E(m) is globally generated and P (m) = h0 (C, E(m))
defines the Hilbert polynomial of E. Therefore, for m ≫ 0, we can view E as a
quotient OC (−m)P (m) ↠ E. There is a Quot scheme QuotP (OC (−m)P (m) ) (see
Theorem 1.1.3) parameterizing quotients with Hilbert polynomial P . There is a
locally closed subscheme Q′m ⊆ QuotP (OC (−m)P (m) ) parameterizing vector bundle
quotients π : OC (−m)P (m) ↠ E such that the induced map Γ(π⊗OC (m)) : CP (m) →
Γ(C, E(m)) is an isomorphism. The universal quotient over Q′m defines a map
Q′m → Bun r,d (C) which is smooth and the collection {Q′m → Bun r,d (C)} for m ≫ 0
defines a smooth cover. This shows that an Bun r,d (C) is an algebraic stack; see
Theorem 3.1.21 for details. It is not a Deligne–Mumford stack.

0.6.5 Quotient stacks

One of the most important examples of a stack is a quotient stack [X/G] arising
from an action of an algebraic group G on a scheme X. The geometry of [X/G]
could not be simpler: it is the G-equivariant geometry of X (see Table 0.6.15).
Similar to how toric varieties provide concrete examples of schemes, quotient
stacks provide concrete examples of algebraic stacks that are useful to gain geometric
intuition of general algebraic stacks, and at the same time provide a fertile testing
ground for conjectural results. On the other hand, it turns out that many algebraic
stacks are quotient stacks or are at least locally quotient stacks, and most properties
that hold for quotient stacks can also be established for many algebraic stacks.

Quotient prestacks. Given an action of an algebraic group G on a scheme X, the

quotient prestack [X/G]pre is the prestack whose fiber category [X/G]pre (S) over a
scheme S is the quotient groupoid (or the moduli groupoid of orbits) [X(S)/G(S)]
as in Example 0.4.6. This will not satisfy the gluing axioms of a stack; even when
the action is free, the quotient functor Sch → Sets defined by S 7→ X(S)/G(S) is
not a sheaf (see Exercise 0.3.37). How can we make it into a stack? Well, instead of
thinking of an object of [X/G]pre over a scheme S as a morphism f : S → X, let us
think of it as a trivial G-bundle together with a map to X:

fe
/ X,  / g · f (s)
G×S (g, s)
p2

S.

Exercise 0.6.9 (easy). Given two maps f1 , f2 : S → X, show that an element

of α ∈ G(S) satisfying f2 = α · f1 is the same data as an isomorphism of trivial
G-bundles G × S → G × S compatible with the maps fe1 and fe2 to X.

From this perspective, it is even more clear that [X/G]pre is not a stack even when
X is a point: given a Zariski cover {Si } of a scheme S, trivial G-bundles G × Si → Si
together with isomorphisms over Si ∩ Sj satisfying a cocycle condition will glue to a
principal G-bundle P → S (Definition B.1.47), but it will not necessarily be trivial.
This suggests that we should define an object of a quotient stack to be a principal
G-bundle together with a G-equivariant map to X.

Quotient stacks. We define the quotient stack [X/G] as the category over Sch/C

39
whose objects over a C-scheme S are diagrams

P
f
/X


S

where P → S is a principal G-bundle and f : P → X is a G-equivariant morphism.

A morphism is the data of a commutative diagram

f′

#
P′
φ
/P f
/X

□
 
S′
g
/S

where the left square is cartesian. There is an object of [X/G] over X given by the
diagram
G×X
σ /X
p2

X,
where σ denotes the action map. By the 2-Yoneda Lemma (2.4.21), this defines a
map X → [X/G]. Even if the action of G on X is not free, the map X → [X/G] is
a principal G-bundle. Let us pause to appreciate that:

The map X → [X/G] is a principal G-bundle even if the action of

G on X is not free.

This is one of the great advantages of working with stacks. At the expense of
enlarging our category from schemes to algebraic stacks, we are able to tautologically
construct the quotient [X/G] as a ‘geometric space’ with desirable properties.
Example 0.6.10 (Classifying stack). We define the classifying stack of an algebraic
group G as the category BG := [Spec C/G] of principal G-bundles P → S. The
projection Spec C → BG is not only a principal G-bundle; it is the universal
principal G-bundle. Given any other principal G-bundle P → S, there is a unique
map S → BG and a cartesian diagram

P / Spec C

□
 
S / BG.

Example 0.6.11 (Quotients by finite groups). Quotients by free actions of finite

groups exist as algebraic spaces! See Corollary 3.1.14.
Exercise 0.6.12. What is the universal family over the quotient stack [X/G]?
Moduli stacks can often be described as quotient stacks, and these descriptions
can be leveraged to establish properties of the moduli stack.

40
Example 0.6.13 (Moduli stack of smooth curves as a quotient). Reexamining
Example 0.6.7, we see that the embedding of a smooth curve C via |Ω⊗3
C | : C ,→ P
5g−6
⊗3 ∼ 5g−5
depends on a choice of basis Γ(C, ΩC ) = C and therefore is only unique up to a
projective automorphism, i.e., an element of PGL5g−5 = Aut(P5g−6 ). The algebraic
group PGL5g−5 acts on the subscheme H ′ ⊂ HilbP (P5g−6 ) parameterizing smooth
tricanonically embedded curves, and there is an isomorphism Mg ∼ = [H ′ / PGL5g−6 ].
Example 0.6.14 (Moduli stack of vector bundles as a quotient). For the moduli
stack of vector bundles (Example 0.6.8), the presentation of a vector bundle E as a
quotient OC (−m)P (m) ↠ E depends on a choice of basis Γ(C, E(m)) ∼ = CP (m) . The
algebraic group GLP (m) acts on the scheme Qm and there is an identification
′

Bun r,d (C) ∼

[
= [Q′m / GLP (m) ];
m≫0

see Theorem 3.1.21.

Geometry of a quotient stack. While the definition of the quotient stack [X/G]
may appear abstract, its geometry is very familiar. The table below provides a
dictionary between the geometry of a quotient stack [X/G] and the G-equivariant
geometry of X. The stack-theoretic concepts on the left-hand side will be introduced
later.

Geometry of [X/G] G-equivariant geometry of X

C-point x ∈ [X/G] orbit Gx of C-point x ∈ X (with x the image
of x under X → [X/G])
automorphism group Aut(x) stabilizer Gx
function f ∈ Γ([X/G], O[X/G] ) G-equivariant function f ∈ Γ(X, OX )G
map [X/G] → Y to a scheme Y G-equivariant map X → Y
line bundle G-equivariant line bundle (or G-linearization)
quasi-coherent sheaf G-equivariant quasi-coherent sheaf
tangent space T[X/G],x normal space TX,x /TGx,x to the orbit
coarse moduli space [X/G] → Y geometric quotient X → Y
good moduli space [X/G] → Y good GIT quotient X → Y

Table 0.6.15: Dictionary between the geometry of [X/G] and the G-equivariant
geometry of X.

0.7 Constructing projective moduli spaces

The spaces Mg,ν are in my eyes (together with the group SL(2)) the
most beautiful and most fascinating objects that I have encountered in
mathematics.
Grothendieck [Gro86, p. 129]

41
 One of our incentives for introducing algebraic stacks is to ensure that a given
moduli problem M is representable and equipped with a universal family. While
many geometric questions can be studied (and arguably should be studied) on the
moduli stack M itself, it is often very convenient to make a trade-off: by sacrificing
the existence of a universal family, we can sometimes construct a more familiar
geometric space, ideally a projective variety. This allows us to utilize the much
larger toolkit of projective geometry (e.g., birational geometry, intersection theory,
Hodge theory, ...) to study the moduli problem.
We highlight two approaches to construct projective moduli spaces:
(1) Geometric Invariant Theory (GIT), and
(2) Intrinsic construction of coarse/good moduli spaces.
There is a beautiful interplay between the intrinsic and extrinsic approaches.
Ideas from GIT have inspired techniques in each of the six steps of the intrinsic
approach below, and conversely the intrinsic approach sheds light back on GIT.
GIT is also deeply intertwined with 19th century invariant theory, and determining
the GIT semistable locus is an interesting and important problem on its own. It is
valuable to keep both approaches in mind.

0.7.1 GIT approach

Outline of the GIT strategy

(A) Express the moduli stack M as a substack

M ⊆ [X/G],

where G is reductive and X ,→ P(V ) is a G-equivariant

embedding into the projectivization of a G-representation V .
(B) Show that a point x ∈ X is GIT semistable if and only if
x ∈ M, or in other words that M = [X ss /G].
For Step A, there are often natural ways to rigidify the moduli problem by
parameterizing additional data. For a smooth curve C, a choice of basis of Γ(C, Ω⊗3
C )
defines a tricanonical embedding C ,→ P5g−6 (Example 0.6.13), or for any k ≥ 3,
a choice of basis of Γ(C, Ω⊗kC ) defines the kth pluricanonical embedding C ,→
P(2k−1)(g−1)−1 . For a vector bundle E on a fixed smooth curve C, after choosing a
sufficiently large integer m, a choice of basis of Γ(C, E(m)) ∼
= CP (m) allows us to
view E as a quotient OC (−m) (m) ↠ E. The rigidified moduli problem should
P

have a compactification which is representable by a projective variety X—which

is HilbP (P5g−6 ) and QuotP (OC (−m)P (m)) in our two examples—and the choice
of additional data should be governed by an action of a group G. For the GIT
approach to succeed, we need that G is reductive and that M is a substack of [X/G].
Finally, we need to choose a G-equivariant embedding X ,→ P(V ) where V is a finite
dimensional G-representation, or equivalently choose a G-linearization of an ample
line bundle on X.
Step B is the hardest: we must show that M is precisely the open substack of
[X/G] of GIT semistable points. Using the Hilbert–Mumford Criterion (7.4.4), we
can translate the problem to the following: a point x ∈ X represents an object of the
moduli problem M if and only if the Hilbert–Mumford index µ(x, λ) ≥ 0 for every
one-parameter subgroup λ : Gm → G. This often reduces the goal to a tractable
(but often still daunting) combinatorial problem.

42
 The GIT quotient M := X ss //G is necessarily projective. One beautiful feature
of GIT is that even if the moduli stack M is not compact, the GIT strategy
provides a compactification! If M has only finite automorphisms or equivalently
there are no strictly semistable points, then X ss → M is a geometric quotient
and M = [X ss /G] → M is a coarse moduli space. In the presence of infinite
automorphisms, X ss → M is a good quotient and M → M is a good moduli space.
The GIT approach is covered in detail in Chapter 7. We sketch the GIT
construction of M g in §5.8 and present a complete GIT construction of Bun r,d (C)
in §??.

0.7.2 Intrinsic approach

Six steps toward projective moduli

① Algebraicity: Express the moduli stack M as a substack

M⊆X

of a larger moduli stack X . Define an object x ∈ X to be

semistable if it is in M; this allows us to think of M as
the semistable locus X ss . Show that X is an algebraic stack
locally of finite type over C.
② Openness of semistability: Show that semistability is an
open condition, i.e., M = X ss ⊆ X is an open substack.
③ Boundedness of semistability: Show that semistability
is bounded, i.e., M = X ss is of finite type over C.
④ Semistable reduction: Show that M satisfies the existence
part of the valuative criterion for properness.
⑤ Existence of a moduli space: Show that there is a
fine/coarse/good moduli space M → M where M is a proper
algebraic space.7
⑥ Projectivity: Show that a tautological line bundle on M
descends to an ample line bundle on M , i.e., M is projective.

GIT magically solves all these steps at once! In Step A of the GIT approach,
expressing the moduli stack M as a substack [X/G] already implies ‘boundedness.’
Since GIT semistability is always an open condition, the identification in Step B
of M with the semistable locus [X ss /G] gives ‘openness of semistability’ and thus
‘algebraicity’ of M. Strikingly, GIT also implies each of the other steps: ‘semistable
reduction,’ ‘existence of a moduli space’, and ‘projectivity.’

Step 1 (Algebraicity). Many moduli stacks have natural enlargements. The stack
Mg of smooth curves and the stack Mg of stable curves are both contained in the
stack of all curves or the stack of polarized curves. The stack of semistable vector
bundles on a smooth curve is contained in the stack of all vector bundles or the
even larger stack of all coherent sheaves. It is usually easier to first show that the
enlargement X is an algebraic stack locally of finite type, and then use Steps 2 to
conclude that M itself is algebraic.
7 The calligraphic font M denotes the stack while the Roman font M denotes the space. This

convention will be followed throughout the text.

43
 To get started, we need to define the stacks M and its enlargement X —this
entails specifying families of objects along with pullbacks and identifications. To
check that X is algebraic requires finding a smooth cover U → X by a scheme. In
many cases, we can even show that X is identified with a quotient stack [U/G] in
which case U → [U/G] provides a presentation. Alternatively, it is often possible
to use Artin’s Criteria (Theorem C.7.4) to establish algebraicity; this essentially
amounts to verifying local properties of the moduli problem and in particular requires
an understanding of the deformation and obstruction theory.

Step 2 (Openness of semistability). This translates to the following condition:

for every family E of objects of X over a scheme S, the subset

{s ∈ S | Es is semistable, i.e., Es ∈ M}, (0.7.1)

where Es is the pullback of E along Spec κ(s) → S, is an open subset of S. This is

precisely what it means for the inclusion M = X ss ,→ X to be representable by open
immersions: for every map S → X (corresponding to the family E), the fiber product
M ×X S (which is identified set-theoretically with (0.7.1)) is an open subscheme of
S. This step ensures that M is also an algebraic stack locally of finite type over C.

Step 3 (Boundedness of semistability). We say that an algebraic stack M

over C is bounded if it is of finite type. Since Step 2 implies that M is locally of
finite type over C, boundedness translates into the quasi-compactness of M. More
concretely, boundedness is equivalent to the existence of a scheme Z of finite type
over C and a family of objects E over Z such that every object E of M is isomorphic
to Ez for some (not necessarily unique) z ∈ Z.
For example, Mg is bounded but the stack of all proper curves of genus g and
the stack g Mg of all smooth curves (of any genus) are not bounded. For vector
`
bundles, we will show that the stack Bun ssr,d (C) of semistable vector bundles of fixed
rank and degree is bounded. reference The stack of all vector bundles Bun r,d (C) of
fixed rank and degree is not bounded, nor is the stack of semistable vector bundles
of arbitrary rank and degree.

Step 4 (Semistable reduction). The existence part of the valuative criterion for
properness is the assertion that for every DVR R (which you can think of as a local
model of a smooth curve) with fraction field K (or punctured curve), every object
E × over K extends to a family of objects E over R after possibly replacing R with
an extension of DVRs. In other words, every diagram

E× /M
Spec K ;
(0.7.2)
E

Spec R,

has an extension after replacing R with an extension. If the extension E over R is

also unique, then we say that M satisfies the valuative criterion for properness, and
this implies properness (Theorem 3.8.7) and in particular separatedness. Arguably
the usefulness of valuative criteria in algebraic geometry is best witnessed in moduli
theory.
The moduli stack of smooth curves is not compact and does not satisfy the
existence part of the valuative criterion.

44
 0 1 λ
Figure 0.7.3: The family of elliptic curves y 2 z = x(x − z)(x − λz) degenerates to
the nodal cubic over λ = 0, 1.

Projective varieties are of course compact and satisfy the valuative criterion.
If there’s any hope to construct a projective moduli space, then the moduli stack
better satisfy the existence part of the valuative criterion. Properness of Mg was
first proven by Deligne and Mumford in their influential paper [DM69]. We prove
semistable reduction in characteristic 0 in §5.5.
For the moduli of vector bundles, semistable reduction was first proved by
Mumford and Seshadri as a consequence of the GIT construction [Ses67]. An
intrinsic geometric argument was later given by Langton [Lan75]. Note that unlike
stable curves, the stack Bun ss r,d (C) is not separated as there may exist several non-
isomorphic extensions of a vector bundle on CK to CR . Nevertheless, the moduli
stack Bun ss
r,d (C) satisfies a weaker notion of separatedness called S-completeness.

Step 5 (Existence of a moduli space). We would like to construct an algebraic

space that is the best possible approximation of the moduli stack. This step depends
on the automorphisms of the moduli problem:

• No automorphisms: in this case, the moduli stack M is already an algebraic

space M . In other words, M is a fine moduli space: for a scheme S, there is a
natural bijection between objects over a scheme S and maps S → M .
• Finite automorphisms: we must show that M is separated or in other words
that M satisfies the uniqueness part (in addition to the existence part) of the
valuative criterion. The Keel–Mori theorem (Theorem 4.4.6) then establishes
the existence of a coarse moduli space M → M where M is a proper algebraic
space. The map M → M induces a bijection of C-points and satisfies the
universal property that any other map M → Y to an algebraic space Y factors
uniquely through M .
• Reductive automorphisms: in addition to show that M satisfies the existence
part of the valuative criterion for properness, we must show that M satisfies
two additional valuative criterion called Θ-completeness and S-completeness,
which requires the existence of extensions of Gm -equivariant families of objects
over certain punctured surfaces with a Gm -action; see §6.9.2. Given these
properties, the Existence Theorem for Good Moduli Spaces (6.10.1) yields a
good moduli space M → M where M is a proper algebraic space. The map
M → M is no longer a bijection of C-points as it identifies points whose

45
 closures intersect in an analogous way to the orbit closure equivalence relation
in GIT. But M → M does induce a bijection between closed C-points of
M (sometimes called polystable objects) and the C-points of M , and it also
satisfies the universal property for maps to algebraic spaces.

Step 6 (Projectivity). This is usually the hardest step. It requires a solid

understanding of the geometry of the moduli problem and sometimes relies on
sophisticated techniques in birational geometry. Kollár introduced a strategy in
[Kol90] to verify projectivity for moduli stacks of varieties and applied it to verify
the projectivity of M g . We cover Kollár’s method in §5.9. Faltings constructed
projective moduli spaces of vector bundles in [Fal93] without using the theory of
GIT, and we borrow several of his ideas in our construction in §??.

46
Chapter 1

Hilbert and Quot schemes

Wir müssen wissen. Wir werden wissen.

David Hilbert

We prove that the Grassmannian, Hilbert and Quot functors are representable by
projective schemes. These results serve as the backbone of many results in moduli
theory and more widely algebraic geometry. In particular, they are essential for
establishing properties about the moduli stacks Mg of stable curves and Bun ss r,d (C)
of vector bundles over a curve C. While the reader could safely treat these results as
black boxes (and we encourage some readers to do this), it is also worthwhile to dive
into the details. We follow Mumford’s simplification [Mum66a] of Grothendieck’s
original construction of Hilbert or Quot schemes [FGAIV].

1.1 The Grassmannian, Hilbert, and Quot functors

1.1.1 Statements of the main theorems
Theorem 1.1.1 (Projectivity of the Grassmannian). Let V be a vector bundle of
rank n on a noetherian scheme S. For an integer 0 < q < n, the functor

Gr(q, V ) : Sch → Sets

(T → S) 7→ {vector bundle quotients VT ↠ Q on T of rank q}

is representable by a projective scheme over S.

⊕n
When S = Spec Z and V = OSpec Z , the functor Gr(q, V ) is identified with the
Grassmannian functor Gr(q, n) introduced in Example 0.3.17, and Gr(q, OS⊕n ) ∼ =
Gr(q, n) ×Z S.
Theorem 1.1.2 (Projectivity of the Hilbert Scheme). For every noetherian scheme
S and every polynomial P ∈ Q[z], the functor

HilbP (PnS /S) : Sch/S → Sets

closed subschemes Z ⊆ PnT flat and finitely presented over T
 
(T → S) 7→
such that Zt ⊆ Pnκ(t) has Hilbert polynomial P for all t ∈ T

is representable by a projective scheme over S.

47
 Observe that HilbP (PnS /S) = HilbP (PnZ /Z) ×Z S.
Theorem 1.1.3 (Projectivity of the Quot Scheme). Let S be a noetherian scheme
and F be a coherent sheaf on PnS that can be written as a quotient of OPnS (−l)⊕r for
some l and r. For every polynomial P ∈ Q[z], the functor

QuotP (F/PnS /S) : Sch/S → Sets

 
 quasi-coherent and finitely presented quotients 
(T → S) 7→ FT ↠ Q on PnT such that Q is flat over T and
 Q := Q| n has Hilbert polynomial P for all t ∈ T 
t Pκ(t)

is representable by a scheme projective over S.

In particular, HilbP (PnS /S) and QuotP (F/PnS /S) are quasi-compact!
Caution 1.1.4. When the base scheme S is clear, we will often shorten our notation to
HilbP (PnS ) and QuotP (F/PnS ). We will also abuse notation by referring to Gr(q, V ),
HilbP (PnS ), and QuotP (F/PnS ) as both the functor and the scheme representing it.
Remark 1.1.5.
(1) The Grassmannian and the Hilbert scheme are special cases of the Quot scheme:
Gr(q, V ) ∼= QuotP (V /P0S ), where P (z) = q is the constant polynomial, and
P
Hilb (PS ) = QuotP (OPnS /PnS ).
n

(2) In the definition of the Grassmannian and Quot functor above, two quotients
p : FT ↠ Q and p′ : FT ↠ Q′ are identified if ker(p) = ker(p′ ) as subsheaves
∼
of FT , or equivalently if there exists an isomorphism α : Q → Q′ such that
p = p′ ◦ α. In the Hilbert functor, two subschemes of XT are identified if they
are equal as subschemes (or equivalently if their ideal sheaves are equal as
subsheaves of OXT ).
(3) When T is noetherian, the conditions that Z be finitely presented over T and
Q be of finite presentation in the definitions of HilbP (PnS ) and QuotP (F/PnS )
are superfluous.
(4) If we do not fix the Hilbert polynomial P , the definitions above extend to
functors Hilb(PnS ) and Quot(F/PnS ), which are representable by schemes locally
of finite type. There are decompositions
a a
Hilb(PnS ) = HilbP (PnS ) and Quot(F/PnS ) = QuotP (F/PnS );
P P

this follows from the Semicontinuity Theorem (A.6.4) as the Hilbert polynomial
of a sheaf Q on PnT flat over T is locally constant. Recall also that over a
reduced base scheme T , the flatness of a quotient sheaf Q over T is equivalent
to the local constancy of the Hilbert polynomial (Proposition A.2.4).
(5) If S is an affine scheme, every coherent sheaf F on PnS can be written as a
quotient of of OPnS (−l)⊕r for some l and r.
(6) The representability and projectivity of the Hilbert and Quot scheme hold
more generally. For instance, the main theorems can also be formulated and
proved in the same way for the functors HilbP (X) and QuotP (F/X), where
X ⊆ PnS is a closed subscheme. They also hold more generally if X is a closed
subscheme of P(E) for a vector bundle E on S, i.e. if X → S is strongly
projective. Generalizations are discussed in §1.4.4, while variants such as the
Chow scheme are discussed in §1.4.5.

48
Roadmap of this chapter. In §1.2, we reduce the projectivity of Gr(q, V ) to the
⊕n
special case of Gr(q, n) = Gr(q, OSpec Z ), which we show is representable by a pro-
Vq ⊕n
jective scheme by using the functorial Plücker embedding Gr(q, n) → P( OSpec Z ):
⊕n
over
Vq ⊕n a scheme
Vq S, a quotient OS → Q is mapped to the line bundle quotient
OS → Q. In §1.3, we introduce Castelnuovo–Mumford regularity and es-
tablish Mumford’s result on Boundedness of Regularity (1.3.8). After reducing to
the case of F = OPn (−l)⊕r , Cohomology and Base Change implies that for d ≫ 0,
the pushforward π∗ F (d) under π : PnS → S is a vector bundle whose construction
commutes with base change, and Boundedness of Regularity further implies that
the morphism of functors

QuotP (F/PnS ) → Gr(P (d), π∗ F (d))

(1.1.6)
[Ft ↠ Q] 7→ [πT,∗ (FT (d)) → πT,∗ (Q(d))] ,

defined on an S-scheme T , is well-defined. Note that for HilbP (PnS ) with F =

OPnS , we have that π∗ (OPnS (d)) ∼
= Symd OS⊕n+1 so that (1.1.6) takes the form of
d ⊕n+1
P n
Hilb (PS ) → Gr(P (d), Sym OS ). We then prove that (1.1.6) is representable by
locally closed immersions (Proposition 1.4.1). Since Gr(q, π∗ F (d)) is representable
by a projective scheme over S, this already establishes the representability and
quasi-projectivity of QuotP (F/PnS ). Finally, by checking the valuative criterion, we
establish that QuotP (F/PnS ) is proper over S (Proposition 1.4.2) which in turns
implies the projectivity over S. This chapter follows the excellent expositions of
[Mum66a, §14-15], [AK80, §2], [Kol96, §1], [Laz04a, §1.8], and [Nit05].

Historical comments. Grothendieck established the projectivity of the Hilbert and

Quot scheme in [FGAIV, Thm. 3.2]. Our exposition largely follows Grothendieck’s
original strategy while incorporating Mumford’s simplification to establish the
boundedness (or equivalently the finite typeness) of the Hilbert and Quot schemes.
Boundedness is one of the hardest parts of the proof, and almost every boundedness
argument for a moduli space in algebraic geometry ultimately relies on the bounded-
ness of Hilb or Quot. Grothendieck’s approach for boundedness relied on Chow’s
boundedness result for the Chow scheme parameterizing reduced, pure-dimensional
subschemes of fixed degree. In [Mum66a], Mumford introduced the regularity of a
coherent sheaf—now called Castelnuovo–Mumford regularity—and proved that for
sufficiently large integers m every subsheaf F ⊂ OP⊕r
n with fixed Hilbert polynomial

is m-regular (Theorem 1.3.8). Mumford used this result to construct the Hilbert
scheme of curves on a surface but his argument applies equally to construct the
Quot scheme.

1.2 Projectivity of the Grassmannian

The art of doing mathematics consists in finding that special case which
contains all the germs of generality.

David Hilbert

The Grassmannian provides a warmup to the functorial approach of constructing

projective moduli spaces, and also plays an essential role in the representability and
projectivity arguments of Hilb and Quot.

49
1.2.1 Representability by a scheme
⊕n
We show that Gr(q, n) = Gr(q, OSpec Z ) is representable by a scheme (Proposi-
tion 1.2.5) by exhibiting a Zariski open cover of Gr(q, n) by representable subfunctors;
see Definition 0.3.41. Given a subset I ⊆ {1, . . . , n} of size q, define the subfunctor
GrI ⊆ Gr(q, n) as:
GrI : Sch → Sets
( )
quotients [p : OT⊕n ↠ Q] ∈ Gr(q, n)(T ) such that the (1.2.1)
T 7→ ⊕I eI ⊕n p ,
composition OT −→ OT ↠ Q is an isomorphism

where eI denotes the canonical inclusion.

Lemma 1.2.2. For each subset I ⊆ {1, . . . , n} of size q, the functor GrI is repre-
q×(n−q)
sentable by affine space AZ .
Proof. We may assume that I = {1, . . . , q}. We define a map
of functors ϕ : Aq×(n−q) →
GrI , where over a scheme T , a q × (n − q) matrix f = fi,j 1≤i≤q , 1≤j≤n−q of global
functions on T is mapped to the quotient
 
1 f1,1 · · · f1,n−q
 1 f2,1 · · · f2,n−q  ⊕q
.. ..  : OT⊕n → OT . (1.2.3)
 
. .

 
1 fq,1 ··· fq,n−q

The injectivity of ϕ(T ) : Aq×(n−q) (T ) → GrI (T ) follows from the fact that any two
quotients written in the form of (1.2.3) which are equivalent in GrI are necessarily
defined by the same equations. To see surjectivity, let [p : OT⊕n ↠ Q] ∈ GrI (T ) where
e p
by definition OT⊕I −→
I
OT⊕n ↠ Q is an isomorphism. The tautological commutative
diagram
O⊕n
p
/Q
T

(p◦eI )−1
(p◦eI )−1 ◦p " 
OT⊕I

shows that [p : OT⊕n ↠ Q] = [(p ◦ eI )−1 ◦ p : OT⊕n ↠ OT⊕I ] ∈ Gr(q, n)(T ). Since the
e (p◦eI )−1
composition OT⊕I −→I
OT⊕n ↠ OT⊕I is the identity, the q×n matrix corresponding
to (p ◦ eI ) ◦ p is necessarily of the form of (1.2.3) for functions fi,j ∈ Γ(T, OT ).
−1

Therefore ϕ(T )({fi,j }) = [p : OT⊕n ↠ Q] ∈ Gr(q, n)(T ).

Lemma 1.2.4. The set of subfunctors {GrI }, where I ranges over all subsets of
size q, is a Zariski open cover of Gr(q, n).
Proof. For a fixed subset I, we first show that GrI ⊆ Gr(q, n) is an open subfunctor.
To this end, we consider a scheme T and a morphism T → Gr(q, n) corresponding to a
quotient p : OT⊕n → Q. Let C denote the cokernel of the composition p◦eI : OT⊕I → Q.
Notice that if C = 0, then p ◦ eI is an isomorphism. The fiber product
TI /T
⊕n p
□ [OT ↠Q]
 
GrI / Gr(q, n)

50
of functors is representable by the open subscheme U = T ∖ Supp(C) (the reader is
encouraged to verify this claim). Note that if T is not noetherian, then Supp(C) ⊆ T
is still closed as C is a finite type quasi-coherent sheaf.
To check the surjectivity of ⨿I TI → T , let t ∈ T be a point. Since p ⊗
κ(t) : κ(t)n ↠ Q ⊗ κ(t) is a surjection of vector spaces, there is a nonzero q × q minor,
given by a subset I, of the q × n matrix p ⊗ κ(t). This implies that [p ⊗ κ(t)] ∈
TI (κ(t)).
Lemmas 1.2.2 and 1.2.4 together imply:
Proposition 1.2.5. The functor Gr(q, n) is representable by a scheme.

Exercise 1.2.6 (easy). Show that Gr(q, n) is a smooth scheme over Z of relative
dimension q(n − q).
Exercise 1.2.7 (good practice). Use the valuative criterion of properness to show
that Gr(q, n) → Spec Z is proper.

1.2.2 Projectivity of the Grassmannian

We show that the Grassmannian scheme Gr(q, n) is projective by explicitly providing
a projective embedding. The Plücker embedding is the map of functors
^q 
⊕n
P : Gr(q, n) → P OSpec Z
 ⊕n h^q ^q i
OT⊕n →

OT ↠ Q 7→ Q

defined above over a scheme T . As both sides are representable by schemes, the
morphism P corresponds to a morphism of schemes via Yoneda’s Lemma (0.3.12).
V 
q ⊕n
Proposition 1.2.8. The morphism P : Gr(q, n) → P OSpec Z is a closed
immersion. In particular, Gr(q, n) is a projective scheme over Z.
Vq ⊕n
Proof. A subset I ⊆ {1, . . . , n} of size q corresponds to a coordinate xI on P( OSpec ),
Vq ⊕n Vq ⊕n Z
and we set P( OSpec Z )I to be the open locus where xI ̸= 0. Note that P( OSpec Z )I ⊆
Vq ⊕n Vq ⊕n
P( OSpec Z ) is the subfunctor parameterizing line bundle quotients OS → L
eI Vq ⊕n
such that the composition OS −→ OS → L (where the first map is the inclusion
of the Ith term) is an isomorphism. It follows that there is a cartesian diagram of
functors
GrI
PI
/ P(Vq O⊕n )I
Spec Z

□
 
Gr(q, n)
P / P(Vq O⊕n ),
Spec Z
Vq ⊕n
where GrI is defined in (1.2.1). Since {P( OSpec Z )I } is a Zariski open cover of
Vq ⊕n Vq ⊕n
P( OSpec Z ), it suffices to show that each PI : Gr(q, n)I → P( OSpec Z )I is a
closed immersion.
For simplicity, assume that I = {1, . . . , q}. Under the isomorphisms GrI ∼ =
n
( ) −1
of Lemma 1.2.2 and P( OSpec Z )I ∼
q×(n−q) Vq ⊕n
AZ = AZ q
, the morphism PI corresponds
to the map
q×(n−q) (n)−1
AZ → AZq

51
 (n)−1
assigning a q × (n − q) matrix {xi,j } to the element of AZq whose Jth coordinate,
where J ⊂ {1, . . . , n} is a subset of length q distinct from I, is the {1, . . . , q} × J
minor of the q × n block matrix
 
1 x1,1 · · · x1,n−q
 1 x2,1 · · · x2,n−q 
.. .. .
 
. .

 
1 xq,1 ··· xq,n−q

q×(n−q)
The coordinate xi,j on AZ is the pullback of the coordinate corresponding to the
subset {1, · · · , bi, · · · , q, q + j} (see Figure 1.5.5). This shows that the corresponding
ring map is surjective thereby establishing that PI is a closed immersion.

Figure 1.2.9: The minor obtained by removing the ith column and all columns
q + 1, . . . , n other than q + j is precisely xi,j .

 
1 x1,1 ··· x1,j ··· x1,n−q
.. .. .. ..
. . . .
 
 
 
xi,j = det  1 xi,1 ··· xi,j ··· xi,n−q 
.. .. .. ..
 
. . . .
 
 
1 xq,1 ··· xq,j ··· xq,n−q

Exercise 1.2.10 (good practice). For a field k, define Gr(q, n)k := Gr(q, n) ×Z k,
and let p ∈ Gr(q, n)k be a point corresponding to a quotient Q = kn /K. Show that
there is a natural bijection of the tangent space
∼
Tp (Gr(q, n)k ) → Homk (K, Q).

with the vector space of k-linear maps K → Q.

52
Exercise 1.2.11 (good practice). Using the valuative criterion for properness for
Gr(q, n) (Exercise 1.2.7), provide an alternative proof of projectivity.

Exercise 1.2.12 (hard). Show that Gr(q, n) is identified with either the quotient
U/ GLq where U ⊆ Aq×n is the subset of q × n matrices of full rank, or the quotient
GLn /H, where
  
A B
H= A ∈ GLq , B ∈ Aq×(n−q) , C ∈ GLn−q ⊆ GLn .
0 C

Exercise 1.2.13 (hard). Show that Pic(Gr(q, n)) ∼

= Z and is generated by the
Plücker line bundle.

Exercise 1.2.14 (Flag varieties). Fix a positive integer n and a decreasing sequence
n > q1 > . . . > qs > 0 of integers. Show that the functor Sch → Sets, taking a
scheme S to isomorphism classes of sequences of quotients

OS⊕n ↠ V1 ↠ · · · ↠ Vs ,

where each Vi is a vector bundle on S of rank qi , is representable by a projective

scheme over Z.

1.2.3 Relative version

We prove that the functor Gr(q, V ) is representable and projective over S for any
vector bundle V on a noetherian scheme S .

Proof of Theorem 1.1.1. If V is a vector bundle over S of rank n, there is the relative
Plücker embedding
^ q 
P : Gr(q, V ) → P V
h^q ^q i
[VT ↠ Q] 7→ VT → Q

defined
Vq above over a S-scheme T , which is a morphism of functors over S. Since
P( V ) is projective over S, it suffices to show that this morphism is representable
by closed immersions. This property can be checked Zariski-locally: if U ⊆ VS is an
q
open subscheme where V is trivial, then the restriction of Gr(q, V
Vq ⊕n ) → P( V ) to
⊕n
U is base change of the Plücker embedding Gr(q, OU ) → P( OU ), which we’ve
already verified to be a closed immersion (Proposition 1.2.8), by U → Spec Z.

Since the Grassmannian functor is representable, there is a universal family

(see Definition 0.3.24), i.e., there is a universal quotient OGr(q,V ) ⊗S V → Quniv ,
where OGr(q,V ) ⊗S V is notation for the pullback of V under the structure morphism
Gr(q, V ) → S. The pullback of O(1) under the Plücker embedding is det(Quniv ),
which we sometimes call the Plücker line bundle. Thus, we obtain:

Corollary 1.2.15. The determinant det(Quniv ) of the universal quotient is a line

bundle on Gr(q, V ) which is relatively very ample over S.

53
1.3 Castelnuovo–Mumford regularity
Some of the deepest results in algebraic geometry concern the problem of
giving criteria for the higher cohomology groups of a sheaf to be 0... We
shall prove here (with the help of techniques developed and used by Nakai,
Matsusaka and Kleiman) only a weak vanishing theorem, but one which
is uniformly applicable to a large class of sheaves.

David Mumford [Mum66a, p. 99]

Serre’s vanishing theorem—sometimes called the Cartan–Serre–Grothendieck

theorem—states that if F is a coherent sheaf on a projective variety (X, OX (1)),
then for d ≫ 0
(1) F (d) is base point free,
(2) Hi (X, F (d)) = 0 for i > 0, and
(3) the multiplication map

H0 (X, F (d)) ⊗ H0 (X, O(k)) → H0 (X, F (d + k))

is surjective for all k ≥ 0.

Castelnuovo–Mumford regularity provides a quantitative measure of the size of d
necessary for the twist F (d) to satisfy these three cohomological properties.

1.3.1 Definition and basic properties

Definition 1.3.1. Let F be a coherent sheaf on projective space Pn over a field k.
For an integer m, we say that F is m-regular if

Hi (Pn , F (m − i)) = 0

for all i ≥ 1. The regularity of F is the smallest integer m such that F (m) is
m-regular.
It follows from the definition of regularity that if F is m-regular, then F (d)
is (m − d)-regular. We will show in Proposition 1.3.5 that if F is m-regular, it
also d-regular for all d ≥ m. While the requirement in the definition that the ith
cohomology of the (m − i)th twist vanishes may appear mysterious at first, this
definition is very convenient for induction arguments on the dimension, as evidenced
by the following result.
Lemma 1.3.2. Let F be an m-regular coherent sheaf on Pn over a field k. If H ⊆ Pn
is a hyperplane avoiding the associated points of F , then F |H is also m-regular.
Proof. The hypotheses imply that over an affine open subscheme U ⊆ Pn , the defining
equation of H is a nonzerodivisor for the module Γ(U, F ). Thus H : F (−1) → F is
injective, and for each integer i > 0, we have a short exact sequence

0 → F (m − i − 1) → F (m − i) → F |H (m − i) → 0,

which induces a long exact sequence on cohomology

· · · → Hi (Pn , F (m − i)) → Hi (H, F |H (m − i)) → Hi+1 (Pn , F (m − i − 1)) → · · · .

If F is m-regular, then Hi (Pn , F (m − i)) = Hi+1 (Pn , F (m − i − 1)) = 0. It follows

that Hi (H, F |H (m − i)) = 0 for all i > 0, and thus F |H is also m-regular.

54
Exercise 1.3.3 (easy, good practice).
(a) Show that O(d) is (−d)-regular on Pn .
(b) Show that the structure sheaf of a hypersurface H ⊆ Pn of degree d is (d − 1)-
regular.
(c) Show that the structure sheaf of a smooth curve C ⊆ Pn of genus g is (2g − 1)-
regular.
Exercise 1.3.4 (easy). Let 0 → K → F → Q → 0 be a short exact sequences of
coherent sheaves on Pn . If K is (m + 1)-regular and F is m-regular, show that Q is
m-regular.
Another advantage of regularity is the following result of Castelnuovo.
Proposition 1.3.5 (Properties of Regularity). Let F be an m-regular coherent
sheaf on Pn over a field k.
(1) For d ≥ m, F is d-regular.
(2) The multiplication map
H0 (Pn , F (d)) ⊗ H0 (Pn , O(k)) → H0 (Pn , F (d + k))
is surjective if d ≥ m and k ≥ 0.
(3) For d ≥ m, F (d) is globally generated and Hi (Pn , F (d)) = 0 for i ≥ 1.
Proof. If k → k′ is a field extension, then Flat Base Change (A.2.12) implies that
Hi (Pnk , F ) ⊗k k′ = Hi (Pnk′ , F ⊗k k′ ). As k → k′ is faithfully flat, the assertions (1)–(3)
can be checked after base change. We can thus assume that k is algebraically closed
and, in particular, infinite. For (1) and (2), we will argue by induction on n with
the base case of n = 0 being clear. If n > 0, since k is infinite, we may choose a
hyperplane H ⊆ Pn avoiding the associated points of F . Since the restriction F |H
is m-regular (Lemma 1.3.2) on H ∼ = Pn−1 , the inductive hypothesis implies that (1)
and (2) hold for F |H .
We prove (1) by also arguing via induction on the integer d. The base case d = m
holds by hypothesis. For d > m, the short exact sequence 0 → F (d − i − 1) →
F (d − i) → F |H (d − i) → 0 induces a long exact sequence on cohomology
· · · → Hi (Pn , F (d − i − 1)) → Hi (Pn , F (d − i)) → Hi (H, F |H (d − i)) → · · · .
For i > 0, the first term vanishes by the induction hypothesis on d since F is
(d − 1)-regular. The third term above vanishes by the inductive hypothesis on n:
F |H is m-regular by Lemma 1.3.2 and thus d-regular by the inductive hypothesis
on n, hence Hi (H, F |H (d − i)) = 0. Thus, the second term vanishes and we have
established (1).
To show (2), we use induction on k in addition to n. We denote the multiplication
map by
µd,k : H0 (Pn , F (d)) ⊗ H0 (Pn , O(k)) → H0 (Pn , F (d + k)).
While the base case k = 0 is clear, the induction argument will require us to directly
establish the case k = 1. To this end, we consider the commutative diagram

H0 (Pn , F (d)) ⊗ H0 (Pn , O(1))

νd ⊗res
/ H0 (H, F |H (d)) ⊗ H0 (H, OH (1))
4
id ⊗H µd,1
 νd+1

H0 (Pn , F (d))
α / H0 (Pn , F (d + 1)) / H0 (H, F |H (d + 1)).
(1.3.6)

55
The map α is given by multiplication by H ∈ H0 (Pn , O(1)), and there is an in-
clusion im(α) ⊆ im(µd,1 ). Since H1 (Pn , F (d)) = 0 by (1), the restriction map
νd : H0 (Pn , F (d)) → H0 (H, F |H (d)) is surjective. Likewise, since H1 (Pn , O) = 0,
res : H0 (Pn , O(1)) → H0 (H, OH (1)) is surjective, and so the top horizontal arrow is
surjective. The inductive hypothesis applied to H = Pn−1 implies that the right
vertical arrow is surjective. Therefore, the composition νd+1 ◦ µd,1 is surjective and
it follows that im(µd,1 ) surjects onto H0 (H, F |H (d + 1)). By exactness of the bottom
row, we have that

H0 (Pn , F (d + 1)) = im(µd,1 ) + ker(νd+1 ) = im(µd,1 ) + im(α) = im(µd,1 ),

which shows that µd,1 is surjective.

If k > 1, we consider the commutative square

H0 (Pn , F (d)) ⊗ H0 (Pn , O(k − 1)) ⊗ H0 (Pn , O(1)) / H0 (Pn , F (d)) ⊗ H0 (Pn , O(k))

µd,k−1 ⊗id µd,k

 µd+k−1,1

H0 (Pn , F (d + k − 1)) ⊗ H0 (Pn , O(1)) / H0 (Pn , F (d + k)).

The left vertical map and bottom horizontal arrow are surjective by the inductive
hypothesis applied to k − 1 and k = 1, respectively. It follows that µd,k is surjective.
To show (3), we know that for k ≫ 0, F (d + k) is globally generated, i.e.,
γF (d+k) : H0 (Pn , F (d+k))⊗OPn → F (d+k) is surjective. Consider the commutative
square
µd,k ⊗id
H0 (Pn , F (d)) ⊗ H0 (Pn , O(k)) ⊗ OPn / H0 (Pn , F (d + k)) ⊗ OPn

γF (d) ⊗id γF (d+k)

 id ⊗γO(k) 
/ F (d) ⊗ O(k).


F (d) ⊗ H0 (Pn , O(k)) ⊗ OPn

Since the top horizontal arrow is surjective by (2), the composition from the top
left to the bottom right is surjective. Since the bottom horizontal map is defined by
tensoring with F (d), the map γF (d) must be surjective. Finally, to see the vanishing
of the higher cohomology of F (d), observe that for each i > 0, the sheaf F is
(d + i)-regular by (1) and thus Hi (Pn , F (d)) = 0.
One direct consequence of (1) is that if F is m-regular, then the restriction map

νd : : H0 (Pn , F (d)) → H0 (H, F |H (d))

is surjective for all d ≥ m. Indeed, (1) implies that F is also d-regular and the
surjectivity follows from the vanishing of H1 (Pn , F (d − 1)). The following lemma—
which will be used in the proof of Theorem 1.3.8—shows that we can still arrange
for the surjectivity of νd under weaker hypotheses.
Lemma 1.3.7. Let F be a coherent sheaf on Pn and H be a hyperplane avoiding
the associated points of F . If F |H is m-regular and νd is surjective for some d ≥ m,
then νp is surjective for all p ≥ d.
Proof. By staring at diagram (1.3.6), we see that the top arrow νd ⊗ res is surjective
(as both νd and res are surjective) and the vertical right multiplication morphism
is surjective (by applying Proposition 1.3.5(2) to the m-regular sheaf F |H ). This
implies that νd+1 is surjective, and the statement follows by induction.

56
1.3.2 Regularity bounds
We prove a boundedness result for the regularity of subsheaves of the trivial vector
bundle first established by Mumford in [Mum66a, p.101]. This is the basis of almost
every boundedness result in algebraic geometry.1

Theorem 1.3.8 (Boundedness of Regularity). For every pair of nonnegative integers

r and n, and for every polynomial P ∈ Q[z], there exists an integer m0 with the
following property: for every field k, every subsheaf F ⊂ OP⊕r
n with Hilbert polynomial
k
P is m0 -regular.

Proof. Since being m-regular is insensitive to field extensions, we can assume that k
is infinite. We will argue by induction on n. The base case of n = 0 holds as every
sheaf F on P0 is m-regular for every integer m.
For n ≥ 1 and a subsheaf F ⊆ OP⊕r n with Hilbert polynomial P , we can choose

a hyperplane H ⊆ Pn avoiding all associated points of OP⊕r n /F . This ensures that

TorO1
Pn
(O H , O ⊕r
P n /F ) = 0 and that the short exact sequence 0 → F → OP⊕r n →
⊕r
OPn /F → 0 restricts to a short exact sequence
⊕r ⊕r
0 → F |H → OH → OH /F → 0. (1.3.9)

As H ∼= Pn−1 , we may apply the inductive hypothesis to F |H ⊆ OH⊕r

. On the other
⊕r
hand, since F ⊆ OPn is torsion free, we have a short exact sequence
H
0 → F (−1) −→ F → F |H → 0, (1.3.10)

and the Hilbert polynomial of F |H is χ(F |H (d)) = χ(F (d)) − χ(F (d − 1)) =
P (d) − P (d − 1). In particular, the Hilbert polynomial of F |H only depends on P ,
⊕r
and the inductive hypothesis applied to F |H ⊆ OH gives an integer m1 such that
F |H is m1 -regular.
For m ≥ m1 − 1, since Hi (H, F |H (m)) = 0 for all i ≥ 1, we have a long exact
sequence

0 → H0 (Pn , F (m − 1)) → H0 (Pn , F (m)) → H0 (H, F |H (m)) →

H1 (Pn , F (m − 1)) → H1 (Pn , F (m)) → 0 (1.3.11)
∼
along with isomorphisms Hi (Pn , F (m − 1)) → Hi (Pn , F (m)) for i ≥ 2. Since
Hi (Pn , F (d)) vanishes for d ≫ 0, we can conclude that Hi (Pn , F (m − 1)) = 0 for
m ≥ m1 − 1 and i ≥ 2.
To handle H1 , we use the inequalities h1 (Pn , F (m1 )) ≥ h1 (Pn , F (m1 + 1)) ≥
· · · , which eventually stabilize to 0. We claim that in fact that the inequalities
h1 (Pn , F (m1 )) > h1 (Pn , F (m1 + 1)) > · · · are strict until they become 0. To see
this, we observe that there is an equality h1 (Pn , F (m − 1)) = h1 (Pn , F (m)) for
m ≥ m1 if and only if νm : H0 (Pn , F (m)) → H0 (H, F |H (m)) is surjective. Suppose
that νm′ is surjective for some m′ ≥ m. Since F |H is m1 -regular, we may apply
Lemma 1.3.7 to conclude that µm′′ is surjective for all m′′ ≥ m′ . Thus h1 (Pn , F (m′′ ))
is constant for m′′ ≥ m′ , and therefore zero. This establishes the claim. Setting
m2 = m1 + 1 + h1 (Pn , F (m1 )), we see that h1 (Pn , F (m2 − 1)) = 0 and that F is
m2 -regular.
1 Even in [SP, Tag 0DPA] where Hilbert and Quot schemes are constructed using Artin’s

Axioms for Algebraicity (C.7.4), it is this result [SP, Tag 08AG] that is applied to establish
quasi-compactness.

57
 It remains to show that m2 is bounded above by a constant m0 independent of
F . Since F ⊆ OP⊕r n , we have that h (P , F (d)) ≤ rh (P , O(d)) = r
n+d
for any
0 n 0 n


n
d ≥ 0. Using the vanishing of hi (Pn , F (m1 )) for i ≥ 2, we have

h1 (Pn , F (m1 )) = h0 (Pn , F (m1 )) − χ(F (m1 ))

 
n + m1
≤r + P (m1 ).
n

Defining m0 := m1 + 1 + r n+m + P (m1 ) ≥ m2 gives the desired constant such

1


n
that every F ⊆ OP⊕r
n with Hilbert polynomial P is m0 -regular.

Remark 1.3.12. The above proof establishes in fact a stronger statement. To

formulate the result, we recall that every numerical polynomial P ∈ Q[z] (i.e.,
P (d) ∈ Z for integers d ≫ 0) of degree n can be expressed uniquely as
n  
X d
P (d) = ai
i=0
i

for ai ∈ Z; this follows from a straightforward inductive argument, c.f., [Har77, Prop.
I.7.3]. For nonnegative integers r and n, there exists a polynomial

Λr,n ∈ Z[x0 , . . . , xn ] (1.3.13)

with the following property: for every field k, every subsheaf F ⊆ OP⊕r n with Hilbert
k

polynomial P (d) = i ai i is m0 -regular for m0 = Λr,n (a0 , . . . , an ). See [Mum66a,

d
P

p. 101].
Exercise 1.3.14. Let P = d be the constant polynomial for a fixed integer d.
What is the optimal m such that every ideal sheaf I ⊆ OPn defining a dimension 0
subscheme of length d is m-regular?
Remark 1.3.15 (Optimal bounds). Although Mumford’s result on Boundedness of
Regularity (Theorem 1.3.8) provides an explicit bound and is sufficient for many
applications including the construction of the Hilbert scheme, there is a more optimal
bound established by Gotzmann: for a projective scheme X ⊆ PN over a field k
with Hilbert polynomial P , there are unique integers λ1 ≥ λ2 ≥ · · · ≥ λr ≥ 1 such
that P can be expressed as
     
d + λ1 − 1 d + λ2 − 2 d + λr − r
P (d) = + + ··· + ,
λ1 − 1 λ2 − 1 λr − 1

and the ideal sheaf IX of X is r-regular. See [Got78], [BH93, §4.3], [Gre89], and
[Gre98, §3].
Exercise 1.3.16. Let C ⊆ Pn be a curve of degree d and
genus g. Show that
Gotzmann’s bound implies that the ideal sheaf IC of C is ( d2 + 1 − g)-regular. Can
you compare this to the bound given by the proof of Theorem 1.3.8, i.e., can you
compute Λ1,n (1 − g, d) from (1.3.13)?
Remark 1.3.17. It was shown in [GLP83] that the ideal sheaf IC of an integral,
non-degenerate curve C ⊆ PN of degree d is (d − N + 2)-regular. It is conjectured
more generally that the ideal sheaf of a smooth, non-degenerate projective subvariety
X ⊆ PN of dimension n and degree d is (d − (N − n)) + 1)-regular; see [GLP83] and
[EG84].

58
Proposition 1.3.18 (Regularity in Families). Let S be a noetherian scheme,
π : PnS → S be relative projective space, and Q be a coherent sheaf on PnS flat
over S. Suppose that there is an integer m > 0 such that for every s ∈ S, the
restriction Qs to Pnκ(s) is m-regular. Then for d ≥ m,
(1) π∗ Q(d) is a vector bundles whose construction commutes with base change by
f : T → S, i.e., f ∗ π∗ Q(d) ∼
= πT,∗ QT (d),
(2) Ri π∗ Q(d) = 0 for i > 0, and
(3) π ∗ π∗ Q(d) → Q(d) is surjective.

Proof. Since Ri π∗ Q(d) ⊗ κ(s) → Hi (Pnκ (s), Qs (d)) = 0 is surjective for all s ∈ S,
by Cohomology and Base Change (A.6.8) yields (1) and (2). For (3), let C =
coker(π ∗ π∗ Q(d) → Q(d)). For each s ∈ S, we have a cartesian diagram

Pnκ(s)
j
/ Pn
S

πs □ π
 
Spec κ(s)
i / S.

Using (1), j ∗ π ∗ π∗ Q(d) = πs∗ i∗ π∗ Q(d) = πs∗ πs,∗ Qs (d), the pullback of the adjunction
map π ∗ π∗ Q(d) → Q(d) under j corresponds to the adjunction map πs∗ πs,∗ Qs (d) →
Qs (d), which we know to be surjective by Properties of Regularity (1.3.5(3)). Thus
C ⊗ κ(s) = coker(πs∗ πs,∗ Qs (d) → Qs (d)) = 0 for all s ∈ S, and π ∗ π∗ Q(d) → Q(d) is
surjective.

1.4 Projectivity of Hilb and Quot

Projective geometry is all geometry.

Arthur Cayley

We prove the Projectivity of Hilbert and Quot Schemes (1.1.2 and 1.1.3).

1.4.1 Special case when F is a direct sum of line bundles

Let S be a noetherian scheme and π : PnS → S be relative projective space. We first
consider the special case of a coherent sheaf

F := OPnS (−l)⊕r

for integers l and r ≥ 1. For each integer d ≥ 0, we can try to define morphisms of
functors
QuotP (F/PnS ) → Gr(P (d), π∗ F (d))
[FT ↠ Q] 7→ [πT,∗ FT (d) → πT,∗ Q(d)],
where [FT ↠ Q] is an object over an S-scheme T .
Step 1: There exists an integer m0 such that QuotP (F/PnS ) → Gr(P (d), π∗ F (d))
is well-defined for d ≥ m0 .

⊕r
Proof. Since F = OPnS (−l)⊕r , for d ≥ l, we have that π∗ F (d) = Symd−l Γ(S, OS )
is a vector bundle whose construction commutes with base change by T → S and

59
Ri π∗ F (d) = 0 for all i > 0. For every field-valued point s ∈ S(k) and quotient
Fs = OPnk (−l)⊕r ↠ Q of Hilbert polynomial P with kernel Ks = ker(Fs → Q), then
K(l) is a subsheaf of OP⊕rn whose Hilbert polynomial is determined by P and the
k
⊕r
Hilbert polynomial of OPn . We may therefore apply Boundedness of Regularity
k
(1.3.8) to choose m0 ≥ l such that every such kernel Ks is m0 -regular. By possibly
increasing m0 , we can arrange that Fs is also m0 -regular. Exercise 1.3.4 implies
that each quotient Q is m0 -regular. By Regularity in Families (1.3.18(1)), for each
d ≥ m0 and quotient sheaf FT ↠ Q flat over T with Hilbert polynomial P , πT,∗ Q(d)
is a vector bundle on T of rank P (d) whose construction commutes with base change.
Since FT and Q are flat over T , so is K := ker(FT → Q). Since every fiber Ks is m0 -
regular, Regularity in Families (1.3.18(2)) implies that for d ≥ m0 , R1 πT,∗ K(d) = 0.
Therefore πT,∗ FT (d) → πT,∗ Q(d) is surjective and (1.4.1) is well-defined.

Step 2: There exists an integer m0 such that QuotP (F/PnS ) → Gr(P (d), π∗ F (d))
is a monomorphism for d ≥ m0 .

Proof. Let FT ↠ Q be a quotient sheaf flat over a scheme T with Hilbert polynomial
P . The kernel K = ker(FT → Q) is also flat over T . For m0 chosen as in Step 1, for
every s ∈ S, the fibers Ks , (FT )s , and Qs are m0 -regular. By Regularity of Families
(1.3.18(2)), for d ≥ m0 , R1 πT,∗ K(d) = 0, which yields a commutative diagram of
short exact sequences

0 / π ∗ πT,∗ K(d) / π ∗ πT,∗ FT (d) / π ∗ πT,∗ Q(d) /0

T T T
α

 '  
0 / K(d) / FT (d) / Q(d) /0

By Regularity of Families (1.3.18(3)), each vertical map is surjective. Thus Q(d) =

coker(α) is uniquely determined by πT,∗ FT (d) → πT,∗ Q(d).

Step 3: There exists an integer m0 such that QuotP (F/PnS ) → Gr(P (d), π∗ F (d))
is a locally closed immersion for d ≥ m0 .

Proof. Let T → Gr(P (d), π∗ F (d)) be a map determined by a vector bundle quotient
(π∗ F (d))T = πT,∗ FT (d) ↠ V , and let K be the kernel. Consider the fiber product

Y /T

□ [πT ,∗ FT (d)↠V ]
 
QuotP (F/PnS ) / Gr(P (d), π∗ F (d)).

On PnT , we have a commutative diagram

0 / π∗ K / π ∗ πT,∗ FT (d) / π∗ V /0
T T T
α

&  
FT (d) / Q(d) / 0,

of sheaves on PnT , where Q is defined such that Q(d) = coker(α). Since K is a vector
bundle on T , K = πT,∗ πT∗ K, and thus πT,∗ α is identified with the inclusion K ,→

60
πT,∗ FT (d). Applying πT,∗ to the bottom sequence and using that R1 πT,∗ (πT∗ K) = 0,
we obtain a short exact sequence 0 → K = πT,∗ πT∗ K → πT,∗ FT (d) → πT,∗ Q(d) → 0.
This gives the identification πT,∗ Q(d) = V .
If Q is flat over T with Hilbert polynomial P , then [FT ↠ Q] : T → QuotP (F/PnS )
provides a lift of the given map [πT,∗ FT (d) ↠ V ] : T → Gr(P (d), π∗ F (d)), or, in
∼
other words, Y → T is an isomorphism. This analysis shows that Y is identified
with the subfunctor of T defined by
Sch/T → Sets,
if QT ′ is flat over T ′ with Hilbert polynomial P

′ {∗}
(T → T ) 7→
∅ otherwise.

Existence of Flattening Stratifications2 (A.2.16) implies that Y is representable by

a locally closed subscheme of T .
We summarize the conclusion as:
Proposition 1.4.1. Let S be a noetherian scheme and π : Pn → S be relative
projective space. If F := OPnS (−l)⊕r and P ∈ Q[z], there exists an integer m0 such
that for all d ≥ m0 ,

QuotP (F/PnS ) → Gr(P (d), π∗ F (d))

is a locally closed immersion. In particular, QuotP (F/PnS ) is representable by a

quasi-projective scheme over S.

1.4.2 Valuative criteria for Hilb and Quot

To establish that Quot is projective, it will be sufficient to know that it is proper.
Proposition 1.4.2. For every noetherian scheme S, coherent sheaf F on PnS , and
polynomial P ∈ Q[z], the functor QuotP (F/PnS ) satisfies the valuative criterion for
properness over S, i.e., for every DVR R over S with fraction field K, every flat
coherent quotient FK → Q on PnK with Hilbert polynomial P extends uniquely to a
flat coherent quotient FR → Q e on Pn with Hilbert polynomial P .
R

Remark 1.4.3. In other words, the proposition implies that for every commutative
diagram
Spec K
Q
/ QuotP (F )
8 X/S
Q
e

 
Spec R / S,

has a unique lift. See §3.8 for a discussion of the valuative criterion for functors and
stacks.
Proof. If we write j : XK ,→ XR as the open immersion, we define Q e as the image of
the composition FR → j∗ FK → j∗ Q, where the first map is given by the adjunction
FR → j∗ j ∗ FR = j∗ FK . Since Q
e is a subsheaf of j∗ Q, it is torsion free over R and
thus flat (as R is a DVR). Finally, since Q if flat over R and Spec R is connected,
its Hilbert polynomial is constant.
2 The Existence of Flattening Stratifications is arguably the deepest component of the proof of

the Projectivity of the Quot Scheme (1.1.3).

61
Remark 1.4.4. For HilbP X/S , the argument translates into the following: the unique
extension of a closed subscheme Z ⊆ XK is the scheme-theoretic image Z e = im(Z →
XK ,→ XR ). The scheme Z is flat over R as all associated points live over the
e
generic point of Spec R.

1.4.3 Projectivity
The following completes the proofs of Theorems 1.1.2 and 1.1.3.
Theorem 1.4.5. Let S be a noetherian scheme and π : Pn → S be relative projective
space. If F := OPnS (−l)⊕r and P ∈ Q[z], there exists an integer m0 such that for all
d ≥ m0 ,
QuotP (F/PnS ) → Gr(P (d), π∗ F (d))
is a closed immersion. In particular, QuotP (F/PnS ) is representable by a projective
scheme over S.

Proof. Let S be a noetherian scheme, π : PnS → S relative projective space, and F

be any coherent sheaf on PnS . Choosing integers l and r together with a surjection
G := OPnS (−l)⊕r ↠ F , we obtain a morphism of functors

QuotP (F/PnS ) → QuotP (G/PnS )

[FT → Q] 7→ [GT ↠ FT ↠ Q],

defined over an S-scheme T . The functor QuotP (G/PnS ) is representable by a

quasi-projective scheme over S by Proposition 1.4.1 and is proper over S by Propo-
sition 1.4.2, hence it is projective.
To prove that QuotP (F/PnS ) is projective over S, it suffices to prove that the
above morphism is representable by closed immersions. This boils down to the claim
that for an S-scheme T and quotient GT ↠ Q with Q flat over T , there is a closed
subscheme Z ⊆ T such that a morphism T ′ → T factors through Z if and only if
GT ′ → QT ′ factors through FT ′ . Defining K = ker(GT → FT ) and considering the
diagram
0 /K / GT / FT /0


Q,
then GT ′ → QT ′ factors through FT ′ if and only if KT ′ → QT ′ is zero. By
Exercise 0.3.22, the subfunctor of T , parameterizing maps T ′ → T such that
KT ′ → QT ′ is zero, is representable by a closed subscheme.
Taking G = OPnS (−l)⊕r ↠ F to be a surjection as in the proof above, for d ≫ 0,
we have a composition of closed immersions
^P (d)
QuotP (F/PnS ) ,→ QuotP (G/PnS ) ,→ Gr(P (d), π∗ G(d)) ,→ P( π∗ G(d)).

Letting FQuotP (F/PnS ) → Quniv denote the universal quotient on PnS ×S QuotP (F/PnS ),
the pullback of O(1) under the above composition is identified with det p2,∗ (Quniv (d)).
We summarize this as follows:
Corollary 1.4.6. For d ≫ 0, the line bundle det p2,∗ (Quniv (d)) on QuotP (F/PnS )

is relatively very ample over S.

62
1.4.4 Generalizations
The first generalization we consider is the Flag Scheme.

Exercise 1.4.7 (Flag Schemes). Let S be a noetherian scheme and F be a coherent

sheaf on PnS that can be written as a quotient of OPnS (−l)⊕r for some l and r. For
every sequence P1 , . . . , Pk ∈ Q[z] of polynomials, show that the functor

FlagP1 ,...,Pk (F/PnS /S) : Sch/S → Sets

filtrations 0 = F0 ⊆ · · · ⊆ Fk = FT of quasi-
 
 
coherent sheaves on PnT such that each quotient F i /F i−1

 

(T → S) 7→
 is a finitely presented OPnT -module and flat over T 
with Hilbert polynomial Pi

 


is representable by a scheme projective over S.

Hint: Construct the Flag scheme as an iteration of relative Quot schemes, i.e.,
consider Flag := FlagP2 ,...,Pk (F/PnS /S) and the universal filtration 0 = Funiv,0 ⊆
Funiv,2 ⊆ · · · ⊆ p∗1 F on PnS ×S Flag, and show that FlagP1 ,...,Pk (F/PnS /S) ∼ =
QuotP1 (Funiv,2 / Flag /S).

The existence and projectivity of the Quot scheme QuotP (F/PnS ) was proven in
Theorem 1.1.3 for coherent sheaves F on PnS that that can be written as a quotient
of OPnS (−l)⊕r . We now consider more general setups. Recall that there are three
distinct notions of projectivity for a morphism X → S, listed below in increasing
generality:

– X → S factors as a closed immersion X ,→ PnS and PnS → S; this is sometimes

referred to as H-projective as this is the definition in [Har77, II.4],
– X → S factors as a closed immersion X ,→ P(E) and P(E) → S, where E is a
vector bundle over S; this is called strongly projective, and
– X → S factors as a closed immersion X ,→ P(E) and P(E) → S, where E
is a finite type, quasi-coherent sheaf on S (i.e., a coherent sheaf when S is
noetherian); this is the definition in [EGA, §II.5], [SP, Tag 01W8].

When S has the resolution property, i.e., every coherent sheaf is the quotient of a
vector bundle, then a projective morphism is strongly projective. This holds for
every smooth or quasi-projective scheme. The definitions of the Hilbert functor
HilbP (X/S) and Quot functor QuotP (F/X/S) extend to any projective morphism
X → S and finite type, quasi-coherent sheaf F on X.

Exercise 1.4.8. Let X → S be a projective morphism of schemes and F be a

coherent sheaf on X.
(a) (easy) Show that QuotP (F/X/S) is H-projective if X → S is H-projective and
F is a quotient of OX (−l)⊕r .
(b) (moderate) Show that QuotP (F/X/S) is strongly projective if f : X → S is
strongly projective and F is a quotient of f ∗ V ⊗ OX (−l) where V is a vector
bundle on S.
(c) (moderate) Show that QuotP (F/X/S) is a proper scheme over S if X → S is
projective.
(d) (hard) Show that QuotP (F/X/S) is projective if X → S is projective and F
is flat over S.

63
 (e) (open) Is QuotP (F/X/S) always projective if X → S is projective?
See also [AK80, Thm. 2.6] and [Nit05, Thm. 5.2].

Exercise 1.4.9 (hard). Suppose that X → S is strongly quasi-projective morphism

of noetherian scheme, i.e., there is a locally closed immersion X ,→ P(V ) where V is
a vector bundle on S. If F is a coherent sheaf on X and P ∈ Q[z] is a polynomial,
we can modify the Quot functor as follows:

QuotP (F/X/S) : Sch/S → Sets

finitely presented, quasi-coherent quotients
 
 
FT → Q on XT which are flat and have

 

(T → S) 7→ .
 proper support over T such that Qt on Xt 
has Hilbert polynomial P for all t ∈ T

 


The Hilbert functor HilbP (X/S) is defined analogously by only parameterizing closed
subschemes with proper support over the base. Show that HilbP X/S and QuotX/S (F )
P

are representable by strongly quasi-projective schemes over S. See also [FGAIV, §4],
[AK80, Thm. 2.6], or [Nit05, §6].

Remark 1.4.10. If X → S is merely a separated morphism of noetherian schemes,

then one can define functors Hilb(X/S) and Quot(F/X/S) as above after dropping
the condition on the Hilbert polynomial P . Artin’s Axioms for Algebraicity (C.7.4)
can be applied to show that these functors are representable by algebraic spaces
separated and locally of finite type over S; see [Art69b, Thm. 6.1] and [SP, Tag 09TQ],
or Theorem C.7.7 for a special case. Examples of Hironaka produce smooth proper
(but not projective) threefolds X over C such that Hilb(X/S) is not a scheme.

1.4.5 Chow varieties and other variants

The Chow variety is a variant of the Hilbert scheme that is easier to construct
but doesn’t afford as nice functorial properties. The Chow variety Chowr,d (Pnk )
parameterizes effective cycles on Pnk of dimension r and degree d, and can be
constructedPby using the Chow form Chowα of an Q effective cycle α. For an effective
cycle α = i ai [Xi ], one can define Chowα = i ChowaXii , so it suffices to define
the Chow form for an integral closed subscheme X ⊆ Pnk of dimension r and
degree d. Letting Pn,∨ denote the dual projective space parameterizing hyperplanes
H ∈ Γ(Pn , O(1)), define the locus

{(H0 , . . . , Hr ) | X ∩ H0 ∩ · · · · · · ∩ Hr ̸= ∅} ⊆ (Pn,∨ )r+1 . (1.4.11)

As this is a divisor, there is a polynomial in the variables ui,j with 0 ≤ i ≤ r and

0≤j≤n
ChowX (uij ) ∈ Symd (kn+1 )⊗(r+1) ,
which is homogenous of degree d in ui,0 , . . . , ui,n for each i, such that ChowX (H0 , . . . , Hr ) =
0 if and only if X ∩ H0 ∩ · · · · · · ∩ Hr ̸= ∅.
The Chow variety Chowr,d (Pn ) is the closure of the set of Chow forms of all
effective cycles on Pn of dimension r and degree d. The main existence theorem in
characteristic 0 asserts that

– Chowr,d (Pn ) is projective and seminormal (i.e. every finite bijective morphism
Y → Chowr,d (Pn ) inducing isomorphisms on residue fields is an isomorphism),

64
 – Chowr,d (Pn ) represents a functor on the category of seminormal k-schemes:
for a seminormal k scheme S, Mor(S, Chowr,d (Pn )) are identified with well-
defined families of effective algebraic cycles on PnS of dimension r and degree
d, and
r
– for a polynomial P (t) = d· tr! +(lower terms), there is a Hilbert–Chow morphism

HilbP (Pn )sn → Chowr,d (Pn ),

from the seminormalization of HilbP (Pn ), taking a closed subscheme Z ⊂ Pn

to the cycle ai [Zi ] were the Zi are the reduced scheme structures of the ith
irreducible components and ai is its multiplicity of Zi at its generic point.
See [CW37], [HP47, §X.6-8], and [Sam55], or the modern treatments in [GIT, §5.4]
and [Kol96, I.3.21, I.6.1]. In our sketch of the GIT construction of M g , we will
utilize the Chow variety for curves in Pn (i.e., the r = 1 case), in which case the
statements are slightly easier to prove.
Example 1.4.12 (Additional variants). There are further variants and general-
izations. For instance, Vistoli constructs a Hilbert stack parameterizing finite and
unramified morphisms to a separated scheme X [Vis91]. Alexeev and Knutson’s
moduli of branch varieties parameterizes finite morphisms from a geometrically
reduced proper scheme to a separated scheme X [AK10].
Exercise 1.4.13 (Schemes of morphisms). For strongly projective morphisms
X → S and Y → S of noetherian schemes, consider the functor

MorS (X, Y ) : Sch/S → Sets

(T → S) 7→ MorT (XT , YT )

assigning an S-scheme T to the set of T -morphisms XT → YT . By using a suitable

closed subscheme of HilbP
X×S Y /X parameterizing graphs X ⊆ X ×S Y of morphisms
X → Y , show that MorS (X, Y ) is representable by a projective scheme over S.

1.5 An invitation to the geometry of Hilbert schemes

Murphy’s Law for Hilbert Schemes: There is no geometric possibility so
horrible that it cannot be found generically on some component of the
Hilbert scheme.
Joe Harris and Ian Morrison [HM98, p.18]

Hilbert schemes are some of the most well-studied moduli spaces, with perhaps
only M g and Bun psr,d (C) having received greater attention over the last 50 years. As
such, we will not attempt a systematic exposition, but merely offer a few interesting
examples and features.

1.5.1 First examples

In this section,
Pd we work over an algebraically closed field k. The Hilbert polynomial
P (z) = i=0 ai z i of a projective scheme X ⊂ Pn encodes invariants of X. For
instance, dim X is the degree d of P and deg X is the normalized leading coefficient
d!ad . Riemann–Roch implies that P (z) = deg(C)z + (1 − g) for a smooth curve

65
C ⊆ Pn and P (z) = 12 (zH · (zH − K)) + (1 − pa ) for a smooth surface S ⊂ Pn , where
H is a hyperplane divisor, K is the canonical divisor, and pa = 1 − χ(OS ) is the
arithmetic
R genus. In arbitrary dimension, Hirzebruch–Riemann–Roch states that
P (z) = X ch(OX (z))td(X), where ch(OX (z)) is the Chern character and td(X) the
Todd class.

Exercise 1.5.1 (Hypersurfaces and linear subspaces).

(a) A hypersurface H ⊆ Pn of degree d has Hilbert polynomial
   
n+z n+z−d
P (z) = χ(OPn (z)) − χ(OPn (z − d)) = − .
n n

Show that HilbP (Pn ) ∼

= P(Γ(Pn , O(d))).
(b) A linear subspace L ⊆ Pn of dimension k has Hilbert polynomial P (z) = z+k


k
and HilbP (Pn ) = Gr(k + 1, n + 1).

Example 1.5.2 (Hilbert scheme of points on a curve). If C is a smooth projective

curve, then the Hilbert scheme Hilbn (C) of n points (viewing n as the constant
polynomial) is a smooth irreducible projective variety isomorphic to the symmetric
product
Symn C := C × · · · × C /Sn ,
| {z }
n

where Sn acts by permuting the factors. The quotient exists as a projective variety
since C × · · · × C is projective; see Exercise 4.2.14.

Example 1.5.3 (Hilbert scheme of points on a surface). If S is a smooth irreducible

projective surface, then the Hilbert scheme of n points Hilbn (S) is a smooth irre-
ducible projective variety [Fog68]. See also [Nak99a] and [Mac07, §4]. There is a
birational morphism

Hilbn (S) → Symn (S) := S × · · · × S /Sn ,

| {z }
n

of projective varieties. The symmetric product Symn (S) is not smooth for n > 1 and
this provides a resolution of singularities. For an unordered collection of (possibly
non-distinct) points (p1 , . . . , pn ) ∈ Symn (S), the fiber consists of all possible scheme
structures on {p1 , . . . , pn } of length n.
The case of n = 2 is the first interesting case, as Hilb1 (S) = S. For any
point p ∈ S, there are many non-reduced scheme structures of length 2 supported
at p. They are parameterized by their “tangent direction”: given [a : b] ∈ P1 ,
k[x, y]/(x2 , xy, y 2 , ay − bx) defines a length 2 non-reduced subscheme. In this case,
Hilb2 (S) → Sym2 (S) is the blowup of the diagonal S ,→ Sym2 (S) given by p 7→ (p, p).
For n > 2, the map Hilbn (S) → Symn (S) is a blowup along some ideal sheaf [Hai98]
but the description of the ideal sheaf is more complicated. When X is of arbitrary
dimension, Hilbn (X) is smooth at (reduced) closed subschemes Z ⊆ X consisting
of n distinct smooth points of X. If X is reduced, there is an open subscheme of
Hilbn (X) dimension n dim(X) parameterizing n distinct smooth points. Another
result of Fogarty is that Hilbn (X) is connected as long as X is connected [Fog68].
Moreover, for every projective scheme X, there is an irreducible component Hilbn (X),
called the “good component,” that can be identified with the blowup of Symn (X)
along an ideal sheaf [ES14].

66
Example 1.5.4 (Twisted cubics). The Hilbert scheme Hilb3z+1 (P3 ) consists of the
union of two smooth rational irreducible components H and H ′ of dimensions 12
and 15 intersecting transversely along a smooth rational subvariety of dimension 11
[PS85]. The locus H is the closure of the locus H0 consisting of twisted cubics, i.e.,
rational smooth curves in P3 of degree 3. Each twisted cubic can be represented by a
map P1 → P3 given by the line bundle OP1 (3) and a choice of basis of Γ(P1 , OP1 (3)),
and this representation is unique up to automorphisms of P1 . All such curves are
projectively equivalent, i.e., differ by an automorphism of P3 , so we see that H0 is
identified with the homogeneous space Aut(P3 )/ Aut(P1 ) = PGL4 / PGL2 , which is
smooth and irreducible of dimension 12. The locus H0 is not proper as it includes
families such as P1 ,→ P3 given by [x, y] 7→ [x3 , x2 y, xy 2 , ty 3 ] parameterized by t ∈ A1
whose limit is a singular curve C0 supported on a nodal cubic in V (w) = P2 (where
w is the 4th coordinate) but with an embedded point at the node; see [Har77, Ex.
9.8.4].

H dim12

ooo

to
H dim

Figure 1.5.5: Hilb3z+1 (P3 )

TIM
sooooo

The locus H ′ is the closure of the locus H0′ consisting of subschemes C ⨿ {p}
where C is a smooth cubic curve contained in a hyperplane H and p ∈ P3 ∖ C. To
count the dimension, observe that the choice of hyperplane H ∈ P(H0 (P3 , O(1))) is
a
given by 3 parameters, the choice of plane cubic C ∈ P(H0 (H, OH (3))) is given by 9
parameters, and the point p ∈ P3 ∖ C is given by 3 parameters. It follows that the
locus H0′ is smooth and irreducible of dimension 15. The locus H0′ is not proper and
its closure contains the degenerations where the point p lies on the curve.
The intersection H ∩ H ′ consists of plane singular cubic curves with an embedded
point at the singular point. This locus contains curves such as C0 above but it also
contains even more degenerate curves such as a triple line with an embedded point.
Every curve C ∈ H ∩ H ′ is in fact projectively equivalent to the curve defined by
V (xz, yz, z 2 , q(x, y, w)) where q(x, y, w) is a homogeneous cubic polynomial with a
singular point at (0, 0, 1). This depends on 11 parameters.

67
1.5.2 Geometric properties
Exercise 1.5.6 (Local properties). Let X be a projective scheme over a field k and
F be a coherent sheaf on X.
(a) Let p ∈ QuotP (F/X) be the point corresponding to a quotient Q = F/K.
Show that Tp QuotP (F/X) ∼= HomOX (K, Q). This generalizes Exercise 1.2.10
computing the tangent space of the Grassmannian.
(b) Conclude that if p ∈ HilbP (X) is a point corresponding to a closed subscheme
Z ⊆ X defined by a sheaf of ideals I, then Tp HilbP (X) ∼ = H0 (Z, NZ/X )
where NZ/X is the normal sheaf HomOZ (IZ /IZ , OZ ). (This recovers Proposi-
2

tion C.1.3.)

Non-emptiness. The Hilbert scheme HilbP (Pn ) is non-empty if and only if the
Hilbert polynomial P can be written as as
     
z + λ1 − 1 z + λ2 − 2 z + λr − r
P (z) = + + ··· + , (1.5.7)
λ1 − 1 λ2 − 1 λr − 1

integers λ1 ≥ λ2 ≥ · · · ≥ λr ≥ 1. This is a result of Hartshorne [Har66b, Cor. 5.7].

Connectedness. Hartshorne’s Connectedness Theorem asserts that the Hilbert
scheme HilbP (Pn ) is connected for every Hilbert polynomial P [Har66b]. The proof
strategy is to exhibit a degeneration from any closed subscheme Z ⊆ Pn to a
subscheme V (I) ⊆ Pn defined by a monomial ideal. This reduces the theorem to
the combinatorial question of connecting any two monomial ideals by a family over
A1 . This turns out to be a purely deformation and combinatorial question, or as
Hartshorne writes: “It also appears that the Hilbert scheme is never actually needed
in the proof.”
Murphy’s Law. The first pathology was exhibited by Mumford: there is an
irreducible component of Hilb14z−23 (P3 ) which is generically non-reduced [Mum62].
Ellia—Hirschowitz—Mezzetti show that the number of irreducible components in
Hilbaz+b (P3 ) is not bounded by a polynomial in a, b [EHM92]. Murphy’s Law was
made precise by Vakil [Vak06]: for every scheme X finite type over Z and point
x ∈ X, there exists a point [Z ⊆ Pn ] ∈ HilbP (Pn ) of some Hilbert scheme such
that (X, x) and (HilbP (Pn ), [Z ⊆ Pn ]) are smooth locally isomorphic, i.e., their
complete local rings become isomorphic after appending power series rings. In fact,
one can take [Z ⊆ Pn ] to be a smooth curve! Many other moduli spaces satisfy
Murphy’s Law: Kontsevich’s moduli space of maps, moduli of canonically polarized
smooth surfaces, moduli of curves with linear systems, and the moduli space of
stable sheaves.
Smoothness. Despite Murphy’s Law, Hilbert schemes are surprisingly often smooth.
We have seen before that the Hilbert scheme of points on a smooth surface is smooth.
A recent theorem of Skjelnes–Smith [SS20] gives necessary and sufficient conditions for
the Hilbert scheme HilbP (Pn ) to be smooth in terms of the partition λ = (λ1 , . . . , λr )
defining P as in (1.5.7).

68
Chapter 2

Sites, sheaves, and stacks

If there is one thing in mathematics that fascinates me more than anything

else (and doubtless always has), it is neither ‘number’ nor ‘size,’ but
always form.

Alexander Grothendieck

This chapter introduces the core categorical constructions—sites, sheaves, and

stacks—necessary to define algebraic spaces and stacks. Grothendieck introduced
stacks in [FGAI, §A.1] and [SGA1, §6] as a way to package objects, e.g., quasi-
coherent sheaves, satisfying descent.

2.1 Descent theory

It is hard to overstate the importance of descent in moduli theory. The central idea
is strikingly simple, and a special case is already familiar to you: quasi-coherent
sheaves, morphisms of schemes, and schemes themselves can be constructed locally
on a Zariski cover, and moreover most of their properties can be checked Zariski
locally. Indeed, a central technique in the development of scheme theory is to reduce
to the case of affine schemes, and then apply results from commutative algebra.
Descent theory implies that each of these objects (quasi-coherent sheaves, morphisms
of schemes, and schemes) can be constructed not only Zariski locally but étale locally
(and even fppf or fpqc locally), and moreover that their properties can be verified
locally. This allows us to prove statements about algebraic stacks by reducing to
the case of schemes (or even affine schemes).
Descent theory was originally developed by Grothendieck in [FGAI], [SGA1, §6],
and [EGA, §IV] as a grand generalization of the theory of Galois descent initiated
by Weil [Wei56]. Applications of descent theory extend far beyond moduli theory.
For instance, since field extensions are faithfully flat, one can reduce properties of a
scheme over a field k to the case of an algebraically closed field. Similarly, since the
map A → A b of a noetherian local ring to its completion is faithfully flat, properties
of A can be reduced to properties of its completion. As there are already wonderful
expositions on descent theory such as [BLR90, §6], [Vis05], and [SP, Tag 0238], our
treatment will sometimes be short on details.

69
2.1.1 Descending quasi-coherent sheaves
The following key algebraic fact is the basis for descent for quasi-coherent sheaves.

Proposition 2.1.1. If ϕ : A → B is a faithfully flat ring map, then

/A ϕ
/B
b7→b⊗1
/
0 / B ⊗A B
b7→1⊗b

is exact (i.e., an equalizer sequence). More generally, if M is an A-module,

/M m7→m⊗1
/ M ⊗A B
m⊗b7→m⊗b⊗1
/
0 / M ⊗A B ⊗A B (2.1.2)
m⊗b7→m⊗1⊗b

is exact.

Proof. Note that A → B and M → M ⊗A B are necessarily injective by Faithfully

Flat Equivalences (A.2.19). Since A → B is faithfully flat, the sequence (2.1.2) is
exact if and only if

m⊗b⊗b′ 7→m⊗b⊗1⊗b′
′ ′
/ M ⊗A B m⊗b 7→m⊗1⊗b / M ⊗A B ⊗A B /
0 / M ⊗A B ⊗A B ⊗A B
m⊗b⊗b′ 7→m⊗1⊗b⊗b′

is exact. The above sequence can be rewritten as

x⊗y7→x⊗y⊗1
/ M ⊗A B x7→x⊗1/ (M ⊗A B) ⊗B (B ⊗A B) /
0 / (M ⊗A B) ⊗B (B ⊗A B) ⊗B (B ⊗A B)
x⊗y7→x⊗1⊗y

which is precisely sequence (2.1.2) applied to the ring map B → B ⊗A B, defined by

b 7→ 1 ⊗ b, and the B-module M ⊗A B. Since B → B ⊗A B has a left inverse given
by b ⊗ b′ 7→ bb′ , we are reduced to proving the proposition when ϕ : A → B has a
left inverse s : B → A with s ◦ ϕ = idA . Let x = i mi ⊗ bi ∈ M ⊗A B such that
P

X X
mi ⊗ bi ⊗ 1 = mi ⊗ 1 ⊗ bi ∈ M ⊗A B ⊗A B.
i i

Applying idM ⊗ idB ⊗s : P

M ⊗A B ⊗A B → P M ⊗A B ⊗A A ∼ = M ⊗A B to this
identity shows that x = i mi ⊗ ϕ(s(bi )) = i ϕ(s(bi ))mi ⊗ 1 is in the image of
M → M ⊗A B.

Exercise 2.1.3. Denoting (B/A)⊗n as the n-fold tensor product B ⊗A · · · ⊗A B,

show that the exact sequence (2.1.2) extends to a long exact sequence

d d d
→ M ⊗A (B/A)⊗1 −
0→M − → M ⊗A (B/A)⊗2 −
→ ···

with differentials

d : M ⊗A (B/A)⊗n → M ⊗A (B/A)⊗(n+1)
n+1
X
m ⊗ b1 ⊗ · · · ⊗ bn 7→ (−1)i m ⊗ b1 ⊗ · · · ⊗ bi−1 ⊗ 1 ⊗ bi+1 ⊗ · · · ⊗ bn .
i=0

70
 Recall from Definition A.2.20 that a morphism of schemes f : S ′ → S is fpqc
if f is faithfully flat and every quasi-compact open subset of S is the image of
quasi-compact open subset of S ′ . Fpqc morphisms include faithfully flat and quasi-
compact morphisms, but the definition is broader as it includes for instance fppf
morphisms (i.e. faithfully flat morphisms locally of finite presentation).
It is always instructive to keep in the mind the special case of an étale morphism
f : S ′ = ⨿i Ui → S induced from a Zariski cover {Ui } of S. In this case, the
fiber product S ′ ×S S ′ is ⨿i,j Ui ∩ Uj and the higher fiber products afford similar
descriptions.
Proposition 2.1.4 (Fpqc Descent for Quasi-Coherent Sheaves). Let f : S ′ → S be
an fpqc morphism of schemes.
(1) Let F and G be quasi-coherent OS -modules. Let p1 , p2 : S ′ ×S S ′ ⇒ S ′ be
the two projections and q : S ′ ×S S ′ → S be the composition f ◦ pi . Then the
sequence

p∗
/ HomO (F, G) f∗
/ HomO ′ (f ∗ F, f ∗ G)
1
/ ∗ ∗
0 S S / HomOS′ ×S S′ (q F, q G)
p∗
2

is exact.
(2) Let H be a quasi-coherent OS ′ -module and α : p∗1 H → p∗2 H be an isomorphism
of OS ′ ×S S ′ -modules satisfying the cocycle condition p∗23 α ◦ p∗12 α = p∗13 α on
S ′ ×S S ′ ×S S ′ . Then there exists a quasi-coherent OS -module G and an
isomorphism ϕ : H → f ∗ G such that p∗1 ϕ = p∗2 ϕ ◦ α on S ′ ×S S ′ . The data
(G, ϕ) is unique up to unique isomorphism.
The following diagram may help to internalize (2):

α
p∗23 α ◦ p∗12 α = p∗13 α p∗1 H −
→ p∗2 H H G

p12 / /
S ′ ×S S ′ ×S S ′ p13 / S ′ ×S S ′ p1
/S
′ f
/ S.
p23 / p2

The cocycle condition p∗23 α ◦ p∗12 α = p∗13 α and the equality p∗1 ϕ = p∗2 ϕ ◦ α should
be understood as the commutativity of

p∗
12 α
p∗12 p∗1 H / p∗12 p∗2 H p∗
1ϕ
/ p∗1 f ∗ G
p∗1 H
p∗13 p∗1 H p∗23 p∗1 H and α
p∗
13 α p∗
23 α
) u  p∗
2ϕ
p∗2 H / p∗2 f ∗ G.
p∗13 p∗2 H p∗23 p∗2 H

Proposition 2.1.4 can be reformulated as an equivalence of categories

∼
QCoh(S) → QCoh(S ′ → S), G 7→ (f ∗ G, can), (2.1.5)

where QCoh(S ′ → S) is the category of descent data for S ′ → S, whose objects

are pairs (H, α) consisting of a quasi-coherent OS ′ -module H and an isomorphism
α : p∗1 H → p∗2 H satisfying the cocycle condition, and a morphism (H ′ , α′ ) → (H, α)
is a morphism β : H ′ → H such that α ◦ p∗1 β = p∗2 β ◦ α′ . Note that if H = f ∗ G
∼
for G ∈ QCoh(S), then there a canonical isomorphism can : p∗1 H → p∗2 H since the

71
compositions f ◦ p1 and f ◦ p2 are equal. In yet other language, this result asserts
that every descent data (H, α) for S ′ → S is effective, i.e., in the essential image of
(2.1.5). Finally, we will see shortly that this is equivalent to the statement that the
prestack QCoh parameterizing quasi-coherent sheaves is a stack in the fpqc topology
(Example 2.5.9).

Proof. If S ′ = Spec A′ and S = Spec A are affine, write F = M f and G = N e.

Proposition 2.1.1 implies that 0 → N → N ⊗A A ⇒ N ⊗A A ⊗A A is exact.
′ ′ ′

Part (1) follows from applying HomA (M, −) and using tensor-hom adjunction, e.g.,
HomA (M, N ⊗A A′ ) = HomA′ (M ⊗A A′ , N ⊗A A′ ). For (2), writing H = M f′ , we
define the A-module M as the equalizer

/M / M′
m7→m⊗1
/ ′ ′
0 / M ⊗A A
m7→α(m⊗1)

Tensoring this sequence of A-modules with A′ expresses M ⊗A A′ as the equalizer

of M ′ ⊗A A′ ⇒ M ′ ⊗A A′ ⊗A A′ . On the other hand, the key algebra result of
Proposition 2.1.1 shows that M ′ is identified with the same equalizer. This gives an
isomorphism ϕ : M ′ → M ⊗A A′ of A′ -modules and one checks that p∗1 ϕ = p∗2 ϕ ◦ α.
The general case is Zariski local on S so we may assume that S is affine. Since f
is fpqc, S is the image of a quasi-compact open subset U ′ ⊂ S ′ . By choosing a finite
affine cover {Ui′ } of U ′ , we can reduce to the case of a faithfully flat map ⨿i Ui′ → S
of affine cases (details left to the reader), where we have already verified the result.
It is, in fact, a general result (see Exercises 2.3.7 and 2.5.5) that to verify properties
(1)–(2) for an fpqc morphism S ′ → S, it suffices to verify them for maps ⨿i Ui → S
induced by a Zariski cover {Ui } of S (which we already know) and for a faithfully
flat map S ′ → S of affine schemes (which we just verified). See also [FGAI, Thm. 1,
p. 315], [BLR90, Thm. 6.4], [Vis05, Thm. 4.23], and [SP, Tag 023T].

Remark 2.1.6. It turns out that descent for modules holds for a class of ring maps
A → B larger than just faithfully flat maps. It holds for universally injective maps
(see Definition A.2.22), and remarkably the converse is true! More precisely, A → B
is universally injective if and only if the functor to the category of descent data
 ∼ 
N ∈ ModB , α : N ⊗B,p1 (B ⊗A B) → N ⊗B,p2 (B ⊗A B)
ModA → (N, α) .
satisfying the cocycle condition p∗23 α ◦ p∗12 α = p∗13 α
M 7→ (M ⊗A B, can)

is an equivalence of categories. See [Mes00] or [SP, Tag 08XA].

2.1.2 Descending morphisms

Proposition 2.1.7 (Fpqc Descent for Morphisms). Let Y be a scheme and f : S ′ → S
be an fpqc morphism of schemes. If g : S ′ → Y is a morphism such that g ◦p1 = g ◦p2 ,
then there exists a unique morphism h : S → Y filling in the commutative diagram
p1
/ f
/S
S ′ ×S S ′ /S
′
p2
h
g

Y.

72
 In other words, an fpqc morphism f : S ′ → S is an effective epimorphism in the
category of schemes, i.e., for every scheme Y , the sequence

Mor(S, Y ) → Mor(S ′ , Y ) ⇒ Mor(S ′ ×S S ′ , Y ), (2.1.8)

is exact; being only an epimorphism translates to the injectivity of the first map.
As we will see shortly, this also translates into the functor Mor(−, Y ) being a sheaf
in the fpqc topology (see Proposition 2.3.8).

Proof. The affine case is straightforward. Writing S ′ = Spec A′ , S = Spec A, and

Y = Spec R, then Proposition 2.1.1 yields that A → A′ ⇒ A′ ⊗A A′ is exact, and
applying Hom(R, −) shows that Hom(R, A) → Hom(R, A′ ) ⇒ Hom(R, A′ ⊗A A′ ) is
also exact, which translates to the exactness of (2.1.8) under the duality between
affine schemes and rings. To reduce to the affine case, observe that question is local
on S so we may assume that S is affine. As S ′ → S is fpqc, there is a quasi-compact
open subset U ′ ⊆ S ′ surjecting onto S. After choosing a finite affine covering {Ui′ }
of U ′ , we can replace S ′ with the affine scheme ⨿i Ui′ (details left to the reader).
Reducing to the case that Y is affine is a little harder. We first observe that

|S ′ ×S S ′ | ⇒ |S ′ | → |S| (2.1.9)

is a coequalizer diagram of sets. Indeed, we already know that |S ′ | → |S| is

surjective, and that if s′1 , s′2 ∈ |S ′ | have the same image in S, then there exists
a point q ∈ |S ′ ×S S ′ | with p1 (q) = s′1 and p2 (q) = s′2 . Thus there exists a map
|h| : |S| → |Y | of sets such that |g| = |h| ◦ |f |. Letting U ⊆ Y be an open affine
subset, then V := |h|−1 (U ) is a subset of S such that f −1 (V ) = g −1 (U ) is an open
subset of S ′ . Since S ′ → S is submersive (i.e., S has the quotient topology) by
Exercise A.4.9, it follows that V ⊆ S is open. Since f −1 (V ) → V is also fpqc, we
may assume that Y is affine. See also [FGAI, Thm. 2, p. 317], [BLR90, Thm. 6(a)],
[Vis05, Thm. 2.55], and [SP, Tag 023Q].

The following generalization also holds.

Corollary 2.1.10. Let f : S ′ → S be an fpqc morphism of schemes.

(1) If S → T is a morphism of schemes and Y is an T -scheme, then

MorT (S, Y ) → MorT (S ′ , Y ) ⇒ MorT (S ′ ×S S ′ , Y ),

is exact.
(2) If X and Y are schemes over S, then

MorS (X, Y ) → MorS ′ (XS ′ , YS ′ ) ⇒ MorS ′′ (XS ′′ , YS ′′ ),

is exact where S ′′ = S ′ ×S S ′ .

Proof. For (1), it follows from Proposition 2.1.7 that MorT (S, Y ) → MorT (S ′ , Y ) is
injective and that if g : S ′ → Y is an T -morphism such that p1 ◦ g = p2 ◦ g, there
exists a map h : S → Y of schemes with g = h ◦ f . Letting pS : S → T , pS ′ : S ′ → T ,
and pY : Y → T denote the structure morphisms, observe that pY ◦ h and pS are
elements of Mor(S, T ) mapping to pY ◦ g = pS ′ ∈ Mor(S ′ , T ). By Proposition 2.1.7,
the inclusion Mor(S, T ) → Mor(S ′ , T ) is injective and we conclude that h : S → Y is
a T -morphism. Part (2) follows from applying (1) to the fpqc morphism f : XS ′ → X
over T = S.

73
2.1.3 Descending schemes
Proposition 2.1.11 (Fpqc Descent for Open/Closed Subschemes). Let f : S ′ → S
be an fpqc morphism of schemes. If Z ′ ⊆ S ′ is a closed (resp., open) subscheme such
that p−1 ′ −1 ′ ′ ′
1 (Z ) = p2 (Z ) as subschemes of S ×S S , then there exists a closed (resp.,
′ −1
open) subscheme Z ⊆ S such that Z = f (Z).

Proof. If Z ′ ,→ S ′ is a closed immersion defined by an ideal sheaf IZ ′ ⊆ OS ′ , then

Fpqc Descent for Quasi-Coherent Sheaves (2.1.4) implies that IZ ′ descends to a
quasi-coherent sheaf IZ on S and the inclusion IZ ′ ,→ OS ′ descends to an inclusion
IZ ,→ OS . It follows that Z ′ descends to closed subscheme Z ⊆ S defined by IZ .
The case of an open immersion is handled by passing to the reduced complement.

For the following results, it will be convenient to denote f ∗ X as the base change
of X → S by a morphism f : S ′ → S.

Proposition 2.1.12 (Fpqc Descent for Affine/Quasi-affine Schemes). Let f : S ′ → S

be an fpqc morphism of schemes. If X ′ → S ′ is an affine (resp., quasi-affine)
∼
morphism and α : p1 ∗ (X ′ ) → p∗2 (X ′ ) is an isomorphism over S ′ ×S S ′ satisfying
∗ ∗ ∗
p23 α ◦ p12 α = p13 α, then there exists an affine (resp., quasi-affine) morphism X → S
of schemes and an isomorphism ϕ : X ′ → f ∗ (X) over S ′ such that p∗1 ϕ = p∗2 ϕ ◦ α.

In other words, there exists dotted arrows completing the diagram

p∗23 α ◦ p∗12 α = p∗13 α p∗1 X ′ −

α
→ p∗2 X ′ X′ /X

p12 /  / ′ 
S ′ ×S S ′ ×S S ′ p13 / S ′ ×S S ′ p1
/S
f
/ S.
p23 / p2

Proof. If X ′ → S ′ is affine, we can write X ′ = SpecX A′ for a quasi-coherent sheaf

A′ of OS ′ -algebras. Fpqc Descent for Quasi-Coherent Sheaves (2.1.4) allows us
to first descend A′ to a quasi-coherent sheaf A on X, and then to descend the
multiplication map A′ ⊗OS′ A′ → A′ to a map A ⊗OS A → A, which by descent
will necessarily satisfy the axioms making A into a quasi-coherent OS -algebra.
It follows that X ′ → S ′ descends to the affine morphism X := SpecS A → S.
The case of quasi-affine morphisms is handled by using the canonical factorization
X ′ ,→ SpecS ′ g∗′ OX ′ → S ′ into an open immersion followed by an affine morphism,
and then combining the affine case above with Fpqc Descent for Open Subschemes
(2.1.11). See also [BLR90, Thm. 6.6] and [Vis05, Thm. 4.33].

Remark 2.1.13 (Cocycle condition). The cocycle condition is necessary for a scheme
to descend. Shimura showed that for a genus 2 curve C ′ over C defined by y 2 =
x6 + ax5 + bx4 + x3 − bx2 + ax − 1 for a general a, b ∈ C is isomorphic to its complex
conjugate but does not descend to a real curve [Shi72]. In other words, under the
∼
cover S ′ = Spec C → Spec R, there is an isomorphism α : p∗1 C ′ → p∗2 C ′ but C ′ → S ′
does not descend to a curve C → S.

Theorem 2.1.14 (Fppf Descent for Locally Quasi-finite and Separated Schemes).
Let f : S ′ → S be an fppf morphism of schemes. If X ′ → S ′ is a locally quasi-finite
∼
and separated morphism of schemes and α : p∗1 (X ′ ) → p∗2 (X ′ ) is an isomorphism
′ ′ ∗ ∗ ∗
over S ×S S satisfying p23 α ◦ p12 α = p13 α, then there exists a locally quasi-finite
and separated morphism X → S of schemes and an isomorphism ϕ : X ′ → f ∗ (X)
over S ′ such that p∗1 ϕ = p∗2 ϕ ◦ α.

74
 This will imply that the prestack SepLQFin over Sch parameterizing locally
quasi-finite and separated morphisms X → S is a stack in the fppf topology.

Proof. Our strategy is to reduce to Fpqc Descent for Quasi-affine Morphisms (2.1.12).
For each quasi-compact open subset U ′ ⊆ X ′ , the composition U ′ ,→ X ′ → S
is quasi-affine by Zariski’s Main Theorem (A.7.3). Since S ′ → S is fppf, so is
p2 : S ′ ×S S ′ → S ′ , and hence the image V ′ := p2 (α(p∗1 U ′ )) is an open subset of X ′ .
The image V ′ is also quasi-compact and contains U ′ , and moreover α restricts to an
∼
isomorphism p∗1 (V ′ ) → p∗2 (V ′ ) satisfying the cocycle condition. As the map V ′ → S ′
is quasi-affine, Fpqc Descent for Quasi-affine Morphisms implies that V ′ → S ′
descends to a quasi-affine morphism V → S. Covering X ′ with quasi-compact open
subsets Ui′ , the subsets Vi′ := p2 (α(p∗1 Ui′ )) also cover X ′ , and since each Vi′ descends
to a scheme Vi quasi-affine over S, we may glue the schemes Vi to a scheme X over
S which pulls back to X ′ . That X → S is locally quasi-finite and separated follows
from a straightforward descent argument; see Proposition 2.1.26. See also [GR71,
Lem. 5.7.2] and [SP, Tag 02W7].

Remark 2.1.15 (Historical comment). If X ′ → S ′ is quasi-compact, then it is

quasi-affine in which case the argument for effective descent is straightforward (see
Proposition 2.1.12). In a 1963 letter to Mumford [Mum10, p. 681], Grothendieck
indicates that he could prove effective descent for locally quasi-finite and separated
morphisms. This result appeared in unpublished work of Murre—later published in
[Mur95, Prop. 1]—under a locally noetherian hypothesis and in [SGA3II , Lem. X.5.4]
under a locally of finite presentation hypothesis. In full generality, this theorem was
proven by Raynaud and Gruson, but unfortunately was disguised in the proof of
[GR71, Lem. 5.7.2]. That this theorem was not widely known at the time may have
been a contributing reason to why algebraic spaces and stacks were first defined in
[Knu71] and [Art74b] with a quasi-compact hypothesis on the diagonal.
We will often apply the above result in the form of Proposition 2.3.17 to show
that a given sheaf in the big étale or fppf topology is representable by a scheme.

Example 2.1.16 (Non-effective descent). An arbitrary morphism X ′ → S ′ of

schemes with descent data along an fppf (or even étale) morphism S ′ → S may
not descend to a morphism X → S of schemes. Hironaka’s example of a proper
three-fold that is not projective can be adapted to give an example of a projective
morphism X ′ → S ′ that does not descend; see [SP, Tag 08KE]. Additionally, Raynaud
constructed a normal noetherian local ring A of dimension 2, an étale cover S ′ →
S = Spec A, and a family C ′ → S ′ of smooth genus 1 curves that does not descend
to a family C → S [Ray70, XIII 3.2]. There is also an example of a DVR R, an
étale cover S ′ → S = Spec R, and a family C ′ → S ′ of nodal genus 1 curves with
smooth generic fiber that does not descend to a family C → S [BLR90, § 6.7], and
an example of a projective surface S, an étale cover S ′ → S, and a family C ′ → S ′
of nodal genus 0 curves with smooth generic fiber that does not descend to a family
C → S [Ful10, Ex. 2.3].
On the other hand, an map X ′ → S ′ of schemes with descent data along an
fppf cover S ′ → S always descends to a morphism X → S of algebraic spaces (see
Theorem 3.4.13 for the étale case, and Corollary 6.3.6 in general). In fact, the
prestack AlgSp, whose objects over a scheme S are morphisms X → S of algebraic
spaces, is a stack in the fppf topology (see Exercise 4.5.15).
Effective descent does, however, hold in some other settings. For instance, we
will see in the next result that it holds for principal G-bundles. It also holds for
pairs (X ′ → S ′ , L′ ), where X ′ → S ′ is a quasi-compact morphism of schemes and L′

75
is a line bundle on X ′ relatively ample over S ′ ; see [BLR90, Thm. 6.7] and [Vis05,
Thm. 4.38].

Proposition 2.1.17 (Fpqc Descent for Principal G-bundles). Let G → T be an

fppf affine group scheme, and let f : S ′ → S be an fpqc morphism of schemes over
∼
T . If P ′ → S ′ is a principal G-bundle and α : p∗1 P ′ → p∗2 P ′ is an isomorphism of
principal G-bundles over S ′ ×S S ′ satisfying p∗12 α ◦ p∗23 α = p∗13 α, then there exists a
principal G-bundle P → S and an isomorphism ϕ : P ′ → f ∗ P of principal G-bundles
such that p∗1 ϕ = p∗2 ϕ ◦ α.

Proof. Since G → T is affine, the principal G-bundle P ′ → S ′ is affine. By Fpqc

Descent for Affine Schemes (Proposition 2.1.12), there is an affine morphism P → S
and an isomorphism ϕ : P → f ∗ Q of schemes with p∗1 ϕ = p∗2 ϕ ◦ α. By Fpqc
Descent for Morphisms (2.1.7), we can descend the action G ×S P ′ → P ′ to an
action G ×S P → P giving P the structure of a principal G-bundle and making
ϕ : P ′ → f ∗ P a G-equivariant isomorphism.

2.1.4 Descending properties

Proposition 2.1.18 (Fpqc Local Properties of Quasi-Coherent Sheaves). Let
f : S ′ → S be an fpqc morphism of schemes.
(1) A homomorphism F → G of quasi-coherent OS -module is an isomorphism
(resp., injective, surjective) if and only if f ∗ F → f ∗ G is.
(2) A quasi-coherent OS -module F is of finite type (resp., of finite presentation,
flat, a vector bundle, a line bundle) if and only if f ∗ G is. If S and S ′ are
noetherian, then the same holds for coherence.
(3) A quasi-coherent OX -module F on an S-scheme X is flat over S if and only
if the pullback of F to X ×S S ′ is flat over S ′ .

In other words, each of these properties is fpqc local on S.

Proof. Part (1) reduces to the algebra statement: if A → A′ is a faithfully flat ring
map, an A-module map M → N is an isomorphism (resp., injection, surjection) if
and only if M ⊗A A′ → N ⊗A A′ is. This follows directly from the faithful exactness
of −⊗A A′ . Part (2) reduces to: if A → A′ is faithfully flat, an A-module M is finitely
generated (resp., finitely presented, flat, locally free of rank r) if and only if M ⊗A A′
is. The (⇒) implications are clear. Conversely, if M ⊗A AP ′
is finitely generated, then
let x1 , . . . , xm ∈ M ⊗A A be generators and write xi = j xij ⊗ a′ij with xij ∈ M
′ ′ ′ ′

and a′ij ∈ A′ . Letting n be the number of the xij , the map (xij ) : A⊕n → M is
surjective since it becomes surjective after base changing by the faithfully flat map
A → A′ . Repeating this argument to the kernel, we see that the property of being
finite presentation descends. For flatness, suppose that M ⊗A A′ is flat. By the
faithful flatness of A → A′ , the exactness of M ⊗A − is equivalent to the exactness
of (M ⊗A A′ ) ⊗A′ (A′ ⊗A −), which follows from the flatness of A → A′ and the
flatness of the A′ -module M ⊗A A′ . As being locally free of finite rank is equivalent
to being finitely presented and flat, the final statement also follows. Part (3) reduces
to: if A → A′ is faithfully flat and A → B is a ring map, a B-module N is flat over
A if and only if N ⊗A A′ is flat over A′ . This is special case of (2). See also [EGA,
IV2 .2.5] and [SP, Tag 05AY].

The following, perhaps surprising, fact that regularity descends under faithful
flatness will come in handy.

76
Lemma 2.1.19. If A → B is a flat local ring map of noetherian local rings and B
is regular, then so is A.

Proof. Recall that a noetherian local ring R is regular if and only if every finitely gen-
erated R-module M has a finite resolution by free modules of finite rank. Moreover,
if R is regular of dimension d and

0 → K → R⊕kd−1 → · · · → R⊕k0 → M → 0 (2.1.20)

is an exact sequence of R-modules, then K is free; see [Eis95, Thm. 19.12] and [SP,
Tag 00OC]. If M is a finitely generated A-module, choose an exact sequence (2.1.20).
Since B is regular, K ⊗A B is free. Since being locally free is an Fpqc Local Property
of Quasi-Coherent Sheaves (2.1.18(2)) and R is local, K is free. Therefore A is
regular. See also [EGA, IV0 .17.3.3] and [SP, Tag 00OF].

Proposition 2.1.21 (Fpqc Descent for Properties of Schemes). Let X → Y be

an fpqc morphism of schemes. If X is quasi-compact (resp., locally noetherian,
noetherian, integral, reduced, normal, regular), then so is Y .

Proof. First, note that quasi-compactness descends under any surjective map. The
other parts reduce to the algebraic statement: if A → B is a faithfully flat map of
rings and B is noetherian (resp., a domain, reduced, normal, or regular), then so
is A. As A → B is faithfully flat, A → B is injective and I = IB ∩ A for every
ideal I ⊆ A. By the injectivity of A → B, the ‘domain’ and ‘reduced’ cases are
clear. For noetherianness, if I1 ⊆ I2 ⊆ · · · is an ascending chain of ideals, then since
I1 B ⊆ I2 B ⊆ · · · terminates, so does I1 = I1 B ∩ A ⊆ I2 = I2 B ∩ A ⊆ · · · . If B is
a normal domain and a/b is integral over A where a, b ∈ A, then a/b ∈ B. This
implies that a ∈ (b) ⊆ B, hence a : B → B/bB is the zero map. As this map is
the base change of a : A → A/bA, faithful flatness implies that a : A → A/bA is the
zero map, hence a ∈ (b) ⊆ A and a/b ∈ A. The regularity statement follows from
Lemma 2.1.19. See also [SP, Tags 033D, 034B, and 06QL].

Proposition 2.1.22 (Fpqc Descent for Properties on the Source). Let X ′ → X

be an fpqc morphism of schemes. If X → Y is a morphism of schemes such that
X ′ → X → Y is smooth (resp., étale), then X → Y is smooth (resp., étale).

Proof. The smooth case follows from Lemma 2.1.19. The étale case follows from the
smooth case with the observation that for y ∈ Y , the map Xy′ → Xy on fibers is
surjective, and hence if dim Xy′ = 0, then dim Xy = 0. See also [EGA, IV4 .17.7.7]
and [SP, Tag 05B5].

Note that smoothness and étaleness are, however, not fpqc (or fppf) local
properties, i.e., properties that hold if and only if they hold after an fpqc (or fppf)
cover. For instance, there are non-smooth (but necessarily flat) schemes of finite
type over a field.

Proposition 2.1.23 (Fppf/Smooth Local Properties of Schemes).

(1) If X → Y is an fppf morphism of schemes, then X is locally noetherian if and
only if Y is.
(2) If X → Y be a surjective smooth morphism of schemes, then X is reduced
(resp., normal, regular) if and only if Y is.

77
Proof. The (⇒) implications follows from Proposition 2.1.21. For (1), if Y is locally
noetherian, so is X by Hilbert’s Basis Theorem. Part (2) reduces to the algebra
statement that if A → B is a smooth ring map and A is reduced (resp., normal,
regular), then so is B, which we leave to the reader. See also [SP, Tag 034D].

Remark 2.1.24. The property of being a domain is not étale local, e.g., there is a
reducible étale neighborhood of the nodal cubic (see Example 0.5.3). Reducedness,
normality, and regularity are not fppf local: there are non-reduced schemes of finite
type over a field. On the other hand, if X → Y is a flat morphism of noetherian
schemes such that Y is normal and every fiber Xy is normal, then X is normal; see
[EGA, IV2 .6.5.4] or [SP, Tag 0C22].
Remark 2.1.25. If A is a noetherian local ring, the map A → A b to its completion
is faithfully flat. If the completion A is reduced (resp., normal, regular), Fpqc
b
Descent for Properties of Schemes (2.1.21) implies that the same holds for A. While
the converse holds for regularity, it does not hold in general for reducedness and
normality. However, if A is essentially of finite type over a field (or more generally
excellent), then A is reduced (resp., normal) if and only if A
b is, and moreover in this
case the normalization commutes with completion. See [SP, Tags 07NZ and 0C23].

Proposition 2.1.26 (Fpqc Local Properties on the Target). Let S ′ → S be an fpqc

morphism of schemes and P be one of the following properties of a morphism of
schemes: surjective, quasi-compact, quasi-separated, isomorphism, open immersion,
closed immersion, monomorphism, affine, quasi-affine, quasi-compact locally closed
immersion, locally of finite type, locally of finite presentation, separated, proper,
universally closed, universally open, universally submersive, finite, locally quasi-finite,
quasi-finite, flat, fppf, smooth, étale, unramified, or syntomic. Then X → S has P
if and only if X ×S S ′ → S ′ does.

In other words, each of these properties is fpqc local on the target.

Proof. To setup the notation, consider the cartesian diagram

p1 / g′
/X
X ′ ×X X ′ p2 /X
′

□ f′ □ f
 / ′ 
S ′ ×S S ′
p1
/S
g
/ S.
p2

We already know that each property is stable under base change. It is not hard to
see that the properties ‘surjective’, ‘quasi-compact’, and ‘quasi-separated’ descend.
For ‘isomorphism’, let h′ : S ′ → X ′ be an inverse of f ′ . As inverses are unique,
p1 h = p∗2 h′ as morphisms S ′ ×S S ′ → X ′ ×X X ′ . By Fpqc Descent for Morphisms
∗ ′

(2.1.10(2)) applied to X ′ → X, the S ′ -morphism h′ descends to a S-morphism

h : S → X. This yields that f ◦ h = idY . To see that h ◦ f = idX , observe that the
two morphisms idX , h ◦ f : X → X become equal after precomposing with X ′ → X,
and thus Fpqc Descent for Morphisms (2.1.7) implies that h ◦ f = idX . For ‘open
immersion’, if f ′ is an open immersion, then f ′ (X ′ ) = g −1 (f (X)) is open. As g is
universally submersive (Exercise A.4.9), f (X) is open and we have reduced to show
that X → f (X) is an isomorphism, which follows from the previous case. The case
of ‘closed immersion’
and ‘quasi-compact locally closed immersion’ follow For ‘affine’ and ‘quasi-affine’,
we can assume that f is quasi-compact and quasi-separated. We appeal to the
canonical factorization f : X → SpecY f∗ OX → S which commutes with flat base

78
change by S ′ → S, and use that f is affine (resp., quasi-affine) if and only if
X → SpecY f∗ OX is an isomorphism (resp., open immersion).
The ‘locally of finite type’ (resp., ‘locally of finite presentation’) case reduces to:
if A → A′ is faithfully flat, then a ring map A → B is of finite type (resp., finite
presentation) if and only if A′ → A′ ⊗B B ′ is. If A′ → A′ ⊗A B is ofP finite type,
there are A′ -algebra generators b′1 , . . . , b′n which we can write as b′i = j a′ij ⊗ bij
with a′ij ∈ A′ and bij ∈ B. If B e ⊆ B denotes the A-subalgebra generated by the
bij , then since B ⊗A A = B ⊗A A′ , the faithful flatness of A → A′ implies that
e ′

Be = B. If A′ → A′ ⊗A B is of finite presentation, then we have just seen that

A → B is of finite type and we can write B = A[x1 , . . . , xn ]/I. Since A → A′ is
flat, B ⊗A A′ = A′ [x1 , . . . , xn ]/I ′ , where I ′ = I ⊗A[x1 ,...,xn ] A′ [x1 , . . . , xn ]. Since I ′
is a finitely generated ideal, Fpqc Descent for Properties of Quasi-Coherent Sheaves
(2.1.18(2)) implies that I is also finitely generated.
The ‘flat’ case is easy and follows from Proposition 2.1.18(2). Since the property
‘smooth’ is equivalent to flat, locally of finite presentation, and smoothness of
every fiber, this case reduces to the algebra fact: a finite type algebra A over an
algebraically closed field K is regular if and only if A ⊗K L is regular for every
algebraically closed field extension L/K. The remaining cases are left to the reader.
See also [EGA, IV2 .2.6–7 and IV4 .17.7.4] and [SP, Tag 02YJ].

Proposition 2.1.27 (Fppf/Smooth/Étale Local Properties on the Source).

(1) If X ′ → X is an fppf morphism of schemes, a morphism X → Y of schemes is
locally of finite presentation, (resp., locally of finite type, surjective, flat, fppf )
if and only if X ′ → X → Y is.
(2) If X ′ → X is a surjective smooth morphism of schemes, a morphism X → Y
of schemes is smooth if and only if X ′ → X → Y is.
(3) If X ′ → X is a surjective étale morphism of schemes, a morphism X → Y
of schemes is étale, (resp., locally quasi-finite, unramified) if and only if
X ′ → X → Y is.

In other words, each property is fppf/smooth/étale local on the source.

Proof. Part (1) reduces to: if A → A′ is a faithfully flat and finitely presented map
of R-algebras, then R → A is of finite type (resp., finite presentation) if and only
if R → A′ is. Using Limit Methods (B.3.2 and B.3.3), there exists a finite type
R-algebra A0 , a faithfully flat and finitely presented map A0 → A′0 of R-algebras,
and a commutative diagram

R / A0 / A′0

⌟
  
A / A′

such that A′ =∼ A ⊗A A′ . If R → A′ is of finite type, after possibly enlarging A′ ,

0 0 0
we may arrange that A′0 → A′ is surjective. As A → A′ is faithfully flat, this implies
that A0 → A is surjective. Hence R → A is also of finite type. We leave the finite
presentation case to the reader. For (2), smoothness descends in fact under fpqc
morphisms by Proposition 2.1.22. Conversely, smooth morphisms are stable under
composition. For (3), it easy to see that locally quasi-finiteness and unramifiedness
descend, and since étaleness is equivalent to smoothness and unramifiedness, étaleness
also descends. See also [EGA, IV4 .17.7.5] and [SP, Tags 036M, 036T, and 036V].

79
2.2 Grothendieck topologies and sites
The introduction of the digit 0 or the group concept was general nonsense
too, and mathematics was more or less stagnating for thousands of years
because nobody was around to take such childish steps...

Alexander Grothendieck, letter to Ronald Brown

To utilize the descent properties of the preceding section, it will be convenient to

generalize the concept of a topological space so that we can view étale/fppf/fpqc
morphisms as ‘opens’, and so that we can formulate the axioms for sheaves and
stacks in these generalized topologies. Grothendieck topologies were introduced in
[SGA4, Def. II.1.3]. Our exposition follows [Art62], [Vis05], [Ols16, §2], and [SP,
Tag 00UZ].

2.2.1 Definitions and first examples

Definition 2.2.1 (Sites). A Grothendieck topology on a category S consists of the
following data: for each object X ∈ S, there is a set Cov(X) consisting of coverings
of X, i.e., collections of morphisms {Xi → X}i∈I in S. We require that:
(1) (identity) If X ′ → X is an isomorphism, then (X ′ → X) ∈ Cov(X).
(2) (restriction) If {Xi → X}i∈I ∈ Cov(X) and Y → X is a morphism, then the
fiber products Xi ×X Y exist in S and the collection {Xi ×X Y → Y }i∈I ∈
Cov(Y ).
(3) (composition) If {Xi → X}i∈I ∈ Cov(X) and {Xij → Xi }j∈Ji ∈ Cov(Xi ) for
each i ∈ I, then {Xij → Xi → X}i∈I,j∈Ji ∈ Cov(X).
A site is a category S with a Grothendieck topology.
Caution 2.2.2. The definition requires that Cov(X) is a set, but this is not true in
many cases of interest, such as the big étale site. We will ignore set-theoretic issues
in this book, but they are usually easy (but annoying) to address by working with a
suitable subcategory containing all morphisms of interest which defines the same
category of sheaves; see [SP, Tag 00VI].
Example 2.2.3 (Topological spaces). If X is a topological space, let Op(X) denote
the category of open sets U ⊆ X, where there is a unique morphism U → V if and
only if U ⊆ V . We say that a covering of
S U (i.e., an element of Cov(U )) is a collection
of open subsets {Ui }i∈I such that U = i∈I Ui . This defines a Grothendieck topology
on Op(X). In particular, if X is a scheme, the Zariski topology on X defines a site
XZar , called the small Zariski site on X.
By replacing Zariski open immersions with étale morphisms, we obtain the small
étale site.
Example 2.2.4 (Small étale site). If X is a scheme, the small étale site on X is the
category Xét of étale morphisms U → X such that a morphism (U → X) → (V → X)
is simply an X-morphism U → V (which is necessarily étale). In other words, Xét
is the full subcategory of Sch/X consisting of schemes étale over X. A covering
of an`object (U → X) ∈ Xét is a collection of étale morphisms {Ui → U } such
that i Ui → U is surjective. Later we will introduce the small étale site Xét of an
algebraic space or Deligne–Mumford stack (Definition 4.1.1), which we use to define
sheaves on X .

80
2.2.2 Big sites
Example 2.2.5 (Big étale site). The big étale site is the category Sch where a
covering
` of a scheme U is a collection of étale morphisms {Ui → U } in Sch such
that i Ui → U is surjective. We denote this site as Schét .

The big étale site Schét is the most important site in this text. It is used to
define the most central notions in this book: an algebraic space is a sheaf on Schét
that is étale locally a scheme (Definition 3.1.2) while an algebraic stack is a stack
over Schét that is smooth locally a scheme (Definition 3.1.6). There are various
analogous sites.

Example 2.2.6 (Big topological site). Let Top be the category of topological spaces.
A covering of U ∈ Top is a collection of open subspaces {Ui ,→ U }i∈I such that
U = i∈I Ui .
S

Example 2.2.7 (Big Zariski site). Replacing étale morphisms in Example 2.2.5
with open immersions defines the big Zariski site SchZar .

Example 2.2.8 (Big fppf site). A morphism of schemes is fppf if it is surjective, flat,
and locally of finite presentation (see Definition A.2.20). The big fppf site Schfppf is
the category
` Sch of schemes where a covering {Ui → U } is a collection of morphisms
such that i Ui → U is fppf.

The big fppf site Schfppf is also important in moduli theory. In [SP], algebraic
spaces and stacks are defined using Schfppf rather than Schét , but these two sites
nevertheless define equivalent notions [SP, Tag 076U]. In this text, we use the big fppf
site in §6.3 to discuss gerbes and quotients stacks by non-smooth groups schemes.

Example 2.2.9 (Big fpqc site). A morphism of schemes is fpqc if it is surjective,

flat, and every quasi-compact subset of the target is the image of a quasi-compact
subset (see Definition A.2.20). The big fpqc site Schfppf is the
` category Sch where a
covering {Ui → U } is a collection of morphisms such that i Ui → U is fpqc.

Caution 2.2.10. There are serious set-theoretic issues in defining an fpqc site, arising
from the presence of too many fpqc covers—given any nonzero ring R, there does
not exist a set of fpqc coverings of Spec R which can refine every fpqc covering; see
[SP, Tags 0BBK and 03NV]. If one defines the big fpqc site ignoring set-theoretic
issues (as we just did!), there are presheaves that do not have a sheafification [Wat75,
Thm. 5.5]. Fortunately, we have no need for fpqc sites in this text. On the other
hand, the notion of a sheaf or stack in the fpqc topology is well-defined, and this
allows us to formulate general statements (but we usually only invoke the étale case).

Example 2.2.11 (Lisse-étale site). On a scheme X, the lisse-étale site Xlis-ét is the
category of schemes smooth over X where morphisms in Xlis-ét are (not necessarily
smooth) morphisms of schemes over X. ` A covering {Ui → U } of an X-scheme U is
a collection of X-morphisms such that i Ui → U is surjective and étale.

We introduce the lisse-étale site of an algebraic stack in Definition 6.1.1 in order

to define quasi-coherent sheaves.

Example 2.2.12 (Restricted categories and sites). If S is a category and S ∈ S,

define the restricted category (sometimes referred to as the localized category) as the
category S/S whose objects are maps T → S in S. A morphism (T ′ → S) → (T → S)
is a map T ′ → T over S. If S is a site, S/S is also a site where a covering of T → S

81
in S/S is a covering {Ti → T } in S. Applying this construction to a scheme S
yields the relative versions of the big Zariski, étale, fppf, and fpqc sites (Sch/S)Zar ,
(Sch/S)ét , (Sch/S)fppf , and (Sch/S)fpqc .

Example 2.2.13 (Grothendieck topologies on the category of affine schemes). In the

literature, authors sometimes use the big sites AffSchZar , AffSchét , and AffSchfppf
on the category of affine schemes. These define the same categories of sheaves as
the corresponding big sites on Sch.

2.3 Presheaves and sheaves

According to A. Grothendieck one really does not need a space to do
geometry, all one needs is a category of sheaves on this would-be space.

Vladimir Berkovich [Ber90]

Recall that if X is a topological space, a presheaf of sets on X is simply a

contravariant functor F : Op(X) → Sets on the category Op(X) of open [Link]
sheaf axiom translates succinctly into the condition that for each covering U = i Ui ,
the sequence Y Y
F (U ) → F (Ui ) ⇒ F (Ui ∩ Uj )
i i,j

is exact (i.e., is an equalizer diagram), where the two maps F (Ui ) ⇒ F (Ui ∩ Uj ) are
induced by restricting along the two inclusions Ui ∩ Uj ⊂ Ui and Ui ∩ Uj ⊆ Uj . Also
note that the intersections Ui ∩ Uj can also be viewed as fiber products Ui ×X Uj .
The definition of presheaves and sheaves on a general site mirrors the topological
case. Using descent, we prove some first examples of sheaves on the big étale site.

2.3.1 Definitions
Definition 2.3.1 (Presheaves). A presheaf on a category S is a contravariant functor
S → Sets.

Remark 2.3.2. If F : S → Sets is a presheaf and f : S → T is a map in S, then the

pullback F (f )(b) of an element b ∈ F (T ) is sometimes denoted as f ∗ b or b|S .

Example 2.3.3 (Representable presheaves). If S is a category and S ∈ S, then

MorS (−, S) : S → Sets defines a presheaf.

Definition 2.3.4 (Sheaves). A sheaf on a site S is a presheaf F : S → Sets such

that for every object S ∈ S and covering {Si → S} ∈ Cov(S), the sequence
Y Y
F (S) → F (Si ) ⇒ F (Si ×S Sj ) (2.3.5)
i i,j

is exact, where the two maps F (Si ) ⇒ F (Si ×S Sj ) are induced by the two maps
Si ×S Sj → Si and Si ×S Sj → Si .

A morphism of presheaves or sheaves is by definition a natural transformation.

Remark 2.3.6. The exactness ofQ(2.3.5) means that it is an equalizer diagram: F (S)
is identified with the subset of i F (Si ) consisting of elements whose images under
the two maps F (Si ) ⇒ F (Si ×S Sj ) are equal.

82
Exercise 2.3.7. Let F be a presheaf on Sch. Show that the following are equivalent:
(1) F is a sheaf on Schét (resp., Schfppf , Schfpqc ),
(2) F sends coproducts to products (i.e., F ( i Ui ) = i F (Ui ) for schemes Ui )
` Q
and for every surjective étale (resp., fppf, faithfully flat) morphism S ′ → S of
schemes, the sequence F (S) → F (S ′ ) ⇒ F (S ′ ×S S ′ ) is exact.
(3) F is a sheaf in the big Zariski topology SchZar and for every surjective étale
(resp., fppf, faithfully flat) morphism S ′ → S of affine schemes, the sequence
F (S) → F (S ′ ) ⇒ F (S ′ ×S S ′ ) is exact.
`
Hint: Given a covering {Si → S}, consider the map i Si → S.

Proposition 2.3.8 (Schemes are Sheaves). If X → S is a morphism of schemes,

then MorS (−, X) : Sch/S → Sets is a sheaf on (Sch/S)fpqc and therefore also a sheaf
on (Sch/S)ét and (Sch/S)fppf .

Proof. As MorS (−, X) is a sheaf in the big Zariski topology, it suffices by Exer-
cise 2.3.7 to show that if T ′ → T is a faithfully flat morphism of affine schemes over
S, then the sequence

MorS (T, X) → MorS (T ′ , X) ⇒ MorS (T ′ ×T T ′ , X)

is exact, which is precisely Fpqc Descent for Morphisms (Corollary 2.1.10).

Exercise 2.3.9.
(a) If F and G are sheaves on a site S, show that the presheaf MorS (F, G), defined
by S 7→ MorSets (F (S), G(S)), is a sheaf on S.
(b) Conclude that if X and Y are schemes over S, the functor MorS (X, Y ) : Sch/S →
Sets, assigning an S-scheme T to MorT (XT , YT ), is a sheaf in the fpqc topology.
(This gives another proof of Corollary 2.1.10(2).)

Exercise 2.3.10 (Gluing sheaves). Let S be a site and (Xi → X) be a covering

in S. If Fi are sheaves on the restricted sites S/Xi and αij : Fi |Xij → Fj |Xij are
isomorphisms of sheaves on S/Xij (where Xij := Xi ×X Xj ) satisfying the cocycle
condition αjk ◦ αij = αik on S/Xijk (where Xijk = Xi ×X Xj ×X Xk ), show that
there exists a unique sheaf F on S and isomorphisms ϕi : F |Xi → Fi satisfying
ϕj |Xij = αij ◦ ϕi |Xij .

Exercise 2.3.11. Show that a surjective smooth (resp., fppf, fpqc) morphism of
schemes is an epimorphism of sheaves on Schét (resp., Schfppf , Schfpqc ).

2.3.2 Fiber products

By Yoneda’s Lemma (0.3.12), if X ∈ S is an object of a category S and F is
a presheaf on S, a morphism α : X → F (which we interpret as a morphism of
presheaves Mor(−, X) → F ) corresponds to an element in F (X), which by abuse of
notation we also denote by α.

Definition 2.3.12. Given morphisms α : F → G and β : G′ → G of presheaves on

a category S, the fiber product of α and β is the presheaf F ×G G′ whose set of
sections over S ∈ S is F (S) ×G(S) G′ (S), i.e.,

F ×G G′ : S → Sets
(2.3.13)
S 7→ {(a, b) ∈ F (S) × G′ (S) | αS (a) = βS (b)} .

83
Exercise 2.3.14.
(a) Show that (2.3.13) is a fiber product F ×G G′ in Pre(S). (This is a generalization
of Exercise 0.3.40 but the same proof should work.)
(b) Show that if F , G, and G′ are sheaves on a site S, then so is F ×G G′ . In
particular, (2.3.13) is also a fiber product F ×G G′ in Sh(S).

2.3.3 Sheafification
Theorem 2.3.15 (Sheafification). Let S be a site. The forgetful functor Sh(S) →
Pre(S) admits a left adjoint F 7→ F sh , called the sheafification.
Proof. A presheaf F on S is Qcalled separated if for every covering {Si → S} of an
object S, the map F (S) → i F (Si ) is injective (i.e., if sections glue, they glue
uniquely). Let Pre(S) and Sh(S) be the categories of presheaves and sheaves, and
let Presep (S) ⊂ Pre(S) be the full subcategory of separated presheaves. We will
construct left adjoints
sh2 sh1
v v 
Sh(S)  / Presep (S)  / Pre(S).

For F ∈ Pre(S), we define sh1 (F ) by S 7→ F (S)/ ∼ where a ∼ b if there exists a

covering {Si → S} such that a|Si = b|Si for all i. For F ∈ Presep (S), we define
sh2 (F ) by

{Si → S} ∈ Cov(S) and ai ∈ F (Si )
 
S 7→ {Si → S}, {ai } /∼
such that ai |Sij = aj |Sij for all i, j

where ({Si → S}, {ai }) ∼ ({Sj′ → S}, {a′j }) if ai |Si ×S Sj′ = a′j |Si ×S Sj′ for all i, j. The
details are left to the reader.
Remark 2.3.16 (Topos). A topos is a category equivalent to the category of sheaves
on a site. Two different sites may have equivalent categories of sheaves, and the
topos is a more fundamental invariant. While topoi are undoubtedly important in
moduli theory, they will not play a role in these notes.

2.3.4 A criterion for a sheaf to be a scheme

The following is a reinterpretation of Fppf Descent for Schemes (2.1.11, 2.1.12, and
2.1.14).
Proposition 2.3.17 (Descent Criterion for an Fppf Sheaf to be a Scheme). Let P
be one of the following properties of morphisms of schemes: open immersion, closed
immersion, affine, quasi-affine, or locally quasi-finite and separated. Let X → Y be a
surjective smooth (resp., fppf ) morphism of schemes. Let F be a sheaf on (Sch/Y )ét
(resp., (Sch/Y )fppf ). Consider the fiber product

FX /X

□
 
F /Y

of sheaves. If FX is a scheme and FX → X has P, then F is a scheme and F → Y

has P.

84
Proof. As FX is the pullback of F , there is a canonical isomorphism α : p∗1 FX → p∗2 FX
on X ×Y X satisfying the cocycle condition p∗23 α ◦ p∗12 α = p∗13 α. By Fppf Descent
for Schemes (2.1.11, 2.1.12, and 2.1.14), there exists a morphism of schemes W → Y
satisfying P that pulls back to FX → X. The sheaf F is identified with the
sheafification of the presheaf on Sch/Y defined by

F pre : (T → Y ) 7→ Eq FX (T → Y ) ⇒ (FX ×X FX )(T → Y ) .

By Schemes are Sheaves (2.3.8), W is a sheaf and there is a morphism F pre → W of

presheaves. By the universal property of sheafification, there is a morphism α : F →
W of sheaves which pulls back under X → Y to an isomorphism αX : FX → WX .
It is not hard to see that this implies that α is an isomorphism. For instance, since
Hom(W, F ) is a sheaf (Exercise 2.3.9) and the inverse βX of αX defines a section
over X → Y whose two pullbacks to X ×Y X agree, the inverse βX descends to
a section β of Hom(W, F ) over id : Y → Y . Since Hom(F, F ) and Hom(W, W ) are
sheaves, it follows that β ◦ α = idF and α ◦ β = idW . See also [SP, Tag 02W5].

2.4 Prestacks
The Red Queen shook her head. “You may call it ‘nonsense’ if you like,”
she said, “but I’ve heard nonsense, compared with which that would be as
sensible as a dictionary!”

Lewis Carroll, Through the Looking-Glass

Prestacks and stacks were introduced by Grothendieck in [FGAI, §A.1] and

[SGA1, §6] to express the categorical structure of objects satisfying fpqc descent.
The language of prestacks was further developed by Giraud [Gir64] and [Gir71].
Motivation for prestacks was provided in §0.6.1: in an effort to keep track of
automorphisms, we were naively led to consider a ‘functor’

F : S → Groupoids .

While this is a good way to think about prestacks, it is more convenient to define a
prestack by packaging the groupoids F (S) for S ∈ S into one massive category X
over S parameterizing pairs (a, S) where S ∈ S and a ∈ F (S).

2.4.1 Definition of a prestack

Let S be a category and p : X → S be a functor of categories. We visualize this data
as
X a
α /b
_ _
p
  
S S
f
/T

where the lower case letters a, b are objects of X and the upper case letters S, T
are objects of S. We say that a is over S and that a morphism α : a → b is over
f : S → T.

Definition 2.4.1 (Prestacks). A functor p : X → S is a prestack over a category S

if

85
 (1) (pullbacks exist) for every diagram

a /b
_ _

 
S /T

of solid arrows, there exists a morphism a → b over S → T ; and

(2) (universal property for pullbacks) for every diagram

a /b /$ c
_ _ _

  
R /S /T

of solid arrows, there exists a unique arrow a → b over R → S filling in the

diagram.

Caution 2.4.2. When defining and discussing prestacks, we often write X instead of
X → S, but when necessary, we denote the projection by pX : X → S. We do not
usually spell out the definition of the functor X → S as it should be clear to the
reader. Moreover, when defining a prestack X , we often only define the objects and
morphisms in X and leave the composition law to the reader.
Remark 2.4.3. Axiom (2) above implies that the pullback in Axiom (1) is unique up
to unique isomorphism. We often write f ∗ b or simply b|S to indicate a choice of a
pullback.

Definition 2.4.4 (Fiber categories). If X is a prestack over S, the fiber category

X (S) over S ∈ S is the category of objects in X over S with morphisms over idS .

Exercise 2.4.5. Show that the fiber category X (S) is a groupoid.

Caution 2.4.6. Our terminology is not standard. Prestacks are usually referred to as
categories fibered in groupoids. In the literature (c.f., [FGI+ 05, §4], [Ols16, §4.6]),
a prestack is sometimes defined as a category fibered in groupoids together with
Axiom 2.5.1(1) of a prestack.
It is also standard to call a morphism b → c in X cartesian if it satisfies the
universal property in Axiom 2.5.1(2) and p : X → S a fibered category if for every
diagram as in Axiom 2.5.1(1), there exists a cartesian morphism a → b over S → T .
With this terminology, a prestack (as we have defined it) is a fibered category
where every arrow is cartesian, or equivalently where every fiber category X (S) is a
groupoid.

2.4.2 Examples
Example 2.4.7 (Presheaves as prestacks). If F : S → Sets is a presheaf, we can
construct a prestack XF as the category of pairs (a, S) where S ∈ S and a ∈ F (S).
A morphism (a′ , S ′ ) → (a, S) in XF is a map f : S ′ → S such that a′ = f ∗ a, where
f ∗ is convenient shorthand for F (f ) : F (S) → F (S ′ ). The projection XF → S is
defined by (a, S) 7→ S. Observe that the fiber categories XF (S) are equivalent (even
equal) to the set F (S). We will often abuse notation by conflating F and XF .

86
Example 2.4.8 (Representable prestacks). If S is an object of category S, then
the restricted category S/S of objects over S defines a prestack over S, where
the projection morphism S/S → S is given by (T → S) 7→ T . This is the same
prestack obtained by applying the previous example to the representable presheaf
Mor(−, S) : S → Sets.

Example 2.4.9 (Schemes as prestacks). A scheme X defines the prestack Sch/X of

schemes over X as in the previous example. The projection Sch/X → Sch is given
by (T → X) 7→ T . We often abuse notation by referring to Sch/X simply as X.

Example 2.4.10 (Prestack of smooth curves). We define the prestack M over

Sch as the category of families of smooth curves C → S, i.e., smooth and proper
morphisms C → S of schemes such that every geometric fiber is a connected curve.
A map (C ′ → S ′ ) → (C → S) is the data of maps α : C ′ → C and f : S ′ → S such
that the diagram
C′
α /C

□
 
S′
f
/S

is cartesian. The projection M → Sch is given by (C → S) 7→ S.

The prestack Mg is defined as the full subcategory of M consisting of families
of smooth curves C → S where every geometric fiber has genus g. Note that the
fiber category Mg (k) over an algebraically closed field k is the groupoid of smooth,
connected, and projective curves C over k of genus g such that MorMg (k) (C, C ′ ) =
IsomSch/k (C, C ′ ).

Exercise 2.4.11 (easy but important). Verify that M and Mg are prestacks.

Example 2.4.12 (Prestack of coherent sheaves and vector bundles). Let X be a

scheme over a field k. We define the prestack QCoh(X) over Sch/k as the category
of pairs (E, S) where S is a scheme over k and E is a quasi-coherent sheaf on
XS = X ×k S flat over S. A morphism (E ′ , S ′ ) → (E, S) consists of a map of
schemes f : S ′ → S together with an isomorphism f ∗ E → E ′ of OXS′ -modules.1
The projection QCoh(X) → Sch/k is defined by (E, S) 7→ S.
The substack Coh(X) ⊆ QCoh(X) is the full subcategory consisting of pairs
(E, S) where E is a finitely presented, quasi-coherent sheaf on XS (or equivalently
a coherent sheaf when XS is noetherian). Similarly, Bun(X) ⊆ Coh(X) is the full
subcategory where E is a vector bundle on XS (i.e., locally free quasi-coherent sheaf
of finite rank).

Exercise 2.4.13. Verify that QCoh(X), Coh(X), and Bun(X) are prestacks.
1 This definition of a morphism is not completely precise as the pullback f ∗ E is not canonical.

Recall that f ∗ E is defined as f −1 E ⊗f −1 OS OS ′ , and while it exists and is unique up to unique

isomorphism, a choice of pullback f ∗ E involves a choice of a limit in the definition of f −1 E,
choices of tensor products, and a choice of sheafification. Instead, one can define a morphism
(E ′ , S ′ ) → (E, S) as an equivalence class of triples (f, F ′ , α) where f : S ′ → S is a map of schemes,
F ′ is a choice of a pullback of E, and an isomorphism α : F ′ → E ′ , where (f, F ′ , α) ∼ (g, G′ , β)
∼
if f = g and the canonical isomorphism γ : F ′ → G′ satisfies α = β ◦ γ. Alternatively, since the
pushforward f∗ E is canonical, a morphism (E , S ′ ) → (E, S) can be defined as a map f : S ′ → S
′ ′

and a morphism E → f∗ E ′ of OXS -modules whose adjoint is an isomorphism.

87
2.4.3 Classifying stacks and quotient stacks
Classifying and quotient stacks were motivated in §0.6.5. Their definitions involve
the notion of a principal G-bundle. For a smooth affine group scheme G → S 2 , a
principal G-bundle over an S-scheme T is a morphism P → T of schemes with an
action of G on P via σ : G ×S P → P such that P → T is a G-invariant smooth
morphism and

(σ, p2 ) : G ×S P → P ×T P, (g, p) 7→ (gp, p)

is an isomorphism (see Definition B.1.47); in other words, G acts freely and transi-
tively on P with quotient T . Equivalently, P → T is a principal G-bundle if there is
étale cover T ′ → T such that P ×T T ′ is G-equivariantly isomorphic to the trivial
principal G-bundle G ×S T ′ (Proposition B.1.49). See §B.1.7 for further background
and many examples.

Definition 2.4.14 (Classifying stacks). The classifying stack BG of a smooth affine

group scheme G → S is the category over Sch/S whose objects are principal G-
bundles P → T and a morphism (P ′ → T ′ ) → (P → T ) is the data of a G-equivariant
morphism P ′ → P such that
P′ /P

□
 
T′ /T
is cartesian.

Definition 2.4.15 (Quotient prestacks and stacks). Let G → S be a smooth affine

group scheme acting on a scheme U over S. The quotient prestack [U/G]pre of an
action of a smooth affine group scheme G → S on an S-scheme U is the category
over Sch/S consisting of pairs (T, u) where T is an S-scheme and u ∈ U (T ). A
morphism (T ′ , u′ ) → (T, u) is the data of a map f : T ′ → T of S-schemes and an
element g ∈ G(T ′ ) such that f ∗ u = g · u′ . Note that the fiber category [U/G]pre (T )
is identified with the quotient groupoid [U (T )/G(T )] from Example 0.4.6.
The quotient stack [U/G] is the category over Sch/S consisting of diagrams

P /U


T

where P → T is a principal G-bundle and P → U is a G-equivariant morphism of

S-schemes. A morphism (T ′ ← P ′ → U ) → (T ← P → U ) consists of a morphism
T ′ → T and a G-equivariant morphism P ′ → P of schemes such that the diagram

/P '/
P′ U
□
 
T′ /T

is commutative and the left square is cartesian.

2 In §6.3.2, we will define principal G-bundles, classifying stacks, and quotient stacks more

generally for fppf group schemes.

88
Exercise 2.4.16 (easy). Verify that [U/G]pre and [U/G] are prestacks over Sch/S.

We show shortly that [U/G] and BG = [S/G] are stacks over (Sch/S)ét (Propo-
sition 2.5.13), which justifies our terminology of a ‘quotient stack’ and ‘classifying
stack’. We also show that [U/G] is identified as the stackification of [U/G]pre
(Exercise 2.5.21), and later show that [U/G] is algebraic (Theorem 3.1.10).

2.4.4 Morphisms of prestacks

Definition 2.4.17.
(1) A morphism of prestacks f : X → Y is a functor f : X → Y such that the
diagram
X
f
/Y

pX pY
 
S
strictly commutes, i.e., for every object a ∈ Ob(X ), there is an equality pX (a) =
pY (f (a)) of objects in S, and for every morphism α : a → b, pX (α) = pY (f (α)).
(2) If f, g : X → Y are morphisms of prestacks, a 2-isomorphism (or 2-morphism)
α : f → g is a natural transformation α : f → g such that for every object
a ∈ X , the morphism αa : f (a) → g(a) in Y (which is an isomorphism) is over
the identity in S. We often describe the 2-isomorphism α schematically as

f
'
X  α
7 Y.
g

(3) We define the category Mor(X , Y) whose objects are morphisms of prestacks
and whose morphisms are 2-isomorphisms.
(4) A 2-commutative diagram (which we often call simply a commutative diagram)
is a diagram
f′
X′ / Y′

g′ | α
g
 
X
f
/Y
∼
together with a 2-isomorphism α : g ◦ f ′ → f ◦ g ′ .
(5) A morphism f : X → Y of prestacks is a monomorphism (resp., epimorphism)
if f is fully faithful (resp., essentially surjective), and f is an isomorphism if
∼
there exists a morphism g : Y → X of prestacks and 2-isomorphisms g◦f → idX
∼
and f ◦ g → idY .

Exercise 2.4.18 (easy). Show that every 2-isomorphism is an isomorphism of

functors, or in other words that Mor(X , Y) is a groupoid.

Exercise 2.4.19 (details). Let f : X → Y be a morphism of prestacks over a

category S.
(a) Show that f is a monomorphism if and only if fS : X (S) → Y(S) is fully
faithful for every S ∈ S.

89
 (b) Show that f is an isomorphism if and only if f is fully faithful and essentially
surjective.
A prestack X is equivalent to a presheaf if there is a presheaf F and an isomorphism
between X and the prestack XF corresponding to F (see Example 2.4.7).
Exercise 2.4.20 (good practice). Show that G acts freely on U (i.e., the action
map (σ, p2 ) : G ×S U → U ×S U is a monomorphism) if and only if [U/G]pre (resp.,
[U/G]) is equivalent to a presheaf. We denote these presheaves by (U/G)pre and
U/G.

2.4.5 The 2-Yoneda Lemma

The Yoneda Lemma (0.3.12) states that for a presheaf F : S → Sets on a category S
∼
and an object S ∈ S, there is a bijection Mor(S, F ) → F (S). In particular, there is
a fully faithful embedding S → Pre(S), from S into the category of presheaves on S,
given by S 7→ Mor(−, S). We will need an analog of Yoneda’s lemma for prestacks.
Recall from Example 2.4.8 that an object S ∈ S can be viewed as a prestack over S,
which we also denote by S, whose objects over T ∈ S are morphisms T → S and a
morphism (T → S) → (T ′ → S) is an S-morphism T → T ′ .
Lemma 2.4.21 (The 2-Yoneda Lemma). Let X be a prestack over a category S
and S ∈ S. The functor

Mor(S, X ) → X (S), f 7→ fS (idS )

is an equivalence of categories.
Proof. We will construct a quasi-inverse Ψ : X (S) → Mor(S, X ) as follows.
On objects: For a ∈ X (S), we define Ψ(a) : S → X as the morphism of prestacks
sending an object (f : T → S) (of the prestack corresponding to S) over T to a
choice of pullback f ∗ a ∈ X (T ) and a morphism (f ′ : T ′ → S) → (f : T → S), given
by an S-morphism g : T ′ → T , to the morphism f ′∗ a → f ∗ a uniquely filling in the
diagram
f ′∗ a / f ∗a /( a
_ _

  
T′
g
/T f
/ S,

using Axiom (2) of a prestack.

On morphisms: If α : a′ → a is a morphism in X (S), then Ψ(α) : Ψ(a′ ) → Ψ(a)
is defined as the morphism of functors which maps a morphism f : T → S (i.e., an
object in S over T ) to the unique morphism f ∗ a′ → f ∗ a filling in the diagram

f ∗ a′ / f ∗a T
over f
 
a′
α / a S

using again Axiom (2) of a prestack.

We leave the verification that Ψ is a quasi-inverse to the reader.
Caution 2.4.22. We will use the 2-Yoneda Lemma, often without mention, throughout
these notes in passing between morphisms S → X and objects of X over S.

90
Remark 2.4.23. If f, g : S → X are morphisms from a scheme S, corresponding
∼
via the 2-Yoneda Lemma to objects a, b ∈ X (S), then a 2-isomorphism α : f → g
∼
corresponds to an isomorphism a → b in X (S).
Remark 2.4.24 (Universal families). When F : Sch → Sets is the functor hM =
Mor(−, M ) representable by a scheme M , the (usual) Yoneda Lemma (0.3.12) gives
a bijection Mor(hM , F ) ∼
= F (M ) and the object U ∈ F (M ) corresponding to the
identity map is a universal family (see §0.3.4). The Yoneda 2-Lemma does not
immediately apply to give a universal family as the category X (X ) of objects of
X over X needs to be defined. Later in §3.1.7, we prove a Generalized 2-Yoneda
Lemma (3.1.25) for algebraic stacks and use it to define universal families.
Remark 2.4.25 (Cleavages and splittings). In the proof of the Yoneda 2-Lemma, we
made a choice of pullbacks f ∗ : X (S) → X (T ) for every morphism T → S as in [SP,
Tag 02XN]. This is often called a cleavage in the literature [SGA1, VI.7], [Vis05,
Def. 3.9]. A choice of pullbacks defines a pseudo-functor S → Groupoids. If for a
f g
composition T − →S− → U , the functors (g ◦ f )∗ and f ∗ ◦ g ∗ are equal, then it is often
said that this choice of pullbacks defines a splitting or that X is split. In this case,
the induced pseudo-functor S → Groupoids is a strict functor. Every prestack (or
more generally fibered category) is equivalent in the 2-categorical sense to a split
prestack [SP, Tag 004A]. Do not worry; we will not make use of this terminology.
Example 2.4.26 (Quotient stack presentations). Consider the prestack [U/G] in
Definition 2.4.15 arising from a group action σ : G ×S U → U . The object of [U/G]
over U given by the diagram

G ×S U
σ /U
p2

U
corresponds via the 2-Yoneda Lemma (2.4.21) to a morphism U → [U/G]. Later
we will verify that U → [U/G] is surjective, smooth, and representable, or in other
words a smooth presentation. This will verify that [U/G] is an algebraic stack
(Theorem 3.1.10).
Exercise 2.4.27 (good practice).
(a) Show that there is a morphism p : U → [U/G]pre and a 2-commutative diagram

G ×S U
σ /U

p2  α
p
 
U
p
/ [U/G]pre .

(b) Show that U → [U/G]pre is a categorical quotient among prestacks, i.e., for
every 2-commutative diagram

G ×S U
σ /U

p2  α
p
  φ
U
p
/ [U/G]pre

φ  τ

/Z

91
 of prestacks, there exists a morphism χ : [U/G]pre → Z and a 2-isomorphism
∼
β : φ → χ ◦ p which is compatible with α and τ (i.e., the two natural transfor-
β◦σ χ◦α τ β◦p2
mations φ ◦ σ −−→ χ ◦ p ◦ σ −−→ χ ◦ p ◦ p2 and φ ◦ σ −
→ φ ◦ p2 −−−→ χ ◦ p ◦ p2
agree. Show that χ is unique up to unique 2-isomorphism.

2.4.6 Warmup: fiber products of groupoids

Fiber products of prestacks brings out many of the 2-categorical subtleties. As
understanding fiber products is essential to working with algebraic stacks, we
recommend the reader spend time working out many examples. It is instructive to
begin with fiber products of groupoids.
Construction 2.4.28. Let f : C → D and g : D′ → D be functors of groupoids.
Define the groupoid C ×D D′ as the category of triples (c, d′ , γ) where c ∈ C and
∼
d′ ∈ D′ are objects, and γ : f (c) → g(d′ ) is an isomorphism in D. A morphism
∼ ∼
(c1 , d′1 , γ1 ) → (c2 , d′2 , γ2 ) is the data of morphisms χ : c1 → c2 and λ : d′1 → d′2 such
that
f (χ)
f (c1 ) / f (c2 )

γ1 γ2
 g(λ) 
g(d′1 ) / g(d′2 )
commutes.
Exercise 2.4.29 (details). Formulate a university property for fiber products of
groupoids and show that C ×D D′ satisfies it.
The following foreshadows important cartesian diagrams involving quotient
stacks.
Exercise 2.4.30 (important, good practice). Let G be a group acting on a set U via
σ : G × U → U . Let [U/G] denote the quotient groupoid as defined in Exercise 0.4.8:
objects are elements u ∈ U and a morphism u → u′ is an element g ∈ G with
u′ = gu. Let p : U → [U/G] denote the projection.
(a) Let P be a set with a free action of G and quotient T = P/G. If f : P → U is
a G-equivariant map, show that there is a cartesian diagram

P
f
/U

□ p
 
T / [U/G],

i.e., that P is equivalent to the fiber product.

(b) Show that there are cartesian diagrams

G×U
σ /U G×U
(σ,p2 )
/ U ×U
p2 □ p and □ p×p
   
U
p
/ [U/G] [U/G]
∆ / [U/G] × [U/G].

Exercise 2.4.31 (good practice). Recall from Example 0.4.3 that the classifying
groupoid BG of a group G is the category with one object ∗ with Mor(∗, ∗) = G.

92
 (a) Let ϕ : H → G be a homomorphism of groups. Show that there is an induced
morphism BH → BG of groupoids and that BH ×BG pt ∼ = [G/H], where pt
denotes the groupoid with one object and one morphism.
(b) If K ⊴ G is a normal subgroup with quotient Q = G/K, show that there is a
cartesian diagram
Q / BK / pt

□ □
  
pt / BG / BQ.

(c) Let G be a group acting on a set U and let [U/G] be the groupoid quotient.
If u ∈ U is a point with stabilizer Gu and orbit Gus, show that there is a
morphism BGu → [U/G] of groupoids and a cartesian diagram

Gu /U

□ p
 
BGu / [U/G].

Exercise 2.4.32 (important). Show that a groupoid C is equivalent to a set if and

only if C → C × C is fully faithful.

2.4.7 Fiber products of prestacks

The fiber product of morphisms X → Y and Y ′ → Y of prestacks over a category
S is the prestack X ×Y Y ′ whose fiber category over S ∈ S is the fiber product
X (S) ×Y(S) Y ′ (S) of groupoids.
Construction 2.4.33. Let f : X → Y and g : Y ′ → Y be morphisms of prestacks
over a category S. Define the prestack X ×Y Y ′ over S as the category of triples
(x, y ′ , γ) where x ∈ X and y ′ ∈ Y ′ are objects over the same object S := pX (x) =
∼
pY ′ (y ′ ) ∈ S, and γ : f (x) → g(y ′ ) is an isomorphism in Y(S). A morphism
(x1 , y1 , γ1 ) → (x2 , y2 , γ2 ) consists of a triple (h, χ, λ) where h : pX (x1 ) = pY ′ (y1′ ) →
′ ′

pY ′ (y2′ ) = pX (x2 ) is a morphism in S, and χ : x1 → x2 and λ : y1′ → y2′ are morphisms

in X and Y ′ over h such that
f (χ)
f (x1 ) / f (x2 )

γ1 γ2
 g(λ) 
g(y1′ ) / g(y2′ )

commutes.
Letting p1 : X ×Y Y ′ → X and p2 : X ×Y Y ′ → Y ′ denote the projections
∼
(x, y ′ , γ) 7→ x and (x, y ′ , γ) 7→ y ′ , define the 2-isomorphism α : f ◦ p1 → g ◦ p2 , which
is defined on an object (x, y ′ , γ) ∈ X ×Y Y ′ by setting α(x,y′ ,γ) := γ : f (x) → g(y ′ ).
This yields a 2-commutative diagram

X ×Y Y ′
p2
/ Y′

p1 α @H g (2.4.34)
 
X
f
/ Y.

93
Theorem 2.4.35. The prestack X ×Y Y ′ together with the morphisms p1 and p2
and the 2-isomorphism α as in (2.4.34) satisfy the following universal property: for
every 2-commutative diagram
q2 . Y′
?
p2 g

KS KS 
T τ X ×Y Y ′ α
?Y
p1  f
q1 0X

∼
with 2-isomorphism τ : f ◦ q1 → g ◦ q2 , there exists a morphism h : T → X ×Y Y ′
and 2-isomorphisms β : q1 → p1 ◦ h and ρ : q2 → p2 ◦ h yielding a 2-commutative
diagram
q2 . Y′
?
p2 g
⇓ρ
KS 
h / X ×Y Y ′
T α
?Y
⇑β
p1  f
q1 0X

such that
f (β)
f ◦ q1 / f ◦ p1 ◦ h

τ α◦h
 g(ρ) 
g ◦ q2 / g ◦ p2 ◦ h

commutes. The data (h, β, ρ) is unique up to unique isomorphism.

∼
Proof. We
define h : T → X ×Y Y on objects
′
by t 7→ q1 (t), q2 (t), τt : f (q1 (t)) →
g(q2 (t)) and on morphisms as (Ψ : t → t′ ) 7→ (pT (Ψ), q1 (Ψ), q2 (Ψ)). There are
equalities of functors q1 = p1 ◦ h and q2 = p2 ◦ h so we can define β and ρ as the
identity natural transformations. The remaining details are left to the reader.
Definition 2.4.36. We say that a 2-commutative diagram

X′ / Y′
α <D
 
X / Y.

is cartesian if it satisfies the universal property of Theorem 2.4.35. We often write a

cartesian diagram of prestacks as

X′ / Y′

□
 
X / Y,

where the existence of the 2-isomorphism α is implicit.

94
 When X and Y are prestacks over Sch, then X ×Sch Y = X ×Spec Z Y, which we
sometimes abbreviate as X × Y. Likewise, when working over a field k, X ×Sch/k Y =
X ×Spec k Y is abbreviated as X ×k Y or, if there is no possible confusion, as X × Y.

2.4.8 Examples
The following exercise is essential for verifying the algebraicity and establishing
properties of quotient stacks (e.g., see Theorem 3.1.10).
Exercise 2.4.37 (important, good practice). Let G → S be a smooth affine group
scheme acting on a scheme U over S via σ : G ×S U → U , and let [U/G] be the
quotient stack (Definition 2.4.15).
(a) Let T → [U/G] be a morphism corresponding via the 2-Yoneda Lemma (2.4.21)
to a principal G-bundle P → T and a G-equivariant map f : P → U . Show
that there is a cartesian diagram

P
f
/U

□
 
T / [U/G].

(We will later see that [U/G] is an algebraic stack and that U → [U/G]
is principal G-bundle (Theorem 3.1.10). This will allow us to identify the
principal G-bundle U → [U/G] together with the identity map U → U as the
universal family over [U/G], corresponding via the 2-Yoneda Lemma to the
identity map [U/G] → [U/G].)
(b) Show that there are cartesian diagrams

G ×S U
σ /U G ×S U
(σ,p2 )
/ U ×S U
p2 □ p and □ p×p
   
U
p
/ [U/G] [U/G]
∆ / [U/G] ×S [U/G].

The diagram in the next exercise is utilized extensively, just as it is in the case
of schemes. It will be used to define stabilizers and the inertia stack in §3.2.2.
Exercise 2.4.38 (important). Let X be a prestack over a category S. Let S, T ∈ S
be objects which we can view as prestacks over S via Example 2.4.9. Show that for
all morphisms a : S → X and b : T → X , there is a cartesian diagram

S ×X T / S×T

□ a×b
 
X
∆ / X × X,

where the fiber products S × T and X × X are taken over S.

Exercise 2.4.39 (Isom presheaves, important). Let X be a prestack over a category
S. For S ∈ S, recall from Example 2.2.12 that the restricted category S/S denotes
the category whose objects are morphisms T → S in S and whose morphisms are
S-morphisms.

95
 (a) Show that for objects a and b of X over S that the functor

IsomX (S) (a, b) : S/S → Sets

f
→ S) 7→ MorX (T ) (f ∗ a, f ∗ b),
(T −

where f ∗ a and f ∗ b are choices of pullbacks, defines a presheaf on S/S.

(b) Show that there is a cartesian diagram

IsomX (S) (a, b) /S

□ (a,b)
 
X
∆ / X × X.

(c) Show that the presheaf AutX (T ) (a) = IsomX (T ) (a, a) is naturally a presheaf
in groups.
(d) Show that X is equivalent to a presheaf if and only if the diagonal X → X × X
is fully faithful.
Exercise 2.4.40 (good practice).
(a) If H → G is a morphism of smooth affine group schemes over a scheme S, define
a morphism of prestacks BH → BG over Sch/S by using Definition B.1.60 to
construct a principal G-bundle from a principal H-bundle.
(b) Show that BH ×BG S ∼ = [G/H].
(c) If 1 → K → G → Q → 1 is an exact sequence of smooth affine algebraic
groups over a field k, show that there is a cartesian diagram

Q / BK / Spec k

□ □
  
Spec k / BG / BQ.

Exercise 2.4.41 (good practice). Let G and H be smooth affine group schemes
over a scheme S.
(a) Show that B(G ×S H) ∼
= BG ×S BH.
(b) If X and Y are S-schemes with a actions by G and H, show that [(X ×S
Y )/(G ×S H)] ∼
= [X/G] ×S [Y /H].
(c) Conclude that [An /Gnm ] ∼
= [A1 /Gm ] × · · · × [A1 /Gm ] over Spec Z.
| {z }
n times
Exercise 2.4.42 (important, good practice). Analogous to the prestack Mg of
smooth curves (Example 2.4.10), let Mg,1 be the prestack, where an object over
a scheme S is a family of smooth curves C → S and a section σ : S → C. Let
Mg,1 → Mg be the morphism of prestacks forgetting the section. Show that if
S → Mg is a morphism corresponding to a family of curves C → S, there is a
cartesian diagram
C / Mg,1

□
 
S / Mg .

96
In other words, Mg,1 → Mg is the universal family, as defined later in Defini-
tion 3.1.27.
Exercise 2.4.43 (good practice). Let H and G be smooth affine group schemes
over a scheme S. Let Hom(H, G) be the sheaf on (Sch/S)ét whose sections over an
S-scheme T are homomorphisms HT → GT , and let Mor(BH, BG) be the prestack
over S whose objects over T ∈ S are morphisms B(HT ) → B(GT ). Show that
Mor(BH, BG) ∼
= [Hom(H, G)/G],
where G acts via conjugation.

2.5 Stacks
An absence of proof is a challenge; an absence of definition is deadly.

Pierre Deligne

A stack over a site S is a prestack X where the objects and morphisms glue
uniquely in the Grothendieck topology of S.

2.5.1 The definition

Given a covering {Si → S} in a site, we will use the convention that Sij denotes
Si ×S Sj and Sijk denotes Si ×S Sj ×S Sk .
Definition 2.5.1 (Stacks). A prestack X over a site S is a stack if the following
conditions hold for all coverings {Si → S} of an object S ∈ S:
(1) (morphisms glue) For objects a and b in X over S and morphisms ϕi : a|Si → b
such that ϕi |Sij = ϕj |Sij as displayed in the diagram

a|Si ϕi
? ? Si
 
a|Sij ?a /= b over Sij
? S,
 
a|Sj ϕj Sj

there exists a unique morphism ϕ : a → b over idS with ϕ|Si = ϕi .

(2) (objects glue) For objects ai over Si and isomorphisms αij : ai |Sij → aj |Sij ,
as displayed in the diagram
5 ai
? Si
ai |Sij

αij
>a over Sij ?S

aj |Sij

) Sj
aj

satisfying the cocycle condition αjk |Sijk ◦ αij |Sijk = αik |Sijk on Sijk , there
exists an object a over S and isomorphisms ϕi : a|Si → ai over idSi such that
ϕj |Sij = αij ◦ ϕi |Sij on Sij .

97
 A morphism of stacks is a morphism of prestacks.
Remark 2.5.2. If the covering consists of a single map S ′ → S and a′ ∈ X is an object
∼
over S ′ , the cocycle condition for an isomorphism α : p∗1 a′ → p∗2 a′ over S ′ ×S S ′
translates to the commutativity of

p∗
12 α
p∗12 p∗1 a / p∗12 p∗2 a

p∗13 p∗1 a p∗ p∗
p∗23 p∗1 a
13 α 23 α
( v
p∗13 p∗2 a p∗23 p∗2 a

over S ′ ×S S ′ ×S S ′ . Axiom (2) requires the existence of an object a ∈ X (S) and an

∼
isomorphism ϕ : a′ → a|S ′ satisfying p∗1 ϕ = p∗2 ϕ ◦ α.
Remark [Link] to Q the sheaf axiom of a presheaf F : S → Sets requiring
that F (S) → i F (Si ) ⇒ i,j F (Si ×S Sj ) is exact for coverings {Si → S}, the
stack axioms can be interpreted as the ‘exactness’ of

/ Q X (Si ) // Q // Q
X (S) i i,j X (Si ×S Sj ) / i,j,k X (Si ×S Sj ×S Sk ).

Exercise 2.5.4. Show that Axiom (1) is equivalent to the condition that for all
objects a and b of X over S ∈ S, the Isom presheaf IsomX (S) (a, b) (see Exercise 2.4.39)
is a sheaf on S/S.
Exercise 2.5.5. Generalizing Exercise 2.3.7 from sheaves to stacks, show that a
prestack X over Sch is a stack over Schét (resp., Schfppf , Schfpqc ) if and only if X is
a stack over SchZar and Axioms (1) and (2) hold for a surjective étale (resp., fppf,
fpqc) morphism Spec A′ → Spec A of affine schemes.
Exercise 2.5.6 (Fiber product of stacks). Show that if X → Y and Y ′ → Y are
morphisms of stacks over a site S, then X ×Y Y ′ is also a stack over S.

2.5.2 First examples of stacks

Example 2.5.7 (Sheaves and schemes as stacks). A presheaf F on a site S defines
a prestack XF over S whose objects are pairs (a, S) where S ∈ S and a ∈ F (S)
(see Example 2.4.7), and F is a sheaf if and only if XF is a stack. We often abuse
notation by writing F also as the stack XF .
Since a scheme X is a sheaf on Schét (Proposition 2.3.8), the prestack Sch/X—
often denoted simply as X—whose objects over a scheme S are morphisms S → X,
is a stack over Schét .
Example 2.5.8 (Stacks of sheaves). Let Sheaves be the prestack over Sch whose
objects are pairs (T, F ) where T is a scheme and F is a sheaf on the Zariski
topology of T . A morphism (T ′ , F ′ ) → (T, F ) is pair (f, α) where f : T ′ → T is a
map of schemes and α : f −1 F → F ′ is an isomorphism of sheaves. The projection
Sheaves → Sch is defined by (T, F ) 7→ T . Because sheaves and their morphisms glue
in the Zariski topology [Har77, Exc. II.1.15 and 22], X is a stack over the big Zariski
site SchZar .
Example 2.5.9 (Stack of quasi-coherent sheaves). Define the QCoh, Coh, and Bun
as the category over Sch consisting of pairs (T, F ) where F is a quasi-coherent sheaf
(resp., finitely presented, quasi-coherent sheaf, vector bundle) on a scheme T and a

98
morphism (T, F ) → (T ′ , F ′ ) is a map f : T → T ′ and an isomorphism α : f ∗ F ′ → F .
To see that QCoh is a stack over Schfpqc , by Exercise 2.5.5, it suffices to verify
Axioms (1) and (2) with respect to an fpqc map T ′ → T , and these translate literally
to the two parts of Fpqc Descent of Quasi-Coherent Sheaves (2.1.4(2)). In particular,
QCoh is a stack over Schét . By Fpqc Descent of Properties of Quasi-Coherent
Sheaves (2.1.18), Coh and Bun are also stacks over Schfpqc and Schét .
Exercise 2.5.10 (not important). Show that the prestack Mod parameterizing
pairs (T, F ) where T is a scheme and F is a sheaf of OT -modules is not a stack over
Schét .
Exercise 2.5.11 (not easy but also not so important). Define the prestack of
sheaves over any site and apply Exercises 2.3.9 and 2.3.10 to conclude that it is a
stack.
Example 2.5.12 (Stack of schemes). Define Schemes as the prestack over Sch
consisting of morphisms X → S of schemes where a morphism (X ′ → S ′ ) → (X → S)
consists of morphisms X ′ → X and S ′ → S that forms a cartesian diagram. The
projection Schemes → Sch takes X → S to S. Since schemes glue in the Zariski
topology [Har77, Exc. II.2.12], Schemes is a stack over SchZar . However, Schemes
is not a stack over Schét ; see Example 2.1.16. Schemes can be glued to algebraic
spaces in the étale topology and there is a stack of algebraic spaces over Schét ; see
Exercise 4.5.15.
On the other hand, the subcategories ClSubSch (resp., OpenSubSch, Aff, QAff,
SepLQFin) parameterizing morphisms X → S which are closed immersions (resp.,
open immersions, affine, quasi-affine, locally quasi-finite and separated) are stacks
over Schét by Fppf Descent for Schemes (2.1.11, 2.1.12, and 2.1.14).

2.5.3 Classifying stacks and quotient stacks

Let G → S be a smooth affine group scheme acting on an S-scheme U , and let
[U/G] be the category over Sch/S defined in Definition 2.4.15: an object over an
S-scheme T is a diagram

P
f
/U


T

where P → T is a principal G-bundle and f : P → U is a G-equivariant morphism

of schemes.
Proposition 2.5.13. If G → S be a smooth affine group scheme acting on an
S-scheme U , then [U/G] is a stack over Schét . In particular, the classifying stack
BG = [S/G] is a stack over Schét .
Proof. We will show in fact that [U/G] is a stack over Schfpqc . Since schemes
and their morphisms glue in the Zariski topology, it suffices by Exercise 2.5.5 to
verify Axioms (1) and (2) with respect to an fpqc map T ′ → T . For Axiom (1),
let (P ′ → T, f ′ : P → U ) and (P → T, f : P → U ) be objects over an S-scheme
T , T ′ → T be an fpqc map, and ϕ′ : PT′ ′ → P be a G-equivariant morphism
∼
inducing an isomorphism PT′ ′ → PT ′ and compatible with f and f ′ . By Fpqc
Descent for Morphisms (2.1.7), there exists a unique morphism P ′ → P compatible

99
with f ′ and f . By Fpqc Local Properties on the Target (2.1.26), P ′ → P is an
isomorphism. For Axiom (2), let (P ′ → T ′ , f ′ : P ′ → U ) be an object over T ′ and
α : p∗1 P ′ → p∗2 P ′ be an isomorphism commuting with f ′ and satisfying the cocycle
condition p∗23 α ◦ p∗12 α = p∗13 α on T ′ ×T T ′ ×T T ′ . The existence of a principal
G-bundle P → T pulling back to P ′ → T follows from Fpqc Descent of Principal
G-bundles (2.1.17). The existence of a G-equivariant morphism P → U follows from
Fpqc Descent for Morphisms (2.1.7).

2.5.4 Moduli stack of curves

Let Mg denote the prestack of families of smooth curves C → S of genus g as defined
in Example 2.4.10.

Proposition 2.5.14 (Moduli stack of smooth curves). If g ≥ 2, then Mg is a stack

over Schét .

Proof. We check that Mg is a stack over Schfpqc . As smooth curves and their
morphisms glue in the Zariski topology, it suffices by Exercise 2.5.5 to verify Axioms
(1) and (2) with respect to an fpqc map S ′ → S. Axiom (1) translates to: for families
of smooth curves C → S and D → S of genus g, every commutative diagram

f′
*
/ / C 99K
f
CS ′ ×S S ′ / CS ′ D

□ □
 / ′  
S ′ ×S S ′ /S /S

of solid arrows can be uniquely filled in. The existence and uniqueness of f follow
from Fpqc Descent for Morphisms (2.1.7). Since the base change CS ′ → DS ′ is
an isomorphism (because of how we’ve defined morphisms in Mg ), the morphism
f : C → D is also an isomorphism by étale descent (Proposition 2.1.26).
Axiom (2) will require some geometry: we must show that given a diagram
)
α
p∗1 C ′ −
→ p∗2 C ′ / C′ /C

□ □
  p1
/ ′ 
S ′ ×S S ′ /S / S,
p2

where C ′ → S ′ is a family of smooth curves and α : p∗1 C ′ → p∗2 C ′ is an isomorphism

satisfying the cocycle condition p∗23 α ◦ p∗12 α = p∗13 α on S ′ ×S S ′ ×S S ′ , there is family
of smooth curves C → S and an isomorphism ϕ : C|S ′ → C ′ such that p∗1 ϕ = p∗2 ϕ ◦ α.
We will apply Properties of Families of Smooth Curves (5.1.16): Ω⊗3 C ′ /S ′ is relatively
⊗3
very ample on S (as g ≥ 2) and E := π∗ (ΩC ′ /S ′ ) is a vector bundle on S ′ of rank
′

5(g − 1) whose construction commutes with base change. This implies that Ω⊗3 C ′ /S ′
defines a closed immersion C ′ ,→ P(E ′ ) over S ′ . The isomorphism α induces an
isomorphism β : p∗1 E ′ → p∗2 E ′ satisfying the cocycle condition p∗23 β ◦ p∗12 β = p∗13 β
on S ′ ×S S ′ ×S S ′ . Fpqc Descent for Quasi-Coherent Sheaves (2.1.4) yields a quasi-
coherent sheaf E on S and isomorphisms ψ : E ′ → E|S ′ such that p∗1 ψ = p∗2 ψ ◦ β. It

100
follows from Fpqc Descent of Properties of Quasi-Coherent Sheaves (2.1.18) that E
is a vector bundle. Pictorially, we have

P(E ′ ) / P(E)
= ;

( .
α
p∗1 C ′ −
→ p∗2 C ′ / C′ / C-

  p1
/  ′  
S ′ ×S S ′ /S / S,
p2

Under the identifications of P(p∗1 E ′ ) and P(p∗2 E ′ ) with P(E) ×S (S ′ ×S S ′ ), the

preimages p∗1 C ′ and p∗2 C ′ are equal. By Fpqc Descent for Closed Subschemes (2.1.11),
there is a closed subscheme C ⊆ P(E) pulling back to C ′ . Smoothness and properness
are Fpqc Local Properties on the Target (2.1.26), and thus C → S is smooth and
proper. Every geometric fiber of C → S is identified with a geometric fiber of C ′ → S ′
and is thus a connected genus g curve.

Exercise 2.5.15 (Moduli of genus 0 and elliptic curves, good practice).

(a) Use the correspondence between families of genus 0 curves and principal PGL2 -
torsors (Exercise B.1.67) to show that the prestack M0 is a stack on Schét
isomorphic to B PGL2 over Spec Z.
(b) A family of elliptic curves over a scheme S is a pair (E → S, σ) where E → S
is smooth proper morphism with a section σ : S → E such that for every s ∈ S,
the fiber (Es , σ(s)) is an elliptic curve over the residue field κ(s). Show that
the moduli stack M1,1 , whose objects are families of elliptic curves, is a stack
on Schét .

Remark 2.5.16 (Moduli of genus 1 curves). The prestack M1 , whose objects are
smooth and proper morphisms C → S of schemes whose geometric fibers are
connected genus 1 curves, is not a stack over Schét . Unlike Mg with g ≥ 2 and
M1,1 , there is no natural line bundle defining an embedding into projective space
that we can use as above to verify Axiom (2). Raynaud constructed an étale cover
S ′ → S and a family C ′ → S ′ of smooth genus 1 curves which does not descend to a
family C → S [Ray70, XIII 3.2]. However, similar to Example 2.5.12, if we redefine
M1 as the category of smooth and proper morphisms C → S from an algebraic space
such that every geometric fiber is a connected curve of genus 1, then M1 is a stack.

2.5.5 Moduli stack of coherent sheaves and vector bundles

If X is a scheme over a field k, the prestacks QCoh(X), Coh(X), and Bun(X)
introduced in Example 2.4.12 parameterizes pairs (T, F ) where T is a scheme and F
is a quasi-coherent, coherent sheaf (more precisely, a finitely presented quasi-coherent
sheaf), or a vector bundle on XT = X ×k T . Note that the prestacks QCoh, Coh,
and Bun of Example 2.5.9 parameterizes pairs (T, F ) where T is a scheme and F is
a quasi-coherent sheaf on T .

Proposition 2.5.17. The prestacks QCoh(X), Coh(X), and Bun(X) are stacks
over (Sch/k)ét .

101
Proof. Just as in the proof of Example 2.5.9, the statement follows directly from
Fpqc Descent for Quasi-Coherent Sheaves (2.1.4) and Fpqc Descent of Properties of
Quasi-Coherent Sheaves (2.1.18).

2.5.6 Stackification
Recall from Sheafification (2.3.15) that for any a presheaf F on a site S, there is a
map F → F sh which is a left adjoint to the inclusion, i.e., Mor(F sh , G) → Mor(F, G)
is bijective for every sheaf G on S. Similarly, there is a stackification X → X st of a
prestack X over S.
Theorem 2.5.18 (Stackification). If X is a prestack over a site S, there exists a
stack X st , which we call the stackification, and a morphism X → X st of prestacks
such that for every stack Y over S, the induced functor
Mor(X st , Y) → Mor(X , Y) (2.5.19)
is an equivalence of categories.
Proof. As in the construction of the sheafification in Theorem 2.3.15, we construct
the stackification in stages. Most details are left to the reader.
First, given a prestack X , we can construct a prestack X st1 satisfying Axiom (1)
and a morphism X → X st1 of prestacks such that
Mor(X st1 , Y) → Mor(X , Y)
is an equivalence for all prestacks Y satisfying Axiom (1). The objects of X st1 are
the same as X , and for objects a, b ∈ X over S, T ∈ S, the set of morphisms a → b in
X st1 over a given morphism f : S → T is the global sections Γ(S, IsomX (S) (a, f ∗ b)sh )
of the sheafification of the Isom presheaf introduced in Exercise 2.4.39.
Second, given a prestack X satisfying Axiom (1), we construct a stack X and a
morphism X → X st2 of prestacks such that (2.5.19) is an equivalence for all stacks Y.
An object of X st2 over S ∈ S is given by a triple consisting of a covering {Si → S},
objects ai of X over Si , and isomorphisms αij : ai |Sij → aj |Sij satisfying the cocycle
condition αjk |Sijk ◦ αij |Sijk = αik |Sijk on Sijk . Morphisms



{Si → S}, {ai }, {αij } → {Tµ → T }, {bµ }, {βµν }
in X st2 over S → T are defined as follows: for a choice of pullbacks ai |Si ×S Tµ and
bµ |Si ×S Tµ with respect to the induced cover {Si ×S Tµ → S}i,µ , a morphism is the
data of maps Ψiµ : ai |Si ×S Tµ → bµ |Si ×S Tµ for all i, µ which are compatible with αij
and βµν (i.e., Ψjν ◦ αij = βµν ◦ Ψiµ on Sij ×S Tµν ).
Exercise 2.5.20 (details). Show that stackification commutes with fiber products:
if X → Y and Z → Y are morphisms of prestacks, then (X ×Y Z)st ∼ = X st ×Y st Z st .
Exercise 2.5.21 (good practice). Let G → S be a smooth affine group scheme
acting on a scheme U over S. Recall from Definition 2.4.15 that the quotient prestack
[U/G]pre and quotient stack [U/G] denote the prestacks over Sch/S classifying trivial
principal G-bundles (resp., principal G-bundles) P → T and G-equivariant maps
P → U.
(a) Show that [U/G]pre satisfies Axiom (1) of a stack over (Sch/S)ét .
(b) Show that the [U/G] is isomorphic to the stackification of [U/G]pre over
(Sch/S)ét , and that [U/G]pre → [U/G] is fully faithful.
Exercise 2.5.22. Extending Exercise 2.4.27, show that U → [U/G] is a categorical
quotient among stacks.

102
Chapter 3

Algebraic spaces and stacks

The notion of stacks came up in the sixties. But to swallow schemes was
already enough for one generation of mathematicians.

Gerd Faltings [Sto95, p. 45]

3.1 Definitions of algebraic spaces and stacks

The most abstract definition, once you are familiar with it, is not abstract
anymore. It’s like a beautiful mountain that you see very well, because
the air is very clear and there is light that lets you see all the details.

Claire Voisin

What are algebraic spaces, Deligne–Mumford stacks, and algebraic stacks? After
giving their definitions, we will verify the algebraicity of quotient stacks [U/G], the
moduli stack of curves Mg , and the moduli stack of vector bundles Bun r,d (C).

3.1.1 Algebraic spaces

Definition 3.1.1 (Morphisms representable by schemes). A morphism X → Y of
prestacks (or presheaves) over Sch is representable by schemes (or sometimes called
schematic) if for every morphism T → Y from a scheme, the fiber product X ×Y T
is a scheme.
If P is a property of morphisms of schemes stable under base change (e.g.,
surjective or étale), a morphism X → Y of prestacks representable by schemes has
property P if for every morphism T → Y from a scheme, the morphism X ×Y T → T
of schemes has property P.

Definition 3.1.2 (Algebraic spaces). An algebraic space is a sheaf X on Schét such

that there exists a scheme U and a surjective étale morphism U → X representable
by schemes.

The map U → X is called an étale presentation. Morphisms of algebraic spaces

are by definition morphisms of sheaves. Every scheme is an algebraic space.

103
3.1.2 Deligne–Mumford stacks
Definition 3.1.3 (Representable morphisms and their properties). A morphism
X → Y of prestacks (or presheaves) over Sch is representable if for every morphism
T → Y from a scheme T , the fiber product X ×Y T is an algebraic space.
Let P be a property of morphisms of schemes stable under base change and étale-
local on the source (i.e., if X ′ → X is a surjective étale morphism, then a morphism
X → Y of schemes has P if and only if X ′ → X → Y has P); examples include
the properties of being surjective, étale, or smooth. We say that a representable
morphism X → Y of prestacks has property P if for every morphism T → Y
from a scheme and étale presentation U → X ×Y T by a scheme, the composition
U → X ×Y T → T has property P.
Definition 3.1.4 (Deligne–Mumford stacks). A Deligne–Mumford stack is a stack X
over Schét such that there exists a scheme U and a surjective, étale, and representable
morphism U → X .
The morphism U → X is called an étale presentation. Morphisms of Deligne–
Mumford stacks are by definition morphisms of stacks. Every algebraic space is a
Deligne–Mumford stack via Example 2.4.7. We show later that a Deligne–Mumford
stack (or even an algebraic stack) that is a sheaf is an algebraic space (Theorems 3.6.6
and 4.5.10).
Remark 3.1.5. While the essential difference between an algebraic space and a
Deligne–Mumford stack is that one is a sheaf while the other is a stack, there
is also the technical difference that an étale presentation of an algebraic space is
representable by schemes while an étale presentation of a Deligne–Mumford stack
is only required to be representable. If the diagonal of a Deligne–Mumford stack
is separated and quasi-compact, then it is representable by schemes and every
presentation U → X is representable by schemes (Corollary 4.5.8). While requiring
that the diagonal be separated and quasi-compact is a very mild condition and
satisfied in almost every moduli problem of interest, you should be aware that there
are Deligne–Mumford stacks whose diagonal is not quasi-compact, not separated, or
not representable by schemes, e.g., BG for an étale group algebraic space that is
not quasi-compact, not separated, or not a scheme; see Examples 3.9.35 and 3.9.37.

3.1.3 Algebraic stacks

Definition 3.1.6 (Algebraic stacks). An algebraic stack is a stack X over Schét such
that there exists a scheme U and a surjective, smooth, and representable morphism
U → X.
The morphism U → X is called a smooth presentation. Morphisms of algebraic
stacks are by definition morphisms of prestacks. Every scheme, algebraic space, or
Deligne–Mumford stack is also an algebraic stack.
Caution 3.1.7. In the literature, most authors add a representability condition on
the diagonal. However, as we show in Theorem 3.2.1, the existence of a smooth
presentation implies the representability of the diagonal: the diagonal of an algebraic
space is representable by schemes and that the diagonal of an algebraic stack is
representable.
Definition 3.1.8 (Open and closed substacks). A substack T ⊆ X of a stack over
Schét is called an open substack (resp., closed substack ) if the inclusion T → X is
representable by schemes and an open immersion (resp., closed immersion).

104
Exercise 3.1.9 (Fiber products). Show that fiber products exist for algebraic
spaces, Deligne–Mumford stacks, and algebraic stacks.

3.1.4 Algebraicity of quotient stacks

Theorem 3.1.10 (Algebraicity of Quotient Stacks). If G → S is a smooth affine
group scheme acting on an algebraic space U → S, the quotient stack [U/G] is an
algebraic stack over S such that U → [U/G] is a principal G-bundle and in particular
surjective, smooth, and affine. In particular, the classifying stack BG = [S/G] is
algebraic.
Remark 3.1.11. An eagle-eyed reader may have noticed that we only defined [U/G]
when U is a scheme. It is not hard to extend the definition. An action of a
smooth affine group scheme G → S on an algebraic space U → S is a morphism
σ : G ×S U → U satisfying the same axioms as in Definition B.1.9, and we define
the quotient stack [U/G] as the stackification of the prestack [U/G]pre , whose
fiber category over an S-scheme T is the quotient groupoid [U (T )/G(T )], as in
Definition 2.4.15. Objects of [U/G] over an S-scheme T are principal G-bundles
P → T and G-equivariant morphisms P → U . Since morphisms to algebraic spaces
glue uniquely in the étale topology, the argument of Proposition 2.5.13 extends
to show that [U/G] is a stack. Finally, we note that saying that U → [U/G] is a
principal G-bundle means that it is representable by schemes and every base change
by a map T → [U/G] from scheme is a principal G-bundle.
Proof. We will use the natural projection U → [U/G] corresponding via the 2-
Yoneda Lemma (2.4.21) to the trivial principal G-bundle p2 : G × U → U and the
G-equivariant map σ : G × U → U given by multiplication. To show that U → [U/G]
is a principal G-bundle, let T → [U/G] be a morphism from an S-scheme classified by
a principal G-bundle P → T and a G-equivariant map P → U . By Exercise 2.4.37,
there is a cartesian diagram
P /U

□
 
T / [U/G].

Since every base change is a principal G-bundle, so is U → [U/G]. The map

U → [U/G] is a smooth presentation if U is a scheme. In general, letting U ′ → U be
an étale presentation with U ′ a scheme, the composition U ′ → U → [U/G] provides
a smooth presentation.
Example 3.1.12 (BGm , BGLn , and B PGLn ). Since principal Gm -bundles cor-
respond to line bundles (Exercise B.1.51), the classifying stack BGm is equivalent
to the category of pairs (S, V ) consisting of a scheme S and a line bundle V on
S. Similarly, BGLn is the stack of pairs (S, V ) where V is a vector bundle of
rank n on a scheme S (Exercise B.1.56). The classifying stack B PGLn can be
described equivalently using either principal PGLn -bundles, Brauer–Severi schemes,
or Azumaya algebras over S (see Exercises B.1.66 and B.1.68).
Exercise 3.1.13 (BO(q)). Let k be a field of char(k) ̸= 2. For a non-degenerate
quadratic form q on an n-dimensional vector space V , the orthogonal group O(q)
is the subgroup of GL(V ) containing matrices preserving q. If q and q ′ are non-
degenerate quadratic forms, show that BO(q) ∼
= BO(q ′ ) even though O(q) and O(q ′ )
may be non-isomorphic.

105
 We obtain the desirable consequence that the category of algebraic spaces is closed
under taking quotients of free actions of finite groups. Recall from Example 0.5.7
that the quotient of a free action by a finite group on a scheme need not exist as a
scheme.
Corollary 3.1.14. Let G be an finite abstract group viewed as a group scheme over
a scheme S. If G acts freely on an algebraic space U over S, then the quotient sheaf
U/G is an algebraic space.
Proof. Since the action is free, the quotient stack [U/G] is equivalent to a sheaf,
which we denote by U/G (see Exercise 2.4.20). Theorem 3.1.10 implies that U/G
is an algebraic stack and that U → U/G is a principal G-bundle so in particular
finite, étale, surjective and representable by schemes. Taking U ′ → U to be an
étale presentation by a scheme, the composition U ′ → U → U/G yields an étale
presentation of U/G.
Remark 3.1.15. In addition, it shows that the category of algebraic spaces itself is
closed under taking quotients by free actions of finite groups so that we do not need
to enlarge our category even further.
Exercise 3.1.16 (easy). Let G → S be a smooth affine group scheme acting on
S-schemes X and Y . Show that a G-equivariant morphism X → Y induces a
morphism [X/G] → [Y /G] of algebraic stacks, and conversely that [X/G] → [Y /G]
is induced by a G-equivariant morphism if and only if [X/G] → [Y /G] is a morphism
over BG.

3.1.5 Algebraicity of Mg
Why is Mg algebraic? Here is one reason: every smooth, connected, and projective
curve C is tricanonically embedded C ,→ P5g−6 by the very ample line bundle Ω⊗3 C
and the locally closed subscheme H ′ ⊆ HilbP (P5g−6 ) parameterizing smooth families
of tricanonically embedded curves provides a smooth presentation H ′ → Mg . The
technical details however are a bit involved. The Algebraicity of Bun r,d (C) (3.1.21)
will be easier.
Theorem 3.1.17 (Algebraicity of Mg ). If g ≥ 2, then Mg is an algebraic stack
over Spec Z. Moreover, Mg ∼ = [H ′ / PGL5g−5 ] where H ′ is a locally closed subscheme
of a projective Hilbert scheme.
Proof. As in the proof that Mg is a stack (Proposition 2.5.14), we will use Properties
of Families of Smooth Curves (5.1.16): for a family of smooth curves p : D → S,
Ω⊗3 ⊗3
D/S is relatively very ample on S and p∗ (ΩD/S ) is a vector bundle of rank 5(g − 1).
It follows that Ω⊗3 ⊗3
D/S defines a closed immersion D ,→ P(p∗ (ΩD/S )) over S. By
Riemann–Roch (5.1.2), the Hilbert polynomial of a fiber Ds ,→ P5g−6
κ(s) is given by

P (n) := χ(ODs (n)) = deg(Ω⊗3n

Ds ) + 1 − g = (6n − 1)(g − 1).

and we define
H := HilbP (P5g−6
Z )
as the (projective) Hilbert scheme parameterizing closed subschemes of P5g−6 with
Hilbert polynomial P (Theorem 1.1.2). Let C ,→ P5g−6 × H be the universal closed
subscheme and let π : C → H be the projection. We claim that there is a locally
closed subscheme H ′ ⊂ H such that the family CH ′ → H ′ satisfies

106
 (a) for every h ∈ H ′ , Ch → Spec κ(h) is smooth and geometrically connected,
(b) the natural map p2,∗ OP5g−6 ×H ′ (1) → p2,∗ OCH ′ (1) induced by the closed immer-
sion CH ′ ,→ P5g−6 × H ′ is an isomorphism (or equivalently H0 (P5g−6
κ(h) , O(1)) →
H0 (Ch , OCh (1)) is an isomorphism for all h ∈ H ′ ), and
(c) the line bundles Ω⊗3
CH ′ /H ′ and OCH ′ (1) differ by a pullback of a line bundle
from H .′

Moreover, if T → H is a morphism of schemes such that (a)–(c) hold for the family
CT → T , then T → H factors through H ′ . In particular, H ′ ⊆ H is unique. Note
that (a)–(c) imply that Ch ,→ P5g−6
κ(h) is embedded by the complete linear series
⊗3
ΩCh /κ(h) for h ∈ H.

Since the condition that a fiber of a proper morphism (of finite presentation)
is smooth is an open condition on the target (Corollary A.3.9), the condition on
H that Ch is smooth is open. Consider the Stein factorization [Har77, Cor. 11.5]
C → H e = SpecH π∗ OC → H, where C → H e has geometrically connected fibers
and H e → H is finite. Since the kernel and cokernel of OH → π∗ OC have closed
support, H e → H is an isomorphism over an open subscheme of H, which is precisely
where the fibers of C → H are geometrically connected. In summary, the set of
h ∈ H satisfying (a) is an open subscheme of H, which we will denote by H1 . To
arrange (b), observe that Cohomology and Base Change (A.6.8) implies that both
p2,∗ OP5g−6 ×H1 (1) and p2,∗ OCH1 (1) are vector bundles whose constructions commute
with base change, and Riemann–Roch (5.1.2) further implies they have the same
rank. Therefore, α : p2,∗ OP5g−6 ×H1 (1) → p2,∗ OCH1 (1) is an isomorphism if and only
if it is surjective. Setting H2 := H1 ∖ Supp(coker(α)), we conclude that (a)–(b)
hold over H2 . It is also clear that any map T → H over which (a)–(b) hold factors
through H2 . To arrange property (c), as CH2 → H2 has geometrically integral
fibers and the relative canonical sheaf Ω := ΩCH2 /H2 is a line bundle, we may use
Proposition A.6.17 to conclude that there exists a closed subscheme H3 ,→ H2 such
that a morphism T → H2 factors through H3 if and only if Ω⊗3 |CT and OC (1)|CT
differ by the pullback of a line bundle on T . The subscheme H ′ := H3 satisfies
(a)–(c) along with the universal property.

The group scheme PGL5g−5 = Aut(P5g−6 Z ) over Z acts naturally on H =

HilbP (P5g−6
Z ): if g ∈ PGL 5g−5 and [D ⊆ P 5g−6
S ] ∈ H(S), then g · [D ⊆ P5g−6
S ]=
5g−6
[g(D) ⊆ PS ]. The closed subscheme H ⊆ H is PGL5g−5 -invariant. We claim
′

that Mg ∼ = [H ′ / PGL5g−5 ]. This claim finishes the theorem by the Algebraicity of

Quotient Stacks (3.1.10). To see the claim, consider the morphism H ′ → Mg defined
by the restriction CH ′ → H ′ of the universal family of the Hilbert scheme. This
morphism forgets the embedding, i.e., assigns a closed subscheme D ⊆ P5g−6 S to the
family D → S. This morphism is PGL5g−5 -invariant and descends to a morphism
[H ′ / PGL5g−5 ]pre → Mg of prestacks. We first show that this map is fully faithful:
let S → H ′ be a map corresponding to a closed subscheme


D / P5g−6 × S
p2
q
$ 
S.

107
We will exploit the equivalences

H0 (P5g−6
Z , O(1)) ⊗ OS ∼
= p2,∗ OP5g−6 ×S (1)
∼
= q∗ OD (1) (property (c))
∼ q∗ (Ω⊗3 ∗
(property (b) for M ∈ Pic(S))
= D/S ⊗ q M )
∼ q∗ (Ω⊗3 (projection formula).
= D/S ) ⊗ M

An automorphism of D → S induces a unique automorphism of Ω⊗3 D/S and thus an

⊗3
unique automorphism of q∗ (ΩD/S )⊗M , which in turn induces a unique automorphism
of P5g−6 × S preserving D. This shows that the natural map StabPGL5g−5 ([D ⊆
∼
P5g−6
S ]) → Aut(D → S) is bijective.
Since Mg is a stack (Proposition 2.5.14), the universal property of stackifi-
cation yields a morphism [H ′ / PGL5g−5 ] → Mg ; this map is fully faithful since
[H ′ / PGL5g−5 ]pre → [H ′ / PGL5g−5 ] is fully faithful (Exercise 2.5.21). It remains to
check that [H ′ / PGL5g−5 ] → Mg is essentially surjective. For this, it suffices to
check that if q : D → S is a family of smooth curves, there exists an étale cover
{Si → S} such that each DSi is in the image of H ′ (Si ) → Mg (Si ). Since Ω⊗3 D/S
defines a closed immersion D ,→ P(q∗ (Ω⊗3 D/S )) over S and q∗ (Ω ⊗3
D/S ) is locally free of
rank 5g −5, we may simply take {Si } to be a Zariski open cover (and thus étale cover)
such that each (q∗ (ΩD/S )⊗3 )|Si is free. We conclude that Mg ∼ = [H ′ / PGL5g−5 ] and
Theorem 3.1.10 therefore implies that Mg is algebraic.

Remark 3.1.18. The `entire stack M of smooth curves (Example 2.4.10) is also
algebraic since M = g Mg .

Exercise 3.1.19 (Moduli of elliptic curves). Recall from Exercise 2.5.15 that M1,1
denotes the stack over Schét parameterizing families of elliptic curves.
(a) Show that M1,1 is an algebraic stack over Z.
(b) Use the Weierstrass form y 2 = x3 + ax + b (see [Sil09, §3.1]) to show that if
we invert the primes 2 and 3, there is an isomorphism

M1,1 ×Z Z[1/6] ∼
= [(A2 ∖ V (∆))/Gm ],

where the action is given by t · (a, b) = (t4 a, t6 b) and ∆ is the discriminant

4a3 + 27b2 . See also [MS72, §3, Thm. 1].
(c) Define a stable elliptic curve over a field k as a pair (E, p) where E is an
irreducible projective curve over k of arithmetic genus 1 with at worst nodal
singularities and p ∈ E(k) is a smooth point. Over a scheme S, a family of
stable elliptic curves over S is a proper, flat, and finitely presented morphism
E → S together with a section σ : S → E such that every fiber is a stable
elliptic curve. Denoting M1,1 as the stack over Sch classifying stable elliptic
curves, show that
M1,1 ×Z Z[1/6] ∼ = [(A2 ∖ 0)/Gm ]
with the same action as above.

Exercise 3.1.20. An n-pointed family of genus 0 smooth curves is a smooth, proper

morphism X → S of schemes with n sections σ1 , . . . , σn : S → X such that for every
s : Spec k → S, Xs is a genus 0 curve with and σ1 (s), . . . , σn (s) ∈ Xs are distinct.
In Exercise 2.5.15, we have identified M0,0 with the classifying stack B PGL2 .

108
 (a) Show that the prestack M0,n parameterizing n-pointed families of genus 0
curves is a stack over Schét .
(b) Show that M0,1 ∼ = BU2 , where U2 ⊆ PGL2 is the two-dimensional subgroup
of upper triangular matrices.
(c) Show that M0,2 = ∼ BGm .
(d) Show that M0,3 ∼= Spec Z.
(e) Show that for n > 3, M0,n ∼
= (P1 ∖ {0, 1, ∞})n−3 ∖ ∆, where ∆ is the closed
subscheme where at least two of the n − 3 points are equal.

3.1.6 Algebraicity of Bun(C)

In Proposition 2.5.17, we showed that QCoh(X), Coh(X), and Bun(X) are stacks
over (Sch/k)ét for a scheme X over a field k. We now specialize to the case of
a curve C, even though the following result holds in far greater generality (see
Exercise 3.1.23). We define

Cohr,d (C) ⊆ Coh(C) and Bun r,d (C) ⊂ Bun(C)

as the full subcategories parameterizing pairs (E, S) such that for every geometric
point Spec K → S, EK is a coherent sheaf on CK of rank r and degree d.
Theorem 3.1.21 (Algebraicity of Bun(C)). If C is a smooth, connected, and
projective curve over an algebraically closed field k, then the stacks Bun(C) and
Coh(C) are algebraic, and Bun(C) ⊆ Coh(C) is an open substack. For integers r ≥ 0
and d, Bun r,d (C) and Cohr,d (C) are algebraic stacks, and Bun r,d (C) ⊆ Bun(C) and
Cohr,d (C) ⊆ Coh(C) are open and closed substacks.
Remark 3.1.22. The theorem yields decompositions
a a
Coh(C) = Cohr,d (C) and Bun(C) = Bun r,d (C).
r,d r,d

While Bun r,d (C) and Cohr,d (C) are not quasi-compact (Definition 3.3.21), the proof
below shows that every quasi-compact open substack of Bun r,d (C) or Cohr,d (C) is
a quotient stack of a quasi-projective scheme (in fact, an open subscheme of a Quot
scheme) by GLN .
Proof. We first note that Cohr,d (C) and Bun r,d (C) are stacks over (Sch/k)ét since
they are defined as the full subcategories of the stacks Coh(C) and Bun(C) by a
condition on the geometric fiber. To see that Bun(C) ⊆ Coh(C) is an open substack,
let S → Coh(C) be a morphism classified by a finitely presented, quasi-coherent
sheaf E on C × S flat over S. If V ⊆ C × S is the open locus where E is a vector
bundle, then S ∖ p2 (C × S ∖ V ) is open and identified with the fiber product
S ×Coh(C) Bun(C). By Riemann–Roch for Coherent Sheaves (8.1.2), the Hilbert
polynomial of a coherent sheaf E on C of rank r and degree d is

P (n) := χ(E(n)) = deg(E(n)) + rk(E(n))(1 − g) = d + rn + r(1 − g).

Since the Hilbert polynomial is locally constant in flat families (Proposition A.2.4),
the inclusions Bun r,d (C) ,→ Bun(C) and Cohr,d (C) ,→ Coh(C) are open and closed
substacks. It therefore suffices to show that Cohr,d (C) is algebraic.
If E is a coherent sheaf C of rank r and degree d, then Serre vanishing implies
that E(N ) is globally generated and H1 (C, E(N )) = 0 for N ≫ 0. This yields a

109
surjection Γ(C, E(N )) ⊗k OC ↠ E(N ) inducing an isomorphism on global sections.
For each integer N , let
UN ⊆ Cohr,d (C)
be the substack parameterizing pairs (S, E) ∈ Cohr,d (C) such that for every s ∈
S, Es (N ) := E(N )|C×Spec κ(s) is globally generated on Cs := C ×k κ(s) and
H1 (Cs , Es (N )) = 0. Note that since P (N ) = h0 (Cs , Es (N )) − h1 (Cs , Es (N )), these
conditions imply that h0 (C, Es (N )) = P (N ) and that the surjection Γ(C, E(N )) ⊗
OC ↠ E(N ) induces an isomorphism on global sections.
We claim that UN ⊆ Cohr,d (C) is an open substack. To verify the claim, let S
be a scheme and E be a finitely presented, quasi-coherent sheaf on C × S flat over S
with Hilbert polynomial P , and consider the diagram

C ×S
p1 p2

| #
C S.

A simple application of Cohomology and Base Change (see Proposition A.6.11)

implies that the locus S ′ ⊆ S of points
s ∈ S such that H1 (C,
Es (N )) = 0
is open, and moreover that R1 p2,∗ E(N ) |S ′ = 0 and p2,∗ E(N ) |S ′ is a vec-
tor bundle of rank P (N ) whose
construction commutes base change. The sheaf
F := coker p∗2 p2,∗ E(N ) → E(N ) has closed support and the open subset S ′′ :=
S ′ ∖ p2 (Supp(F)) ⊆ S is the locus of points s ∈ S such that Es (N ) is globally
generated and H1 (C, Es (N )) = 0. If S → Cohr,d (C) is the map classifying E, the
base change UN ×Cohr,d (C) S is identified with S ′′ , and the claim is established.
For each N , consider the Quot scheme

QN := QuotP (OC (−N )⊕P (N ) /C/k)

parameterizing quotients OC (−N )⊕P (N ) ↠ F with Hilbert polynomial P (The-

orem 1.1.3). A similar argument as above shows that there is an open sub-
scheme Q′N ⊆ QN parameterizing quotients q : OC (−N )⊕P (N ) ↠ F such that
H0 (q(N )) : H0 (C, OC )⊕P (N ) → H0 (C, F (N )) is surjective and H1 (C, F (N )) = 0.
The Quot scheme QN inherits a natural action from GLP (N ) : given g ∈ GLP (N )
and [q : OC (−N )⊕P (N ) ↠ F ] ∈ QN ,

g −1 q
g · q := [OC (−N )⊕P (N ) −−→ OC (−N )⊕P (N ) −
→ F ] ∈ QN .

The locus Q′N ⊆ QN is clearly GLP (N ) -invariant. The morphism

Q′N → UN , [OC (−N )⊕P (N ) ↠ F ] 7→ F,

is GLP (N ) -invariant (i.e., GLP (N ) -equivariant with respect to the trivial action on
UN ) and thus descends to a morphism Ψpre : [Q′N / GLP (N ) ]pre → UN of prestacks. To
see that Ψpre is fully faithful, consider a quotient [q : OC (−N )⊕P (N ) ↠ F ] ∈ QN (k)
and the induced map StabGLP (N ) (q) → Aut(F ). An automorphism α ∈ Aut(F )
induces an automorphism of H0 (C, F (N )) ∼ = kP (N ) fixing the kernel of H0 (q(N )).
This defines an inverse to Stab(q) → Aut(F ), and it is not hard to extend this
construction to families of quotients.
Since Cohr,d (C) is a stack (Proposition 2.5.17), so is UN , and there is an in-
duced morphism Ψ : [Q′N / GLP (N ) ] → UN of stacks which is fully faithful (by

110
Exercise 2.5.21) and essentially surjective (by construction). We conclude that
UN = [Q′N / GLP (m ] and that
[
Q′N / GLP (N ) .

Cohr,d (C) =
N

The algebraicity of quotient stacks (Theorem 3.1.10) implies the algebraicity of

Cohr,d (C).
Exercise 3.1.23.
(1) (easy) Modify the above argument to show that Bun(X) and Coh(X) are
algebraic stacks if X is a projective scheme over a field k of arbitrary dimension.
(2) (moderate) More generally, show that if X → S is a strongly projective (see
§1.4.4) morphism of noetherian schemes, then the stack Coh(X/S), whose
objects over an S-scheme T are finitely presented, quasi-coherent sheaves on
XT flat over T , is an algebraic stack. (See also [LMB00, Thm. [Link]].)
See also Theorem C.7.7 for a proof using Artin’s Axioms.
Exercise 3.1.24 (hard). Let X be a projective scheme over a field k and G be an
affine algebraic group over k. Define a stack Bun G (X) of principal G-bundles over
X, and prove that it is algebraic.
Hint: Choose a faithful embedding G ,→ GLn and show that inducing principal
bundles defines a representable morphism Bun G (X) → Bun GLn (X) to the stack of
vector bundles on X.

3.1.7 Universal families

Lemma 3.1.25 (Generalized 2-Yoneda Lemma). Let X be a stack over Schét . If T
is an algebraic stack and U → T is a smooth presentation, define the category X (T )
as the equalizer
//
// X (U ×T U )


X (T ) := Eq X (U ) / X (U ×T U ×T U ) ,
∼
i.e., an object is the data of a pair (a, α) where a ∈ X (U ) and α : p∗1 a → p∗2 a is an
isomorphism satisfying the cocycle condition p∗23 α ◦ p∗12 α = p∗13 α, while a morphism
(a, α) → (a′ , α′ ) is the data of a morphism β : a → a′ satisfying p∗2 β ◦ α = α′ ◦ p∗1 β.
There is a natural equivalence of categories

Mor(T , X ) → X (T ). (3.1.26)

In particular, X (T ) is independent of the presentation.

Proof. Let q : U → T denote the smooth presentation, and consider the 2-commutative
diagram
U ×T U
p2
/U
p1  γ
q
 
U
q
/T.
Given a morphism f : T → X , let a ∈ X (U ) be the object corresponding via the
2-Yoneda Lemma (2.4.21) to the composition f ◦ q : U → T → X . Composing the
∼
2-isomorphism γ : q ◦ p2 → q ◦ p1 with f induces a 2-isomorphism α : p∗1 a → p∗2 a.
The map (3.1.26) is defined by taking the morphism f to the pair (a, α) ∈ X (T ).

111
 To show that (3.1.26) is an equivalence, we first assume that it holds for algebraic
spaces, and so it holds for U , U ×T U , and U ×T U ×T U . To construct an inverse
Ψ : X (T ) → Mor(T , X ) of (3.1.26), let (a, α) ∈ X (T ) and fa : U → X be the map
corresponding to a. To define a morphism Ψ(a, α) : T → X of stacks, let b ∈ T
be an object over a scheme S classified by a map fb : S → T , and consider the
commutative diagram


S2  / S1 / S0 /U /X
fa

□ q

'  fb 
/T,
S

where S0 = S ×T U is an algebraic space, S1 → S0 is an étale presentation, and

S2 → S is an etale surjection of schemes factoring through S1 , which exists because
smooth morphisms have sections étale locally (Corollary A.3.5). The composition in
∼
the top row defines an object c2 of X over S2 . The isomorphism α : p∗1 a → p∗2 a over
∼
U ×T U induces an isomorphism β : p∗1 c2 → p∗2 c2 over S2 ×S S2 . Since α satisfies the
cocycle condition and (3.1.26) is an equivalence for U ×T U ×T U , β also satisfies
the cocycle condition, and thus there is an object c of X over S pulling back to
c2 . We set Ψ(a, α)(b) = c ∈ X (S). The details of verifying that is an inverse are
left to the reader. Finally, the same argument applies to show that (3.1.26) is an
equivalence when T is an algebraic space by using an étale presentation U → T .

Definition 3.1.27 (Universal family). If X is an algebraic stack, the universal family

over X is the object u ∈ X (X ) (unique up to unique isomorphism) corresponding to
the identity morphism idX : X → X under the equivalence (3.1.26).

Exercise 3.1.28 (details). Let X be an algebraic stack and u ∈ X (X ) be the

universal family. If g : S → T is a morphism of algebraic stacks, show that there is
a natural pullback functor g ∗ : X (T ) → X (S), and that if fa : T → X is a map from
scheme classified by an object a ∈ X (T ), then a is isomorphic to the pullback fa∗ u.

In practice, for an algebraic stack X arising from a moduli problem, the geometric
significance of objects of X (S) is usually clear.

Example 3.1.29 (Universal family of Mg ). A family of smooth curves C → T

over an algebraic stack T is a morphism of algebraic stacks which is representable
by schemes, proper, flat, and finitely presented such that for every geometric point
Spec k → T , the fiber Cs := C ×S k is a smooth, connected, and projective curve of
genus g. Descent theory provides an identification of Mg (T ) with the category of
family of smooth curves over T . Let Ug → Mg be the family of curves corresponding
to the universal family u ∈ Mg (Mg ). By Exercise 2.4.42, the universal family is
identified with the map Mg,1 → Mg forgetting a section, where Mg,1 is the stack of
smooth 1-pointed curves. For every morphism S → Mg corresponding to a family
C → S of smooth curves, there is a cartesian diagram

C / Ug = Mg,1

□
 
S / Mg .

See Figure 0.3.26 for a visualization of Ug → Mg .

112
Example 3.1.30 (Universal family of Bun r,d (C)). Later we will define vector
bundles on an algebraic stack using the theory of quasi-coherent sheaves (see §6.1).
By decent theory, Bun r,d (C)(T ) is identified with the groupoid of vector bundles
on C × T of rank r and degree d. The universal family is a vector bundle Euniv on
C × Bun r,d (C) with the property that if S is a scheme S and E is a vector bundle
on C × S classified by a map f : S → Bun r,d (C), then E ∼ = (id ×f )∗ Euniv .

3.1.8 Desideratum
We will develop the foundations of algebraic spaces and stacks in the forthcoming
chapters but we first share some of the highlights.

The importance of the diagonal. When overhearing others discussing algebraic

stacks, you may have wondered what is all the fuss about the diagonal. Well, I will
tell you—the diagonal encodes the stackiness!
First and foremost, the diagonal X → X ×X of an algebraic stack is representable
and the diagonal X → X × X of an algebraic space is representable by schemes
(Theorem 3.2.1). The automorphism group or stabilizer of a field-valued point
x : Spec k → X is defined as the sheaf

Gx := AutX (k) (x) = IsomX (k) (x, x)

and is identified with the fiber product X ×X ×X Spec k by Exercise 2.4.39. By

Representability of the Diagonal (3.2.1), the stabilizer Gx is representable by a group
algebraic space over k. See §3.2.2 for a further discussion of stabilizers.
For schemes (resp., separated schemes), the diagonal is an immersion (resp.,
closed immersion). For algebraic spaces, the diagonal need not be an immersion,
and for algebraic stacks, the diagonal need not even be a monomorphism as the
fiber over (x, x) : Spec k → X × X (that is, the stabilizer Gx ), may be non-trivial.
Theorems 3.6.4 and 3.6.6 characterize algebraic spaces and Deligne–Mumford stacks
in terms of the diagonal, as expressed by the following table:

Type of space Property of the diagonal Property of stabilizers

algebraic space monomorphism trivial

Deligne–Mumford stack unramified reduced finite groups1
algebraic stack arbitrary arbitrary

Table 3.1.31: Characterization of algebraic spaces and Deligne–Mumford stacks. See

Table 0.0.1, Figure 0.0.2, and Table 0.6.5 for further diagrammatic explanations of
the trichotomy of moduli.

This characterization will allow us to conclude that Mg is Deligne–Mumford

(Corollary 3.6.10). We show later in Corollaries 3.6.9 and 4.5.11 that a morphism
f : X → Y of algebraic stacks is representable if and only if the induced map on
stabilizer groups Gx → Gf (x) is injective for every field-valued point x ∈ X (k).

Properties of algebraic spaces.

1 If the diagonal is not quasi-compact, the stabilizers will only be discrete and reduced.

113
• (Algebraicity of Quotients by Equivalence Relations) If R ⇒ U is an étale equiv-
alence relation of schemes, the quotient sheaf U/R is an algebraic space (Theo-
rem 3.4.13). This is extended to smooth equivalence relations of algebraic spaces
in Corollary 4.5.12 and fppf equivalence relations in Corollary 6.3.6. In particular,
the quotient of a free action by an algebraic group on an algebraic space exists as
an algebraic space.
• (Zariski’s Main Theorem) Generalizing Zariski’s Main Theorem for Schemes
(A.7.3), if X → Y is a quasi-finite and separated morphism of noetherian algebraic
spaces, then there exists a factorization X ,→ X e → Y where X ,→ X e is an
open immersion and X → Y is finite (Theorem 4.5.9). In particular, X → Y is
e
quasi-affine. Zariski’s Main Theorem also holds for representable morphisms of
Deligne–Mumford stacks, and is extended to representable morphisms of algebraic
stacks in Theorem 6.1.18.
• (Affine Criteria) By Serre’s Criterion for Affineness (4.5.16), an algebraic space X
is an affine scheme if and only if Γ(X, −) is exact on the category of quasi-coherent
sheaves. By Chevalley’s Criterion for Affineness (4.5.20), if X → Y is a finite
surjection of noetherian algebraic spaces and X is affine, then Y is also affine.
• (Algebraic spaces vs schemes) If X is a quasi-separated algebraic space, there
exists a dense open subspace U ⊆ X which is a scheme (Theorem 4.5.1). A
quasi-separated group algebraic space locally of finite type over a field is a
scheme (Theorem 4.5.28); in particular, the stabilizer of a field-valued point of an
algebraic stack with quasi-separated diagonal is a group scheme. A separated one
dimensional algebraic spaces is a scheme (Theorem 4.5.32).

Properties of Deligne–Mumford stacks.

• (Algebraicity of Quotients by Groupoids) If R ⇒ U is an étale groupoid of schemes,

the quotient stack [U/R] is a Deligne–Mumford stack (Theorem 3.4.13).
• (Local Structure of Deligne–Mumford Stacks) If X is a separated Deligne–Mumford
stack and x ∈ X is a finite type point with stabilizer Gx , there exists an affine
étale neighborhood [Spec A/Gx ] → X of x (Theorem 4.3.1).
• (Keel–Mori Theorem) If X is an Deligne–Mumford stack separated and of finite
type over a noetherian scheme S, there exists a coarse moduli space X → X where
X is an algebraic space separated over S (Theorem 4.4.6).
• (Le Lemme de Gabber) If X is a Deligne–Mumford stack (e.g., algebraic space)
separated and of finite type over a noetherian scheme S, there exists a scheme Z
and a finite surjection Z → X (Theorem 4.6.1).

Properties of algebraic stacks.

• (Algebraicity of Quotients by Groupoids) If R ⇒ U is a smooth groupoid of

schemes, the quotient stack [U/R] is an algebraic stack (Theorem 3.4.13). This is
extended to fppf groupoids of algebraic spaces in Corollary 6.3.6.
• (Residual Gerbes and Minimal Presentations) If X is a noetherian algebraic stack
and x ∈ |X | is a finite type point, then the residual gerbe Gx exists and Gx ,→ X
is a locally closed immersion (Proposition 3.5.16), and if in addition the stabilizer
Gx is smooth, there is a smooth presentation (U, u) → (X , x) of relative dimension
dim Gx such that Gx ×X U ∼ = Spec κ(u) (Theorem 3.6.1). Later we establish the
existence of residual gerbes for any point x ∈ |X | (Proposition 6.4.26).

114
• (Infinitesimal Lifting Criteria) We establish infinitesimal lifting criteria charac-
terizing morphisms that are smooth, étale, unramified, or that have unramified
diagonal (Theorem 3.7.1).
• (Valuative Criteria) We provide valuative criteria characterizing morphisms that
are proper, universally closed, separated, or that have separated diagonal (Theo-
rem 3.8.7).
• (Local Structure of Algebraic Stacks) If X is an algebraic stack of finite type over an
algebraically closed field k with affine diagonal, every point x ∈ X (k) with linearly
reductive stabilizer Gx has an affine étale neighborhood [Spec(A)/Gx ] → X of x
(Theorem 6.7.1).
• (Existence of Good Moduli Spaces) Let X be an algebraic stack, of finite type
over an algebraically closed field k of characteristic 0, with affine diagonal. If X is
S-complete and Θ-complete, then there exists a good moduli space X → X where
X is a separated algebraic space of finite type over k (Theorem 6.10.1).

Historical comments
Deligne–Mumford and algebraic stacks were first introduced in [DM69] and [Art74b]—
and in both cases referred to as algebraic stacks—with conventions slightly different
than ours. Namely, [DM69, Def. 4.6] assumed that algebraic stacks had an étale
presentation U → X which is representable by schemes (which as they point out
in a footnote gives the “right” definition when the diagonal is separated and quasi-
compact). On the other hand, [Art74b, Def. 5.1] assumed that algebraic stacks are
locally of finite type over an excellent Dedekind domain. The term Artin stack —
which we refrain from using—is sometimes reserved for stacks that satisfy Artin’s
Axioms (C.7.1 or C.7.4) as first established in [Art74b, Thm. 5.3].
We follow the conventions of [LMB00], [Ols16], and [SP] with the following
exceptions:

– [LMB00] imposes a separated and quasi-compact hypothesis on the diagonal,

but we do not as in [Ols16] and [SP].
– We define algebraic spaces and stacks over Schét , while [LMB00] works over
the big étale site AffSchét of affine schemes and [SP] works over Schfppf . These
gives equivalent notions of algebraic stacks (c.f., [SP, Tag 076U]).
– What we call representable (resp., representable by schemes) for a morphism
of stacks is called representable by algebraic spaces (resp., representable) in
[SP, Tags 02ZW and 02Y7].

3.2 Representability of the diagonal

The diagonal gives a three-dimensional feeling to dancers that cannot
be achieved when they are only front and back. On the diagonal, more
movement is automatically visible.

Lucinda Childs

We prove the diagonals of algebraic spaces and stacks are necessarily representable.
The diagonal is used to define stabilizers and the inertia stack.

115
3.2.1 Representability
Theorem 3.2.1 (Representability of the Diagonal).
(1) The diagonal of an algebraic space is representable by schemes.
(2) The diagonal of an algebraic stack is representable.

Proof. Let X be an algebraic space and U → X be an étale presentation. Define

the scheme R := U ×X U . If T → X × X is a morphism from a scheme, we need to
show that the sheaf QT = X ×X×X T is, in fact, a scheme. Since U → X is étale,
surjective, and representable by schemes, so is U × U → X × X. The base change
of T → X × X by U × U → X × X is a scheme T ′ which is surjective étale over T .
In the cartesian cube

QT ′ / T′

} {
R / U ×U

(3.2.2)
 
QT /T

 }  {
X / X × X,

QT is a sheaf on Schét while QT ′ is a scheme. Since R → U × U is a locally quasi-

finite and separated morphism of schemes, so is QT ′ → T ′ . (If X has quasi-compact
diagonal, then by Zariski’s main theorem R → U × U is quasi-affine and thus so is
QT ′ → T ′ .) Since QT is a sheaf in the étale topology that pulls back to a scheme
QT ′ locally quasi-finite and separated over T ′ , we may apply the Descent Criterion
for an Fppf Sheaf to be a Scheme (2.3.17) to conclude that QT is a scheme.
If X is an algebraic stack and U → X is a smooth presentation, we may imitate
the above argument. The fiber product R := U ×X U is an algebraic space. If
T → X × X is a morphism from a scheme, its base change along U × U → X × X
yields an algebraic space T1 which is surjective smooth over T . Choose an étale
presentation T2 → T1 . Then T2 → T is a surjective smooth morphism of schemes
which has a section after an étale cover T ′ → T (Proposition A.3.4). The composition
T ′ → T2 → T1 → U × U provides a lift of T → X × X . We obtain a diagram similar
to (3.2.2) but where the left and right squares are not necessarily cartesian. Since
QT is identified with IsomX (T ) (a, a) where a : T → X (see Exercise 2.4.39), Axiom
(1) for X being a stack implies that QT is a sheaf. The morphism QT ′ → QT is
étale, surjective, and representable by schemes (as T ′ → T is). Choosing an étale
presentation V → QT ′ of the algebraic space QT ′ , the composition V → QT ′ → QT
yields an étale presentation showing that QT is an algebraic space.

Corollary 3.2.3.
(1) If the diagonal of a stack X is representable (resp., representable by a scheme),
then every morphism U → X from a scheme is representable (resp., repre-
sentable by a scheme).
(2) Every morphism from a scheme to an algebraic stack (resp., algebraic spaces)
is representable (resp., representable by schemes).

116
Proof. The first part follows from the cartesian diagram of Exercise 2.4.38
T1 ×X T2 / T1 × T2

□
 
X / X × X.

associated to any two maps T1 → X and T2 → X from schemes to an algebraic

stack. The second part follows directly from the first part and the Representability
of the Diagonal (Theorem 3.2.1).
Exercise 3.2.4.
(a) If X → Y is a representable morphism of algebraic stacks (e.g., a morphism of
algebraic spaces), show that X → X ×Y X is representable by schemes.
(b) If X → Y is a morphism of algebraic stacks, show that X → X ×Y X is
representable.

3.2.2 Stabilizer groups and the inertia stack

Now that we know the diagonal is representable, we can discuss its properties.
Crucially, the diagonal encodes the automorphism or stabilizer groups. stabilizer
groups.
Definition 3.2.5 (Stabilizers). If X is an algebraic stack and x : Spec k → X is a
field-valued point, the stabilizer of x (or automorphism group of x) is defined as the
group algebraic space Gx := AutX (k) (x).
The stabilizer Gx is identified with the fiber product

Gx = AutX (k) (x) / Spec k

□ (x,x)
 
X
∆ / X ×X

by Exercise 2.4.39, and is representable by an algebraic space over k by Repre-

sentability of the Diagonal (3.2.1). The stabilizer Gx is a group algebraic space, i.e.,
an algebraic space Gx with multiplication, inverse, and identity morphisms satisfying
the commutativity conditions of Definition B.1.1 (or equivalently a group object in
the category of algebraic spaces). In fact, Gx is actually a group scheme locally of
finite type as long as the diagonal of X is quasi-separated (Corollary 4.5.31).
Remark 3.2.6. Let G be a group scheme over a field k acting on a k-scheme U via
σ : G × U → U , and let u ∈ U (k). The stabilizer of the image of u in [U/G] is the
usual stabilizer group scheme, i.e., the fiber product of (σ, p2 ) : G × U → U × U
along (u, u) : Spec k → U × U .
Exercise 3.2.7.
(a) Show that the stabilizer of a field-valued point of a fiber product of algebraic
stacks is the fiber product of stabilizers, i.e., for x′ ∈ (X ×Y Y ′ )(k), then
Gx′ ∼
= Gx ×Gy Gy′ , where x, y and y ′ are the images of x′ .
(b) Let f : X → Y be a morphism of algebraic stacks and x ∈ X (k) be a field-
valued point. Show that the fiber of the diagonal X → X ×Y X over the point
(x, x, id) ∈ (X ×Y X )(k) is identified with ker(Gx → Gy ). What is the fiber of
the diagonal over an arbitrary field-valued point of X ×Y X ?

117
Exercise 3.2.8. Let X be a Deligne–Mumford stack .
(a) For a field-valued point x ∈ X (k), show that Gx is a separated étale group
scheme over k. If in addition Gx is quasi-compact2 , show that Gx is a finite
étale group scheme over k.
Hint: First show that Gx is an etale group algebraic space over k which
becomes a scheme after the base change by a finite field extension k → k′ .
Apply Proposition B.1.8 to conclude that Gx ×k k′ is separated. Then apply
the Descent Criterion for an Fppf Sheaf to be a Scheme (2.3.17).
(b) Show that the diagonal of X is unramified.3
We will see later that these properties characterize Deligne–Mumford stacks; see
Theorem 3.6.4.
Remark 3.2.9. If X is a quasi-separated Deligne–Mumford stack, the above exercise
implies that the stabilizer Gx of x ∈ X (k) is a finite étale group scheme. When k is
algebraically closed, then Gx corresponds to the finite group scheme corresponding
to the abstract finite group AutX (k) (k).
The stabilizer group Gx varies as x ∈ X varies, and the stabilzer groups nat-
urally form a family. We have already seen this: if a : T → X is an object, then
IsomX (T ) (a) → S is a group algebraic space such that the fiber over a point s ∈ S
is the stabilizer of the restriction a|Spec κ(s) of a to Spec κ(s). Applying this to the
identity map idX : X → X produces the inertia stack.
Definition 3.2.10 (Inertia stack). The inertia stack of an algebraic stack X is the
fiber product
IX /X

□
 
X / X × X.
In the relative setting of a morphism X → Y of algebraic stacks, the relative inertia
stack is IX /Y := X ×X ×Y X X .
The inertia stack IX is equivalent to the category of pairs (x, α), where x ∈ X is
an object over S ∈ Sch and α ∈ AutX (S) (x) (see Exercise 3.2.15(a)). The fiber of
IX → X over a field-valued point x ∈ X (k) is precisely the stabilizer Gx . We can
therefore think of IX as a group scheme (or really group algebraic space) over X
incorporating all of the stabilizers of X . If we let (Sch /X )ét be the big étale site of
schemes over X , then IX is a sheaf of groups on (Sch /X )ét where IX (a) = AutS (a)
for a ∈ X (S).
If α : a′ → a is a morphism over S ′ → S, there is a natural pullback functor
α : AutS (a) → AutS ′ (a′ ) defined as follows: for β ∈ AutS (a), the image α∗ (β) is
∗

the unique dotted arrow (provided by Axiom (2) of a prestack (Definition 2.4.1))
making the diagram
β◦α

a′ / a′ /& a (3.2.11)
α∗ (β) α

commute. Note that if α : a′ → a is an isomorphism, then α∗ (β) = α−1 ◦ β ◦ α is

conjugation by α.
2 This is implied by the quasi-separatedness of X ; see Definition 3.3.11.
3 Since unramified morphisms are schemes are stable under base change and étale local on the
source, the notion is well-defined for representable morphism of algebraic stacks.

118
Exercise 3.2.12. Let G → S be a group scheme acting on a scheme U → S, and
let X = [U/G] be the quotient stack. Show that there is a cartesian diagram
SU /U

□
 
IX /X

where SU → U is the stabilizer group scheme, i.e., the fiber product of the action
map G × U → U × U and the diagonal U → U × U .
Example 3.2.13. The inertia stack of the classifying stack BGm is IBGm ∼ = Gm ×
BGm . Similarly, if we let Gm act on Gm × A1 via the product of the trivial and
the scaling action and we let V (x(t − 1)) ⊆ Gm × A1 be the Gm -invariant closed
subscheme, then I[A1 /Gm ] ∼
= [V (x(t − 1))/Gm ].
Exercise 3.2.14.
(a) If G is a smooth affine algebraic group over a field k, show that the inertia
stack of BG is the quotient [G/G] where G acts on itself via conjugation. In
particular, if G is abelian then IBG ∼
= G × BG.
(b) More generally, show that if G acts on a k-scheme U , show that I[U/G] = ∼
[(G × U )/G] where the action is given by g · (h, u) = (ghg −1 , gu).
(c) Let G be a group scheme over k corresponding to a finite abstract group. If G
acts on a k-scheme U , then
a
I[U/G] = [U g /Cg ],
g∈Conj(G)

where Conj(G) is the set of conjugacy classes of elements of G, Cg is the

centralizer of g, and U g := {x ∈ U | gx = x} is the closed locus fixed by g.
(d) Explicitly compute the inertia stack for the quotient stack [A3 /S3 ] of the
permutation action.
Exercise 3.2.15. Let f : X → Y be a morphism of algebraic stacks.
(a) Show that the relative inertia stack IX /Y is equivalent to the category of pairs
∼
(x, α) where x ∈ X and α : x → x is an isomorphism such that f (α) = idf (x) ,
and show that the identity section X → IX /Y takes an object x to (x, idx ).
(b) Show that there are morphisms IX /Y → IX → IY ×Y X of algebraic stacks
over X which corresponds over a field-valued point x ∈ X (k) to a left exact
sequence 1 → Kx → Gx → Gf (x) of group algebraic spaces.
(c) Show that there is a cartesian diagram
IX / IY ×Y X

□
 
X / X ×Y X .

Hint for (c): An object of IY ×Y X over a scheme S is a quadruple (y, α, x, β) where

∼ ∼
y ∈ Y(S), α : y → y, x ∈ X (S), and β : y → f (x). On the other hand, an object of
∼
X ×Y X over S is a triple (x1 , x2 , γ) where x1 , x2 ∈ X (S) and γ : f (x1 ) → f (x2 ).
Define IY ×Y X → X ×Y X on fiber categories by (y, α, x, β) 7→ (x, x, β ◦ α ◦ β −1 ).
Construct a map IX (S) to the fiber product of X (S) and (IY ×Y X )(S) over (X ×Y
X )(S), and show that it is an equivalence.

119
3.3 First properties
‘Of course, here I’m working with the moduli stack rather
than with the moduli space. For those of you who aren’t
familiar with stacks, don’t worry: basically, all it means is
that I’m allowed to pretend that the moduli space is smooth
and that there’s a universal family over it.’
Who hasn’t heard these words, or their equivalent, spoken in a talk? And
who hasn’t fantasized about grabbing the speaker by the lapels and shaking
him until he says what — exactly — he means by them?

Joe Harris and Ian Morrison [HM98, p.140]

We define our first properties of morphisms of algebraic stacks (e.g., closed

immersion, affine, locally of finite type, smooth, quasi-separated, quasi-finite, and
étale) and some first properties of algebraic stacks (e.g., locally noetherian, quasi-
compact, connected, reduced, regular, and noetherian). We also introduce the
underlying topological space |X | of an algebraic stack.

3.3.1 Properties of morphisms

Recall that a morphism of prestacks X → Y over Schét is representable by schemes
(resp., representable) if for every morphism T → Y from a scheme, the base change
X ×Y T is a scheme (resp., algebraic space); see Definitions 3.1.1 and 3.1.3. Both
notions are clearly stable under base change. Morphisms representable by schemes
are also clearly stable under composition, and the following lemma shows the same
for representable morphisms.
Lemma 3.3.1.
(1) If X → Y is a representable morphism of prestacks over Schét and Y is an
algebraic space, then X is an algebraic space.
(2) The composition of representable morphisms is representable.
Proof. For the first statement, if V → Y is an étale presentation by a scheme V ,
the base change XV is an algebraic space. Since the diagonal X → X ×Y X base
changes under V → Y to a monomorphism XV → XV ×V XV , étale descent implies
that X → X ×Y X is also a monomorphism. Hence, X is equivalent to a sheaf
(Exercise 2.4.39). Moreover, the base change XV → X is a morphism of algebraic
spaces which is étale, surjective and representable by schemes. Letting U → XV be
an étale presentation, then the composition U → XV → X is étale, surjective, and
representable by schemes, and thus X is an algebraic space. The second statement
follows from the first.
Definition 3.3.2. Let P be a property of morphisms of schemes.
(1) We say that P is étale local on the source if for every étale surjection X ′ → X
of schemes, a morphism X → Y satisfies P if and only if X ′ → X → Y does,
and that P is étale local on the target if for every étale surjection Y ′ → Y of
schemes, a morphism X → Y satisfies P if and only if X ×Y Y ′ → Y does.
The notions of being smooth (resp., fppf, fpqc) local on the source/target are
defined similarly.
(2) If P is étale local (resp., smooth local) on the source and target and is stable
under composition and base change, a morphism X → Y of Deligne–Mumford

120
 stacks (resp., algebraic stacks) has property P if for all étale (resp., smooth)
presentations V → Y and U → X ×Y V yielding a diagram

U / X ×Y V /V

□
 
X / Y,

the composition U → V has P. It is enough to check this for specific presenta-

tions V → Y and U → X ×Y V .
(3) A morphism X → Y of algebraic stacks representable by schemes has property
P if for every morphism T → Y from a scheme, the base change X ×Y T → T
has P. In particular, a morphism X → Y of algebraic stacks is an isomorphism,
open immersion, closed immersion, locally closed immersion, affine, or quasi-
affine if it is representable by schemes and has the corresponding property.
The properties of flatness, smoothness (resp., smoothness of relative dimension
n), surjectivity, locally of finite presentation, and locally of finite type are smooth
local on the source and target. Using (2), these properties extend to morphisms
of algebraic stacks. Likewise, étaleness and unramifiedness are étale local on the
source and target, and thus extend to morphisms of Deligne–Mumford stacks. These
properties are stable under composition and base change.
Example 3.3.3. If G → S is a smooth affine group scheme acting on an algebraic
space U → S, then [U/G] → S is flat (resp., smooth, surjective, locally of finite
presentation, locally of finite type) if and only if U → S is. In particular, using the
quotient stack presentations in the proofs of Theorems 3.1.17 and 3.1.21, we conclude
that Mg is locally of finite type over Z and that both Cohr,d (C) and Bun r,d (C) are
locally of finite type over k.
Exercise 3.3.4 (easy). Let G = Z/2 act on A2 over a field k via (−1) · (x, y) =
(−x, −y). Show that [A2 /G] is smooth but that the scheme quotient A2 /G :=
Spec k[x, y]G is singular.
Exercise 3.3.5 (easy). Show that the diagonal of a morphism X → Y of algebraic
stacks is locally of finite type.
We now show that certain morphisms of algebraic stacks are smooth local on
the target. They are even fppf local, but this is postponed until Exercise 6.3.5.
Proposition 3.3.6. Let P be one of the following properties of morphisms of
algebraic stacks: representable, isomorphism, open immersion, closed immersion,
affine, or quasi-affine. Consider a cartesian diagram

X′ / Y′

□
 
X /Y

of algebraic stacks where Y ′ → Y is smooth and surjective. Then X → Y has P if

and only if X ′ → Y ′ has P.
Proof. We will show the (⇐) implications as the other directions are clear. For
representability, we may assume that Y and Y ′ are schemes. It suffices to show that

121
the every automorphism α : a → a of an object a ∈ X over a scheme T is trivial.
The base change T ′ of a : T → X by X ′ → X is identified with T ×Y Y ′ , and thus
is a scheme smooth over T . Since smooth morphisms étale locally have sections
(Corollary A.3.5), there is an étale cover g : Te → T that factors through T ′ . The
automorphism α defines a section of AutT (a) over T . Since AutT (a) is a sheaf on
(Sch/T )ét and g ∗ α = id, we have that α = id.
For the other properties, we already know that X → Y is representable, and it
thus suffices to assume that Y, Y ′ and X ′ are schemes and that X is an algebraic
space. Fortunately, we can apply Descent Criterion for an Fppf Sheaf to be a Scheme
(Proposition 2.3.17) to conclude that X is a scheme. The result therefore follows as
these properties of morphisms of schemes are Fpqc Local Properties on the Target
(2.1.26)

3.3.2 Properties of algebraic spaces and stacks

Definition 3.3.7 (Properties of algebraic spaces and stacks). Let P be a property
of schemes which is étale (resp., smooth) local, i.e., if X → Y is an étale (resp.,
smooth) surjection of schemes, then X has P if and only if Y has P. We say that
a Deligne–Mumford stack (resp., algebraic stack) X has property P if for an étale
(resp., smooth) presentation (equivalently for all presentations) U → X , the scheme
U has P.
The properties of being locally noetherian, reduced, or regular are smooth local
(Proposition 2.1.21).
Example 3.3.8. Let G → S be a smooth affine group scheme acting on a scheme
U over S. Then [U/G] is locally noetherian, reduced, or regular if and only if U is.
Definition 3.3.9 (Substacks). If X is an algebraic stack, a substack Z ⊆ X is closed
(resp., open, locally closed ) if the induced morphism Z → X is a closed immersion
(resp., open immersion, locally closed immersion).
Exercise 3.3.10. For an action of a smooth affine group scheme G → S on a scheme
U over S, show that there is an equivalence between closed (resp., open) substacks
of [U/G] and G-invariant closed (resp., open) subschemes of U .

3.3.3 Separation properties

Separation properties for algebraic stacks are defined in terms of the diagonal.
Definition 3.3.11.
(1) A morphism of algebraic stacks X → Y has affine diagonal (resp., quasi-affine
diagonal, separated diagonal ) if the diagonal X → X ×Y X is affine (resp.,
quasi-affine, separated). An algebraic stack X has affine diagonal (resp.,
quasi-affine diagonal, separated diagonal ) if X → Spec Z does.
(2) A morphism of algebraic stacks X → Y is quasi-separated if the diagonal
X → X ×Y X and second diagonal X → X ×X ×Y X X are quasi-compact. An
algebraic stack X is quasi-separated if it is quasi-separated over Spec Z.
(3) A morphism of algebraic stacks X → Y is of finite presentation if it locally of
finite presentation, quasi-compact, and quasi-separated.
(4) A representable morphism X → Y of algebraic stacks is separated if the
morphism X → X ×Y X , which is representable by schemes (Exercise 3.2.4),
is proper. Separated morphisms are defined in general in Definition 3.8.1.

122
 Conditions on the diagonal translate to conditions on the Isom sheaves since the
base change of X → X × X by a morphism (a, b) : S → X × X from a scheme S
is identified with IsomX (S) (a, b) (see Exercise 2.4.39), which is an algebraic space
by Representability of the Diagonal (Theorem 3.2.1(2)). In particular, X has
affine diagonal if and only if IsomX (S) (a, b) is affine for every scheme S and a, b ∈
X (S). Every algebraic stack with affine or quasi-affine diagonal is necessarily
quasi-separated.
Remark 3.3.12. For morphisms of schemes, the definition of separatedness above
agrees with the usual notion as the diagonal of a morphism of schemes is a closed
immersion if and only if it is proper.
Remark 3.3.13. A quasi-separated Deligne–Mumford stack has finite and reduced
stabilizer groups (see Exercise 3.2.8).
Lemma 3.3.14. Let S be an affine scheme and G → S be a smooth affine group
scheme acting on an algebraic space U over S. If U has affine diagonal (resp.,
quasi-affine diagonal), then so does [U/G].
Proof. Recall that we established that [U/G] is an algebraic stack in Theorem 3.1.10.
Representability of the Diagonal (Theorem 3.2.1(2)) implies that [U/G] → [U/G] ×S
[U/G] is a representable morphism. Applying smooth descent (Proposition 3.3.6) to
the cartesian diagram

G ×S U / U ×S U

□
 
[U/G] / [U/G] ×S [U/G]

of Exercise 2.4.37, it suffices to show that G ×S U → U ×S U is affine (resp., quasi-

p2
affine). Since G is affine, so is the composition G ×S U → U ×S U −→ U . On the
other hand, p2 : U ×S U → U has affine diagonal (resp., quasi-affine diagonal) since
U does. The cancellation property for morphisms implies that G ×S U → U ×S U is
affine (resp., quasi-affine).
The condition of having affine diagonal is satisfied by most moduli problems
(except for example M1 ).
Example 3.3.15. The moduli stacks Mg and Bun r,d (C) have affine diagonal and
are thus quasi-separated. The statement for Mg follows from the above lemma and
the quotient presentation Mg = [H ′ / PGL5g−5 ] in Theorem 3.1.17 as H ′ is locally
closed subscheme of a projective Hilbert scheme. We will show later that Mg is
separated (Proposition 5.5.20).
Similarly in Theorem 3.1.21, we expressed every quasi-compact open substack
of Bun r,d (C) as a quotient stack [Q′ / GLN ] where Q′ is an open subscheme of a
projective Quot scheme. To see that Bun r,d (C) has affine diagonal, it suffices to show
that the base change of the along a morphism Spec A → Bun r,d (C) × Bun r,d (C) is
affine. But such a morphism factors through U × U for some quasi-compact open
substack U ⊆ Bun r,d (C), and we know that U has affine diagonal.
Example 3.3.16. The non-separated union A∞ A∞ ∖0 A∞ is a prototypical example
S
of a non-quasi-separated scheme. For algebraic spaces and stacks, there are additional
pathologies coming from actions of non-quasi-compact group schemes. For instance,
[A1 /Z] is a non-quasi-separated algebraic space (see Example 3.9.35), while BZ is a
non-quasi-separated algebraic stack (see Example 3.9.35).

123
3.3.4 The topological space of a stack
We can associate a topological space |X | to every algebraic stack X .

Definition 3.3.17 (Topological space of an algebraic stack). If X is an algebraic

stack, we define the topological space of X as the set |X | consisting of field-valued mor-
phisms x : Spec K → X , where morphisms x1 : Spec K1 → X and x2 : Spec K2 → X
are identified in |X | if there exists field extensions K1 → K3 and K2 → K3 such
that x1 |Spec K3 and x2 |Spec K3 are isomorphic in X (K3 ). A subset U ⊆ |X | is open if
there exists an open substack U ⊆ X such that U = |U| ⊆ |X |.

A morphism of algebraic stacks X → Y induces a continuous map |X | → |Y|.

Exercise 3.3.18. Show that if X is an algebraic stack and U ⊆ |X | is an open

subset, then there exists a reduced closed substack Z ,→ X such that |Z| = |X | ∖ U .

Example 3.3.19. The topological space of the quotient stack |[A1 /Gm ]| over a field
k, where Gm acts via the standard scaling action on A1 , consists of two points with
0 1
representatives x0 : Spec k − → A1 → [A1 /Gm ] and x1 : Spec k −
→ A1 → [A1 /Gm ]. In
particular, the inclusion of the generic point Spec k(x) → A → [A1 /Gm ] is equivalent
1

to x1 .

Given x ∈ |X |, the stabilizer group Gx1 depends on the choice of representative

x1 : Spec k → X of x but its dimension—which we denote by dim Gx —is independent
of this choice. Similarly, the properties of being smooth, unramified, affine, finite,
and reduced are also independent of this choice.

Exercise 3.3.20 (easy). Let x ∈ |X | be a point of an algebraic stack with two

representatives x1 : Spec k1 → X and x2 : Spec k2 → X .
(1) Show that the stabilizer group Gx1 is smooth (resp., étale, unramified, affine,
finite) if and only if Gx2 is.
(2) Show that dim Gx1 = dim Gx2 .
(3) If X is Deligne–Mumford and both k1 and k2 are algebraically closed, show
that the abstract discrete groups corresponding to Gx1 and Gx2 are isomorphic.

As a consequence of the above exercise, it makes sense to say that x ∈ |X | has

smooth (resp., étale, unramified, affine, finite) stabilizer. For a Deligne–Mumford
stack X , we define the geometric stabilizer of x as the discrete group G = Gx (k)
where x : Spec k → X is a geometric point representing x.
We can now define topological properties of algebraic stacks and their morphisms.

Definition 3.3.21. We say that an algebraic stack X is quasi-compact, connected,

or irreducible if |X | is, and we say that X is noetherian if it is locally noetherian
(Definition 3.3.7), quasi-separated (Definition 3.3.11(2)), and quasi-compact.

Exercise 3.3.22 (easy). Show that an algebraic stack X is quasi-compact if and

only if there exists a smooth presentation Spec A → X from an affine scheme, and
that a quasi-separated algebraic stack X is noetherian if and only if there exists a
smooth presentation Spec A → X where A is a noetherian ring.

Example 3.3.23. The moduli stack Mg is noetherian and in particular quasi-

compact. This follows from the above exercise using the quotient presentation
Mg = [H ′ / PGL5g−5 ] from Theorem 3.1.17. While Bun r,d (C) is quasi-separated
and locally noetherian, it is not quasi-compact, and hence not noetherian.

124
Exercise 3.3.24 (easy).
(a) Show that a morphism X → Y of algebraic stacks is surjective if and only if
|X | → |Y| is surjective.
(b) Show that if X → Y and Y ′ → Y are morphisms of algebraic stacks, then
|X ×Y Y ′ | → |X | ×|Y| |Y ′ | is surjective.

Exercise 3.3.25 (easy). If X is a quasi-compact and locally noetherian algebraic

stack, show that |X | is a noetherian topological space.

Exercise 3.3.26 (moderate). If U → X is a smooth presentation of an algebraic

stack, show that |U | → |X | is submersive, i.e., is surjective and |X | has the quotient
topology.
Hint: If V ⊆ |X | is a subset whose preimage U ′ in U is open, define V ⊆ X as the
substack of objects T → X such that UT′ = UT .

Exercise 3.3.27 (easy). Universally open morphisms of algebraic stacks can be

defined using Definition 3.3.2(1) since they are smooth local on the source and target.
For instance, faithfully flat morphisms locally of finite presentation are universally
open. If f : X → Y is a universally open morphism of algebraic stacks, show that
f (|X |) ⊆ |Y| is open. Conclude that for every morphism Y ′ → Y of algebraic stacks,
the map |X ×Y Y ′ | → |Y ′ | is open.

Definition 3.3.28. A morphism X → Y of algebraic stacks is quasi-compact if for

every morphism Spec B → Y, the fiber product X ×Y Spec B is quasi-compact. We
say that X → Y is of finite type if X → Y is locally of finite type and quasi-compact.

Example 3.3.29. The moduli stack Mg is finite type over Z. On the other hand,
Bun r,d (C) is locally of finite type over k but not of finite type.

Remark 3.3.30. A quasi-compact morphism X → Y induces a quasi-compact mor-

phism |X | → |Y| on topological spaces. The converse is true if Y is quasi-separated
but not in general, e.g., Spec k → BZ (see Example 3.9.35).

Exercise 3.3.31 (challenging).

(a) Let f : X → Y be a quasi-compact morphism of algebraic stacks. For y ∈ |Y|,
show that y ∈ f (|X |) if and only if there exists x ∈ |X | and a specialization
f (x) ⇝ y.
(b) Show an open morphism f : X → Y of algebraic stacks (i.e., |X | → |Y| is
open) satisfies the following lifting property: if x ∈ |X | is a point, then every
specialization y ′ ⇝ f (x) lifts to a specialization x′ ⇝ x. Show that the
converse is true for finitely presented morphisms.
(c) If X is a quasi-separated algebraic stack, show that |X | is a sober topological
space, i.e., every irreducible closed subset has a unique generic point.
(d) Generalize Chevalley’s criterion to algebraic stacks: if f : X → Y is a finitely
presented morphism of algebraic stacks, then the image f (|X |) ⊆ |Y| is
constructible.
See also [LMB00, §5.7-5.9] and [SP, Tags 0DQQ and 0GVY].

Exercise 3.3.32 (Generic Flatness). Extend Generic Flatness (A.2.13) from mor-
phisms of schemes to algebraic stacks: if X → Y is a finite type morphism of
algebraic stacks with Y reduced, then there exists a dense open substack U ⊆ Y
such that the base change XU → U is flat and of finite presentation.

125
Exercise 3.3.33 (technical). Extend the characterization of locally of finite pre-
sentation morphisms given in Proposition A.1.1 to algebraic stacks: a morphism
f : X → Y of algebraic stacks is locally of finite presentation if and only if for every
directed system {Spec Aλ }λ∈I of affine schemes over Y, the natural map

colimλ MorY (Spec Aλ , X ) → MorY (Spec(colimλ Aλ ), X )

is an equivalence of categories.

3.3.5 Quasi-finite morphisms

A morphism of schemes is locally quasi-finite if it is locally of finite type and every
fiber is discrete, i.e., has the discrete topology. Since this property is étale local on
the source and target, we can extend this property to morphisms of algebraic spaces
using Definition 3.3.2.

Definition 3.3.34.
(1) A representable morphism X → Y of algebraic stacks is locally quasi-finite
if for every morphism T → Y from a scheme, the algebraic space X ×Y T is
locally quasi-finite over T .
(2) A morphism X → Y of algebraic stacks is locally quasi-finite if is locally of
finite type, the diagonal X → X ×Y X is locally quasi-finite, and for every
morphism Spec k → Y from a field, the topological space |X ×Y Spec k| is
discrete.
(3) A morphism X → Y of algebraic stacks is quasi-finite if it is locally quasi-finite
and quasi-compact.

To understand condition (2), recall that the diagonal X → X ×Y X is always a

representable morphism (Exercise 3.2.4). The diagonal is quasi-finite (resp., locally
quasi-finite) if and only if for every field-valued point x ∈ X (k) with image y ∈ Y(k),
the kernel ker(Gx → Gy ) of the induced map of stabilizer groups is finite (resp.,
discrete); see Exercise 3.2.7. In particular, if Y is a scheme, the diagonal is quasi-
finite if and only if all stabilizers of Y are finite. For instance, if G is a finite group
scheme over a field k (e.g., µp ), then BG → Spec k is quasi-finite. On the other
hand, BGm → Spec k is not quasi-finite despite that |BGm | is a single point.

Exercise 3.3.35 (easy). Show that a finite type morphism f : X → Y of algebraic

stacks is quasi-finite if and only if |X | → |Y| has finite fibers and for every field-valued
point x ∈ X (k), the map AutX (k) (x) → AutY(k) (f (x)) has finite cokernel.

We will later establish that every representable, quasi-finite, and separated

morphism is quasi-affine (Proposition 4.5.5).

3.3.6 Étale and unramified morphisms

We will require that étale and unramified morphisms are relatively Deligne–Mumford.

Definition 3.3.36. A morphism of stacks X → Y over Schét is relatively Deligne–

Mumford if for every morphism T → Y from a scheme, the fiber product X ×Y T is
a Deligne–Mumford stack.

We will see in Corollary 3.6.5 that relatively Deligne–Mumford morphisms are

characterized by the unramifiedness of the diagonal. A morphism X → Y satisfying

126
the weaker condition that the diagonal X ×Y X is locally quasi-finite is called
quasi-DM [SP, Tag 04YW].
For morphisms of schemes, étaleness and unramifiedness are étale local on the
source and smooth local on the target. These notions thus extend to relatively
Deligne–Mumford morphisms.
Definition 3.3.37. A morphism X → Y of algebraic stacks is étale (resp., unrami-
fied ) if it is relatively Deligne–Mumford and for every smooth presentation V → Y
and étale presentation U → X ×Y V , the induced morphism U → V of schemes is
étale (resp., unramified).
A morphism is étale if and only if it is smooth and unramified, and a morphism is
unramified if and only if the diagonal is étale. These follows from the analogous facts
for morphisms of schemes (Theorems A.3.2 and A.3.3), noting that for a morphism
of schemes, the diagonal is étale if and only if it is an open immersion.
While étale morphisms are smooth and locally quasi-finite, the converse is not
true, e.g., over a field k of characteristic p, the map Bµ µp → Spec k is smooth and
quasi-finite but is not étale as Bµµp is not Deligne–Mumford (see Exercise 6.3.12).
Similarly, étale morphisms are smooth of relative dimension 0, but again the converse
doesn’t hold, e.g., Bµ µp → Spec k in characteristic p or [A1 /Gm ] → Spec k in any
characteristic.
Exercise 3.3.38 (technical). For a morphism f : X → Y of algebraic stacks, show
that the following are equivalent:
(1) the diagonal ∆f : X → X ×Y X is unramified (resp., separated, quasi-separated),
(2) the relative inertia IX /Y → X is unramified (resp., separated, quasi-separated),
and
(3) the double diagonal (or equivalently the identity section of the relative inertia)
X → IX /Y = X ×X ×Y X X is an open immersion (resp., closed immersion,
quasi-compact).

Hint: Extend the characterization of unramified (resp., separated, quasi-separated)

group schemes of Proposition B.1.8(4) to group algebraic spaces.

3.4 Equivalence relations and groupoids

Different people at different times will use the different techniques for
different purposes.

Serge Lang

We show that quotients of equivalence relations and groupoids can be constructed

as algebraic stacks and spaces (Theorem 3.4.13). Moreover, any presentation of an
algebraic space or stack induces a equivalence relation or groupoid of schemes. In
addition to giving a useful way to construct algebraic spaces and stacks, equivalence
relation and groupoids provide an alternative way to think about algebraic spaces
and stacks.

3.4.1 Definitions
Definition 3.4.1. An étale (resp., smooth) groupoid of schemes is a pair of schemes
U and R together with étale (resp., smooth) morphisms s : R → U called the source

127
and t : R → U called the target, and a composition morphism c : R ×s,U,t R → R
satisfying:
(1) (associativity) the following diagram commutes

R ×s,U,t R ×s,U,t R
c×id
/ R ×s,U,t R

id ×c c
 
R ×s,U,t R
c / R,

(2) (identity) there exists a morphism e : U → R (called the identity) such that
the following diagrams commute
(e◦s,id)
U R / R ×s,U,t R o (e◦t,id)R
id id
e c
   id
$  z
id
U o /U
s t
R R,

(3) (inverse) there exists a morphism i : R → R (called the inverse) such that the
following diagrams commute

R
i /R i /R R
s /U R
t /U

t (i,id) e (id,i) e
s s
      
U R ×s,U,t R
c /R R ×s,U,t R
c / R.

We often abuse notation by denoting this data simply as s, t : R ⇒ U .

If (s, t) : R → U × U is a monomorphism, then we say that s, t : R ⇒ U is an
étale (resp., smooth) equivalence relation.
If U and R are algebraic spaces, and the source, target, and composition are
morphisms of algebraic spaces, we obtain the notion of an étale (resp., smooth)
groupoid of algebraic spaces and similarly an étale (resp., smooth) equivalence relation
of algebraic spaces.
Groupoids of schemes were introduced in [SGA3I , Exp. V].
Remark 3.4.2 (relations). We can view R as a scheme of relations on U : a point
r ∈ R specifies a relation on the points s(r), t(r) ∈ U , which we sometimes write as
r : s(r) → t(r). For every scheme T , the morphisms R(T ) ⇒ U (T ) define a groupoid
of sets, i.e., there is composition morphism R(T ) ×s,U (T ),t R(T ) → R(T ) satisfying
axioms analogous to (1)–(3). We can think of an element r ∈ R(T ) as specifying
a relation r : u → v between elements u, v ∈ U (T ). The composition morphism
composes relations r : u → v and r′ : v → w to the relation r′ ◦ r : u → w, while the
identity morphism takes u ∈ U (T ) to id : u → u and the inverse morphism takes
r : u → v to r−1 : v → u. When R ⇒ U is an equivalence relation, the morphism
R(T ) → U (T ) × U (T ) is injective, and there is at most one relation between any
two elements of U (T ).
The prototypical examples of groupoids arise from group actions.
Example 3.4.3 (Group actions). If G → S is an étale (resp., smooth) group scheme
with multiplication m : G ×S G → G acting on a scheme U over S via multiplication
σ : G × U → U , then
p2 , σ : G ×S U ⇒ U

128
is an étale (resp., smooth) groupoid of schemes. The inverse G ×S U → G ×S U is
given by (g, u) 7→ (g −1 , gu) and the composition is

(g ′ , u′ ), (g, u) | u′ = gu 7→ (g ′ g, u).


(G ×S U ) ×p2 ,U,σ (G ×S U ) → G ×S U,

A T -valued point (g, u) of G ×S U as the relation u → gu.

The special case when U = S defines a groupoid s, t : G ⇒ S with both arrows
equal to the structure morphism G → S.

Example 3.4.4. For an interesting example of an étale equivalence relation, consider

the action of Z/2 = {±1} on A1 with multiplication σ : Z/2 × A1 → A1 defined by
−1 · x = −x. Removing the non-identity element of the stabilizer of the origin, we
obtain a scheme R = (Z/2×A1 )∖{(−1, 0)} and an equivalence relation p2 , σ : R ⇒ A1 .
We will show later that the quotient X = A1 /R is an algebraic space that is not a
scheme (Example 3.9.2).

identitysection
R 7 2 14121,0
Pa o

1A a x Palx

Figure 3.4.4: The quotient X = A1 /R looks like A1R except that the origin has residue
field C.

Example 3.4.5 (Groupoids induced from presentations). Let X be a Deligne–

Mumford stack (resp., algebraic stack) and p : U → X be an étale (resp., smooth)
noidentification
presentation which we assume is not only representable but representable by schemes.
identified
Define the algebraic spaceA R := U ×X U (which is a scheme if the diagonal of X
is representable by schemes). By the definition of fiber products, a T -valued point of
R is a triple (u1 , u2 , α) where α : p(u1 ) → p(u2 ) is an isomorphism in X (T ), and this
ee
is viewed as the relation α : u1 → u2 . If we let the source morphism s = p1 : R → U
be the first projection and the target morphism t = p2 : R → U be the second
projection, the composition A morphism can be defined as R ×s,U,t R → U ×X U
taking (α2 : u2 → u3 , α1 : u1 → u2 ) to α2 ◦ α1 : u1 → u3 . This gives the structure
of an étale (resp., smooth)Caption
groupoid R ⇒ A Uis of algebraic
a bag eyedcoverspaces. Observe that if X
is an algebraic space, then the diagonal X → X × X is a monomorphism. Hence,
(s, t) : R → U × U is a monomorphism, and R ⇒ U is an étale equivalence relation
of schemes.
Choosing different presentations yields different groupoids which are equivalent
under a notion called Morita equivalence. While
an this is an important notion, we will
not use it in this text.
If U → [U/G] is a presentation of a quotient stack, then the induced groupoid is
precisely the groupoid p2 , σ : G ×S U ⇒ U discussed in the previous
have example. Indeed,
∼ Is
there is a natural identification U ×[U/G] U → G ×S U defined by u1 ×g u2 7→ (g, u2 ),
where u1 ×g u2 is shorthand notation for the triple (u1 = gu2 , u2 , g) defining an
∼
element of the fiber product. Similarly U ×[U/G] U ×[U/G] U → G ×S G ×S U is given
by u1 ×g1 u2 ×g2 u3 7→ (g1 , g2 , u3 ). The various projection maps in the groupoid can

129
be identified with the maps arising from the group action:

U ×[U/G] U ×[U/G] U
∼ / G ×S G ×S U

p12 p13 p23 id ×σ m×id p23

     
U ×[U/G] U
∼ / G ×S U

p1 p2 p2 σ
   
U U


[U/G].

Exercise 3.4.6. Show that there are identifications

∼
(U/[U/G])n := U ×[U/G] · · · ×[U/G] U → Gn−1 ×S U,
| {z }
n times
u1 ×g1 · · · ×gn−1 un 7→ (g1 , . . . , gn−1 , un ),

where ui = gi ui+1 , such that the projection map (U/[U/G])n+1 → (U/[U/G])n

forgetting the kth term is identified with that map

Gn × U → Gn−1 × U,
(g2 , . . . , gn , un+1 ) if k = 1


(g1 , . . . , gn , un+1 ) 7→ (g1 , . . . , gk−2 , gk−1 gk , gk+1 , . . . , gn , un+1 ) if k = 2, . . . , n
(g1 , . . . , gn , gn un+1 ) if k = n + 1.


Definition 3.4.7 (Orbits and stabilizers). Let R ⇒ U be a smooth groupoid of

algebraic spaces, and let x : Spec k → U be a field-valued point. The stabilizer Gx
of x (or the automorphism group Aut(x) of x) is defined as the fiber product of
(s, t) : R → U × U by (x, x) : Spec k → U × U . The orbit O(x) is defined as the set
s(t−1 (x)) ⊆ U .

Remark 3.4.8. Observe that if the groupoid R ⇒ U arises from a smooth presentation
U → X of an algebraic stack, then the stabilizer of x ∈ U (k) is identified with the
stabilizer of its image in X (k). On the other hand, the orbit O(x) is the set of points
∼
y ∈ U such that there exists a relation x → y in R, or equivalently the set of points
y ∈ U whose image in X is isomorphic to the image of x.

Exercise 3.4.9 (not important). In Definition 3.4.1, show that the identity and
inverse morphism are uniquely determined.

3.4.2 Algebraicity of quotients by groupoids

Definition 3.4.10 (Quotient stack of a smooth groupoid). Let s, t : R ⇒ U be a
smooth groupoid of algebraic spaces. Define [U/R]pre as the prestack whose objects
are morphisms T → U from a scheme T , where a morphism (a : S → U ) → (b : T →
U ) is the data of a morphism of schemes f : S → T and an element r ∈ R(S) such
that s(r) = a and t(r) = f ◦ b. Define [U/R] to be the stackification of [U/R]pre in
the big étale topology Schét . If in addition R ⇒ U is an equivalence relation, then
[U/R] is isomorphic to a sheaf (Exercise 3.4.12) and we denote it as U/R.

130
 The fiber category [U/R]pre (T ) is the groupoid whose objects are U (T ) and
morphisms are R(T ). The identity morphism id : U → U defines a map U → [U/R]pre
and therefore a map p : U → [U/R].
Exercise 3.4.11 (important). Extend Exercise 2.4.37 to show that if s, t : R ⇒ U
is a smooth groupoid of algebraic spaces, the following diagrams are cartesian:

R
s /U R
(s,t)
/ U ×U
t □ p and □ p×p
   
U
p
/ [U/R] [U/R]
∆ / [U/R] × [U/R].

Exercise 3.4.12. Let R ⇒ U be a smooth groupoid of algebraic spaces. Show that

[U/R] is equivalent to a sheaf if and only if R ⇒ U is an equivalence relation.
Theorem 3.4.13 (Algebraicity of Quotients by Groupoids).
(1) If R ⇒ U is an étale (resp., smooth) groupoid of algebraic spaces. Then [U/R]
is a Deligne–Mumford stack (resp., algebraic stack) and U → [U/R] is an étale
(resp., smooth) presentation.
(2) If R ⇒ U be an étale equivalence relation of schemes, then U/R is an algebraic
space and U → U/R is an étale presentation.
Remark 3.4.14. In Corollary 4.5.12, we show that in fact the quotient U/R of an
étale equivalence relation of algebraic spaces is an algebraic space, establishing that
one doesn’t obtain new algebro-geometric objects by considering sheaves which are
étale locally algebraic spaces. This result is delayed until §4.5 as it takes more work
to show that the diagonal of U/R is representable by schemes. More generally, if
R ⇒ U is an fppf groupoid (resp., fppf equivalence relation) of algebraic spaces, then
[U/R] is an algebraic stack (resp., U/R is an algebraic space); see Corollary 6.3.6.
Proof. For (1), we will show that U → X := [U/R] is surjective, smooth, and
representable. Let T → X be a morphism from a scheme T . It follows from the
definition of [U/R] as the stackification of [U/R]pre that there exists an étale cover
T ′ → T and a commutative diagram

T′ /U

 
T / X.

In the commutative cube

UT ′ / T′

} ~
R /U
(3.4.15)
 
UT /T

 }  ~
U /X

131
of prestacks, the front, back, top, and bottom squares are cartesian. We need to
show that the sheaf UT is an algebraic space, but this isn’t hard: UT ′ is a scheme
and UT ′ → UT is a surjective étale morphism representable by schemes (as it is the
base change of T ′ → T ). Since R → U is surjective and étale (resp., smooth), so is
UT ′ → T ′ , and therefore by étale descent, UT → T is also.
For (2), we already know that U/R is a sheaf and that U → U/R is surjective,
étale, and representable. We need to show that U → U/R is representable by
schemes. It suffices to show that the diagonal of X is representable by schemes; see
Corollary 3.2.3(1). Let T → X × X be a morphism from a scheme and define QT
and QT ′ using the cartesian cube

QT ′ / T′

} {
R / U ×U

(3.4.16)
 
QT /T

 }  {
X / X × X.

Since R → U × U is locally quasi-finite and separated, so is QT ′ → T ′ . The Descent

Criterion for an Fppf Sheaf to be a Scheme (Proposition 2.3.17) implies that QT is
a scheme.

Remark 3.4.17 (Quotient stacks revisited). As a consequence, wee see that the
hypothesis in Algebraicity of Quotient Stacks (3.1.10) that hypothesis that the group
scheme G → S is affine is not necessary for the quotient stack [X/G] or classifying
stack BG to be algebraic.

Exercise 3.4.18 (important). Show that if X is an algebraic stack (resp., algebraic

space) and U → X is a smooth presentation, then X is isomorphic to the quotient
stack [U/R] (resp., quotient sheaf U/R) of the étale groupoid (resp., equivalence
relation) R ⇒ U where R = U ×X U .

3.4.3 Inducing and slicing presentations

We provide here two useful techniques to build new presentations from given ones.
First, consider a quotient stack X = [X/H] where H is a smooth affine algebraic
group acting on a scheme X over k. If G is another group that contains H ⊆ G as a
subgroup, then H acts freely on G × X via h · (g, x) = (gh−1 , hx), and we can define
the quotient algebraic space

G ×H X := (G × X)/H,

which is sometimes referred to as the Borel construction. So far, we only that this is
a sheaf representable by an algebraic stack ( Theorem 3.4.13(1)), but it will follow
from Corollary 3.6.8 that it is, in fact, an algebraic space. (If H is a finite abstract
group, then Theorem 3.4.13(2) already implies that it is an algebraic space.) In any
case, there is an action of G on G ×H X via g · (g ′ , x) = (gg ′ , x).

132
Exercise 3.4.19. Show that [X/H] ∼
= [(G ×H X)/G].
The quotient presentation [G ×H X/G] is called the induced presentation.

The second technique we introduce is called slicing a groupoid. Let U → X

be a smooth presentation of an algebraic stack with the corresponding groupoid
s, t : R = U ×X U ⇒ U . If g : U ′ → U is a morphism, we define the restriction of
R ⇒ U along U ′ → U to be the groupoid R|U ′ ⇒ U ′ defined by the fiber product

(t′ ,s′ )
R|U ′ / U′ × U′

□
 (t,s) 
R / U ×U

Exercise 3.4.20.
(a) Show that R|U ′ fits into a cartesian diagram

R|U ′ / R ×s,U U ′ / U′

□ □ g
  
U ′ ×U,t R /R s /U

□ t □
  
U′
g
/U / [U/R]

s
Assume in addition that U ′ ×U,t R → R −
→ U is étale (resp., smooth).
(2) Show that R|U ′ ⇒ U ′ is an étale (resp., smooth) groupoid.
(3) Show that there is an open immersion [U ′ /R|U ′ ] → [U/R].
(4) Show that [U ′ /R|U ′ ] → [U/R] is an isomorphism if and only if for every point
u ∈ U , there exists a point u′ ∈ U and a relation u → g(u′ ) in R.

Exercise 3.4.21 (Moduli of elliptic curves revisited). Consider the moduli stack
M1,1 ∼ = [(A2 ∖ V (∆))/Gm ] of elliptic curves from Exercise 3.1.19, where k is a
field with char(k) ̸= 2, 3 and ∆ = 4a3 + 27b2 is the discriminant (with a and b the
coordinates on A2 ). Slice along the closed immersion V (∆ − 1) ,→ A2 ∖ V (∆) to
show that M1,1 ∼ = [V (∆ − 1)/µ12 ] is a quotient of an affine scheme by a finite group.

3.5 Dimension, tangent spaces, and residual gerbes

A point is that which has no part.

Euclid

After extending the notions of dimension and tangent spaces to algebraic stacks,
we show that every finite type point has a residual gerbe, which serves as a replacement
for the notion of the residue field of a point of a scheme.

133
3.5.1 Dimension
Recall that the dimension dim X of a scheme X is the Krull dimension of the
underlying topological space while the dimension dimx X at a point x ∈ X is the
minimum dimension of open subsets containing x. This is equal to dim OX,x when
X is of finite type over a field and x ∈ X is a closed point. We now extend these
definitions to algebraic spaces and stacks.

Definition 3.5.1 (Dimension).

(1) Let X be a noetherian algebraic space and x ∈ |X|. We define the dimension
of X at x to be
dimx X = dimu U ∈ Z≥0 ∪ ∞

where U → X is an étale presentation and u ∈ U is a preimage of x.

(2) Let X be a noetherian algebraic stack and x ∈ |X |. If U → X is a smooth
presentation and s, t : R ⇒ U is the induced smooth groupoid, and u ∈ U is a
preimage of x ∈ |X |. We define the dimension of X at x to be

dimx X = dimu U − dime(u) Ru ∈ Z ∪ ∞,

where Ru is the fiber of s : R → U over u and e : U → R denotes the identity

morphism in the groupoid.
(3) If X is a noetherian algebraic space or stack, we define the dimension of X to
be
dim X = sup dimx X ∈ Z ∪ ∞.
x∈|X |

Proposition 3.5.2. The definition of dimx X is independent of the choices of

presentation U → X and the preimage u of x.

Proof. The dimension of an algebraic space at a point is well-defined as étale

morphisms have relative dimension 0. If U → X is a smooth presentation (with U a
scheme) and u ∈ U is a preimage of x with residue field κ(u), then the fiber Ru is
identified with the fiber product

Ru /R t /U

□ s □
  
Spec κ(u) /U / X,

and is a smooth algebraic space over κ(u). Let U ′ → X be a second presentation

inducing a smooth groupoid R′ ⇒ U ′ and u′ ∈ U ′ be a preimage of x. Since we can
form the smooth presentation U ×X U ′ of X dominating both U and U ′ , we may
assume that there is a smooth morphism U ′ → U of schemes, taking u′ to u. By
the dimension formula for smooth morphisms,

dimu′ U ′ = dimu U + dimu′ Uu′ ; (3.5.3)

see Proposition A.3.10. To relate dimu′ Uu′ to dime(u) Ru and dime′ (u′ ) Ru′ ′ , we form

134
the cartesian diagram

Ru′ ′ / Uu′′ / U ′′ / U′

□ □ □
   
Ru ×κ(u) κ(u′ ) / Ru /R t /U

□ □ s □
   
Spec κ(u′ ) / Spec κ(u) /U / X.

We will apply again the dimension formula for smooth morphisms to Uu′′ → Ru , but
this time using its generalization (Exercise 3.5.5(a)) to algebraic space. Using that
the fiber over e(u) ∈ |Ru | is identified with Uu′ and letting u′′ ∈ Uu′′ is a preimage of
e(u) ∈ |Ru | and u′ ∈ U ′ , we have that
dimu′′ Uu′′ = dime(u) Ru + dimu′ Uu′ . (3.5.4)
Since dimension is preserved under field extensions (Exercise 3.5.5(b)), dime′ (u′ ) Ru′ ′ =
dimu′′ Uu′′ . Combining (3.5.3) and (3.5.4) gives the desired result.
Exercise 3.5.5 (details).
(a) Show that if X → Y is a smooth morphism of noetherian algebraic spaces and
x ∈ |X| is a point with image y ∈ |Y |, then
dimx X = dimy Y + dimx Xy .
This generalizes Proposition A.3.10 to algebraic spaces.
(b) If X be a noetherian algebraic stack over a field k, k → k′ is a field extension,
and x′ ∈ |Xk′ | is a point with image x ∈ |X |, show that dimx′ Xk′ = dimx X .
Example 3.5.6. If G is a smooth affine algebraic group over a field k acting on a
pure dimensional scheme U of finite type over k, then
dim [U/G] = dim U − dim G.
In particular, the dimension of the classifying stack BG is
dim BG = − dim G,
which can be negative! Similarly, dim [A1k /Gm ] = 0.

3.5.2 Tangent spaces

The dual numbers is the ring k[ϵ] := k[ϵ]/ϵ2 defined over a field k.
Definition 3.5.7. For an algebraic stack X and a point x : Spec k → X , we define
the Zariski tangent space (or simply the tangent space of X at x) as the set
 

 Spec _ k 


 

x
TX ,x := 2-commutative diagrams α
∼
  s{ # 
τ /X

 

 Spec k[ϵ] 

∼
In other words, TX ,x is the set of pairs (τ, α) with τ : Spec k[ϵ] → X and α : x → τ |k ,
∼
where two pairs are equivalent (τ, α) ∼ (τ ′ , α′ ) if there is an isomorphism β : τ → τ ′
in X (k[ϵ]) compatible with α and α , i.e., α = β|Spec k ◦ α.
′ ′

135
Example 3.5.8. Consider a smooth, connected, and projective curve [C] ∈ Mg (k)
defined over k of genus g ≥ 2. Deformation Theory (C.1.11) implies that TMg ,[C] =
H 1 (C, TC ). Since deg TC < 0, H0 (C, TC ) = 0 and Riemann–Roch implies
dim TMg ,[C] = dim H1 (C, TC ) = −χ(TC ) = −(deg TC + (1 − g)) = 3g − 3.
Example 3.5.9. Let C be a smooth, connected, and projective curve over k and E ∈
Bun r,d (C)(k) be a vector bundle on C of rank r and degree d. Deformation Theory
(C.1.18) implies that TBun r,d (C),[E] = Ext1OC (E, E) = H 1 (C, E ⊗ E ∨ ). By Riemann–
Roch (8.1.2), χ(E ⊗ E ∨ ) = r2 (1 − g). Since dim Aut(E) = dimk HomOC (E, E) =
H0 (C, E ⊗ E ∨ ), we compute that
dim TBun r,d (C),[E] = dim Ext1OC (F, F ) = dim Aut(F ) + r2 (g − 1).
Proposition 3.5.10. If X is an algebraic stack with affine diagonal and x ∈ X (k),
then TX ,x is naturally a k-vector space.
Proof. Scalar multiplication of c ∈ k on (τ, α) ∈ TX ,x is defined as the composition
τ
→ X , where the first map is defined by ϵ 7→ cϵ and with the
Spec k[ϵ] → Spec k[ϵ] −
same 2-isomorphism α. To define addition, we will show that there is an equivalence
of categories
X (k[ϵ1 ] ×k k[ϵ2 ]) → X (k[ϵ1 ]) ×X (k) X (k[ϵ2 ]),
or, in other words, that

Spec _ k  / Spec k[ϵ1 ]
_
⌟
 
 / Spec(k[ϵ1 ] ×k k[ϵ2 ])
Spec k[ϵ2 ]

is a pushout among algebraic stacks with affine diagonal. (This is related to the
homogeneity conditions in Schlessinger’s Criteria (C.4.6) and Artin’s Axioms for
Algebraicity (C.7.4); see also §B.4 for a discussion of pushouts of schemes.) Once
this is established, we define addition of (τ1 , α1 ) and (τ2 , α2 ) by the composition
Spec k[ϵ] → Spec(k[ϵ1 ] ×k k[ϵ2 ]) → X , where the first map is defined sending both
(ϵ1 , 0) and (0, ϵ2 ) to ϵ.
To establish the equivalence, choose a smooth morphism (U, u) → (X , x) from
an affine scheme U . Since X has affine diagonal, U → X is an affine morphism. Let
Spec A0 = Spec k ×X U , Spec A1 = Spec k[ϵ1 ] ×X U and Spec A2 = Spec k[ϵ2 ] ×X U .
Since Spec(A1 ×A A2 ) is the pushout of Spec A0 ,→ Spec A1 and Spec A0 ,→ Spec A2
in the category of affine schemes, there are unique morphisms Spec(A1 ×A A2 ) →
Spec(k[ϵ1 ] ×k k[ϵ2 ]) and Spec(A1 ×A A2 ) → U completing the diagram

Spec A0 / Spec A2

x t
Spec k / Spec k[ϵ1 ]
τ1
 
Spec A1 / Spec(A1 ×A A2 )

 x  t '
Spec k[ϵ2 ] / Spec(k[ϵ1 ] ×k k[ϵ2 ]) U


τ2
*0 X w

136
By the Flatness Criterion over Artinian Rings (A.2.3), we see that the map
Spec(A1 ×A A2 ) → Spec(k[ϵ1 ] ×k k[ϵ2 ]) is faithfully flat. Repeating this argu-
ment on U ×X U shows that Spec(A1 ×A A2 ) → U descends uniquely to a map
U → X . These details and the verification that scalar multiplication and addition
give TX ,x the structure of a vector space are left to the reader.

Exercise 3.5.11 (hard).

(a) Show that TX ,x is naturally a representation of Gx which is given set-theoretically
by: g · (τ, α) = (τ, g ◦ α) for g ∈ Gx and (τ, α) ∈ TX ,x .
(b) Show that the affine diagonal in Proposition 3.5.10 is superfluous.

3.5.3 Residual gerbes

Attached to every point x ∈ X of a scheme is a residue field κ(x) and a monomorphism
Spec κ(x) → X with image x. The residual gerbe will provide us with an analogous
property for algebraic stacks. Note that non-trivial stabilizers prevent field-valued
points from being monomorphisms, e.g., Spec k → BG is not a monomorphism for a
non-trivial algebraic group.

Definition 3.5.12. Let X be an algebraic stack and x ∈ |X | be a point. We say

that the residual gerbe at x exists if there is a reduced, locally noetherian algebraic
stack Gx and a monomorphism Gx ,→ X such that |Gx | is a point mapping to x. The
algebraic stack Gx is called the residual gerbe at x.

We define gerbes later in Definition 6.4.6 and show that Gx is a gerbe over a
field κ(x), called the residue field (Proposition 6.4.26). While not apparent from
the definition, it is not hard to see that residual gerbes are in fact unique (see
Proposition 3.5.16). These facts justify our terminology of calling Gx the residual
gerbe. We will discuss the existence of residual gerbes more generally in §6.4.4, but
in the meantime, we will be content with verifying the existence of residual gerbes
at finite type points.

Definition 3.5.13. A point x ∈ |X | in an algebraic stack is of finite type if there

exists a representative Spec k → X of x that is locally of finite type.4

Remark 3.5.14. If X is a noetherian scheme, a point x ∈ X is of finite type if and

only if x ∈ X is locally closed. More generally, a morphism Spec k → X from a
field with image x is of finite type if and only if the image x ∈ X is locally closed
and κ(x)/k is a finite extension. Indeed, to see the non-trivial implication (⇒),
we replace X with {x}, and since Spec k → X is of finite type with dense image,
Generic Flatness (A.2.13) implies that Spec k → X is fppf and thus its image is
open.
An example of a finite type point of a scheme that is not closed is the generic
point of a DVR. On the other hand, if X is a scheme of finite type over a field k,
then every finite type point is a closed point. The analogous fact is not true for
algebraic stacks of finite type over k, e.g., 1 : Spec k → [A1 /Gm ] is an open finite
type point.

Exercise 3.5.15. Let X be an algebraic stack.

4 If X has quasi-compact diagonal, e.g., X is quasi-separated (Definition 3.3.11), then every

field-valued point Spec k → X is automatically quasi-compact, and thus the locally of finite typeness
of Spec k → X is equivalent to finite typeness.

137
 (a) Show that a point x ∈ |X | is of finite type if and only if there exists a scheme
U , a closed point u ∈ U , and a smooth morphism (U, u) → (X , x).
(b) Show that any algebraic stack (resp., quasi-compact algebraic stack) has a
finite type point (resp., closed point).

Proposition 3.5.16 (Existence of Residual Gerbes). If X is a noetherian algebraic

stack and x ∈ |X | is a finite type point, then there exists a unique residual gerbe Gx
at x. Moreover, Gx satisfies the following:
(1) The algebraic stack Gx is regular and the morphism Gx ,→ X is a locally closed
immersion.
(2) If in addition X is of finite type over a field k and x ∈ X (k) has a smooth
affine stabilizer Gx , then Gx ∼
= BGx .
(3) If in addition X is a noetherian algebraic space, then Gx ∼= Spec κ(x) for a
field κ(x), called the residue field of x.

Proof. After replacing X with {x}, we may assume that X is reduced and x ∈ |X |
is dense. Let Spec k → X be a finite type representative of x. By Generic Flatness
(3.3.32), Spec k → X is flat. Therefore, the image {x} ⊆ |X | of Spec k → X is open
(Exercise 3.3.27). The corresponding open substack Gx ⊆ X satisfies the properties
of being a residual gerbe. Since Spec k → Gx is fppf and the property of being
regular descends under fppf morphisms (Proposition 2.1.21), Gx is regular.
For the uniqueness, suppose that G and G ′ are reduced, locally noetherian
algebraic stacks with monomorphisms G ,→ X and G ′ ,→ X such that |G| and
|G ′ | are singletons mapping to x. Then G ′′ := G ×X G ′ is a nonempty algebraic
stack with monomorphisms G ′′ → G and G ′′ → G ′ . Let Spec k → G be a finite type
morphism from a field, which exists by Exercise 3.5.15 and is fppf by Generic Flatness
(3.3.32). The base change G ′′ ×G Spec k is a nonempty noetherian algebraic space
equipped with a monomorphism to Spec k, which by Exercise 3.5.17 implies that
G ′′ ×G Spec k ∼ = Spec k. By fppf descent, G ′′ → G is also an isomorphism. Similarly,
G → G is an isomorphism.
′′ ′

For (2), there is a monomorphism of prestacks BGpre x → X : for a k-scheme

T , there is a unique object of BGprex over T , and this object gets mapped to the
x
composition T → Spec k − → X . Similarly, a morphism over T ′ → T corresponds
to a map T → Gx , and this gets mapped to the corresponding morphism in
′

X . Under stackification, this induces a monomorphism BGx → X , and we see

that BGx satisfies the properties of a residual gerbe. Finally, (3) follows from
Exercise 3.5.17.

Exercise 3.5.17 (details). Let X be a reduced noetherian algebraic space such

that |X| is a point. Show that X ∼ = Spec k for a field k. (This will also follow from
later facts, e.g., Theorem 4.5.1.)

Exercise 3.5.18 (technical). Let X be a (possibly non-noetherian) algebraic stack

and x ∈ X be a finite type point such that the stabilizer is unramified (i.e., the
stabilizer group scheme of any representative is unramified). Show that the residual
gerbe exists and is unique. See also [SP, Tag 06G3].

This corollary shows that the orbit O(u) ⊆ U of a finite type point in a smooth
groupoid is locally closed.

Corollary 3.5.19. Let x ∈ |X | be a finite type point of a noetherian algebraic stack

X . If (U, u) → (X , x) is a smooth morphism from a scheme U with u ∈ U a finite

138
type point, then there is a cartesian diagram

O(u)  /U

□ (3.5.20)
  
Gx  /X

where O(u) is identified set-theoretically with the orbit s(t−1 (u)) of the induced
groupoid s, t : R := U ×X U ⇒ U .

Remark 3.5.21. If X = [U/G] is the quotient stack of a smooth affine algebraic

group over a field k acting on a noetherian k-scheme U and u ∈ U (k), there is a
cartesian diagram

Gu  /U

□
  
BGx  / [U/G],

and we recover the familiar fact that orbit Gu ,→ U is locally closed (Proposi-
tion B.1.17(5)).

Exercise 3.5.22. Let X be a noetherian algebraic stack and x ∈ |X | be a finite

type point with smooth stabilizer. Let x : Spec k → X be a representative of x.
Show that dim Gx = − dim Gx .

3.6 Characterization of Deligne–Mumford stacks

After proving the existence of minimal presentations around finite type points
with smooth stabilizers, we prove characterizations of Deligne–Mumford stacks and
algebraic spaces. We use this conclude that Mg is Deligne–Mumford.

3.6.1 Existence of minimal presentations

Theorem 3.6.1 (Existence of Minimal Presentations). Let X be a noetherian
algebraic stack and let x ∈ |X | be a finite type point with smooth stabilizer Gx .
Then there exists a scheme U , a closed point u ∈ U , and a smooth morphism
(U, u) → (X , x) of relative dimension dim Gx from a scheme U such that the diagram

Spec κ(u)  /U

□
  
Gx /X

is cartesian. In particular, if Gx is finite and reduced, there is an étale morphism

(U, u) → (X , x) from a scheme.

Proof. Let (U, u) → (X , x) be a smooth morphism of relative dimension n from a

scheme U such that u ∈ U is a finite type point. By Proposition 3.5.16, the residual
gerbe Gx at x exists, the inclusion Gx ,→ X is locally closed, and Gx is regular of

139
dimension − dim Gx (Exercise 3.5.22). We obtain a cartesian diagram

O(u)  /U

□
  
Gx  / X,

where O(u) is a regular scheme of dimension c := n−dim Gx . Let f1 , . . . , fc ∈ OO(u),u

be a regular sequence generating the maximal ideal at u. After replacing U with
an open affine neighborhood of u, we may assume that each fi is a global function
on U . The closed subscheme W := V (f1 , . . . , fc ) intersects O(u) transversely at
U , i.e., W ∩ O(u) = Spec κ(u) scheme-theoretically. We claim that inductively
applying the Slicing Criterion for Flatness (A.2.9) implies that the composition
W ,→ U → X is flat at u. Indeed, by smooth descent, it suffices to verify that
p2
W ×X U ,→ U ×X U −→ U is flat at a preimage u′ of u under p1 : W ×X U → W . Since
the induced map Spec κ(u) → Gx is flat, so is the base change U ×X Spec κ(u) → O(u),
and therefore the pullbacks of f1 , . . . , fc define a regular sequence in the local ring of
U ×X Spec κ(u) at u′ . Inductively applying the Slicing Criterion for Flatness gives
the claim.
Since Gx is smooth, so is Spec κ(u) → Gx . For flat morphisms, smoothness is a
property that can be checked on fibers and thus (again arguing via smooth descent)
W → X is smooth at u. The statement follows after replacing W with an open
neighborhood of u.
Remark 3.6.2. A smooth presentation p : U → X is called a miniversal at u ∈ U (k)
if TU,u → TX ,p(u) is an isomorphism of k-vector spaces. We will see that the above
presentations are miniversal in Proposition 3.7.6.
If the stabilizer Gx is not smooth, there are two candidates for ‘minimal pre-
sentations.’ There still exists a miniversal presentation (U, u) → (X , x), but its
relative dimension is equal to the dimension of the Lie algebra of Gx (rather than
dim Gx ), and the fiber product Gx ×X U may thus be positive dimensional. For
example, Bµ µp is an algebraic stack in characteristic p (Proposition 6.3.10) and it
can be realized as the quotient of Gm acting on itself via t · x = tp x; here Gm → Bµ
µp
is a miniversal presentation. On the other hand, the above argument extends to
show that there exists an fppf (but not always smooth) morphism (U, u) → (X , x)
such that Gx ×X U ∼ = Spec κ(u). If in addition X has quasi-finite diagonal, U → X
is an fppf and quasi-finite presentation. See also [SP, Tag 06MC].
Exercise 3.6.3 (technical). If X is a (possibly non-noetherian) algebraic stack and
x ∈ X is a finite type point with discrete and unramified stabilizer Gx , show that
there is an étale morphism (U, u) → (X , x) from a scheme U where u ∈ U is a closed
point.
Hint: Replicate the above argument using Exercise 3.5.18.

3.6.2 Equivalent characterizations

Theorem 3.6.4 (Characterization of Deligne–Mumford Stacks). Let X be an
algebraic stack. The following are equivalent:
(1) the stack X is a Deligne–Mumford;
(2) the diagonal X → X × X is unramified; and
(3) every point of X has a discrete and reduced stabilizer group.

140
 If X has quasi-compact diagonal (e.g., X is quasi-separated), then (3) is equivalent
to requiring that every stabilizer is finite and reduced.

Proof. The equivalence (2) ⇔ (3) is essentially the definition of unramified: since
the diagonal X → X × X is always locally of finite type (Exercise 3.3.5), it is
unramified if and only if every geometric fiber (which is either empty or isomorphic
to a stabilizer) is discrete and reduced. It is not hard to see that a Deligne–Mumford
stack has unramified diagonal (Exercise 3.2.8). For the converse, Existence of
Minimal Presentations (Theorem 3.6.1 and Exercise 3.6.3) shows that for every
finite type point x ∈ X , there is an étale morphism U → X from a scheme whose
image contains x. Thus X is Deligne–Mumford. See also [LMB00, Thm 8.1] and
[SP, Tag 06N3].

Corollary 3.6.5. A morphism X → Y of algebraic stacks is relatively Deligne–

Mumford (Definition 3.3.36) if and only if X → X ×Y X is unramified.

Theorem 3.6.6 (Characterization of Algebraic Spaces). Let X be an algebraic stack

whose diagonal is representable by schemes. The following are equivalent:
(1) the stack X is an algebraic space,
(2) the diagonal X → X × X is a monomorphism, and
(3) every point of X has a trivial stabilizer.

Remark 3.6.7. We will remove the pesky hypothesis that ∆X is representable by

schemes in Theorem 4.5.10.

Proof. Condition (2) is equivalent to the condition that X is a sheaf (Exercise 2.4.39(d)).
The implication (1) ⇒ (2) follows from the definition of an algebraic space. For the
converse, if X is a sheaf, then Existence of Minimal Presentations (3.6.1) implies
that there exists a surjective, étale, and representable morphism U → X from a
scheme. Since ∆X is representable by schemes, so is U → X . The equivalence (2)
⇔ (3) follows from the fact that a group scheme of finite type is trivial if and only
if every fiber is trivial (Proposition B.1.8).

Corollary 3.6.8. Let G → S be a smooth affine group scheme acting on an algebraic

space U over S. Then
(1) [U/G] is Deligne–Mumford if and only if every point of U has a discrete
and reduced stabilizer group, or equivalently if and only if the action map
G × U → U × U is unramified.
(2) [U/G] is an algebraic space if and only if every point of U has a trivial
stabilizer group, or equivalently if and only if the action map G × U → U × U
is a monomorphism.

Corollary 3.6.9 (Characterization of Representable Morphisms). Let X → Y be a

morphism of noetherian algebraic stacks whose diagonal is representable by schemes.
Then X → Y is representable if and only if for every geometric point x ∈ X (k), the
map Gx → Gf (x) on automorphism groups is injective.

Corollary 3.6.10. For g ≥ 2, Mg is a Deligne–Mumford stack of finite type over

Z with affine diagonal.

141
Proof. It only remains to show that Mg is Deligne–Mumford, and by the Charac-
terization of Deligne–Mumford Stacks (3.6.4), it suffices to show that every smooth,
connected, and projective curve C over a field k has a discrete and reduced automor-
phism group scheme Aut(C). In other words, we need to show that the dimension
of the Lie algebra TAut(C),e is zero. Consider the diagrams

Spec _ k
e / Aut(C) / Mg Spec _ k
[C]
/3 M g
9 C
τ
□ and 
τ
    
Spec k[ϵ] / Spec k ([C],[C])/ Mg × Mg Spec k[ϵ] / Spec k.
p

A lifting τ : Spec k[ϵ] → Aut(C) of the left diagram, i.e., a tangent vector τ ∈
TAut(C),e , translates on the right diagram to an automorphism τ of the trivial first
order deformation [C] ◦ p. By Deformation Theory (C.2.4), the automorphism group
of the trivial first order deformation is identified with H0 (C, TC ), but this vector
space is zero since the degree of TC = Ω∨ C is 2 − 2g < 0.

3.7 Smoothness and the Infinitesimal Lifting Criteron

Truth is ever to be found in simplicity, and not in the multiplicity and
confusion of things.

Isaac Newton

The Infinitesimal Lifting Criteria (A.3.1(3), A.3.2(5), and A.3.3(4)) for schemes
provides functorial characterizations of morphisms that are smooth, étale, or unram-
ified. We provide below a generalization to algebraic stacks (Theorem 3.7.1). As an
application, we show that the moduli stacks Mg of smooth curves and Cohr,d (C)
of coherent sheaves on a fixed smooth curve C are smooth (Propositions 3.7.7
and 3.7.9).

3.7.1 Infinitesimal Lifting Criteria

Recall that smoothness is well-defined for morphisms of algebraic stacks since
smoothness is a smooth local property on the source and target. On the other hand,
étaleness and unramifiedness are smooth local on the target, but only étale-local
on the source, and thus well-defined for morphisms X → Y that are relatively
Deligne–Mumford, i.e., if every base change X ×Y T by a map T → Y from a scheme
is Deligne–Mumford; see Definition 3.3.36. A morphism X → Y of algebraic stacks
is étale (resp., unramified) if it is relatively Deligne–Mumford and every base change
X ×Y T → T is étale (resp., unramified); see Definition 3.3.37. We also recall that
X → Y is relatively Deligne–Mumford if and only if the diagonal X → X ×Y X is
unramified (Corollary 3.6.5).

Theorem 3.7.1 (Infinitesimal Lifting Criteria for Smoothness/Étaleness/Unramifedness).

Let f : X → Y be a locally of finite type morphism of locally noetherian algebraic
stacks with quasi-compact and separated diagonals. Consider a 2-commutative dia-

142
gram
/
Spec A
_ 0 ;X
f (3.7.2)
 
Spec A / Y,

of solid arrows where A → A0 is a surjection of artinian local rings with residue

field k such that ker(A → A0 ) ∼
= k. Then
(1) f is smooth if and only if there exists a lifting of every diagram (3.7.2);
(2) f is étale if and only if there exists a lifting, which is unique up to unique
isomorphism, of every diagram (3.7.2),
(3) f is unramified if and only if every two liftings of a diagram (3.7.2) are uniquely
isomorphic; and
(4) f has unramified diagonal if and only if every automorphism of a lifting of a
diagram (3.7.2) is trivial.
Remark 3.7.3. To be explicit, a lifting of a 2-commutative diagram

S
x /X
g | α
f (3.7.4)
 
T / Y,
y

∼ ∼
is a triple (e
x, β, γ) where x
e : T → X is a map and β : x → x
e ◦ g and γ : f ◦ x
e→y
are 2-isomorphisms such that

f ◦x
f (β) α

/ y ◦ g
∗
g γ
f ◦x
e◦g

commutes. We may view a lifting as a 2-commutative diagram

S
x
/X
>
β

g x
e f
 γ
 
T / X.
y

∼
A morphism (ex, β, γ) → (ex′ , β ′ , γ ′ ) of liftings is a 2-isomorphism Θ : x e′ such
e→x
that β = g Θ ◦ β and γ = γ ◦ f (Θ).
∗ ′ ′

We can also interpret liftings using the map

Ψ : X (T ) → X (S) ×Y(S) Y(T )

of groupoids. A diagram (3.7.4) defines an object (x, y, α) ∈ X (S) ×Y(S) Y(T ) and
the category of liftings is the fiber category over this object. The criteria above
states that f : X → Y is smooth (resp., is étale, is unramified, has unramified
diagonal) if and only if X (A) → X (A0 ) ×Y(A0 ) Y(A) is essentially surjective (resp.,
an equivalence, fully faithful, faithful).

143
Proof. We handle the criterion for smoothness leaving the remaining cases to the
reader. Suppose that f : X → Y is smooth and that we are given a diagram
(3.7.2). By replacing Y with Spec A and X with X ×Y Spec A, we may assume that
Y = Spec A is affine, and we need to show that a section over Spec A0

9X]

 
Spec A0  / Spec A

extends to a section over Spec A. If X is a scheme, the existence of a lifting is provided

by the Infinitesimal Lifting Criterion for Smoothness (A.3.1) for schemes. If X is an
algebraic space, we may apply Proposition 4.3.4 (proven later but independently) to
construct a smooth presentation U → X from a scheme and a lifting Spec k → U of
Spec k ,→ Spec A0 → X . This gives a diagram

9 A UX


9X

  
Spec k  / Spec A0  / Spec A.

Since U → X is representable by schemes and smooth, applying the Infinitesimal

Lifting Criterion (A.3.1) inductively to the square-zero extensions A0 /mk+1 →
A0 /mk gives a lifting Spec A0 → U of Spec A0 → X . Since U → Spec A is a smooth
morphism of schemes, another application of the Infinitesimal Lifting Criterion gives
a lifting Spec A → U of Spec A0 → U , and the composition Spec A → U → X gives
the desired extension. Finally, if X is an algebraic stack, we can repeat the above
argument by applying Proposition 4.3.4 to construct a smooth presentation U → X
with a lifting Spec k → U , where we use the representability of U → X and the
algebraic space case to construct the lifting Spec A0 → U .
Conversely, suppose that every diagram (3.7.2) for f : X → Y has a lifting.
Choose smooth presentations V → Y and U → XV giving a commutative diagram

U / XV /X

□ f
 
V / Y.

Since U → XV is smooth, by the implication already proven, every diagram (3.7.2)

for U → XV has a lifting. Since every diagram (3.7.2) for XV → V also has a lifting
(as it is the base change of X → Y), so does the composition U → V . As U → V is
a morphism of schemes, the Infinitesimal Lifting Criterion (A.3.1) implies that it
is smooth. Since smoothness is a smooth local property on the source and target,
X → Y is smooth. See also [LMB00, Prop. 4.15] and [SP, Tag 0DP0].
Exercise 3.7.5. Prove the remaining parts of the Infinitesimal Lifting Criteria for
étaleness, unramifiedness, and unramified diagonal.
Hint: For the case of unramified diagonal, relate an automorphism of a lifting for the
morphism f : X → Y to a lifting of the diagonal ∆f : X → X ×Y X . Reduce the case

144
of étaleness to unramifiedness. For unramifiedness, use Corollary 3.6.5 to reduce
to showing that a relatively Deligne–Mumford morphism f : X → Y is unramified if
and only if every two liftings of a diagram (3.7.2) are isomorphic, and establish this
using similar methods to the proof of the smoothness criterion.

As a first application, we see that the presentation produced by the Existence of

Minimal Presentations (Theorem 3.6.1) is miniversal, i.e., induces an isomorphism
on tangent spaces at the chosen preimage, and we can also express the dimension of
a smooth algebraic stack in terms of its tangent space and stabilizer.

Proposition 3.7.6. Let X be a noetherian algebraic stack and x ∈ |X | be a finite

type point with smooth stabilizer. Let f : (U, u) → (X , x) be a smooth morphism
from a scheme such that Gx ×X U ∼ = Spec κ(u). Then U → X is miniversal at u,
i.e., TU,u → TX ,f (u) is an isomorphism of κ(u)-vector spaces. In particular, if X is
smooth over a field k and x ∈ X (F) is a point with smooth stabilizer over a finite
extension F/k, then
dimx X = dim TX ,x − dim Gx .

Proof. Surjectivity of TU,u → TX ,f (u) follows from applying the Infinitesimal Lifting
Criterion (3.7.1) to the smooth morphism U → X . Injectivity follows from the fact
that

Spec κ(u)  /U

□
  
Gx /X

is cartesian. Indeed, if τ : Spec κ(u)[ϵ] → U is an element of TU,u mapping to

0 ∈ TX ,f (u) , then the composition Spec κ(u)[ϵ] → U → X factors through the
residual gerbe Gx and therefore also factors through Spec κ(u). We conclude that
τ = 0. For the final statement, Existence of Minimal Presentations (3.6.1) produces
a smooth morphism f : (U, u) → (X , x) of relative dimension dim Gx such that
Gx ×X U ∼ = Spec κ(u). Therefore, dimx X = dimu U − dim Gx , and since U is smooth
at u, dimu U = dim TU,u = dim TX ,x .

3.7.2 Smoothness of moduli stacks

Combining the Infinitesimal Lifting Criterion for Smoothness (3.7.1) with deformation
theory allows both for the verification of smoothness of a moduli stack and the
computation of its dimension.

Proposition 3.7.7. For g ≥ 2, the Deligne–Mumford stack Mg is smooth over

Spec Z of relative dimension 3g − 3.

Proof. Let Spec k → Mg be a morphism from a field k corresponding to smooth,

connected, and projective curve C over k. Consider a diagram

[C]

Spec k / Spec A0 /* Mg
_ 9
(3.7.8)
 
Spec A / Spec Z,

145
where A → A0 is surjection of artinian local rings with residue field k such that
k = ker(A → A0 ). The map Spec A0 → Mg corresponds to a family of curves
C0 → Spec A0 , and a lifting in (3.7.8) translates to an extension
 
C / C0  /C

    
Spec k  / Spec A0  / Spec A

of solid arrows. By Deformation Theory (C.2.4), there is a cohomology class

obC ∈ H2 (C, TC ) such that the above diagram has an extension if and only if
obC = 0. Since C is a curve, H2 (C, TC ) = 0; thus, obC = 0 and there is a
lifting. Finally, Deformation Theory (C.2.4) also gives the identification TMg ,[C] =
H1 (C, TC ), which has dimension 3g − 3 by a Riemann–Roch computation (see
Example 3.5.8). Since dim Aut(C) = 0, we conclude using Proposition 3.7.6 that
dim[C] (Mg ×Z k) = 3g − 3.

Proposition 3.7.9. For a smooth, connected, and projective curve C over an

algebraically closed field k, the algebraic stack Cohr,d (C) is smooth over Spec k of
dimension r2 (g − 1).

Proof. By Exercise 3.7.10, Cohr,d (C) is nonempty. Let [F ] ∈ Cohr,d (C)(F) be a

coherent sheaf on CF = C ×k F of rank r and degree d. Let A → A0 be a surjection of
artinian local rings with residue field F such that F = ker(A → A0 ). We need to check
that every coherent sheaf F0 on CA0 that restricts to F extends to a coherent sheaf
F on CA . By Deformation Theory (C.2.11), there is an element obF ∈ Ext2OC (F, F )
such that obF = 0 if and only if there exists an extension. Since C is a smooth
curve, Ext2OC (F, F ) = H2 (CF , F ⊗ F ∨ ) = 0. Deformation theory also provides an
identification TCohr,d (C),[F ] = Ext1OC (F, F ) and a Riemann–Roch calculation yields
that dim Ext1OC (F, F ) = dim Aut(F ) + r2 (g − 1) (see Example 3.5.9). Therefore,
dim[F ] Cohr,d (C) = dim Ext1OC (F, F ) − dim Aut(F ) = r2 (g − 1).

Exercise 3.7.10. For every line bundle L and rank r > 0, show that there exists a
vector bundle E of rank r with det E ∼
= L.

3.8 Properness and the Valuative Criterion

Working with noncompact spaces is like trying to keep change with holes
in your pockets.

Angelo Vistoli

After defining universally closed, separated, and proper morphisms of algebraic

stacks, we prove valuative criteria (Theorem 3.8.7) providing a generalization of the
usual valuative criteria for schemes (Theorem A.4.5). These valuative criteria are
essential in moduli theory. They are applied in this book to show that Mg is proper
(Theorem 5.5.23) and Bun ss r,d (C) is universally closed.

146
3.8.1 Definitions
With some care, we define separatedness and properness for morphisms of algebraic
stacks. Recall from Definition 3.3.11 that a representable morphism X → Y of
algebraic stacks is separated if the diagonal X → X ×Y X (which is representable by
schemes) is proper.
Definition 3.8.1.
(1) A morphism X → Y of algebraic stacks is universally closed if for every
morphism Y ′ → Y of algebraic stacks, the morphism X ×Y Y ′ → Y ′ induces a
closed map |X ×Y Y ′ | → |Y ′ |.
(2) A representable morphism X → Y of algebraic stacks is proper if it is universally
closed, separated, and of finite type.
(3) A morphism X → Y of algebraic stacks is separated if the representable
morphism X → X ×Y X is proper.
(4) A morphism X → Y of algebraic stacks is proper if it is universally closed,
separated, and of finite type.
For a morphism of schemes, properness is equivalent to the diagonal being a
closed immersion. This is also true for algebraic spaces as proper monomorphisms of
algebraic spaces are closed immersions (Corollary 4.5.14). This fails for morphisms
of algebraic stacks as the diagonal need not be a monomorphism. Recall that the
stabilizer Gx of a field-valued point x : Spec k → X is given by the cartesian diagram
Gx / Spec k

□ (x,x)
 
X / X × X.

If X is a separated algebraic stack over a scheme S, then Gx is a proper group

algebraic space over k, and even a group scheme by Theorem 4.5.28. If X is separated
and has affine diagonal, then Gx is proper and affine, thus finite.
Example 3.8.2. If G is an abstract finite group, then BG → Spec Z is proper. If
G → S is a proper group scheme (e.g., an abelian scheme), then BGm → S proper.
On the other hand, if G → S is a non-finite affine group scheme (e.g., Gm → Spec Z),
then BG → S is universally closed but not separated.
Exercise 3.8.3 (easy). Show that M(0,0) , M(0,1) , M(0,2) , and M(1,0) are not
separated.
Example 3.8.4 (Bun r,d (C) is not separated). Since Bun r,d (C) has affine diagonal
(Example 3.3.15) and infinite automorphism groups, Bun r,d (C) is not separated over
the base field k.
Remark 3.8.5. As universal closedness is a smooth local property on the target,
X → Y is universally closed if and only if X ×Y T → T is closed for all maps
T → Y from schemes, or equivalently X ×Y V → V is universally closed for a smooth
presentation V → Y. In the proof of Theorem 3.8.7, we will show that when Y is
noetherian, it suffices to consider finite type base changes T → Y from schemes.
Exercise 3.8.6. An action of an algebraic group G over a field k on an algebraic
space U is called proper if the action map
Ψ : G × U → U × U, (g, u) 7→ (gu, u)
is proper.

147
 (a) Show that the action of G on U is proper if and only if [U/G] is separated.
(b) For u ∈ U (k), let Ψu : G → U be the map defined by g 7→ gu (viewing Ψ as a
morphism over U via the projections on the second components, then Ψu is
the fiber of Ψ over u). Show that the following are equivalent:
(i) Ψu : G → U is proper,
(ii) u : Spec k → [U/G] is proper,
(iii) Gu ⊆ U is closed and Gu is proper.

Hint for (b): To show that (i) or (ii) implies (iii), replace U with the reduced orbit
(Gu)red , use Generic Flatness (3.3.32) to show that the morphisms Φu or u are
faithfully flat, and then use fppf descent. See also [GIT, Lem. 0.3].

3.8.2 Valuative Criteria

For moduli problems, the valuative criterion for properness translates to the geometric
question of whether an object over the fraction field of a DVR R (which we think of
as a punctured curve) extends uniquely to a family over the DVR (or curve). In
the formulation below, we will use the notions of liftings of 2-commutative diagrams
and their morphisms as defined formally in Remark 3.7.3.
Theorem 3.8.7 (Valuative Criteria for Properness/Universal Closedness/Separatedness).
Let f : X → Y be a quasi-compact and quasi-separated morphism of locally noetherian
algebraic stacks. Consider a 2-commutative diagram

Spec K /X

 α f (3.8.8)
 
Spec R /Y

where R is a DVR with fraction field K. Then

(1) f is universally closed if and only if for every diagram (3.8.8), there exists an
extension R → R′ of DVRs with K → K ′ = Frac(R) of finite transcendence
degree and a lifting

Spec K ′ / Spec K 5/ X
f (3.8.9)
  
Spec R′ / Spec R / Y;

(2) f is proper if and only if f is finite type and for every diagram (3.8.8), there
exists an extension R → R′ of DVRs with the map K → K ′ on fraction fields
having finite transcendence degree and a lifting as in (3.8.9), which is unique
up to unique isomorphism;
(3) f is separated if and only if every two liftings of a diagram (3.8.8) are uniquely
isomorphic; and
(4) f has separated diagonal if and only if every automorphism of a lifting of a
diagram (3.8.8) is trivial.
Before proving this theorem in the next subsection, we provide some examples.
Exercise 3.8.10 (good practice). Use the valuative criterion to show the following:

148
 (a) If G is an abstract finite group, then BG → Spec Z is proper.
(b) The map BGm → Spec Z is universally closed but not separated.

Exercise 3.8.11 (hard). Show that the stack M1,1 of stable elliptic curves intro-
duced in Exercise 3.1.19(c) is proper over Spec Z.
We will prove later that Mg,n is proper as long as g ≥ 2, or g = 1 and n ≥ 1, or
g = 0 and n ≥ 3 (Theorem 5.5.23).

Example 3.8.12 (Properness of Mg,n ). The moduli stack Mg,n is proper over
Spec Z (Theorem 5.5.23) as long as 2g − 2 + n > 0 (i.e., we exclude (g, n) = (0, 0),
(0, 1), (0, 2), and (1, 0)). We prove this in §5.5.1 in characteristic 0 by verifying the
valuative criterion.

Exercise 3.8.13 (good practice). For a smooth, geometrically connected, and

projective curve C over a field k, show that the algebraic stacks Cohr,d (C) and
Bun r,d (C) satisfy the valuative criterion for universally closedness. Observe that in
this case the valuative criterion holds without requiring an extension of DVRs.

Remark 3.8.14. Note that the above exercise does not imply that Bun r,d (C) is
universally closed. Indeed, the Valuative Criterion for Universal Closedness requires
quasi-compactness and Bun r,d (C) is not quasi-compact (see Exercise 8.2.3). In
fact, a universally closed morphism of schemes is necessarily quasi-compact [SP,
Tag 04XU], and presumably the same holds for morphisms of algebraic stacks. The
stack of vector bundles over a curve and the stack of all curves are not universally
closed (see Exercises 3.8.15 and 5.4.18).

Exercise 3.8.15.
(a) Show that Bun r,d (C) is not universally closed
(b) For an integral proper scheme X over k of arbitrary dimension, show that the
stack Cohtf of torsion free sheaves on X satisfies the valuative criterion for
universally closedness.

Are base changes necessary? For the Valuative Criterion for Properness (A.4.5)
for a morphism X → Y of schemes, it is not necessary to allow extensions of the
DVR, i.e., there exists a unique lift

Spec K /X
;

 
Spec R / Y.

The same holds for morphisms of algebraic spaces [SP, Tag 0A40]. For general
morphisms of algebraic stacks, it is necessary to allow extensions.

Example 3.8.16. Consider the Deligne–Mumford stack X = Bµ µn over a field k

whose characteristic is prime to n. Let R = k[x](x) with fraction field K = k(x).
Then Spec K → Spec K, given by x 7→ xn , is a non-trivial principal µ n -torsor. The
classifying map Spec K → Bµ µn does not extend to a map Spec R → Bµ µn (as all
µ n -torsors over Spec R are trivial). Note however that the principal µ n -bundle
Spec K → Spec K becomes trivial after base changing by the field extension K(x1/n )
of K, and therefore the composition Spec K(x1/n ) → Bµ µn extends to the map
Spec R[x1/n ] → Bµ µn classifying the trivial bundle.

149
Example 3.8.17. Let X = Spec R be the spectrum of a DVR and X be the nth
root stack X( n OX /π) (Example 3.9.22). The morphism X → X is an isomorphism
p

over the generic point, but the section Spec K → X does not extend to a global
section X → X .
If G is a special algebraic group over a field (i.e., every principal G-bundle is
Zariski-locally trivial) such as SLn or GLn , then BG satisfies the valuative criterion
for universal closedness without a base change: any map Spec K → BG corresponds
to the trivial principal G-bundle and thus extends to a map Spec R → BG. On the
other hand, base changes are necessary for BPGLn .
Exercise 3.8.18 (hard). Show that there is a principal PGLn -bundle over the
fraction field of a DVR that does not extend to the DVR.
For the valuative criterion of properness for Mg , extensions of the DVR are
necessary (see Example 5.5.11). On the other hand, the valuative criterion for
universal closedness for Bun r,d (C) holds without extensions (Exercise 3.8.13).

3.8.3 Proof of the Valuative Criteria

We modify the proof of the valuative criterion for schemes given in §A.4. The
starting point is a lifting criterion for closed morphisms generalizing Lemma A.4.1.
Lemma 3.8.19. Let f : X → Y be a quasi-compact morphism algebraic stacks.
Then f is closed if and only for every point x ∈ |X |, every specialization f (x) ⇝ y0
lifts to a specialization x ⇝ x0 .
Proof. If f is closed, then f ({x}) is closed and equal to {f (x)}. Every point
y0 ∈ {f (x)} = f ({x}) is therefore the image of a specialization of x. For the
converse direction, after replacing X with any closed substack, it suffices to show
that f (|X |) is closed. By Exercise 3.3.31(a), it suffices show that f (|X |) is closed
under specialization.
Specializations are induced by maps from DVRs, just as in the case of schemes
(Proposition A.4.4).
Proposition 3.8.20. If f : X → Y is a finite type morphism of noetherian algebraic
stacks, x ∈ |X |, and f (x) ⇝ y0 is a specialization, then there exists a diagram

Spec K /X x
_
f
  
Spec R /Y f (x) / y0 ,

where R is a DVR with fraction field K, the image of Spec K → X is x, and

Spec R → Y realizes the specialization f (x) ⇝ y0 . In particular, every specialization
x ⇝ x0 in a noetherian algebraic stack is realized by a map Spec R → X from a
DVR.
Proof. Let V → Y be a smooth presentation and v0 ∈ V be a preimage of y0 . Since
V → Y is smooth, it is an open morphism (Exercise 3.3.27), and thus there exists
a specialization v ⇝ v0 over f (x) ⇝ y0 (Exercise 3.3.31(b) ). Let x′ ∈ |XV | be a
preimage of v ∈ V and x ∈ |X |. Let U → XV be a smooth presentation and u ∈ U
be a preimage of x′ . Applying Proposition A.4.4 to the morphism U → V of schemes
and u 7→ v ⇝ v0 gives the desired diagram.

150
Proof of Theorem 3.8.7. The quasi-separatedness of f : X → Y means that the
diagonal ∆f : X → X ×Y X and the double diagonal ∆∆f : X → X ×X ×Y X X
are quasi-compact. The quasi-compactness of f , ∆f , and ∆∆f are needed in the
proof for the implication that the valuative criterion implies universally closedness,
separatedness, and separated diagonal, respectively.
We first establish the criterion for universal closedness. Suppose that the valuative
criterion holds and that f : X → Y is not universally closed. Since universal
closedness is a smooth local property on the target, we may assume that Y = Y is a
noetherian scheme and that there is a map T → Y of schemes such that fT : XT → T
is not closed. We will reduce to the case that T → Y is a finite type morphism. By
Lemma 3.8.19, there exists z ∈ |XT | and a specialization fT (z) ⇝ t0 which doesn’t lift
to a specialization z ⇝ z0 . This implies that Z = {z} ⊆ XT has trivial intersection
with the fiber (XT )t0 . If p : X → X is a smooth presentation, then the preimage Z
of Z under XT → XT does not meet the fiber (XT )t0 . Applying Lemma A.4.8 to
XT → T shows that after replacing T with an open neighborhood of t0 , there exists
g
a factorization T −→ T ′ → Y and a closed subscheme Z ′ ⊆ XT ′ such that T ′ → Y is
of finite type, Z ∩ (XT ′ )g(t0 ) = ∅, and im(Z ,→ XT → XT ′ ) ⊆ Z ′ . Letting z ′ ∈ |XT ′ |
be the image of z ∈ |XT |, we have that z ′ maps to g(fT (z)) ∈ T ′ and that there is
a specialization g(fT (z)) ⇝ g(t0 ) which does not lift to a specialization of z ′ . By
Lemma 3.8.19, XT ′ → T ′ is not closed.
If T → Y is a finite type morphism, the base change XT → T is a finite type
morphism of noetherian algebraic stacks which also satisfies the valuative criterion
for universal closedness. It therefore suffices to show that f : X → Y is closed. By
Lemma 3.8.19, we need to show that given a point x ∈ |X |, every specialization
f (x) ⇝ y0 lifts to a specialization x ⇝ x0 . By Proposition 3.8.20, there exists a
diagram

Spec K /X x
_
f (3.8.21)
  
Spec R /Y f (x) / y0 .

The valuative criterion implies that after replacing R with an extension there exists
a lift Spec R → X , which in turn yields a specialization x ⇝ x0 lifting f (x) ⇝ y0 .
Conversely, assume that f : X → Y is universally closed and that we are given a
diagram as in (3.8.21). By replacing Y with Spec R and X with X ×Y Spec R, we
may assume that Y = Spec R and that we have a diagram

Spec K /X

f
% 
Spec R.

By replacing X with {x}, we may assume that X is integral with generic point
x. Since X → Spec R is closed, there exists a specialization x ⇝ x0 mapping to
the specialization of the generic point to the closed point in Spec R. Since f is
quasi-separated, Spec K → X is quasi-compact. Applying Proposition 3.8.20 to

151
Spec K → X yields a DVR R′ with fraction field K ′ and a commutative diagram

Spec K ′ / Spec K

  %
Spec R′ /X f
/ Spec R

such that Spec R′ → X realizes the specialization x ⇝ x0 . As Spec R′ → Spec R

is surjective,R → R′ is an extension of DVRs and Spec R′ → X provides a lift of
(3.8.21).
The criterion for separated diagonal is left for the reader (Exercise 3.8.22).
Assuming that f has separated diagonal, the valuative criterion for the separatedness
of f : X → Y translates to the valuative criterion for the universal closedness of
the diagonal ∆f : X → X ×Y X . Thus, the valuative criteria for properness and
separatedness follow from the valuative criterion for universal closedness. See also
[LMB00, Thm. 7.10], [Fal03, §4], and [SP, Tags 0CLQ, 0CLS, 0CLV, and 0CLY].

Exercise 3.8.22. Verify the valuative criterion for separated diagonal.

Try to use a similar argument as for the Infinitesimal Lifting Criterion for Unramified
Diagonal (Exercise 3.7.5), using instead that ∆f is separated if and only if ∆∆f
is a closed immersion (Exercise 3.3.38), and noting that since ∆∆f is a quasi-
compact monomorphism representable by schemes, the Valuative Criterion (A.4.5)
for schemes applies to verify that it is a closed immersion.

3.9 Further examples

The extreme possibilities are the most illuminating.

David Mumford

This section provides examples of algebraic spaces (§3.9.1), Deligne–Mumford

stacks (§3.9.2), and algebraic stacks (§3.9.3), as well as some extremely non-separated
examples (§3.9.4).

3.9.1 Examples of algebraic spaces

Example 3.9.1. As discussed in Example 0.5.7, there exists a smooth proper
complex 3-fold U with a free action of Z/2 = {±1} such that there is an orbit
{u, −1 · u} not contained in an affine open subscheme. The quotient sheaf X =
U/(Z/2) is an algebraic space (Corollary 3.1.14), but is not a scheme. If X were a
scheme, there would be an affine open around the image of u and the preimage of
this open under the finite morphism U → X would be an affine open containing the
orbit.

Example 3.9.2 (Bug-eyed cover). Consider the étale equivalence relation (p2 , σ) : R =
(Z/2 × A1 ) ∖ {(−1, 0)} ⇒ A1 defined in Example 3.4.4 obtained by restricting the
Z/2-action on A1 where (−1) · x = −x. Let X = A1 /R be the algebraic space
quotient.

152
 noidentification

A
identified

ee

A
Figure 3.9.3: X → A1 is a bug-eyed cover.

Caption A
is a eyedcover
bag
We claim that the algebraic space quotient X = A1 /R is not a scheme. To
achieve this, we will show that the diagonal X → X × X is not a locally closed
immersion. This follows from the cartesian diagram

(A1 ∖ 0) ⨿ {0} /R /X

□ (p2 ,σ) □ ∆
 t7→(t,−t)
 
A1 / A1 × A1 / X ×X
an
as the vertical arrow on the left is transparently not a locally closed immersion.
The algebraic space X is smooth of dimension one and looks eerily similar to
the affine line. This example first appeared in [Art74a, p. 311]. The terminology
ic where more general
of a bug-eyed cover was coined in [Kol92, §4.2.2], covers were
have
ates
considered: if Y → Z is a generically étale morphism of schemes with étale locus
Y ét , then Z bug = Y /R → Z is a bug-eyed cover, where R = Y ⨿ (Y ét ×Z Y ét ∖ ∆ét ),
where ∆ét is the diagonal of Y ét → Z.

Exercise 3.9.4 (Alternative descriptions of the bug-eyed cover). Let X be the

bug-eyed cover from the previous example.
(a) Show that X is isomorphic to the quotient U/(Z/2) of the free Z/2-action on
the non-separated affine line U = A1 A1 ∖0 A1 over k by swapping the origins
S
and by (−1) · x = −x for x ̸= 0.
(b) Show that there is a universal homeomorphism X → A1 which is an isomor-
phism away from the origin but ramified over the origin.
(c) Show that every map to a scheme X → Z factors through X → A1 . (In other
words, A1 is the categorical quotient of U by Z/2 in the category of schemes
but not in the category of algebraic spaces.)
(d) (hard) Consider the SL2 action on Vd = Symd k2 , the space of homogeneous
polynomials in x and y of degree d. Let W ⊆ V1 × V4 be the reduced, locally
closed subscheme defined as the set of pairs (L, F ) such that L ̸= 0 and F is
the square of a homogeneous quadratic form of discriminant 1. Show that the
induced SL2 -action on W is free and that X ∼= W/ SL2 .
While the descriptions of X as A1 /R and U/(Z/2) may seem pathological, this
exercise shows that X also arises as a quotient of a quasi-affine variety by SL2 .
See also [GIT, Ex. 0.4] and [Kol97, Ex. 2.18].

153
 A

ee

2 2 2 11121,0
section
identity

A
(e) (hard) Let π : Y → A1 be a proper flat
a family of surfaces which is smooth over
A1 ∖ 0. Assume also that theiscentral fiber Y0 has an isolated nodal singularity
at a point p given locally by z 2 + xy = t, where t is a local coordinate of A1 .
1AShow that the functor outx
F : (Sch /A1 ) → SetsA bageyedcover
Caption is a
(T → A1 ) 7→ {simultaneous resolutions Ye → Y ×A1 T },
where a simultaneous resolution is a proper morphism which is a minimal
resolution on every geometric fiber, is representable by the bug-eyed cover X.
See also [Art74a, p. 331].
Exercise 3.9.5. Let V = Spec k[x, y]/(xy) and U = V V ∖0 V be the non-separated
S
union. Let Z/2 act on U exchanging the branches and swapping the origins, and let
X = U/(Z/2) be the quotient algebraic space.

noidentification
an
Figure 3.9.6: The affine line with a nodal singularity at the origin.
A
identified

(a) Show that the diagonal of X is not a locally closed immersion, and conclude
that X is not a scheme.
ic
(b) Show that X has dimension one with a singularity at the origin, and that
have
ee
there is a universal homeomorphism X → A1 which is ates
an isomorphism away
from the origin.
(c) Generalize this example by constructing an algebraic space, which looks like

A
the affine line with an Ak singularity (i.e., y 2 = xk + 1) at the origin.
(d) Construct an affine line with the (non-lci) singularity k[[x, y, z]]/(xy, yz, xz)
(three coordinate axis in A3 ) at the origin.
Exercise Caption A bag
3.9.7. Let Z/2 = {±1}isacta on A1C eyed cover
via conjugation over Spec R. Let
R = (Z/2 × AC ) ∖ {(−1, 0)} ⇒ U be the induced étale equivalence relation. Show
1

that the algebraic space X = A1C /R is not a scheme. Observe also that, unlike
the bug-eyed cover and the previous example, the diagonal of X is a locally closed
immersion. (The condition that the diagonal be a locally closed immersion is
sometimes referred to as locally separated, which is a different notion that being
Zariski locally separated.)
S Identify X with the quotient U/(Z/2) of the non-separated affine line U =
Hint:
A1C A1 ∖0 A1C by the Z/2-action that acts via conjugation and swaps the origins. If
C an
there were an affine neighborhood of the origin, show that the preimage would be an
affine neighborhood containing both neighborhoods.

ic have
ates

Figure 3.9.8: The quotient X = A1 /R looks like A1R except that the origin has residue
field C.

Remark 3.9.9. Typical constructions in algebraic geometry often only exist in the
category of algebraic spaces. The quotient of a free action of a finite group on a

154
 A
Caption A is a bageyedcover
scheme exists as a scheme if every orbit is contained in an affine (Exercise 4.2.14),
but not in general (Example 3.9.1). Similarly, the pushout X ⨿X0 Y0 of a closed
immersion X0 ,→ X and finite (or even affine) morphism X0 → Y0 always exists as
an algebraic space [Art70, Thm. 6.1], but is not always a scheme, unless for instance
the preimage of any point in Y0 in X0 is contained in an affine (Theorem B.4.1).
While Castelnuovo proved that curves with negative self-intersection in smooth
surfaces can be contracted (Theorem B.2.6), contractions in higher dimension do
not always exist as scheme. Artin [Art70, Thm. 3.1] (see also [Maz75]) provided
conditions for when they exist as algebraic spaces. Likewise, simultaneous resolutions
of singularities of families of surfaces (as in Exercise 3.9.4(e)) do not always exist as
schemes, but do exist as algebraic spaces an[Art74a, Thm. 1].

3.9.2 Examples of Deligne–Mumford stacks

Example 3.9.10 (Classifying stacks). If G is an finite abstract group scheme over
a field k, then the classifying stack BG of G is the stack defined as the category of
pairs (T, P ) where T is a scheme and P c have
→ T is a G-torsor (Definition 2.4.15). Then
BG is a smooth and proper Deligne–Mumford stack over k of dimension 0. Is
Example 3.9.11. Suppose char(k) ̸= 2. Let Z/2 act on A2 via −1·(x, y) = (−x, −y).
Show that [A2 /(Z/2)] is a smooth Deligne–Mumford stack over a field k and that
there is a proper and bijective morphism [A2 /(Z/2)] → Y where Y is the singular
variety Spec k[x2 , xy, y 2 ] defined by the Z/2-invariants of Γ(A2 , OA2 ).

Example 3.9.12 (Quasi-finite but non-finite inertia). Let G = A1 ⨿ (A1 ∖ 0) → A1

be the group scheme where every fiber is Z/2 except over the origin where it is
trivial. Then BG → A1 is a Deligne–Mumford stack with quasi-finite but not finite
diagonal, and hence is not separated. Similarly, the Deligne–Mumford locus of the
moduli stack [Sym4 P1 / PGL2 ] of four unordered points in P1 is not separated (see
Example 4.4.17).

742 trivial
Figure 3.9.13: The stabilizer group jumps down over the origin.

Example 3.9.14 (Weighted projective stacks). For a tuple of positive integers

(d0 , . . . , dn ), let Gm act on An+1 via t · (x0 , . . . , xn ) = (td0 x0 , . . . , tdn xn ). We define
the weighted projective stack as

P(d0 , . . . , dn ) = [(An+1 ∖ 0)/Gm ].

If the di are all 1, then we recover projective space Pn ; otherwise, P(d0 , . . . , dn ) is

not an algebraic space. For example, over Z[1/6] the stack of stable elliptic curves
M1,1 is isomorphic to P(4, 6) by Exercise 3.1.19(c).
More generally, if R is a finitely generated positively graded k-algebra, we can
define stacky proj as Proj R = [(Spec(R) ∖ 0)/Gm ], where Gm acts such that the
weight of xi is the same as its degree.

Exercise 3.9.15.

155
 (a) If k is a field of characteristic p, show that P(d0 , . . . , dn ) (or more precisely
P(d0 , . . . , dn )) ×Z k) is a Deligne–Mumford stack over k if and only if p doesn’t
divide each di .
(b) Classify all the points of P(3, 3, 4, 6) that have non-trivial stabilizers.
(c) We say that an algebraic stack X has generically trivial stabilizer if there exists
a dense open substack U ⊂ X which is an algebraic space. Provide conditions
for when P(d0 , . . . , dn ) has generically trivial stabilizer.
(d) Show that there is a bijective morphism P(d0 , . . . , dn ) to weighted projective
space Proj k[x0 , . . . , xn ], where xi has degree di . (This is an example of a
coarse moduli space.)
Example 3.9.16 (Stacky curves). A stacky curve is a one dimensional Deligne—
Mumford stack of finite type over a field k. For positive integers m and n, the
weighted projective stack P(m, n) is an example of a smooth and proper stacky
curve as long as the characteristic is relatively prime to m and n.
Exercise 3.9.17.
(a) Show that a smooth stacky curve has abelian stabilizers.
(b) Generalize this to nodal stacky curves.
Example 3.9.18 (Football curves). For integers m and n, the football curve F(m, n)
is the proper stacky curve obtained by gluing [A1 /µm ] and [A1 /µn ] along [(A1 ∖
0)/µm ] ∼
= A1 ∖ 0 ∼
= [(A1 ∖ 0)/µn ]. The topological space |F(m, n)| is identified with
|P |, and F(m, n) → P1 is an example of a coarse moduli space.
1

Exercise 3.9.19. Show that F(m, n) ∼

= P(m, n) if and only if m and n are relatively
prime.
Remark 3.9.20 (Uniformization of stacky curves). In [BN06], it is shown that every
smooth and separated stacky curve over C has a universal cover which is a simply
connected, smooth, and separated stacky curve with generically trivial stabilizers
and is isomorphic to H, C, or P(m, n) for gcd(m, n) = 1.
Root gerbes and root stacks are important examples that were first introduced
in [Cad07, §2].
Example 3.9.21 (Root gerbes). Let X be a scheme and L be a line bundle.
This data determines a morphism [L] : X → BGm . For a positive integer r, let
r : BGm → BGm be the morphism induced from the rth power map r : Gm → Gm ,
where t 7→ tr ; alternatively r : BGm → BGm is defined functorially on objects
√ by
the assignment M 7→ M ⊗r on p line bundles. Define the rth root gerbe X( r L) of X
and L (sometimes denoted as r L/X) as the fiber product
√
X( r L) / BGm

π □ r
 [L] 
X / BGm .
√
Gerbes are discussed later in §6.4.3. The map X( r L) → X is an example of a
banded µ r -gerbe (Exercise 6.4.25) and, in particular, is a coarse moduli space.
Example 3.9.22 (Root stacks). Let X be a scheme, L be a line bundle, and
s ∈ Γ(X, L) be a section. This data determines a morphism [L, s] : X → [A1 /Gm ]

156
(see Example 3.9.25 below). Let r : [A1 /Gm ] → [A1 /Gm ] be the morphism induced
from the rth power map r : A1 → A1 , given by x 7→ xr , which is equivariant under
the rth power map r : Gm → Gm ; alternatively r : [A1 /Gm ] → [A1 /Gm ] is defined
functorially by (M, t) 7→ (M ⊗r , t⊗r )√on line bundles and sections. For a positive
integer r, define the rth root stack X( r L, s) of X and L along s (sometimes denoted
as (L, s)/X) as the fiber product
pr

√ / [A1 /Gm ]
X( r L, s)
π □ r
 [L,s]

X / [A1 /Gm ].

√
The structure map π : X( r L, s) → X is an example of a coarse moduli space.
√
if s = 0 is the zero section, then X( r L, 0) is not isomorphic to the root
Caution: √
gerbe X( r L), even though they have the same reduced structures and same coarse
moduli space.

Exercise 3.9.23. Let S be a scheme and r be an integer invertible in Γ(S, OS ).

(This hypothesis ensures that µ r,S → S is an étale group scheme; it will be removed
in Exercise 6.4.25.)
√ √
(a) Show that X( r L) and X( r L, s) are Deligne–Mumford stacks.
√
(b) Show that X( r L) has the equivalent description as the category of tuples
(f : T → X, M, α) where f : T → X is a morphism from a scheme, M is a line
∼
bundle on T and α : M ⊗r
√ → f ∗ L is an isomorphism. In particular, there is a
∼
line bundle L1/r on X( L) and an isomorphism (L1/r )⊗r → π ∗ L.
r

√
(c) Show that X( r L, s) has the equivalent description as the category of triples
f
(T −→ X, M, α, t) where f : T → X is a morphism from a scheme, M is a line
bundle on T , α : M ⊗r → f ∗ L is an isomorphism, and t ∈ Γ(T, M ) is a section
√
such that α(t⊗r ) = f ∗ s. In particular, there is a line bundle L1/r on X( r L, s)
∼
with a section s1/r together with an isomorphism (L1/r )⊗r → π ∗ L identifying
(s ) with π s.
1/r ⊗r ∗

(d) If X = Spec A is an affine scheme over S and L = OX is trivial, show that

√
X( L) ∼ µr ] and X( r L, s) ∼
p
= [Spec A[x]/(xr − s) /µ
r


= [X/µ µr ]

where µ r acts trivially on X and acts on Spec A[x]/(xr − s) via t · x = tx.

√
(e) Show that the fiber of X( r L) → X at a point x ∈ X is isomorphic to Bµ µr,κ(x) .
√
(f) Show that X( L, s) → X is an isomorphism over Xs = {s ̸= p
r
0} and that it
restricts to an infinitesimal extension of the root gerbe V (s)( r L|V (s) ) over
V (s).

3.9.3 Examples of algebraic stacks

Example 3.9.24. The classifying stack BGLn over Spec Z classifies vector bundles
of rank n. When n = 1, BGm = BGL1 classifies line bundles. The stack BGLn is a
universally closed and smooth algebraic stack over Spec Z of relative dimension −n2
with affine diagonal. However, BGLn is not separated nor Deligne–Mumford. The
structure morphism BGLn → Spec Z is an example of a coarse moduli space and
trivial GLn -banded gerbe.

157
Example 3.9.25. If Gm acts on A1 over Spec Z via scaling, the objects of the
quotient stack [A1 /Gm ] over a scheme T are pairs (L, s), where L is a line bundle
on T and s ∈ Γ(T, L). The stack [A1 /Gm ] is an algebraic stack universally closed
and smooth over Spec Z of relative dimension 0 with affine diagonal. The stack
[A1 /Gm ] is, however, not separated nor Deligne–Mumford. Over a field k, [A1 /Gm ]
has two points—one open and one closed—corresponding to the two Gm -orbits (see
Figure 0.4.7). There is an open immersion and closed immersion

1 : Spec k  / [A1 /Gm ] o ? _ BGm : 0
op cl

The morphism [A1 /Gm ] → Spec k identifies the two orbits and is an example of a
good moduli space.

Exercise 3.9.26. Show that [P1 /Gm ] ∼

= [(A1 A1 ∖0 A1 )/Gm ].
S

Example 3.9.27. Analogous to [A1 /Gm ], the quotient stack [P1 /Ga ] of the action
t · [x : y] = [x + ty : y] has two points with one open and one closed

 op
/ [P1 /Ga ] o cl ? _ BGa : ∞.
0 : Spec k

Example 3.9.28. Working over a field k, let Gm act on A2 via t · (x, y) = (tx, t−1 y).
The quotient stack [A2 /Gm ] is a smooth algebraic stack of dimension 1. An object of
[A2 /Gm ] over a scheme T is a triple (L, s, t) where L is a line bundle on T , s ∈ Γ(T, L)
and t ∈ Γ(T, L−1 ). The complement [A2 /Gm ] ∖ 0 of the origin is isomorphic to
the non-separated affine line. There is a morphism π : [A2 /Gm ] → A1 defined by
(x, y) 7→ xy, which is an isomorphism over A1 ∖ 0 and identifies the three orbits
defined by xy = 0; this is an example of a good moduli space.

xy t
t t

IT
o
Gm
stabilizer

A AYE 1A
Example 3.9.30 (Toric stacks). A fan Σ on a lattice L = Zn defines a toric variety
X(Σ), i.e., a normal separated variety with an action of Gnm such that there is a
dense orbit with trivial stabilizer; see [Ful93]. Meanwhile,
refence this picture in GIT
a stacky fan is a pair
(Σ, β) where Σ is a fan on a lattice L and β : L → N is a homomorphism of lattices.
As L and N are lattices (i.e., finitely generated free abelian groups), the Z-linear
duals define tori TL := D(L∨ ) and TN := D(N ∨ ) (Example B.1.6) where TL is a
torus for the toric variety X(Σ). The map β induces a homomorphism Tβ : TL → TN ,
naturally identifying β with the induced map on lattices of 1-parameter subgroups.
We can then define Gβ = ker(Tβ ) and the toric stack

X(Σ, β) := [X(Σ)/Gβ ].

158
Similar to the role of toric varieties, toric stacks provide a rich source of combinatorial
examples that can serve as a testing ground for general conjectures. See [FMN10]
and [GS17].
Example 3.9.31 (Picard schemes and stacks). Let X be a proper integral scheme
over an algebraically closed field k, the Picard functor of X and Picard stack of X
are defined as the sheaf Pic(X) and stack Pic (X) on Schét by

Pic(X) = sheafification of T 7→ Pic(XT )

Pic (X)(T ) = {groupoid of line bundles L on XT }

A morphism (T, L) → (T ′ , L′ ) in Pic (X) is the data of a morphism f : T → T ′ of

schemes and an isomorphism α : L → f ∗ L′ . The sheaf Pic(X) is representable by
a projective scheme and the tensor product of line bundles provides it with the
structure of a group scheme, hence an abelian variety. Moreover, Pic (X) is a smooth
algebraic stack over k. The map Pic (X) → Pic(X) forgetting automorphism groups
is an example of a banded Gm -gerbe. See §6.4.7.
d
Example 3.9.32 (Vector bundle stacks). Let K • : K 0 − → K 1 be a two-term complex
of vector bundles on a scheme X. The vector bundle stack C(K • ), sometimes called
a cone stack and written as [K 1 /K 0 ] or [h0 /h1 ], is defined as the quotient stack
over (Sch/X)ét
C(K • ) = [A(K 1,∨ )/A(K 0,∨ )],
where A(K 0,∨ ) is viewed as the additive group scheme and the action σ : A(K 0,∨ ) ×
A(K 1,∨ ) → A(K 1,∨ ) is defined by σ(x0 , x1 ) = x1 + d(x0 ). When X = Spec k with
k a field, then C(K • )(k) is the groupoid whose objects are A(K 1,∨ )(k) = K 1 and
where a morphism v → v ′ is an element α ∈ A(K 0,∨ )(k) = K 0 such that d(α) =
v − v ′ . Moreover, C(K • ) is identified with the quotient [A(H1 (K • )∨ )/A(H0 (K • )∨ )]
of the trivial action, and in particular the groupoid C(K • )(k) is equivalent to
[H1 (K • )/H0 (K • )].
For general X, the fiber of C(K • ) → X over a geometric point Spec k → X is
C(K • ⊗OX k) ∼ = [A(H1 (K • ⊗OX k)∨ )/A(H0 (K • ⊗OX k)∨ )]. If g : T → X is a morphism
from an affine scheme, then since isomorphism classes of principal A(K 0,∨ )-bundles
over T are in bijection with H1 (T, K 0,∨ ) = 0, every principal bundle is trivial. Hence,

C(K • )(T ) = [A(K 1,∨ )(T )/A(K 0,∨ )(T )] ∼

= [H0 (T, g ∗ K 1 )/H0 (T, g ∗ K 0 )],

where we have used the bijections

g
A(K i,∨ )(T −
→ S) = MorS (T, A(K i,∨ ))
= HomOS −alg (Sym∗ K i,∨ , g∗ OT )
= HomOS (K i,∨ , g∗ OT )
(3.9.33)
= H0 (S, K i ⊗ g∗ OT )
= H0 (S, g∗ g ∗ K i ) (projection formula)
0 ∗ i
= H (T, g K )

By construction, we have that C(K • ) → X is a smooth morphism of relative

dimension h1 (K • ⊗OX k) − h0 (K • ⊗OX k) for any k-point Spec k → X.
Exercise 3.9.34. Show that a quasi-isomorphism K • → L• of two-term complexes
∼
on a scheme X induces an isomorphism C(K • ) → C(L• ) of vector bundle stacks.

159
 The exercise above shows that the vector bundle stack C(K • ) is well-defined for
a complex K • ∈ Db (X) with amplitude in [0, 1]. We will use this construction in
Proposition 8.2.11 to describe the stack of extensions C(Rp2,∗ H omOC×S (F1 , F2 )) →
S of vector bundles F1 , F2 on C × S, where C is a smooth curve. For a syntomic
morphism X → Y , LX/Y is perfect complex with amplitude in [−1, 0], one can
consider the vector bundle stack C(L∨X/Y ), which plays a role in defining the intrinsic
normal cone in [BF97].

3.9.4 Extremely non-separated examples

The following are often counterexamples for properties that hold for algebraic stacks
with separated and quasi-compact diagonal, but fail in general. For instance, if
the diagonal of a Deligne–Mumford stack is quasi-compact and separated diagonal,
then it is quasi-affine (Corollary 4.5.8) and in particular representable by schemes,
and the examples below illustrate that both the quasi-compactness and separated
hypotheses are necessary.

Example 3.9.35 (Deligne–Mumford stacks with non-quasi-compact diagonals). Let

Z denote the constant group scheme over Spec Z associated to the abstract discrete
group Z. The classifying stack BZ is a smooth algebraic stack of dimension 0 whose
diagonal is not quasi-compact. In particular, BZ is not quasi-separated.
Let Z act on A1 over a characteristic zero field k via n · x = x + n for x ∈ A1
and n ∈ Z. The quotient G := A1 /Z is a quasi-compact, locally noetherian group
algebraic space with non-quasi-compact diagonal that is not a scheme. Any quasi-
separated group algebraic space locally of finite type over a field k is a scheme
(Theorem 4.5.28), and this example shows that the quasi-separatedness hypothesis
is necessary.
The classifying stack BG is a Deligne–Mumford stack whose diagonal ∆BG is
quasi-compact but whose second diagonal ∆∆BG is not quasi-compact; in particular
BG is not quasi-separated. Moreover, the diagonal of BG is not representable by
schemes.
Similarly, one can consider the algebraic space quotient A1C /Z2 where (a, b) · x =
x + a + ib. While the analytic quotient C/Z2 of this action is an elliptic curve over
C, the algebraic space quotient is not quasi-separated and not a scheme.

Exercise 3.9.36. Let G = A1 /Z.

(a) Show that G is not a scheme.
(b) Show that the generic point Spec k(x) → A1 → X is fixed under the Z-
action, and that the composition Spec k(x) → G does not factor through a
monomorphism Spec L → G for a field L, i.e., the generic point of G does not
have a residue field.

Example 3.9.37 (Deligne–Mumford stacks with non-separated diagonal). The

non-separated affine line Q := A1 A1 ∖0 A1 is a group scheme over A1 , where every
S
fiber is trivial except over the origin. The classifying stack BQ is a Deligne–Mumford
stack whose diagonal is non-separated.
Here is a construction to produce similar examples. Let G → S be a finite étale
group scheme. If H ⊆ G is a subgroup scheme over S, then G/H is separated if and
only if H ⊂ G is closed. If H is normal in G, then G/H → S is a group algebraic
space. For instance, taking G = Z/2 × A1 → A1 and the subgroup H = G ∖ {−1, 0},
the quotient Q = G/H is the non-separated affine line ver A1 .

160
 Likewise, consider µ3,Q = Spec Q ⨿ G′ over Q where G′ = Spec Q[x]/(x2 + x + 1)
is the non-identity component , and let G = µ3,Q × A1Q . Then Q = (µ3,Q × A1Q )/H
is a non-separated étale group algebraic space over A1Q which is not a scheme. It
is not
S a scheme S for several reasons to Exercise 3.9.7: Q → A1 is a form of the
AQ A1 ∖0 AQ A1 ∖0 AQ of the group scheme which is trivial except over the origin
1 1 1
Q Q
where it is Z/2. If the non-identity point in Q0 contained an affine open, its preimage
under A1Q[x]/(x2 +x+1) → A1Q would be an affine open containing two of the origins, a
contradiction.
In this case, BQ is Deligne–Mumford stack which is quasi-compact and quasi-
separated (i.e., BG, the first diagonal ∆BG , and second diagonal ∆∆BG are quasi-
compact), but whose diagonal is not separated and not representable by schemes.
While the vast majority of algebraic stacks that appear in moduli theory have
separated diagonal, there are a few exceptions, e.g., the stack Logalg of log structures
[Ols03].

161
Chapter 4

Geometry of Deligne–Mumford
stacks

Geometry is the cilantro of math.

Jordan Ellenberg

4.1 Quasi-coherent sheaves and cohomology

4.1.1 Sheaves
The small étale site of a Deligne–Mumford stack can be defined analogously to the
small étale site of a scheme (Example 2.2.4).
Definition 4.1.1. The small étale site of a Deligne–Mumford stack X is the category
Xét of schemes étale over X . A covering`of an X -scheme U is a collection of étale
morphisms {Ui → U } over X such that i Ui → U is surjective.
We can therefore discuss sheaves of abelian groups on Xét and their morphisms.
We denote Ab(Xét ) as the category of abelian sheaves on Xét . For an abelian sheaf
F on Xét , the sections over an étale X -scheme U are denoted by F (U → X ).
Example 4.1.2 (Structure sheaf). The structure sheaf OX on a Deligne–Mumford
stack is defined by OX (U → X ) = Γ(U, OU ) on an étale X -scheme U .
Example 4.1.3 (Differentials). If X is a Deligne–Mumford stack over a scheme S,
the relative sheaf of differentials ΩX /S is defined by ΩX /S (U → X ) = Γ(U, ΩU/S ).
Example 4.1.4 (Hodge bundle). Define the sheaf H on Mg (for g ≥ 2) as follows:
for an étale morphism U → Mg classified by a family C → U of smooth curves, we set
H(U → Mg ) = Γ(C, ΩC/U ). We will see shortly that H is a coherent OMg -module
and, in fact, a vector bundle of rank g.
Exercise 4.1.5 (easy). If k is an algebraically closed field, show that the functor
Ab((Spec k)ét ) → Ab, defined by F 7→ F (Spec k), is an equivalence of categories.
While sections of a sheaf F on Xét are only defined on étale X -schemes, we can
extend the definition to a Deligne–Mumford stack U étale over X as follows: choose
étale presentations U → U and R → U ×U U by schemes and define
F (U → X ) := Eq(F (U → X ) ⇒ F (R → X )).

162
One checks that this is independent of the choice of presentation. In particular, it
makes sense to discuss global sections
id
Γ(X , F ) := F (X −→ X ).

Exercise 4.1.6 (easy). If F is an abelian sheaf on a Deligne–Mumford stack X ,

show that Γ(X , F ) = HomAb(Xét ) (Z, F ) where Z is the constant sheaf.
Given a morphism f : X → Y of Deligne–Mumford stacks, there are functors
f∗
,
Ab(Xét ) l Ab(Yét )
f −1

where the pushforward f∗ F (V → Y) := F (X ×Y V → X ) and the inverse image

f −1 G is the sheafification of the presheaf

(U → X ) 7→ colim(V →Y,U →X ×Y V ) G(V → Y),

where the colimit is over pairs of étale morphisms V → Y and U → X ×Y V , i.e.,

commutative diagrams
U /V

ét ét
 
X / Y.

Exercise 4.1.7 (easy). Show that if f : X → Y is an étale morphism of Deligne–

Mumford stacks and G is a sheaf on Yét , then f −1 G = G|X is the restriction of
G to Xét , i.e., the sheaf (U → X ) 7→ G(U → X → Y). In particular, show that
f −1 OY = OX .
Exercise 4.1.8 (easy). Let X be a scheme and x ∈ X.
(a) Denote ix : Spec κ(x) → X, show that the global sections of i−1
x OX is the
henselization OX,x
h
of OX,x .
(b) Denoting ix : Spec κ(x) → X, show that ix−1 is identified with the strict
henselization OX,x
sh
of OX,x .
See §B.5.3 for background on (strict) henselizations.
Exercise 4.1.9. Let k be a field, ksep be a separable closure, and G := Gal(ksep /k)
be its Galois group with the profinite topology.
(a) Show that a covariant functor

F : {étale k-algebras} → Ab

defines a sheaf of abelian groups on (Spec k)ét if and only if F ( i Ai ) =

Q
F (Ai ) for every finite set of étale k-algebras {Ai } and F (K) = F (L)Gal(L/K)
L
for every finite Galois extension L/K of finite separable field extensions of k.
(b) Show that the category of sheaves on (Spec k)ét is equivalent to the category
of discrete G-modules, i.e., abelian groups with a continuous G-action.
Hint: Given F , define a discrete G-module by MF := colimk′ /k F (k′ ), where
the colimit is over finite separable field extensions. Conversely, given M , show
sep ′
that the functor on finite field extensions k′ 7→ M Gal(k /k ) extends to a sheaf
FM on (Spec k)ét . Show that M 7→ MF and M 7→ FM are inverse functors.

163
Exercise 4.1.10. Let X be a Deligne–Mumford stack and x : Spec k → X be a
geometric point (i.e., k is an algebraically closed field). Define the stalk of an abelian
sheaf F on Xét at x as Fx := x−1 F .
(a) Show that Ab(Xét ) → Ab, defined by F 7→ Fx , is an exact functor.
(b) Show that a sequence of abelian sheaves on Xét is exact if and only if the
induced sequence of stalks is exact for every geometric point.
(c) Show that f −1 is exact for a morphism f : X → Y of Deligne–Mumford stacks.

Exercise 4.1.11 (details). Show that f −1 is left adjoint to f∗ .

Exercise 4.1.12 (technical). If X is a Deligne–Mumford stack, define instead the

site Xét′ as the category of algebraic spaces étale over X where coverings are étale
coverings. Show that the categories of sheaves on Xét and Xét′ are equivalent.

4.1.2 OX -modules
The structure sheaf OX is a ring object in Ab(Xét ), and OX -modules are defined as
module objects over OX .

Definition 4.1.13. If X is a Deligne–Mumford stack, a sheaf of OX -modules (or

simply an OX -module) is a sheaf F on Xét which is a module object for OX in the
category of sheaves, i.e., for every étale X -scheme U , F (U → X ) is an module over
OX (U → X ) = Γ(U, OU ) and the module structure is compatible with respect to
restriction along morphisms V → U of étale X -schemes.

We denote Mod(OX ) as the category of OX -modules. Given two OX -modules F

and G, we can define the tensor product F ⊗ G := F ⊗OX G as the sheafification
of the OX -module given by (U → X ) 7→ F (U → X ) ⊗OX (U →X ) G(U → X ). The
Hom sheaf H omOX (F, G) has sections HomOU (F |U , G|U ) over an étale morphism
f : U → X from scheme, where F |U = f −1 F denotes the restriction of F to Uét .

Exercise 4.1.14 (easy). If F is an OX -module, show that Γ(X , F ) = HomOX (OX , F ).

Given a morphism f : X → Y of Deligne–Mumford stacks, there are functors

f∗
,
Mod(OX ) Mod(OY )
m
f∗

where for an OX -module F , f∗ F is the pushforward as sheaves and is naturally

an OY -module. For an OY -module G, since there is a morphism f −1 OY → OX of
sheaves of rings in Xét and f −1 G is a f −1 OY -module, it makes sense to define the
pullback as the OX -module

f ∗ G := f −1 G ⊗f −1 OY OX .

If f is étale, then f ∗ G = f −1 G = G|X since f −1 OY = OX (by Exercise 4.1.7)

Exercise 4.1.15 (details). Show that f ∗ is left adjoint to f∗ .

Exercise 4.1.16 (details). Show that Mod(OX ) is an abelian category.

164
4.1.3 Quasi-coherent sheaves
Let F be an OX -module on a Deligne–Mumford stack X . For an étale X -scheme
U , we have the restriction F |Uét to the small étale site Uét of U and the further
restriction F |UZar restricted to the Zariski topology of U . Note that when X is a
scheme, OX could refer to the structure sheaf either in Xét or XZar , and we write
either OXZar or OXét if there is any possibility of confusion.
Definition 4.1.17. Let X be a Deligne–Mumford stack. An OX -module F is
quasi-coherent if
(1) for every étale X -scheme U , the restriction F |UZar is a quasi-coherent OUZar -
module, and
(2) for every étale morphism f : U → V of étale X -schemes, the natural morphism
f ∗ (F |VZar ) → F |UZar is an isomorphism.
A quasi-coherent OX -module F on X is a vector bundle (resp., vector bundle of
rank r, line bundle) if F |UZar is for every étale morphism U → X from a scheme. If
in addition X is locally noetherian, we say F is coherent if F |UZar is coherent for
every étale morphism U → X .
We denote by QCoh(X ) and Coh(X ) (in the noetherian setting) the categories
of quasi-coherent and coherent sheaves. In a moment, we will see that this agrees
with the usual definition when X is a scheme, but first we give some examples:
Example 4.1.18. The structure sheaf OX is quasi-coherent. It is a line bundle,
which is coherent when X is locally noetherian.
Example 4.1.19. For a Deligne–Mumford stack X over a scheme S, the relative
sheaf of differentials ΩX /S of Example 4.1.3 is quasi-coherent since for an étale
morphisms f : U → V of étale X -schemes, f ∗ ΩV /S → ΩU/S is an isomorphism. It is
a vector bundle when X → S is smooth.
Example 4.1.20. For Mg with g ≥ 2, the Hodge bundle H of Example 4.1.4 is a
vector bundle of rank g. This follows from Proposition 5.1.16(2): for a smooth family
π : C → V of genus g curves corresponding to a Mg -scheme V , the construction of
∼
π∗ ΩC/V commutes with the base change along a map f : U → V , i.e., f ∗ (π∗ ΩC/V ) →
πU,∗ ΩCU /U (which implies that H is quasi-coherent), and moreover π∗ ΩC/V is a
vector bundle on V of rank g (which implies that H is a vector bundle of rank g).
Example 4.1.21. If G is a finite abstract group viewed as a group scheme over a
field k, a quasi-coherent sheaf on BG corresponds to a representation V of G. If G
acts on an affine k-scheme Spec A, a quasi-coherent sheaf on [Spec A/G] is the data
of an A-module M equipped with a group homomorphism G → EndA (M ). These
descriptions follow from Exercise 4.1.24(b).
Exercise 4.1.22 (Equivalences I). Let X be a scheme.
(a) Show that the following are equivalent for an an OXét -module F :
(i) F is quasi-coherent as in Definition 4.1.17.
(ii) There is a quasi-coherent OXZar -module F ′ and an isomorphism F ∼
= α∗ F ′ ,
where
α∗ : Mod(OXZar ) → Mod(OXét )
is the functor taking an OXZar -module F ′ to the OXét -module α∗ F ′
g
defined by α∗ F ′ (U −
→ X) = Γ(U, g ∗ F ′ ).

165
 (iii) F is quasi-coherent as a module object in the ringed site (Xét , OXét )
using the general definition in a ringed site (S, O) [SP, Tag 03DL]: an
O-module F is quasi-coherent if for every object U ∈ S, there is a covering
{Ui → U } such that the restriction F |Ui to the restricted site S/Ui has a
free presentation
⊕J ⊕I
OU i
→ OU i
→ F |Ui → 0.

Hint: To show that (iii) implies the other conditions, use Fpqc Descent for
Quasi-Coherent Sheaves (2.1.4) to construct a quasi-coherent sheaf.
(b) Show that α∗ restricts to an equivalence of categories QCoh(XZar ) → QCoh(Xét ).
Hint: Show that restriction F |XZar of an OXét -module to the Zariski topology
defines a left adjoint α∗ to α∗ . Use adjunction and the identification α∗ α∗ F ∼
=
F for a quasi-coherent OXZar -module F to show that α∗ is fully faithful on
quasi-coherent sheaves.

Exercise 4.1.23. Provide an example of a scheme X and a non-quasi-coherent

OXét -module F such that F |XZar is quasi-coherent.
This shows why condition (2) is necessary in Definition 4.1.17.

Exercise 4.1.24 (Equivalences II). Let X be a Deligne–Mumford stack.

(a) Show that the following are equivalent for an OXét -module F :
(i) F is quasi-coherent as in Definition 4.1.17.
(ii) There is an étale presentation U → X by a scheme such that F |Uét is
quasi-coherent.
(iii) F is quasi-coherent as a sheaf of modules in the ringed site (Xét , OXét ).
(b) Let U → X be an étale presentation and set R = U ×X U . Show that the
category of quasi-coherent sheaves on X is equivalent to the category of pairs
∼
(G, α) where G is a quasi-coherent sheaf on U and α : p∗1 G → p∗2 G is an
isomorphism on R := U ×X U satisfying the cocycle condition p∗23 α ◦ p∗12 α =
p∗13 α.
(c) Show that a quasi-coherent sheaf on X is equivalent to the data of a quasi-
coherent sheaf FU on every scheme U over X and compatible isomorphisms
f ∗ FV → FU for every map f : U → V of schemes over X .

Exercise 4.1.24(c) provides a convenient and functorial way to think about quasi-
coherent sheaves on moduli stacks. For instance, the Hodge bundle on Mg can be
viewed as the data of the sheaf π∗ ΩC/S for every smooth family of curves π : C → S.

Exercise 4.1.25 (Equivalences III). Given a Deligne–Mumford stack X , define

the big étale site (Sch /X )ét , big fppf site (Sch /X )fppf , and lisse-étale site Xlis-ét
analogously to the definitions on a scheme as in §2.2.2. In each site, there is structure
sheaf O and we say that an O-module is quasi-coherent if it is quasi-coherent as a
sheaf of modules in the ringed site. Show that there are equivalences
∼ ∼ ∼
QCoh(Xét ) → QCoh(Xlis-ét ) → QCoh((Sch/X )ét ) → QCoh((Sch/X )fppf ).
∼
defined analogously to the equivalence QCoh(XZar ) → QCoh(Xét ) in Exercise 4.1.22
over scheme X.

166
4.1.4 Pushforwards and pullbacks
Exercise 4.1.26 (Pushforward–Pullback Adjunction). Let f : X → Y be a morphism
of Deligne–Mumford stacks.
(a) Show that f ∗ preserves quasi-coherence.
(b) If f is quasi-compact and quasi-separated, show that f∗ preserves quasi-
coherence and conclude that

f∗
,
QCoh(X ) QCoh(Y)
m
∗
f

are adjoint functors (with f∗ the right adjoint).

Exercise 4.1.27 (good practice). Let G be a finite group and k be a field.

(a) Consider a G-representation V viewed also as a quasi-coherent sheaf on BG
p π
using Example 4.1.21. Under the composition Spec k − → BG − → Spec k,
show that π∗ V = V G , where V G is the subspace of G-invariants, and that
p∗ V = V forgetting the G-action. Moreover, for a k-vector space W , show
that π ∗ W = W with the trivial G-action and that p∗ W = W ⊗ p∗ k, where
p∗ k is the regular representation Γ(G, OG ).
(b) Given an action of G an an affine k-scheme Spec A, consider the diagram

Spec A
p
/ [Spec A/G] π / Spec AG

q

BG

and recall from Example 4.1.21 that a quasi-coherent sheaf on [Spec A/G] is an
A-module M with a group homomorphism G → EndA (M ). Provide explicit
descriptions of the functors p∗ , p∗ , π∗ , π ∗ , q∗ and q ∗ on quasi-coherent sheaves.

Exercise 4.1.28 (hard). Let X be a noetherian Deligne–Mumford stack. Prove the

following two statements:
(a) Every quasi-coherent sheaf on X is a directed colimit of its coherent subsheaves.
(b) If U ⊆ X is an open substack, then every coherent sheaf on U extends to a
coherent sheaf on X .
This exercise extends [Har77, Exc. II.5.15] from schemes to Deligne–Mumford stacks;
see also [LMB00, Prop. 15.4], [Ols16, Prop. 7.1.11], and [SP, Tag 0GRE].

4.1.5 Quasi-coherent constructions

A quasi-coherent OX -algebra on a Deligne–Mumford stack is a quasi-coherent OX -
module with the compatible structure of a ring object in Ab(Xét ). We define the
relative spectrum SpecX A as the stack whose objects over a scheme S consists of a
morphism f : S → X and a morphism f ∗ A → OS of OS -algebras.

Exercise 4.1.29 (easy, good practice). Show that SpecX A is an algebraic stack
affine over X .

167
Example 4.1.30 (Reduction). Let X be a Deligne-Mumford stack and let OX red

be the sheaf of OX -algebras defined by OX (U → X ) = Γ(U, OU )red for an étale

red

X -scheme U . Then OX red

is a quasi-coherent OX -algebra and Xred := SpecX OX
red
is
a reduced algebraic stack equipped with a closed immersion Xred ,→ X such that
|Xred | = |X |. We call Xred the reduction of X .
Example 4.1.31 (Normalization). Let X be an integral Deligne-Mumford stack
and let A be the sheaf of OX -algebras defined by setting A(U → X ) to be the
normalization of Γ(U, OU ). Since normalization commutes with étale extensions
(Proposition A.7.4), A is a quasi-coherent OX -algebra. The normalization of X is
defined as Xe := SpecX A.
Exercise 4.1.32 (good practice). Let f : X → Y be a quasi-compact and quasi-
separated morphism of Deligne–Mumford stacks.
(a) Show that there is factorization f : X → Spec f∗ OX → Y.
(b) Show that f is affine if and only if X → Spec f∗ OX is an isomorphism.
(c) Show that f is quasi-affine if and only if X → Spec f∗ OX is an open immersion.
Exercise 4.1.33 (details). Use Exercise 4.1.28 to show that every quasi-coherent
sheaf of algebras on a noetherian Deligne–Mumford stack is a directed colimit of
finite type subalgebras.

4.1.6 Picard groups

We denote by Pic(X ) the set of isomorphism classes of line bundles on a Deligne–
Mumford stack X . It is an abelian group under tensor product.
Exercise 4.1.34.
(a) Show that Pic(Bµn ) = Z/nZ over a field whose characteristic is relatively
prime to n.
(b) More generally, if G is finite group whose order is prime to the characteristic
of a field k, show that Pic(BG) ∼ = X∗ (G), where X∗ (G) denotes the group of
characters of G, i.e., group homomorphisms G → Gm .
The hypothesis on the characteristic is only to ensure that BG is Deligne–Mumford
stack.
Exercise 4.1.35 (hard). Show that Pic(M1,1 ) = Z/12 over a field with char(k) ̸=
2, 3 and is generated by the Hodge bundle (see Example 4.1.4).
This was the main conclusion of Mumford’s influential paper [Mum65], where the
result was formulated without the language of Deligne–Mumford stacks (which had
not yet been introduced). This exercise will be easier once we have more technology,
namely quasi-coherent sheaves; see Exercise 6.1.22.

4.1.7 Cohomology
We develop a cohomology theory for abelian sheaves on Deligne–Mumford stacks.
Despite utilizing the cohomology of quasi-coherent sheaves on schemes throughout
these notes, we have surprisingly little need for cohomology on algebraic spaces
and Deligne–Mumford stacks, and many of the results here are included only for
completeness.
The existence of enough injective objects is shown analogously to the case of
schemes [Har77, Prop. 2.2].

168
Lemma 4.1.36. If X is a Deligne–Mumford stack, the categories Ab(Xét ) and
Mod(OX ) have enough injectives. If in addition X is quasi-separated, then QCoh(X )
has enough injectives.
Proof. Recall that a functor R : A → B between abelian categories with an exact
left adjoint L preserves injectives since for an injective I ∈ A, HomB (−, R(I)) =
HomA (L(−), I) is exact. We claim that for every sheaf of rings Λ on Xét , the category
Mod(Λ) of Λ-modules has enough injectives. Taking Λ to be the constant sheaf
Z (resp., the structure sheaf OX ) establishes that Ab(Xét ) (resp., Mod(OX )) has
enough injectives. Let F be a Λ-module and let U → X be an étale presentation. For
each geometric point Spec k → U , consider the composition ju : Spec k → U → X .
Since an étale sheaf on Spec k corresponds to an ordinary group (Exercise 4.1.5), the
preimage Fu := ju−1 F corresponds to an ordinary module over the ring Λu := ju−1 Λ.
Choose an inclusion Fu ,→ Iu into an injective Λu -module. Adjunction gives a map
F → ju,∗ Iu , and since ju−1 is exact (Exercise 4.1.10),Q
ju,∗ Iu ∈ Ab(Xét ) is injective.
By taking the product, we obtain an injection F ,→ u∈U ju,∗ Iu into an injective
Λ-module.
For the final statement, let F ∈ QCoh(X ) and let p : U = i Spec Ai → X be
`
an étale presentation. Choose an injection p∗ F ,→ I into an injective quasi-coherent
OU -module. The composition F ,→ p∗ p∗ F ,→ p∗ I is injective and since p∗ is exact,
p∗ I is injective.
Remark 4.1.37. The above argument extends to any ringed site with enough
points (see [Ols16, Thm. 2.3.2]), and the result holds even in any ringed site [SP,
Tag 01DP]. The category of quasi-coherent sheaves on a Deligne–Mumford stack X
is a Grothendieck abelian category [SP, Tag 0781] and any such category has enough
injectives [Gro57b], [SP, Tag 079H].
Definition 4.1.38 (Cohomology). Let X be a Deligne–Mumford stack and F a
sheaf of abelian groups on Xét . The cohomology group Hp (Xét , F ) is defined as the
pth right derived functor of the global sections functor Γ : Ab(Xét ) → Ab.
Given a morphism f : X → Y of Deligne–Mumford stacks, the higher direct image
Rp f∗ F is defined as the pth right derived functor of f∗ : Ab(Xét ) → Ab(Yét ).
In order to show that the cohomology of a quasi-coherent sheaf on a scheme
can be computed using either the usual Zariski topology or the small étale site
(Proposition 4.1.43), we will use Čech cohomology. Just as with schemes, Čech
cohomology provides an extremely effective tool to compute cohomology groups.
Čech cohomology in the étale topology is defined similarly to the case of the Zariski
topology [Har77, III.4] replacing intersections Ui0 ∩ · · · ∩ Uin with fiber products
Ui0 ×X · · · ×X Uin and considering all (possibly non-distinct) indices i0 , . . . , in in
any order.
Definition 4.1.39 (Čech cohomology). Let F be an abelian sheaf on a Deligne–
Mumford stack X . If U = {Ui → X }i∈I is an étale covering, the Čech complex of F
with respect to U is Cˇ• (U, F ) where
Y
Cˇn (U, F ) = F (Ui0 ×X · · · ×X Uin )
(i0 ,...,in )∈I n+1

with differential
 n+1
X 
dn : Cˇn (U, F ) → Cˇn+1 (U, F ), (si0 ,...,in ) 7→ (−1)k pb∗k si0 ,...,ibk ,...,in+1
k=0 (i0 ,...,in+1 )

169
where pbk : Ui0 ×X · · · ×X Uin+1 → Ui0 ×X · · · ×X U
d ik ×X · · · ×X Uin+1 is the map
forgetting the kth component (with indexing starting at 0). The Čech cohomology
of F with respect to U is

Ȟp (U, F ) := Hp (Cˇ• (U, F )).

The following is a standard result in Čech cohomology whose proof is analogous

to topological spaces. It is often referred to as Cartan’s criterion.
Lemma 4.1.40. Let X be a Deligne–Mumford stack and let F be an abelian sheaf
on Xét . Suppose that Ob′ is a subset of the set of schemes étale over X and that
Cov′ is a subset of all the coverings in Xét such that for each {Ui → U )} ∈ Cov′ , we
have that U , each Ui , and each fiber product Ui1 ×U · · · ×U Uin is in Ob′ . Assume
that every covering of an étale X -scheme U has a refinement in Cov′ . If for every
covering U ∈ Cov′ , Ȟp (U, F ) = 0 for p > 0, then Hp (U, F ) = 0 for every U ∈ Ob′ .
Proof. This holds in any site; see [God58, II.5.9.2], [Mil80, Prop. III.2.12], [SP,
Tag 03F9], or [Ols16, Prop. 2.3.15].
Cohomology of an étale sheaf can be computed as the Čech cohomology with
respect to a covering with vanishing cohomology, just as in the case of ordinary
topological spaces [Har77, Exc. III.4.11].
Lemma 4.1.41. Let F be an abelian sheaf on Xét and (Ui → X )i∈I an étale
covering. If Hp (Ui0 ×U · · · ×U Uin , F ) = 0 for all p > 0, n ≥ 0 and i0 , . . . , in ∈ I,
then Ȟp (U, F ) = Hp (Xét , F ).
Proof. This also holds in any site and follows from Čech-to-derived functor spectral
sequence; see [SP, Tag 03F7].
Theorem 4.1.42. For a quasi-coherent sheaf F on an affine scheme X, Hp (Xét , F ) =
0 for all p > 0.
Proof. Let X = Spec A, F = M f be a quasi-coherent OX -module, which we
Zar
can also view as a quasi-coherent OXét -module (Exercise 4.1.22). The set of étale
coverings of the form U = {Spec B → Spec A} is sufficient to refine any other
covering. For the covering U, faithfully flat descent (Exercise 2.1.3) gives a long
exact sequence

0 → M → M ⊗A B → M ⊗A B ⊗A B → M ⊗A B ⊗A B ⊗A B → · · · ,

which is identified with the Čech complex Cˇ• (U, F ). This shows that Ȟp (U, F ) = 0
for p > 0 and thus Lemma 4.1.40 implies that Hp (Xét , F ) = 0.
Proposition 4.1.43. If X is a scheme with affine diagonal and F is a quasi-coherent
sheaf, then Hp (X, F ) = Hp (Xét , F ) for all p.
Proof. Let U := {Ui ,→ X} be a Zariski covering of X by affine scheme. Since X has
affine diagonal, each fiber product Ui0 ×X · · ·×X Uin is also affine. By Theorem 4.1.42,
Hp (Ui0 ×U · · · ×U Uin , F ) = 0, and therefore we may apply Lemma 4.1.41 to conclude
that Hp (X, F ) = Ȟp (U, F ) = Hp (Xét , F ). The same result holds more generally
without the affine diagonal hypothesis and in other sites, e.g., the big fppf site
(Sch/X)fppf ; see [Mil80, Prop. 3.7] and [SP, Tag 03DW].
We now compare the cohomologies computed in Ab(Xét ), Mod(OXét ), and
QCoh(X ).

170
Proposition 4.1.44. Let X be a Deligne–Mumford stack.
(1) If F is an OX -module, then the cohomology Hp (Xét , F ) of F as an abelian
sheaf agrees with the pth right derived functor of Γ : Mod(OX ) → Ab.
(2) If X is quasi-compact with affine diagonal and F is a quasi-coherent sheaf on
X , then the cohomology Hp (Xét , F ) of F as an abelian sheaf agrees with the
pth right derived functor of Γ : QCoh(X ) → Ab.
The same holds for the higher direct images Rp f∗ F of a morphism f : X → Y of
Deligne–Mumford stacks.

Proof. For (1), we need to show that an injective object in Mod(OX ) is acyclic
in Ab(Xét ). This uses a standard technique in Čech cohomology. We will use the
notation that if U = {Ui → X }i∈I is an étale covering and j = (j0 , . . . , jn ) ∈ I n+1 ,
then Uj := Uj0 ×X · · · ×X Ujn with structure morphism qj : Uj → X . There is a
chain complex ZU ,• of presheaves on X defined by
M
ZU ,n := qj,! Z,
j∈I n+1

where Z denotes the constant presheaf and qj,! Z(V → X ) := MorX (V,Ui ) Z, and
L
where the differentials are the alternating sums of the natural maps. This complex of
presheaves is exact in positive degrees and has the property that for every presheaf
F
ˇ F ) = MorPAb(X ) (Z , F ) = MorPMod(O ) (Z ⊗Z OX , F ),
C(U, ét U ,• X U ,•

where morphisms are computed in the categories PAb(Xét ) and PMod(OX ) of

presheaves. If F ∈ Mod(OX ) is injective, then it is also injective as a presheaf of OX -
modules. It follows that C(U, ˇ F ) is exact in positive degrees and thus Ȟp (U, F ) = 0
for p > 0. Therefore, Lemma 4.1.40 implies that Hp (Xét , F ) = 0. This result holds
more generally in any ringed site; see [Ols16, Cor. 2.3.16] or [SP, Tag 03FD].
For (2), let F ∈ QCoh(X ) be an injective object. Let q : U = Spec A → X be
an étale presentation and choose an injection q ∗ F ,→ G into an injective object
G ∈ QCoh(U ). Then pushforward q∗ G is injective (as the right adjoint q ∗ is
exact), and we have an inclusion F ,→ q∗ q ∗ F ,→ q∗ G of injectives which necessarily
splits. It thus suffices to show that q∗ G is acyclic in Ab(Xét ). Since X has affine
diagonal, p : U → X is an affine morphism. By Flat Base Change (A.2.12) and étale
descent, q∗ is exact on the category of quasi-coherent sheaves. By Theorem 4.1.42,
Hp (Xét , q∗ G) = Hp (Uét , G) = 0 .

Exercise 4.1.45 (Flat Base Change). Consider a cartesian diagram

g′
X′ /X

f′ □ f
 
Y′
g
/Y

of Deligne–Mumford stacks, and let F be a quasi-coherent sheaf on X . If g : Y ′ → Y

is flat and f : X → Y is quasi-compact and quasi-separated, the natural adjunction
map
g ∗ Rp f∗ F → Rp f∗′ g ′∗ F
is an isomorphism for all p ≥ 0.

171
Exercise 4.1.46. If X is a Deligne–Mumford stack and Fi is a directed system
of abelian sheaves in Xét , show that colimi Hp (Xét , Fi ) → Hp (Xét , colimi Fi ) is an
isomorphism.
Example 4.1.47 (Group cohomology). Let G be a finite abstract group viewed as
a group scheme over a field k, and let V be a G-representation which we can also
view as a quasi-coherent sheaf on BG. The group cohomology Hp (G, V ) is defined
as the pth right derived functor of

Rep(G) → Vectk , V 7→ V G .

Since Rep(G) ∼
= QCoh(BG), Γ(BG, V ) = V G , and Hp ((BG)ét , V ) can be computed
in QCoh(BG) (Proposition 4.1.44(2)), we have the identification

Hp (G, V ) ∼
= Hp ((BG)ét , V ).

The Čech complex Č • (U, V ) of V with respect to the étale cover U = {Spec k → BG}
has terms
Č n (U, V ) = Γ(G, OG )⊗n ⊗ V
and differentials
dn : Γ(G, OG )⊗n ⊗ V →Γ(G, OG )⊗(n+1) ⊗ V
n
X (4.1.48)
f ⊗ v 7→ (−1)k µ∗k (f ) ⊗ v + (−1)n+1 f ⊗ σ(v)
k=0

where σ : V → Γ(G, OG ) ⊗ V is the coaction and µk : Gn+1 → Gn is defined by send-

ing (g1 , . . . , gn+1 ) to (g2 , . . . , gn+1 ) for k = 0 and to (g1 , . . . , gk−1 , gk gk+1 , gk+2 , . . . , gn+1 )
for k = 1, . . . , n; see Exercise 3.4.6. In low degrees, we have d0 (v) = v − σ(v) and
d1 (f, v) = 1 ⊗ f ⊗ v − µ∗ (f1 ) ⊗ v + f1 ⊗ σ(v) where µ = µ1 is group multiplication
G × G → G. This is called the standard resolution or bar resolution, and provides
an effective way to compute group cohomology. Since G is finite, this complex can
also be interpreted using the identifications Γ(Gn , OGn ) ⊗ V ∼ = Map(Gn , V ) with
set-theoretic maps. See also [Wei94, §6] and [Bro94, §1].
The following example illustrates that coherent sheaf cohomology on a Deligne–
Mumford stack can be nonzero in arbitrary high degrees. This doesn’t happen in
characteristic 0 or for tame Deligne–Mumford stacks (see Corollary 4.4.23).
Exercise 4.1.49. Let k be a field of characteristic p > 0 and let Z/pZ be the
group scheme over Spec k corresponding to the abstract cyclic group. Show that
Hi (B(Z/pZ), OB(Z/pZ) ) = k for each i ≥ 0.
Exercise 4.1.50 (hard). Generalize Grothendieck’s Vanishing Theorem [Har77,
Thm. III.2.7] to algebraic spaces: if X is a noetherian algebraic space and F is a
quasi-coherent sheaf, then Hi (X, F ) = 0 for i > dim X. See also [SP, Tag 0A4Q].

4.2 Quotients by finite groups

Symmetry, as wide or narrow as you may define its meaning, is one idea
by which man through the ages has tried to comprehend and create order,
beauty, and perfection.

Hermann Weyl

172
 Quotient stacks [Spec A/G] of affine schemes by finite abstract groups form
a particularly nice class of Deligne–Mumford stacks. Their geometry is the G-
equivariant geometry of Spec A (see Table 0.6.15). In this section, we show that the
natural map [Spec A/G] → Spec AG is a coarse moduli space (Theorem 4.2.3). In
the next section, we will show that every Deligne–Mumford stack is étale locally
isomorphic to a quotient stack of the form [Spec A/G] (Theorem 4.3.1). Combined
these two theorems will be applied to establish the existence of a coarse moduli
spaces, i.e., the Keel–Mori Theorem (4.4.6).

4.2.1 Coarse moduli spaces and geometric quotients

Definition 4.2.1 (Coarse moduli spaces). A morphism π : X → X from an algebraic
stack to an algebraic space is a coarse moduli space if
(1) for every algebraically closed field k, the induced map X (k)/∼ → X(k), from
the set of isomorphism classes of objects of X over k, is bijective, and
(2) π is universal for maps to algebraic spaces, i.e., every map X → Y to an
algebraic space factors uniquely as

X
π

X / Y.

Definition 4.2.2 (Geometric quotients). If G is an fppf group scheme over a scheme

S (e.g., a finite group scheme associated to a finite abstract group) acting on an
algebraic space U over S, a G-invariant morphism U → X over S to an algebraic
space is a geometric quotient if [U/G] → X is a coarse moduli space, i.e.,
(1) for every algebraically closed field k, the map U → X induces a bijection
∼
U (k)/G → X(k), and
(2) U → X is universal for G-invariant maps to algebraic spaces, i.e., for every
G-invariant map U → Y to an algebraic space, there is a unique map X → Y
such that U → Y factors as U → X → Y .

If π : U → X is a geometric quotient, we often write X = U/G. A geometric

quotient U → U/G factors as

U / [U/G]
coarse moduli
geometric space
quotient ! 
U/G

In the case that G acts freely on U (i.e., the action map G × U → U × U is a

monomorphism), then we have already defined the algebraic space quotient U/G
(Corollary 3.1.14) and this is identified with the geometric quotient (Corollary 4.2.13).
If a finite abstract group G acts on an affine scheme Spec A, then G also acts on
the ring A. We define the invariant ring as

AG = {f ∈ A | g · f = f for all g ∈ G}.

The main result of this section is that Spec A → Spec AG is a geometric quotient.

173
Theorem 4.2.3. If G is a finite abstract group acting on an affine scheme Spec A,
then π : [Spec A/G] → Spec AG is a coarse moduli space. Moreover,
(1) π is a universal homeomorphism;
(2) the base change of π along a flat morphism X ′ → Spec AG of algebraic spaces
is a coarse moduli space;
(3) the natural map OX → π∗ OX is an isomorphism; and
(4) if A is finitely generated over a noetherian ring R, then AG is finitely generated
over R and π is proper.
Quotients of schemes by finite groups is covered in many places in the literature,
e.g., [Bou89, §V.9], [Ser88, §III.12], and [Mum70a, p. 66]. Our proof below follows
standard arguments for the existence of the quotient scheme U/G, but it requires
additional effort to prove that it is universal in the category of algebraic spaces.
Example 4.2.4. Assume char(k) ̸= 2. If G = Z/2 acts on A1 = Spec k[x] via
−1 · x = −x, then k[x]G = k[x2 ]. The geometric quotient is the map A1 =
Spec k[x] → Spec k[x2 ] = A1 sending t to t2 . The map [A1 /G] → Spec k[x2 ] is the
coarse moduli space.
Similarly, if G = Z/2 acts on A2 = Spec k[x, y] via −1 · (x, y) = (−x, −y), then
k[x, y]G = k[x2 , xy, y 2 ] and the geometric quotient is A2 → A2 /G = Spec k[x2 , xy, y 2 ].
By setting A = x2 , B = xy and C = y 2 , the invariant ring can be identified with
k[A, B, C]/(B 2 − AC) so the quotient A2 /G is a cone over a conic and in particular
singular.

[Link]
IA x̅

ALTA
IT
11 44 7422 IN722
A

4.2.2 Properties of invariant rings

Lemma 4.2.6. If G is a finite abstract group acting on a R-algebra A via R-algebra
automorphisms, then AG → A is integral. If R is noetherian and A is finitely
generated over R, then AG → A is finite and AG is finitely generated over R.
Proof. ToQsee that AG → A is integral, observe that for every element a ∈ A, the
product g∈G (x − ga) ∈ AG [x] is polynomial with invariant coefficients which has
a as a root. If R is noetherian and R → A is of finite type, then AG → A is also of
finite type. Because AG → A is integral and of finite type, it is finite (c.f., [AM69,
Cor. 5.2]). Since R is noetherian, we may conclude by the Artin–Tate Lemma (c.f.,
[AM69, Prop. 7.8]) that R → AG is of finite type.

174
 The invariant ring is compatible with flat base change.
Lemma 4.2.7. Let G be a finite abstract group acting on an affine scheme Spec A. If
AG → B is a flat ring homomorphism, then G acts on the affine scheme Spec(B ⊗AG
A) and B = (B ⊗AG A)G .
Proof. By definition, the invariant ring is the equalizer
p1 Y
0 → AG → A −
−⇒ A
p2
g∈G

where p1 (f ) = (f )g∈G and p2 (f ) = (gf )g∈G . Since AG → B is flat, we have that

p1 Y
0 → B → A ⊗AG B −
−⇒ A ⊗AG B
p2
g∈G

is also exact, and we conclude that B = (B ⊗AG A)G .

Exercise 4.2.8. Let AG → B be a ring homomorphism and consider the commuta-
tive diagram

Spec B ⊗AG A / Spec A

□
u  
Spec(B ⊗AG A)G / Spec B / Spec AG .

(a) Show that Spec(B ⊗AG A)G → Spec B is an integral homeomorphism.

(b) If |G| is invertible in A, show that B → (B ⊗AG A)G is an isomorphism.
(c) Provide an example where B → (B ⊗AG A)G is not an isomorphism.
Hint: In characteristic p, consider the action of Z/p on A = k[x, y] where a
generator acts via (x, y) 7→ (x + y, y) and take B = AG /(y).

4.2.3 Properties of coarse moduli spaces

Lemma 4.2.9. Let π : X → X be a coarse moduli space such that for every étale
morphism X ′ → X from an affine scheme, the base change X ×X X ′ → X ′ is a
coarse moduli space. Then the natural map OX → π∗ OX is an isomorphism.
Proof. As π is universal for maps to algebraic spaces, we have that Map(X, A1 ) →
Map(X , A1 ) is bijective or in other words Γ(X, OX ) ∼ = Γ(X , OX ). For every étale
map X ′ → X, the base change X ′ = X ×X X ′ → X ′ is also a coarse moduli space and
thus Γ(X ′ , OX ′ ) ∼
= Γ(X ′ , OX ′ ). This shows that OX → π∗ OX is isomorphism.
The property of being a coarse moduli space is étale local.
Lemma 4.2.10. Let π : X → X be a morphism to an algebraic space. Suppose that
there is an étale covering {Xi → X} such that X ×X Xi → Xi is a coarse moduli
space for each i. Then π : X → X is a coarse moduli space.

Proof. Axiom (1) of a coarse moduli space is a condition on geometric fibers and can
thus be checked étale locally, while Axiom (2) follows from the fact that algebraic
spaces are sheaves in the étale topology.

175
4.2.4 Spec A → Spec AG is a geometric quotient
Proposition 4.2.11. If G is a finite abstract group acting on an affine scheme
Spec A, then Spec A → Spec AG is a geometric quotient.

Proof. Consider the commutative diagram

U = Spec A
π
e

 (
X = [U/G]
π / X = Spec AG .

Since πe is integral and dominant, it is surjective. To see that π e is injective on

G-orbits of geometric points, let k be an algebraically closed field and suppose that
x, x′ ∈ U (k) are points such that under the induced G-action on U ×X k, the G-orbits
of x, x′ ∈ U ×X k are distinct. As the orbits are disjoint and
Q there exists a function
f ∈ A ⊗AG k with f |Gx = 0 and f |Gx′ = 1. Then fe = g∈G gf ∈ (A ⊗AG k)G
is a G-invariant function with fe(x) = 0 and fe(x′ ) = 1, and this implies that
e(x) ̸= π
π e(x′ ) ∈ X(k).
It remains to show that the natural map

Map(X, Y ) → Map(X , Y ) (4.2.12)

is bijective for every algebraic space Y . We note that this is immediate when Y
is affine as Γ(X , OX ) = Γ(X, OX ). (The case when Y is a scheme can be reduced
to this case without much effort: if g : X → Y is a map, an affine covering Yi of Y
induces an open covering Xi = X ∖ π(X ∖ g −1 (Yi )) of X, and g restricts to a map
π −1 (Xi ) → Yi which factors uniquely through Xi since π∗ OX = OX ; see also [GIT,
§0.6].)
For the injectivity of (4.2.12), let h1 , h2 : X → Y be two maps such that h1 ◦ π =
h2 ◦ π. Let E → X be the equalizer of h1 and h2 , i.e., the pullback of the diagonal
Y → Y × Y along (h1 , h2 ) : X → Y × Y . The equalizer E → X is a monomorphism
and locally of finite type. By construction π : X → X factors through E → X
and since π is universally closed and schematically dominant (i.e., OX → π∗ OX
is injective), so is E → X. As every universally closed and locally of finite type
monomorphism is a closed immersion (see Remark A.7.6), we conclude that E → X
is an isomorphism.
For the surjectivity of (4.2.12), let g : X → Y be a morphism. We will begin
with the easy reductions. By Lemmas 4.2.7 and 4.2.10, the question is étale-local on
X and, in particular, we can assume that X is affine. Moreover, since X is quasi-
compact, we may assume that Y is quasi-compact as g : X → Y factors through a
quasi-compact open algebraic subspace of Y . Let Y ′ → Y be an étale presentation
from an affine scheme and let X ′ := X ×Y Y ′ .
We claim that after replacing X with an étale cover V → X and X with the
base change X ×X V , there is a section s : X → X ′ of X ′ → X in the commutative
diagram
s
w
X′ /X π /X

g′ □ g
  ~
Y′ / Y.

176
The surjectivity of (4.2.12) follows from this claim: since X and Y ′ are affine,
s g′
the equality Γ(X, OX ) = Γ(X , OX ) implies that X − → X ′ −→ Y ′ factors through
π : X → X via a morphism X → Y . The composition X → Y ′ → Y yields the
′

desired dotted arrow above.

To establish the claim, we will use limit methods to further reduce to the case
that X = Spec AG is the spectrum of a strictly henselian local ring. For a closed
point u of U := Spec A over x ∈ |X |, the strict henselization X sh := OX,π(x)
sh
is the
limit limλ Xλ over all affine étale neighborhoods Xλ → X of π(x). The base change
U sh := U ×X X sh is the limit of the affine schemes Uλ := U ×X Xλ . We also set
X sh := X ×X X sh = [U sh /G] and Xλ := X ×X Xλ = [Uλ /G]. Since X ′ → X is
locally of finite presentation, the natural map

colimλ MorX (Xλ , X ′ ) → MorX (X sh , X ′ )

is an equivalence; this follows from Exercise 3.3.33 (which asserts the equality when
the first terms Xλ and X sh are affine) by expressing MorX (X sh , X ′ ) as the equalizer
of MorX (U sh , X ′ ) ⇒ MorX (G × U sh , X ′ ) and similarly for the left-hand side. A
section of X ′ ×X X sh → X sh is determined by a map X sh → X ′ over X . This map
extends to a morphism Xλ → X ′ for some λ, giving us the desired section.
Let κ be the residue field of AG . As AG → A is finite, A = A1 × · · · × Ar is a
product of strictly henselian local rings, each finite over AG (Proposition B.5.10).
If u ∈ Spec A1 ⊆ Spec A is a closed point, then Spec A1 is Gu -invariant and the
orbit Gu is in bijection with the r connected components of Spec A. There is an
isomorphism X ∼ = [Spec A1 /Gu ]; this can be verified directly by for instance slicing
the groupoid G × Spec A ⇒ Spec A by Spec A1 ,→ Spec A (as in Exercise 3.4.20).
We may thus replace X = [Spec A/G] with [Spec A1 /Gu ], and we can assume that
there is a unique closed point u ∈ Spec A which is set-theoretically fixed by G. As
Y ′ → Y is representable by schemes, we can write X ′ = [U ′ /G] for a scheme U ′ . Let
u′ ∈ U ′ be a preimage of u ∈ Spec A. As A is strictly henselian and the G-equivariant
morphism U ′ → U is the base change of the étale morphism Y ′ → Y , we see that
κ(u′ ) = κ(u) and Gu′ = Gu = G, and moreover the stabilizers act trivially on the
residue fields. Again using that A is strictly henselian, there is a unique section
s : Spec A → U ′ with s(u) = u′ (Proposition B.5.9). This section is G-invariant
g −1
because for every g ∈ G, both s ◦ g and g ◦ s are sections of U ′ → Spec A −−→ Spec A
with u′ 7→ u, and therefore s ◦ g = g ◦ s by the uniqueness of the sections. It follows
that s : Spec AS → U descends to a section X = [Spec A/G] → [U ′ /G] = X ′ of
X ′ → X . This finishes the proof that Spec A → Spec AG is a geometric quotient.

We now conclude not only that [Spec A/G] → Spec AG is a coarse moduli space,
but that it enjoys properties (1)–(4) of Theorem 4.2.3.

Proof of Theorem 4.2.3. Proposition 4.2.11 implies that π : [Spec A/G] → Spec AG
is a coarse moduli space. To show property (1), i.e., that π is a universal homeo-
morphism, observe that since π is bijective and closed, its set-theoretic inverse is
continuous, and thus π is a homeomorphism. The base change of π along a morphism
Spec B → Spec AG factors as [Spec(B ⊗AG A)/G] → Spec(B ⊗AG A)G → Spec B
where the first map is a homeomorphism by the above argument and the second
is a homeomorphism by Exercise 4.2.8. Property (2), i.e., the base change of π
by a flat morphism X ′ → Spec AG is also a coarse moduli space, follows from
Lemma 4.2.7 after reducing to the case that X ′ is affine using Lemma 4.2.10. Prop-
erty (3) that OX → π∗ OX is an isomorphism follows direct from (2) by Lemma 4.2.9.

177
Finally, Lemma 4.2.6 implies (4): if R is noetherian and R → A is of finite type,
then R → AG is of finite type. In particular, AG → A is of finite type and
π : [Spec A/G] → Spec AG is separated, universally closed, and of finite type, i.e.,
proper.

Corollary 4.2.13. Let G be a finite abstract group acting freely on an affine scheme
U = Spec A, then the algebraic space quotient U/G is isomorphic to Spec AG .

Proof. Both the algebraic space quotient and the geometric quotient are universal
for maps to algebraic spaces.

Exercise 4.2.14. Let G be a finite abstract group acting on a scheme U .

(a) Show that the geometric quotient U/G is a scheme if and only if every G-orbit
is contained in an affine open subscheme.
(b) If U is affine (resp., quasi-affine, quasi-projective, projective) scheme over a
noetherian ring R, show that U/G is affine (resp., quasi-affine, quasi-projective,
projective) scheme over R.

Exercise 4.2.15. Suppose that G is a finite abstract group acting on an affine

scheme Spec A of finite type over a noetherian ring R. If x ∈ Spec A is a closed
point, show that there is an isomorphism

bGx ∼
A =AcG

between the Gx -invariants of the completion at Spec A at x and the completion of

Spec AG at the image of x.

The following exercise generalizes Theorem 4.2.3 from quotients of finite groups
to quotients of finite flat groupoids.

Exercise 4.2.16 (hard). Let s, t : R ⇒ U be a finite flat groupoid of affine schemes,

and define AR ⊆ A as the subring of R-invariants, i.e., the subring of elements a ∈ A
such that s∗ a = t∗ a ∈ Γ(R, OR ). Show that [U/R] → X := Spec AR is a coarse
moduli space with properties (1)–(4) of Theorem 3.4.13.
Hint: ToQshow that AR → A is integral, for an element a ∈ A, rather than using the
product g∈G (x − ga) ∈ AG [x] as in the proof of Lemma 4.2.6, use the characteristic
polynomial of δ0 (a) ∈ B as a locally free module over A via δ1 where δ0 , δ1 : A ⇒
B = Γ(U, OU ) are the two finite flat ring maps. See also [SGA3I , Thm. V.4.1],
[KM97, Prop. 5.1], [Con05b, Thm. 3.1], [Ols16, Thm. 6.2.2], and [SP, Tag 0DUR].

4.3 The local structure of Deligne–Mumford stacks

If only I had the theorems! Then I should find the proofs easily enough.

attributed to Bernhard Riemann

We show that a Deligne–Mumford stack X near a point x is étale locally the

quotient stack [Spec A/Gx ] of an affine scheme by the stabilizer group scheme.
Conceptually, this tells us that just as schemes (resp., algebraic spaces) are obtained
by gluing affine schemes in the Zariski topology (resp., étale topology), Deligne–
Mumford stacks are obtained by gluing quotient stacks [Spec A/G] in the étale

178
topology.1 This has the practical application of allowing one to reduce many
properties of Deligne–Mumford stacks to quotient stacks [Spec A/G]. We will exploit
the Local Structure Theorem to construct a coarse moduli space (Theorem 4.4.6).

4.3.1 The Local Structure Theorem

The geometric stabilizer of a point x ∈ |X | of a Deligne–Mumford stack X is the
abstract group defined as the stabilizer of any geometric point Spec k → X with
image x.

Theorem 4.3.1 (Local Structure Theorem of Deligne–Mumford Stacks). Let X

be a quasi-separated Deligne–Mumford stack and x ∈ |X | be a finite type point with
geometric stabilizer Gx . There exists an étale morphism

f : ([Spec A/Gx ], w) → (X , x)

where w ∈ [Spec A/Gx ] such that f induces an isomorphism of geometric stabilizer

groups at w. Moreover, if X has separated (resp., affine) diagonal, then it can be
arranged that f is representable and separated (resp., affine).

Proof. We first handle the case when X has affine diagonal. Choose a geometric point
Spec k → X of representing x, and let d be the cardinality of Gx . Viewing Gx as a
group scheme over Spec k, let BGx → X be the induced map. Let (U, u) → (X , x) be
an étale representable morphism from an affine scheme. Since X has affine diagonal,
U → X is affine. Define the quasi-affine scheme

SECd := U ×X · · · ×X U ∖∆,
| {z }
d times

where ∆ is the union of all pairwise diagonals A map S → SECd from a scheme is
classified by a morphism S → X and d sections s1 , . . . , sd of US := U ×X S → S
which are disjoint (i.e., the intersection of si and sj is empty for i ̸= j). There is an
action of Sd on SECd given by permuting the sections and we define the quotient
stack
ETd := [SECd /Sd ].
By the correspondence between principal Sd -bundles and finite étale covers of degree
d (Exercise B.1.52), an object of ETd over a scheme S corresponds to a diagram

Z / US /U

□
 
S /X

where Z ,→ US is a closed subscheme and Z → S is finite étale of degree d. There is

an induced étale morphism ETd → X and a commutative diagram

SEC _ d / ETd

 !
U ×X · · · ×X U /U /X
1 Of course, Deligne–Mumford stacks are also étale locally schemes but the étale neighborhoods

([Spec A/Gx ], w) → (X , x) produced by Theorem 4.3.1 preserve the stabilizer group at w.

179
To construct a preimage w ∈ ETd (k) of x, we observe that the fiber product
BGx ×X U is a finite disjoint union of Spec k’s and we may choose one of them.
This leads to a commutative

Spec k  / ` Spec k /U

□
%  
BGx /X

with Spec k → BGx finite étale of degree d, and this defines a lift BGx → ETd of
BGx → X . The composition w : Spec k → BGx → ETd is a preimage of x. We
claim the map ETd → X induces an isomorphism of geometric stabilizers at w.
To see this, it suffices to assume that X = BGx and U = Spec k. In this case,
there is an isomorphism U ×X · · · ×X U ∼ x , which we can further identify
= Gd−1
with the quotient Gx /Gx of the diagonal action. Then SECd ⊆ U ×X · · · ×X U is
d

identified with the quotient (Gdx ∖ ∆)/Gx . The permutation Sd -action is transitive
with stabilizer isomorphic to Gx . In fact, SECd is Sd -equivariantly isomorphic to
the quotient Sd /G of the regular representation G ⊆ Sd , and thus ETd ∼ = BGx .
Since ETd is separated and X has separated diagonal, the relative inertia stack
IETd /X → ETd is finite. As ETd → X induces an isomorphism of stabilizers at w,
IETd /X → ETd is an isomorphism over an open substack [W/Sd ] ⊆ ETd around w,
where W ⊆ SECd is a quasi-affine scheme. It follows that [W/Sd ] → X is a separated,
étale, and representable morphism inducing an isomorphism on stabilizer groups
at w. By quotienting out by Gx ⊆ Sd instead, [W/Gx ] → X is also a separated,
étale, and representable morphism inducing an isomorphism of stabilizer groups at
w. Letting W ′ ⊆ W be an affine open subschemeT containing w, we may replace W
with the Gx -invariant affine open subscheme g∈Gx g · W ′ . It remains to show that
[W/Gx ] → X is affine. Since X has affine diagonal, the morphism W → X from the
affine scheme W is affine. The fiber product

[W/Gx ] ×X W /W

□
 
[W/Gx ] /X

is affine over [W/Gx ] and thus isomorphic to a quotient stack [Spec B/Gx ]. On the
other hand, since [W/Gx ] → X is representable, the quotient stack [Spec B/Gx ]
is an algebraic space and the action of Gx on Spec B is free. By Corollary 4.2.13,
[Spec B/Gx ] is isomorphic to the affine scheme Spec B Gx . By étale descent [W/Gx ] →
X is affine.
If X has separated diagonal, then the above argument produces a separated
scheme SECd with an action of Gx and an separated, étale, and representable
morphism ([SECd /Gx ], w) → (X , x) inducing an isomorphism at w. By applying
the argument again to the separated Deligne–Mumford stack [SECd /Gx ] gives the
desired local quotient presentation. If X is only quasi-separated, the argument
produces a (not necessarily representable) morphism ([SECd /Gx ], w) → (X , x)
inducing an isomorphism at w where SECd is a quasi-separated algebraic space. As
SECd has separated diagonal (as the diagonal is a monomorphism), [SECd /Sd ] is a
Deligne–Mumford stack with separated diagonal, and we may apply the statement
in the case of separated diagonal to obtain a local quotient presentation by the
quotient of an affine scheme. See also [LMB00, Thm. 6.2].

180
Exercise 4.3.2. Let X be a Deligne–Mumford stack. Show that X is isomorphic to
a quotient stack [U/G], where U is an affine scheme (resp., scheme, algebraic space)
and G is a finite abstract group, if and only if there exists a finite étale morphism
V → X from an affine scheme (resp., scheme, algebraic space).
Hint: If V → X is a finite étale cover of degree d, consider the associated principal
Sd -torsor V ×X · · · ×X V ∖∆ → X as in Exercise B.1.52.
| {z }
d times

Proposition 4.3.3. If R ⇒ U is a finite étale equivalence relation of affine schemes,

then the algebraic space quotient U/R is an affine scheme.
Proof. By Exercise 4.3.2, the algebraic space U/R is isomorphism to V /G for the free
action of a finite group G on an affine scheme V = Spec B. Theorem 4.2.3 shows that
V /G → Spec B G is universal for maps to algebraic spaces and thus an isomorphism.
Alternatively, this follows from Exercise 4.2.16: if U = Spec A, then U/R → Spec AR
is universal for maps to algebraic spaces and thus an isomorphism.

4.3.2 Lifting rational points to presentations

With a similar technique to the proof of the Local Structure Theorem (4.3.1), we
can prove the existence of presentations with a lift of a given field-valued point.
Proposition 4.3.4. If X is an algebraic stack with separated diagonal (resp.,
Deligne–Mumford stack with separated and quasi-compact diagonal) and x ∈ X (k) is
a field-valued point, then there exists a smooth (resp., étale) morphism U → X from
an affine scheme and a point u ∈ U (k) over x.
Proof. Let U → X be a smooth morphism from an affine scheme such that x is
contained in its image. Consider the fiber product
Ux /U

□
 
Spec k
x / X.

Since X has separated diagonal, the morphism U → X is separated. As Ux is finite

type and quasi-separated over Spec k, it is a noetherian algebraic space. Choosing a
closed point u ∈ Ux , by the Existence of Residual Gerbes (3.5.16), there is a closed
immersion Spec κ(u) → Ux from the residue field. The field extension k → κ(u) is
finite and separable, and we let d be its degree.
Following the notation of the proof of Theorem 4.3.1, we consider the open
subspace SECd ⊆ U ×X · · · ×X U parameterizing d disjoint sections of U → X , which
is a separated algebraic space. Define the separated Deligne–Mumford stack ETd =
[SECd /Sd ] and consider the induced morphism ETd → X . As Spec κ(u) → Spec k
is finite étale of degree d, the closed immersion Spec κ(u) ,→ Ux defines a k-point v
of ETd . This gives a commutative diagram
Spec k
v
x
{ 
SECd / ETd / X.

The point v ∈ ETd (k) does not necessarily lift to SECd , but we can use the
following trick to choose a different quotient presentation of ETd where v does lift.

181
Namely, choose a faithful representation Sd ⊆ GLn and write ETd ∼ = [V / GLn ] where
V = SECd ×Sd GLn is a separated algebraic space. Then v : Spec k → [V / GLn ]
corresponds to a principal GLn -bundle P → Spec k and a GLn -equivariant map P →
V . Since principal GLn -bundles are in bijection to vector bundles (Exercise B.1.56),
P is the trivial principal GLn -bundle and there is a section Spec k → P . The
composition V → [V / GLn ] → X is smooth and the composition Spec k → P → V
is a lift of x.
As V is a separated algebraic space, we are reduced to showing the second
statement, i.e., a k-point of a Deligne–Mumford stack X with separated and quasi-
compact diagonal lifts to an étale neighborhood by a scheme. We will use the fact that
X has quasi-affine diagonal, which is proven independently later in Corollary 4.5.8
(the proof only uses Proposition 4.3.3 above and the theory of quasi-coherent sheaves
in §4.1). We repeat the above argument by choosing an étale map U → X from an
affine scheme such that the image contains x. Then the space SECd of d disjoint
sections with respect to U → X is a quasi-affine scheme with a free action of Sd .
The quotient ETd = SECd /Sd is also quasi-affine (Exercise 4.2.14). The induced
map ETd → X is étale and by construction the k-point x lifts to a k-point of ETd .
See also [LMB00, Thm. 6.3].

There is also a version of the Local Structure Theorem formulated étale locally
on the coarse moduli space (see Corollary 4.4.13).

4.4 Coarse moduli spaces and the Keel–Mori Theo-

rem
As for Mg there is virtually no doubt that it can be provided with the
structure of an algebraic variety.

André Weil [Wei58, p. 383]

We prove that every separated Deligne–Mumford stack X of finite type over a

noetherian scheme admits a separated coarse moduli space π : X → X, which we
refer to as the Keel–Mori Theorem (4.4.6). One can view this theorem as a way
to remove the stackiness of a Deligne–Mumford stack: at the expense of sacrificing
universal properties of X (e.g., existence of a universal family), one can replace X
with an algebraic space without changing the underlying topological space. Once we
introduce the Deligne–Mumford stack Mg of stable curves (Theorem 5.4.14) and
show that it is separated (Proposition 5.5.20), the Keel–Mori Theorem implies that
there is a coarse moduli space Mg → M g , where M g is a separated algebraic space.
Moreover, Semistable Reduction (5.5.9), i.e., the properness of Mg , implies further
that M g is proper. In §5.8 and §5.9, we give two proofs that M g is projective.
Coarse moduli spaces were introduced in Definition 4.2.1. If G is a finite abstract
group acting on an affine scheme Spec A, then [Spec A/G] → Spec AG is a coarse
moduli space (Theorem 4.2.3), or in other words Spec A → Spec AG is a geometric
quotient. To prove the Keel–Mori Theorem (4.4.6), we will apply the Local Structure
Theorem (4.3.1) to construct étale neighborhoods [Spec Ai /G] → X and then glue
the geometric quotients Spec AG i in the étale topology to construct a coarse moduli
space of X .

182
4.4.1 Descending étale morphisms to geometric quotients
In order to glue geometric quotients in the étale topology, we will need a condition
ensuring that an étale G-equivariant morphism Spec A → Spec B induces an étale
morphism Spec AG → Spec B G on geometric quotients. The following proposition
can be viewed as a baby version of Luna’s Fundamental Lemma (6.5.28).

Proposition 4.4.1. Let G be a finite abstract group and f : Spec A → Spec B be a

G-equivariant morphism of affine schemes of finite type over a noetherian ring R.
Let x ∈ Spec A be a closed point. Assume that
(a) f is étale at x and
(b) the induced map Gx → Gf (x) of stabilizer group schemes is bijective.
Then there is an open affine neighborhood W ⊆ Spec AG of the image of x such that
W → Spec AG → Spec B G is étale and πA −1
(W ) ∼
= W ×Spec B G [Spec B/G], where
G
πA : [Spec A/G] → Spec A .

Remark 4.4.2. In other words, after replacing Spec AG with an affine neighborhood
−1
W of πA (x) and Spec A with πA (W ), it can be arranged that the diagram

[Spec A/G]
f
/ [Spec B/G]

πA □ πB (4.4.3)
 
Spec AG / Spec B G

is cartesian where both horizontal maps are étale.

Proof of Proposition 4.4.1. Set y = f (x). We first claim that the question is étale
local around πB (y) ∈ Spec B G . Indeed, if Y ′ → Y := Spec B G is an affine étale
neighborhood of πB (y), we let X ′ , X ′ and Y ′ denote the base changes of X :=
Spec AG , X := [Spec A/G], and Y := [Spec B/G]. By Lemma 4.2.7, we know that
Y′ ∼= [Spec B ′ /G] with Y ′ ∼ = Spec B ′G and similarly for X ′ and X ′ . If the result
holds after this base change, there is an open neighborhood W ′ ⊆ X ′ containing a
preimage of πA (x) such that W ′ ,→ X ′ → Y ′ is étale and such that the preimage
of W ′ in X ′ is isomorphic to W ′ ×Y ′ Y ′ . Taking W as the image of W ′ under
X ′ → Spec AG and applying étale descent yields the desired claim.
We now claim that this allows us to assume that B G is strictly henselian. To see
this, define Y sh = Spec OY,π
sh
B (y)
, and let X sh , X sh , and Y sh be the base changes of
X, X , and Y along Y → Y . We can write Y sh = limλ Yλ as the limit of affine étale
sh

neighborhood Yλ → Y of y, and we set Xλ , Xλ , and Yλ to be the base changes of

X, X , and Y along Yλ → Y . Suppose U sh ⊆ X sh is an open affine subscheme of the
unique point in X sh over x and the closed point of Y sh such that U sh → Y sh is étale
−1 sh ∼
with πX sh (U ) = U sh ×Y sh Y sh . By Proposition B.3.3, the morphism U sh → X sh
descends to Uη → Xη for some η. Setting Uλ = Uη ×Xη Xλ for λ > η, it follows
from Proposition B.3.7 that for λ ≫ 0 (a) Uλ → Xλ is an open immersion, (b) the
composition Uλ → Xλ → Yλ is étale, and (c) πX −1
λ
(Uλ ) ∼
= Uλ ×Yλ Yλ (by arguing on
the étale presentations of X and Y).
Finally, As B G → B is finite (Lemma 4.2.6), B = B1 × · · · × Br is a product
of strictly henselian local rings (Proposition B.5.10). As in the proof of The-
orem 4.2.3, we may replace [Spec B/G] with [Spec B1 /Gy ] and [Spec A/G] with
[f −1 (Spec B1 )/G] to assume that G fixes x and y while acting trivially on the residue
fields κ(x) = κ(y). Thus Spec A → Spec B has a unique section s : Spec B → Spec A

183
taking y to x. The section s is necessarily G-invariant (as in the proof Theorem 4.2.3).
Thus s descends to a section of Spec AG → Spec B G which gives our desired open
and closed subscheme W ⊆ Spec AG .

Remark 4.4.4. Here is a conceptual reason why we should expect the induced map
of quotients to be étale. For simplicity, assume that R = k is an algebraically closed
field. Let A
b and Bb be the completions of the local rings at x and f (x). The stabilizers
Gx and Gf (x) act on Spec A b and Spec B,
b respectively, and the map Spec A b → Spec B b
is equivariant with respect to the map Gx → G . The completion A of A at the
f (x)
cG G

image of x is isomorphic to A
bGx (Exercise 4.2.15) and similarly B
d b Gf (x) . Since
G =B

f is étale at x, B
b→A b is an isomorphism and since Gx → Gf (x) is bijective, the
induced map B → A is an isomorphism which shows that Spec AG → Spec B G is
d G cG

étale at the image of x.

The above proposition will be applied in the following form in the proof of the
Keel–Mori Theorem (4.4.6).

Corollary 4.4.5. Let G be a finite abstract group and f : Spec A → Spec B be a

G-equivariant morphism of affine schemes of finite type over a noetherian ring R.
Assume that for every closed point x ∈ Spec A,
(a) f is étale at x and
(b) the induced map Gx → Gf (x) of stabilizer group schemes is bijective.
Then Spec AG → Spec B G is étale and (4.4.3) is cartesian.

4.4.2 The Keel–Mori Theorem

We now state and prove the Keel–Mori Theorem.

Theorem 4.4.6 (Keel–Mori Theorem). Let X be a Deligne-Mumford stack separated

and of finite type over a noetherian algebraic space S. Then there exists a coarse
moduli space π : X → X with OX = π∗ OX such that
(1) X is separated and of finite type over S,
(2) π is a proper universal homeomorphism, and
(3) for every flat morphism X ′ → X of algebraic spaces, the base change X ×X
X ′ → X ′ is a coarse moduli space.
Moreover, X is proper if and only if X is.

This theorem extends to (possibly non-separated) Deligne–Mumford stacks X

such that IX → X is finite (Exercise 4.4.10), except that the coarse moduli space
may not be separated. More impressively, the Keel–Mori Theorem [KM97] also
holds if we replace the hypothesis that X is Deligne–Mumford with the hypothesis
that X is an algebraic stack with finite inertia IX → X ; see Remark 4.4.11. This
requires all stabilizers to be finite but not necessarily reduced. The theorem also
holds without any noetherian or finiteness conditions; see [Con05b], [Ryd13], and
[SP, Tag 0DUK].
Remark 4.4.7 (Historical remarks). Around roughly the same time as [KM97] was
published, Kollár proved an important special case, which is sufficient for many
applications: if G is an affine group scheme over an excellent schemes S acting
acting properly on a separated algebraic space U over S such that either G → S is a

184
reductive group scheme or S = Spec k with char(k) > 0, then there exists a separated
geometric quotient U/G [Kol97, Thm. 1.5]. Before these articles, the existence of a
coarse moduli spaces was viewed as a folklore theorem, but its statement and proof
for separated Deligne–Mumford stacks was surely known to experts. For instance,
the theorem was was explicitly formulated (but not proven) in [FC90, Thm. I.4.10]
and [Kol90, Thm. 2.2]. An arguments using analytic techniques was sketched by
Popp in the setting of a group action over C in [Pop74, Thm. p. 55], and both
analytic and algebraic arguments were sketched in [MF82, §5.A].
Remark 4.4.8. We emphasize that this theorem not only produces a coarse moduli
space, but that it satisfies properties (1)-(3). It can be difficult to work with the notion
of a coarse moduli space that only satisfies the defining properties (Definition 4.2.1).
For instance, it is not clear in general that coarse moduli spaces are stable under
étale base change (or even open immersions) or that π∗ OX = OX .

Proof. We first handle the case when S = Spec R is affine. The question is Zariski-
local on X : if {Xi } is a Zariski open covering of X with coarse moduli spaces
Xi → Xi , then since coarse moduli spaces are unique (Definition 4.2.1(2)), each
intersection Xi ∩Xj admits a coarse moduli space Uij identified with open subschemes
of both Xi and Xj , we may glue to form an algebraic space X and a map X → X,
which is a coarse moduli space by Lemma 4.2.10. It thus suffices to show that every
closed point x ∈ |X | has an open neighborhood that admits a coarse moduli space.
By the Local Structure Theorem of Deligne–Mumford Stacks (4.3.1), there exists
an affine étale morphism


f : W = [Spec A/Gx ], w → (X , x)

such that f induces an isomorphism of geometric stabilizer groups at w. We claim

that the locus U consisting of points z ∈ |W|, such that f induces an isomorphism
of geometric stabilizer groups at z, is open. To establish this, we will exploit that
the relative inertia IX → X is finite, which follows from the separatedness of X (as
the diagonal is both proper and quasi-finite). The fiber of the natural morphism
IW → IX ×X W of relative group schemes over W over z ∈ W(k) is precisely the
morphism Gz → Gf (z) of stabilizers. We will use the cartesian diagram

IW
Ψ / IX ×X W

□
 
W / W ×X W

from Exercise 3.2.15. Since W → X is representable, étale, and separated, the

diagonal W → W ×X W is an open and closed immersion and thus so is Ψ. Since
IX → X is finite, so is p2 : IX ×X W → W. Thus p2 (|IX ×X W| ∖ |IW |) ⊆ |W|
is closed and its complement, which is identified with the locus U, is open. Let
πW : W → W = Spec AGx be the coarse moduli space (Theorem 4.2.3). Choose
−1
an affine open subscheme X1 ⊆ π(U) containing πW (w). Then X1 = πW (X1 )
is isomorphic to a quotient stack [Spec A1 /Gx ] such that X1 = Spec A1 . This
Gx

provides an affine étale morphism

g : (X1 = [Spec A1 /Gx ], w) → (X , x)

which induces a bijection on all geometric stabilizer groups.

185
 We now show that the open substack X0 := im(f ) admits a coarse moduli
space. Since g is affine and X1 = [Spec A1 /Gx ], we can express X2 := X1 ×X X1
as [Spec A2 /Gx ], and there is a coarse moduli space π2 : X2 → X2 = Spec AG
2 . By
x

universality of coarse moduli spaces, there is a diagram

X2 // X g
/ X0 = im(g)
1

π2 □ π1 π0 (4.4.9)
  
X2 // X / X0
1

where the natural squares commute. Since g induces bijections of geometric stabilizer
groups at all points, the same is true for each projection X2 → X1 . Corollary 4.4.5
implies that each map X2 → X1 is étale, and the natural squares of solid arrows in
(4.4.9) are cartesian.
The universality of coarse moduli spaces induces an étale groupoid structure
X2 ⇒ X1 . To check that this is an étale equivalence relation, it suffices to check that
X2 → X1 × X1 is injective on geometric points, but this follows from the observation
the |X2 | → |X1 | × |X1 | is injective on closed points. Therefore there is an algebraic
space quotient X0 := X1 /X2 and a map X1 → X0 . By étale descent along X1 → X0 ,
there is a map π0 : X0 → X0 making the right square in (4.4.9) commute.
To argue that π : X0 → X0 is a coarse moduli space, we will use the commutative
cube
X2 / X1
} |
X1 / X0
 
X2 / X1
 }  |
X1 / X0 ,

where the top, left, and bottom faces are cartesian. Since X1 → X0 is étale and
surjective, and the natural map X1 → X0 ×X0 X1 pulls back to an isomorphism,
X1 ∼= X0 ×X0 X1 and the right square is also cartesian. Since being a coarse moduli
space is étale local on X0 (Lemma 4.2.10), we conclude that π0 : X0 → X0 is a
coarse moduli space. The condition that OX0 → π0,∗ OX0 and each of the properties
(1)–(3) are also étale local and since they hold for [Spec A1 /Gx ] → Spec AG 1
x
by
Theorem 4.2.3, they also hold for X0 → X0 .
The case when S is a noetherian algebraic space can be reduced to the affine
case by imitating the above argument to étale locally construct the coarse moduli
space of X . Finally, since X0 → X0 is proper, the separatedness of X0 is equivalent
to the separatedness of X0 .
Exercise 4.4.10. If X is a Deligne–Mumford stack of finite type over a noetherian
algebraic space S such that IX → X is finite. Then there exists a coarse moduli
space π : X → X with OX = π∗ OX satisfying properties (1)–(3) of Theorem 4.4.6.
Remark 4.4.11. The more general case when X is an algebraic stack with finite
inertia IX → X (but not necessarily Deligne–Mumford) is proven in an analogous
but more technical manner:
Step 1: Construct a quasi-finite and flat surjection U → X from a scheme. This
is proven as in the Existence of Minimal Presentations (3.6.1); see Remark 3.6.2,
[SGA3I , Lem. V.7.2], [KM97, Lem. 3.3], [Con05b, Lem. 2.1], and [SP, Tag 06MC].

186
Step 2: Show that every point has an étale neighborhood W → X where W admits
a finite flat presentation V → W from an affine scheme. This should be viewed
as a replacement to the Local Structure Theorem for Deligne–Mumford Stacks
(4.3.1), where instead of W being the quotient [Spec A/G] by a finite group, W is
the quotient [V /R] of a finite flat groupoid R := V ×W V ⇒ V of affine schemes.
The proof is also similar: instead of using ETd := [(U ×X · · · ×X U ∖ ∆)/Sd ] to
parameterize d unordered points in fibres of U → X , one uses a relative Hilbert stack
Hilbd over X , whose objects over S → X consist of a closed subscheme Z ,→ U ×X S
finite and flat (rather than finite and étale) of degree d over S. See [MF82, §5.A],
[KM97, Prop. 4.2], [Con05b, Lem. 2.2], and [SP, Tag 0DU4].
Step 3: Prove the existence of groupoids for the quotient [R/V ] of a finite flat
groupoid R ⇒ V of affine schemes. This is proven analogously to Theorem 4.2.3;
see Exercise 4.2.16, [SGA3I , Thm. V.4.1], [KM97, Prop. 5.1], [Con05b, Thm. 3.1],
[Ols16, Thm. 6.2.2], and [SP, Tag 0DUR]
Corollary 4.4.12. If G is an affine fppf group schemes over a noetherian scheme S
acting properly on a separated algebraic space U , i.e., the action map G × U → U × U
is proper, such that all stabilizer groups are reduced, then there exists a geometric
quotient π : U → X to an algebraic space X separated and finite type over S such
that OX = (π∗ OU )G .
Proof. Because G is affine over S, σ : G×U → U ×U is affine. Hence, the properness
of σ is equivalent to the finiteness of U , and it implies that [U/G] has finite diagonal.
Since the stabilizer groups are reduced, we see that [U/G] is a separated Deligne–
Mumford stack so we may apply the Keel–Mori Theorem (4.4.6).
The construction of the coarse moduli space in the proof provides a local structure
theorem of coarse moduli spaces. (Alteratively, the following theorem also follows
directly from the Local Structure Theorem (4.3.1) and Exercise 4.4.14.)
Corollary 4.4.13 (Local Structure of Coarse Moduli Spaces). Let X be a Deligne–
Mumford stack of finite type and separated over a noetherian algebraic space S,
and let π : X → X be its coarse moduli space. For every closed point x ∈ |X | with
geometric stabilizer group Gx , there exists a cartesian diagram

[Spec A/Gx ] /X

□ π
 
Spec AGx /X

such that Spec AGx → X is an étale neighborhood of π(x) ∈ |X|.

Exercise 4.4.14. Establish the following generalization of Proposition 4.4.1: if

f : X → Y is a morphism of Deligne–Mumford stacks separated and of finite type
over a noetherian algebraic space S fitting in a commutative diagram

X
f
/Y
πX πY
 
X /Y

where πX : X → X and πY : Y → Y are coarse moduli spaces and x ∈ |X | is a closed

point such that

187
 (1) f is étale at x and
(2) the induced map Gx → Gf (x) of geometric stabilizer groups is bijective,
then there exists an open neighborhood U ⊆ X of πX (x) such that U → X → Y is
étale and πX (U ) ∼
= U ×Y Y.

4.4.3 Examples
The basic example of a coarse moduli space is the map [Spec A/G] → Spec AG
induced by an action of a finite group; see Example 4.2.4 for some elementary
examples.

Example 4.4.15 (Elliptic curves). Consider the moduli stack M1,1 ∼ = [(A2 ∖
V (∆))/Gm ] of elliptic curves from Exercise 3.1.19, where k is a field with char(k) ̸=
2, 3, A2 has coordinates a and b with the action given by t · (a, b) = (t4 a, t6 b), and
∆ = 4a3 + 27b2 is the discriminant. The rings of invariants k[a, b]G ∆ is k[a /∆],
m 3

where β := b /∆ is generated by α := a /∆ under the relation 4α + 27β = 1.

2 3

Therefore, the coarse moduli space M1,1 → A1 is given by (a, b) 7→ a3 /∆. An elliptic
curve Eλ expressed by its Legendre equation y 2 z = x(x − z)(x − λz) in P2 for λ ∈ k
is projectively equivalent to y 2 z = x3 +aλ xz 2 +bλ z 3 for some aλ , bλ ∈ k, and one can
2
−λ+1)3
check that a3λ /∆λ is a scalar multiple of the j-invariant 28 (λλ2 (λ−1) 3 . Alternatively,
∼
using the quotient presentation M1,1 = [V (∆ − 1)/µ12 ] from Exercise 3.4.21, the
coarse moduli space is again given by (a, b) 7→ a3 /∆.

Exercise 4.4.16. (hard) Let char(k) ̸= 2 and G = Z/2.

(a) Let G act on the non-separated union X = A1 x̸=0 A1 by exchanging the
S

copies of A1 . The quotient [X/G] is a Deligne–Mumford stack with quasi-finite

but not finite inertia, and in particular non-separated. Show nevertheless that
there is a coarse moduli space [X/G] → A1 .
(b) Let X be the non-separated union A2 x̸=0 A2 . Let G = Z/2 act on X by
S

simultaneously exchanging the copies of A2 and by acting via the involution

y 7→ −y on each copy. Show that [X/G] does not admit a coarse moduli space.

Example 4.4.17. Consider the action of PGL2 on the scheme Sym4 P1 = ∼ (P1 )4 /S4
(which is the coarse moduli space of [(P ) /S4 ]) parameterizing four unordered
1 4

points in P1 . Let X ⊆ [Sym4 P1 / PGL2 ] be the open substack parameterizing tuples

(p1 , p2 , p3 , p4 ) where at least three points are distinct. Consider the family (0, 1, λ, ∞)
with λ ∈ P1 . If λ ∈ / {0, 1∞}, then we claim that Aut(0, 1, λ, ∞) = Z/2 × Z/2. To
see this, there is a unique element σ ∈ PGL2 such that σ(0) = ∞, σ(∞) = 0 and
σ(1) = λ which acts on P1 via σ([x, y]) = [y, λ, x] and thus σ(λ) = 1. Similarly, there
is an element interchanging 0 with 1 and λ with ∞ and an element interchanging
0 with λ and 1 with ∞. However, if λ ∈ {0, 1∞}, then Aut(0, 1, λ, ∞) = Z/2. We
therefore see that the inertia IX → X while quasi-finite is not finite and that X is
not separated. Nevertheless, the map X → P1 taking (p1 , p2 , p3 , p4 ) to its cross-ratio
is a coarse moduli space.

Exercise 4.4.18. Let π : X → X be the coarse moduli space of a separated and

finite type Deligne–Mumford stack X over a noetherian affine scheme.

Exercise 4.4.19. Let X be a Deligne–Mumford stack of finite type and separated

over a noetherian algebraic space S, and let π : X → X be its coarse moduli space.

188
 (a) If x ∈ |X | is a finite type point, show that the residual gerbe Gx of x is identified
with the reduction of the fiber Xπ(x) , and provide an example where the fiber
Xπ(x) is non-reduced.
(b) Show that if X is normal, then so is X.
(c) If X is regular, show that X → X is flat if and only if X is regular.
(d) Provide an example of a coarse moduli space X → X that is not flat.

4.4.4 Tame coarse moduli spaces

Definition 4.4.20. A Deligne–Mumford stack X is tame if for every geometric
point x ∈ X (k), the order of AutX (k) (x) is invertible in Γ(X , OX ). If in addition X
admits a coarse moduli space X → X, then we say that X → X is a tame coarse
moduli space.

Remark 4.4.21. If X is defined over a field k, then this means that the order of every
geometric stabilizer group is prime to the characteristic of k.

Lemma 4.4.22. Let X be a Deligne-Mumford stack separated and of finite type

over a noetherian algebraic space S. If π : X → X is a tame coarse moduli space,
then π∗ is exact.

Proof. The question is étale-local on X: if g : X ′ → X is an étale cover inducing a

cartesian diagram
g′
X′ /X

π′ □ π
 
X′
g
/ X,

then by Flat Base Change (4.1.45), there is an identification g ∗ π∗ = π∗′ g ′∗ of functors

on quasi-coherent sheaves. Since g ∗ is faithfully exact, we see that π∗ is exact if and
only π∗′ is. We can therefore use the Local Structure Theorem (4.4.13) to reduce to
the case that X = [Spec A/G] and X = Spec AG , where the exactness of π∗ follows
from its interpretation in terms of taking invariants (Exercise 4.1.27) and the linear
reductivity of G, i.e., Maschke’s Theorem (B.1.38).

Corollary 4.4.23. Let X be a Deligne-Mumford stack separated and of finite type

over a noetherian algebraic space S, and π : X → X its coarse moduli space. If F is
a quasi-coherent sheaf on X , then Hi (X , F ) = Hi (X, F ) and is 0 if i > dim X .

Proof. The first statement follows from the exactness of π∗ and the second from
Exercise 4.1.50.

Exercise 4.4.24 (Base change). Let X be a Deligne–Mumford stack of finite

type and separated over a noetherian algebraic space S. If π : X → X is a tame
coarse moduli space and X ′ → X is a morphism of algebraic spaces, show that
X ×X X ′ → X ′ is also a coarse moduli space.

On the other hand, Exercise 4.2.8(c) provides an example of a coarse moduli

space X → X and a map X ′ → X such that X ×X X ′ → X ′ is not a coarse moduli
space. In particular, it is does not automatically follow that M g ×Z Fp is the coarse
moduli space of Mg ×Z Fp ; see Question 5.5.25.

189
Remark 4.4.25 (Canonical stacks and a bottom-up classification of coarse moduli
spaces). The coarse moduli space X of a smooth separated tame Deligne–Mumford
stack over a field k. X has tame quotient singularities, i.e., every point of X
has an étale neighborhood which is a geometric quotient V /G of a finite group
of order relatively prime to char(k) acting on a smooth scheme V . This follows
from the Local Structure of Coarse Moduli Spaces (4.4.13). Conversely, Vistoli
showed for every algebraic space X with tame quotient stacks, there is a canonical
smooth separated tame Deligne–Mumford stack X can with coarse moduli space
X [Vis89, Prop. 2.8]. There is also a relative canonical stack X can → X for any
separated Deligne–Mumford with tame quotient singularities, and it is a theorem
of Geraschenko and Satriano that every smooth separated tame Deligne–Mumford
stack X with generically trivial stabilizer can be recovered from its coarse moduli
space X as the composition
√ √
X ∼
= (X can ( D))can → X can ( D) → X can → X,
√
where X can ( D) → X can is a composition of root stacks (Example 3.9.22) along
the irreducible components of the ramification divisor of X → X where each order
is the ramification degree [GS15].

4.4.5 Descending vector bundles to the coarse moduli space

We begin with a Nakayama lemma for coherent sheaves.

Lemma 4.4.26. Let X be a Deligne-Mumford stack separated and of finite type

over a noetherian algebraic space S, and let π : X → X be its coarse moduli space.
Let x ∈ |X | be a closed point and Gx be its residual gerbe.
(1) If F is a coherent sheaf on X such that F |Gx = 0, then there exists an open
neighborhood U ⊆ X of π(x) such that F |π−1 (U ) = 0.
(2) If ϕ : F → G is a morphism of coherent sheaves (resp., vector bundles of
the same rank) on X such that ϕ|Gx is surjective, then there exists an open
neighborhood U ⊆ X of π(x) such that ϕ|π−1 (U ) is surjective (resp., an isomor-
phism).

Proof. For (1), the support Supp(F ) ⊆ |X | of F is a closed subset (which follows
from using descent along a presentation) and the open set U = X ∖ π(Supp(F ))
satisfies the conclusion. For (2), apply (1) to the coherent sheaf coker(ϕ) noting that
a surjection of vector bundles of the same rank is an isomorphism.

We say that a vector bundle F on X descends to its coarse moduli space π : X → X

if there exists a vector bundle G on X and an isomorphism F ∼ = π ∗ G. Observe
that one necessary condition is that for every geometric point x : Spec k → X and
commutative diagram
  ix /
BGx X
p π
  
Spec k
j
/ X,

the pullback i∗x F = i∗x π ∗ G = p∗ j ∗ G is trivial or, in other words ,that Gx acts trivially
on the fiber F ⊗ k := i∗x F . This is equivalent to requiring that the restriciton F |Gx
to the residual gerbe of x is trivial.

190
Proposition 4.4.27. Let X be a tame Deligne-Mumford stack separated and of
finite type over a noetherian algebraic space S, and let π : X → X be its coarse
moduli space. A vector bundle F on X descends to a vector bundle on X if and only
if for every geometric point x : Spec k → X with closed image, the action of Gx on
the fiber F ⊗ k is trivial. In this case, π∗ F is a vector bundle and the adjunction
map π ∗ π∗ F → F is an isomorphism.

Proof. To see that the condition is sufficient, consider the commutative diagram

Gx  /X

p π
 
Spec κ(x)  /X

where Gx is the residual gerbe. We break down the proof into three steps.
Step 1: π ∗ π∗ F → F is surjective. It suffices by Lemma 4.4.26 to show that
(π ∗ π∗ F )|Gx → F |Gx is surjective for every closed point x ∈ |X |. Since F → F |Gx
is surjective and π∗ is exact (Lemma 4.4.22), π∗ F → π∗ (F |Gx ) is surjective and
thus (π ∗ π∗ F )|Gx → π ∗ (π∗ (F |Gx ))|Gx ∼
= p∗ p∗ (F |Gx ) is also surjective. The hypotheses
imply that the adjunction p∗ p∗ (F |Gx ) → F |Gx is an isomorphism, and it follows that
(π ∗ π∗ F )|Gx → p∗ p∗ (F |Gx ) ∼
= F |Gx is surjective.
Step 2: π∗ F is a vector bundle. We can assume that the rank r of F is constant. Since
being a vector bundle is an étale-local property, we can assume L
that X = Spec A.
The surjection s∈Γ(X,π∗ F ) A → π∗ F pulls back to a surjection s∈Γ(X ,F ) OX →
L

π ∗ π∗ F and by Step 1, the composition s∈Γ(X ,F ) OX → π ∗ π∗ F → F is surjective.

L

As F |Gx ∼ = OGr x is trivial, for each closed point x ∈ |X |, we can find r sections
⊕r
ϕ : OX → F such that ϕ|Gx is an isomorphism. By Lemma 4.4.26, there exists an
open neighborhood U ⊆ X of π(x) such that ϕ|π−1 (U ) is an isomorphism. Thus
⊕r
π∗ ϕ : O X → π∗ F is an isomorphism over U and we conclude that π∗ F is a vector
bundle of the same rank as F .
Step 3: π ∗ π∗ F → F is an isomorphism. Since π ∗ π∗ F → F is a surjection of vector
bundles of the same rank, it is an isomorphism.

Remark 4.4.28. The analogous statement for coherent sheaves is not true. For
example, if the characteristic is not 2, then letting Z/2 on A1 via x 7→ −x, we
have a tame coarse moduli space [A1 /(Z/2)] → A1 = Spec k[x2 ]. The inclusion
B(Z/2) ,→ [A1 /(Z/2)] of 0 is a closed substack and OB(Z/2) is a coherent sheaf which
does not descend.
When X is not tame, we have the following variant for descending line bundles.

Exercise 4.4.29. Let X be a Deligne-Mumford stack separated and of finite type

over a noetherian algebraic space S, and let π : X → X be its coarse moduli space.
Let L be a line bundle on X .
(1) Show that L⊗N descends to X for N sufficiently divisible.
Hint: Prove that for N sufficiently divisible π∗ (L⊗N ) is a line bundle and the
adjunction map π ∗ π∗ (L⊗N ) → L⊗N is an isomorphism. Reduce to the case
that X = [U/G] such that L pulls back to the trivial line bundle on U , and use
2
the spectral sequence Epq = Hp (BG, Hq (U, Gm )) ⇒ Hp+q ([U/G], Gm ) induced
by the composition [U/G] → BG → Spec Z.

191
 (2) Suppose that X is proper and representable over a Deligne–Mumford stack
Y of finite type over S. If L is relatively ample over Y, show that for N
sufficiently divisible L⊗N descends to a line bundle relatively ample over the
coarse moduli space Y of Y.
(Ampleness is an étale local property on the base, and we say that a line bundle
L on X is relatively ample over Y if X → Y is representable by schemes and
there exists an étale presentation V → Y such that the restriction LV to XV
is relatively ample.)

4.5 Algebraic spaces versus schemes

But who can quantify
the algebra of space,
or weigh those worlds that swim
each in its place?
Who can outdo the dark?
And what computer knows
how beauty comes to birth -
shell star and rose?
Jean Kenward, Technicians

We prove various results providing conditions for an algebraic space to be a

scheme:

• a quasi-separated algebraic space is a scheme on a dense open subspace (Theo-

rem 4.5.1);
• Zariski’s Main Theorem for algebraic spaces: a representable, quasi-finite, and
separated morphism factors as a composition of an open immersion and a finite
morphism (Theorem 4.5.9). This has some nice consequences:

– an algebraic space locally quasi-finite and separated over a scheme is a scheme

(Corollary 4.5.7),
– if the diagonal of a Deligne–Mumford stack is separated and quasi-compact,
then the diagonal is quasi-affine (Corollary 4.5.8) and in particular repre-
sentable by schemes, and
– an algebraic stack with trivial stabilizers is an algebraic space (Theorem 4.5.10)
generalizing Theorem 3.6.6;
• Serre’s and Chevalley’s Criteria for Affineness (Theorems 4.5.16 and 4.5.20) for
algebraic spaces;
• if X is a quasi-separated algebraic space locally of finite type over a field k
such that every finite set of points in Xk is contained in an affine (e.g., Xk is
quasi-projective), then X is a scheme (Proposition 4.5.27);
• a quasi-separated group algebraic space locally of finite type over a field is a
scheme (Theorem 4.5.28); and
• separated one dimensional algebraic spaces are schemes (Theorem 4.5.32)

4.5.1 Algebraic spaces are schemes over a dense open

Theorem 4.5.1. Every quasi-separated algebraic space has a dense open subspace
which is a scheme.

192
Proof. We may assume that X is a quasi-compact and quasi-separated algebraic
space. Let f : V → X be an étale presentation with V an affine scheme. Since X
is quasi-separated, f : V → X is quasi-compact and there exists an open algebraic
subspace U ⊆ X such that f −1 (U ) → U is finite. By Exercise 4.3.2, U is isomorphic
to a quotient stack [V /G] for the free action of a finite abstract group G on a
scheme V . If V1 ⊆ V is a dense affine open subscheme, then V2 = g∈G gV1 is a
T
G-invariant quasi-affine open subscheme of V and in particular separated. Repeating
this argument, we can choose a dense affine open subscheme V3 ⊆ V2 and now
V4 = g∈G gV3 is a G-invariant affine open subscheme. Proposition 4.3.3 implies
T

that V4 /G ∼= Spec AG is a dense affine open algebraic subspace of U . See also

[Knu71, II.6.7] and [SP, Tag 06NN].
Remark 4.5.2. The above result is not necessarily true if X is not quasi-separated,
e.g., A1 /Z (Example 3.9.35).
Corollary 4.5.3. An integral quasi-separated algebraic space has a well-defined
fraction field.

Exercise 4.5.4. Let G be a finite abstract group acting on a quasi-compact and

quasi-separated algebraic space U . Show that there is a dense G-invariant affine
open subscheme of U .

4.5.2 Zariski’s Main Theorem

We prove Zariski’s Main Theorem (4.5.9) for Deligne–Mumford stacks, which has the
important application that locally quasi-finite and separated morphisms of algebraic
spaces are representable by schemes (Corollary 4.5.7). Its proof relies on the theory
of quasi-coherent sheaves, and, specifically, on the factorization

f : X → SpecY f∗ OX → Y

of a quasi-compact and quasi-separated morphism See Definition 3.3.34 for the

definition of a quasi-finite morphism of algebraic spaces and §A.7 for a discussion of
Zariski’s Main Theorem for schemes.
Proposition 4.5.5. A representable, quasi-finite, and separated morphism f : X →
Y of Deligne–Mumford stacks factors as the composition of an open immersion
X ,→ SpecY f∗ OX and an affine morphism SpecY f∗ OX → Y. In particular, f is
quasi-affine.
Proof. Since the construction of f∗ OX commutes with flat base change on Y (Exer-
cise 4.1.45), so does the formation of the factorization f : X → SpecY f∗ OX → Y.
The statement is thus étale-local on Y. In particular, we can assume that Y = Y
is an affine scheme and that X = X is an algebraic space. After replacing Y with
SpecY f∗ OX , we can assume that f∗ OX = OY and we must show that f : X → Y
is an open immersion.
Since X is quasi-compact, there is an étale presentation π : U → X from an
affine scheme. Since X is separated, U → X is also separated. As the composition
π f
U−
→X−
→Y

is a quasi-finite morphism of schemes, we can apply Étale Localization of Quasi-

Finite Morphisms (A.7.1) around every point y ∈ Y : after replacing Y with an étale

193
neighborhood, we can assume that U = U1 ⨿ U2 with U1 → Y finite and (U2 )y = ∅.
Then π(U1 ) is open (as π is étale) and closed (as U1 → Y is finite and X → Y is
separated). Thus X = X1 ⨿ X2 with X1 = π(U1 ) and (X2 )y = ∅. This shows that
OY = f∗ OX is the product A1 × A2 of quasi-coherent OX -algebras, and thus we
can also decompose Y as Y1 ⨿ Y2 such that y ∈ Y1 and f (Yi ) ⊆ Xi for i = 1, 2. After
replacing Y with Y1 , the composition U → X → Y is finite and Lemma 4.5.6 implies
that X is affine. Thus X = Y = SpecY f∗ OX .
Lemma 4.5.6. Suppose that U → X is a surjective étale morphism of algebraic
spaces and that X → Y is a separated morphism of algebraic spaces. If the composi-
tion U → X → Y is finite, so is X → Y .
Proof. The statement is étale-local on Y so we can assume that Y and U are affine.
As X → Y is separated, U → X is also finite. Since X is identified with the
quotient U/R of the finite étale equivalence relation R := U ×X U ⇒ U of affine
schemes, Proposition 4.3.3 implies that X is affine. As U → Y is proper, so is
X → Y . As X → Y is a proper and quasi-finite morphism of schemes, it is finite
(Corollary A.7.5).
Corollary 4.5.7. A morphism of algebraic spaces which is locally quasi-finite and
separated is representable by schemes. In particular, an algebraic space locally
quasi-finite and separated over a scheme is a scheme.
Proof. It suffices to show that if X → Y = Spec A is a locally quasi-finite and
separated, then X is a scheme. Since being a scheme is a Zariski-local property, we
can assume that X is quasi-compact. In this case, Proposition 4.5.5 applies.
Corollary 4.5.8. The diagonal of a Deligne–Mumford stack with separated and
quasi-compact diagonal is quasi-affine. In particular, a quasi-separated algebraic
space has quasi-affine diagonal.
Proof. Since the diagonal is representable, quasi-finite, and separated, Proposi-
tion 4.5.5 applies. Note that the diagonal of an algebraic space is a monomorphism,
hence separated.
As in the case for schemes, we can refine Proposition 4.5.5 to obtain a version of
Zariski’s Main Theorem.
Theorem 4.5.9 (Zariski’s Main Theorem). A representable, quasi-finite, and sep-
arated morphism f : X → Y of noetherian Deligne–Mumford stacks factors as the
composition of a dense open immersion X ,→ Ye and a finite morphism Ye → X .
Proof. Let A ⊆ f∗ OX be the integral closure of OY → f∗ OX : for an étale map
T → Y from a scheme, A(T → Y) is the integral closure of Γ(T, OT ) → Γ(XT , OXT ).
Since the integral closure is compatible with étale extensions (Proposition A.7.4), A
is a quasi-coherent sheaf of OY -algebras. Using Exercise 4.1.33, write A = colim Aλ
as the colimit of finite type OY -algebras. As Y is quasi-compact, there exists an
étale presentation p : U → Y from an affine scheme. Then the base change XU → U
is a quasi-finite and separated morphism of algebraic spaces, thus a morphism of
schemes by Corollary 4.5.7. Since p∗ A = colim p∗ Aλ is identified with the integral
closure of OU → fU,∗ OXU , it follows from Zariski’s Main Theorem (A.7) for schemes
that for λ ≫ 0 XU → SpecU p∗ Aλ is an open immersion and SpecU p∗ Aλ → U
is finite. By étale descent, for λ ≫ 0, X → SpecY Aλ is also an open immersion
and SpecY Aλ → Y is finite. See also [Knu71, II.6.15], [LMB00, Thm. A.2], [Ols16,
Thm. 7.2.10], and [SP, Tag 05W7].

194
4.5.3 Characterization of algebraic spaces
We now can remove the hypotheses in Theorem 3.6.6 and Corollary 3.6.9 that the
diagonal be representable by schemes.

Theorem 4.5.10 (Characterization of Algebraic Spaces II). For an algebraic stack

X , the following are equivalent:
(1) the stack X is an algebraic space,
(2) the diagonal X → X × X is a monomorphism, and
(3) every point of X has a trivial stabilizer.

Proof. We only need to show (2) ⇒ (1). As the diagonal of X is a monomorphism,

it is locally quasi-finite and separated. Corollary 4.5.7 implies that the diagonal X
is representable by schemes and thus Theorem 3.6.6 applies.

Corollary 4.5.11 (Characterization of Representable Morphisms II). A morphism

X → Y of algebraic stacks is representable if and only if for every geometric point
x ∈ X (k), the map Gx → Gf (x) on automorphism groups is injective.

If X is a sheaf admitting a surjective étale morphism representable by schemes

U → X from a scheme, then X is an algebraic space (by definition). It turns out
that X is also an algebraic sapce if we only require that U is an algebraic space
and that U → X be smooth and representable (by algebraic spaces). Moreover,
we can extend the existence of quotients of étale equivalence relations of schemes
(Theorem 3.4.13(2)) to étale equivalence relations of algebraic spaces.

Corollary 4.5.12.
(1) If X is a sheaf on Schét such that there exists a surjective, étale (resp., smooth),
and representable morphism U → X from an algebraic space, then X is an
algebraic space.
(2) If R ⇒ U is an étale (resp., smooth) equivalence relation of algebraic spaces,
then the quotient U/R is an algebraic space.

Remark 4.5.13. The above statement holds with with ‘étale’ replaced with ‘fppf’;
see Theorem 6.3.1 and Corollary 6.3.6.

Proof. By Theorem 3.4.13(1), the sheaf X in (1) and the quotient sheaf U/R in (2)
are algebraic stacks, and the statement follows from Theorem 4.5.10.

Corollary 4.5.14. A proper and quasi-finite morphism (resp., proper monomor-

phism) of algebraic spaces is finite (resp., a closed immersion).

Proof. Proper and quasi-finite morphisms are representable by schemes. Thus the
statement follows from the corresponding result for schemes (Corollary A.7.5) and
étale descent.

Exercise 4.5.15. Show that the prestack AlgSp over Schét , whose objects over a
scheme T are algebraic spaces over T and whose morphisms correspond to cartesian
diagrams of algebraic spaces, is a stack.

195
4.5.4 Affineness criteria
Theorem 4.5.16 (Serre’s Criterion for Affineness). Let X be a quasi-compact and
quasi-separated (resp., noetherian) algebraic space. If the functor Γ(X, −) is exact on
the category of quasi-coherent (resp., coherent) sheaves, then X is an affine scheme.
Proof. We will show that the canonical morphism π : X → Y := Spec Γ(X, OX ) is a
proper monomorphism. This will imply that X is affine as a proper monomorphism
is a closed immersion (Corollary 4.5.14). The proof is inspired by properties of good
moduli spaces (see §6.5).
Claim: If g : Y ′ → Y is a morphism of algebraic spaces, then the base change
π ′ : X ′ := X ×Y Y ′ → Y ′ has the following properties:
(a) π∗′ induces an equivalence of the categories of quasi-coherent sheaves on X ′
and Y ′ .
(b) OY ′ → π∗′ OX ′ is an isomorphism.
(c) X ′ → Y ′ is a homeomorphism.
By Flat Base Change (Exercise 4.1.45), properties (a) and (b) are étale local on
Y ′ so we may assume Y ′ = Spec B. We will show that the adjunction morphisms
G → π∗′ π ′∗ G and π ′∗ π∗′ F → F are isomorphisms for G ∈ QCoh(Y ′ ) and F ∈
QCoh(X ′ ). This immediately gives (a) and also (b) (by taking G = OY ′ ). For the
first adjunction, choose a free presentation OY⊕J → OY⊕I → g∗ G → 0 of G as an
OY -module. As π∗ is exact, we have a morphism of right exact sequences

OY⊕J / O⊕I / g∗ G /0
Y

  
π∗ π ∗ (OY⊕J ) / π∗ π ∗ (O⊕I ) / π∗ π ∗ g∗ G /0
Y

The two left vertical arrows are isomorphisms since π∗ OX = OY . Therefore g∗ G →

π∗ π ∗ g∗ G ∼
= g∗ π∗′ π ′∗ G is an isomorphism. Since g∗ is faithfully exact, G → π∗′ π ′∗ G
is also an isomorphism. To see that π ′∗ π∗′ F → F is an isomorphism, let K and
Q be the kernel and cokernel of π ′∗ π∗′ F → F . As π∗ is exact and g∗ is faithfully
exact, it follows that π∗′ is exact. Since π∗′ π ′∗ π∗′ F → π∗′ F is an isomorphism (by
applying the first adjunction to π∗′ F ), we see that π∗′ K = π∗′ Q = 0. It thus suffices
to show that for a quasi-coherent sheaf F ′ on X ′ , then F ′ ̸= 0 implies π∗′ F ′ ̸= 0.
If x : Spec k → X ′ is a geometric point such that F ⊗ k := x∗ F ̸= 0, then by base
changing by the composition π ′ ◦ x : Spec k → Y ′ , we may assume that Y ′ = Spec k
and that x : Spec k → X ′ is a section of π ′ . Since every k-point of an algebraic
space defined over k is a closed point, x : Spec k → X ′ is a closed immersion, and
hence F → x∗ x∗ F = F ⊗ k is surjective. It follows from the exactness of π∗′ that
π∗′ F → π∗′ x∗ x∗ F = F ⊗ k is surjective and hence π∗′ F ̸= 0.
To see (c), if y : Spec k → Y is a geometric point, then by (b) Γ(Xy , OXy ) = k
as thus the fiber Xy is non-empty. On the other hand, if x, x′ ∈ Xy (k) were distinct
points each necessarily closed, then OXy → O{x,x′ } is surjective. Since π∗ is exact,
we also get a surjection k = Γ(Xy , OXy ) → k ⊕ k, a contradiction. To see that π ′ is
closed, let Z ⊆ X ′ be a closed subspace and q : Z → im(Z) denote the morphism
to its scheme-theoretic image. Then OZ → q∗ Oim(Z) is an isomorphism and q∗ is
exact. Applying the surjectivity result above to q, we see that q is surjective, and
hence π ′ (Z) is closed.
The claim implies that X → Y is universally closed. To see that X → Y is a
monomorphism, i.e., the diagonal ∆ : X → X ×Y X is an isomorphism, observe

196
that (b) implies that the pushforward of OX×Y X → ∆∗ X along the first projection
p1 : X ×Y X → X is an isomorphism. Thus, (a) applied to p1 shows that OX×Y X →
∆∗ X is an isomorphism. Zariski’s Main Theorem (A.7.3) implies that ∆ is an
open immersion. Applying (c) to p1 shows that p1 : |X ×Y X| → |X| is bijective.
Hence, ∆ is an isomorphism. It remains to show that X → Y is of finite type. Let
p : U = Spec A → X be an étale presentation. Since X is separated, R := U ×X U is
a closed subscheme of U ×Y U = Spec(A ⊗Γ(X,OX ) A). Hence R = Spec B is affine.
Letting s and t denote the two maps A ⇒ B, we have a commutative diagram

?A
s

Γ(X, OX ) A ⊗Γ(X,OX ) A / */5 B.
?
 t
A

Since p : U → X is étale, t : A → B is of finite type and Pwe can choose A-algebra

generators b1 , . . . , bn ∈ B. For each i, choose a preimage j aij ⊗a′ij ∈ A⊗Γ(X,OX ) A
of bi . Note that s(aij ) also generate B as an A-algebra via t. The elements aij ∈
Γ(X, p∗ OU ) = A define a homomorphism OX [zij ] → p∗ OU of OX -algebras taking zij
to aij . Its pullback via p is identified with OU [zij ] → p∗ p∗ OU ∼
= t∗ OR , where the last
equivalence comes from Flat Base Change (Exercise 4.1.45), and this map is surjective
precisely because s(aij ) generate B over t. By étale descent, OX [zij ] → p∗ OU is
surjective and therefore so is Γ(X, OX )[zij ] → A. Thus, Γ(X, OX ) → A is of finite
type and, by étale descent, so is X → Y .
Finally, in the noetherian case, every quasi-coherent sheaf is a colimit of coherent
sheaves (Exercise 4.1.28) and Γ(X, −) commutes with colimits. Assume that Γ(X, −)
is exact on coherent sheaves. Given a surjection p : F ↠ G of quasi-coherent sheaves
on X, write G = colimi Gi as a colimit of coherent sheaves and choose coherent
subsheaves Fi ⊆ p−1 (Gi ) surjecting onto Gi . Then Γ(X, Fi ) ↠ Γ(X, Gi ) and the
composition colimi Γ(X, Fi ) → Γ(X, F ) → Γ(X, G) = colimi Γ(X, Gi ) is surjective.
Thus Γ(X, F ) → Γ(X, G) is surjective and we conclude that Γ(X, −) is exact on
quasi-coherent sheaves. See also [Knu71, Thm. III.2.5], [Ryd15, Thm. 8.7], and [SP,
Tag 07V6].

Corollary 4.5.17. Let X be a quasi-compact and quasi-separated (resp., noetherian)

algebraic space. Then X is an affine scheme if and only if Hi (X, F ) = 0 for every
quasi-coherent (resp., coherent) sheaf F and i > 0.

Proof. If X is affine, then Theorem 4.1.42 establishes the vanishing of quasi-coherent

cohomology. Conversely, the vanishing of quasi-coherent (resp., coherently) cohomol-
ogy implies that Γ(X, −) is exact on the category of quasi-coherent (resp., coherent)
sheaves: if 0 → F1 → F2 → F3 → 0 is exact, then Γ(X, F2 ) → Γ(X, F3 ) is surjective
as H1 (X, F1 ) = 0.

Remark 4.5.18. Given a quasi-compact and quasi-separated morphism f : X → Y of

Deligne–Mumford stacks, the relative versions of Theorem 4.5.16 and Corollary 4.5.17
also hold: f is affine if and only if f∗ : QCoh(X ) → QCoh(Y) is exact if and only if
Ri f∗ F = 0 for all i > 0 and F ∈ QCoh(X ).

197
Proposition 4.5.19. Let X be a noetherian algebraic space. If Xred is a scheme
(resp., quasi-affine, affine), then so is X.
Proof. If Xred is affine, then one uses Corollary 4.5.17 to show that X is affine
exactly as in [Har77, Exc. III.3.1]: if F is a coherent sheaf on X and I ⊆ OX denotes
the nilpotent ideal defining Xred , then one shows the vanishing of Hi (X, F ) using
the filtration 0 = I N F ⊆ I N −1 F ⊆ · · · ⊂ IF ⊆ F , whose factors I k F/I k+1 F are
supported on Xred . If Xred is a scheme, then every point x ∈ |X| has an open
neighborhood U such that Ured is affine. Thus U is affine and X is a scheme. If
Xred is quasi-affine, then Xred → Spec Γ(X, OX )red is an open immersion. Thus
X → Spec Γ(X, OX ) is an open immersion and X is quasi-affine.
Theorem 4.5.20 (Chevalley’s Criterion for Affineness). Let X → Y be a finite
surjective morphism of noetherian algebraic spaces. If X is affine, then so is Y .
Proof. One can argue as in [Har77, Exc. 4.1] using Corollary 4.5.17.
Exercise 4.5.21. Prove the following generalization: if X → Y is an integral
surjective morphism of algebraic spaces with Y noetherian and X is affine, then Y
is affine.
Hint: Write X = limλ Xλ as a limit of finite Y -scheme and use Limit Methods
(B.3).
The cohomological criterion for ampleness [Har77, Prop. 5.3] also extends to
algebraic spaces. A line bundle L on an algebraic space X is defined to be ample if
X is a scheme and L is ample in the ordinary sense.
Exercise 4.5.22. Let X be a proper algebraic space over a noetherian ring. For a
line bundle L on X, show that L is ample if and only if for every coherent sheaf F
on X, there is an integer n0 such that Hi (X, F ⊗ Ln ) = 0 for i > 0 and n ≥ n0 . See
also [SP, Tag 0D2W].
As in the case of schemes (Proposition B.2.9), ampleness can be checked on a
finite cover.
Exercise 4.5.23. Let X be a proper algebraic space over a noetherian ring and L
be a line bundle on X. If f : X ′ → X is a finite surjective morphism, L is ample if
and only if f ∗ L is. See also [SP, Tag 0GFB].
Exercise 4.5.24. A noetherian scheme has the Chevalley–Kleiman property if every
finite set of points is contained in an affine. Show that if X → Y is a finite surjective
morphism of noetherian algebraic spaces such that X has the Chevalley-Kleiman
property, then Y also has the Chevalley-Kleiman property. See [Kol12, Cor. 48].

4.5.5 Effective descent along field extensions

Lemma 4.5.25. Let X be a quasi-separated algebraic space locally of finite type
over a field k. If Xk is an affine scheme, then so is X.
Proof. As Xk → X is an integral surjective morphism, the statement follows Cheval-
ley’s Criterion for Affineness (4.5.21).
Proposition 4.5.26. Let X be a quasi-separated algebraic space of finite type over a
field k. If Xk is a scheme, then there exists a finite separable field extension k → K
such that XK is a scheme.

198
Proof. Choose an étale presentation U → X from an affine scheme and set R =
U ×X U . Write k = colim kλ as the colimit of finite field extensions, and set
Xλ := Xkλ , Uλ = Ukλ and Rλ = Rkλ . Let V ⊆ Xk be an open affine subscheme.
We claim that for λ ≫ 0, there exists an open subscheme Vλ ⊆ Xλ such that
V = Uλ ×kλ k. Indeed, the preimage Z ⊆ Uk of V has the property that its two
preimages in Rk are equal. Using Proposition B.3.3 and Proposition B.3.7, for λ ≫ 0
there is an open subscheme Zλ ⊆ Uλ with Z = Zλ ×kλ k such that the two preimages
of Zλ in Rλ are equal. By étale descent, Zλ descends to the desired closed subscheme
Vλ ⊆ Xλ . Lemma 4.5.25 implies that Vλ is a scheme. By covering Xk with finitely
many affines and choosing λ sufficiently large, we obtain a finite field extension
K = kλ of k such that Xλ is a scheme. Now factor k → kλ as a separable field
extension k → K and a purely inseparable field extension K → kλ . Since Xλ → XK
is a finite univeral homemorphism, Chevalley’s Theorem for Affineness (4.5.20)
implies that the image of each Vλ in XK is affine, and hence XK is a scheme.

With an additional condition on Xk , we can conclude that X is a scheme.

Proposition 4.5.27. Let X be a quasi-separated algebraic space locally of finite type

over a field k. If Xk is a scheme such that every finite set of k-points is contained
in an affine (e.g., Xk is quasi-projective), then X is a scheme.

Proof. We may assume that X is quasi-compact. We will show that every closed
point x ∈ X has an affine open neighborhood. Let Spec κ(x) ,→ X be the inclusion
of the residue field of x (Proposition 3.5.16(3)) and consider the cartesian diagram
`n  /X / Spec k
i=1 Spec Ai k

   
Spec κ(x)  /X / Spec k
x

where n is the degree of the separable closure of k in K, and each Ai is a artinian

local k-algebra. The hypotheses on Xk ensure that there is an affine open subscheme
U ⊆ Xk containing the images of each Spec Ai . By Proposition 4.5.26, there is a finite
field extension k → K such that XK is a scheme. After enlarging K, we can arrange
that U descends to an affine open subscheme U ′ ⊆ UK by using Proposition B.3.3
to descend the morphism U → X, Proposition B.3.7 to arrange that it is an open
immersion, and Proposition B.3.5 to arrange affineness. Observe that U ′ contains all
preimages of x under XK → X. By taking the normal closure of K, we can assume
K is normal over k. Let G = Aut(K/k) so that K G is a purely inseparable field
extension of k. Then G acts on XK freely such that XK /G = XK G . The intersection
of the translates of U ′ by elements of G is a G-invariant quasi-affine variety U ′′ .
Choosing an affine open subscheme in U ′′ containing all of the preimages of x and
intersecting again the translates of G, we obtain a G-invariant affine V ⊆ XK
containing the preimages of x. Then the quotient V /G is an affine subscheme of
XK G containing the unique preimage of x (Theorem 4.2.3). Letting W be the image
of V /G under the finite universal homeomorphism XK G → X, Chevalley’s Criterion
for Affineness (Theorem 4.5.20) implies that W is an affine neighborhood of x.

4.5.6 Group algebraic spaces are schemes

Every quasi-separated group algebraic space over a field k is a scheme. When k is
algebraically closed, this follows easily from Theorem 4.5.1 as we know there is a

199
dense open that is a scheme, and we can translate this around by rational points.
The general case relies on Proposition 4.5.27.

Theorem 4.5.28. A quasi-separated group algebraic space G locally of finite type

over a field k is a separated scheme. The connected component of the identity G0 is
quasi-projective.

Remark 4.5.29. If G is not quasi-separated, then the above corollary does not hold,
e.g., G = Ga /Z over k (Example 3.9.35). Note that Proposition 4.5.19 implies
that the result also holds over an artinian base. Over a general base scheme, even
separated group algebraic spaces need not be schemes; see [Ray70, Lem. X.14].

Proof. It suffices to show that G is a scheme. Indeed, a group scheme locally of finite
type over a field is necessarily separated (B.1.16(1)) and the identity component G0
is quasi-compact (B.1.16(4)), thus quasi-projective (B.1.16(7)). Assume first that k
is algebraically closed. There is a non-empty open subscheme U of G (Theorem 4.5.1)
with a point h ∈ U (k). For every g ∈ G(k), left multiplication by gh−1 defines an
∼
isomorphism G → G and the image gh−1 U of U is a scheme containing g. The
general case follows from Proposition 4.5.27 using that Gk is a scheme with the
property that every finite set of points is contained in an affine (Lemma 4.5.30). See
also [Art69b, Lem. 4.2] and [SP, Tag 0B8D].

Lemma 4.5.30. Every group scheme G locally of finite type over an algebraically
closed field k has the property that every finite set of k-points is contained in an
affine open subscheme.

Proof. Let g1 , . . . , gn ∈ G(k). We first use induction on n to reduce to the case

that the elements gi are in the same connected component. If not, we can write
G = W1 ⨿ W2 with r points in W1 and n − r points in W2 for 0 < r < n. By
induction, there are affine opens U1 ⊆ W1 and U2 ⊆ W2 containing the r and n − r
points, respectively. Then U1 ⨿ U2 is an affine containing each gi . By translating by
g1−1 , we may further assume that g1 , . . . , gn ∈ G0 (k). Let U ⊆ G0 be an affine open
neighborhood of the identity. Since G0 is irreducible (B.1.16(4)), U g1−1 ∩ · · · ∩ U gn−1
is non-empty and contains a closed point h. Since h ∈ U gi−1 , each gi is contained in
the affine open h−1 U .

Corollary 4.5.31. Let X be an algebraic stack with quasi-separated diagonal. Then

the stabilizer of every field-valued point is a group scheme locally of finite type.

Proof. By Exercise 3.3.5, the diagonal of X is locally of finite type. As the stabilizer
is the base change of the diagonal, the statement follows from Theorem 4.5.28.

4.5.7 Separated one dimensional algebraic spaces are schemes

Theorem 4.5.32. A separated noetherian algebraic space X with dim X ≤ 1 is a
scheme.

Proof. We may assume that X has pure dimension 1. By Proposition 4.5.19, we may
assume that X is reduced. By Theorem 4.5.1, there exists an open dense subscheme
U ⊆ X which is regular. Let D ⊆ U be a union of reduced points such that every
irreducible component of X contains at least four points. Then OU (D) defines a line
bundle on U which extends to a line bundle OX (D). Using Le Lemme de Gabber
(4.6.1) proved in the next section, there exists a finite surjection f : Z → X from a
scheme Z. After replacing Z with its normalization, we may assume that Z is regular.

200
As Z is a separated scheme of dimension 1, Z is quasi-projective and f ∗ OX (D) is
ample. By Exercise 4.5.23, OX (D) is also ample. See also [SP, Tags 0ADD and
09NZ].

Exercise 4.5.33.
(a) (hard) Show that every smooth separated algebraic space of finite type over a
field of dimension two is a scheme.
(b) (moderate) Construct a non-normal separated algebraic space of finite type
over a field of dimension two that is not a scheme.
(c) (open) Is every separated regular algebraic space of dimension two a scheme?

4.6 Finite covers of Deligne–Mumford stacks

It is obvious if you think about it.

Angelo Vistoli

We prove Le Lemme de Gabber (4.6.1): a separated Deligne–Mumford stack

has a finite cover by a scheme. As a consequence, we obtain a Chow’s Lemma
for Algebraic Spaces (4.6.6) and Deligne–Mumford Stacks (4.6.5). We deduce
cohomological applications in §4.6.3.

4.6.1 Le Lemme de Gabber

A celebrated result attributed to Gabber asserts that a Deligne–Mumford stack has
a finite cover by scheme. Arguments were first published in [Del85] and [Vis89].

Theorem 4.6.1 (Le Lemme de Gabber). Let X be a Deligne-Mumford stack

separated and of finite type over a noetherian scheme S. Then there exists a finite,
generically étale, and surjective morphism Z → X from a scheme Z.

Proof. By replacing X with the disjoint union of the irreducible components with
their reduced stack structure, we may assume that X is irreducible and reduced.
Every étale presentation U → X is representable, quasi-finite, and separated, and
thus it factors as the composition of an open immersion U ,→ Xe and a finite
morphism Xe → X by Zariski’s Main Theorem (4.5.9). After replacing X with
Xe, we may assume that X has a dense open subscheme. If p : U → X is an
étale presentation, there is therefore a dense open subscheme V ⊆ X such that
p−1 (V ) → V is finite étale of degree d. We may choose a finite étale covering
V ′ → V such that p−1 (V ) ×V V ′ → V ′ is a trivial étale covering; indeed as in
Proposition A.3.11, we may take V ′ to be the complement of the (open and closed)
pairwise diagonals in
(V ′ /V )d = V ′ ×V · · · ×V V ′ .
| {z }
d

Applying Zariski’s Main Theorem (4.5.9) again, this time to the composition V ′ ↠
V ,→ X , gives a finite surjective morphism Xe → X restricting to V ′ → V . Thus
after replacing X with Xe, we may further assume that there is an étale presentation
U → X which over a dense open subscheme j : V ,→ X is a trivial étale covering,

201
i.e., there is a cartesian diagram


V
j′
`d /U
i=1

p
  
V / X.
j

We will construct a finite surjective morphism Z → X from a scheme that is an

isomorphism over V . Let A ⊆ j∗ OZ be integral closure of OX → j∗ OZ . Then p∗ A is
the integral closure of OU in j∗′ O⨿V = p∗ j∗ OV (Proposition A.7.4). The idempotent
ei ∈ Γ(U, j∗′ O⨿V ) = Γ(⨿V, O⨿V ), defining the ith copy of V , is integral over OU
and thus defines a global section ei ∈ Γ(U, p∗ A). Now write A = colimλ Aλ as a
filtered colimit of finite type OX -subalgebras (Exercise 4.1.33). Since A is integral
over OX , each Aλ is a finite OX -algebra. For λ ≫ 0, we have ei ∈ Γ(U, p∗ Aλ ). The
Deligne–Mumford stack Z := SpecX Aλ is finite over X and we claim that Z is a
scheme. To see this, consider the cartesian diagram

Z′ /U
p
 
Z / X cartesian,

where Z ′ is a scheme (as it is finite over U ). Each idempotent ei defines a global

`d
section of OZ ′ and thus yields a decomposition Z ′ = i=1 Zi′ . Each morphism
Zi′ → Z is étale, separated and birational, thus an open immersion. Since Z ′ → Z is
surjective, {Zi′ } defines an open covering of Z and Z is a scheme. See also [LMB00,
Thm. 16.6].
Exercise 4.6.2. Provide an alternative proof following [Vis89, Prop. 2.6]:
(i) Use limit methods to reduce to the case that S is of finite type over Z (which
ensures that normalizations are finite).
(ii) Reduce to the case that X is irreducible, reduced, and normal.
(iii) Let X → X be the coarse moduli space and let U → X be an étale presentation
by a scheme U = i Ui , where each Ui is integral and affine. Let L be a
`
normal finite field extension of K(X) containing each function field K(Ui ),
and let Y → X and Yi → Ui be the normalizations of X and Ui in L. Prove
that Y is a scheme and that Y → X Zariski locally factors through X by
showing that each Yi → Y is an open immersion and that Y is the union of
the translates σ · Yi over all i and σ ∈ Aut(L/K(X)).
(iv) By replacing X with Y and X with X ×X Y , this reduces to the case
S that the
coarse moduli space X is a scheme with an open covering X = Xα such
that each Xα ,→ X factors through X . Show that after replacing X with a
finite cover, the sections Xα → X glue to a global section X → X .
Remark 4.6.3. More generally, every separated algebraic stack of finite type over S
has a proper cover by a quasi-projective scheme [Ols05].
Exercise 4.6.4 (good practice). Let X be a normal algebraic space of finite type
over a noetherian scheme S. Show that there is a normal scheme U with an action
of a finite abstract group G such that X is the quotient of U by G, i.e., [U/G] → X
is a coarse moduli space.

202
Hint: After reducing to the case that X is integral, choose a finite, generically
étale and surjective morphism U → X from a scheme. Let L be the Galois closure
of K(U )/K(X), and take U to be the integral closure of X in L and take G =
Gal(L/K(X)). See also [LMB00, Cor. 16.6.2].

4.6.2 Chow’s Lemma

Corollary 4.6.5 (Chow’s Lemma for Deligne–Mumford Stacks). Let X be a Deligne-
Mumford stack separated and of finite type over a noetherian scheme S. Then there
exists a projective, generically étale, and surjective morphism Z → X from a scheme
Z quasi-projective over S.

Proof. Le Lemme de Gabber (4.6.1) reduces the statement to the case of schemes,
c.f., [Har77, Exc. II.4.10].

There is also a birational version for algebraic spaces.

Exercise 4.6.6 (Chow’s Lemma for Algebraic Spaces). Let X be an algebraic

space separated and of finite type over a noetherian scheme S. Then there exists
a projective, birational, and surjective morphism Z → X from a scheme Z quasi-
projective over S. See also [Knu71, IV.3.1] and [SP, Tag 088U].

Exercise 4.6.7 (Valuative criteria can be checked on dense opens).

(a) Let X be a Deligne–Mumford stack separated and of finite type over a noethe-
rian scheme S, and let U ⊆ X be a dense open substack. Show that X → S is
proper if and only if for every DVR R defined over S with fraction field K,
every map Spec K → U extends to a map Spec R → X over S, after replacing
R with an extension of DVRs.
Hint: Use Corollary 4.6.5 to reduce to the case that X → S is quasi-projective.
(b) Let X be a Deligne–Mumford stack with separated and quasi-compact diagonal.
Show that a finite type morphism X → S to a noetherian scheme is separated
if and only if for every DVR R over S with fraction field K and every pair
∼
h, g : Spec R → X of morphisms over S, any isomorphism hK → gK of their
∼
generic fibers extends to a unique isomorphism h → g.
Hint: Reduce to Part (a).

4.6.3 Applications to cohomology

Most of the foundational theorems for coherent sheaf cohomology on proper schemes—
Finiteness of Cohomology (A.5.3), Cohomology and Base Change (A.6.2), Semi-
continuity (A.6.4), Formal Functions (A.5.4), Grothendieck’s Existence Theorem
(C.5.3), and Stein Factorization [Har77, Cor. III.11.5]— also hold for proper Deligne–
Mumford stacks. They imply Zariski’s Connectedness Theorem (4.6.15), which is an
essential ingredient in the proof of the irreducibility of Mg in positive characteristic
(Theorem 5.7.33).

Theorem 4.6.8 (Finiteness of Cohomology). Let f : X → Y be a proper morphism

of noetherian Deligne–Mumford stacks. For any coherent sheaf F on X and any
i ≥ 0, Ri f∗ F is coherent.

Proof. By Flat Base Change (4.1.45), we can assume that Y = Spec A is an affine
scheme, and we need to show that Hi (X , F ) is a finite A-module for i ≥ 0. We will

203
use the following generalization of Dévissage (A.5.1), which is proved in the same
way: if P is a property of coherent sheaves on X such that
(a) in a short exact sequence in Coh(X ), if two out of the three satisfies P, then
the third does too, and
(b) for all integral closed substacks Z ⊆ X with ideal sheaf IZ ⊆ OX and a
coherent sheaf F on X with IZ F = 0, there exists a coherent sheaf G on X
satisfying P with IZ G = 0 and a morphism F → G which is an isomorphism
over an open substack of Z,
then every coherent sheaf on X satisfies P. Taking P to be the property that
Hi (X , F ) is a finite A-module for i ≥ 0, we see that (a) holds. For (b), using Le
Lemme de Gabber (4.6.1), let g0 : V0 → Z be a finite surjective morphism from a
scheme. Letting V1 = V0 ×Z V0 with projection g1 : V1 → Z, there is a complex
0 → F → g0,∗ g0∗ F → g1,∗ g1∗ F , which is exact if V0 → Z is flat (Proposition 2.1.1).
Defining G := ker(g0,∗ g0∗ F → g1,∗ g1∗ F ), there is a map F → G. By Generic Flatness
(3.3.32), V0 → Z is flat over a nonempty open substack U ⊆ Z, and it follows that
F → G is an isomorphism over U. Since V0 and V1 are proper over Spec A, g0,∗ g0∗ F
and g1,∗ g1∗ F have finite cohomology, and thus so does G. See also [Fal03, Thm. 1],
[Ols05, Thm. 1.2], [LMB00, Thm. 15.6], and [Ols16, Thm. 11.6.1].
Remark 4.6.9. Unlike schemes, Hi (X , F ) can be nonzero for arbitrary large i; see
Exercise 4.1.49.
Exercise 4.6.10. If X is a noetherian algebraic space, show that there exists an
integer N such that for every coherent sheaf F on X, Hi (X, F ) = 0 for i > N .
Proposition 4.6.11 (Semicontinuity). Let X → Y be a proper morphism of noethe-
rian Deligne–Mumford stacks, and let F be a coherent sheaf on X which is flat over
Y. For each i ≥ 0, the function

|Y| → Z, y 7→ dimK Hi (XK , FK ) (4.6.12)

where Spec K → Y is any field-valued point representing y, is upper semicontinuous.

Proof. We leave the reader to check that the function (A.6.5) is constructible.
It therefore suffices to check that it is upper semicontinuous with respect to a
specialization y ⇝ y0 in |Y|. By Proposition 3.8.20, this specialization is realized
by a morphism Spec R → Y from a DVR, and we may therefore assume that
Y = Spec R. Letting K and κ be the fraction and residual field of R, we need to
show that dimK Hi (XK , FK ) ≤ dimκ Hi (Xκ , Fκ ). If π ∈ R denotes a uniformizer,
then π : F → F is injective since F is flat over R. Apply Hi (X , −) to the short exact
sequence 0 → F → F → Fκ → 0, we obtain an exact sequence
π
· · · → Hi (X , F ) −
→ Hi (X , F ) → Hi (Xκ , Fκ ) → · · · .

This gives an injection Hi (X , F )/πHi (X , F ) ,→ Hi (Xκ , Fκ ). On the other hand,

Hi (X , F ) is a finite R-module by Finiteness of Cohomology (4.6.8) with Hi (X , F ) ⊗R
K = Hi (X , FK ) by Flat Base Change (4.1.45). This gives

dimK Hi (XK , FK ) ≤ dimκ Hi (X , F )/πHi (X , F ) ≤ dimκ Hi (Xκ , Fκ ).

Exercise 4.6.13 (hard). Let X be a noetherian Deligne–Mumford stack proper

over a complete noetherian local ring (A, m). Let Xn = X ×A A/mn+1 .

204
 (a) (Formal Functions) If F is a coherent sheaf on X , there is a natural isomorphism
∼
Hi (X , F) → lim Hi (Xn , F |Xn )
← −
n

for every i ≥ 0.
(b) (Grothendieck’s Existence Theorem) There is an equivalence of categories

Coh(X ) → lim Coh(Xn ), F 7→ {F |Xn }.

←−
See also [Ols05, Thm. 1.4] and [Con05a, Cor. 3.2, Thm. 4.1].

Theorem 4.6.14 (Stein Factorization/Zariski’s Connectedness Theorem). A proper

morphism f : X → Y of noetherian Deligne–Mumford stacks factors as
f′ g
f : X −→ Y ′ = SpecY f∗ OX −
→ Y,

where f ′ has geometrically connected fibers and g is finite. In particular, if f∗ OX =

OY , then f has geometrically connected fibers.

Proof. This is proved just as in the case of schemes—[Har77, Cor. III.11.5], [EGA,
III1 .4.3.1], and [SP, Tag 03H2]—using Formal Functions. By Finiteness of Cohomol-
ogy (4.6.8), f∗ OX is coherent and g is finite. We are thus reduced to showing that f
has geometrically connected fibers if f∗ OX = OY . Since the question is étale local,
we can further assume that Y = S is affine. Moreover, since geometric connectedness
of a fiber Xs can be checked on separable finite field extensions of κ(s), it suffices to
verify that for every étale morphism S ′ → S, the fibers of f ′ : XS ′ → S ′ are connected.
By Flat Base Change (4.1.45), f∗′ OXS′ = OS ′ , so it is enough to verify that f has con-
nected fibers. If a fiber Xs is disconnected, then setting Sn = Spec OS,s /mn+1 s , there
are compatible isomorphisms H0 (X ×S Sn , OX ×S Sn ) ∼ = An × Bn for nonzero coherent
OSn -modules An and Bn . By Formal Functions (4.6.13(b)), O bS,s ∼
= lim A × lim B ,
←− n ←− n
contradicting that OS,s is a local ring.
b

The following result will be applied to the morphism Mg → Spec Z to show

that Mg ×Z Fp is connected (Theorem 5.7.33). This application requires only the
lower semicontinuity below, which relies on Stein Factorization (4.6.14) but not on
Semicontinuity (4.6.11).

Corollary 4.6.15 (Zariski’s Connectedness Theorem II). Let f : X → Y be a proper

flat morphism of noetherian Deligne–Mumford stacks. The function

|Y| → Z, y 7→ #connected components of X ×Y Spec K, (4.6.16)

where Spec K → Y is any geometric point representing y, is lower semicontinuous.

If in addition f has geometrically reduced fibers, then (4.6.16) is locally constant.
f′
Proof. By étale descent, we can assume that Y = S is a scheme. Let X −→ S ′ → S
be the Stein Factorization (4.6.14). As S ′ → S is finite, there are only finitely
many points s′1 , . . . , s′m over a point s ∈ S. By Étale Localization of Quasi-Finite
Morphisms (A.7.1), after replacing S with an étale neighborhood of s, we can
arrange that S ′ = S1′ ⨿ · · · ⨿ Sm ′
with s′i ∈ Si′ and κ(s′i )/κ(s) purely inseparable.
Since X → S has geometrically connected fibers, each fiber Xs′i is geometrically
′

connected, and thus the geometric fiber of X → S over s has precisely m connected

205
components. As T X → S is flat, the image Ui ⊆ S of f (Si ) is open. In the open
′−1 ′

neighborhood i Ui of s, every geometric fiber has at least m connected components.

Assuming now that every fiber is geometrically reduced, the number of geo-
metric components over s ∈ S is precisely dimκ(s) H0 (Xs , OXs ). By Semicontinuity
(4.6.11), (4.6.16) is also upper semicontinuous. See also [EGA, IV3 .15.5.3-7] and
[SP, Tags 0BUI and 0E0N] for the case of schemes, [SP, Tags 0E1D and 0E1E] for
the case of algebraic spaces, and [DM69, Thm. 4.17] (for a statement in the case of
Deligne–Mumford stacks).
We also record the following application to the openness of ampleness, which
extends Proposition B.2.10 to algebraic spaces.

Proposition 4.6.17 (Openness of Ampleness). Let X be an algebraic space proper

and finitely presented over a scheme S, and L be a line bundle on X. If for some
geometric point s : Spec k → S, the restriction Ls of L to the fiber Xs is ample, then
there exists an open neighborhood U ⊆ S of s such that XU is a scheme and the
restriction LU to XU is relatively ample over U . In particular, for all u ∈ U , Lu is
ample on Xu .
Proof. Apply Le Lemme de Gabber (4.6.1) and the invariance of ampleness un-
der finite covers (Exercise 4.5.23), the statement reduces to the case of schemes
(Proposition B.2.10).

206
Chapter 5

Moduli of stable curves

The point is, I love maps, that is “maps” in the sense of “maps of the
world,” “charts of the ocean,” “atlases of the sky”! I think one of the key
things that attracted me to the group of problems was the hope of making
a map of some parts of the world of algebraic varieties. An algebraic
variety felt like a tangible thing in the lectures of Oscar Zariski, so why
shouldn’t you venture out, like Magellan, and uncover their geography?

Mumford [Mum04, p. vii]

This chapter proves Theorem A: for g ≥ 2, Mg is a smooth, proper, and

irreducible Deligne–Mumford stack of dimension 3g − 3 which admits a projective
coarse moduli space Mg → M g . In terms of the six-step strategy to construct
projective moduli spaces outlined in §0.7.2, we proceed as follows.
① (Algebraicity) We express Mg as a substack of the stack of all curves Mall g ,
and we prove that Mall g is an algebraic stack locally of finite type over Z
(Theorem 5.4.6).
② (Openness of Stability) Mg ⊆ Mall g is an open substack (Proposition 5.3.24).
③ (Boundedness) Mg is finite type (Theorem 5.4.14).
④ (Stable reduction) Mg is proper (Theorem 5.5.23).
⑤ (Existence of a moduli space) Applying the Keel–Mori Theorem (4.4.6) gives a
coarse moduli space Mg → M g with M g a proper algebraic space.
⑥ (Projectivity) Using Kollár’s Criterion for Ampleness (5.9.2), we show that
det π∗ (ωU⊗k/M ) is ample for k ≫ 0, where π : Ug → Mg is the universal family
g g
(Corollary 5.9.5).
Along the way, we develop the foundational properties of nodal curves (§5.2) and
stable curves (§5.3 and §5.6). The application to moduli theory necessitates that we
prove most properties for families of curves, which is one of the most technically
challenging facets of the exposition. The irreducibility of Mg is established in §5.7,
and a GIT construction of M g is sketched in §5.8.

5.1 Review of smooth curves

Everyone knows what a curve is, until he has studied enough mathematics
to become confused through the countless number of possible exceptions.

Felix Klein

207
 We review basic properties of smooth curves, which we will generalize to nodal
curves in the next section.

5.1.1 Curves
Definition 5.1.1. A curve over a field k is a one dimensional scheme C of finite
type over k.
Proper curves are projective [Har77, Prop. I.6.7] and moreover every one dimen-
sional proper algebraic space is, in fact, a projective scheme (Theorem 4.5.32). If C
is a proper curve over a field k, we define the arithmetic genus of C or simply the
genus of C as
g(C) = 1 − χ(C, OC ),
which is equal to h1 (C, OC ) if C is geometrically connected and reduced. The
geometric genus of a reduced proper curve C is the (arithmetic) genus of the
normalization C.e
For a reduced, connected, and projective curve C over an algebraically closed
field k, the degree of a very ample line bundle L on C is defined as the number of
zeros (counted with multiplicity) of any section of L. In other words, if C ,→ Pn is
the projective embedding defined by L, then deg L = dimk Γ(C ∩ H, OC∩H ), where
H is any hyperplane and C ∩ H is the scheme-theoretic intersection. Any line bundle
on C can be written as the difference of two very ample line bundles: if M is very
ample on C, then M ′ := L ⊗ M n is very ample for n ≫ 0, and L ∼= M ′ ⊗ (M ⊗n )∨ . In
this way, we can define deg L = deg M S− n deg M . Note that the degree is additive
′

under tensor product and that if C = i Ci denotes the irreducible decomposition,

then deg L = i deg L|Ci .
P

Theorem 5.1.2 (Riemann–Roch). Let C be a reduced, connected, and projective

curve of genus g over an algebraically closed field k. If L is a line bundle on C, then
χ(C, L) = deg L + 1 − g.
Proof. We can write L = OC (D) for a divisor D supported on the smooth locus.
Since Riemann–Roch holds for OC (by the definition of genus), it suffices by adding
and subtracting points to show that Riemann–Roch holds for OC (D) if and only if
it holds for OC (D + p) for a smooth point p ∈ C(k). This follows by considering the
short exact sequence
0 → OC (D) → OC (D + p) → κ(p) → 0
and the identity χ(C, OC (D + p)) = χ(C, OC (D)) + 1. See also [Har77, Thm IV.1.3,
Exc. IV.1.9] and [Vak17, Exers. 18.4.B and S].

5.1.2 Smooth curves

If C is a smooth curve, then the sheaf of differentials ΩC is a line bundle. Serre
Duality is the statement that ΩC is a dualizing sheaf on C, and is a deep result
indispensable in the study of curves.
Theorem 5.1.3 (Serre Duality for Smooth Curves). If C is a smooth projective
curve over a field k, then ΩC is a dualizing sheaf, i.e., there is a linear map
tr : H1 (C, ΩC ) → k such that for every coherent sheaf F, the natural pairing
tr
HomOC (F, ΩC ) × H1 (C, F) → H1 (C, ΩC ) −
→k
is perfect.

208
Proof. See [Ser88, Thm. II.8.2] and [Har77, Cor. III.7.12].
Remark 5.1.4. The pairing being perfect means that the HomOC (F, ΩC ) is identified
with the dual H1 (C, F)∨ . If F is a vector bundle, HomOC (F, ΩC ) ∼
= H0 (C, F ∨ ⊗ ΩC )
and Serre Duality gives an isomorphism

H0 (C, F ∨ ⊗ ΩC ) ∼
= H1 (C, F)∨ .

Taking F = ΩC , we see that H1 (C, ΩC ) ∼

= H0 (C, OC )∨ and that tr : H1 (C, ΩC ) → k
is an isomorphism if C is geometrically connected and reduced.
Serre Duality yields a more powerful version of Riemann–Roch.
Theorem 5.1.5 (Riemann–Roch II). Let C be a smooth, connected, and projective
curve of genus g over an algebraically closed field. If L is a line bundle on C, then

h0 (C, L) − h0 (C, ΩC ⊗ L∨ ) = deg L + 1 − g.

Remark 5.1.6. This is often written in divisor form as h0 (C, L) − h0 (C, K − L) =

deg L + 1 − g, where K denotes a canonical divisor, i.e., ΩC = OC (K).
The theorem of Riemann–Hurwitz (5.7.4) also plays an essential role in the study
of smooth curves, describing how the sheaf of differentials behaves under finite covers.
Our treatment is deferred until we discuss branched covers in §5.7.1.
Exercise 5.1.7 (easy). For any g ≥ 0, show that there exists a smooth, connected,
and projective curve of genus g.

Automorphisms. If C is a smooth curve over a field k and p1 , . . . , pn ∈ C(k),

we let Aut(C, pi ) denote the abstract group of automorphisms of (C, pi ) over k
∼
consisting of automorphisms α : C → C such that α(pi ) = pi for all i.
Proposition 5.1.8. Let C be a smooth, connected, and projective curve of genus
g over an algebraically closed field k, and let p1 , . . . , pn ∈ C(k) be distinct points.
The automorphism group Aut(C, pi ) is finite if 2g − 2 + n > 0, i.e., either g ≥ 2, or
g = 1 and n ≥ 1, or g = 0 and n ≥ 3. Moreover, if g ≥ 2 and char(k) = 0, then
| Aut(C)| ≤ 84(g − 1).
Proof. This is a classical result of Hurwitz [Hur92]. See also [Har77, Exc. IV.5.2],
[ACGH85, Exc. I.F.4], and [Mir95, Thm. 4.18] for the finiteness and [Har77, Exc. IV.2.5],
[ACGH85, Exc. I.F.10], and [Mir95, Thm.3.9] for the explicit bound.
Exercise 5.1.9. Let C be a smooth, connected, and projective curve over an
algebraically closed field k of genus g ≥ 2. Show that there are no non-trivial
automorphisms of C fixing more than 2g + 2 points.
While every hyperelliptic curve (and thus every curve of genus 2) has a non-trivial
hyperelliptic involution, the following exercise will be applied later to show that
general curves of higher genus are automorphism free.
Exercise 5.1.10. Show that if g ≥ 3 and k is an algebraically closed field, there
exists a smooth, connected, and projective curve C over k with trivial automorphism
group.
For further background on smooth curves, we recommend [Har77, §IV], [Vak17,
§20], [Liu02, §7], [ACGH85], and [Mir95].

209
5.1.3 Positivity of divisors on smooth curves
The following consequence of Riemann–Roch provides useful criteria to determine
whether a given line bundle is base point free (equivalently globally generated),
ample, or very ample.

Corollary 5.1.11. Let C be a smooth, connected, and projective curve over an

algebraically closed field k, and let L be a line bundle on C.
(1) if deg L < 0, then h0 (C, L) = 0;
(2) if deg L > 0, then L is ample;
(3) if deg L ≥ 2g, then L is base point free; and
(4) if deg L ≥ 2g + 1, then L is very ample.

Proof. See [Har77, Cor. IV.3.2].

Remark 5.1.12. If g > 1, we can use Riemann–Roch and Serre Duality to compute
that: (a) h0 (C, ΩC ) = h1 (C, OC ) = g, (b) h1 (C, ΩC ) = h0 (C, OC ) = 1, and (c)
ΩC has degree 2g − 2 and is thus ample on C. Similarly, if k > 1, we have: (a)
h0 (C, Ω⊗k ⊗k ⊗k
C ) = (2k − 1)(g − 1), (b) h (C, ΩC ) = 0, and (c) ΩC has degree 2k(g − 1)
1

and is very ample if k ≥ 3. Note that ΩC is not very ample precisely when C is
hyperelliptic. On the other hand, if g = 1 then ΩC ∼
= OC , and if g = 0 then C = P1
and ΩC = O(−2).

5.1.4 Classification of rational curves

Over an algebraically closed field k, every connected, smooth, and proper curve of
genus 0 is isomorphic to P1k . Over an arbitrary field, the classification is slightly
more involved.

Exercise 5.1.13 (Classificiation of the projective line). Show that the following
are equivalent for a proper curve C over a field k:
(a) C ∼
= P1 ;
k
(b) C is smooth and geometrically irreducible over k, C has genus 0, and C(k) ̸= ∅.
(c) C is Gorenstein and geometrically integral over k, C has genus 0, and C has a
line bundle of odd degree;
(d) H0 (C, OC ) = k, C has genus 0, and C has a line bundle of degree 1;
(e) H1 (C, OC ) = 0 and C has a line bundle of degree 1.
See also [SP, Tag 0C6U].

There is also a classification of singular rational curves that will be convenient

to understand nodal rational curves as well as the classification of rational tails and
bridges in prestable curves.

Exercise 5.1.14 (Classification of singular rational curves). Let C be a singular

proper curve of genus 0 over a field k with H0 (C, OC ) = k, and let π : C ′ → C be
the normalization. Show the following:
(a) C has a unique singular point p, which is a k-rational point;
(b) π : C ′ → C is an isomorphism over C ∖ p and π −1 (p) = {p′ } for a k′ -rational
point p′ ∈ C ′ ;
∼ P1′ for a finite field extension k′ /k; and
(c) C ′ = k

210
 (d) C is identified with the Ferrand Pushout (B.4.1)

p′
Spec k′ / C′

π
 
Spec k
p
/ C.

See also [SP, Tag 0DJB].

5.1.5 Families of smooth curves

Definition 5.1.15. A family of smooth curves (of genus g) over a scheme S is a
smooth and proper morphism C → S of schemes such that every geometric fiber is a
connected curve (of genus g).

Recall that the relative sheaf of differentials ΩC/S is a line bundle on C such that
for s ∈ S, the restriction ΩC/S |Cs is identified with ΩCs . More generally, for every
morphism T → S of schemes, the pullback of ΩC/S to CT := C ×S T is canonically
isomorphic to ΩCT /T . Generalizing Corollary 5.1.11, Ω⊗k C/S is relatively very ample
for k ≥ 3 and its pushforward is a vector bundle on S.

Proposition 5.1.16 (Properties of Families of Smooth Curves). Let π : C → S be a

family of smooth curves of genus g ≥ 2.
(1) π∗ OC = OS ;
(2) The pushforward π∗ (Ω⊗k
C/S ) is a vector bundle of rank

g if k = 1
r(k) :=
(2k − 1)(g − 1) if k > 1.

whose construction commutes with base change (i.e., for a morphism f : T → S

of schemes, f ∗ π∗ (Ω⊗k ∼ ⊗k
C/S ) = πT,∗ (ΩCT /T ));

(3) R1 π∗ Ω⊗k
C/S is isomorphic to OS if k = 1 and zero otherwise; and

(4) For k ≥ 3, Ω⊗k

C/S is relatively very ample. In particular, C → S is projective.

Proof. Items (1)–(3) follows from Cohomology and Base Change (A.6.8) as detailed
in Proposition A.6.10. For (4), observe that for every point s ∈ S, the fiber
Ω⊗k ⊗k ⊗k
C/S ⊗ κ(s) = ΩCs is very ample by Corollary 5.1.11 as deg ΩCs = k(2g − 2) > 0.
Since H1 (Cs , Ω⊗k ⊗k
Cs ) = 0, we may apply Proposition B.2.10 to conclude that ΩC/S is
relatively very ample.

It is also true that the relative sheaf of differentials ΩC/S is a relative dualizing
sheaf, i.e., satisfies a relative version of Serre Duality; see [Har66c] or [Liu02, §6.4].

5.2 Nodal curves

[Doing] mathematics for me is like being on a long hike with no trail and
no end in sight.

Maryam Mirzakhami

211
 After providing a Characterization of Nodes (5.2.4), we discuss the Genus Formula
(5.2.12), the dualizing sheaf (Definition 5.2.18), and the Local Structure of Nodal
Families (5.2.25).

5.2.1 Nodes
Definition 5.2.1 (Nodes). Let C be a curve over a field k.
bC,p ∼
• We say that p ∈ C(k) is a split node if there is an isomorphism O =
k[[x, y]]/(xy).
• We say that a closed point p ∈ C is a node if there exists a split node p ∈ Ck
over p.
We say that a curve C is a nodal (or has at-worst nodal singularities) if C is pure
one dimensional and every closed point is either smooth or nodal.

Figure 5.2.2: A node of a curve over C viewed algebraically (left-hand side) or

analytically (right-hand side).

Example 5.2.3.
(1) The curve C = Spec k[x, y]/(xy) has a split node at 0. The normalization
e∼
C = A1 ⨿A1 has coordinate ring k[x]×k[y] and Γ(C, OC ) = {(f, g) ∈ Γ(C,
e O e) |
C
f (0) = g(0)}, or in other words C is the Ferrand Pushout (B.4.1)

Spec k ⨿ Spec k
0⨿0 /C
e

 
Spec k
0 / C.

(2) The nodal cubic C = Spec k[x, y]/(y 2 − x2 (x + 1)) has a split node at 0. The
e∼
normalization C = A1 has a coordinate t = y/x with x = t2 − 1 and y = t3 − t,
and C is again the Ferrand Pushout obtained from C e by gluing the two
preimages of 0.
(3) The curve C = Spec R[x, y]/(x2 + y 2 ) has a node at 0, but it is not a split node
as the quadratic form x2 + y 2 does not split into linear factors.
(4) The curve Spec Q[x, y]/(x2 − 2)(y 2 − 3) has a non-split node at the point p
defined by the maximal ideal (x2 − 2, y 2 − 3). Note that unlike the previous
examples where the nodes are rational points, the node p is not a rational
point and the field extension Q → κ(p) has degree 4.
(5) Let k → k′ be a separable field extension of degree 2, and let C be the affine
curve over k defined by the k-algebra {f ∈ k′ [x] | f (0) ∈ k}. In other words,
C is the Ferrand Pushout of the inclusion of the origin 0 : Spec k′ ,→ A1k′ along
Spec k′ → Spec k. The curve C has a non-split node at a k-rational point.

212
5.2.2 Equivalent characterizations of nodes
Recall from §A.3.4 that the singular locus Sing(C) of a curve C is defined scheme-
theoretically as the vanishing of the first fitting ideal sheaf of ΩC : locally if C =
V (f1 , . . . , fm ) ⊆ An , then Sing(C) is defined by the vanishing of all (n − 1) × (n − 1)
∂f
minors of the Jacobian matrix J = ( ∂xji ); note that if C = V (f ) ⊆ A2 is a plane
affine curve, then Sing(C) = V ( ∂f ∂x , ∂y ). We will also use properties of local complete
∂f

intersections as discussed in §A.3.5.

Proposition 5.2.4 (Characterization of Nodes). Let C be a pure one dimensional

curve over a field k, and let p ∈ C be a closed point with maximal ideal m ⊆ OC,p .
The following are equivalent:
(1) p ∈ C is a node;
(2) C is a local complete intersection at p, and Sing(C) is unramified over k at
p ∈ Sing(C);
(3) k → κ(p) is separable, OC,p is reduced, dim m/m2 = 2, and there is a nonde-
generate quadratic form q ∈ Sym2 m/m2 mapping to 0 in m2 /m3 ;
(4) k → κ(p) is separable and ObC,p ∼= κ(p)[[x, y]]/(q) where q is a nondegenerate
quadratic form; and
(5) there exists a finite separable field extension k → k′ and a point p′ ∈ Ck′ such
bC ′ ,p′ ∼
that O = k′ [[x, y]]/(xy).
k

Proof. Assuming (1), let p ∈ Ck be a node over p and let Sing(C) ⊆ C be

the scheme-theoretic singular locus. By properties of fitting ideals (see §A.3.4),
Sing(C) ×k k = Sing(Ck ) and the preimage of Sing(Ck ) under Spec O bC ,p → C is
k k
b b ∼
Sing(Spec OCk ,p ). Since OCk ,p = k[[x, y]]/(xy), Sing(Spec OCk ,p ) = V (x, y) = Spec k.
b
Therefore Sing(C) → Spec k is unramified at p. Since O bC ,p is a complete inter-
k
section, C is a local complete intersection at p (Proposition A.3.16). This gives
(2).
Assuming (2), since Sing(C) is unramified at p, the field extension k → κ(p)
is separable and there is an open neighborhood U ⊆ C of p such that Sing(U ) =
Sing(C) ∩ U = {p}. In particular, OC,p is generically reduced. On the other hand,
since C is a local complete intersection, OC,p is a one dimensional Cohen–Macaulay
local ring and thus has no embedded primes. It follows that OC,p is reduced. Using
that C is a local complete intersection, we can write O bC,p = R/(f1 , . . . , fn−1 ) where
R is a regular complete local ring. As R contains the base field k, the Cohen
Structure Theorem (B.5.7) implies that R ∼ = κ(p)[[x1 , . . . , xn ]]. Since Sing(C)
  is
∂fj
unramified at p, the (n − 1) × (n − 1) minors of the Jacobian matrix ∂xi
i,j
generate the maximal ideal m = (x1 , . . . , xn ) ⊆ O bC,p . If ∂fj ∈ R is a unit for
∂xi
some i and j, then the sequence x1 , . . . , xbi , xn , fj also generates m/m2 . We may use
Complete Nakayama’s Lemma (B.5.6(3)) to change coordinates by replacing the
generators x1 , . . . , xn with x1 , . . . , xbi , xn , fj . Eliminating fj allows us to write R =
κ(p)[[x1 , . . . , xbi , xn ]]/(f1 , . . . , fbj , . . . , fn−1 ). After finitely many such replacements, we
∂f
can assume that ∂xji ∈ m for every i, j. This implies that every (n−1)×(n−1) minor
is in mn−1
, but since these minors generate m, we must have that n = 2. Therefore,
ObC,p = κ(p)[[x, y]]/(f ) with f = f2 + f3 + · · · and each fi homogeneous of degree i.
Since the partials fx and fy generate (x, y), the quadratic form q := f2 ∈ Sym2 m/m2
must be nondegenerate. This gives (3).

213
 Assuming (3), we have that dimκ(p) md /md+1 = 2 for every d ≥ 1 since q maps
to 0 in m2 /m3 . A choice of elements x0 , y0 ∈ m mapping to a basis in m/m2 induces
a surjection κ(p)[[x, y]] → O bC,p by Complete Nakayama’s Lemma (B.5.6(3)). Since
OC,p is reduced, O bC,p is also reduced (Proposition 2.1.21). Since κ(p)[[x, y]] is a
UFD, we can write κ(p)[[x, y]]/(f ) for an element f expressed as a product of distinct
irreducible elements such that the quadratic component q = ax2 + bxy + cy 2 of
f is a nondegenerate quadratic form. We claim that we can modify our choice
of coordinates x0 , y0 ∈ m so that f = q, i.e., q(x0 , y0 ) = 0 ∈ ObC,p . We will show
inductively that for each N , there exists elements xi , yi ∈ mi+1 for i = 0, . . . , N
such that q(x0 + · · · + xN , y0 + · · · + yN ) ∈ m
P . Since
N +3
P OC,p is complete, this
b
would enable us to replace x0 and y0 with x
i i and i yi and conclude that
∼
OC,p = κ(p)[[x, y]]/(q). Supposing that we have already chosen x′ = x0 + · · · + xN −1
b
and y ′ = y0 + · · · + yN −1 , then for every xN and yN ∈ mN +1 , we have that
q(x′ + xN , y ′ + yN ) = q(x′ , y ′ ) + (2ax0 + by0 )xN + (bx0 + 2cy0 )yN mod mN +3
The nondegeneracy of q = ax2 + bxy + y 2 implies that 2ax0 + by0 and bx0 + 2cy0
are linearly independent. Since dimκ(p) mN +2 /mN +3 = 2, we may choose xN and
yN such that Q(x′ + xN , y ′ + yN ) ∈ mN +3 . This completes (4).
Assuming (4) and using that q is nondegenerate, we may choose a degree 2
separable field extension κ(p) → k′ such that q splits as a product of linear forms.
Thus O bC,p ⊗k k′ ∼
= k′ [[x, y]]/(xy), yielding (5). Finally, (5) implies that p is a node.
See also [SP, Tags 0C49, 0C4D, and 0C4E].
Exercise 5.2.5. (easy) Show that  Spec k[x, y]/(f) has a node at 0 if and only if
f (0) fxy (0)
f (0) = fx (0) = fy (0) = 0 and det xx is nonzero.
fyx (0) fyy (0)
Exercise 5.2.6. (good practice) Let C be a pure one dimensional reduced curve
over a field k with normalization π : C
e → C. Show that p ∈ C is a node if and only if
k → κ(p) is separable, dimκ(p) (π∗ OCe /OC ) ⊗ κ(p) = 1, and π(q)=p [κ(q) : κ(p)] = 2.
P

Hint: Identify (π∗ OCe /OC ) ⊗ κ(p) with the quotient A/A
e where A = O bC,p and
Ae is its normalization, using that normalization commutes with completion (see
Remark 2.1.25). To show (⇐), use that A e is a product of complete DVRs to derive
the structure of A. See also [SP, Tag 0C4A].
Remark 5.2.7. The quantityPdimκ(p) (π∗ OCe /OC ) ⊗ κ(p) is referred to as the δ-
invariant of p, and the sum π(q)=p [κ(q) : κ(p)]sep is referred to as the number of
geometric branches over p. A cusp k[[x, y]]/(y 2 − x3 ) has δ-invariant one but has only
one geometric branch.
Proposition 5.2.8 (Local Structure of Nodes). Let C be a curve over a field k. If
p ∈ C is a node, then there exists a finite separable field extension k → k′ and étale
neighborhoods

(U, u)

(5.2.9)
{ (
(C, p) (Spec k′ [x, y]/(xy), 0).

Proof. Using the Characterization of Nodes (5.2.4(5)), there is a finite separable

bC,p ⊗k k′ ∼
field extension k → k′ such that O = k′ [[x, y]]/(xy). The result is now a
consequence of Artin Approximation (B.5.21).

214
 We will shortly generalize this to families of nodal curves (Theorem 5.2.25).

Example 5.2.10. Let us construct an explicit étale neighborhood for the nodal
cubic C = Spec C[x, y]/(y 2 − x2 (x − 1)). In fact, in Example√ 0.5.3 we essentially
already showed how do this: if we add a square root t = x − 1, then y 2 − x3 +
x2 = (y − xt)(y + xt). Therefore, we can take the elementary étale neighborhood
U = Spec C[x, y, t]t /(y 2 − x3 + x2 , t2 − x + 1) → C given by (x, y, t) 7→ (x, y) and
the étale neighborhood U → Spec C[x, y]/(xy) defined by (x, y, t) 7→ (y − xt, y + xt).

Exercise 5.2.11. Provide a proof of the Local Structure of Nodes (5.2.8) without
appealing to Artin Approximation.
sh
Hint: Use that the normalization of a strict henselization OC,p has two components
to find an affine étale neighborhood (Spec R, u) → (C, p) with R e = R1 × R2 . Use
the exact sequence 0 → R → R1 × R2 → κ(u) → 0 to construct elements x, y ∈ R
mapping to (1, 0), (0, 1) ∈ R1 × R2 , and argue that κ(u)[x, y]/(xy) → R is étale.

5.2.3 Genus formula

Proposition 5.2.12 (Genus Formula). Let C be a connected, nodal, and projective
S an algebraically closed field k with δ nodes ande ν irreducible components.
curve over
Let C = i Ci be the irreducible decomposition and let C i be the normalization of
Ci with genus g(Cei ). The genus g of C satisfies

ν
X
g= ei ) + δ − ν + 1.
g(C
i=1

Proof. Let p1 , . . . , pδ ∈ C denote the nodes of C. We claim that the normalization

e → C induces a short exact sequence
π: C

δ
M
0 → OC → π∗ OCe → κ(pj ) → 0.
j=1

It suffices to verify this étale-locally around a node pi ∈ C, and so by the Local

Structure of Nodes (5.2.8), we can assume that C = Spec k[x, y]/(xy). In this case,
Ce = Spec(k[x] × k[y]) and the sequence above corresponds to 0 → k[x, y]/(xy) →
k[x] × k[y] → k → 0. Alternatively, normalization commutes with completion and a
direct calculation as above shows that if A := O bC,p ∼ e ∼
= k[[x, y]]/(xy), then A/A = k.
The short exact sequence induces a long exact sequence on cohomology
M
0 → H0 (C, OC ) → H0 (C,
e O e) → κ(pj ) → H1 (C, OC ) → H1 (C,
e O e) → 0
| {z } | {z C } j
| {z } | {z C }
1 ν g P
i g(Ci )
| {z } e
δ

where the labels indicate the dimensions. The statement follows.

215
 Figure 5.2.13: An example of a nodal curve of genus 14.

Remark 5.2.14. Notice that δ − ν + 1 is precisely the number of connected regions

bounded by the curve C as in Figure 5.2.13. Thus, the genus of a nodal curve can be
easily computed from the picture by summing the geometric genera of the irreducible
components and adding the number of bounded regions. See Definition 5.3.8
for another interpretation using dual graphs. For non-nodal curves, there is an
analogous formula for the genus involving the delta invariants (see Remark 5.2.7) of
the singularities.

5.2.4 The dualizing sheaf

Since a nodal curve C over a field k is a locally a complete intersection, C is Gorenstein
∼
and there is a dualizing line bundle ωC with a trace map trC : H1 (C, ωC ) → k; see
[Har77, III.7.11] or [Ser88, §IV]. In other words, for every coherent sheaf F , the
natural pairing
tr
HomOC (F, ωC ) × H1 (C, F ) → H1 (C, ωC ) −
→k

is perfect. Due to its importance in the study of stable curves, we provide an explicit
description of ωC below over an algebraically closed field k. Let Σ := C sing be the
singular locus and U = C ∖ Σ. Let π : C e → C be the normalization of C, and let Σ e
and U be the preimages of Σ and U as in the diagram
e

e 
U /C
eo ? _Σ
e
(5.2.15)
∼

π
   
U /Co ? _ Σ.

Let Σ = {z1 , . . . , zn } be an ordering of the points and π −1 (zi ) = {pi , qi }. Since C

e
is smooth, the sheaf of differentials ΩCe is a dualizing line bundle. There is a short
exact sequence
0 → ΩCe → ΩCe (Σ)
e → Oe → 0
Σ (5.2.16)
obtained by tensoring the sequence 0 → OCe (−Σ) e → O e → O e → 0 with Ω e (Σ).
C Σ C
e
As ΩCe (Σ)|Ue = ΩUe , we can interpret sections of ΩCe (Σ) as rational sections of ΩCe
e e
with at worst simple poles along Σ.e Evaluating (5.2.16) on an open Ve ⊆ C e yields

/ Γ(Ve , Ω e ) / Γ(Ve , Ω e (Σ)) /

L
0 e e ∩Σ κ(y),
C C y∈V
(5.2.17)
e
s 7→ (resy (s))

216
where the last map can be interpreted as taking a rational section s ∈ Γ(Ve ∩ Ue, Ω e)
C
to the tuple whose coordinate at y ∈ Ve ∩ Σ
e is the residue resy (s) of s at y.

Definition 5.2.18. Let C be a nodal curve over an algebraically closed field k.

Using the notation of (5.2.15) and (5.2.17), we define the subsheaf ωC ⊆ π∗ ΩCe (Σ) e
by declaring that sections along V ⊆ C consist of rational sections s of ΩCe along
π −1 (V ) with at worst simple poles along Σ e such that for every node zi ∈ V ∩ Σ with
preimages pi , qi ∈ π −1 (V ), respi (s) + resqi (s) = 0.
The definition implies that ωC sits in the following two exact sequences:

0 / ωC / π∗ Ω e (Σ) /L /0
zi ∈Σ k
e
C
s 7→ (respi (s) + resqi (s))

0 / π∗ Ω e / ωC /L / 0.
C zi ∈Σ k
s 7→ (respi (s))

Example 5.2.19 (Local calculation). Let C = Spec k[x, y]/(xy). Then C e = A1 ⨿ A1

with coordinates x and y respectively. The singular locus of C is Σ = {0} with
preimage Σ e consisting of the two origins. Then Γ(C, e Ω e ) = Γ(A1 , ΩA1 ) × Γ(A1 , ΩA1 )
C
and ( x , − y ) is a rational section of ΩCe with opposite residues at p and q. In fact,
dx dy

every section of Γ(C, ωC ) is of the form

 
dx −dy dx −dy
f (x) , g(y) = (f (x) + g(y) − f (0)) · ( , )
x y x y
for polynomials f (x) and g(y) such that f (0) = g(0), which is precisely the condition
for (f, g) ∈ Γ(C,
e O e ) to descend to a global function f (x) + g(y) − f (0) ∈ Γ(C, OC ).
C
In other words, ωC ∼ = OC with generator ( dx dy
x , − y ).

Example 5.2.20. Let C be the nodal projective plane cubic and P1 → C be the
normalization with coordinates [x : y] such that 0 and ∞ are the fibers of the node.
The rational differential η := dx
x = − y on P satisfies res0 η + res∞ η = 0, and it
dy 1

is not hard to see that every local section of ωC is a multiple of η; in other words,
ωC ∼= OC with generator η.
Exercise 5.2.21 (details). Let C be a connected, nodal, and projective curve over
an algebraically closed field k. Define temporarily the subsheaf
!
M
e →
KC = ker π∗ Ω e (Σ) k
C
zi ∈Σ

as in Definition 5.2.18.
(a) Show that if π : C ′ → C is an étale morphism, then π ∗ KC ∼
= KC ′ .
Hint: Use the fact that normalization commutes with étale base change.
(b) Show that KC is a line bundle.
Hint: Use the Local Structure of Nodes (5.2.8) and the local calculation in
Example 5.2.19.
(c) (hard) Show that KC is a dualizing sheaf.
Hint: Reduce to the case of a smooth curve by using the normalization:

217
 (i) Use Yoneda’s Lemma to show that it suffices to exhibit an isomorphism
HomOC (−, KC ) ∼
= H1 (C, −)∨ of functors Pic(C) → Vectk .
(ii) For any L ∈ Pic(C), show that tensoring the exact sequence defining KC
by L∨ and taking cohomology gives identifications
!
HomOC (L, KC ) ∼
= H0 (C, L∨ ⊗KC ) ∼
M
e π ∗ L∨ ⊗ Ω e (Σ))
= ker H0 (C, e → k .
C
zi ∈Σ

(iii) Use the short exact sequence 0 → π∗ π ∗ (L(Σ))

e →L→ L
zi ∈Σ k → 0 to
deduce that
!
M
1 ∨ 1 ∗ ∨
H (C, L) = ker H (C,
e π L(Σ))
e → k .
zi ∈Σ

(iv) Show that the isomorphism H0 (C, e ∼

e π ∗ L∨ ⊗ Ω e (Σ)) e π ∗ L(Σ))
= H1 (C, e ∨
C L
given by Serre Duality on C
e commutes with the two maps to
zi ∈Σ k.

See also [Ols16, Prop 13.2.9]. For other approaches, see [Ser88, Ch. IV §3],
[Liu02, Lem. 10.3.12], and [ACG11, Ch. X §2].
(d) If T ⊆ C is a subcurve with complement T c := C ∖ T , show that

ωC |T = ωT (T ∩ T c ).

Exercise 5.2.22 (good practice). Let C be a connected, nodal, and projective curve
over an algebraically closed field k. Let π : C
e → C be the normalization and Σ
e ⊆Ce
the set of preimages of nodes. Show that there is an identification
e O e) ∼
π∗ H omOCe (ΩCe (Σ), C = H omOC (ΩC , OC ),

or equivalently that π∗ (TCe (−Σ))

e = TC . In other words, regular vector fields on C
correspond to regular vector fields on C
e vanishing at the preimages of nodes.

5.2.5 Nodal families

Recall that the relative singular locus Sing(C/S) of a morphism C → S with one
dimensional fibers is defined by the first fitting ideal sheaf of ΩC/S ; see Defini-
tion A.3.14. Syntomic morphisms are fppf morphisms (i.e., flat and locally of finite
presentation) whose fibers are local complete intersections; see §A.3.5.

Proposition 5.2.23. Let C → S be an fppf morphism of schemes and s ∈ S a point

such that the fiber Cs is pure one dimensional. A point p ∈ Cs is a node if and only if
C → S is syntomic at p and the relative singular locus Sing(C/S) → S is unramified
at p.

Proof. The conditions that C → S is syntomic at p and Sing(C) → S is unramified

at p are both conditions on the fibers over s. Since Sing(C/S)s = Sing(Cs ), the
result follows from the equivalence of (1) ⇔ (4) of the Characterization of Nodes
(5.2.4).

The above characterization implies that the property of being a nodal family
descends under limits (Definition B.3.6).

218
Lemma 5.2.24. The following property of morphisms of schemes descends under
limits: an fppf morphism such that every fiber is a pure one dimensional nodal curve.

Proof. From Descent of Properties of Morphisms under Limits (B.3.7), we know that
the properties of being fppf, syntomic, unramified, and having connected pure one
dimensional fibers descend under limits. Since the relative singular locus commutes
with base change, the result follows from Proposition 5.2.23.

5.2.6 Local structure of nodal families

By the Local Structure of Smooth Morphisms (A.3.4), if C → S is a family of smooth
curves, then every point p ∈ C over s ∈ S is étale locally isomorphic to relative one
dimensional affine space. More precisely, there is a commutative diagram

(C, p) o
op
? _ (C ′ , p′ ) ét / (S ′ ×Z A1 , (s′ , 0))
Z

  w
(S, s) o
op
? _ (S ′ , s′ ),

where the left horizontal maps are open immersions, the right-hand map is étale
map, and S ′ ×Z A1Z = A1S ′ → S ′ is the base change of A1Z → Spec Z. We establish an
analogous local structure for a family of nodal curves providing a relative version of
the Local Structure of Nodes (5.2.8).

Theorem 5.2.25 (Local Structure of Nodal Families). Let π : C → S be an fppf

morphism such that every geometric fiber is a curve. Let p ∈ C be a node in the fiber
Cs over a point s ∈ S. There is a commutative diagram

(C, p) o / (Spec A[x, y]/(xy − f ), (s′ , 0))

ét ét
(C ′ , p′ )
(5.2.26)
  t
(S, s) o
ét
(Spec A, s′ ),

where each horizontal map is étale and f ∈ A is a function vanishing at s′ .

Remark 5.2.27. In other words, every family of nodal curves étale locally on the
source and target has the structure of the top horizontal arrow in the fiber product

Spec A[x, y]/(xy − f ) / Spec A

f
 
Spec Z[x, y, t]/(xy − t) / Spec Z[t]

induced by a function f ∈ A.

Proof. We offer a local-to-global proof.

Step 1: Reduce to the case where S is of finite type over Z. Use limit methods and
Lemma 5.2.24.
Step 2: Reduce to the case where O ∼ κ(s)[[x, y]]/(xy). By Proposition 5.2.4,
bC ,p =
s
there is a finite separable field extension κ(s) → k′ and a point p′ ∈ Cs ×κ(s) k′

219
whose completion is isomorphic to k′ [[x, y]]/(xy). Letting (S ′ , s′ ) → (S, s) be an étale
morphism such that κ(s′ ) ∼ = k′ over κ(s), we replace S with S ′ .
Step 3: Show that O bC,p ∼=ObS,s [[x, y]]/(xy − fb) for a function fb ∈ m bs ⊆ O bS,s . We
claim that there exists elements xb, yb ∈ ObC,p and fb ∈ m
bs ⊆ ObS,s such that x byb − fb = 0.
To achieve this, we will inductively construct elements xn , yn ∈ OC,p and fn ∈ m
b bs
for n ≥ 0 which are compatible (i.e., xn+1 ≡ xn (mod mn+1 ), etc.) such that
b n+1
xn yn − fn ∈ m s O
bC,p . (5.2.28)
The claim follows by defining x
b = limn xn , yb = limn yn , and fb = limn fn .
The base case n = 0 is handled by Step 2: letting x, y ∈ O bC ,p be the images of
s
b ∼
x and y under the isomorphism OCs ,p = κ(s)[[x, y]]/(xy), choose x0 , y0 ∈ O bC,p to be
any lifts of x and y under the surjection ObC,p → O bC ,p and set f0 = 0. Assuming
s
that we have constructed xn , yn , and fn satisfying (5.2.28), write
X
xn yn − fn = ai bi , where ai ∈ m b n+1
s and bi ∈ ObC,p .
i

Since xn and yn generate the maximal ideal of p in the fiber Cs , we can use the
identity κ(s) = κ(p) on residue fields to find a′i ∈ O
bS,s and b′ , b′′ ∈ O
i i
bC,p such that

bi − (xn b′i + yn b′′i + a′i ) ∈ m

b sO
bC,p .

We then define
X X X
xn+1 = xn − ai b′′i , yn+1 = yn − ai b′i , and fn+1 = fn + ai a′i ,
i i i

and check that

X X X
xn+1 yn+1 − fn+1 = (xn − ai b′′i )(yn − ai b′i ) − (fn + ai a′i )
i i i
X X X X
= (xn yn − fn ) − xn ai b′i − yn ai b′′i − ai a′i + ai aj b′′i b′j
i i i i,j
X X X X X
= ai bi − xn ai b′i − yn ai b′′i − ai a′i + ai aj b′′i b′j
i i i i i,j
X X
= ai (bi − xn b′i − yn b′′i − a′i ) + ai aj b′′i b′j
|{z} | {z } |{z}
i i,j
b n+1
m s b sO
m bC,p b 2(n+1)
m s

is an element of m
b n+2
s
bC,p .
O
With the claim established, we have a well-defined O
bS,s -algebra homomorphism

O
bS,s [[x, y]]/(xy − fb) → O
bC,p ,

defined by x 7→ x b and y 7→ yb. This map is surjective by Complete Nakayama’s

Lemma (B.5.6(2)) as it is surjective modulo m b s , and it is injective by a version of
the local criterion for flatness (Lemma A.2.8); it is thus an isomorphism.
Step 4 (formal-to-étale): Extend the isomorphism in Step 3 on formal neighorhoods
to étale neighborhoods. From Step 3, we have a diagram

C ×S Spec O
bS,s Spec O
bS,s [x, y]/(xy − fb)

' v
Spec O
bS,s

220
such that the points (p, s) ∈ C ×S Spec O
bS,s and (s, 0) ∈ Spec O
bS,s [x, y]/(xy − fb) have
isomorphic completions. A consequence of Artin Approximation (B.5.21) implies
that there are étale morphisms

(U, u)

v )
(C ×S Spec O
bS,s , (p, s)) (Spec O
bS,s [x, y]/(xy − fb), (s, 0))
(5.2.29)

defined over Spec O

bS,s . After replacing S with an open affine neighborhood of s, we
can assume that S = Spec A is affine. By Neron–Popescu (B.5.15), we may write
ObS,s = colim Bλ as a directed colimit of smooth A-algebras. Set Sλ = Spec Bλ ,
Cλ = C ×S Sλ , and Uλ = U ×S Sλ . For λ ≫ 0, fb ∈ O bS,s is the image of an element
fλ ∈ Bλ , and the pullbacks of x and y to Γ(U, OU ) are the pullbacks of elements in
Γ(Uλ , OUλ ) under U → Uλ . This yields a commutative diagram

Uλ

~ (
Cλ Spec Bλ [x, y]/(xy − fλ )

over Sλ which base changes to (5.2.29) under Spec O bS,s → Sλ . By Descent of

Properties of Morphisms under Limits (B.3.7), étaleness descends under limits and
so the maps Uλ → Cλ and Uλ → Spec Bλ [x, y]/(xy − fλ ) are étale for λ ≫ 0. Letting
uλ = (u, sλ ) ∈ Uλ , we have a commutative diagram

(C, p) o / (Spec Bλ [x, y]/(xy − fλ ), (sλ , 0))

sm ét
(Uλ , uλ )

  t
(S, s) o
sm
(Spec Bλ , sλ )

This gives our desired diagram (5.2.26) except that the left horizontal arrows are
smooth rather than étale. Since smooth maps étale locally have sections (Corol-
lary A.3.5), there is an étale map (Spec A, s′ ) → (S, s) and a map (Spec A, s′ ) →
(Sλ , sλ ) over S. The result follows from setting C ′ := Uλ ×Sλ Spec A and p′ = (uλ , s′ ).
See also [SP, Tag 0CBY].

Exercise 5.2.30. Provide an alternative proof avoiding Artin Approximation/Neron–

Popescu as follows:
(i) Reduce to the case where S is of finite type over Z.
(ii) Reduce to the case that C → S is a syntomic morphism of affine schemes whose
geometric fibers are connected with at most two irreducible components and
that Sing(C/S) → S is either an isomorphism or a closed immersion defined
by a nonzerodivisor.
(iii) Handle the two cases that Sing(C/S) = S and Sing(C/S) = V (f ) by using
properties of normalization and henselization.

221
5.3 Stable curves

The only problem is to pick the “boundary” components shrewdly, i.e. to

decide which non-generic varieties to allow.

Alan Mayer and David Mumford [MM64a, p. 3]

Stable curves were introduced in unpublished joint work by Mayer and Mumford
[MM64a, MM64b].

5.3.1 Definition and equivalences

An n-pointed curve is a curve C over an algebraically closed field k together with

an ordered collection of k-points p1 , . . . , pn ∈ C; we call the points pi ∈ C marked
points. A point q ∈ C of an n-pointed curve is called special if q is a node or a
marked point.

Definition 5.3.1 (Stable, semistable, and prestable curves). An n-pointed connected,

nodal, and projective curve (C, p1 , . . . , pn ) of genus g over an algebraically closed
field k is stable if

(1) p1 , . . . , pn ∈ C(k) are distinct smooth points,

(2) C is not of genus 1 without marked points, i.e., (g, n) ̸= (1, 0), and
(3) every smooth irreducible rational subcurve P1 ⊆ C contains at least 3 special
points.

If (1)–(2) hold, and (3) is replaced with the condition that every smooth rational
subcurve contains at least 2 (rather than 3) special points, we say that (C, pi ) is
semistable. If only (1)–(2) hold, we say that (C, pi ) is prestable.

We have the implications:

stable ⇒ semistable ⇒ prestable ⇒ n-pointed nodal.

In the unpointed case, a curve is a prestable curve if it is nodal. When k is not

algebraically closed, condition (3) in the definition of stability needs to be amended;
see Proposition 5.3.14

Remark 5.3.2. Note that there are no n-pointed stable curves of genus g if (g, n) ∈
{(0, 0), (0, 1), (0, 2), (1, 0)}, or equivalently 2g − 2 + n ≤ 0. We often impose the
condition that 2g − 2 + n > 0 to exclude these special cases.

222
 NEETU

I in
a te
Figure 5.3.3: Examples of stable, semistable, and prestable curves

∼
An automorphism of a stable curve (C, p1 , . . . , pn ) is an automorphism α : C → C
such that α(pi ) = pi . We denote by Aut(C, p1 , . . . , pn ) the abstract group of
automorphisms. Recall also that if C is a geometrically smooth, connected, and
not prestablecurve of genus g ≥ 2, then Aut(C) is finite (Proposition 5.1.8).
projective
Proposition 5.3.4. Let (C, p1 , . . . , pn ) be an n-pointed prestable curve over an
algebraically closed field k. The following are equivalent:
(1) (C, p1 , . . . , pn ) is stable, not semistable
(2) Aut(C, p1 , . . . , pn ) is finite, and
(3) ωC (p1 + · · · + pn ) is ample.
Proof. The equivalence (1) ⇐⇒ (2) follows from Exercise 5.3.6 and the fact that
the only way that a smooth prestable n-pointed curves (C, pi ) can have a positive
dimensional automorphism group is if C = P1 with n ≤ 2 or. if C is a genus 1 curve
with n = 0 (see Proposition 5.1.8). To see the equivalence with (3), we will use the
fact that for a subcurve T ⊆ C, we have ωC |T = ωT (T ∩ T c ) (Exercise 5.2.21). The
line bundle ωC (p1 + · · · + pn ) is ample if and only if its restriction to each irreducible
component T ⊆ C
X
ωC (p1 + · · · + pn )|T = ωT ( pi + (T ∩ T c )) (5.3.5)
pi ∈T

is ample. If the genus g(T ) of T is at least two, then ωT is ample and thus so is
(5.3.5). If g(T ) = 1, then (5.3.5) is ample if and only if n ≥ 1 or T must meet the
complement T C . If g(T ) = 0, then (5.3.5) is ample if and only if T contains at least
three special points.
Exercise 5.3.6 (Pointed normalization). Let (C, p1 , . . . , pn ) be an n-pointed prestable
curve. Let π : C e → C be the normalization of C, pei ∈ C e be the unique preimage of
pi , and qe1 , . . . , qem ∈ C be an ordering of the preimages of nodes. We call (C,
e e pei , qej )
the pointed normalization of (C, pi ). Show that (C, pi ) is stable if and only if every
pointed connected component of (C, e pei , qej ) (i.e., each connected component of C e
together with the points pei and qej lying on it) is stable.

223
Exercise 5.3.7. Let (C, p1 , . . . , pn ) be an n-pointed prestable curve.
(a) Show that the automorphism group scheme Aut(C, pi ) is an algebraic group.
(b) Show that Aut(C, pi ) is naturally a closed subgroup of the automorphism
group Aut(C, e pei , qej ) of the pointed normalization (C, e pei , qej ).
(c) Show that Aut(C, pi )0 = Aut(C, e pei , qej )0 but that in general Aut(C, pi ) ̸=
Aut(C,
e pei , qej ).

Dual graphs. A vertex-weighted, n-marked graph Γ = (G, w, m) is the data of

a finite, connected, and undirected graph G with vertices V (G) and edges E(G)
(where loops and parallel edges are allowed), a weight function w : V (G) → Z≥0 ,
and an n-marking m : {1, . . . , n} → V (G). Each vertex m(i) ∈ V (G) is viewed as a
half-edge. The genus of Γ is
X
g(Γ) = g(G) + w(v),
v∈V (G)

where
g(G) = |E(G)| − |V (G)| + 1

denotes the genus of G, i.e., the first Betti number of G considered as a 1-dimensional
CW complex. We say that a vertex-weighted, n-marked graph Γ is stable if for every
vertex v ∈ V (G)
2w(v) − 2 + val(v) + |m−1 (v)| > 0,

where val(v) is the valence of v, defined as the number of edges containing v with
loops counted twice. In other words, stability means that for every vertex v either
(a) w(v) ≥ 2, (b) w(v) = 1 and v is contained in an edge or half-edge, or (c) w(v) = 0
and val(v) + |m−1 (v)| ≥ 3.

0
0 3

3
dual graph
5

5 2

Figure 5.3.7: A genus 13 curve and its dual graph.

Definition 5.3.8 (Dual graph). The dual graph of an n-pointed prestable curve
(C, p1 , . . . , pn ) is defined as Γ = (G, w, m), where the vertices vi of G correspond to
irreducible components Ci of C and the weight w(vi ) is the geometric genus of Ci .
For every node q of C lying on components Ci and Cj there is an edge eq between
vi and vj , and the marking m(i) is the vertex vj with pi ∈ Cj .

Exercise 5.3.9 (easy). Show that an n-pointed prestable curve (C, pi ) is stable if
and only if its dual graph Γ is stable, and that the genus of C is equal to the genus
of Γ.

224

I
 The stack Mg,n of stable curves admits a stratification according to the dual
graph (Exercise 5.6.23).

1 1 1

1 1 1

Figure 5.3.9: There are seven genus 2 vertex-weighted graphs with no marked points.
Curves are displayed on the top while their dual graphs are displayed below. The
arrows indicate specialization. Going from the left column to the right, the dimension
of the substacks of M2 parameterizing curves with the given dual graph are 3, 2, 1,
I
and 0.

Exercise 5.3.10.
(a) Classify stable, vertex-weighted 2-marked graphs of genus 1.
(b) Determine the number of isomorphism classes of stable, vertex-weighted 0-
marked graphs of genus 3.

5.3.2 Rational tails and bridges

Definition 5.3.11 (Rational tails and bridges). Let (C, p1 , . . . , pn ) be an n-pointed
prestable curve over a field k. We say that an irreducible smooth subcurve E ⊆ C
of genus 0 with nonempty complement E c = C ∖ E is

• a rational tail if the scheme-theoretic intersection E ∩ E c is a single reduced

point x, H0 (E, OE ) ∼= κ(x), and E contains no marked points, and
• a rational bridge if the scheme-theoretic intersection E ∩ E c has degree 2 over
H0 (E, OE ) and E contains no marked points, or if E contains a single marked
point pj and E is a rational tail of (C, p1 , . . . , pbj , . . . , pn ).

Over an algebraically closed field, every irreducible smooth rational subcurve E

is isomorphic to P1 . The subcurve E is a rational tail if E · E c = 1 and E has no
marked points, and is a rational bridge if either E · E c = 2 and E has no marked
points or if E · E c = 1 and E has precisely one marked point.

225
.-
as
0 To 0 To

EP P

yep
L a 2
Figure 5.3.11: (A) features a rational tail while (B) and (C) feature rational bridges.

Over an arbitrary field, the classification of rational tails and bridges is more
subtle, which is unsurprising given that the classification of rational curves is also
more involved (see Exercises 5.1.13 and 5.1.14).
Lemma 5.3.12. Let (C, pi ) be an n-pointed prestable curve over a field k.
(1) If E is a rational tail with k′ := H0 (E, OE ) = κ(x), then k′ is a finite separable
field extension of k and E ∼ = P1k′ .
(2) If E is a rational bridge with no marked points, then k′ := H0 (E, OE ) is a
finite separable field extension of k and k′′ := H0 (E ∩ E c , OE∩E c ) is either a
degree 2 separable field extension of k′ or k′′ = k′ × k′ .
Proof. In both cases, since E is reduced and connected, k′ = H0 (E, OE ) is a field.
For (1), the intersection x = E ∩ E c is a node and hence k′ = κ(x) is a separable field
extension of k (Proposition 5.2.4). Viewing E as a scheme over k′ via structure map
E → Spec Γ(E, OE ) = Spec k′ , we see that E ∼ = P1k′ since it is irreducible, smooth,
genus 0, and contains a rational point. For (2), the points of E ∩ E c are nodes in C
but smooth points in both E and E c . As dimk′ k′′ = 2, we see that either k′′ is a
field, in which case it is separable over k by Proposition 5.2.4, or k′′ = k′ × k′ .
Exercise 5.3.13. If K/k is a field extension, (C, pi ) has a rational tail (resp.,
rational bridge) if and only if (C ×k K, pi ×k K) has a rational tail (resp., rational
bridge).
Stability is equivalent to not containing a rational tail or bridge.
Proposition 5.3.14. Let (C, pi ) be an n-pointed prestable curve of genus g over a
field k. If K is an algebraically closed field containing k, then (C ×k K, pi ×k K) is
stable (resp., semistable) if and only if (C, pi ) contains no rational tails or rational
bridges (resp., no rational tails).
Proof. Over an algebraically closed field, stability (resp., semistability) is equivalent
to not containing a rational tail or bridge (resp., rational tail). Therefore, the
statement follows from the fact that having a rational tail or bridge is insensitive to
field extensions (Exercise 5.3.13).
Remark 5.3.15 (Relationship to −1 and −2 curves). Suppose that C → ∆ = Spec R
is a family of nodal curves over a DVR R with algebraically closed residue field
k such that the generic fiber is smooth. If E ∼ = P1 ⊆ C0′ is a smooth rational
subcurve in the central fiber, then E = −E · E (which follows from the identity
2 c

0 = E · C0 = E · E + E · E c ). Thus if E is a rational tail (resp., rational bridge

without a marked point), then E 2 = −1 (resp., E 2 = −2).

226
 EP
Ez E 2 p

sas sas

To
Figure 5.3.15: In (A) (resp., (B)), the exceptional component E meets the rest of
the curve at one point (resp., two points) and E 2 = −1 (resp., E 2 = −2).

EP
The following exercise will be generalized later by the Stable Contraction of a
Prestable Family (5.6.6).
yep
Exercise 5.3.16 (hard). Let C → ∆ = Spec R be a family of nodal curves over a

4 a 2
DVR R with smooth generic fiber. For any rational tail or bridge E ⊂ C0 in the
central fiber, show that there is a contraction C → C ′ of E where C ′ → ∆ is a family
of nodal curves.
Hint: Castelnuovo’s Contraction Theorem (B.2.6) gives the existence of the contrac-
tion C → C ′ . The challenge here is to show that the central fiber C0′ is nodal. You
may want to appeal to a fact in the minimal model program that C0′ is nodal if and
only if the pair (C ′ , C0′ ) is log canonical [Kol13, Cor. 2.32, Thm. 4.9(2)], and show
that the latter property holds because (C, C0 ) (resp., (C, C0 − E)) is log canonical.

5.3.3 Positivity of the dualizing sheaf

Exercise 5.3.17 (moderate). Let (C, p1 , . . . , pn ) be an n-pointed prestable curve
over an algebraically closed field k, and let L := ωC (p1 + · · · + pn ).
(a) If (C, pi ) is stable, show that L⊗k is very ample for k ≥ 3 and that H1 (C, (ωC (p1 +
· · · + pn ))⊗k ) = 0 for k ≥ 2.
(b) For k ≥ 2, show that (C, pi ) is semistable if and only if L⊗k is base point free.
Hint: For (a), show that the global sections of L⊗k separate points and tangent
vectors or, in other words, that the maps

H0 (C, L⊗k ) → L⊗k ⊗ κ(x) ⊕ L⊗k ⊗ κ(y) H0 (C, L⊗k ) → L⊗k ⊗ OC,x /m2x

are surjective. Establish this using Serre Duality and a case analysis on whether
x, y are smooth or nodal. See also [DM69, Thm. 2], [ACG11, Lem. 10.6.1], [SP,
Tag 0E8X], and [Ols16, Prop. 13.2.17].

Exercise 5.3.18 (good practice). If C is the nodal union C1 ∪ C2 of genus i and

g − i curves along a single node p = C1 ∩ C2 , show that ωC has a base point at p.

5.3.4 Families of stable curves

Definition 5.3.19. A family of n-pointed curves over a scheme S is a proper, flat,
and finitely presented morphism C → S of algebraic spaces together with n sections

227
σ1 , . . . , σn : S → C such that every geometric fiber Cs is a curve over Spec κ(s) (e.g.,
a finite type κ(s)-scheme of dimension 1).
A family of n-pointed stable curves (resp., semistable curves, prestable curves,
nodal curves) is a family of n-pointed curves such that every geometric fiber
(Cs , σ1 (s), . . . , σn (s)) is stable (resp., semistable, prestable, nodal).
Caution 5.3.20. If (C → S, σi ) is a stable family over a scheme S, C → S is necessarily
projective (Proposition 5.3.21), hence C is a scheme. While every one dimensional
separated algebraic space of finite type over a field is a scheme (Theorem 4.5.32), in
the relative setting, the total family C may not be a scheme. There exists a families
of prestable genus 0 curves [Ful10, Ex. 2.3] and smooth genus 1 curves [Ray70, XIII
3.2] where the total families are not schemes; see also Example 2.1.16. In the next
section, we show that the stack of all curves (resp., nodal curves, semistable curves,
prestable curves) is algebraic and in particular satisfies the descent condition for
families of curves, and this result requires that the total family is an algebraic space.
For a family of n-pointed curves (resp., nodal curves), there is no condition on
whether the marked points are distinct or land in the relative smooth locus of C → S.
For stable, semistable, and prestable families, the marked points are distinct and
avoid the nodes. If (C → S, σi ) is a family of n-pointed prestable curves, then C → S
is a local complete intersection morphism. Thus there is a relative dualizing line
bundle ωC/S that is compatible with base change T → S and in particular restricts
to the dualizing line bundle ωCs on every fiber of C → S; see [Har66c] or [Liu02,
§6.4]. The image of each section σi is a divisor contained in the smooth locus, and
we can form the line bundle ωC/S (σ1 + · · · + σn ).
The following statement extends Proposition 5.3.4 to families. We will denote by
Aut(C/S, σ1 , . . . , σn ) the locally of finite type group scheme over S parameterizing
automorphisms of a family of n-pointed prestable curves.
Proposition 5.3.21. Let (C → S, σi ) be a family of n-pointed prestable curves of
genus g over a scheme S. The following are equivalent:
(1) (C → S, σi ) is a family of stable curves,
(2) Aut(C/S, σ1 , . . . , σn ) → S is quasi-finite, and
(3) ωC/S (σ1 + · · · + σn ) is relatively ample over S.
In particular, a family of n-pointed stable curves is a projective morphism of schemes.
Proof. Each condition can be checked on geometric fibers so the statement follows
from Proposition 5.3.4.
The separatedness of Mg,n (Proposition 5.5.20) implies that Aut(C/S, σ1 , . . . , σn ) →
S is, in fact, a finite group scheme.
Exercise 5.3.22. Let (π : C → S, σi ) be a family of n-pointed prestable curves and
let L = ωC/S (σ1 + · · · + σn ). For k ≥ 2, show that (C → S, σi ) is semistable if and
only if π ∗ π∗ L → L is surjective.
The following generalization of Properties of Families of Smooth Curves (5.1.16)
is proven in the same way using the very ampleness of ωC (p1 + · · · + pn )⊗3 for a
stable n-pointed curve (C, pi ) over an algebraically closed field (Exercise 5.3.17).
Proposition 5.3.23 (Properties of Families of Stable Curves). Let (C P→ S, σi )
be a family of n-pointed stable curves of genus g, and set L := ωC/S ( i σi ). If
k ≥ 3, then L⊗k is relatively very ample and π∗ (L⊗k ) is a vector bundle of rank
(2k − 1)(g − 1) + kn.

228
Proposition 5.3.24 (Openness of Stability). Let (π : C → S, σi ) be a family of
n-pointed curves. The locus of points s ∈ S such that (Cs , σi (s)) is stable (resp.,
semistable, prestable, nodal) is open.
Proof. We first claim that the nodal locus is open. When C is a scheme, this follows
from the Local Structure of Nodes (5.2.25): the locus C ≤nod ⊆ C of points which
are smooth or nodal in their fiber is open, and therefore the locus of points s ∈ S
such that Cs has worse than nodal singularities is identified with the closed locus
π(C ∖ C ≤nod ) ⊆ S. In general, we can choose an étale cover C ′ → C by a scheme and
use the fact that a point p ∈ C is a node in its fiber Cπ(p) if and only if a preimage
p′ ∈ C ′ of p is a node in its fiber Cπ(p) ′
. Since the locus in S where σ1 (s), . . . , σn (s)
are distinct and smooth points in Cs is open, the condition that Cs is prestable is
open. For an n-pointed prestable family (π : C → S, σi ), the locus of points s ∈ S
where Cs is semistable is identified with the open locus over which π ∗ π∗ L → L is
surjective, where L := ωC/S ( i σi ) (see Exercise 5.3.22).
P
To see that stability is an open condition in an n-pointed prestable family, we pro-
vide two arguments. First, observe that the fiber dimension of Aut(C/S, σ1 , . . . , σn ) →
S is upper semicontinuous (B.1.8), and therefore, the locus of points s ∈ S such
that Aut(Cs , σ1 (s), . . . , σn (s)) is finite is open. As this condition categorizes stability
(Proposition 5.3.4(2)), this open subset is identified with the stable locus. Second,
by Openness of AmplenessP (Proposition 4.6.17), the locus of points s ∈ S such that
ωC/S ( i σi )|Cs ∼= ωCs ( i σi (s)) is ample is open, and this condition also categorizes
P
stability (Proposition 5.3.4(3)).

5.3.5 Deformation theory of stable curves

If C is a smooth curve over a field k, then every first-order deformation is locally
trivial (Proposition C.1.8) and the set Def(C) of isomorphism classes of first-order
deformations is naturally in bijection with H1 (C, TC ) (Proposition C.1.11). More gen-
erally, automorphisms, deformations, and obstructions of higher-order deformations
are classified by Hi (C, TC ) for i = 0, 1, 2 (Proposition C.2.4). Nodal singularities,
on the other hand, have first-order deformations that are not locally trivial, e.g.,
Spec k[x, y, ϵ]/(xy − ϵ) → Spec k[ϵ].
Proposition 5.3.25. Let A′ ↠ A be a surjection of artinian local rings with residue
field k. Suppose that J = ker(A′ → A) satisfies mA′ J = 0. Let (C → Spec A, σi ) be
a family of prestable curves over A, and let (C, pi ) be its base change to the residue
field k.
′
Pautomorphisms of a deformation of (C → Spec A, σi ) over A is
(1) The group of
0
ExtOC (ΩC ( i pi ), OC ⊗k J).
(2) There is no obstruction to deforming (C → Spec A, σi ) over A′ .
′
P of deformations of (C → Spec A, σi ) over A is
(3) The set of isomorphism classes
1
a torsor under ExtOC (ΩC ( i pi ), OC ⊗k J) and

dimk Ext1OC (ΩC ( i pi ), OC ) − dimk Ext0OC (ΩC ( i pi ), OC ) = 3g − 3 + n.

P P

Moreover, if (C, pi ) is stable, then Ext0OC (ΩC ( i pi ), OC ) = 0, i.e., there are no

P

non-trivial automorphisms of a deformation, and dimk Ext1OC (ΩC ( i pi ), OC ) =

P
3g − 3 + n.
Proof. Since C is generically smooth and a local complete intersection and the
marked points pi ∈ C are smooth, it is a consequence of Proposition C.2.4 (unpointed

229
case) and Proposition C.2.8 (pointed case) that automorphisms, deformations, and
obstructions of a nodal curve C are classified by
ExtiOC (ΩC , I)
for i = 0, 1, 2, where I ⊆ OC is
Pthe ideal sheaf of {p1 , . . . , pn } ⊆ C. Since I defines
a Cartier divisor, I = OC (− i pi ), and thus automorphisms, deformations, and
obstructions are classified by
ExtiOC (ΩC ( i pi ), OC ).
P

To compute the dimensions of these Ext groups, we first handle the unpointed case.
Since HomOC (ΩC , −) is the composition Γ ◦ H omOC (ΩC , −) of left exact functors,
there is a Grothendieck spectral sequence
E2p,q = Hp (C, E xtqOC (ΩC , OC )) ⇒ ExtO
p+q
C
(ΩC , OC ),
c.f., [Wei94, Thm. 5.8.3]. Since dim C = 1, we have that E2p,q = 0 if p ≥ 2. We can
thus draw the E2 -page as:

Setting TC = H omOC (ΩC , OC ), the associated exact sequence of low-degree terms

is

0 → H1 (C, TC ) → Ext1OC (ΩC , OC ) → H0 (C, E xt1OC (ΩC , OC ))

| {z } | {z }
E21,0 E20,1

→ H2 (C, TC ) = 0 . (5.3.26)
| {z }
E22,0

As ΩC is locally free away from the nodes, E xt1OC (ΩC , OC ) and E xt2OC (ΩC , OC )
are zero-dimensional sheaves supported only at the nodes Σ = {q1 , . . . , qs } of C. It
follows that E21,1 = H1 (C, E xt1OC (ΩC , OC )) = 0 and
M
E20,1 = H0 (C, E xt1OC (ΩC , OC )) = Ext1Ob (ΩObC,q , O
bC,q )
j
C,qj j
j
M
E20,2 = H0 (C, E xt2OC (ΩC , OC )) = Ext2Ob (ΩObC,q , O
bC,q ).
j
C,qj j
j

To compute ExtiOb (Ω1Ob bC,q ), we may assume that k is algebraically closed.

,O j
C,qj C,qj

Consider the locally free resolution of O

bC,q = k[[x, y]]/(xy)
j
 
y
x (dx,dy)
0→O
bC,q −− b⊕2 −
→O
j C,qj −−−→ ΩO bC,q → 0.
j

230
 (x,y)
This shows that Ext1Ob (ΩObC,q , O b⊕2 −
bC,q ) = coker(O −−→ ObC,q ) = k and
C,qj j
j C,qj j

Ext2Ob bC,q ) = 0. The former implies that dimk E 0,1 = |Σ|, the number of
(ΩObC,q , O
C,qj j
j 2

nodes, while the latter implies that Ext2OC (ΩC , OC ) = 0 (as E20,2 = E21,1 = E22,0 = 0).
By appealing to the short exact sequence (5.3.26) and the identities H 0 (C, TC ) =
Ext0OC (ΩC , OC ) and Hi (C, TC ) = Hi (C, e (see Exercise 5.2.22), we have that
e T e (−Σ))
C

dimk Ext1OC (ΩC , OC ) − dimk Ext0OC (ΩC , OC ) = |Σ| − χ(C, TC )

 
= |Σ| − χ C,
e T e (−Σ)
C
e .

Write the normalization Ce = `ν C

i=1 i as a union of its connected components
e
and set Σ to be the preimage of the nodal locus Σ ⊆ C with Σ
e e We may
ei ∩ Σ.
ei = C
apply Riemann–Roch to compute that
  ν
X  
|Σ| − χ C, C
e = |Σ| −
e T e (−Σ) χ C
ei , T e (−Σ
Ci
e i)
i=1
ν 
X 
= |Σ| + ei ) − 3 + |Σ
3g(C e i|
i=1
ν
!
X
=3 g(Ci ) − ν + |Σ|
e
i=1
= 3g − 3,
Pν
where we have used the Genus Formula (5.2.12) g = i=1 g(C ei ) − ν + |Σ| + 1. This
completes the proof of (1)–(3) in the case of an unpointed prestable curve.
In the case of marked points, let D = P
{p1 , . . . , pn } ⊆ C. Applying HomOC (ΩC , −)
to the short exact sequence 0 → OC (− i pi ) → OC → OD → 0 gives a long exact
sequence

0 → Ext0OC (ΩC ( i pi ), OC ) → Ext0OC (ΩC , OC ) → Ext0OC (ΩC , OD ) →

P

Ext1OC (ΩC ( i pi ), OC ) → Ext1OC (ΩC , OC ) → 0 → Ext2OC (ΩC ( i pi ), OC ) → 0,

P P

where we’ve used that ExtiOC (ΩC , OD ) = 0 for i = 1, 2 (since D is zero-dimensional).

This yields that Ext2OC (ΩC ( i pi ), OC ) = 0. Moreover, since dimk HomOC (ΩC , OD ) =
P
n and since we’ve already computed the difference of the Ext groups in the unpointed
case, the above long exact sequence implies that

dimk Ext1OC (ΩC ( i pi ), OC ) − dimk Ext0OC (ΩC ( i pi ), OC )

P P

= dimk Ext1OC (ΩC , OC ) − dimk Ext0OC (ΩC , OC ) + n = 3g − 3 + n.

Finally, suppose that (C, pi ) is a stable curve with pointed normalization (C,
e pei , Σ).
e
Exercise 5.2.22 implies that

HomOC (ΩC ( i pi ), OC ) ∼
P P e O e ).
= HomOCe (ΩCe ( i pei + Σ), C

Since the pointed normalization is smooth and each pointed connected component
is stable (Exercise 5.3.6), the degree of the restriction of TCe (− i pei − Σ)e to each
P

connected component of C e is strictly negative. Thus, HomO (ΩC ( P

C i pi ), OC ) = 0.
See also [DM69, Prop. 1.5] and [ACG11, §11.3].

231
 The proof above shows more, namely it reveals an important relationship between
global and local deformations. We denote by Def(C, pi ) (resp., Def lt (C, pi )) the
vector space of first-order deformations (resp., locally trivial first-order deformations)
of (C, pi ).

Proposition 5.3.27 (Local-to-global Deformation Sequence). Let (C, pi ) be an

n-pointed prestable curve over an algebraically closed field k, and let Σ ⊆ C be the
e → C be the normalization, pei the unique preimage of pi , and
set of nodes. Let π : C
e = π −1 (Σ). There is an exact sequence
Σ
M
0 → Def lt (C, pi ) → Def(C, pi ) → Def(O
bC,q ) → 0 (5.3.28)
q∈Σ

and identifications

Def lt (C, pi ) ∼
= H1 (C, TC ) ∼
= H1 C, e ∼


e T e (− P pei − Σ = Def(C,
e pi , Σ)
e
C i
Def(C, pi ) ∼ = Ext1 (ΩC (
P
OC ipi ), OC )
bC,q ) ∼
Def(O = Ext1Ob 1
(ΩOb , O bC,q ) ∼
= k, for q ∈ Σ.
C,q C,q

Exercise 5.3.29 (Deformation of stable curves revisited).

(a) Provide a direct argument that deformations of an unpointed stable curve are
unobstructed.
Hint: Letting qi ∈ C be the nodes, choose a cover C = U1 ∪ U2 of affine opens
with each qi ∈ U2 ∖ U1 . Using that U1 and U12 = U1 ∩ U2 are unobstructed
(as they are smooth) and U2 is also unobstructed (as it is lci and affine, c.f.,
Proposition A.3.7), show that deformations of U1 , U12 , and U2 can be glued to
a deformation of C. See also [SP, Tag 0DZQ].
(b) Extend this argument to the pointed case.
(c) Can you show this instead using properties of the cotangent complex?

Remark 5.3.30 (Consequences of deformation theory). Very shortly we will know

that Mg,n is an algebraic stack. We will then be able to use deformation theory to
argue that Mg,n is a smooth Deligne–Mumford stack of dimension 3g − 3 + n (see
Theorem 5.4.14). Here is the central idea:

• Ext0 : We have already seen that a stable curve (C, pi ) has finitely many
automorphism (Proposition 5.3.4). The vanishing of Ext0 implies that an
n-pointed stable curve (C, pi ) has no infinitesimal automorphisms, i.e., the
automorphism group scheme Aut(C, pi ) is reduced, finite, and étale. This will
allow us to use the Characterization of Deligne–Mumford Stacks (3.6.4) to
conclude that Mg,n is Deligne–Mumford.
• Ext1 : Since Ext1 parametrizes isomorphism classes of deformations of a stable
curve (C, pi ) over a field k, it is identified with the Zariski tangent space of
Mg,n ×Z k at the k-point corresponding to (C, pi ). The computation of Ext1
implies that Mg,n has relative dimension 3g − 3 + n over Spec Z.
• Ext2 : The vanishing of Ext2 implies that there are no obstructions to deforming
(C, pi ), and thus the Infinitesimal Lifting Criterion (3.7.1) implies that Mg,n
is smooth over Spec Z.

232
5.4 The stack of all curves
It’s better to work with a nice category containing some nasty objects,
than a nasty category containing only nice objects.

John C. Baez

We show that the stack Mall

g,n of all n-pointed proper curves is algebraic (The-
orem 5.4.6) and that the open substack Mg,n ⊆ Mall g,n of stable curves is a quasi-
compact Deligne–Mumford stack smooth over Spec Z of dimension 3g − 3 + n
(Theorem 5.4.14).

5.4.1 Families of arbitrary curves

Recall that a curve over a field k is a one dimensional scheme of finite type over k,
and that the genus of a projective curve C is

g(C) = 1 − χ(C, OC ).

Curves may be non-pure dimensional, non-connected, and arbitrarily singular (e.g.,

non-reduced). While there is an algebraic stack parameterizing geometrically reduced
curves (Remark 5.4.17), once we allow non-reduced curves, we must also allow non-
pure dimensional curves, as otherwise the stack would fail to be algebraic. For
instance, the plane nodal curve with an embedded point at the node, arising from
the flat degeneration of a rational normal curve P1 ,→ P3 [Har77, Ex. 9.8.4], deforms
in a flat family to the disjoint union of a plane nodal cubic and a point in P3 .
A family of n-pointed curves over a scheme S is a proper, flat, and finitely pre-
sented morphism C → S of algebraic spaces together with n sections σ1 , . . . , σn : S →
C such that every fiber Cs is a curve. We say that a family (C → S, σi ) has genus g
if every fiber Cs has genus g. Note that the marked points σi (s) may be non-distinct
and singular in a fiber Cs . For the stack of curves to be algebraic, it is necessary
that we allow the total family C to be an algebraic space (see Caution 5.3.20).

Proposition 5.4.1. If C → S is a family of curves over a scheme S, there exists

an étale cover S ′ → S such that CS ′ → S ′ is projective.

Proof. We offer a local-to-global argument. We first reduce to the case that S is

of finite type over Spec Z. By Noetherian Approximation (B.3.2), we can write
S = limλ∈Λ Sλ as a limit of finitely presented Z-schemes with affine transition maps.
By Descent of Morphisms under Limits (B.3.3), there is an index 0 ∈ Λ and a
finitely presented morphism C0 → S0 such that C ∼ = C0 ×S0 Sλ , and moreover if
we set Cλ = C0 ×S0 S for λ ≥ 0, then C = limλ≥0 Cλ . By Descent of Properties of
Morphisms under Limits (B.3.7), Cλ → Sλ is a family of curves for λ ≫ 0. Since
projectivity is stable under base change, we may replace S with Sλ .
For a point s ∈ S, define Sn = Spec OS,s /mn+1s and Sb = Spec O bS,s . Consider
the cartesian diagram
 / C1   / · · · 
Cs = C0  / Cb /C

□ □
   
 / S1   
Spec κ(s) = S0  / ··· / Sb / S.

233
Since separated one dimensional algebraic spaces are schemes (Theorem 4.5.32) and
proper one dimensional schemes are projective, there exists an ample line bundle
L0 on C0 . By Proposition C.2.11, the obstruction to deforming a line bundle Ln
on Cn to Ln+1 on Cn+1 lives in H 2 (C0 , OC0 ⊗κ(s) mn ). The obstruction vanishes
since dim C0 = 1, and thus there is a compatible sequence of line bundles Ln on Cn .
By Grothendieck’s Existence Theorem (C.5.3), there exists a line bundle L b on C
b
extending Ln .
Applying Artin Approximation (Theorem B.5.18) to the functor

Sch/S → Sets, (T → S) 7→ Pic(CT ),

we obtain an étale neighborhood (S ′ , s′ ) → (S, s) of s and a line bundle L′ on CS ′

extending L0 . Since L0 is ample and ampleness is an open condition in families
(Proposition B.2.10), after replacing S ′ with an open neighborhood of s′ , we can
arrange that L′ is relatively ample over S ′ .

Exercise 5.4.2. Provide an alternative argument by explicitly extending an ample

line bundle:
(i) Use limit methods to reduce to the case that S is the spectrum of a strictly
henselian local ring R with closed point s, and reduce the statement to the
surjectivity of Pic(C) → Pic(Cs ).
(ii) Apply Le Lemma de Gabber (4.6.1) to reduce to the case that C is a scheme.
(iii) Reduce to the case that C is reduced, and show that Pic(C) → Pic(Cs ) is
surjective in this case by explicitly extending a line bundle OCs (−xs ) for each
closed point xs ∈ Cs .
See also [SGA4 12 , IV.4.1], [Hal13, Lem. 1.2], and [Ols16, Cor. 13.2.5].

Remark 5.4.3. Raynaud gives an example of a family of smooth g = 1 curves over

an affine curve which is Zariski-locally projective but not projective [Ray70, XIII
3.1]. The examples in Caution 5.3.20 are not even Zariski-locally projective.

5.4.2 Algebraicity of the stack of all curves

Definition 5.4.4. Let Mall g,n be the prestack over Schét , where an object over a
scheme S is a family of curves C → S of genus g with n sections σ1 , . . . , σn : S → C.
A morphism (C ′ → S ′ , σ1′ , . . . , σn′ ) → (C → S, σ1 , . . . , σn ) is the data of a cartesian
diagram
′ g
/C
AC @
σi′ σi
 
S′
f
/S

such that g ◦ σi′ = σi ◦ f .

As a stepping stone to the algebraicity of Mall

g,n , we show that the diagonal is
representable.

Lemma 5.4.5. The diagonal Mall all all

g,n → Mg,n × Mg,n is representable.

Proof. For simplicity, we handle the case when n = 0. Let S be a scheme and
S → Mallg × Mg be a morphism corresponding to families of curves C1 → S and
all

234
C2 → S. Considering the cartesian diagram

IsomS (C1 , C2 ) /S

□
 
Mall
g
/ Mall
g × Mall
g ,

we need to show that IsomS (C1 , C2 ) is an algebraic space. By Proposition 5.4.1,

there exists an étale cover S ′ → S such that CS ′ → S ′ is projective. Since
IsomS (C1 , C2 ) ×S S ′ = IsomS ′ (C1,S ′ , C2,S ′ ), the morphism IsomS ′ (C1,S ′ , C2,S ′ ) →
IsomS (C1 , C2 ) is surjective, étale, and representable. We may thus assume that C1
and C2 are projective over S.
Consider the inclusion of functors:

IsomS (C1 , C2 ) ⊆ MorS (C1 , C2 ) ⊆ Hilb(C1 ×S C2 /S)

where the second inclusion assigns a morphism α : C1 → C2 to the graph Γα : C1 ,→

C1 ×S C2 (and is similarly defined on T -valued points). Since the subfunctor of
Mor(−, S), parameterizing maps T → S where (C1 )S → (C2 )S is an isomorphism,
is open (Exercise 0.3.44), the first inclusion is a representable open immersion.
Analyzing the second inclusion, we see that a subscheme [Z ⊆ C1 ×S C2 ] ∈ Hilb(C1 ×S
C2 /S)(S) is contained in the image of Mor(C1 , C2 )(S) if and only if the composition
p1
Z ,→ C1 ×S C2 −→ C1 is an isomorphism (and similarly for T -valued points). Therefore,
Exercise 0.3.44 also establishes that the second inclusion is a representable open
immersion.

Theorem 5.4.6. Mall

g,n is an algebraic stack locally of finite type over Spec Z.

Proof. To see that Mall g,n is a stack over Schét , suppose that {Si → S} is an étale
cover of schemes, (Ci → Si , σi,1 , . . . , σi,n ) is a family of n-pointed curves for each i,
and αij : Ci |Sij → Cj |Sij is an isomorphism over Sij = Si ×S Sj compatible with the
sections and satisfying the cocycle condition. The quotient of the étale equivalence
relation a a
Cij ⇒ Ci ,
i,j i

with maps defined by p1 and p2 ◦ αij , is an algebraic space C. Moreover, by étale

descent of morphisms, there are sections σ1 , . . . , σn : S → C such that σk,i = σk |Si .
Thus (C → S, σk ) is a n-pointed family of curves that restricts to (Ci → Si , σi,k ) for
each i.
To see algebraicity, we claim that it suffices to handle the n = 0 case. Observe
that the map Mall g,n+1 → Mg,n , defined by forgetting the last marked point, is
all

well-defined and representable: if S → Mall g,n is a map corresponding to a n-pointed

family (C → S, σk ), then S ×Mall g,n
Mg,n+1 ∼
all
= C. (In fact, Mallg,n+1 → Mg,n is
all

identified with the universal family; see Exercise 5.4.8.) Therefore, if U → Mall g,n is
a smooth presentation by a scheme, then U := U ×Mall
′
g,n
Mg,n+1 → Mg,n+1 is a
all all

surjective, étale, and representable map from an algebraic space U ′ . Choosing an

étale presentation V → U ′ by a scheme, the composition V → U ′ → Mall g,n+1 is a
smooth presentation.
To show algebraicity of Mall g , it suffices show that for every projective curve
X over a field k, there exists a smooth representable morphism U → Mall g from

235
a scheme U of finite type over Z with [X] ∈ |Mall g | in the image. Choose an
embedding X ,→ PN such that H1 (X, OX (1)) = 0, and let P (t) be its Hilbert
polynomial. Let H := HilbP (PN
Z /Z) be the Hilbert scheme, which is projective over
Z by Theorem 1.1.2. Considering the universal family

C / PN
H

 ~
H,

there is a point h ∈ H(k) such that Ch = X as closed subschemes of PN k . Cohomology

and Base Change (A.6.8) implies that there exists an open neighborhood H ′ ⊆ H of
h such that for all s ∈ H ′ , H1 (Cs , OCs (1)) = 0. Consider the morphism

H ′ → Mall
g , [C ,→ Pn ] 7→ [C],

defined by forgetting the embedding. The representability of the diagonal of Mall g

(Lemma 5.4.5) implies that H ′ → Mall g is representable (Corollary 3.2.3).
We claim that H ′ → Mall g is smooth. Even though we haven’t yet established
the algebraicity of Mall
g , we do know that H ′ → Mall g is representable and this is
suffices to apply the Infinitesimal Lifting Criterion (3.7.1) to verify smoothness. To
this end, let A′ → A be a surjection of artinian local rings with residue field k such
that k = ker(A′ → A). For every embedded curve [C ⊆ PN ] ∈ H ′ and diagram

[C⊆PN
k ]


N %
Spec k / Spec A [C⊆PA ] / H ′
_ :
(5.4.7)
[C ′ ⊆PN
A′
]
 
Spec A′ / Mall
g ,
′
[C ]

we need to show that there is a dotted arrow extending the diagram. This translates
to the existence of a dotted arrow in the diagram

N N
: Pk 9 PA PN
: A′

,  ,
C / C+   / C′

    
Spec k  / Spec A   / Spec A′ .

The closed immersion C ,→ PN A is defined by a very ample line bundle L on C and

sections s0 , . . . , sN ∈ Γ(C, L). As the obstruction to deforming L to C ′ lives in
H2 (C, OC ) = 0 (Proposition C.2.11), we may extend L to a line bundle L′ on C ′ . We
now argue that the sections si deform to sections s′i ∈ Γ(C ′ , L′ ). As ker(A′ → A) = k,
the ideal sheaf defining C ,→ C ′ is isomorphic to OC , and we have a short exact
sequence
0 → OC → OC ′ → OC → 0.

236
Tensoring with L′ yields a short exact sequence

0 → L|C → L′ → L → 0.

Since [C ⊆ PN ] ∈ H ′ , we have that H1 (C, L|C ) = 0. Thus, H0 (C ′ , L′ ) → H0 (C, L) is

surjective and we may lift the sections si to sections s′i . The sections si are base
point free, and Nakayama’s Lemma implies that so are the sections s′i . This gives
a morphism j ′ : C ′ → PNA′ restricting to C ,→ PA . The map j : C → PA′ is proper
N ′ ′ N

and quasi-finite, thus finite. To see that it is a closed immersion, it suffices to show
that OPN′ → j∗′ OC ′ is surjective. The cokernel is a coherent sheaf whose restriction
A
to A vanishes, and thus Nakayama’s lemma implies that the cokernel vanishes. See
also [Hal13, Thm. 1.1] and [JHS11, Prop. 3.3]. This can also be established using
Artin’s Axioms; see Theorem C.7.7 and [SP, Tag 0D5A].
Exercise 5.4.8 (easy). Show that map Mall g,n+1 → Mg,n forgetting the last marked
all

point is the universal family as defined in Definition 3.1.27. (We will see in Propo-
sition 5.6.12 that Mg,n+1 is also the universal family of Mg,n , but this is a more
remarkable fact since an n-pointed stable curve can become unstable if a marked
point is forgotten.)
Exercise 5.4.9 (technical). Show that Mall g,n has quasi-compact and separated
diagonal, and in particular Mg,n is quasi-separated.
all

Exercise 5.4.10 (good practice). Show that the stack P, parameterizing a family
of curves C → S and a line bundle L on C relatively ample over S, is algebraic, and
that the natural map P → Mallg is smooth and representable.

Remark 5.4.11. Moduli stacks of varieties of higher dimension are often not algebraic.
For instance, the stack parameterizing abstract K3 surfaces is not algebraic (see
Example C.7.15); on the other hand, there is an analytic stack parameterizing K3
surfaces (see [BHPV04, §VII.12]) and an algebraic stack parameterizing polarized
K3 surfaces (i.e., K3 surfaces with a primitive ample line bundle).

5.4.3 Algebraicity and boundedness of Mg,n

Consider the inclusions of prestacks

Mg,n ⊆ Mg,n ⊆ Mss pre ≤nodal

g,n ⊆ Mg,n ⊆ Mg,n ⊆ Mall
g,n , (5.4.12)

where Mg,n (resp., Mss g,n , Mg,n , and Mg,n

pre ≤nodal
) denotes the full subcategory of
Mg,n consisting of n-pointed families (C → S, σ1 , . . . , σn ) of stable curves (resp.,
all

semistable, prestable, and nodal curves).

Corollary 5.4.13. The sequence of inclusions in (5.4.12) are open immersions,
and each prestack is an algebraic stack locally of finite type over Spec Z.
Proof. Theorem 5.4.6 establishes that Mall g,n is algebraic and locally of finite type.
Openness of the stable (resp., semistable, prestable, nodal) locus was proved in
Proposition 5.3.24, and this implies that each inclusion is an open immersion.
At this point, we know the following properties of Mg,n .
Theorem 5.4.14. Assuming that 2g − 2 + n > 0, the stack Mg,n is a non-empty,
quasi-compact, and Deligne–Mumford stack smooth over Spec Z such that Mg,n ×Z k
has pure dimension 3g − 3 + n for every field k.

237
Proof. By Corollary 5.4.13, Mg,n is an algebraic stack of finite type over Spec Z.
As there are smooth curves of any genus (Exercise 5.1.7), Mg,n is non-empty. For
the boundedness of Mg,n (i.e., finite typeness or equivalently quasi-compactness),
we will appeal to the fact that if (C, p1 , . . . , pn ) is an n-pointed stable curve over a

⊗3
field k, then L := (ωC/k (p1 + · · · pn ) is very ample (Exercise 5.3.17). Let P (t)
be the Hilbert polynomial of C ,→ PN k embedded via L; this is independent of
[C, pi ] ∈ Mg,n . Consider the closed subscheme

H ⊆ HilbP (PN N n
Z /Z) × (P )

parameterizing pairs (C ,→ PN , p1 , . . . , pn ) such that pi ∈ C. Since HilbP (PN

Z /Z) is
a projective scheme (Theorem 1.1.2) and in particular quasi-compact, so is H. The
image of the forgetful morphism

H → Mall
g,n [C ,→ PN , p1 , . . . , pn ] 7→ [C, p1 , . . . , pn ]

contains Mg,n , and we conclude that Mg,n is quasi-compact.

To see the final assertions, we invoke each part of Proposition 5.3.25 characterizing
automorphisms, deformations, and obstructions of stable curve. This is analogous
to our proof of Proposition 3.7.7 asserting the same properties for the stack Mg
of smooth curves. Let (C, pi ) be an n-pointed stable curve of genus g. Since
Ext0OC (ΩC ( i pi ), OC ) = 0, the Lie algebra of Aut(C, pi ) is trivial and Aut(C, pi )
P
is a finite and reduced group scheme. By the Characterization Pof Deligne–Mumford
stacks (3.6.4), Mg,n is Deligne–Mumford. Since Ext2OC (ΩC ( i pi ), OC ) = 0, there
are no obstructions to deforming stable curves, and the Infinitesimal Lifting Criterion
(3.7.1) implies that Mg,n is smooth over Spec Z. Finally, P since the Zariski tangent
space of [C, pi ] ∈ Mg,n ×Z k is bijective to Ext1OC (ΩC ( i pi ), OC ) and this vector
space has dimension 3g − 3 + n, we conclude that Mg,n → Spec Z has relative
dimension 3g − 3 + n.
Exercise 5.4.15. Show that Mg is algebraic by explicitly presenting it as a quotient
stack of a locally closed subscheme of the Hilbert scheme.
Hint: Follow the proof of Theorem 3.1.17.
Exercise 5.4.16. Show that each of the stacks Mss g,n , Mg,n , Mg,n
pre ≤nodal
, and Mall
g,n
are not quasi-compact.
Hint: Use the presence of rational bridges of arbitrary length.
Remark 5.4.17. There are various other open substacks of Mall
g parameterizing
particular classes of curves such as

MDM
g = {curves with finite, reduced automorphism group}
MCM
g = {Cohen–Macaulay curves}
Mggeom-red = {geometrically reduced curves}
MGor
g = {Gorenstein curves}
Mlci
g = {local complete intersection curves}
Miso
g = {curves with isolated singularities};

see [SP, Tags 0DST–0E0K]. The open substack MDMg is identified with the maximal
open Deligne–Mumford substack of Mg . For any g, there are curves in Mall
all
g ∖Mg
DM

with positive dimensional and non-reduced automorphism groups. While Mg,n isall

not smooth over Spec Z, the open substack Mlci

g ∩ Mg is smooth [SP, Tag 0DZX].
iso

238
The open substack Mg ⊆ Mall g is not dense: Mumford provided examples of reduced,
irreducible proper curves that do not deform to a smooth curve [Mum75b]. However,
we show in Proposition 5.7.26 that Mg is dense in Mg , and it is also true that Mg
is dense in Mlci
g ∩ Mg [SP, Tag 0E86].
iso

Exercise 5.4.18 (not important). Show that Mall g → Spec Z satisfies the existence
part of the valuative criterion for properness but that it is not universally closed.
Hint: Consider a closed substack of Mall 1
g × AZ consisting of the disjoint union of
([Cn ], n) over positive integers n, where Cn is the nodal union of a genus g curve
and a tree of P1 ’s with n nodes. The argument should also show that Mpre g is not
universally closed over any field.

5.5 Stable reduction and the properness of Mg,n

Algebraic curves were created by God and algebraic surfaces by the Devil.

Max Noether

In this section, we discuss Deligne and Mumford’s celebrated theorem that Mg,n
is proper [DM69]. The key ingredient is the stable reduction theorem.

Theorem 5.5.1 (Stable Reduction). Let R be a DVR with K = Frac(R), and set
∆ = Spec R and ∆∗ = Spec K. If (C ∗ → ∆∗ , σ1∗ , . . . , σn∗ ) is a family of n-pointed
stable curves, then there exists an extension of DVRs R → R′ and an n-pointed
family (C → ∆′ = Spec R′ , σ1 , . . . , σn ) of stable curves extending the base change of
(C ∗ → ∆∗ , σ1∗ , . . . , σn∗ ) to K ′ = Frac R′ .

While the theorem holds for any DVR, we give a complete proof in this section
only in characteristic 0, following [KKMS73, Ch. II], [HM98, §3.C], and [ACG11,
§X.4].
Remark 5.5.2. In characteristic 0, the stable reduction theorem was first stated in
[MM64b, Lem. A] with a proof given in [May69]. The general version first appeared
in Deligne and Mumford’s seminar paper [DM69]. Their proof relied on semistable
reduction for abelian varieties, which had been established in [SGA7-I, SGA7-II], by
embedding the generic fiber C ∗ into its Jacobian. Gieseker offerered a different proof
using GIT [Gie82], which as expressed in [SP, Tag 0C2Q] “is quite an amazing feat:
it seems somewhat counterintuitive that one can prove such a result without ever
truly studying families of curves over a positive dimensional base.” Later arguments
due to Artin–Winters [AW71] and Saito [Sai87] follow roughly the same strategy
as we provide here but require additional techniques in positive characteristic; see
Remark 5.5.10.

5.5.1 Proof of Stable Reduction

Throughout, we use the notation that ∆ = Spec R for a DVR R defined over Q,
∆∗ = Spec K where K = Frac(R), t ∈ R is a uniformizer, and 0 = (t) ∈ Spec R is the
unique closed point. We are given an n-pointed stable family (C ∗ → ∆∗ , σ1∗ , . . . , σn∗ )
of curves of genus g.

239
 Proof strategy

① Reduce to the case where C ∗ → ∆∗ is smooth.

② Construct a flat extension C → ∆.
③ Replace C with a resolution of singularities such that the reduced
central fiber (C0 )red is nodal.
④ Take a ramified base extension ∆′ → ∆ such that the normalization of
C ×∆ ∆ has reduced central fiber.
⑤ Arrange that the marked points are smooth and distinct.
⑥ Contract rational tails and bridges in the central fiber.

It is useful to keep examples in mind while reading the proof. For a very simple
example, consider the family of elliptic curves (C ∗ → ∆∗ , σ ∗ ) defined by the equation

y 2 z = x(x − z)(x − tz) (5.5.3)

in P2 ×∆∗ and the section σ ∗ (t) = [0, 1, 0]. In this case, the stable limit is transparent
from our description of C ∗ : the family (C → ∆, σ), where C ⊆ P2 × ∆ is defined
by (5.5.3) and σ(t) = [0, 1, 0], is a stable family extending (C ∗ → ∆∗ , σ ∗ ); see
Figure 0.7.3. The stable limit C0 is the nodal cubic y 2 z = x2 (x − z). Additional
examples are given in the proof, and more involved examples are described in §5.5.2.

Proof of Stable Reduction (5.5.1) in characteristic 0.

Step 1: Reduce to the case where C ∗ → ∆∗ is smooth. If C ∗ has δ nodes, then after
replacing K and R with extensions, we can arrange that the jth node is given by a
K-point n∗j ∈ C ∗ whose preimage under the normalization Ce∗ → C ∗ consists of two K-
points qj∗ and rj∗ . We call (Ce∗ → ∆∗ , σi∗ , qj∗ , rj∗ ) the pointed normalization, and we let
(Cek∗ → ∆∗ , qkl
∗
) be the pointed connected components, where {qkl ∗
} = {σi∗ , qj∗ , rj∗ } ∩ Cek∗ .
Since C is stable, each (Cek → ∆ , qkl ) is also stable (Exercise 5.3.6).
∗ ∗ ∗ ∗

Assuming that stable reduction holds when the generic fiber C ∗ is smooth, we can
construct stable families (Cek → ∆, qkl ) extending (Cek∗ → ∆∗ , qkl ∗
) after replacing K
and R with extensions. For each j = 1, . . . , δ, we use a Ferrand Pushout (B.4.1) to
glue the sections qil and qi′ l′ corresponding to qj∗ and rj∗ . By the étale local structure
of smooth morphisms (Proposition A.3.4) and the étale local nature of pushouts
(Proposition B.4.8(3)), there is an étale neighborhood of the pushout of the form

 / A1 ⨿ A1
a
0⨿0
Spec R ⨿ Spec R R R
A=R R[x] × R[y]
where R×R

= {(f, g) ∈ R[x] × R[y] | f (0) = g(0)}

 
Spec R / Spec A, = R[x, y]/(xy).

The sections qil and qi′ l′ are glued to a node. This produces an n-pointed family
(C → ∆, σi ) of nodal curves with δ additional sections picking out the nodes. The
pointed normalization of the central fiber C0 is the disjoint union of the central fibers
(Cek )0 , and it follows from Exercise 5.3.6 that (C0 , σi (0)) is stable. We conclude that
(C → ∆, σi ) is a stable family.

240
 gluefandi

F
E

I
d Y
? q
111 I
Figure 5.5.3: (Step 1) Consider a family C ∗ → ∆∗ of stable curves with a single
node, e.g., a node degenerating to a cusp locally defined by y 2 = x3 + tx2 . The
stable limit is obtained by normalizing along the nodes, extending the 2-pointed
family, blowing up where the two sections q and r intersect, and then gluing
the proper transforms qe and re. If the normalization Ce∗ = C × ∆∗ is a constant
family, then the stable limit is the nodal union of C and a rational nodal curve.

Alternatively, using that Mg,n ⊆ Mg,n is dense (Theorem 5.7.30) Exercise 4.6.7
asserts that it suffices to check the valuative criterion for properness under the
condition that ∆∗ → Mg,n factors through Mg,n .

⊗3
Step 2: Find a flat extension (C → ∆, σi ). Using that ωC ∗ /∆∗ ( i σi∗ ) is very
P
ample (Proposition 5.3.23), we may embed C ∗ as a closed subscheme of PN × ∆∗ .
By the Flatness Criterion over Smooth Curves (A.2.2), the scheme-theoretic image
C of C ∗ ,→ PN × ∆ is flat over ∆ (as the closure doesn’t introduce any embedded
points in the central fiber). This gives a family of curves C → ∆ extending C ∗ → ∆∗ .
(This is the same argument we used in Proposition 1.4.2 to show the properness of
the Hilbert scheme using the valuative criterion.) The sections σi∗ : ∆∗ → C ∗ extend
to sections σi : ∆ → C by the Valuative Criterion of Properness (A.4.5).
At this stage, the central fiber C0 may be very singular and the marked points
STEP 2
σi (0) ∈ C0 may be singular and non-distinct.

pi p
C E IPNxD
Ct ? Pz
C

3 7

E o
Figure 5.5.4: (Step 2) The closure C := C ∗ is a flat family over ∆. The central fiber
C0 may be generically non-reduced with embedded points; the numbers 3, 2, and
7 indicate the multiplicity of the irreducible components. The marked points
σi (0) ∈ C0 may be singular and non-distinct.

Step 3: Replace C with a resolution of singularities to arrange that the reduced central
fiber (C0 )red is nodal. By combining Existence of Resolutions (B.2.1) and Existence
of Embedded Resolutions (B.2.3), there is a projective birational morphism Ce → C
from a regular scheme which is an isomorphism over ∆∗ and such that the central
fiber Ce0 has set-theoretic normal crossings, i.e., (Ce0 )red is nodal. By the Flatness
Criterion over Smooth Curves (A.2.2), Ce → ∆ is flat. We then replace C with Ce and
the sections σi with their strict transform.

241
 YEP blowup blowup blowup
E
Z
2 3
3 2 6

C E E E3
Eng Ez
e E
yp blowup blowup blowup
Ez
2

3 2 3
2 6
Figure 5.5.5: (Step 3) Suppose C → ∆ = k[t](t) is a generically smooth family
degenerating to a cusp y 2 = x3 in the central fiber such that the local equation
in C around the singular point is y 2 = x3 + t. We repeatedly blow up the singular
point in the central fiber using local coordinates x, y on the original surface and
e, ye on the new surface, where in the first chart of the blowup x
x e = x, ye = y/x
with exceptional divisor x e = 0, while in the second chart, xe = x/y, ye = y with
exceptional divisor ye = 0.

➝ For the first blowup, the preimage of the singularity in the first chart is
given by x
e2 (e e) and in the second chart by ye2 (1 − x
y2 − x e3 ye). The exceptional
divisor E1 has multiplicity 2. The normalization C of C has genus g − 1.
e
➝ The second blowup has charts defined by x xye3 − 1) and x
e3 (e e2 ye3 (e e), and
y−x
the new exceptional divisor E2 has multiplicity 3.
➝ The final blowup has charts x y − 1) and x
e6 ye3 (e e), and the new
e2 ye6 (1 − x
exceptional divisor E3 has multiplicity 6. The central fiber is non-reduced
and its reduction is nodal.

Step 4: Take a ramified base extension ∆′ = Spec R′ → Spec R = ∆ such that the
central fiber of the normalization of C ×∆ ∆′ becomes reduced and nodal. This is the
most difficult step and is where the characteristic 0 assumption is used. The argument
can be viewed as a version of Abhyankar’s lemma on tame ramification (see [SGA1,
§XIII.5] and [SP, Tag 0EXT]). Since C is regular and t ∈ R is a nonzerodivisor, the
central fiber C0 is Cohen–Macaulay, and thus has no embedded points. On the other
hand, (C0 )red is an effective Cartier divisor and thus is locally cut out by a single
equation. For every p ∈ C0 , we can find an étale neighborhood and local coordinates
x, y such that the map C → ∆ is given explicitly by:

– if p ∈ (C0 )red is a smooth point, then (x, y) 7→ xa and the multiplicity of the
irreducible component of C0 containing p is a, and
– if p ∈ (C0 )red is a node , then (x, y) 7→ xa y b and the two components of C0
containing p have multiplicities a and b. If p ∈ (C0 )red is a non-separating node
(i.e., C0 ∖ p is connected), then a = b.

Let N be the least common multiple of the multiplicities of the irreducible components
of C0 . After replacing R with an extension, we can assume that R contains a primitive
N th root of unity ρ. Let ∆′ = Spec R′ → Spec R = ∆ be a totally ramified extension
of DVRs of degree N (i.e., the image of the uniformizer t ∈ R is t′N for a uniformizer
t′ ∈ R′ ). Let C ′ := C ×∆ ∆′ and p′ ∈ C ′ be the unique preimage of p, and let Ce′ be
the normalization of C ′ .
If p ∈ (C0 )red is smooth, then C ′ is defined étale locally by xa = tN near p′ . Since

242
R is characteristic 0, there is a factorization

a−1
Y
a N
x −t = (x − (ρi t)N/a )
i=0

into distinct factors. (If the characteristic were positive and divided N , then C ′
would be non-reduced.) In the normalization Ce′ , the point p has a preimages, each
etale locally defined by x = ρi tN/a . Each preimage is a smooth point in both the
total family Ce′ and the central fiber Ce0′ . On the other hand, if p ∈ (C0 )red is a node,
then C ′ is defined étale locally by xa y b = tN near p′ . Let d = gcd(a, b). If d > 1,
there is a factorization (again using the characteristic 0 hypothesis)

d−1
Y
xa y b − t N = (xa/d y b/d − (ρi t)N/d ).
i=0

Étale locally near p, the normalization factors as Ce′ → C ′′ → C ′ such that p has
d preimages in C ′′ , each described by xa/d y b/d − (ρi t)N/d . Unless d = a = b, the
central fiber C0′′ is still non-reduced. The reduction (C0′′ )red is nodal at each preimage
of p but now the multiplicities of the two branches are relatively prime. Therefore,
we can assume that gcd(a, b) = 1. After possibly exchanging x and y, we can write
1 = αa − βb for positive integers α and β. The normalization of the integral domain
R[x, y]/(xa y b − tN ) is given by

x 7→ ub
R[x, y]/(xa y b − tN ) ,→ R[u, v]/(uv − tN/(ab) ),
y 7→ v a ,

where u = tαN/b /(xβ y α ) and v = xβ y α /tβN/a . The central fiber Ce0′ is reduced with a
node at a unique preimage of p. If N = ab, Ce′ is smooth at this preimage; otherwise
Ce′ has an An−1 -singularity where n = N/(ab).

We have now arranged the central fiber to be reduced and nodal, but we have
also introduced An−1 -singularities (i.e., xy − tn ) into the total family. Repeatedly
blowing up each An−1 -singularity replaces the singularity with the nodal union of
⌊ n2 ⌋ smooth rational curves; this explicitly describes the minimal resolution of Ce′ as
in Theorem B.2.2. We now replace C with the minimal resolution so that the total
family C is regular and the central fiber C0 is reduced and nodal.

243
 STEP 4
E E E E's
basechangeby
tht andthen
E normalize É
Ez É
Z Es
E
3 y
6
Figure 5.5.6: (Step 4) Continuing from Figure 5.5.5, we base change C → ∆ =
Spec k[t](t) by ∆′ = Spec k[t](t) → Spec k[t](t) = ∆, given by t 7→ t6 , and then
normalize. The central fiber is now reduced and nodal. Each preimage of E1 and
E2 are smooth rational curves, while the preimage E3′′ of E3 is a genus 1 curve.
step
To understand this description, it is convenient to break the base change into the
composition of the normalized base change by t 7→ t2 and the normalized base
change by t 7→ t3 , which we describe in Example 5.5.13.

he

111111
contractE

11111
blowup blowup
Step 5: Arrange that the marked points σi (0) ∈ C0 are smooth and distinct. After
repeatedly blowing up closed points in the central fiber C0 where the marked points
σi (0) are singular or collide, the strict transform ofethe sections become distinct
and smooth points of the central fiber. After replacing C with the blowup and the
STEP
sections σi with their strict transform, 5 a prestableE family (C → ∆, σi ) with
we have
regular total family.
E

blowup STEP blowup 6 contracte

B f
E
E E E E E

Figure 5.5.7: (Step 5) Consider a constant

contract each family C × ∆ → ∆ with sections locally
, −t2 , 4t).tail
defined by (σ1 , σ2 , σ3 ) = (t2rational After blowing up twice, the sections become
disjoint but the central fiber is not stable as the exceptional component E1 ∼ = P1
only has one node and one marked point. As explained in the next step, the
stable limit is obtained by contracting the rational bridge E1 .

T
genus
T
genus
Step 6: Contract rational tails and bridges in the central fiber. The central fiber C0
of the prestable family (C → ∆, σi ) will be stable, unless there are rational tails and
bridges (see Definition 5.3.11). If E ⊆ C0 is a rational tail or rational bridge with a
marked point, then E 2 = −1, while if E is a rational bridge without a marked point,
then E 2 = −2 (see Remark 5.3.15). By Exercise 5.3.16, contracting each rational
tail and bridge yields a morphism C → C ′ of families of nodal curves over ∆ . Letting
σi
σi′ : ∆ −→ C → C ′ , the n-pointed family (C ′ → ∆, σi′ ) is now stable! Alternatively,
we can construct the stable family (C ′ → ∆, σi′ ) using the Stable Contraction of a
Prestable Family (5.6.6).

244
 E E

STEP 6

E Es
E contracteach
E rationaltail
Ei
Ea
E
T
genus

Figure 5.5.8: (Step 6) Continuing from Figure 5.5.6, the contraction of each rational
tail in red produces a stable family. The central fiber is the nodal union of the
normalization C e of the original central fiber (with a cusp) and an elliptic curve
E.

If we proceed with the six-step procedure above, but stop in Step 6 after contract-
ing only rational tails (but not rational bridges), then Castelnuovo’s Contraction
Theorem (B.2.6) implies that the total family is still regular. This important variant
is called semistable reduction.

Theorem 5.5.9 (Semistable Reduction). Let R be a DVR with K = Frac(R), and

set ∆ = Spec R and ∆∗ = Spec K. If (C ∗ → ∆∗ , s∗1 , . . . , s∗n ) is a family of n-pointed
smooth curves, then there exists an extension of DVRs R → R′ and an n-pointed
family (C → ∆′ = Spec R′ , s1 , . . . , sn ) of semistable curves with regular total family
C extending the base change of (C ∗ → ∆∗ , s∗1 , . . . , s∗n ) to K ′ = Frac R′ .

Remark 5.5.10 (Proof in characteristic p). Our proof of stable reduction fails in
Step 4 if the residue field of R has characteristic p > 0 and any of the multiplicities
of the components of the central fiber are divisible by p. A different approach is
needed in positive characteristic. After resolving the singularities of C and arranging
that (C0 )red is nodal (as we do in Step 3 above), Artin and Winters arrange that the
l-torsion of Pic(CK∗
′ ) is isomorphic to (Z/lZ)
2g
for a sufficiently large prime l ̸= p,
and they show that this magically forces the central fiber to be reduced and nodal!
See [AW71], [Liu02, §10.4], and [SP, Tag 0C2P].

Example 5.5.11 (Base changes are necessary). The stable reduction of the cusp
worked out in Figures 5.5.5, 5.5.6 and 5.5.8 demonstrates the necessity of allowing
for extensions of the DVR. Indeed, after Step 3, we have a generically smooth
family C → ∆ with a regular total family C and with a non-reduced central fiber C0 .
Suppose that there is a stable family C ′ → ∆ such that C|∆∗ ∼= C ′ |∆∗ . After resolving
the An -singularities in C , there is semistable family C → ∆ with C ′′ regular and
′ ′′

C0′′ semistable (and in particular reduced). Since C and C ′′ are birational, by taking
a resolution of singularities of the closure of the graph of a rational map C 99K C ′ ,
there is a regular two-dimensional scheme D and birational morphisms D → C and
D → C ′′ over ∆, each which factors as a composition of blowups by Factorization of
Birational Maps (B.2.4). However, any blowup of C will have a non-reduced central
fiber, while any blowup of C ′′ will have a reduced central fiber, and this yields a
contradiction.

245
5.5.2 Explicit stable reduction

In applications to the geometry of curves, it is often essential to explicitly describe

the stable limit. While the proof of Stable Reduction offers a strategy, additional
care is needed to get an explicit handle. The main challenge is determining the
normalization C e ′ of the base change C ′ = C ×∆ ∆′ by a totally ramified extension
∆ → ∆ in Step 4. It is often simpler to factor ∆′ → ∆ as a composition of prime
′

order base changes, because it is straightforward to determine the ramification locus

of prime order normalized base changes Ce′ → C.

Proposition 5.5.12. Let p be a prime integer. Let C → ∆ = Spec R be a generically

smooth family of curves, where R is a DVR of characteristic 0 containing a pth root
of unity. Assume that P the reduced central fiber (C0 )red is nodal. As a divisor on C,
we may write C0 = mi Di where mi is the multiplicity of the irreducible component
Di . Let ∆′ = Spec R′ → Spec R = ∆ be a totally unramified extension of degree
p, and set C ′ := C ×∆ ∆′ with normalization Ce′ . Then Ce′ → C is ramified over a
divisor Di if and only if mi is relatively prime to p. If q ′ ∈ Ce0′ is a preimage of a
point q ∈ Di which is a smooth point of (C0 )red , then the multiplicity of Ce0′ around
q ′ is mi /p if p divides mi and mi otherwise.

Proof. The point q ∈ C has an étale neighborhood with local coordinates x and y
such that C → ∆ is described as (x, y) 7→ xmi . The base change C ′ is described étale
locally by Spec k[x, y, t]/(xmi − tp ) near q ′ . If p divides mi ,

p−1
Y
x mi p
−t = (xmi /p − ρi t), where ρ is a pth root of unity,
i=0

and its normalization has p smooth components. The central fiber Ce0′ has multiplicity
mi /p near each of the p preimages of q. If mi = pm′i + r with 0 < r < p, then
k[x, y, t]/(xmi − tp ) is an integral domain with normalization

′
k[x, y, t′ ]/(xr − t′p ), where t′ = t/xmi .

′
Thus, Ce′ → C is ramified at q. The central fiber Ce′ → ∆ is defined by t = t′ xmi and
the multiplicity at the unique preimage of q can be computed as dimk k[x, t′ ]/(xr −
′
t′p , t′ xmi ). Since the union of t′i xj for i = 0, . . . , p − 1 and j = 0, . . . , m′i − 1 and
′ ′
xmi , . . . , xmi +r−1 forms a basis, the multiplicity is mi .

Example 5.5.13 (Stable reduction of a cusp). Let C → ∆ = Spec k[t](t) be a

generically smooth family degenerating to a cusp y 2 = x3 in the central fiber C := C0
such that the local equation in C around the singular point is y 2 = x3 + t. As
described in Figure 5.5.5, after three blowups of C, we obtain a family C ′ → ∆ such
that (C0′ )red is the nodal union of the normalization C,
e which has multiplicity 1, and
three exceptional components E1 , E2 , and E3 , which have multiplicities 2, 3, and 6.

246
 Expect
C
E Es E Es E Es
E Ei Ei
3blowups Eat Eat
yp E
Ez
E E
Z
Es
E
3 3
6 3 y
genus

Figure 5.5.14: The numbers indicate the multiplicity of a component; if no number is

given, the multiplicity is 1. The thickened components indicate the ramification
locus under the normalized base change.

Using Proposition 5.5.12, the ramification locus of the first normalized base
change t 7→ t2 is the union of C e and E2 . The preimage of E1 is the disjoint union
E1 ⨿ E1 of two smooth rational curves. The preimage E3′ of E3 is a curve which is
′ ′′

a two-to-one cover of E3 = P1 ramified at two points, and thus E3 = P1 but now

with multiplicity 2. For the second normalized base change, the ramification locus
is Ce ⨿ E ′ ⨿ E ′′ (again by Proposition 5.5.12). The preimage of E2 is the union of
1 1
three smooth rational curves, while the preimage E3′′ of E3′ is a smooth curve which
is a three-to-one cover of E3′ = P1 ramified at three points, each with ramification
index two. The genus gE3′′ of E3′′ can be computed to be 1 using Riemann–Hurwitz
(5.7.4): 2gE3′′ − 2 = 3(2(0) − 2) + 3(2) = 0. The final step contracts the five rational
tails as pictured in Figure 5.5.8, and the stable limit is the nodal union Ce ∪ E ′′ of a
3
smooth genus g − 1 curve C (the normalization of the original central fiber) and a
e
genus 1 curve.
Remark 5.5.15. Precisely which elliptic curve E3′′ appears in the stable limit and
how does the stable limit depend on the choice of degeneration? The deformation
space of a cusp is y 2 = x3 + a1 (t)x + a0 (t) and, in other words, we are asking how
the stable limit depends on ai (t). For instance, what happens when the total family
of the surface is singular (e.g., y 2 = x2k+1 + t2 )? These questions are addressed in
detail in [HM98, §3.C].
Exercise 5.5.16 (good practice).
(a) Find the stable limit of a generically smooth family degenerating to a tacnode
y 2 = x4 .
(b) More generally, show that the stable limit of a generically smooth family
degenerating to a A2k singularity y 2 = x2k+1 (resp., y 2 = x2k+2 ) is the nodal
union of a genus g − k curve and a genus k hyperelliptic curve attached
at a Weierstrass point (resp., at two Weierstrass conjugate points). Here a
Weierstrass point of a hyperelliptic curve H is a ramification point under the
double cover H → P1 , while two points are Weierstrass conjugate points if
their union is a fiber of H → P1 .
Example 5.5.17 (Stable reduction of a double conic). Consider a generically smooth
family C → ∆ = Spec k[t](t) , where C ⊆ P2 × ∆ is defined by F 2 + tG where F is
a smooth conic and G is a smooth quartic. The central fiber is the double conic
defined by F 2 . The total space has an A1 -singularity with local equation x2 + yt
at each of the 8 intersection points p1 , . . . , p8 of F ∩ G. Each σi is resolved with a
single blowup with an exceptional divisor Ei P = P1 of multiplicity 1. This gives a
family C2 → ∆ where the central fiber is 2C + i Ei as a divisor. We then take the

247
 C
E Es E Es E Es
E Ei
3blowups Eat É Eat
yp
Ez
E É
2
Ea
E
3
g 3 y
normalization C3 of the base change C2 ×∆ ∆′ by the ramified covergenus
∆′ → ∆, t 7→ t2 .
By Proposition 5.5.12, C3 → C2 is ramified over the disjoint union of the Ei ’s. The
Double conic
preimage of C is a two-to-one cover of P1 branched over the 8 points σi , and hence
is smooth hyperelliptic genus 3 curve.

C VLF C H H

8blowups E E't E Ei
contract

Es Es
Z Z

Figure 5.5.18: Stable reduction of a generically smooth family F 2 + tG degenerating

to a double conic C = V (F 2 ).

Meta-exercise 5.5.19. Read the exposition of [HM98, §3.C] and do the exercises.

5.5.3 Properness of Mg,n

Stable Reduction (5.5.1) implies the existence part of the valuative criterion for
properness for Mg,n . We must also show that the stable limit is unique, i.e., Mg,n
Z
is separated.
Proposition 5.5.20. Let R be a DVR with K = Frac(R), and set ∆ = Spec R
and ∆∗ = Spec K. If (C → ∆, σ1 , . . . , σn ) and (D → ∆, τ1 , . . . , τn ) are families of
n-pointed stable curves, then every isomorphism α∗ : C ×∆ ∆∗ → D ×∆ ∆∗ over ∆∗
compatible with the restriction of the sections (i.e., τi∗ = α∗ ◦ σi∗ ) extends to a unique
isomorphism α : C → D over ∆ with τi = α ◦ σi .
Proof. For simplicity, we handle the case without marked points (n = 0). We claim
that we can reduce to the case where the generic fiber C ∗ ∼ = D∗ is smooth over ∆∗ ,
where C := C ×∆ ∆ . We will employ the same strategy as in the reduction to
∗ ∗

the generically smooth case in Step 1 of the proof of Stable Reduction (5.5.1). By
flat descent, it suffices to construct α : C → D after an extension of DVRs ∆′ → ∆;
indeed, an isomorphism α′ : C ×∆ ∆′ → D ×∆ ∆′ will satisfy the cocycle condition (by
the separatedness of D → ∆) and thus descend to an isomorphism α. Therefore, we
may assume that each node of C ∗ ∼ = D∗ is given by a K-point whose preimage under
the normalization consists of two K-points. Each node extends to sections ∆ → C
and ∆ → D. The pointed normalizations Ce and D e are each disjoint unions of pointed
stable families. If Ce∗ → D e ∗ extends to an isomorphism α e then α descends
e : Ce → D,
to an isomorphism C → D. Alternatively, using that Mg,n ⊆ Mg,n is dense (proven
later in Theorem 5.7.30), it suffices to check the valuative criterion for separatedness
under the condition that ∆∗ → Mg,n factors through Mg,n (Exercise 4.6.7).
Let Ce → C and D e → D be the minimal resolutions (Theorem B.2.2). Let Γ be
the closure of the image of (id, α∗ ) : C ∗ → Ce ×∆ D,
e and let Γ e → Γ be its minimal
resolution. This gives a commutative diagram

: Ce /C

$ (5.5.21)
Γ
e :∆
$ e /D
D

248
 We will give two arguments that there is an isomorphism α : C → D extending
α∗ . First, the local structure of C (resp., D) around a node in the central fiber is an
An singularity of the form xy = tn+1 , where t ∈ R is a uniformizer. The preimage
of each node under Ce → C (resp., D e → D ) is a chain E1 ∪ · · · ∪ En of rational
bridges with Ei = −2. Since C and D are families of stable curves, they each have
2

no smooth rational −1 curves and thus neither do Ce and D. e By the Factorization of

Birational Maps (Theorem B.2.4), both Γ → C and Γ → D
e e e e are the compositions of
finite sequences of blowups at closed points. Since Γ e → Γ is a minimal resolution, Γ e
has no smooth rational −1 curves that get contracted under both Γ e → Ce and Γe → D.
e
We claim that both Γ e → Ce and Γe→D e are isomorphisms. This would finish the
proof as both C and D are obtained by contracting the smooth rational −2 curves
of Ce ∼
= D.
e
To see the claim, suppose for instance that Γ e → Ce is not an isomorphism. Then
there is a smooth rational −1 curve E ⊆ Γ e not contracted under Γ e → D.e Let
EDe ⊆ D e be its image. Since blowing up only decreases the self-intersection number
(indeed, if we write the pre-image of EDe in Γ e as E + F , then the projection formula
implies that EDe = E · (E + F ) = E + E · F ), we have that ED
2 2 2
≥ E 2 = −1. On
the other hand, the Hodge Index Theorem for Exceptional Curves (B.2.5) implies
that ED e ≤ −1. Hence ED
2
e = −1. But ED
2
e is not a smooth rational −1 curve so it
must be singular, and one of the blowups in the composition Γ e→D e must be at a
singular point of EDe . But this implies that exceptional locus F of Γ → D
e e intersects
E non-trivially, and thus EDe ≥ E + 1, a contradiction.
2 2

Alternatively, we could argue as follows. The birational maps Γ e → Ce and Γe→D e

of smooth projective surfaces are isomorphisms in codimension 2. As the relative
dualizing sheaves are line bundles, there are identifications of the pluricanonical
sections
e ω ⊗k ) ∼
Γ(C, = Γ(Γ,e ω ⊗k ) ∼ e ω ⊗k )
= Γ(D,
C/∆
e Γ/∆
e D/∆
e

for each nonnegative integer k (c.f., [Har77, Thm. II.8.19]). Using that C and D are
the stable contraction of the families Ce and D
e over ∆ (Theorem 5.6.6) arising as the
Proj of the graded ring of pluricanonical sections, we obtain an isomorphism
∼ ∼ ∼
M M
C → Proj e ω ⊗k ) →
Γ(C, Proj e ω ⊗k ) →
Γ(D, D
C/∆
e D/∆
e
k k

extending α∗ : C ∗ → D∗ . See also [DM69, Lem. 1.12], [ACG11, Lem. 10.5.1], and
[SP, Tag 0E97].

Exercise 5.5.22 (details). Extend the proof above to the case of marked points.

Even though we have only proved Stable Reduction (5.5.1) in characteristic 0,

we state the next result over Z.

Theorem 5.5.23 (Properness of Mg,n and M g,n ). If 2g − 2 + n > 0, the Deligne–

Mumford stack Mg,n is proper over Spec Z. Moreover, there is a coarse moduli space
Mg,n → M g,n , where M g,n is a proper algebraic space over Spec Z.

Proof. Using the Valuative Criterion (3.8.7), Stable Reduction (5.5.1) gives the
existence of limits, while Proposition 5.5.20 gives the uniqueness of a limit. As
Mg,n is separated, the Keel–Mori Theorem (4.4.6) gives a coarse moduli space
Mg,n → M g,n , where M g,n is an algebraic space separated and of finite type over
Spec Z. As Mg,n is proper over Spec Z, so is M g,n .

249
Exercise 5.5.24. Show that the coarse moduli space of Mg,n ×Z Fp is the normal-
ization of M g,n ×Z Fp .

Question 5.5.25 (Open). Is M g,n ×Z Fp the coarse moduli space of Mg,n ×Z Fp ?1

5.6 Contraction, forgetful, and gluing morphisms

Old theorems never die; they turn into definitions.

Edwin Hewitt

In this section, we construct several important morphisms between moduli spaces

of curves.

– (Stable Contraction) There is a morphism Mpre g,n → Mg,n , which maps a

prestable family (C → S, σi ) to a stable family (C st , σist ) by contracting all
rational tails and bridges (Theorem 5.6.6).
– (Forgetful) There is a morphism Mg,n+1 → Mg,n assign an n + 1-pointed
stable family (C → S, σ1 , . . . , σn+1 ) to the stable contraction of the n-pointed
prestable family (C → S, σ1 , . . . , σn ) (Proposition 5.6.10). Moreover, we iden-
tify Mg,n+1 → Mg,n with the universal family (Proposition 5.6.12).
– (Gluing) There is a morphism Mi,k × Mg−i,n−k+2 → Mg,n , gluing a k-pointed
stable family of genus i curves to a n − k + 2-pointed stable family of genus
g − i curves along the final sections, and a morphism Mg−1,n+2 → Mg,n ,
gluing the final two sections of an n + 2-pointed stable family of genus g − 1
(Corollary 5.6.16).

As with the Local Structure of Nodal Families (5.2.25), the biggest challenge is
ensuring that the constructions hold for families over an arbitrary base. We conclude
this section in §5.6.4 with a discussion of boundary divisors and line bundles on
Mg,n .

5.6.1 Contracting rational tails and bridges

Rational tails and bridges of a prestable curve over a field were defined in Defini-
tion 5.3.11 and characterized in Lemma 5.3.12. We first show that rational tails and
bridges can be contracted to a stable curve over a field (Corollary 5.6.3), and then
we extend the construction to families (Theorem 5.6.6).

Proposition 5.6.1 (Contracting a rational tail or bridge). Let (C, pi ) be an n-

pointed prestable curve over a field k, and E be a rational tail or rational bridge.
Then there is a canonical morphism

c : C → C′

contacting E to a point such that c∗ OC = OC ′ and R1 c∗ OC = 0. Moreover, C ′ is

identified with the pushout Spec Γ(E, OE ) ⨿E C, and the formation of c : C → C ′
commutes with field extensions of k.
1 See [Link]
and [Link]

250
Proof. In both cases, we construct C ′ as the Ferrand Pushout (B.4.1)

E ∩ Ec / Ec

⌟
 
Spec k′ / C ′,

where k′ = Γ0 (E, OE ). Since scheme-theoretic unions are pushouts (Exercise B.4.6),

we have that C = E ⨿E∩E c E c . This induces a larger commutative diagram

E ∩ Ec / Ec
 ⌟ 
E /C
 ⌟ c
Spec k′ / C ′,

where c : C → C ′ is the unique map making the diagram commute. Since the top
and outer squares are pushouts, so is the bottom square, i.e., C ′ = E ⨿Spec k′ C.
Since Ferrand pushouts commute with field extension (Proposition B.4.8(4)), so
does the construction of c : C → C ′ .
If E is a rational tail or a rational bridge with a marked point, then E ∩ E c =
Spec k′ = Spec κ(x) for a point x ∈ E ∩ E c (see Lemma 5.3.12), and thus C ′ = E c .
The map c : C → C ′ = E c is the identity on E c and contracts E to the point x ∈ E c
via the structure morphism E → Spec Γ(E, OE ) = Spec κ(x). Consider the short
exact sequence
0 → OC → OE ⊕ OE c → κ(x) → 0.

Applying c∗ yields a long exact sequence

(E, OE ) ⊕ OC ′ → κ(x) → R1 c∗ OC → ix,∗ H1^

0 → c∗ OC → ix,∗ H0^ (E, OE ) → 0,

where ix : Spec κ(x) ,→ C ′ is the inclusion of x. Since H0 (E, OE ) = κ(x) and

H1 (E, OE ) = 0, we see that c∗ OC = OC ′ and R1 c∗ OC = 0.
If E is a rational bridge without marked points, set k′′ = H0 (E ∩ E c , OE∩E c ). By
Lemma 5.3.12, E ∩ E c is either a single point with k′′ a degree 2 separable extension
of k′ or two points with k′′ = k′ × k′ . Since the construction of the pushout is étale
local (Proposition B.4.8(3)), the affine pushouts

/ Spec k′′ [y]

`
0 0
Spec k′′
0
Spec k′ × k′ / Spec k′ [y] × k′ [z]

⌟ and ⌟
   
Spec k′ / Spec{f ∈ k′′ [y] | f (0) ∈ k′ }, Spec k′ / Spec k′ [y, z]/(yz)

are étale neighborhoods of C at the image of E, depending on whether E ∩ E c is one

or two points. In both cases, the image of E is a node. One shows that c∗ OC = OC ′
and R1 c∗ OC = 0 as above using the short exact sequence 0 → OC → OE ⊕ OE c →
OE∩E c → 0. See also [SP, Tags 0E3H and 0E3M].

A curve can also contain a chain of rational tails and bridges of arbitrary length.

251
 L a 2

Figure 5.6.2: Chains of rational tails and bridges

Corollary 5.6.3 (Stable Contraction). Let (C, pi ) be an n-pointed prestable curve of

genus g over a field k such that 2g − 2 + n > 0. Then there is a canonical morphism

c : C → C st ,

called the stable contraction, contacting all rational tails and rational bridges to
points, such that (C st , c(pi )) is an n-pointed stable curve of genus g, c∗ OC = OC st ,
and R1 c∗ OC = 0. Moreover, the formation of c commutes with field extensions of k.
Proof. If E denotes the scheme-theoretic union of all rational tails and bridges, then
H0 (E, OE ) is a finite product of fields. Iteratively applying Proposition 5.6.1 yields
a pushout diagram
E /C

⌟ c
 
Spec H0 (E, OE ) / C st ,

such that c∗ OC = OC st and R1 c∗ OC = 0, with the construction of c commuting

with field extensions. Since (C st , c(pi )) has no rational tails and bridges, it is stable
(Proposition 5.3.14). See also [SP, Tag 0E7Q].

44
a f

si a
Figure 5.6.4: Stable contractions

252
Exercise 5.6.5. Let (C, pi ) be an n-pointed prestable curve over a field k.
(a) Show that the stabilization morphism c : C → C st is the unique morphism
such that (C st , c(pi )) is stable, c∗ OC = OC st , and R1 c∗ OC = 0.
(b) If (C, pi ) is semistable P and L := ωC ( i pi ), show that C st ∼
= Proj d≥0 H0 (C, L⊗d ))
P L
and that L = ∼ c∗ ωC st (
i c(pi )).

The construction of the stable contraction extends to families of prestable curves.

Theorem 5.6.6 (Stable Contraction of a Prestable Family). If (C → S, σi ) is a

family of n-pointed prestable curves of genus g such that 2g − 2 + n > 0, then there
exists a unique morphism c : C → C st over S such that
(1) (C st → S, σist ) is an n-pointed family of stable curves of genus g where σist =
c ◦ σi ;
(2) OC st = c∗ OC and R1 c∗ OC = 0;
(3) the construction of c : C → C st is compatible with base change S ′ → S; and
(4) for each s ∈ S, (Cs , σi (s)) → (Csst , σist (s)) is the stable contraction of rational
bridges and tails as in Corollary 5.6.3.
P
Moreover, if (π : C → S, σi ) is a semistable family, then P ω C/S ( i σi ) is the pull-
back of the relatively ample line bundle L := ωC st /S ( i σist ); in particular, C st ∼ =
ProjS d≥0 π∗ (L⊗d ).
L

Proof. This will be a local-to-global argument. For any s ∈ S, Corollary 5.6.3 yields
the stable contraction cs : Cs → Y0 over κ(s), and satisfies the uniqueness property by
Exercise 5.6.5(a). The key idea of the proof is the following: since cs,∗ OCs = OY0 and
R1 cs,∗ OCs = 0, any infinitesimal deformation of Cs extends uniquely to a deformation
of Cs → Y0 (Exercise C.2.9). Setting Sn = Spec OS,s /mn+1 , this yields compatible
morphisms
C ×S Sn → Yn
over Sn . Artin Approximation (B.5.18) then yields a morphism C ×S S ′ → Y ′ after
an étale cover S ′ → S, which, by the uniqueness properties, descends to a morphism
C → C st .
To make this argument work, we first reduce to the case that S is finite type
over Z using Limit Methods (§B.3). As in the first paragraph, we find compatible
morphisms C ×S Sn → Yn over Sn . The central fibers Cs and Y0 are schemes by
Proposition 4.5.19, thus so are C ×S Sn and Yn . Since Sb := Spec O bS,s is noetherian,
Grothendieck’s Existence Theorem (Corollary C.5.8) yields a projective morphism
Yb → Sb := Spec O bS,s extending Yn → Sn . By the Infinitesimal Criterion for Flatness
(A.2.5), Y → S is flat. A consequence of Grothendieck’s Existence Theorem (C.5.11)
b b
asserts that there is a morphism C ×S Sb → Yb over Sb extending C ×S Sn → Yn .
σi ×S S
Letting τbi be the composition Sb −− −−→ C ×S Sb → Yb , then (Yb → S, b τbi ) is an
b

n-pointed family of curves. Since the central fiber Y0 is stable, Openness of Stability
(5.3.24) implies that (Yb → S,
b τbi ) is a family of stable curves.
Since S is finite type over Z, Artin Approximation (B.5.18) gives an étale
neighborhood (S ′ , s′ ) → (S, s), a family of stable curves (Y ′ → S ′ , τi ), and a
morphism c′ : C ×S S ′ → Y ′ of algebraic spaces over S ′ whose fiber at s′ corresponds
to Cs → Y0 . By Exercise 5.6.8, after replacing S ′ with an open neighborhood of
s′ , we can assume that c∗ OC×S S ′ = OY ′ and R1 c∗ OC×S S ′ = 0, and that this holds
after base change. Therefore, the existence of the desired family of stable curves
will follow from étale descent once we establish uniqueness.

253
 To show uniqueness, let (Y → S, τi ) and (Y ′ → S, τi′ ) be families of stable curves,
let c : C → Y and c′ : C → Y ′ be morphisms over S, and suppose that there is an
∼
isomorphism αs : Ys → Ys′ compatible with cs and c′s such that cs,∗ OCs = OYs and
R1 cs,∗ OCs = 0. We need to show that there exists an open neighborhood U ⊆ S of
s and an isomorphism α : YU → YU′ extending αs , which is compatible with cU and
c′U . By Limit Methods (§B.3), we may assume that S = Spec A is the spectrum of a
noetherian local ring. Letting Y ′′ be the scheme-theoretic image of C → Y ×S Y ′ , it
suffices to show that the projections Y ′′ → Y and Y ′′ → Y ′ are isomorphisms. Since
the scheme-theoretic image commutes with flat base change, we may further assume
that A is complete. By Exercise C.2.9, the restrictions Yn′′ → Yn and Yn′′ → Yn′ to
the base change to Spec A/mn+1 are isomorphisms. A consequence of Grothendieck’s
Existence Theorem (C.5.11) yields the desired isomorphism Y → Y ′ . See also [SP,
Tag 0E8A].
If (C → S, σi ) is a family of semistable curves, then we claim that the natural map
c∗ ωC st /S → ωC/S is an isomorphism. Indeed, since the relative dualizing sheaves are
line bundles, it suffices to show that this map is surjective. Since the relative dualizing
sheaves and the cokernel of the map are compatible with base change, by Nakayama’s
lemma, it suffices to show that the claim holds when S is the spectrum of a field, which
is the assertion in Exercise 5.6.5(b). By the projection formula, ωC st /S ∼ = c∗ c∗ ωC st /S ,
and thus the composition ∼ ∗
ωC st /S = c∗ c PωC st /S → c∗ ωC/S is an isomorphism. It
follows that ωC st /S ( i σist ) ∼= c∗ (ωC/S ( i σi )), which implies the statement as
P
ωC st /S is relatively very ample (Proposition 5.3.21). Alternatively, one can explicitly
construct the stable contraction globally by showing that the L natural maps π∗ (L⊗2 )⊗
⊗d
π∗ (L ) → π∗ (L ⊗d+2
) are surjective for d ≥ 4. Thus, d≥0 π∗ (L
⊗4d
) is a finite
type OS -algebra, and we can define C st := ProjS d≥0 π∗ (L⊗4d ). Since for all d ≥ 0
L

the pushforward π∗ (L⊗4d ) is a vector bundle and its construction commutes with
base change (Proposition 5.3.23), it follows that C → C st is well-defined and C st → S
is a family of stable curves. See [Knu83a, Prop. 2.1] and [ACG11, Prop. 10.6.7].

Corollary 5.6.7 (Stable Contraction Morphism). There is a morphism of algebraic

stacks
Mpre
g,n → Mg,n , (C, pi ) 7→ (C st , c(pi )),

which is the identity on the open substack Mg,n ⊆ Mpre

g,n .

Exercise 5.6.8 (details). Let f : X → Y be a morphism of families of curves over a

noetherian scheme S. Suppose that for a point s ∈ S, the morphism fs : Xs → Ys
satisfies fs,∗ OXs = OYs and R1 fs,∗ OXs = 0. Show that after replacing S with an
open neighborhood of s, f∗ OX = OY and R1 f∗ OX = 0, and that this remains true
after base change by a morphism S ′ → S. See also [SP, Tag 0E88].

5.6.2 The forgetful morphism and the universal family of

Mg,n
We give two consequences of the Stable Contraction Morphism (5.6.7): we show
that there is a map Mg,n+1 → Mg,n which is the stable contraction of forgetting
the last marked point, and that this map is identified with the universal family.

254
 P Pi
R R
Pritt

3 3

P
p
P
R
p Pntl

Figure 5.6.9: In (A), the n+1th point is simply forgotten. In (B), if pn+1 is forgotten,
the curve is no longer stable, and we must contract the rational bridge.

Proposition 5.6.10 (Forgetful Morphism). There is a morphism of algebraic stacks

q
Mg,n+1 → Mpg,n , (C, p1 , . . . , pn+1 ) 7→ (C st , c(p
i 1
), . . . , c(pn )),

where c : C → C st is the stable

P contraction of (C, p1 , . . . , pn ).
Proof. The desired morphism is constructed as the composition
i
Mifg,n+1 → Mpre
g,n → Mg,n ,

where Mg,n+1 → Mpre g,n is the morphism taking an n + 1-pointed stable family
(C → S, σ1 , . . . , σn+1 ) to the n-pointed prestable family (C → S, σ1 , . . . , σn ) and
g,n → Mg,n is the Stable Contraction Morphism (5.6.7).
Mpre

By the Generalized 2-Yoneda Lemma (3.1.25), the identity morphism Mg,n →

Mg,n corresponds to an object of Mg,n over the algebraic stack Mg,n , which in
turn corresponds via descent to an n-pointed family

(Ug,n → Mg,n , σ1univ , . . . , σnuniv )

of stable curves, called the universal family. An object of Ug,n over a scheme S is an
n-pointed family of stable curves (C → S, σi ) with an additional section τ : S → C
(that may land in the nodal locus C → S and may intersection non-trivially with
the sections σi ). Given an n-pointed family of stable curves (C → S, σi ), there is a
morphism S → Mg,n , unique up to unique isomorphism, and a cartesian diagram

C] / Ug,n
Z
σi σiuniv
 
S / Mg,n .

On the other hand, there is the Forgetful Morphism (5.6.10) Mg,n+1 → Mg,n
along with a morphism Mg,n+1 → Ug,n of algebraic stacks taking an n + 1-pointed
stable family (C → S, σi ) to the n+1-pointed family (C st , σ1st , . . . , σnst , c◦σn+1 ), arising
as the Stable Contraction (5.6.6) c : C → C st of the prestable family (C, σ1 , . . . , σn ).

255
 3

P
p P

P p pies
This yields a commutative diagram

Mg,n+1 / Ug,n

(A) (C)
(B)

Mg,n .

(A) P p
3 3

(B) P
pl
o

(C)
P p pies

Figure 5.6.11: Examples of the map Mg,n → Ug,n .

Proposition 5.6.12. The morphism Mg,n+1 → Ug,n is an isomorphism over Mg,n .

In other words, the forgetful morphism Mg,n+1 → Mg,n is the universal family.

Proof. The stacks Mg,n+1 and Ug,n are both proper and representable over Mg,n
by Stable Reduction (5.5.1). Hence, the morphism Mg,n+1 → Ug,n is proper and
representable. For every algebraically closed field k, the induced map Mg,n+1 (k)/∼→
U g,n (k)/∼ on isomorphism classes is bijective. Hence, Mg,n+1 → Ug,n is proper
and quasi-finite, hence finite. On the other hand, Mg,n+1 → Ug,n is birational as
it induces an isomorphism between the open substack of Mg,n+1 parameterizing
pointed curves (C, p1 , . . . , pn+1 ) such that (C, p1 , . . . , pn ) does not contain a rational
bridge and the open substack of Ug,n parameterizing pointed curves (C, p1 , . . . , pn+1 )
such that pn+1 ∈ C is a smooth point that doesn’t coincide with pi for i = 1, . . . , n.
By Theorem 5.4.14, Mg,n+1 is smooth. Since Ug,n → Mg,n has Cohen–Macaulay
fibers and Mg,n is smooth, Ug,n is Cohen–Macaulay. The stack Ug,n is regular
in the open locus where pn+1 ∈ C is a smooth point, and since the complement
of this locus is codimension 2, Serre’s criterion for normality implies that Ug,n is
normal. We thus have a finite birational morphism Mg,n+1 → Ug,n between normal
Deligne–Mumford stacks, which is necessarily an isomorphism.

Remark 5.6.13. One can also explicitly construct the inverse morphism Ug,n →
Mg,n+1 . Let (C → S, σi ) be an n-pointed family of stable curves and τ : S → C be
an additional section defined by an ideal sheaf Iτ . Define the coherent OC -module
K by
(α,β)
0 → OC −−−→ Iτ∨ ⊕ OC (σ1 + · · · + σn ) → K → 0,

256
where α is the dual of the inclusion Iτ ,→ OC and β is the natural inclusion. Define

Ce = ProjS Sym K → C.

With some work, one can prove that τ ∗ (Iτ∨ /OC ) is a line bundle [Knu83a, Lem. 2.2].
The surjection τ ∗ K → τ ∗ (K/OC ) ∼ = τ ∗ (Iτ∨ /OC ) defines a section τe : S → C. e The
cokernel of the injection Iτ ,→ K is identified with OC (σ1 + · · · + σn )| i σi , and the
∨ S

surjections σi∗ K P→ σi∗ OC (σ1 + · · · + σn ) defines sections P σ

ei : S → C.e One checks
that (C → S, σ
e Pz Pz
en , τe) is an n + 1-pointed family of stable curves such that
e1 , . . . , σ
c : Ce → C is the stable contraction
Pna of the n-pointed prestable family (C, eσe1 , . . . , σ
en )
with σi = c ◦ σ ei and τ = c ◦ τe. In many cases, Ce → C can be constructed more
directly as a blow up: for instance, if C → S is a generically smooth family of stable 3
curves over the spectrum of a DVR such that C is regular and τ : S → C is a section
such that τ (0) ∈ C0 is a node, then Ce → C is simply the blow up at τ (0) and τe is
the strict transform P of τ . See [Knu83a, Thm. 2.4] and [ACG11, §X.8].
P P
5.6.3 Gluingpmorphisms
Pml
o
After showing how sections of families of curves can be glued to nodal families
(Proposition 5.6.15), we show that there are well-defined finite morphisms Mi,k ×
Mg−i,n−k+2 → Mg,n and Mg−1,n+2 → Mg,n (Corollary 5.6.16), called gluing
morphisms (or sometimes clutching morphisms).

i
i

q
p

Figure 5.6.14: The nodal gluing of marked points p and q.

Proposition 5.6.15 (Gluing Families Along Sections). Let (C → S, σ, τ ) be a

2-pointed projective family of (possibly disconnected) curves over a scheme S such
that for every point s ∈ S, σ(s) and τ (s) are distinct smooth points of Cs . Then
there is a canonical finite morphism

g : C → C′

of schemes over S such that

(1) C ′ → S is a family of curves with a section ν : S → C ′ such that ν = g ◦σ = g ◦τ
and ν(s) ∈ Cs is a node for all s ∈ S;
∼
(2) the morphism g restricts to an isomorphism C ∖ (σ(S) ∪ τ (S)) → C ′ ∖ ν(S),
and there is an identification of the topological space |C ′ | with the quotient of
|C| under the equivalence relation σ(s) ∼ τ (s) for s ∈ S;

257
 (3) C ′ is identified with the Ferrand Pushout (B.4.1) of σ ⨿ τ : S ⨿ S → C and the
projection S ⨿ S → S; in particular, for every open subset U ⊆ C ′ ,
Γ(U, C ′ ) = {f ∈ Γ(g −1 (U ), C) | σ ∗ f = τ ∗ f };
and
(4) the construction is compatible with base change T → S.
Proof. Since C → S is projective, every two points of a fiber Cs are contained in an
affine open subscheme of C. Therefore, the Ferrand Pushout (B.4.1)

S
`
S
σ⨿τ /C

 
S
ν / C′

exists as a scheme C ′ .`By the universal property of pushouts,

` there is a natural
map C ′ → S. Since S S → S is finite, so is C → C ′ . Since S S → S is flat, the
construction is compatible with arbitrary base change by Properties of Pushouts
(B.4.8(4)). This gives (2)–(4).
To see (1), first observe that C ′ → S is proper since C → C ′ is finite. By
the Local Structure of Smooth Morphisms (A.3.4), for every s ∈ S, there is an
affine open neighborhood Spec A ⊆ S and étale neighborhoods C → Spec A[x] and
C → Spec A[y] of σ(s) and τ (s). By Properties of Pushouts (B.4.8(3)), the pushout
C ′ has an étale neighborhood of ν(s) isomorphic to the pushout

Spec A × A
0⨿0 / Spec A[x] × A[y]

 
Spec A / Spec A[x, y]/(xy).

It follows that C ′ → S is flat and ν(s) ∈ Cs is a node.

Alternatively, one can construct C ′ using a global Proj construction in the case
that ωC/S (σ +τ ) is a relatively very ample line bundle (as it will be in our application
to stable curves). Namely, there are well-defined morphisms qσL : ωC/S (σ +τ ) → Oσ(S)
and qτ : ωC/S (σ + τ ) → Oτ (S) . We then define C ′ := ProjS d≥0 Ad , where Ad is
the fiber product
Ad / OS

∆
 ⊗d
π∗ (qσ )×π∗ (qτ⊗d ) 
π∗ (ωC/S (σ + τ )⊗d ) × π∗ (ωC/S (σ + τ )⊗d ) / OS × O S

and π : C → S denotes the structure morphism. One can check C ′ → S satisfies

the desired properties by essentially the same argument as above. See [Knu83a,
Thm. 3.4] and [ACG11, §X.7].
Corollary 5.6.16 (Gluing Morphisms). Assume that 2g − 2 + n > 0.
(1) If 2g1 − 2 + n1 > 0 and we set g2 = g − g1 and n2 = n − n1 + 2, there is a
finite morphism of algebraic stacks
Mg1 ,n1 × Mg2 ,n2 → Mg,n


(C, pi ), (D, qi ) 7→ (C ⨿pn1 ∼qn2 D, p1 , . . . , pn1 −1 , q1 , . . . , qn2 −1 ).

258
 (2) If g ≥ 1, there is a finite morphism of algebraic stacks

Mg−1,n+2 → Mg,n
(C, pi ) 7→ (C/pn+1 ∼pn+2 , p1 , . . . , pn−2 ).

Proof. In both cases, Gluing Families Along Sections (5.6.15) directly yields mor-
phisms of algebraic stacks to Mpre g,n . Since stability can be checked on geometric
fibers, it suffices to show that the nodal gluing C ⨿pn1 ∼qn2 D and C/pn+1 ∼pn+2 are
stable, but this is clear as stability can be determined by the pointed normalization
(Exercise 5.3.6). Since both maps are quasi-finite and representable morphisms of
proper Deligne–Mumford stacks, they are finite.
More generally, for a finite index set I, we can define the notion of an I-pointed
stable curve of genus g, and we denote Mg,I as the stack of such curves. A bijection
I∼= {1, . . . , n} induces an isomorphism Mg,I ∼
= Mg,n . With this notation, we have
the natural generalization of Corollary 5.6.16: there are finite morphisms

Mg1 ,I1 × Mg2 ,I2 → Mg1 +g2 ,(I1 ∪I2 )∖{i1 ,i2 } and Mg−1,J → Mg,J∖{j1 ,j2 } ,

gluing the marked points i1 ∈ I1 to i2 ∈ I2 (resp., j1 , j2 ∈ J), subject to the

numerical conditions ensuring that each stack is non-empty.

5.6.4 Boundary divisors and line bundles on Mg,n

We define the boundary divisors δi,I of Mg,n and show that the total boundary
divisor δ is a normal crossings divisor (Proposition 5.6.19).
Definition 5.6.17 (Boundary Divisors). Suppose 2g − 2 + n > 0. Let 0 ≤ i ≤ g be
an integer and I ⊆ {1, . . . , n} be a subset such that 1 − 2i < |I| < 2(g − i) + n − 1 > 0
(which ensures that both Mi,I∪{p} and Mg−i,I c ∪{q} are nonempty). The boundary
divisors are defined as the closed substacks of Mg,n

δi,I = im Mi,I∪{p} × Mg−i,I c ∪{q} → Mg,I∪I c ∼

= Mg,n


δ0 = im Mg−1,n+2 → Mg,n
[
δ = δ0 ∪ δi,I
i,I

with the convention that δ0 is empty if g = 0.

Note that δi,I = δg−i,I c . If n = 0, then δ = δ0 ∪ · · · ∪ δ⌊ g2 ⌋ . If g = 0, then δ is
the union of δI over subsets I of size 2 ≤ |I| ≤ n − 2.

i g i i
i

do di I do di I

Figure 5.6.18: Examples of stable curves in the boundary.

259
 A closed substack Z ⊆ X of a Deligne–Mumford stack is called a divisor (resp.,
normal crossings divisor ) if there is an étale presentation U → X such that Z ×X U ⊆
U is a divisor (resp., normal crossings divisor). Recall that a divisor U ⊆ X of a
regular scheme has normal crossings if for every u ∈ U , O bU,u ∼
=ObX,x /(f1 · · · fk )
where the sequence f1 , . . . , fk ∈ ObX,x extends to a regular system of parameters
f1 , . . . , fn .
Proposition 5.6.19. Over a field k, δ ⊆ Mg,n is a normal crossings divisor.
Proof. An easy dimension count shows that δ has pure dimension 3g − 2 + n. To
see that δ has normal crossings, we may assume that k is algebraically closed.
Let (C, pi ) ∈ Mg,n be a stable curve with nodes q1 , . . . , qs ∈ C, and let (U, u) →
(Mg,n , [C, pi ]) be an étale neighborhood where U is a scheme. By the Local-to-global
Deformation Sequence (5.3.27), there is a surjection
M
Def(C, pi ) → Def(O bC,q )
j
(5.6.20)
j

of first order deformations spaces with Def(ObC,q ) ∼

= k.
j
Letting Artk be the category of local artinian k-algebras with residue field k,
define the functors
F, Gj : Artk → Sets,
by setting F (A) to be the set of isomorphism classes of pairs (C → Spec A, σi , α)
∼
where (C → Spec A, σi ) is a family of n-pointed stable curves and α : (C, pi ) →
(C ×A k, σi ×A k) is an isomorphism, and by setting Gj (A) to be the set of isomorphism
classes of pairs (B, β) where B is a flat A-algebra and β : O bC,q → B ⊗A A/mA is an
j

isomorphism. The map Spec OU,u → Mg,n induces a miniversal formal deformation
b
of F over k[[x1 , . . . , x3g−3+n ]]. By Rim–Schlessinger’s Criterion (C.4.6), Gj admits
a miniversal formal deformation over k[[zj ]] (see Exercise L C.4.16). Since (5.6.20)
is surjective, the natural morphism of functors F → j Gj is versal (or formally
smooth), or in other words the natural map

bU,u ∼
O = k[[x1 , . . . , x3g−3+n ]] → k[[z1 , . . . , zs ]] (5.6.21)

is surjective. We may therefore write O bU,u ∼

= k[[z1 , . . . , zs , x′s+1 , . . . , x′3g−3+n ]] so
that (5.6.21) maps each zi to itself. Letting δU = δ ×Mg,n U , we conclude that
O ∼ O
bδ ,u = bU,u /(z1 · · · zs ). See also [DM69, Thm. 5.2], where it is shown more
U
generally that δ is a relative normal crossings divisor over Z.
Exercise 5.6.22. Show that M0,n has 2n−1 − n − 1 irreducible boundary divisors.
Exercise 5.6.23. The dual graph Γ = (G, w, m) of a stable curve C was defined in
Definition 5.3.8.
(a) Show that for every dual graph Γ, there is a locally closed substack MΓ ⊆ Mg,n
parameterizing stable curves with dual graph Γ, and conclude that there is a
stratification a
Mg,n = MΓ ,
Γ

called the dual graph stratification.

(b) Given two dual graphs Γ and Γ′ , provide a combinatorial condition for when
MΓ′ ⊆ MΓ .

260
 (c) The dual graph stratification for M2 was described in Figure 5.3.9. Describe
the stratifications for M2,1 and M3 .
Definition 5.6.24 (Hodge line bundle). Letting π : Ug,n → Mg,n denote the
universal family, the Hodge vector bundle is defined as the pushforward E :=
π∗ (ωUg,n /Mg,n ). It is a vector bundle by Properties of Families of Stable Curves
(5.3.23). The Hodge line bundle is defined as the determinant of the Hodge vector
bundle λ := det E.
Exercise 5.6.25. Show that under every Forgetful Morphism (5.6.10) and Gluing
Morphism (5.6.16), λ pulls back to λ.
Definition 5.6.26 (The Psi line bundles). Let σi : Mg,n → Ug,n be the ith section
of the universal family. Define the line bundle ψi on Mg,n as the conormal bundle
of σi , i.e., ψi = Iσi /Iσ2i where Iσi ⊆ OUg,n is the ideal sheaf defining σi . The line
bundle ψ is defined (using additive notation) as

ψ = ψ1 + · · · + ψn .

There are identifications

ψi ∼
= σi∗ ΩUg,n /Mg,n ∼
= σi∗ ωUg,n /Mg,n ∼
= σi∗ OUg,n (−σi (Mg,n )).

The Psi classes are defined as c1 (ψi ) ∈ CH1 (Mg,n ); see §6.1.6 for the definition of
the Chow group.
Exercise 5.6.27.
(a) If C is the nodal union C1 ∪ C2 of two smooth curves along p1 ∈ C1 and
p2 ∈ C2 , show that there is a natural identification of vector spaces

δ ⊗ κ([C]) ∼
= Tp1 C1 ⊗ Tp2 C2 .

(b) Show that δ pulls back under the gluing morphism Mg,n+2 → Mg,n to
δ − ψn+1 − ψn+2 and pulls back under Mg1 ,n1 × Mg2 ,n2 → Mg,n to (δ −
ψn1 ) ⊠ (δ − ψn2 ).
(c) Conclude that δ − ψ pulls back to itself under every gluing morphism.
See also [HM98, Prop. 3.32].
Definition 5.6.28 (Kappa classes). The ith kappa class is defined as
i+1
κi := π∗ (ψn+1 ) ∈ CHi (Mg,n ).

The first kappa class κ := κ1 is the class of a line bundle.

Definition 5.6.29 (Canonical line bundle). Over a field k, let ΩMg,n /k be the sheaf
of differentials (Example 4.1.3), which is a vector bundle since Mg,n is smooth.
Define the canonical line bundle as

K = KMg,n := det ΩMg,n /k .

Remark 5.6.30 (Mumford’s Formula). An application of Grothendeick–Riemann–

Roch gives the relations

κ = 12λ − δ and K = 13λ − 2δ;

see [Mum83, §5] and [HM98, §3.E].

261
Exercise 5.6.31 (Comparison with coarse moduli space). Let ∆i , ∆ ⊆ M g,n be the
image of δi , δ ⊆ Mg,n under the coarse moduli space map π : Mg,n → M g,n . Show
that
δi if i ̸= 1

π ∗ ∆i =
2δ1 ifi = 1
π ∗ ∆ = δ + δ1
∗
π KM g,n = KMg,n − δ1 .

5.7 Irreducibility
In your “appendix”, you refer to a result of Matsusaka I did not hear
of before, namely the connectedness or irreducibility of the variety of
moduli for curves of genus g, in any characteristic. I did not know there
was any algebraic proof for this (whatever way you state it). Yet I have
some hope to prove the connectedness of the Mg,n (arbitrary levels) using
the transcendental result in char. 0 and the connectedness theorem; but
first one should get a natural “compactification” of Mg,n which should be
simple2 over Z.

Grothendieck, letter to Mumford, 1961 [Mum10, p. 638]

We provide several arguments that Mg,n is irreducible:

– the classical topological argument due to Clebsch, Lüroth, and Hurwitz (Theo-
rem 5.7.19),
– a purely algebraic argument in characteristic 0 using degenerations of smooth
curves and the inductive nature of the boundary δ = Mg ∖ Mg (Theo-
rem 5.7.30), and
– a proof in characteristic p > 0 by reduction to characteristic 0 (Theorem 5.7.33)

These proofs rely fundamentally on the theory of branched covering as discussed in

§5.7.1.

Equivalences. We begin with a few remarks regarding the equivalences between

the connectedness/irreducibility of Mg,n , Mg , and their coarse moduli spaces. Since
Mg,n is a smooth algebraic stack over a field k, its irreducibility is equivalent
to its connectedness. Moreover, since Mg,n+1 → Mg,n is the universal family
(Proposition 5.6.12) and has connected fibers, it suffices to verify the connectedness
of Mg . We thus have equivalences

Mg,n irreducible over k ⇐⇒ Mg,n connected over k

⇐⇒ Mg connected over k
⇐⇒ Mg connected over k and dense in Mg .

Finally, we note that since the coarse moduli space Mg,n → M g,n induces a
∼
homeomorphism |Mg,n | → |M g,n | on topological spaces, each statement can be
formulated using the coarse moduli space.
2 Grothendieck originally called smooth morphisms simple.

262
 5.7.1 Branched coverings
Let f : C → D be a finite morphism of smooth, connected, and projective curves
over an algebraically closed field k. Assume that f is separable, i.e., the induced
map K(D) → K(C) of functions fields is a separable extension. For a point p ∈ C(k)
with image q ∈ D(k), the ramification index at p is the integer ep such that s 7→ utep
under the map OD,Q → OC,p , where s and t are uniformizers and u is a unit. We
say that f is
if ep > 1

 ramified at p
tamely ramified at p if ep > 1 and either char(k) = 0 or char(k)̸ | ep
unramified at p if ep = 1


If f is unramified at p, then the scheme-theoretic fiber over f (p) at p is isomorphic

to Spec κ(p), and thus this agrees with the usual definition of unramified (Theo-
rem A.3.3). Moreover, since f is flat, f is unramified at p if and only if f is étale at
p.
There is a short exact sequence of differentials

0 → f ∗ ΩD → ΩC → ΩC/D → 0. (5.7.1)

Indeed, the sequence above is always right exact. Since f ∗ ΩD and ΩC are line bundles,
the left map is injective if and only if it is nonzero. However, K(D) → K(C) is
separable so ΩC/D ⊗ K(C) = ΩK(C)/K(D) = 0, and thus f ∗ ΩD → ΩC is nonzero at
the generic point. Examining the sequence above at the stalks over a point p ∈ C(k),
the differential dt maps to d(usep ) = eusep −1 ds + sep du. If f is tamely ramified at p,
then (ΩC/D )p ∼ = OC,p ⟨ds⟩/(sep −1 ds) and length(ΩC/D )p = dim ΩC/D ⊗κ(p) = ep −1.
Definition 5.7.2. Let k be an algebraically closed field.
(1) A branched covering is a finite separable morphism f : C → D of smooth,
connected, and projective curves over k.
(2) A simply branched covering is a branched covering such that there is at most
one ramification point in every fiber and every ramification point p ∈ C(k) is
tamely ramified with ramification index ep = 2.

Figure 5.7.3: Examples of branched coverings over P1 : (A) is simply branched while
(B) and (C) are not. While the picture may suggest that the source curve C is not
smooth, C is, in fact, smooth over the base field k. However, the map C → P1 is not
smooth, and the pictures above are designed to reflect the singularities of C over P1 .

If f : C → D is a branched covering, the ramification divisor is defined as

if f is tamely
ramified
(A) X X
length(ΩC/D )p · p (ep − 1).
z}|{
R= =
p∈C(k) p∈C(k)

263

(B)
Theorem 5.7.4 (Riemann–Hurwitz). If f : C → D is a branched covering with
ramification divisor R, then ΩC ∼
= f ∗ ΩD ⊗ OC (R) and

2g(C) − 2 = deg(f )(2g(D) − 2) + deg R.

In particular, f : C → P1 is simply branched, then it is ramified over 2g + 2d − 2

distinct points.

Proof. This follows directly from the exact sequence (5.7.1). See also [Har77, Prop.
IV.2.3]

Example 5.7.5. For a local model of a branched cover, consider the map f : A1 → A1
defined by x 7→ xn . The relative sheaf of differentials is ΩA1 /A1 = k[x]⟨dx⟩/(nxn−1 dx).
Thus, if char(k) does not divide n, f is a branched cover étale over A1 ∖ 0 and tamely
ramified at 0 with index n − 1.

Exercise 5.7.6. Show that every branched covering is étale locally isomorphic to
A1 → A1 , x 7→ xn around a branch point of index n − 1.

Algebraic vs. holomomorphic vs. topological branched covers. A holomor-

phic (resp., topological) branched covering of P1 is a non-constant and holomorphic
(resp., continuous) morphism f : C → P1 of connected and compact Riemann sur-
faces (resp., connected, Hausdorff, and compact topological spaces such that f is a
covering space over the complement of finitely many points of P1 ).

Proposition 5.7.7. Over C, there are natural bijections

{C → P1 algebraic branched coverings} ←→ {C → P1 holomorphic branched coverings}

←→ {C → P1 topological branched coverings}

Proof. An algebraic branched covering is holomorphic and a holomorphic branched

covering is topological. Conversely, if f : C → P1 is a topological covering, then the
holomorphic structure on P1 induces naturally a holomorphic structure on C such
that f : C → P1 is analytic. It is a classical fact relying on the Implicit Function
Theorem that there are local charts of C and P1 such that f is described by z 7→ z k
(c.f., [Mir95, Prop. II.4.1]), which implies that C is algebraic.

Monodromy actions. Let f : C → P1 be a (topological) branched covering of

degree d over C and B ⊆ P1 its ramification locus, i.e., the smallest set of points such
that f −1 (P1 ∖ B) → P1 ∖ B is a covering space. Choose a base point q ∈ P1 ∖ B.
The monodromy action of π1 (P1 ∖ B, q) on the fiber f −1 (q) is defined as follows: for
γ ∈ π1 (P1 ∖ B, q) and p ∈ f −1 (q), then the path γ : [0, 1] → P1 lifts uniquely to a
path γ
e : [0, 1] → C such that γ
e(0) = p, and the action is defined by γ · p = γ e(1).

264
 8P r

r
e
g g

Figure 5.7.8: Monodromy action

A choice of a bijection f −1 (p) ∼

= {1, . . . , d} defines a group homomorphism

ρ : π1 (P1 ∖ B, q) → Sd ,

which we call the monodromy representation. The converse is also true: every such
p group homomorphism is induced by a branched covering.

Proposition 5.7.9. Let B ⊆ P1 be a finite subset of size b, q ∈ P1 ∖ B be a point,

and d > 0 a positive integer. The fundamental group π1 (P1 ∖ B, q) is identified with
the free group generated by the simple loops σi for i = 1, . . . , b around the points of
B. There is a natural bijection of isomorphism classes
   
 simply branched 1
 
  group homomorphisms
 

ρ : π1 (P1 ∖ B, q) → Sd such that

coverings C → P of
  
∼ ←→ ∼,
 of degree d   im(ρ) ⊆ Sd is a transitive subgroup 
fp q
   
branched over B and each ρ(σi ) is a transposition
   

where two branched covers are equivalent if they are isomorphic over P1 , and two
homomorphisms ρ and ρ′ are equivalent if they differ by an inner automorphism of
Sd , i.e., ∃ h ∈ Sd such that ρ′ = h−1 ρh.

Proof. We have already explained the natural map from left to right. Conversely,
given a group homomorphism ρ : π1 (P1 ∖ B, q) → Sd , we let H ⊆ π1 (P1 ∖ B, q) be
the subgroup containing elements γ fixing 1. Since this subgroup has index d, it

p corresponds to a covering space C ∗ → P1 ∖ B, which one can show extends to a

finite morphism C → P1 of degree d. The connectedness of C translates into the
Gi condition that im(ρ) ⊆ Sd is transitive, and the cover C → P1 being simply branched
8
translates into each ρ(σi ) being a transposition. See [Mir95, Prop. III.4.9].

5.7.2 Hurwitz moduli spaces

Vi For positive integers g and b, we define and study the Hurwitz moduli space

fitt
Hurg,b := {simply branched coverings C → P1 of genus g over b ordered points}

of simply branched covers and its relation to Mg . By Riemann–Hurwitz (5.7.4), if

C → P1 has degree d, then b = 2g + 2d − 2. In the literature, Hurwitz spaces are

265
sometimes indexed as Hurd,b or Hurd,g . The key diagram is

Hurg,b

(5.7.10)
{ #
Mg M0,b

where a simply branched covering (C → P1 ) ∈ Hurg,b gets mapped to C ∈ Mg and

the b ordered branch points in M0,b .
The moduli space Hurg,b over C can be viewed analytically as the topological space
with the coarsest topology such that Hurg,b → M0,b is continuous. Algebraically, we
define it using, as usual, our functorial approach. To this end, we define a family
(C → D → S, σi ) of coverings of P1 of genus g simply branched over b ordered points
over a scheme S as a family C → S of smooth genus g curves and a b-pointed
family (D → S, σi ) of smooth genus 0 curves together with a finite, flat morphism
C → D of schemes over S of degree d such that for every geometric point s ∈ S(k),
Cs → Ds ∼ = P1k is simply branched over σ1 (s), . . . , σb (s).
Definition 5.7.11 (Hurwitz functor). For positive integers g and d, set b = 2g+2d−2.
The Hurwitz functor is defined as the functor

Hurg,b : Sch → Sets

families (C → D → S, σi ) of simply branched coverings 
 
S 7→ ∼,
of P1 of genus g and degree d

where two families (C → D → S, σi ) and (C ′ → D′ → S, σi′ ) are equivalent if there are

∼ ∼
isomorphisms C → C ′ and D′ → D′ compatible with the sections and the structure
morphisms C → D and C → D′ .
′

There are morphisms Hurg,b → Mg and Hurg,b → M0,b (as in Diagram (5.7.10))
of stacks taking a family of coverings (C → D → S, σi ) to C → S and (D → S, σi )
giving Diagram (5.7.10).
Remark 5.7.12 (Variants). There are variants of the above definition that are also
referred to as Hurwitz moduli spaces in the literature. First, one can rigidify the
moduli problem by considering the functor Hurrig g,b : Sch → S, where an object
over S is a finite, flat morphism C → P1S over S from a family of smooth curves
of genus g together with disjoint ordered sections σ1 , . . . , σb : S → P1S such that
for every geometric point s ∈ S(k), Cs → P1k is a covering simply branched over
σ1 (s), . . . , σb (s). Here two families are equivalent if they are isomorphic over P1S and
the sections are equal. There is a morphism Hurrig g,b → (P ) ∖ ∆, where ∆ is the
1 b

union of all pairwise diagonals. The algebraic group PGL2 acts freely on Hurrig g,b and
the quotient is identified with Hurg,b , while the quotient of the PGL2 -equivariant
morphism Hurrig g,b → (P ) ∖ ∆ is identified with Hurg,b → M0,b
1 b

Another common variant, which was the version originally considered by Hurtwitz
in [Hur91], is to consider the branch locus as a set of unordered points. Let
Hurrig,unord
g,b : Sch → Sets be the functor, where an object over S is a finite, flat
morphism f : C → P1S over S from a family of smooth curves of genus g such that
for every geometric point s ∈ S(k), Cs → P1k is a covering simply branched over b
points. We claim that there is a morphism Hurrig,unordg,b → Symb P1 ∖ ∆ to the space
parameterizing b unordered distinct points. To see this, observe that the ramification
divisor Rf ⊆ C, defined as the relative singular locus of f : C → P1S , is finite and

266
étale over S of degree b = 2g + 2d − 2. The complement Rf ×S · · · ×S Rf ∖ ∆ of the
pairwise diagonals is a principal Sb -torsor over S (see Exercise B.1.52) and the map
Rf ×S · · · ×S Rf ∖ ∆ → P1S ×S · · · × P1S ∖ ∆ is an Sb -equivariant morphism. This
defines an S-valued point of the quotient stack [ (P1 )b ∖∆ /Sb ] over S, which in turn

is identified with the scheme Symb P1 ∖ ∆. Note also that Symb P1 ∖ ∆ ∼ = Pb ∖ ∆,

where the latter ∆ is the discriminant hypersurface. The symmetric group Sb acts
freely on Hurrig rig,unord
g,b with quotient Hurg,b , and the Sb -quotient of Hurrig 1 b
g,b → (P ) ∖∆
is identified with Hurrig,unord
g,b → Symb P1 ∖ ∆.
The following exercise allows us to conclude that the map Hurg,b → M0,b is
a finite, étale, and surjective morphism, i.e., an algebraic covering space. When
working over a field k, we will abuse notation by also referring to Hurg,b as the base
change Hurg,b ×Z k.
Exercise 5.7.13. Let k be an algebraically closed field of characteristic 0. Let g, b,
and d be positive integers with b = 2g + 2d − 2. Assume that d > 2.
(a) Show that the diagonal of Hurg,b over k is representable by schemes.
(b) If C → P1k is a simply branched covering of degree d, show that every automor-
phism of C over P1 is trivial. Conclude that the automorphism group scheme
of a family of simply branched coverings of P1 of degree d is also trivial.
Hint: Use the fact that there are no non-trivial automorphisms of a smooth
curve fixing more than 2g + 2 points (Exercise 5.1.9).
(c) (hard) Show that Hurg,b → M0,b is a finite, étale, and representable morphism.
In particular, the functor Hurg,b is representable by a scheme. See also [Ful69,
Thm. 7.2], where it is shown more generally that Hurg,b is representable over
Spec Z, and the map Hurg,b → M0,b is étale (resp., finite étale over bases of
characteristic greater than d).
(d) Verify that the functors Hurrig rig,unord
g,b and Hurg,b defined in Remark 5.7.12 are
also representable by schemes.
Remark 5.7.14. For the proof of the Clebsch–Hurtwitz–Lüroth Theorem (5.7.19) in
the next section, we only need to know that Hurg,b → M0,b is a topological covering
F
space, and this fact has an elementary argument. Consulting Figure 5.7.15, given a
simply branched covering f : C → P and a branched point p ∈ C, we can choose
1

an open neighborhood U ⊆ P1 around f (p) such that f −1 (U ) → U is isomorphic to

an open neighborhood of the map C → C given by x 7→ xn . For every other point
q ′ ∈ U , we can construct a branched cover C ′ → P1 which outside U is the same
as C → P1 and over U is locally isomorphic to x 7→ xn but centered over q ′ (rather
than f (p)).

flp q
Figure 5.7.15

267
 We turn now to studying the other map: Hurg,b → Mg .

Lemma 5.7.16. Let C be a smooth, connected, and projective curve of genus g

over an algebraically closed field k of characteristic 0. If L is a line bundle of degree
d ≥ 2g + 3 and V ⊆ H0 (C, L) is a linear system of dimension 2, then a choice of
basis of V induces a simply branched covering C → P1 .

Proof. We proceed with a dimension count. Since h1 (C, L) = h0 (C, ωC ⊗ L∨ ) = 0 as

deg(ωC ⊗ L∨ ) < 0, Riemann–Roch implies that h0 (C, L) = d + 1 − g, and it follows
that the dimension of the Grassmannian Gr(2, H0 (L)) of 2-dimensional subspaces
is 2(d − g − 1). Since char(k) = 0, every finite morphism C → P1 is automatically
separable. Thus, if V does not induce a simply branched covering C → P1 , then
one of the following three conditions must hold:
(a) V has a base point,
(b) there exists a ramification point with index greater than 2, or
(c) there exists two ramification points in the same fiber.
We claim that each condition is closed of codimension at least one. For condi-
tion (a), if p ∈ C(k) is a base point, then V ⊆ H0 (C, L(−p)). Since d ≥ 2g,
we have that deg(ωC ⊗ (L(−p))∨ ) < 0, and thus h0 (C, L(−p)) = d − g and
dim Gr(2, H0 (C, L(−p))) = 2(d − g − 2). Varying p ∈ C, we see that the locus
defining (a) has dimension 2(d − g − 2) + 1 = dim Gr(2, H0 (L)) − 1. We leave cases
(b) and (c) for the reader. See also [Sev21] and [Ful69, Prop. 8.1].

Exercise 5.7.17 (details).

(a) Verify that conditions (b) and (c) above are closed of codimension at least one.
(b) Show that the argument holds as long as char(k) ̸= 2 and explain why it fails
if char(k) = 2.

Corollary 5.7.18. Let k be an algebraically closed field of characteristic 0. If g ≥ 2

and d ≥ 2g + 3 with b = 2g + 2d − 2, the morphism

Hurg,b → Mg , (C → P1 , σi ) 7→ C

is surjective.

5.7.3 The Clebsch–Hurwitz–Lüroth Theorem

We provide the classical argument due to Clebsch [Cle73], Hurwitz [Hur91], and
Lüroth [Lür71] that the Hurwitz moduli space Hurg,b is connected over C. From
the surjectivity of the map Hurg,b → Mg , this allows us to conclude that Mg is
irreducible, a classical result of Klein [Kle82, §19] and Severi [Sev21, §F]. For a
modern treatment, see [Ful69, Prop. 1.5].

Theorem 5.7.19 (The Clebsch–Hurwitz–Lüroth Theorem). For g ≥ 2 and d ≥ 2

with b = 2g + 2d − 2, the Hurwitz moduli space Hurg,b over C is connected.

Proof. We utilize the morphism β : Hurg,b → M0,b , sending a simply branched cover
to the ordered branch locus. By Exercise 5.7.13(c), this map is a covering space. For
every finite ordered set B = {q1 , . . . , qb } ⊆ P1 of b distinct points and q ∈ P1 ∖ B,
the fundamental group π1 (P1 ∖ B, q) = ⟨σi | σ1 · · · σb = 1⟩, where σi is a simple

268
loop around qi , acts on the fiber f −1 (p) of a simply branched covering f : C → P1 .
Similarly, since β is a covering space, π1 (M0,b , B) acts on the fiber

Hurd,B := β −1 (B) ⊆ Hurg,b

over B ∈ M0,b . (Note the distinction between the use of the uppercase subscript ‘B’
and the lowercase ‘b’.) Using Proposition 5.7.9, we have bijections

Hurg,B = {genus g coverings C → P1 simply branched over B}/∼

 group homomorphisms ρ : π1 (P1 ∖ B, q) → Sd  
 
∼
= such that im(ρ) ⊆ Sd is a transitiver Psubgroup conjugation by Sd
and each ρ(σi ) is a transposition F
 
p
sequences (τ1 , . . . , τb ) ∈ (Sd )b of transpositions
 
∼
= conjugation by Sd .
with product 1 generating a transitive subgroup
r

The connectedness of Hurg,b is equivalent to gthe transitivity of the action of

π1 (M0,b , B) on the fiber Hurd,B = β −1 (B). Consider the sequence


τ := (12), (12), (13), (13), . . . , (1 d − 1), (1 d − 1), (1d), (1d), . . . , (1d) ∈ Hurd,B .
| {z } | {z }
2(d−2) 2g+2
p
It suffices to show that every orbit of the action of π1 (M0,b , B) on the fiber Hurd,B =
β −1 (B) contains τ , and, to this end, we define the loop

Γi : [0, 1] → M0,b
t 7→ (q1 , . . . , qi−1 , γi (t), γi′ (t), qi+2 , . . . , qb ),
flp q
where γi and γi′ are paths as in Figure 5.7.20.

G G
P
fi
ri
ri
Girl

Figure 5.7.20: Paths in P1

For an element (λ1 , . . . , λb ) ∈ Hurg,b , the action of Γi is given by

Γi · (λ1 , . . . , λb ) = (λ1 , . . . , λi−1 , λ−1

i λi+1 λi , λi , λi+2 , . . . , λb ).

It is now a combinatorial problem that we leave to the reader to show that there
exists a sequence Γi1 , . . . , Γik of loops such that τ = Γi1 Γi2 · · · Γik · (λ1 , . . . , λb ).

Exercise 5.7.21. Solve the combinatorial problem at the end of the proof.

Corollary 5.7.22 (Irreducibility of Mg ). For any field k of characteristic 0, Mg

is irreducible.

269
Proof. Since the morphsim Hurg,b → Mg is surjective (Corollary 5.7.18), the con-
nectedness of Hurg,b over C implies the connectedness of Mg over C. This, in turn,
implies that Mg is geometrically connected over Q, and thus connected over any field
k of characteristic 0. Since Mg is smooth over k (Theorem 5.4.14), its connectedness
is equivalent to its irreducibility.

The irreducibility of Mg can also be established using the Teichmüller space Tg

parameterizing complex structures on a topological surface Σg of genus g. It was
first proved in [Ber60] that Tg is homeomorphic to a ball in C3g−3 , and that Mg is
identified with the quotient Tg /Γg by the mapping class group Γg .

5.7.4 Irreducibility via degeneration

We now give a completely algebraic argument for the irreducibility of Mg in charac-
teristic 0. The key idea is to show every smooth curve degenerates to a singular stable
curve (Proposition 5.7.26)—this is the most challenging part of the argument and is
achieved with a technique inspired by the theory of admissible covers. This reduces
the connectedness of Mg to the connectedness of the boundary δ = Mg ∖Mg , which
we establish using the inductive structure and the Gluing Morphisms (5.6.16). This
argument is similar in spirit to Deligne and Mumford’s first proof of the irreducibility
of M g in positive characteristic (see Remark 5.7.34). We follow the treatment in
Fulton’s appendix of Harris and Mumford’s paper [HM82].
We begin with a warmup—the genus 0 case.

Proposition 5.7.23. For every algebraically field k and every integer n ≥ 3, M 0,n
is irreducible.

Proof. As M 0,n is smooth, it suffices to show it is connected. Observe that M0,n

is connected as it is identified with (P1 ∖ {0, 1, ∞})n−3 ∖ ∆, where ∆ is the union
of the pairwise diagonals (see Exercise 3.1.20). Therefore, it suffices to show that
the boundary δ = M 0,n ∖ M0,n does not have a connected component consisting
entirely of singular curves. Given a singular curve (C, pi ) ∈ M 0,n (k), it is easy
to directly construct a stable family of curves (C → Spec R, σi ) over a DVR R
whose generic fiber is smooth and whose special fiber is (C, pi ): indeed, every node
of C has a Zariski-open neighborhood of the form Spec k[x, y]/(xy) and we can
glue the deformations Spec k[x, y]/(xy − π), where π ∈ R is a uniformizer, to a
family C → R. Alternatively, by deformation theory Proposition 5.3.25, there are
compatible stable families (Cn → Spec k[t]/(tn+1 ), σn,i ) with (C0 , σ0,i ) = (C, pi ),
which by Grothendieck’s Existence Theorem (C.5.8) effectives to a stable family
(C → Spec k[[t]]), σi ) with smooth generic fiber.
We now give an alternative proof, which uses the identical strategy that we will
use shortly to show the connectedness of Mg,n . First, we claim that an n-pointed
curve (P1 , pi ) degenerates in a one-parameter family to a singular n-pointed stable
curve. Indeed, if we let the points p1 and p2 approach each other at different rates
and blow up the limit in the central fiber where they intersect, then the total family
together with the strict transform of the sections defines a family of stable n-pointed
genus 0 curves with singular central fiber. It therefore suffices to show that the
boundary δ = M 0,n ∖ M0,n is connected. The boundary divisor δ decomposes as a
union
[
where δ0,I = im M0,I∪{p} × M0,I c ∪{q} → M0,n


δ= δ0,I ,
I

270
 i j c

d d o 8

over subsets I ⊆ {1, . . . , n} of size 2 ≤ |I| ≤ n − 2, subject to the relation δ0,I =

δ0,I c . By induction, we can assume that M 0,k is irreducible for k < n, where the
base case holds as M 0,3 = Spec k. This implies that each δ0,I is connected. Let
I, J ⊆ {1, . . . , n} be two distinct subsets. After possibly replacing I with I c , we can
T
assume that K := I ∩ J has size at least 2. If K ⊊ I and K ⊊ J, then by consulting
Figure 5.7.24, we see that δ0,I ∩ δ0,K ̸= ∅ and δ0,J ∩ δ0,K ̸= ∅. If I ⊆ J or J ⊆ I,
then δ0,I ∩ δ0,J ̸= ∅ for the same reason. The connectedness of δ follows.

T's t
a

Figure 5.7.24: A curve in δ0,I ∩ δ0,K as long as K ⊊ I.

Exercise 5.7.25. Show that every stable curve C over an algebraically closed field
k deforms a smooth curve. More precisely, show that there is a family of stable
curves π : C → T over a connected curve3 T over k and a point t ∈ T (k) such that
Ct ∼
= C and π −1 (T ∖ {t}) → T ∖ {t} is smooth.

The converse to the above exercise, i.e., that every smooth curve degenerates to
a stable curve, is more difficult.

Proposition 5.7.26. Let C be a smooth, connected, and projective curve of genus

g ≥ 2 over an algebraically closed field k of characteristic 0. There exists a family
C → T of stable curves over a smooth connected curve T over k with points s, t ∈ T (k)
such that Cs ∼
= C and Ct is a singular stable curve.

Proof. The essential strategy is as follows:

① For d ≫ 0, choose a simply branched covering C → P1 of degree d branched

over distinct points q1 , . . . , qb ∈ P1 where b = 2g + 2d − 2.
② Deform the covering C → P1 branched over qi to a covering C ′ → P1 branched
over general points qi′ .
③ Degenerate (P1 , qi′ ) ∈ M0,b to the stable curve (D, di ) ∈ M 0,b featured in
Figure 5.7.27.

d
de d day
d

Figure 5.7.27: The degenerate curve (D, di ) ∈ M 0,b

④ Degenerate the cover C ′ → P1 to a cover C ′′ → D branched over di such that

C ′′ is singular and stable.
3 By our conventions, a curve (Definition 5.1.1) is necessarily of finite type over k.

271
 For Step 1, Lemma 5.7.16 guarantees the existence of a simply branched covering
C → P1 of degree d ≫ 0 simply branched over b = 2g + 2d − 2 distinct points qi .
This defines a b-pointed stable curve (P1 , qi ) ∈ M0,b . For the remaining steps, we
first claim that it suffices to find a family C → T over a connected scheme T of
finite type over k. Indeed, this would show that C is in the same connected (hence
irreducible) component M′ ⊆ Mg as a singular stable curve C ′′ . By Le Lemme de
Gabber (4.6.1), there is a finite cover Z → Mg by a scheme, and Z must have an
irreducible component Z ′ surjecting onto M′ . As any two k-points of an irreducible
scheme Z ′ of finite type over k are contained in an integral curve Q ⊆ Z ′ , the claim
follows by considering the map Q e → Z ′ → Mg from the normalization of Q.
For Step 2, let H ⊆ Hurg,b be the connected component containing C → P1 .
Since Hurg,b → M0,b is finite and étale (Exercise 5.7.13(c)) and thus both an open
and closed morphism, the composition H ,→ Hurd,b → M0,b is surjective. As H is
connected, it suffices to find a single simply branched cover (C ′ → P1 ) ∈ H and a
family of stable curves C → T over a connected base T connecting C ′ to a singular
stable curve. This brings us to Step 3.
We claim that there is a commutative diagram

T ∖  {t} / H   op & cl / Hurg,b

_
fin ét (5.7.28)
 ## 
T / M 0,b o ? _ M0,b
t7→(D,di )

where T is a smooth connected curve and t ∈ T (k) maps to the stable curve (D, di ) of
Figure 5.7.27. To see this, observe that since M 0,b is irreducible (Proposition 5.7.23),
M0,b ⊆ M 0,b is dense. We can thus choose a map Y → M 0,b from a connected
reduced curve and a point y ∈ Y (k) such that y 7→ (D, di ) and such that the image
of every y ′ ̸= y is smooth. Let Y ′ be a connected component of the base change
(Y ∖ {y}) ×M0,b H, and let T be the normalization of the closure of the image
Y ′ → M 0,b . There is a point t ∈ T mapping to (D, di ) ∈ M 0,b , and after replacing
T with an open neighborhood of t, we can arrange that T ∖ {t} → M 0,b factors
through H.
Diagram 5.7.28 gives a family (D → T, τi ) of b-pointed stable curves of genus
0 together with a simply branched covering C ∗ → D∗ over T ∗ := T ∖ {t} (where
D∗ ⊆ D is the preimage of T ∗ ) such that the sections τi∗ : T ∗ → D∗ pick out the
branch locus. We now claim that, after replacing (T, t) by a ramified cover, there
are dotted arrows completing the cartesian diagram

C∗ / D∗ / T∗

□ □
  
C
f
/D / T,

where C → T is a family of nodal curves. If we define C as the integral closure of

OD in K(C ∗ ), then f : C → D is a finite morphism. Letting Dsm = D ∖ {di } be the
relative smooth locus of D → T , then Purity of the Branch Locus (A.3.13) implies
that the ramification locus of f −1 (Dsm ) → Dsm is a divisor. Therefore, C → D is
ramified only over the sections τ1 , . . . , τb and possibly over the nodes di of D = Dt .
To see that C → S is nodal, observe that étale locally around each di ∈ D, C → D
has the form of a finite morphism U → V := Spec k[x, y]/(xy − π k ) where k ≤ d

272
and π is a local coordinate on an étale neighborhood of t ∈ T . The map U → V is
finite étale over V ∖ 0. Since Spec k[x′ , y ′ ]/(x′k y ′k − π k ) ∖ 0 → V ∖ 0, defined by
(x′ , y ′ ) 7→ (xk , y k ), is the universal cover, we see that the preimage of V ∖0 in U must
be isomorphic to Spec k[x′′ , y ′′ ]/(x′′l y ′′l − π k ) ∖ 0 for some integer l dividing k. Thus,
INCLUDE
U = Spec k[x′′ , y ′′ ]/(x′′l y ′′l − π k ), and we see that the reduced scheme structure of
the fiber Ut , defined by π = 0, is nodal. It follows that (Ct )red is nodal. By the same
argument as in Step 4 of the proof of Stable Reduction (5.5.1), there is ramified
cover T ′ → T such that the central fiber of the normalization Ce′ of C ′ = C ×T T ′ is
reduced and nodal. After replacing C with Ce′ and T with T ′ , we have arranged that
C → T is a family of nodal curves.

Ce DE D
E
n
I
nooo
E D

see I

T
t

Figure 5.7.29: Picture of C → D → T .

The family C → D may not be stable, but using the Stable Contraction of a
Prestable Family (Theorem 5.6.6), we can contract rational tails and bridges to
obtain a stable family C st → T . To show that C0st is singular, it suffices to show that
every smooth irreducible component of C0 has genus 0. Letting X ⊆ C0 be such a
component, the image of X under C0 → D0 = D is one of the P1 ’s. Let d′ be the
degree of the induced map X → P1 . We know that X → P1 is ramified over the
marked points of the P1 and possibly ramified over the nodes of D contained in the
P1 with ramification index at most d′ − 1. If the P1 is at the end of the chain in D,
it contains two marked points and one node, and Riemann–Hurwitz (5.7.4) implies
that 2gX − 2 ≤ −2d′ + 2 + (d′ − 1), which implies that gX = 0. If the P1 is in the
middle, it contains one marked point and two nodes, and Riemann–Hurwitz implies
that 2gX − 2 ≤ −2d′ + 1 + 2(d′ − 1), which also implies that gX = 0.

Theorem 5.7.30 (Irreducibility of Mg,n ). Let g and n be nonnegative integers

satisfying 2g − 2 + n > 0. For any field k of characteristic 0, Mg,n is irreducible
and contains Mg,n as a nonempty dense open substack.

Proof. We may assume that k is algebraically closed. There exists smooth curves
of every genus (Exercise 5.1.7). As P1 with n ≥ 3 distinct points is stable and a
smooth, connected, and projective genus 1 curve with n ≥ 1 distinct points is stable,
Mg,n is nonempty as long as 2g − 2 + n > 0. As Mg,n is smooth (Theorem 5.4.14), it
suffices to show that Mg,n is connected. As Mg,n+1 → Mg,n is the universal family
(Proposition 5.6.12), the connectedness of Mg,n implies the connectedness of Mg,n′
for n′ > n. We already know that M0,n is connected for n ≥ 3 (Proposition 5.7.23).

273
We also know that M1,1 is isomorphic to [(A2 ∖ 0)/Gm ] and thus is connected
Exercise 3.1.19(c), and it follows that M1,n is connected for n ≥ 1.
We are thus reduced to show that Mg is connected for g ≥ 2, and by induction we
can assume that Mg′ is connected for g ′ < g. Since every smooth curve degenerates
to a singular stable curve in the boundary δ = Mg ∖ Mg (Proposition 5.7.26), we
are further reduced to showing that the boundary δ is connected. We write δ =
δ0 ∪ · · · ∪ δ⌊g/2⌋ where δ0 = im(Mg−1,2 → Mg ) and δi = im(Mi,1 × Mg−i,1 → Mg ),
using the Gluing Morphisms (5.6.16). By the inductive hypotheses, Mg−1,2 and
Mi,1 × Mg−i,1 are connected, and thus so is each δi . But on the other hand, the
boundary divisors δi intersect! Namely, for every i, j = 0, . . . , ⌊g/2⌋, the intersection
δi ∩ δj contains curves as in Figure 5.7.31. See also Fulton’s appendix [HM82].

g i j
i i
g

i j c

d d o S
Figure 5.7.31: The boundary divisors δi and δj have a nonempty intersection.

Remark 5.7.32 (Admissible Covers). The above argument was motivated by the
theory of admissible covers introduced by Harris and Mumford [HM82] to compute
the Kodaira dimension of M g . See also the exposition in [HM98, §3.G]. Admissible
covers are a generalization of simply branched covers C → P1 where the source and
target curves may have nodal singularities. The inspiration behind admissible covers
is to define a moduli stack Hurg,b fitting into the diagram

Hurg,b Tn Hurg,b

compactifying
| # { #
Mg M 0,b Mg M0,b .

7
An admissible cover of genus g is a surjective finite morphism f : C → B from a

É
prestable genus g curve to a stable b-pointed genus 0 curve (B, p1 , . . . , pb ) such that
(a) the preimage of the smooth locus B sm under f is the smooth locus C sm and
the induced morphism C sm → B sm is simply branched of degree d over the
points pi , i.e., each ramification index is 2 and there is at most one ramification
point in every fiber; and
(b) for every node q ∈ B and every node r ∈ C over q, the étale local structure
of C → B at r is of the form k[x, y]/(xy) → k[x, y]/(xy) defined by (x, y) 7→
(xm , y m ) for some m, i.e., the two branches have equal ramification indices at
the node.
This definition extends to families and defines a proper Deligne–Mumford stack
Hurg,b . (Note that admissible covers may have non-trivial automorphism groups

274
unlike simply branched coverings, and consequently Hurg,b is not a scheme.) The
assignment (C → B) 7→ (B, pi ) defines a morphism Hurg,b → M 0,b . On the other
hand, the assignment (C → B) 7→ C st defines Hurg,b → Mpreg , noting that the source
curve C of an admissible cover C → B need not be stable. Composing with the
Stable Contraction Morphism (5.6.7) Mpre → Mg gives a morphism Hurg,b → Mg .
The argument degenerating a smooth curve to a singular stable curve (Propo-
sition 5.7.26) can be rewritten in this language. For d ≫ 0, given a smooth curve
C ∈ Mg , we choose a preimage (C → P1 ) ∈ Hurg,b (Lemma 5.7.16). Let H ⊆ Hurg,b
be the connected component containing (C → P1 ), and let H ⊆ Hurg,b be its closure.
Since Hurg,b → M0,b is finite and étale (Exercise 5.7.13(c)), H surjects onto M 0,b .
Choose a commutative diagram

T ∖  {t} /H
_ :
proper
 
T / M 0,b ,
t7→(D,di )

as in the proof of Proposition 5.7.26, where T is a smooth connected curve, t ∈ T (k)

maps to the stable curve (D, di ) ∈ M 0,b of Figure 5.7.27, and there is some t′ ̸=
t ∈ T (k) mapping to a simply branched covering (C ′ → P1 ) ∈ H. By the valuative
criterion of properness, after replacing T with a ramified cover, there exists a lift
T → H where t maps to an admissible cover (C ′′ → D). One shows that every
smooth irreducible component of C ′′ has genus 0 as in the proof of Proposition 5.7.26.
Thus C ′ degenerates to the singular stable curve (C ′′ )st , and since (C → P1 ) and
(C ′ → P1 ) are in the same connected component in Hurg,b , the original smooth
curve C also degenerates to (C ′′ )st . (Note that the argument constructing the family
C → D → T of covers in the proof of Proposition 5.7.26 is one of the essential
ingredients in the proof of the properness of Hurg,b .)

5.7.5 Irreducibility in positive characteristic

We prove that Mg is irreducible in positive characteristic following the historical
arguments of Deligne–Mumford [DM69] and Fulton [Ful69]. A central ingredient
in each argument is Zariski’s Connectedness Theorem (4.6.15): for a flat, proper
morphism of noetherian Deligne–Mumford stacks, the number of geometrically
connected components of a fiber is lower semicontinuous. This connectedness
theorem, stated but unproven in [DM69, Thm. 4.17], requires a surprisingly large
amount of the theory of Deligne–Mumford stacks.
Theorem 5.7.33 (Irreducibility of Mg,n ). For g and n satisfying 2g − 2 + n > 0,
Mg,n is irreducible over any field.
Proof. As we already know that Mg,n is irreducible in characteristic 0 by Theo-
rem 5.7.19 (transcendental proof) or Theorem 5.7.30 (algebraic proof), it suffices
to show irreducibility over Fp . The morphism Mg,n → Spec Z is smooth (Theo-
rem 5.4.14) and proper (Theorem 5.5.23), and thus Zariski’s Connectedness Theorem
(4.6.15) implies that all geometric fibers have the same number of connected com-
ponents. As Mg,n is connected over Q, it is also connected, hence irreducible, over
Fp .
Remark 5.7.34 (Deligne and Mumford’s proofs). The above proof was the second
argument given by Deligne and Mumford [DM69, §5]. Keep in mind that this is

275
the same remarkable paper that introduced stacks (now called Deligne–Mumford
stacks), introduced stable curves, and proved Stable Reduction (5.5.1). Their first
irreducibility argument [DM69, §3], which they called an “elementary derivation
of the theorem”, was very similar in spirit to the proof of Theorem 5.7.30 using
the degeneration of a smooth curve to a singular stable curve and the inductive
nature of the boundary. The proof relied on Stable Reduction and the topological
irreducibility of Mg . We now sketch their argument that Mg ×Z k is irreducible for
any field k.
Let Hg (resp., H g ) denote the locally closed subscheme of HilbP (P5g−6Z /Z)
parameterizing smooth (resp., stable) curves. In [GIT, Thms. 5.11, 7.13], Mumford
had constructed the coarse moduli scheme Mg over Spec Z as the geometric quotient
Hg / PGL5g−6 , and had shown that it is quasi-projective over Spec Z[1/p] for every
prime p. (It is also true that Mg admits a projective coarse moduli scheme M g over
Spec Z which is identified with the geometric quotient H g / PGL5g−6 , but this was
only shown later.)
Step 1: No connected component of Mg ×Z k is proper over k. Let W (k) be the
Witt vectors for k; this is a complete noetherian local ring whose generic point η
has characteristic 0 and whose closed point 0 has residue field k. (For example,
W (Fp ) = Zp is the ring of p-adics.) As Mg ×Z W (k) is quasi-projective, we can
choose a projective compactification

Mg ×Z W (k)  /X

' 
Spec W (k)

containing Mg ×Z W (k) as a dense open subscheme. By the characteristic 0 result,

we know that the generic fiber Xη is connected. Since X is flat and proper over
W (k), Zariski’s Connectedness Theorem (4.6.15) for schemes implies that the special
fiber X0 is also connected.
Suppose Y ⊆ Mg ×Z k is a connected component proper over k. Then Y is an
open subscheme of X0 , but it’s also a closed subscheme as Y is proper. Since X0 is
connected, we conclude that Y must be all of Mg ×Z k, hence Mg ×Z k is proper and
irreducible. To obtain a contradiction, denote by Ag,k the moduli space of principally
polarized g-dimensional abelian varieties over k and consider the morphism

Θ : Mg ×Z k → Ag,k , C 7→ Jac(C)

assigning to a smooth curve C its Jacobian Jac(C). The properness of Mg ×Z k

implies that the image is closed, but it was well known at the time that the closure
of the image of Θ contains products of lower dimensional Jacobians.
Step 2: There is no connected component of H g ×Z k consisting entirely of smooth
(1) (r)
curves. Let H g , . . . , H g be the connected components of H g ×Z k. Step 1 implies
(i) (i)
that Hg := H g ∩ (Hg ×Z k) is not proper for each i. Therefore, there is a morphism
(i)
∆ = Spec k((t)) → Hg that does not extend to ∆ = Spec k[[t]]. By Stable Reduction
∗
(i)
(5.5.1), after possibly replacing ∆ with a ramified extension, ∆∗ → Hg extends to a
(i) (i) (i)
morphism ∆ → H g . This shows that H g ∖ Hg is non-empty. In other words, we
have shown that every smooth curve degenerates to a singular stable curves, giving
a proof of Proposition 5.7.26 in positive characteristic.

276
Step 3: The boundary δ ×Z k is connected, where δ = M g ∖ Mg . By Step 2, the
connectedness of M g ×Z k follows from the connectedness of δ ×Z k. We will show
that δi ×Z k is connected for each i and that each pairwise intersection (δi ∩ δj ) ×Z k
is nonempty. This is precisely what we showed in the proof of Theorem 5.7.30
using induction on the genera, the Gluing Morphisms (5.6.16), and the curves in
Figure 5.7.31 in (δi ∩ δj ) ×Z k. Deligne and Mumford gave essentially the same
argument using ad hoc techniques rather than the formalism of the moduli space
M g,n of n-pointed stable curves and the Gluing Morphisms (5.6.16), which were
only introduced later by Knudsen.
Interestingly, neither of Deligne and Mumford’s proofs relies essentially on the
theory of the stacks. Their second proof—see Theorem 5.7.33—is valid as long as
one knows the existence of a proper and flat coarse moduli space M g,n → Spec Z.
However, both proofs rely fundamentally on the compactification Mg and Stable
Reduction (5.5.1).
Remark 5.7.35 (Fulton’s proof). In [Ful69], Fulton studied the variant of the Hurwitz
moduli space Hurrig,unord
g,b parameterizing families C → P1S of simply branched
coverings of degree d over b unordered points (see Remark 5.7.12), and showed that
the diagram

Hurrig,unord
g,b
ét

zz (
Mg Symb P1 ∖ ∆ = Pb ∖ ∆

is defined over Z such that Hurrig,unord

g,b → Mg surjective if d ≥ g + 1 (which is a
better bound than we proved in Corollary 5.7.18) and that Hurrig,unord g,b → M0,n
is finite étale over Zd! , i.e., after inverting all primes p ≤ d. Fulton established a
“reduction theorem”: if X is a smooth and projective scheme over a complete DVR
R with algebraically closed residue field k and characteristic 0 fraction field K and
∆ ⊆ X is relative divisor over R which is simple (i.e., each fiber of ∆ has no multiple
components), and if Y → X ∖ ∆ is a finite étale covering such that YK is irreducible,
then Yk is also irreducible. Using the irreducibility of the Hurwitz moduli space in
characteristic 0, Fulton’s reduction theorem applied to Hurrig,unord
g,b → Symb P1 ∖ ∆
gives the irreducibility of Hurrig,unord
g,b over fields of characteristic p > d. Taking
d = g + 1 gives the irreducibility of Mg over fields of characteristic p > g + 1.

5.8 Projectivity following Mumford

It would not be an exaggeration to say that [Theorem A] has played as
fundamental a role in the theory of algebraic curves in the last thirty
years as the theory of abstract curve did in the preceding sixty.

Joe Harris and Ian Morrison [HM98, p. 48]

Geometric Invariant Theory (GIT) was developed by Mumford in [GIT] as a

method to construct quotients and moduli spaces in algebraic geometry. The central
idea is to represent a moduli stack M (such as Mg ) into a quotient stack [U/G],
where G is a reductive group and U ⊆ P(V ) is a locally closed subscheme of the
projectivization of a G-representation V , and to use the Hilbert–Mumford Criterion

277
(7.4.4) to show that a point u ∈ U is in U if and only if u is GIT stable. Ironically,
Mumford’s first construction of Mg as a scheme over Z [GIT, Thms. 5.11 and 7.13]
used ad hoc techniques rather than the Hilbert–Mumford Criterion. Shortly after,
Mumford [Mum77] and Gieseker [Gie82] constructed M g as a projective variety
using GIT. The first proof of projectivity of M g,n , however, was due to Knudsen
and Mumford in [KM76, Knu83a, Knu83b] by relying on Torelli map M g → Ag to
the Satake compactification of the moduli space of principally polarized abelian
varieties.
In this section, we cover Mumford’s construction of M g over an algebraically
closed field k of characteristic 04 by applying GIT, as developed in Chapter 7, to
Chow and Hilbert schemes.

5.8.1 GIT outline using Chow and Hilbert schemes

The GIT construction of the Hilbert scheme depends on two integers, while the
construction using the Chow scheme depends only on the integer k:

– k ≥ 5, the multiple of the dualizing sheaf defining the pluricanonical embedding

⊗k
|ωC | : C ,→ PNk , where Nk = (2k − 1)(g − 1) − 1.

⊗k
We need k ≥ 3 so that the kth multiple ωC of the dualizing sheaf of a stable
curve C is ample, but we need k ≥ 5 for the GIT construction using the Chow
or Hilbert scheme to yield M g .
– m ≫ 0, the degree of the equations that we use to embed the Hilbert scheme
of k-canonically embedded curves into a Grassmannian. We need m ≫ 0 to
obtain an embedding of the Hilbert scheme.

The Hilbert scheme. For k > 1, the Hilbert polynomial of a k-pluricanonically

embedded stable curve C ⊆ PNk of genus g is
⊗kt
P (t) = χ(C, ωC ) = (2kt − 1)(g − 1).

The Hilbert scheme HilbP (PNk ) is projective over k (Theorem 1.1.2) and inherits
an action of PGLNk +1 . For k ≥ 3, there is a locally closed PGLNk +1 -invariant
subscheme U ⊆ HilbP (PNk ) such that Mg ∼ = [U/ PGLNk +1 ] (Exercise 5.4.15). Let

H := U ⊆ HilbP (PNk ) (5.8.1)

be the closure of U . By Theorem 1.4.5, for m ≫ 0, there is a PGLNk +1 -equivariant

closed immersion

H ,→ Gr P (m), Γ(PNk , O(m))

 
Nk Nk
 
C⊆P 7→ Γ(P , O(m)) ↠ Γ(C, O(m)) .
| {z } | {z }
⊗k ⊗mk
Symm Γ(C,ωC ) Γ(C,ωC ) if C ∈ U

By Proposition 1.2.8, the Grassmannian is embedded into projective space via the
4 See Remark 5.8.7 for how the construction extends to positive and mixed characteristics.

278
Plücker embedding
 
P (m)
^
Gr P (m), Γ(PNk , O(m)) ,→ P  Γ(PNk , O(m))

 
P (m) P (m)
^ ^
Γ(PNk , O(m)) ↠ Γ(C, O(m)) 7→  Γ(PNk , O(m)) ↠
 
Γ(C, O(m)) .

In fact, for any Hilbert scheme HilbP (Pn ), there are PGLn+1 -equivariant closed
immersions
 
P (m)
^
HilbP (Pn ) ,→ Gr(P (m), Γ(Pn , O(m))) ,→ P  Γ(Pn , O(m))

for m ≫ 0, and it will be just as easy to investigate the stability of curves C ⊆ Pn

that are not necessarily pluricanonically embedded. Let OGr (1) be the very ample
line bundle on the Grassmannian, and let

Lm = OGr (1)|HilbP (Pn )

be its restriction, noting its dependence on m. For m ≫ 0, Lm is a very ample

line bundle on HilbP (Pn ) which inherits an action by PGLn+1 -action, and thus also
GLn+1 and SLn+1 . We will later restrict to one-parameter subgroups of SLn+1 , but
it
Pwill be convenient to also consider one-parameter subgroups λ = diag(λi ) with
λi ̸= 0.

Definition 5.8.2. We denote the Hilbert–Mumford index (Definition 7.4.1) of

[C ⊆ Pn ] ∈ HilbP (Pn ) as
µHilb
Lm (C, λ)

with respect to a one-parameter subgroup λ : Gm → GLn+1 .

To offer an explicit
P description of µLm (C, λ), we define the λ-weight of a mono-
Hilb

mial x0 · · · xn as i mi λi and the λ-weight of a polynomial f (x0 , . . . , xm ) as the

m0 mn

largest weight of
Pmonomial in f . If β = {f1 , . . . , fp } is a set of polynomials, we write
λ-weight(β) = i λ-weight(fi ). With these definitions in place,

µHilb
Lm (C, λ) = min λ-weight(β), (5.8.3)
β

where the minimum is over subsets β ⊆ Γ(Pn , OPn (m)) mapping to a basis of
Γ(C, OC (m)).
The Chow scheme. The Chow form

fC ∈ Symd (kn+1 )⊗2

of an integral curve C ⊆ Pn of degree d was defined in §1.4.5: it is a multihomogeneous

polynomial of degree d in the variables u0i and u1i for i = 0, . . . , n with the property
that fC (a0 , b0 , . . . , an , bn ) = 0 if and only if

C ∩ {a0 x0 + · · · + an xn = 0} ∩ {b0 x0 + · · · + bn xn = 0} =
̸ ∅.

279
The definition of a Chow form extends to effective 1-cycles of degree d, and the
Chow scheme  
Chow1,d (Pn ) ⊆ P Symd (kn+1 )⊗2
is defined as the closure of the locus of Chow forms of all effective 1-cycles of degree
d.
Definition 5.8.4. We denote the Hilbert–Mumford index of the Chow form of a
reduced curve C ⊆ Pn of degree d as

µChow (C, λ)

with respect to a one-parameter subgroup λ : Gm → GLn+1 .

Given an integral curve C ⊆ Pn of degree d and its Chow form fC , for each
t ∈ Gm (k), we can write
ν
X
fλ(t)·C = λ(t) · fC = ti fC,i ∈ k[t, ui,j ],
i=µ

where fC,i has λ-weight i and where we assume that fC,µ =

̸ 0. Then limt→0 λ(t)·fC =
fC,µ ∈ Chow1,d (Pn ) and µChow (C, λ) = µ.
Chow, Hilbert, and asymptotic Hilbert stability. We say that curve a C ⊆ Pn
of degree d is
– Chow stable/semistable if its Chow form in Chow1,d (Pn ) is stable/semistable
with respect to the action by SLn+1 .
– m-Hilbert stable/semistable if [C ⊆ Pn ] ∈ HilbP (Pn ) is stable/semistable with
respect to the action by SLn+1 and the line bundle Lm , and
– asymptotically Hilbert stable/semistable if it is m-Hilbert stable/semistable for
m ≫ 0.
The GIT construction M g using the Hilbert and Chow scheme depends on the
following stability analysis.
Theorem 5.8.5. Let k ≥ 5.
⊗k
(1) A smooth curve C ⊆ PNk embedded by the complete linear series |ωC | is Chow
stable.
(2) If [C ⊆ PNk ] ∈ HilbP (PNk ) is a Chow semistable curve contained in the
closure of the locus of smooth k-pluricanonically embedded curves, then C is a
k-pluricanonically embedded (Deligne–Mumford) stable curve.
Moreover, there exists a m ≫ 0 such that every k-pluricanonically embedded smooth
curve is m-Hilbert stable and such that if [C ⊆ PNk ] ∈ H is m-Hilbert semistable,
then C is a k-pluricanonically embedded stable curve.
Some comments are in order:
– Part (1) is conceptually the most challenging ingredient as it requires ver-
ifying the negativity of the Hilbert–Mumford index with respect to every
one-parameter group. Surprising however, this part has a very clean proof due
to Mumford’s ingenuity in reinterpreting the Hilbert–Mumford index in terms
of the multiplicity e(C, λ) of an ideal on A1 × C (see §5.8.2). This multiplicity
can be bounded in terms of multiplicities on C (see Proposition 5.8.19), which
is sufficient to verify the stability of smooth curves (see Theorem 5.8.22).

280
 – Unfortunately, the bounds on the multiplicity are not sufficient to prove the
Chow stability of a k-pluricanonically embedded stable curve. Nevertheless,
as we show shortly, parts (1) and (2) are sufficient to indirectly conclude that
stable curves are Chow stable.
– While (2) is the conceptually easier part (as we just need to exhibit destabilizing
one-parameter subgroups), it is nevertheless the most technically demanding.
In §5.8.5, we verify only some of the details, while referring the reader to more
complete accounts.
Theorem 5.8.5 allows us to wrap up the proof of Theorem A.
Theorem 5.8.6. Over a field k of characteristic 0, the coarse moduli space M g,n is
projective if 2g − 2 + n > 0.
Proof. By Theorem 5.5.23, Mg,n is a proper Deligne–Mumford stack admitting a
proper coarse moduli space M g,n . Since Mg,n+1 → Mg,n is the universal family
(Proposition 5.6.12), it is a family of n-pointed stable curves, and hence there is
a line bundle L onP Mg,n+1 which is relatively ample over Mg,n . (One can take
L = ωMg,n+1 /Mg,n ( σi ) where σi are the universal sections.) By Exercise 4.4.29(2),
L descends to a line bundle on M g,n+1 which is relatively ample over M g,n . It thus
suffices to show that M g is projective for g ≥ 2. Note that we already know that
M 0,3 = Spec k and M 1,1 = P1 are projective.
For k ≥ 5, let P (t) = (2kt − 1)(g − 1) be the Hilbert polynomial of a k-
⊗k
pluricanonically embedded stable curve C of genus g and let Nk + 1 = h0 (C, ωC )=
(2k − 1)(g − 1). Using Exercise 5.4.15, we can write Mg ∼ = [U/ PGLNk +1 ], where
U ⊆ HilbP (PNk ) is a locally closed PGLNk +1 -invariant subscheme. Letting H = U ⊆
HilbP (PNk ), we need to show that there is an integer m ≫ 0 such that U = HLs m ,
where HLs m is stable locus with respect to the action of PGLNk +1 and the line
bundle Lm . We first note that GIT stability with respect to PGLNk +1 is the same
as stability with respect to SLNk +1 as the map SLNk +1 → PGLNk +1 has finite
cokernel: they have the same one-parameter subgroups up to scaling and therefore
the Hilbert–Mumford Criterion (7.4.4) holds for PGLNk +1 if and only if it holds for
SLNk +1 .
After applying Theorem 5.8.5, it remains to show that every k-pluricanonically
embedded stable curve C is m-Hilbert stable. Let C → Spec R be a generically
smooth family of stable curves over a DVR R such that the central fiber C0 is
isomorphic to C. Since the generic fiber is smooth, it is also m-Hilbert stable. By
semistable reduction in GIT (Remark 7.1.4), after possibly replacing R with an
extension, there exists a family C ′ → Spec R of m-Hilbert semistable curves such
that the generic fibers CK and CK ′
are isomorphic. Since C0′ is m-Hilbert semistable,
it is stable. As C and C are two families of stable curves that are generically
′

isomorphic, the separatedness of Mg (Proposition 5.5.20) implies that C ∼ = C ′ . Thus

C = C0 = C0 is m-Hilbert stable. This establishes the identity U = HLs m . We
′

now apply the first fundamental theorem of GIT (Theorem 7.2.6) to conclude that
there is a geometric quotient U → U/ PGLNk +1 with U/ PGLNk +1 projective. In
other words, Mg = [U/ PGLNk +1 ] → U/ PGLNk +1 is a coarse moduli space. By
uniqueness of coarse moduli spaces, M g = U/ PGLNk +1 is projective.
This argument can also be made using the Chow scheme Chow1,d (PNk ) with
d = 2k(g − 1) and the closure Z ⊆ Chow1,d (PNk ) of the locus of k-pluricanonically
embedded smooth curves. One can write Mg = [U/ PGLn+1 ] where U ⊆ Z is the
locus of k-pluricanonically embedded Deligne–Mumford stable curves; this follows for
instance from the fact that the Hilbert-to-Chow morphism Z → H is an isomorphism

281
on the locus of nodal curves,. The same argument shows that U = Z s is the stable
locus and that Mg → Z s / PGLn+1 has a projective coarse moduli space. See also
[Mum77, Cor. 5.2], [Gie82, Thm. 2.0.2], and [HM98, Cor. 4.42].
Remark 5.8.7. The restriction to characteristic 0 in Theorem 5.8.6 is because we
have only developed GIT in Chapter 7 for linearly reductive groups. GIT applies
equally to reductive groups in positive characteristic (see Remark 6.5.12) and in
fact to reductive group schemes over Z [Ses77]. As Theorem 5.8.5 holds in arbitrary
characteristic, this allows one to use GIT to construct M g as a projective scheme
over Z.
GIT constructions of M g,n directly verifying the Chow stability of a Deligne–
Mumford stable curve have been completed more recently in [BS08] and [LW15],
which also handle the case of marked points. The stability of arbitrary polarized
curves C ⊆ Pn , i.e., curves that are not necessarily k-pluricanonically embedded,
has been analyzed in [Cap94] and [BFMV14].

5.8.2 Interpretation of the Hilbert–Mumford indices

We identify the Hilbert–Mumford indices µChow (C, λ) and µHilb Lm (C, λ) with other
algebro-geometric invariants. This leads us to a useful criterion for Chow or Hilbert
stability (Proposition 5.8.17), which we employ in the next section to prove the
stability of smooth curves. For simplicity, we keep the exposition below focused on
dimension one, even though much of it extends to arbitrary dimension.
The ideal sheaf Jλ ⊆ OC×A1 . Let C ⊆ Pn be a curve; in applications later, we will
take n = Nk to be the dimension of the projective space of the kth pluricanonically
embedding. Let λ : Gm → GLn+1 be a one-parameter subgroup defined by λ(t) =
diag(tλ0 , . . . , tλn ) for integers λ0 ≥ · · · ≥ λn = 0. Given coordinates xi on Pn and t
on A1 , we define the ideal sheaf

Jλ = (tλi xi ) ⊆ OC×A1 .

The rational map A1 × C 99K A1 × Pn , where (t, x) 7→ λ(t) · x is resolved by

blowing up Jλ , leading to a commutative diagram

BlJλ (C × A1 )
v '  / 1
A1 × C X A × Pn
( w
A1 ,

where X is defined as the scheme-theoretic image of the map BlJλ (C ×A1 ) → A1 ×Pn .
The projection X → A1 is flat by construction, and the central fiber [X0 ⊆ Pn ] is
identified with limt→0 λ(t) · [C ⊆ Pn ] ∈ HilbP (Pn ), which follows from the proof of
the valuative criterion for properness of HilbP (Pn ) as in Remark 1.4.4. Observe that
X0 is fixed by Gm , and that Γ(X0 , OX0 (m)) inherits a Gm -action for each m.
Asymptotic weights of the Hilbert function. The λ-weight of a GLn+1 -
representation V is λ-weight(V ) = i i · dimk Vi , where Vi is the ith eigenspace of
P
the induced Gm -action on V .
Definition 5.8.8 (r(C, λ)). Let C ⊆ Pn be a curve and λ : Gm → GLn+1 be a
one-parameter subgroup defined by λ(t) = diag(tλ0 , . . . , tλn ) for integers λ0 , . . . , λn .

282
We define the invariant r(C, λ) as the normalized leading coefficient of the quadratic
polynomial
m2
λ-weight(Γ(X0 , OX0 (m))) = r(C, λ) + c1 m + c0
2
for m ≫ 0.
The fact that r(C, λ) is a quadratic polynomial for m ≫ 0, and therefore that
r(C, λ) is well-defined, is a consequence of Theorem 5.8.11. The definition of r(C, λ)
is analogous to the Futaki invariant in K-stability, and it turns out that K-stability
⊗k
corresponds to asymptotic Chow stability, i.e., the Chow stability of |ωC | : C ⊆ PNk
for k ≫ 0.
Multiplicity of contact with a weighted flag. If X is a separated scheme of
finite type over k of dimension 2, J ⊆ OX is a sheaf of ideals defining a subscheme
C ⊆ X proper over k, and L is a line bundle on X , then the Euler characteristic
m2
χ(X , L⊗m /J m L⊗m ) = eL (J) + (lower terms) (5.8.9)
2
is a polynomial of degree 2 with rational coefficients. We define the multiplicity of
J measured by L as the normalized leading coefficient eL (J). This is a special case
of the Hilbert–Samuel polynomial ; see [Ram74], [Mum77, Prop. 2.1], and [Eis95,
§12.1].
If J = 0 and L is very ample on X , then χ(X , L⊗m /J m L⊗m ) = χ(X , L⊗m )
is the Hilbert polynomial and eL (0) is the degree of X . Similarly, if J = 0 and
L = p∗ OPm (1) is the pullback under a map p : X → Pn , then eL (0) is the degree of
the Chow cycle defined by the image p(X ) (with the coefficients of the Chow cycle
being defined as usual using the degrees of the induced function field extensions).
On the other hand, if J is the ideal sheaf of a subscheme supported at a point, then
eL (J) is the multiplicity of the singularity. See also Proposition 5.8.21 for how the
multiplicity measures the difference of degrees under a linear projection.
Since the ideal sheaf Jλ = (tλi xi ) is only defined if each λi ≥ 0, the definition of
e(C, λ) below also requires that each λi ≥ 0.
Definition 5.8.10 (e(C, λ)). Let C ⊆ Pn be a curve and λ : Gm → GLn+1 be a
one-parameter subgroup defined by λ(t) = diag(tλ0 , . . . , tλn ) for integers λ0 ≥ · · · ≥
λn ≥ 0. We define
e(C, λ) = eOA1 ⊠OC (1) (Jλ )
as the multiplicity of the ideal sheaf Jλ = (tλi xi ) measured by OA1 ⊠ OC (1).
The one-parameter subgroup λ defines a weighted flag
{x1 = · · · = xn = 0} ⊆ {x2 = · · · = xn = 0} ⊆ · · · ⊆ {xn = 0} ⊆ Pn .
| {z } | {z } | {z }
L0 , weight λ0 L1 , weight λ1 Ln−1 , weight λn−1

Roughly speaking, the closed subscheme A1 × C defined by the ideal Jλ = (tλi xi ) is

a union of the λi th-order thickenings of Li ∩ C, and e(C, λ) measures the degree of
contact between this weighted flag and C.
Equivalent interpretations of the Hilbert–Mumford indices.
Theorem 5.8.11. Let C ⊆ Pn be a curve and λ : Gm → GLn+1 be a one-parameter
subgroup defined by λ(t) = diag(tλ0 , . . . , tλn ) for integers λ0 ≥ · · · ≥ λn . Then
µChow (C, λ) = r(C, λ). Moreover, if each λi ≥ 0, then
µChow (C, λ) = r(C, λ) = e(C, λ).

283
Proof. To show that µChow (C, λ) = r(C, λ), we will reduce to case of a union of lines
and then the simpler case of a line. First, if C is a line, we can change coordinates
to assume that C = V (x2 , . . . , xn ), in which case we can directly compute that its
Chow form fC is u00 u11 − u01 u10 and µChow (C, λ) = λ-weight(fC ) = λ0 + λ1 . Since
2
1 ⟩, its λ-weight is (λ0 + λ1 )
m−1 m+1
= (λ0 + λ1 ) m2 +


Γ(C, O(m)) = ⟨xm 0 , x0 x1 , . . . , x m 2
(lower terms). Thus r(C, S λ) = λ0 + λ1 = µ
Chow
(C, λ). If C is set-theoretically
contained in the union ij Lij of the lines Lij = V (x0 , . . . , x bj , . . . , xn ). Let
bi , . . . , x
aij be the multiplicity of C at the generic point of Lij . Note that µChow (C, λ) and
r(C, λ) only depend on the scheme structuresQat the generic points of Lij and not
a
on the embedded points of C. Since fC = ij fLij ij
, we see that µChow (C, λ) =
Chow
(Lij , λ). We leave the equality r(X, λ) = ij aij r(Lij , λ) to the reader.
P P
ij aij µ
The case of a line then implies that µChow (C, λ) = r(C, λ).
To reduce to the case of a union of lines, observe that since both µChow (C, λ)
and r(C, λ) are determined by the limit curve limt→0 λ(t) · [C ⊆ Pn ] ∈ HilbP (Pn ),
we may assume that C is fixed by λ. The maximal torus Gn+1 m ⊆ GLn+1 acts on the
proper scheme HilbP (Pn ), where P is the Hilbert polynomial of C. Since Gn+1 m is
solvable, the Borel Fixed Point Theorem (c.f., [Mil17, Cor. 17.3]) implies that the
closure of the orbit Gn+1 m · [C ⊆ Pn ] contains a Gn+1 m -fixed point [X0 ⊆ P
n+1
], which
is necessarily contained set-theoretically in the union of coordinate lines. This yields
a diagram

X / Pn × A1

π
 {
A1
where X ⊆ Pn × A1 is a Gm -invariant closed subscheme (under the λ action on
Pn and the trivial action on A1 ) and X → A1 is a Gm -equivariant flat family such
that the fiber over 1 is C and over 0 is X0 . We leave the reader to verify that
µ(C, λ) = µ(X0 , λ) and r(C, λ) = r(X0 , λ).
We now show that e(C, λ) = r(C, λ) assuming that each λi ≥ 0. We can write
M
X = Proj Rm ,
m≥0

where Rm is the k[t]-submodule of k[t] ⊗k H0 (C, OC (m)) generated by degree m

polynomials in tλi xi . By definition, r(C, λ) is computed as the normalized leading
coefficient of a quadratic polynomial

r(C, λ) = norm. lead. coef. of λ-weight H0 (X0 , OX0 (m))

= norm. lead. coef. of λ-weight(Rm /tRm ) (5.8.12)

= norm. lead. coef. of dimk (k[t] ⊗k H (C, OC (m))/Rm ),
0

where the last equality follows the identification from Lemma 5.8.14 that λ-weight(Rm /tRm ) =
dimk (k[t] ⊗k H0 (C, OC (m))/Rm ) for all m. On the other hand, letting X = A1 × C
and L = OA1 ⊠ OC (1), the multiplicity e(C, λ) = eL (Jλ ) is defined as the normalized
leading coefficient of χ(X, L⊗m /Jλm L⊗m ). We leave the reader to verify that the
quantity can be computed as

e(C, λ) = norm. lead. coef. of h0 X, L⊗m /Jλm L⊗m

= norm. lead. coef. of dimk H0 (X, L⊗m )/H0 (X, Jλm L⊗m ) (5.8.13)
= norm. lead. coef. of dimk k[t] ⊗k H0 (C, OC (m))/Rm .

284
It follows that r(C, λ) = e(C, λ). See also [Mum77, Prop. 2.6 and Thm. 2.9].
A key ingredient of the proof above was the identification of the quantity
λ-weight(Rm /tRm ), which depends on the limit curve X0 , and the quantity dimk (k[t]⊗k
Γ(C, OC (m))/Rm ), defined on the surface A1 × C.
Lemma 5.8.14. Let V be a Gm -representation of dimension n with weights λn ≥
· · · λ0 ≥ 0. Let Vi be the eigenspace of weight λi , and define R ⊆ k[t] ⊗ V as the
k[t]-submodule generated by tλi Vi for i = 0, . . . , n. Then


λ-weight(R/tR) = dimk (k[t] ⊗ V )/R .
Proof. Writing k[t] ⊗ V = i,d td Vi , R/tR is freely generated by tλi Vi , while (k[t] ⊗
L

V )/R is freely generated by Vi , tVi , . . . , tλi −1 Vi . Thus

X

λ-weight(R/tR) = λi dim Vi = dimk (k[t] ⊗ V )/R .
i

Exercise 5.8.15 (details). Fill in the missing details in the proof of Theorem 5.8.11.
The Hilbert–Mumford index µHilb
Lm (C, λ) also has a nice interpretation, allowing
us to compare it with µChow
(C, λ).
Proposition 5.8.16. Let C ⊆ Pn be a curve and λ : Gm → GLn+1 be a one-
parameter subgroup defined by λ(t) = diag(tλ0 , . . . , tλn ) for integers λ0 ≥ · · · ≥ λn ≥
0. Then for m ≫ 0
µHilb 0 1


Lm (C, λ) = dimk H A × C, OA1 ⊠ OC (m) /Rm ,

where Rm is the subspace generated by monomials of degree m in tλi xi , while

µChow (C, λ) = norm. lead. coef. of dimk H0 A1 × C, OA1 ⊠ OC (m) /Rm .

In particular, µChow (C, λ) is the normalized leading coefficient of µHilb

Lm (C, λ).

Proof. Letting t be the coordinate on A1 , we can write

M
Rm = Rm,i · ti ⊆ H0 (C, OC (m)) · ti = H0 (A1 × C, OA1 ⊠ OC (m)).
i≥0

By construction, we have inclusions

Rm,0 ⊆ Rm,1 ⊆ · · · ⊆ Rm,I = H0 (C, OC (m))
for I ≫ 0. We use the description from (5.8.3) of µHilb
Lm (C, λ) as the minimum of
λ-weight(β) over subsets β ⊆ H0 (Pn , OPn (m)) mapping to a basis of H0 (C, OC (m)).
A basis of minimum weight corresponds to first choosing a basis of Rm,0 and
successively extending it to each Rm,i . Therefore,

µHilb
Lm (C, λ) = 0 · dimk Rm,0 + 1 · (dimk Rm,1 − dimk Rm,0 ) + 2 · (dimk Rm,2 − dimk Rm,1 ) + · · ·
= I · dimk H0 (C, OC (m)) − dimk Rm,0 − · · · − dimk Rm,I−1
I−1
X
= dim H0 (C, OC (m))/Rm,i
i=0
= dimk H 0 A1 × C, OA1 ⊠ OC (m) /Rm .

The final statement follows from the equalities of (5.8.12). See also [Mor80, Prop. 3.2].

285
5.8.3 Criteria for Chow and Hilbert stability
Proposition 5.8.17 (Criterion for Chow Stability). A curve C ⊆ Pn is Chow stable
(resp., semistable) if and only if r(C, λ) < 0 (resp. ≤) for every one-parameter
subgroup λ : Gm → SLn , or equivalently if
2 deg C X
e(C, ρ) < ρi (resp., ≤)
n+1 i

for every one-parameter subgroup ρ : Gm → GLn with weights ρ0 ≥ · · · ≥ ρn = 0.

Proof. The first equivalence follows directly from the equality µChow (C, λ) = r(C, λ)
in Theorem 5.8.11 and the Hilbert–Mumford Criterion (7.4.6). Moreover, we know
that r(C, λ) is the normalized leading coefficient of the λ-weight of Γ(X0 , OX0 (m)),
where X0 is the limit of C under λ in the Hilbert scheme. After a change of basis,
we can write
ρ 
t 0 P
i ρi
α . .. with α = −
λ(t) = t  n+1

and ρ0 ≥ · · · ≥ ρn = 0

tρn
| {z }
ρ(t)

for a one-parameter subgroup ρ : Gm → GLn . Comparing weights under λ and ρ,

we have
λ-weight (Γ(X0 , OX0 (m))) = ρ-weight (Γ(X0 , OX0 (m))) + αm dimk (Γ(X0 , OX0 (m)))
As the leading coefficient of dimk (Γ(X0 , OX0 (m))) is deg(C), we obtain that
r(C, λ) = r(C, ρ) + 2 deg(C)α
2 deg(C) X
= r(C, ρ) − ρi .
n+1 i

Proposition 5.8.18. Let C ⊆ Pn be a curve. If C is Chow stable (resp., not

semistable), then C is asymptotically Hilbert stable (resp., not semistable).
Proof. If C is Chow stable (resp., not semistable), then µChow (C, λ) < 0 (resp., >).
By Proposition 5.8.16, the normalized leading coefficient of the quadratic polynomial
Lm (C, λ) is strictly negative (resp., strictly positive), and therefore µ
µHilb Hilb
(C, λ) < 0
(resp. >) for d ≫ 0.
In other words, we have the implications:
Chow stable =⇒ asymptotically Hilbert stable =⇒ asymptoti-
cally Hilbert semistable =⇒ Chow semistable.

5.8.4 Chow stability of smooth curves

To verify that the Chow form of a smooth curve is stable, we will exploit the following
bound.
Proposition 5.8.19. Let C ⊆ Pn be a smooth curve and λ : Gm → GLn+1 be a
one-parameter subgroup defined by λ(t) = diag(tλ0 , . . . , tλn ) for integers λ0 ≥ · · · ≥
λn = 0. For k = 0, . . . , n, let Ik ⊆ OC be the ideal generated by xk , . . . , xn . Then
n−1
X

e(C, λ) ≤ (λk − λk+1 ) eOC (1) (Ik ) + eOC (1) (Ik+1 ) .
k=0

286
Proof. As before, we set X = A1 × C, L = OA1 ⊠ OC (1), and Jλ ⊆ OX be the ideal
generated by tλi xi . Recall from the identities (5.8.13) that e(C, λ) = eL (Jλ ) is the
normalized
P λkleading coefficient of dimkλ0H (X, L )/H (X, Jλ L m ). Observe
0 ⊗m 0 m ⊗m
that
Jλ = k t Ik and that it contains t · OC . Setting Jm,d := J λ ∩ t d
· OC ⊆ OC ,
we have that Jm,mλ0 = OC . The expansion Jλm = ( k tλk Ik )m contains the sum
P

Inm · (t0 + · · · + tλn−1 −1 ) + Inm−1 In−1 · (tλn−1 + · · · + t2λn−1 −1 ) + · · · +

m
In−1 · (tmλn−1 + · · · + t(m−1)λn−1 +λn−2 −1 ) + · · · +
I1 I0m−1 · (tλ1 +(m−1)λ0 + · · · + tmλ0 −1 ) + OC · (tmλ0 + · · · ).

In other words, we have

 
n−1 λn−k−1 −λn−k −1
M M m−1
M m−j j
M
Jλm = td ·Jm,d ⊇ In−k In−k−1 · tλn−k (m−j)+λn−k−1 j+ν · OC  .
d≥0 k=0 j=0 ν=0

Each integer d = 0, . . . , λ0 m−1 can be written uniquely as λn−k (m−j)+λn−k−1 j +ν,

m−j j
and Jm,d ⊇ In−k In−k−1 . Thus,
 
n−1
X m−1
X m−j j
dimk H0 (X, L⊗m )/H0 (X, Jλm L⊗m ) ≤ h0 C, OC (m)/(In−k


(λn−k − λn−k−1 ) In−k−1 OC (m)) 
k=0 j=0
 
n−1
X m−1
X
h0 C, OC (m)/(Ikm−j Ik+1
j

≤ (λk − λk+1 ) OC (m))  .
k=0 j=0

If Z = V (I) ⊆ C is a finite subscheme defined by an ideal sheaf I ⊆ OC , then the

multiplicity mOC (1) (I) is equal to the length dimk Γ(Z, OZ ) of Z, which is also equal
to Γ(C, OC (m)/IOC (m)) for m ≫ 0. Therefore, using that C is smooth, we have
that

h0 C, OC (m)/(Ikm−j Ik+1j

OC (m)) = (m − j)eOC (1) (Ik ) + jeOC (1) (Ik+1 )

for m ≫ 0. The sum of these dimensions over j = 0, . . . , m − 1 is a quadratic poly-

nomial in m with leading term (eOC (1) (Ik ) + eOC (1) (Ik+1 ))m2 /2 and the proposition
follows. See [Gie82, Thm. 1.0.0] and [Mum77, Prop. 4.10-11].
Remark 5.8.20. A slightly more involved argument establishes the proposition for
arbitrary curves (no smoothness required), but the bounds are unfortunately not
sufficient to verify the Chow stability of a stable curve. There are also similar
bounds for higher dimensional varieties, but they involve the mixed multiplicities
eOC (1) (Irk , Irk+1 ).
As a consequence of Proposition 5.8.19, we have an upper bound for e(C, λ)
depending on multiplicities on the curve C rather than the surface X = A1 × C.
Multiplicities have the following convenient geometric interpretation, which we state
and prove in arbitrary dimension.
Proposition 5.8.21. Let Y ⊆ Pn be an integral subscheme. Let Λ ⊆ Γ(Pn , OPn (1))
be a linear subspace of codimension r, let LΛ ⊆ Pn be the r-dimensional hyperplane
defined by Λ, and let IΛ ⊆ OY be the ideal sheaf of Z := Y ∩ LΛ . If

pΛ : Y ∖ Z → Pr

287
denotes the induced projection and pΛ (Y ) denotes the Chow cycle defined by the
image of pΛ , then
eOY (1) (IΛ ) = deg Y − deg pΛ (Y ).

Proof. Let H be the divisor class of OY (1). There is a commutative diagram

BlIΛ (Y )
q
π
 #
Y
pΛ
/ Pr

where π is the blowup morphism and q is the morphism extending pΛ defined by

π −1 (H) − E with E the exceptional divisor. The degree of Y is

deg(Y ) = norm. lead. coef. of χ(OY (m)),

while the degree of pΛ (Y ) is

deg pΛ (Y ) = norm. lead. coef. of χ (BlIΛ (Y ), q ∗ OPr (m))

= norm. lead. coef. of χ (BlIΛ (Y ), π ∗ OY (m)(−mE))
= norm. lead. coef. of χ (Y, IΛm OY (m)),

where in the final equality we used the identification H0 (BlIΛ (Y ), π ∗ OY (m)(−mE)) =

H0 (Y, IΛm OY (m)) along with the consequence of Asymptotic Riemann–Roch (B.2.22)
that the dimension of the higher cohomology of each sheaf is O(mdim(Y )−1 ). We
conclude that

eOY (1) (IΛ ) = norm. lead. coef. χ(Y, OY (m)/IΛm OY (m))

= norm. lead. coef. χ(Y, OY (m)) − norm. lead. coef. χ(Y, IΛm OY (m))
= deg Y − deg pΛ (Y ).

See also [Mum77, Prop. 2.5].

Theorem 5.8.22. Let C be a smooth, connected, and projective curve of genus

g ≥ 1 over an algebraically closed field k. If C ⊆ Pn is embedded by a complete
linear series of degree d > 2g, then C ⊆ Pn is Chow stable and thus asymptotically
Hilbert stable.

Proof. The proof of Chow stability breaks down into two steps:
(a) A smooth curve C ⊆ Pn embedded by a complete linear series of degree d > 2g
is linearly stable, i.e., for every linear projection pΛ : Pn ∖ LΛ 99K Pr ,

deg pΛ (C) deg C

> ,
r n
where pΛ (C) denotes the image Chow cycle. Using the identity eOC (1) (IΛ ) =
deg C − deg pΛ (C) of Proposition 5.8.21, where IΛ is the ideal sheaf of C ∩ LΛ ,
this is equivalent to
n−r
eOC (1) (IΛ ) < deg C.
n
(b) A linearly stable curve C ⊆ Pn is Chow stable.

288
 For (a), we consider all morphisms f : C → PN whose images are not contained in
a hyperplane, and we consider the corresponding pairs (D, N ), where D is the degree
of the image Chow cycle f (C). If f ∗ OPN (1) is non-special, i.e., h1 (C, f ∗ OPN (1)) = 0,
then by Riemann-Roch (5.1.2),

N = h0 (OPN (1)) − 1 ≤ h0 (f ∗ OPN (1)) − 1 = deg φ∗ OPN (1) − g = D − g.

If f ∗ O(1) is special, Clifford’s Theorem (c.f., [Har77, Thm. IV.5.4]) implies that

N = h0 (OPN (1)) − 1 ≤ h0 (f ∗ OPN (1)) − 1 ≤ (deg φ∗ O(1))/2 = D/2.

Since C ⊆ Pn is embedded by a complete linear series of degree d > 2g, OC (1)

is non-special and Riemann-Roch implies that n + 1 = h0 (C, OC (1)) = d + 1 − g,
so that n = d − g > g. As illustrated by Figure 5.8.23, the point (d, n) lies on the
line N = D − g to the right of (2g, g). It follows that the slope of the line to (d, n)
is greater than the slope to any point in the shaded area, or, in other words, that
D/N > d/n for every map f : C → PN of degree D whose image is not contained in
a hyperplane. In particular, this implies that C is linearly stable.

N Dg
din

N DL

29,9

g d D

Figure 5.8.23: The slope of the line to (d, n) is greater than the slope to any point
in the shaded areas.
NED
g OR NEDA
FFial special
For (b), if λ : Gm → GLn+1 be a one-parameter subgroup, we can choose coor-
dinates x0 , . . . , xWe assume C P complete linear series w 4721
n on P with weights λ0 ≥ · · · ≥ λn = 0. For each k = 0, . . . , n,
n

the subspace ⟨xk , . . . , xn ⟩ ⊆ Γ(C, O c defines 0

hC (1)) a hyperplane Lk = V (xk , . . . , xn )
n'God 1
and induces a linear projection Pn ∖ Lk → Pn−k . The ideal sheaf Ik ⊆ OC defining
C ∩ Lk is generated by xk , . . . , xn . Thedeglinear 2 440
29stability of C ⊆ Pn implies
impling that
So dell is nonspecial and RR
289
n 6,041 n d
not
dttg g
eOC (1) (Ik ) < k
n deg C. The bound of Proposition 5.8.19 yields
n−1
X

e(C, λ) ≤ (λk − λk+1 ) eOC (1) (Ik ) + eOC (1) (Ik+1 )
k=0
n−1  
2 deg C X 1
< (λk − λk+1 ) k +
n 2
k=0
 
2 deg C 1 1
= λ0 + λ1 + · · · + λn−1 + λn
n 2 2
! !
2 deg C X 1
= λi − (λ0 + λn )
n i
2
! !!
2 deg C X 1 X
≤ λi − λi
n i
n+1 i
2 deg C X
= λi .
n+1 i

We may now apply the Criterion for Chow Stability (5.8.17) to conclude that C ⊆ Pn
is Chow stable, and Proposition 5.8.18 to conclude that it is asymptotically Hilbert
stable.
As a corollary, we can deduce that a k-pluricanonical embedded smooth curve is
Chow stable.
⊗k
Proof of Theorem 5.8.5(1). The k-pluricanonical embedded of a smooth curve |ωC |: C ⊆
P Nk
has degree d = 2k(g − 1) with Nk + 1 = (2k + 1)(g − 1). If k ≥ 5 (in fact
k ≥ 3 suffices), then d > 2g and Theorem 5.8.22 implies that C ⊆ PNk is Chow
stable and asymptotically Hilbert stable. With more work, it is possible to exhibit
a sufficiently large m such that every k-pluricanonical embedded smooth curve is
m-Hilbert stable; see [Gie82, Thm. 1.0.0].
Exercise 5.8.24. Show that a curve C ⊆ Pn with a node is not linearly stable.
Remark 5.8.25. There are smooth surfaces that are not Chow stable, e.g., the Steiner
surface defined by the image of P2 → P4 , [x : y : z] 7→ [xz : yz : x2 : xy : y 2 ] [Mor80,
Ex. 3.6].

5.8.5 Destabilizing non-nodal curves

Proposition 5.8.26. Let C ⊆ Pn be a generically reduced curve satisfying deg C 8
n+1 < 7 .
n
If C ⊆ P is Chow semistable, then C is not contained in a hyperplane and C has
at worst nodal singularities.
Proof. If C is contained in a hyperplane H, then we can choose coordinates x0 , . . . , xn
such that H = V (x0 ). Letting λ(t) = diag(t−n , t, . . . , t), the Hilbert–Mumford index
µChow (C, λ) is strictly positive and hence C is not Chow semistable.
Suppose that p ∈ C is a singularity with multp (C) ≥ 3. Choose coordi-
nates x0 , . . . , xn with p = (1, 0, . . . , 0), and define λ(t) = (t, 1, . . . , 1). Then
Jλ = (tx0 , x1 , . . . , xn ) is the maximal ideal of (0, p) ∈ A1 × C and
16 X deg C X
e(C, λ) = mult(0,p) A1 × C = multp C ≥ 3 > λi > 2 λi .
7 i n+1 i

290
The Criterion for Chow Stability (5.8.17) shows that C is Chow non-semistable.
Suppose that p ∈ C is a double point that is not a node. Letting L ⊆ OC be
the ideal defining the reduced tangent line at p, then m2p ⊊ m2p + L ⊊ mp , and
dimk mp /m2p = 2. We can choose coordinates x0 , . . . , xn such that (i) x0 (p) ̸= 0,
(ii) v = x1 /x0 and u = x2 /x0 span mp /m2p , (iii) u ∈ L with u2 ∈ m3p , and
(iv) x3 /x0 , . . . , xn /x0 ∈ m2p . Setting λ(t) = diag(t4 , t2 , t, 1, . . . , 1), then Jλ =
(t4 x0 , t2 x1 , tx2 , x3 , . . . , xn ) defines a closed subscheme of A1 × C supported at (0, p).
Thus e(C, λ) = e(Jλ ) is the Hilbert-Samuel multiplicity of (0, p) and it doesn’t
depend on whether we compute it using the line bundle OA1 ⊠ OC (1) or OA1 ×C .
Defining J0 = (t4 , t2 v, tu, m2p ), then Jλ ⊆ J0 and thus e(Jλ ) ≥ e(J0 ). Consider
J1 := (t4 , m2p ) ⊆ J0 . Since

(t2 v)2 = t4 v 2 ∈ J12 and (tu)4 = t4 (u2 )2 ∈ t4 (m3p )2 ⊆ J14 ,

we have that J0m = J1m for m ≫ 0 and thus e(J0 ) = e(J1 ). We conclude that
16 X deg C X
e(C, λ) = e(Jλ ) ≥ e(J0 ) = e(J1 ) = 4 · 2 · multp C = 16 = λi > 2 λi ,
7 i n+1 i

and applying the Criterion for Chow Stability (5.8.17) again implies that C is Chow
non-semistable. See also [Mum77, Prop. 3.1] and [Gie82, Prop. 1.0.4].
Remark 5.8.27. Interestingly, there are low degree Chow stable curves with singu-
larities that are worst than nodes, but these curves become Chow non-semistable
when re-embedded under a higher degree embedding. For example, there are Chow
stable cuspidal curves C ⊆ P2 of degree 4 (see Exercise 7.5.4) which are Chow
non-semistable when when re-embedded by the degree 8 line bundle OC (2).
Proof of Theorem 5.8.5(2). Observe that if C ⊆ Pn is a stable curve embedded by
⊗k
the complete linear series |ωC | for k ≥ 5, then

deg C 2k(g − 1) 2k 8
= = < .
n+1 (2k − 1)(g − 1) 2k − 1 7
Thus Proposition 5.8.26 applies to show that if C is generically reduced and Chow
semistable, then it is not contained in a hyperplane and has at worst nodal singular-
ities. While this goes part of the way to establishing that C is a k-pluricanonically
embedded stable curve, a rather lengthy and intricate analysis is needed to destabilize
curves that are either not generically reduced or not k-pluricanonically embedded. We
refer the reader to [Mum77, §5], [Gie82, §1], and [HM98, Thm. 4.45] for details.

5.9 Projectivity following Kollár

Still you may have a sentimental attachment to familiar old varieties.
It would appear especially that projective varieties play such a central
technical role in algebraic geometry that it may be virtually impossible to
eliminate their use even if you wanted to. In any case, it is very interesting
to prove, when possible, that [the space] is a projective variety.

Mumford [Mum76, p. 441]

We offer a second proof that M g,n is projective following the approach introduced
by Kollár in [Kol90], which builds on earlier ideas of Viehweg [Vie95]. To introduce

291
the general strategy for projectivity, we need some terminology. Let π : Ug → Mg be
the universal family over Spec Z. For each integer k ≥ 1, define the kth pluricanonical
bundle as
π∗ (ωU⊗k/M )
g g

on Mg , which by Proposition 5.3.23 is a vector bundle of rank g if k = 1 or rank

(2k − 1)(g − 1) if k > 1. The determinants

λk := det π∗ (ωU⊗k/M )
g g

are line bundles on Mg . For k = 1, this is the Hodge line bundle from Defini-
tion 5.6.24.
Projectivity strategy: Show that for k ≫ 0, a positive power of λk descends
to an ample line bundle on the coarse moduli space M g .

5.9.1 Kollár’s Criterion and Projectivity of M g,n

Kollár’s strategy is to employ the multiplication map on pluricanonical bundles

W := Symm π∗ (ωU⊗k/M ) → π∗ (ωU⊗mk ) =: Q, (5.9.1)

g g g /Mg
| {z }
V

depending on the very same two integers k and m as in the GIT construction using the
Hilbert scheme. The fiber of the multiplication map over a stable curve C ∈ Mg (k)
⊗k ⊗mk
is the map Symm H0 (C, ωC ) → H0 (C, ωC ), whose kernel consists of degree m
⊗k ⊗k ⊗k
equations cutting out the image of |ωC | : C → P(H0 (C, ωC )). If k ≥ 3, ωC is
very ample and C can be recovered from the kernel of the multiplication map (5.9.1)
for m ≫ 0, and thus the multiplication map (5.9.1) is surjective. Let w = rk W ,
q = rk Q, and v = rk V . Since W is expressed as the symmetric product Symm V ,
W has a reduction of the structure group by GLv → GLw (see Definition B.1.62).
⊗k ∼ v
For a stable curve C, fixing a basis H0 (C, ωC ) = k defines a quotient [Symm kv ↠
⊗mk
H (C, ωC )] ∈ Gr(q, Sym k ), and modifying the choice of basis changes the quo-
0 m v

tient up to the action by PGLv . This constructions extends to families of stable

curves and defines a morphism of algebraic stacks, which we call the classifying map:

Mg → [Gr(q, Symm kv )/ PGLv ]

[C] 7→ Symm kv ∼ ⊗k
= Symm H0 (C, ωC ⊗mk
) ↠ H0 (C, ωC
 
) .

The main idea is to leverage the projectivity of Gr(q, Symm kv ) to show that M g is
projective. This will require two properties: the quasi-finiteness of the classifying
map and the nefness of the pluricanonical bundle V = π∗ (ωU⊗k/M ).
g g

Theorem 5.9.2 (Kollár’s Criterion for Ampleness). Let X be a proper algebraic

space over a field k. Let W = Symm V ↠ Q be a surjection of vector bundles of
rank w = rk W and q = rk Q with v = rk V . Suppose that
(a) the classifying map

X → [Gr(q, Symm kv )/ PGLv ]

x 7→ [W ⊗ κ(x) ↠ Q ⊗ κ(x)]

is quasi-finite, and

292
 (b) V is nef.
Then det Q is ample.
We will elaborate on these hypotheses in §5.9.3, where we also prove the criterion,
but let’s first explain how it implies the projectivity of M g,n .
Theorem 5.9.3 (Projectivity of M g ). For g ≥ 2 and N sufficiently divisible, the
line bundle λ⊗N
18g−18 on Mg descends to a line bundle on M g which is relatively
ample over Spec Z. In particular, M g is projective over Spec Z.
Proof. By Properness of Mg (5.5.23), Mg is a Deligne–Mumford stack proper over
Spec Z and the coarse moduli space M g is an algebraic space proper over Spec Z.
By Exercise 4.4.29, λ⊗N
18g−18 descends to a line bundle L on M g for N sufficiently
divisible. Since Ampleness is Open (4.6.17), it suffices show the result in the case
that Mg is defined over a field k. By Proposition 5.3.23, ωU⊗3/M is relatively very
g g

ample and π∗ (ωUg /Mg )⊗3 is a vector bundle of rank 5g − 5. For a stable curve C of
⊗3
genus g, the complete linear series |ωC | embeds C as a degree m := 6g − 6 curve in
P5g−6
. As C is cut out by degree m equations, the multiplication map
W := Symm π∗ (ωU⊗3/M ) ↠ π∗ (ωU⊗3m
/M
) =: Q
g g g g

is a surjection of vector bundles on Mg where W has rank w := 5g−6+m and Q

m
has rank q := (6m − 1)(g − 1).
The classifying map Mg → [Gr(q, kw )/G] is injective on k-points since the
kernel of the multiplication map uniquely determines a stable curve. Since the
automorphism group Aut(C) of a stable curve is identified with the stabilizer of
[C ⊆ P5g−6 ] ∈ HilbP (P5g−6 ) under the action of PGL5g−5 , the multiplication map
induces isomorphisms of automorphism groups. Thus, the classifying map is quasi-
finite. The nefness of V is established in Theorem 5.9.21. By Le Lemme de Gabber
(4.6.1), we may choose a finite cover Z → Mg by a scheme. Applying Kollár’s
Criterion (5.9.2) on Z shows that the pullback of λ18g−18 = det Q to Z is ample.
This shows that the pullback of L under the finite morphism Z → Mg → M g is
ample, and therefore L is ample on M g (see Exercise 4.5.23).
⊗3
Remark 5.9.4. In fact, the tricanonical embedding |ωC | : C ,→ P5g−6 of a stable
curve C is projectively normal and defined by quadratic equations; see [Mum70b,
p. 58]. Therefore, the multiplication map Sym2 π∗ (ωU⊗3/M ) ↠ π∗ (ωU⊗6/M ) is
g g g g
surjective, and the same argument shows that a sufficiently divisible tensor power of
λ6 descend to an ample line bundle on M g .
Corollary 5.9.5 (Projectivity of M g,n ). For integers g and n with 2g − 2 + n > 0,
M g,n is projective over Spec Z.
Proof. We know that M 0,3 = Spec Z and M 1,1 = P1 . By the above theorem,
M g is also projective for g ≥ 2. Since Mg,n+1 → Mg,n is the universal family
(Proposition 5.6.12), it is a family of stable curves and L := ωMg,n+1 /Mg,n is relatively
ample over Mg,n (Proposition 5.3.23). Therefore, some power of L descends to
a line bundle on M g,n+1 relatively ample over M g,n (Exercise 4.4.29(2)). Thus,
M g,n+1 → M g,n is a projective morphism, and we obtain by induction that M g,n is
projective if 2g − 2 + n > 0.
Remark 5.9.6. Alternatively, the projectivity of M g,n can be shown using Kollár’s
Criterion (5.9.2) after establishing that π∗ (L⊗3 ) is nef, where L := ωUg,n /Mg,n (σ1 +
· · · + σn ) and σ1 , . . . , σn are sections of the universal family; see [Kol90, Prop. 4.7].

293
 In recent years, Kollár’s Criterion has been applied in more and more general
settings, e.g., Hassett’s moduli space of weighted pointed curves [Has03], the moduli
of stable varieties of any dimension [KP17], and the moduli of K-polystable Fano
varieties [CP21, XZ20]
Outline of this section. We discuss ampleness criteria for algebraic spaces in §5.9.2,
we prove Kollar’s Criterion for Ampleness in §5.9.3, and establish the nefness of
pluricanonical bundles in §5.9.4.

5.9.2 Positivity and ampleness criteria

We extend the notions of positivity for line bundles introduced in §B.2.2 from
schemes to algebraic spaces, and we also extend the notion of nef vector bundles
from §B.2.4 to algebraic spaces.

Definition 5.9.7. A line bundle L on a quasi-compact algebraic stack X is called:

(1) base point free if for every x ∈ |X |, there exists s ∈ Γ(X, L) with s(x) ̸= 0, and
(2) semiample if L⊗n is base point free for some n > 0.
A line bundle L on a proper algebraic space X over a field k is called:
(3) ample if X is a scheme and L is ample in the usual sense (see Proposition B.2.8),
(4) nef (resp., strictly nef ) if for every integral closed curve5 C ⊆ X, L · C =
deg L|C ≥ 0 (resp., > 0), and
(5) big if there is a constant C such that h0 (Z, L|⊗m
Z ) ≥ Cm
dim(Z)
for all m ≥ 0.
A vector bundle E on a proper algebraic space X is nef if OP(E) (1) is nef on P(E).

If f : X ′ → X is a finite surjective map of algebraic spaces, then a line bundle L

on X is ample if and only if f ∗ L is (Exercise 4.5.23).

Lemma 5.9.8. Let X be a proper algebraic space over a field k.

(1) If f : X ′ → X is a surjective proper morphism of algebraic spaces, then a
vector bundle E is nef if and only if f ∗ E is.
(2) If a vector bundle E is nef, then so is det E.
(3) If f : X ′ → X is a generically quasi-finite and proper morphism of integral
algebraic spaces, then a line bundle L is big if and only if f ∗ L is.

Proof. It suffices to prove (1) for a line bundle L. This follows from the equality
deg(f ∗ L)|C ′ = deg(C ′ → C) deg L|C for an integral subcurve C ′ ⊆ X ′ with image
C ⊆ X. For (2) , we apply Le Lemme de Gabber (4.6.1) to obtain a finite cover
f : X ′ → X by a scheme. By part (1), it suffices to show that f ∗ det(E) = det(f ∗ E)
is nef, but this follows from the case of schemes (Proposition B.2.15). Part (3) can be
shown using the same argument as in the case of schemes (Proposition B.2.24).

The following exercise extending Lemma B.2.26 provides clarity into our ample-
ness strategy even though we do not apply it.

Exercise 5.9.9 (easy). Let X be a proper Deligne-Mumford stack over a field k

with coarse moduli space X → X. If a line bundle L on X is semiample and strictly
nef, then L⊗N descends to an ample line bundle on X for some N > 0, and X is
projective.
5 By Theorem 4.5.32, every such closed integral curve is a projective scheme.

294
 The semiampleness hypothesis in the above exercise can be very challenging
to verify in practice. In the GIT approach, semiampleness is hard-coded into the
definition of semistability (Definition 7.2.1): given a reductive group G acting linearly
on projective space P(V ), then v ̸= 0 ∈ V is semistable if and only if there exists
s ∈ Γ([P(V )/G], O(d)) with s(v) ̸= 0 and d > 0. In other words, the semistable
locus P(V )ss is the largest open subscheme such that O(1) is semiample on [(P(V ) ∖
P(V )ss )/G], i.e., [P(V )ss /G] is the stable base locus of O(1) on [P(V )/G]. The
Hilbert–Mumford Criterion (7.4.4) provides an effective way to verify semistability.
In the Nakai–Moishezon Criterion for Ampleness (B.2.28) for a scheme X, the
semiampleness of a line bundle L = OX (D) is obtained from the nefness of L and
from the existence of sections of L|⊗NZ for every integral closed subscheme Z. It also
holds for algebraic spaces.

Theorem 5.9.10 (Nakai–Moishezon Criterion for Ampleness). Let X be a proper

algebraic space over an algebraically closed field k, and let L be a line bundle on X.
The following are equivalent:
(1) L is ample,
(2) L is nef and for every integral closed subscheme Z ⊆ X, L|Z is big, and
(3) L is strictly nef and for every integral closed subscheme Z ⊆ X, L|⊗m
Z is
effective for some m > 0.

Proof. The same argument as in the case of schemes (Theorem B.2.28) proves the
non-trivial direction (3) ⇒ (1).

Remark 5.9.11 (Classical formulation). It is not hard to extend the definition of

intersection numbers to algebraic spaces. Namely, one shows that χ(Z, L|⊗m Z ) is a
rational polynomial, and one defines c1 (L)dim Z · Z as the normalized coefficient of
mdim Z ; see [SP, Tag 0DN3]. For a nef line bundle L, χ(Z, L|⊗m ⊗m
Z ) = h (Z, L|Z ) for
0

m ≫ 0, and so the bigness of L is equivalent to c1 (L)dim Z

· Z > 0. This implies the
classical formulation of the Nakai–Moishezon Criterion for an algebraic space X:

L is ample ⇐⇒ c1 (L)dim Z · Z > 0 for every integral closed subspace Z ⊆ X.

Seshadri’s Criterion (B.2.30) holds for proper algebraic spaces [Cor93], while
Kleiman’s Criterion (B.2.29) is unknown in general for proper algebraic spaces (and
even proper schemes).

5.9.3 Proof of Kollár’s Criterion for Ampleness

We discuss and prove Kollár’s Criterion (5.9.2): if X is a proper algebraic space over
a field k and W = Symm V ↠ Q is a surjection of vector bundles of rank w = rk W
and q = rk Q with v = rk V such that the classifying map

X → [Gr(q, Symm kv )/ PGLv ]

x 7→ [W ⊗ κ(x) ↠ Q ⊗ κ(x)]

is quasi-finite and V is nef, then det Q is ample.

Remark 5.9.12 (More general formulation). We offer an alternative formulation,
which in addition to giving a more general statement offers a clue into the proof.
Suppose that W ↠ Q is a surjection of vector bundles on X of rank w and q such
that

295
 (a) W has a reduction of structure group by a homomorphism G → GLw from
an affine algebraic group G such that the closure of the image of G in the
projective space P(Matw,w ) of w × w matrices is normal;
(b) W is nef;
(c) the classifying map X → [Gr(q, kw )/G] is quasi-finite; and
(d) either G is linearly reductive (e.g., char(k) = 0 and G is reductive), or G =
GLv1 × · · ·×GLvd with G → GLw a semipositive representation (i.e., it extends
to a map Matv1 ,v1 × · · · × Matvd ,vd → Matw,w );
then det Q is ample [Kol90, Prop. 3.6, Lem. 3.9]. Kollár’s Criterion deduces the
ampleness of det Q from the ampleness of the Plücker line bundle on Gr(q, kw ),
generalizing the fact that a proper scheme quasi-finite over Gr(q, kw ) is projective.
Indeed, when W is the trivial vector bundle, there is a reduction of structure
group to the trivial group G = {1} and the quasi-finiteness of the classifying map
X → Gr(q, kw ) implies that it is finite, and hence the pullback det(Q) of the Plücker
line bundle is ample. The main idea of the proof is to trivialize W , essentially
reducing to this case.
Remark 5.9.13 (Unwinding quasi-finiteness). From Definition 3.3.34, the quasi-
finiteness of the classifying map X → [Gr(q, kw )/G] translates to: (i) X(k) →
Gr(q, kw )(k) has finite fibers for every algebraically closed field k, and (ii) for
w
x ∈ X(k), only finitely many elements of G(k) leave ker(k ∼ = W ⊗ k → Q ⊗ k)
invariant.
In fact, as observed in [KP17, Thm. 5.1], the quasi-finiteness of the classifying map
in (c) can be replaced with the set-theoretic quasi-finiteness of X(k) → Gr(q, kw )(k).
Letting Γ : X → X × [Gr(q, kw )/G] be the graph of the classifying map, the set-
theoretic quasi-finite is equivalent to the quasi-finiteness of im(Γ) → [Gr(q, kw )/G],
which is enough to make the proof below work.
Remark 5.9.14 (Stability). The above theorem does not require that the image
of X lands in the G-stable locus of Gr(q, kw ). However, if this happens, then the
composition X → [Gr(q, kw )s /G] → Gr(q, kw )//G is a quasi-finite morphism of
proper algebraic spaces. It is hence finite and det Q is ample as it is the pullback of
an ample line bundle on Gr(q, kw )//G.
We now prove Kollár’s Criterion for Ampleness [Kol90, Lem. 3.9]. See also
[KP17, Thm. 5.1] and the excellent exposition in [CLM22]. The proof will pro-
ceed by reducing the ampleness of det(Q) to its bigness, which we establish in
Proposition 5.9.15.

Proof of Theorem 5.9.2. By the Nakai–Moishezon Criterion for Ampleness (5.9.10),

it suffices to show that det(Q) is nef and that det(Q)|Z is big for each integral
subspace Z ⊆ X. Since W is nef, so is the quotient Q, and Lemma 5.9.8(1) implies
that det(Q) is also nef. If Z ⊆ X is an integral subspace, by Chow’s Lemma (4.6.5),
there exists a generically quasi-finite and proper morphism Z ′ → Z from a normal
projective scheme. Considering the induced surjection WZ ′ = Symm VZ ′ → QZ ′ of
the restrictions to Z ′ , the classifying map Z ′ → X → [Gr(q, Symm kv )/ PGLv ] is
generically quasi-finite and VZ ′ is nef. Proposition 5.9.15 implies that det(QZ ′ ) is
big, and Lemma 5.9.8(3) further implies that det(Q) is big.

Assuming that X is projective, we establish the weaker conclusion that det Q

is big under the weaker hypothesis that the classifying map is generically finite.
The proof below exploits the birational invariance of bigness by blowing up a

296
compactification of the projectivized frame bundle PFrW to Gr(q, kw ) to construct
sections.

Proposition 5.9.15. Let X be a normal projective scheme over a field k. Let

W = Symm V ↠ Q be a surjection of vector bundles of rank w = rk W and q = rk Q
with v = rk V . Suppose that
(a) the classifying map X → [Gr(q, Symm kv )/ PGLv ] is generically quasi-finite,
and
(b) V is nef.
Then det Q is big.

Proof. Step 1: Trivialize V . Define the relative projective space

⊕v
P := P(H omOX (V, OX ))

over X, and let PFrV ⊆ P be the open subscheme defined by the non-vanishing of
the determinant of the natural map

OP⊕v → VP ⊗ OP (1), (5.9.16)

induced from the universal quotient H omOP (VP , OP⊕w ) → OP (1), where VP is the
pullback of V via P → X. The projection PFrV → X is a principal PGLv -bundle,
and in fact is precisely the projectivized frame bundle of V (see Exercise B.1.55). The
map (5.9.16) is surjective over PFrV and defines a morphism PFrV → Gr(q, Symm kv )
which sits in a cartesian diagram

Po ? _ PFrW / Gr(q, Symm kv )

p
□
!  
X / [Gr(q, Symm kv )/ PGLw ] .

Observe that since P ∼ = P((V ∨ )⊕v ), the pushforward p∗ OP (N ) is identified with

Sym ((V ) ). Since V is nef, it follows from properties of nefness (Proposi-
N ∨ ⊕v

tion B.2.33(3)) that (p∗ OP (N ))∨ is nef, which is how we use hypothesis (b) below.
Step 2: Blow up P to extend PFrW → Gr(q, Symm kv ) to a map P′ → Gr(q, Symm kv ).
If the composition

Symm OP⊕v → Symm VP ⊗ OP (m) → QP ⊗ OP (m)

induced from (5.9.16) were surjective, there would

Vq be m
no need to blow up. Otherwise,
taking the qth wedge power gives a map Sym OP⊕v → det(QP ) ⊗ OP (mq).
Letting I be the image subsheaf, define π : P′ → P as the blowup of the ideal sheaf
I ⊗ det(QP )∨ ⊗ OP (−mq) ⊆ OP . Then P′ → P is an isomorphism over PFrW and
there is a map P′ → Gr(q, Symm kv ) extending PFrW → Gr(q, Symm kv ) yielding a
commutative diagram

E H L
,
P′ /P / Gr(q, Symm kv )
π
p
 
X / [Gr(q, Symm kv )/ PGLw ],

297
where E ⊆ P′ is the exceptional divisor, H ⊆ P is the hyperplane class, and L is the
ample divisor on Gr(q, Symm kv ) corresponding to the Plücker embedding. Abusing
notation, we also view det(Q) as a divisor on X. By using subscripts to denote the
pullback of divisors, we have the formula

LP′ = det(Q)P′ + mqHP′ − E. (5.9.17)

Step 3: Use the generic quasi-finiteness to construct sections of OX (m det(Q) −

A) ⊗ SymN mq ((V ∨ )⊕v ) for an ample divisor A on X and some N > 0. Since
X → [Gr(q, Symm kv )/ PGLw ] is generically quasi-finite, so is P′ → Gr(q, Symm kv ).
It follows from Proposition B.2.24 that LP′ is big. For every very ample divisor A on
X, the pullback AP′ is effective, and Kodaira’s Lemma (B.2.20) implies that there is
an N > 0 such that N LP′ − AP′ is effective. Using the identity (5.9.17), we obtain
an inclusion of line bundles
OP′ (N LP′ − AP′ ) ∼
= OP′ (N det(Q)P′ + N mqHP′ − N EP′ − AP′ )
⊆ OP′ (N det(Q)P′ + N mqHP′ − AP′ )
∼
= π ∗ p∗ OX (N det(Q) − A) ⊗ OP (N mq)

Choose a nonzero section OP′ → π ∗ p∗ OX (m det(Q) − A) ⊗ OP (mq) . Since P is

normal and π : P′ → P is birational, π∗ OP′ = OP . Applying π∗ to the chosen section

and using the projection formula yields a nonzero section

OP → p∗ OX (N det(Q) − A) ⊗ OP (N mq).

Applying p∗ gives a nonzero section OX → OX (N det(Q) − A) ⊗ p∗ OP (N mq), which

we rearrange as
(p∗ OP (N mq))∨ → OX (N det(Q) − A). (5.9.18)

Step 4: Blow up X so that (5.9.18) is surjective, and use the nefness of V to

conclude that det(Q) is big. As pointed out earlier, the nefness of V implies that
(p∗ OP (N mq))∨ ∼ = SymN mq ((V ∨ )⊕v )∨ is nef by Proposition B.2.33(3). Therefore, if
(5.9.18) is surjective, then N det(Q) − A is nef which expresses N det(Q) as a nef and
ample line bundle, which implies that det(Q) is even ample. Otherwise, if I is the
image subsheaf of (5.9.18), we blow up the ideal sheaf I ⊗ OX (N det(Q) − A)∨ ⊆ OX
to obtain a birational morphism f : X ′ → X with exceptional divisor E ′ and a
surjection

f ∗ (p∗ OP (N mq))∨ ↠ M := f ∗ OX (N det(Q) − A) ⊗ OX ′ (−E ′ ).

Since f ∗ (p∗ OP (N mq))∨ is nef, so is M . This expresses N det(QX ′ ) as the sum of a

nef divisor M , a big divisor AX ′ , and an effective divisor E ′ . By the Characterization
of Bigness (B.2.21), a multiple of AX ′ is the sum of an ample divisor and effective
divisor. Therefore, a multiple of det(QX ′ ) is the sum of a nef, ample, and effective
divisors. Since nef plus ample is ample, and ample plus effective is big, we conclude
that det(QX ′ ) is big. As X ′ → X is birational, it follows that det(Q) is big.

5.9.4 Nefness of pluricanonical bundles

We prove the nefness of the pluricanonical bundles π∗ (ωU⊗k/M ) on Mg by showing
g g
that a negative line bundle quotient contradicts Ekedahl–Mumford Vanishing.

298
Theorem 5.9.19 (Ekedahl–Mumford Vanishing). Let X be a smooth projective
surface over an algebraically closed field k which is minimal and of general type. If
⊗−n ⊗−n
n ≥ 1, then h1 (X, KX ) ≤ 1. If n > 1 or char(k) ̸= 2, then h1 (X, KX ) = 0.
Proof. The characteristic zero version is [Mum67, Thm. 2], while the positive char-
acteristic case is the main theorem of [Eke88].
Corollary 5.9.20. Let X be a smooth projective surface over an algebraically closed
field k which is minimal and of general type. Let D be a reduced effective Cartier
divisor such that each connected component of D has genus at least 2. If n ≥ 2, then
⊗n ⊗n
h1 (X, KX (D)) ≤ 1. If n > 2 or char(k) ̸= 2, then h1 (X, KX (D)) = 0.
⊗n ⊗n ⊗n
Proof. By the short exact sequence 0 → KX → KX (D) → KX (D)|D → 0 and
⊗n
the inequalities of Theorem 5.9.19, it suffices to show that h1 (X, KX (D)|D ) = 0
for n ≥ 2. By adjunction, KX (D)|D ∼
⊗n
= Dω ⊗n
. Since each connected component of D
⊗n ⊗n
has genus at least 2, h1 (X, KX (D)|D ) = h1 (D, ωD ) = 0.

The nefness of π∗ (ωU⊗k/M ) for k ≥ 2 reduces to the following properties of

g g
families of stable curves over a smooth curve.
Theorem 5.9.21. If T is a smooth, connected, and projective curve over a field k
⊗k
and π : C → T is a family of stable curves, then π∗ (ωC/T ) is nef for k ≥ 2.

Proof. Step 1: Reduction to characteristic p. Assume that char(k) = 0. Since T and

C are of finite type over k, their defining equations involve finitely many coefficients
of k. Thus there exists a finitely generated Z-subalgebra A ⊆ k and a cartesian
diagram
C / Ce
 □ 
T / Te
 □ 
Spec k / Spec A

where Ce and Te are schemes of finite type over A. By Limit Methods (B.3.7 and
5.2.24), we may further arrange that Te → Spec A is a smooth family of curves
and that Ce → Te is a family of stable curves. Finally, by restricting along a map
Spec R → Spec A, we may assume that A is a DVR whose closed and generic points
have characteristic p > 0 and 0, respectively. Since Nefness for Bundles is Stable
under Generization (B.2.34), it suffices to prove the theorem when char(k) = p > 0.
Step 2: Further reductions. We claim that we may also assume that
(a) π : C → T is generically smooth,
(b) the genus of T is at least 2, and
(c) C is a smooth minimal surface of general type.
The reduction to (a) is handled in Exercise 5.9.23. For (b), if g : T ′ → T is any finite
cover where T ′ is a smooth connected curve of genus g ≥ 2 and C ′ := C ×T T ′ , then
⊗k
g ∗ π∗ (ωC/T )∼
= π∗′ (ωC⊗k
′ /T ′ ) by properties of the dualizing sheaf. By Lemma 5.9.8(1),

the nefness of ω ′ ′ implies the nefness of ω ⊗k . For (c), if f : Ce → C is a resolution

⊗k
C /T C/T
⊗k
of singularities, then f∗ (ωC/T ∼ ⊗k
e ) = ωC/T , and thus we can assume that C is smooth. If
C contains a smooth rational −1 curve E, then E must be contained in a fiber of
C → T as otherwise there would be a finite cover E = P1 → T of a genus g ≥ 2 curve.

299
By Castelnuovo’s Contraction Theorem (B.2.6), there is a morphism f : C → C ′
⊗k
contracting E. Since f∗ ωC/T = ωC⊗k
′ /T and the process of contracting smooth rational

−1 curves terminates in finitely many steps, we can assume that C is minimal. As

both T and the generic fiber of C → T are smooth and of general type, C is also of
general type.

⊗k
Step 3: Positive characteristic case. If π∗ (ωC/T ) is not nef, then there exists a
⊗k
quotient line bundle π∗ (ωC/T ) ↠ M where d = deg M < 0. Consider the absolute
Frobenius morphisms F : C → C and F : T → T which fit into a commutative
diagram

C
F /C
π π
 
T
F / T.

⊗k ⊗k
Note that F ∗ π∗ (ωC/T ) = π∗ (ωC/T ). Since deg F ∗ M = pd, we can apply Frobenius
repeatedly to arrange that d is as small as we want. Specifically, we can arrange
that M ∨ ∼= ωT⊗k ⊗ L where L is a very ample line bundle on T . (This was the entire
point of reducing to characteristic p: to repeatedly apply Frobenius to arbitrarily
decrease the degree of M .)
⊗k
The surjection π∗ (ωC/T )↠M ∼ = (ωT⊗k ⊗ L)∨ yields a surjection

⊗k
π∗ (ωC/T ) ⊗ ωT⊗k ⊗ L ↠ OT

of vector bundles on T . Since dim T = 1, we obtain that

⊗k
h1 (T, π∗ (ωC/T ) ⊗ ωT⊗k ⊗ L) ≥ h1 (T, OT ) ≥ 2. (5.9.22)

To obtain a contraction, we examine h1 (C, ωC⊗k ⊗ π ∗ L), which by the degeneration of

the Leray spectral sequence on the E2 -page is bounded below by h1 (T, π∗ (ωC⊗k ⊗π ∗ L)).
We obtain an inequalities

h1 (C, ωC⊗k ⊗ π ∗ L) ≥ h1 (T, π∗ (ωC⊗k ⊗ π ∗ L))

⊗k
= h1 (T, π∗ (ωC/T ⊗ π ∗ ωT⊗k ⊗ π ∗ L)) (since ωC ∼
= ωC/T ⊗ π ∗ ωT )
⊗k
= h1 (T, π∗ (ωC/T ) ⊗ ωT⊗k ⊗ L) (projection formula)
1
≥ h (T, OT ) ≥ 2 (by (5.9.22)).

As L is very ample, there is an effective divisor D which is the union of smooth fibers of
π : C → T such that OC (D) ∼ = π ∗ L. The above inequality contradicts Corollary 5.9.20
applied to D. See also [Kol90, Thm. 4.3] and [CLM22, Thm. 6.10].

Exercise 5.9.23 (details). Let T be a smooth, connected, and projective curve over
⊗k
a field k. Show that if π∗ (ωC/T ) is nef for every generically smooth stable family,
then it is also nef for every stable family.

300
5.10 Glimpse of the geometry of M g
When [Zariski] spoke the words algebraic variety, there was a certain
resonance in his voice that said distinctly that he was looking into a secret
garden. I immediately wanted to be able to do this too... Especially, I
became obsessed with a kind of passion flower in this garden, the moduli
spaces of Riemann.

David Mumford [AI97, p. 225]

The existence, irreducibility, and projectivity (Theorem A) is of course just the

beginning in the study of M g . We can now begin asking all sorts of geometric
questions:
– What is its ample or effective cone?
– What is its Kodaira dimension or canonical model?
– What are the singularities of the coarse moduli space?
– What is its singular, Chow, Hodge, or crystalline cohomology?
– For integers r and d, does a general curve of genus g have a gdr ?
Note that the projectivity of M g is necessary to even discuss ample line bundles and
other birational properties. On the other hand, the irreducibility of M g allows us to
translate a property about a general curve into the existence of a non-empty open
locus of curves in M g with the given property, and when the property is an open
condition, it suffices to exhibit a single smoothable curve with the given property.
We can’t possibly give a comprehensive summary; after all, M g is perhaps the
single most studied variety over the last sixty years. We refer the reader to [Mum83],
[ACGH85], [HM98], [ACG11], and [FM13]. We will however make a few cursory
comments.
Ample cone. For each k ≥ 5 and m ≫ 0, the GIT construction of M g using the
Hilbert scheme constructs a line bundle on Mg which descends to an ample line
bundle on M g . This class on Mg is proportional to

r(k)λmk − r(mk)λk

where if we set Ek = π∗ (ωU⊗k/M ), then r(k) = rk Ek and λk = det Ek . Grothendieck–

g g

Riemann–Roch can be used to express each of the line bundles λk = (6k 2 − 6k +

1)λ − 2 δ as a linear combination of λ := λ1 and δ, the boundary divisor. The
k

asymptotic limit of this class as d → ∞ is proportional to

4
(12 − )λ − δ,
k
which is also proportional to the ample class induced by the GIT construction using
the Chow scheme; see [Mum77, Thm. 5.15] and [HH13, Prop. 5.2]. Taking k = 5,
shows that 11.2λ − δ is ample. On the other hand, Kollár’s construction shows
that λkm descends to ample line on M g for k ≥ 3 and m ≥ 2, and this shows that
(12 + ϵ)λ − δ is ample for any 0 < ϵ ≪ ∞.
However, even more is true! By bootstrapping the positivity deduced from
GIT, Cornalba and Harris showed that aλ − δ is ample if and only if a > 11,
thus determining the ample cone of M g in the λ-δ-plane of Pic(M g ) [CH88]. The

301
full ampleness cone is not known, but there is an interesting conjecture called the
F-conjecture, which provides a finite list of test curves and asserts that a line bundle
is ample if and only if it intersects each of these test curves positively. We will not
enumerate these test curves here, but it includes for instance the rational curves in
Mg parameterizing elliptic tails obtained by fixing a one-pointed (C, p) curve of
genus g − 1 and varying the j-invariant of a nodally-attached elliptic curve (E, p).
Kodaira dimension. Severi showed that M g is unirational for g ≤ 10 and
conjectured that it unirational for all g; see the discussion §0.1.2. Harris and
Mumford showed that this conjecture not only fails, but that it fails spectacularly
[HM98]. Quite the opposite is true: M g is general type for odd g ≥ 25. Their
strategy was to find an ample divisor A and effective divisor D such that the
canonical bundle KM g is proportional to A + D, as this implies that KM g is big.
Defining the slope of D as the infimum of a/b such that aλ − bδ − D is effective and
appealing to the formula KMg = 13λ − 2δ, it suffices to find an effective divisor
of slope less than 13/2. Writing g + 1 as a product (r + 1)(s + 1), they define the
Brill–Noether divisor Dr,s as the closure of curves possessing a gdr , and show that its
slope is less than 13/2. Eisenbud and Harris extend this to all genera g ≥ 24 [EH87].
Given that M g is general type, it is natural to ask what is its canonical model
can ⊗d
), which by [BCHM10], is a projective variety. Given
L
M g = Proj d Γ(Mg , KM
g

that M g is a particular interesting higher dimensional variety and that its modular
description provides a handle on its geometry, it is tempting to test drive the
machinery of the minimal model program on M g . Quite spectacularly, Hassett and
Hyeon realized that the first contraction and flip also carry moduli descriptions
[HH09], [HH13]. For instance, the line bundle KMg + 9/11δ, which is proportional
to 11λ − δ, is nef and has zero intersection with precisely the curves of elliptic tails,
and these curves are replaced in the first contraction with cuspidal singularities
y 2 = x3 .
Singularities of M g . First, since Mg is smooth, the Local Structure of Coarse
Moduli Spaces (4.4.13) implies that M g has finite quotient singularities. By appealing
to the Reid-Tai criterion, Harris and Mumford showed that M g has canonical
singularities [HM82], or, in other words, that pluricanonical sections extend to a
desingularization of M g , a property necessary for their proof that M g is general
type.
A general smooth curve C ∈ Mg (k) of genus g ≥ 3 has a trivial automorphism
group. Since Mg is irreducible, this follows from exhibiting a single smooth curve
with a trivial automorphism groups. It implies that there is an open subset U ⊆ Mg
which is a smooth quasi-projective variety. If g ≥ 4, the locus of curves with a
non-trivial automorphism has codimension at least 2 and a curve [C] ∈ Mg is a
singular point if and only if Aut(C) ̸= 1. See [Rau62] and [Pop69].

302
Chapter 6

Geometry of algebraic stacks

The Red Queen shook her head. “You may call it ‘nonsense’ if you like,”
she said, “but I’ve heard nonsense, compared with which that would be as
sensible as a dictionary!”

Lewis Carroll, Through the Looking-Glass

6.1 Quasi-coherent sheaves and cohomology theories

We define quasi-coherent sheaves on an algebraic stack in essentially the same way
as we did for Deligne–Mumford stacks in §4.1, but using the lisse-étale site on X
instead of the small étale site.

6.1.1 Sheaves and OX -modules

The smallest site suitable to define sheaves on a general algebraic stack is the
lisse-étale topology. This site has some technical issues, most notably the lack of
functoriality; see Caution 6.1.5.
Definition 6.1.1 (Lisse-étale site). The lisse-étale site Xlis-ét on an algebraic stack
X is the category of schemes smooth over X , where morphisms are arbitrary maps
of
` schemes over X . A covering {Ui → U } is a collection of morphisms such that
i Ui → U is surjective and étale.

We denote by Sh(Xlis-ét ) (resp., Ab(Xlis-ét )) the category of sheaves (resp., sheaves

of abelian groups) Xlis-ét . The structure sheaf OX is defined as OX (U ) = Γ(U, OU ),
and allows us to define the notion of a OX -module. For F ∈ Ab(Xlis-ét ), we can
define sections over an arbitrary smooth morphism U → X of algebraic stacks by
the formula F (U → X ) = Eq(F (U → X ) ⇒ F (R → X )) where U → X is a smooth
presentation and R → U ×X U is an étale presentation by schemes. As in the case of
Deligne–Mumford stacks in §4.1.1, this is independent of the choice of presentation.
id
In particular, the global sections Γ(X , F ) := F (X −→ X ) is well-defined.
Given a morphism f : X → Y of algebraic stacks, there are adjoint functors
f∗ f∗
- -
Sh(Xlis-ét ) Sh(Ylis-ét ) and Ab(Xlis-ét ) Ab(Ylis-ét ),
m m
f −1 f −1

303
where the pushforward is defined by f∗ F (V → Y) := F (X ×Y V → X ) and the
inverse image f −1 G is the sheafification of the presheaf

(f −1 G)pre : (U → X ) 7→ colim(U →V ) G(V → Y), (6.1.2)

where the colimit is over the category of morphisms U → V over f such that V is a
smooth Y-scheme. See [SP, Tag 00WX] where this is established more generally for
any continuous functor of sites.
Exercise 6.1.3. If f : X → Y is a smooth morphism of algebraic stacks and F is a
sheaf on Ylis-ét , then f −1 F = F |X and f −1 is exact.
Exercise 6.1.4. Let f : X → Y be a morphism of schemes.
(a) Let Y ′ be a smooth Y -scheme and hY ′ be the sheaf on Ylis-ét representable by
Y ′ , i.e., hY ′ (V → Y ) = MorY (V, Y ′ ). Show that f −1 hY ′ = hX×Y Y ′ .
(b) Show that f −1 OYlis-ét = OXlis-ét .
Hint: use part (a) and the fact that OYlis-ét is representable by A1Y .
(c) Conclude that if ix : Spec κ(x) → X is the inclusion of a point x ∈ X, then
x OXlis-ét = κ(x). (This should be contrasted to the fact that ix OXZar =
i−1 −1

OX,x is the local ring and that ix OXét = OX,x is its henselization.)
−1 h

(d) Show that f −1 is not exact in general.

Hint: consider the case when f = ix is the inclusion of a residue field and use
part (c).
Exercise 6.1.4(b) implies that for a morphism f : X → Y of algebraic stacks
that f −1 OY = OX . Therefore, the pullback of an OY -module G is the OX -module
f ∗ G = f −1 G ⊗f −1 OY OX = f −1 G.
Caution 6.1.5 (Lack of functoriality). The non-exactness of f −1 is related to the
lack of fiber products in the lisse-étale site Xlis-ét : if U → V and U ′ → V are
arbitrary morphisms of smooth X -schemes, U ×V U ′ is not necessarily smooth over
X . Because there are no fiber products, the category indexing the colimit in (6.1.2)
is not filtering. An index category I is filtering if and only if colimits over I commute
with finite limits [SP, Tag 002W], and therefore f −1 doesn’t commute with finite
limits or equivalently is not left exact. See also [Beh03, Warning 5.3.12], [Ols07,
Ex. 3.4], and [SP, Tag 07BF].
Unfortunately, this means that the lisse-étale site is not functorial: a morphism
f : X → Y does not necessarily induce a morphism of sites Xlis-ét → Ylis-ét , as the
definition [SP, Tag 00X0] of a morphism of sites is the data of a functor Ylis-ét →
Xlis-ét (in the opposite direction!) which is continuous, i.e., the preimage of a cover in
Ylis-ét is a cover in Xlis-ét (which does hold here), and such that the preimage functor
f −1 : Sh(Ylis-ét ) → Sh(Xlis-ét ) on categories of sheaves is exact (which fails here).
Because of the lack of functoriality, constructions in the lisse-étale site sometimes
require extra care, especially the derived category and the cotangent complex; see
[Ols07]. Nevertheless, quasi-coherent sheaves in the lisse-étale site (which is all
we will need in this text) can be quickly introduced and enjoy the same familiar
properties as in the case of schemes.
Remark 6.1.6 (The big fppf topology). One can define sheaves on an algebraic stack
in the big fppf topology. This leads to a different category of sheaves but to an
equivalent category of quasi-coherent sheaves, with the same reasoning as in the
case of a Deligne–Mumford stack (see Exercise 4.1.25).

304
 The big fppf topology is functorial but has other technical issues. As this
is the approach in the Stacks Project, we quickly summarize its salient features,
beginning with the case of schemes. A morphism f : X → Y of schemes induces
a continuous functor (Sch/Y )fppf → (Sch/X)fppf between big fppf sites defined by
(V → Y ) 7→ (V ×Y X → X). The pushforward is defined as before by f∗ F (V →
Y ) = F (V ×Y X → X), and the pullback is given by the formula f −1 G(U → X) =
F (U → X → Y ). The simplicity of the pullback formula is because we are working
in a big site: the identity map U → V = U is an initial object in the index category
of (6.1.2) defining the presheaf (f −1 G)pre , and moreover (f −1 G)pre = f −1 G is
already a sheaf. It is a general fact that a continuous functor C → D commuting
with both fiber products and the final objects induces an exact preimage functor
Sh(C) → Sh(D), and therefore defines a morphism of sites D → C [SP, Tag 00X6].
This applies in this case to show that f −1 is exact so that (Sch/Y )fppf → (Sch/X)fppf
defines a morphism of sites (Sch/X)fppf → (Sch/Y )fppf .1
A similar story holds for morphisms of algebraic stacks f : X → Y. The big
fppf site Xfppf is the category X of all schemes over X where coverings are fppf
coverings. In this case, the pullback f −1 G(U → X ) = G(U → X → Y) is defined
as above (no sheafification required), and the pushforward f∗ F can be defined as
f∗ F (V → Y) = limU →V F (U → X ), where the index category of the limit is the
category of morphisms U → V of schemes over X → Y; if f is representable by
schemes, then V ×Y X → V is an final object and f∗ F (V → Y) = F (V ×Y X → X ).
Alternatively, we can define sections of F over any morphism U → Y of algebraic
stacks by smooth descent, and then define the pushforward using the usual formula
f∗ F (V → Y) = F (V ×Y X → X ). Interpreting f∗ and f −1 via the continuous and
cocontinuous functor f : Xfppf → Yfppf of sites as in Footnote 1, it follows that f −1
is exact. This means that (f∗ , f −1 ) defines a morphism of sites Xfppf → Yfppf , so
that the big fppf site of an algebraic stack is functorial (unlike the lisse-étale site).
However, we will see shortly that the pushforward functor on O-modules does not
preserve quasi-coherence; see Remark 6.1.9.

6.1.2 Quasi-coherent sheaves

Definition 6.1.7. Let X be an algebraic stack. An OX -module F is quasi-coherent
if
(1) for every smooth X -scheme U , the restriction F |UZar to the small Zariski site
of U is a quasi-coherent OUZar -module, and
(2) for every morphism f : U → V of smooth X -schemes, the natural morphism
f ∗ (F |VZar ) → F |UZar is an isomorphism.
A quasi-coherent sheaf F on X is a vector bundle (resp., vector bundle of rank r,
line bundle) if F |UZar is for every smooth X -scheme U . If in addition X is locally
noetherian, we say F is coherent if F |UZar is coherent for every smooth X -scheme U .

Equivalently, an OX -module F is quasi-coherent if it is quasi-coherent as a

module in the ringed site (Xlis-ét , OX ) as formulated in Exercise 4.1.22(iii). If X
1 The functor v : (Sch/Y )
fppf → (Sch/X)fppf has a continuous left adjoint u : (Sch/X)fppf →
(Sch/Y )lis-ét defined by (U → X) 7→ (U → X → Y ), which can also be used to define f∗ and
f −1 : on the level of presheaves, f∗ = v p = p u and (f −1 )pre = vp = up ; see [SP, Tags 00VC and
00XF] for the general definitions of p u, up , and up , and [SP, Tag 09VQ] for their relationships.
The functor u is not only continuous but also cocontinuous (a cocontinuous functor u : C → D of
sites requires that for every object c ∈ C and covering {di → u(c)}, there is a covering {ci → c}
such that the covering {u(ci ) → u(c)} refines {di → u(c)}). This also automatically implies that
f −1 is exact [SP, Tag 06NW].

305
is a scheme (resp., Deligne–Mumford stack), then Exercise 4.1.25 implies that the
above definition is consistent with Definition 4.1.17. It is also not hard to see
that quasi-coherence in the lisse-étale site of an algebraic stack is equivalent to
quasi-coherence in the big fppf topology. In any case, we have well-defined categories
QCoh(X ) and Coh(X ) (in the noetherian setting) of quasi-coherent and coherent
sheaves. The tensor product F ⊗ G := F ⊗OX G and Hom sheaf H omOX (F, G) of
quasi-coherent sheaves are defined in the usual way, and are quasi-coherent sheaves
that enjoy the familiar properties.
Given a morphism f : X → Y of algebraic stacks, the pullback f ∗ = f −1 preserves
quasi-coherence. If f is quasi-compact and quasi-separated, the pushforward f∗ also
preserves quasi-coherence and is a right adjoint to f ∗

f∗
-
QCoh(X ) QCoh(Y).
l
∗
f

Exercise 6.1.8 (details). Verify that f∗ and f ∗ preserve quasi-coherence and that
they are adjoint functors.

Remark 6.1.9 (Pushforward of quasi-coherent sheaves in the big fppf site). Given a
morphism f : X → Y of algebraic stacks, the pushforward f∗ of sheaves in the big
fppf sites as defined in Remark 6.1.6 does not preserve quasi-coherence. This even
fails in the case of a map of schemes f : X → Y . If F ∈ QCoh(X) and

f′
U /V

g′ □ g
 
X
f
/Y

is a cartesian diagram of schemes, then it follows from the definitions in the big
site that g ∗ f∗ F = g −1 f∗ F = f∗′ g ′−1 F = f∗′ g ′∗ F but this formula does not hold in
general (unless for example g is flat). In other words, f∗ F has the correct sections of
the desired quasi-coherent pushforward fQCoh,∗ F only over flat maps V → Y . With
some care (see [SP, Tag 070A]), the quasi-coherent pushforward can be defined as
the composition
f∗ Q
fQCoh,∗ : QCoh(X ) → LQCohfbc (OX ) −→ LQCohfbc (OY ) −
→ QCoh(Y),

where LQCohfbc (OX ) denotes the full subcategory of locally quasi-coherent OX -

modules (i.e., for every flat map U → X from a scheme, the restriction F |Uét is
quasi-coherent) which have the flat base change property (i.e., for every flat morphism
f : U → V over X , the induced map f ∗ (F |Vét ) → F |Uét is an isomorphism), and Q
is a the quasi-coherator functor. The functor fQCoh,∗ is a right adjoint of f −1 = f ∗ .
See also [SP, Tag 08MW] for the development of the derived category using the fppf
topology.

Exercise 6.1.10. Let X be a scheme and Gm be the multiplicative group over

Spec Z. Show that a quasi-coherent sheaf on X ×BGm is the same data as a Z-graded
quasi-coherent sheaf on X.
Hint: Extend the argument of Proposition B.1.15.

306
Exercise 6.1.11. Let G be an affine algebraic group over a field k. Recall that
a G-representation is a k-vector space with a dual action σ : V → Γ(G, OG ) ⊗k V
satisfying two natural compatibility conditions (see §B.1.12).
(a) Show that QCoh(BG) is equivalent to the category Rep(G) of G-representations.
More generally, if Spec A is an affine k-scheme with a G-action, show that a
quasi-coherent sheaf on [Spec A/G] is the data of an A-module M together with
a coaction σ : M → Γ(G, OG )⊗k M over k (i.e., a map of k-vector spaces giving
M the structure of a G-representation) such that multiplication A ⊗k M → M
is a map of G-representations. This extends Example 4.1.21 which was the
case when G is finite.
(b) Considering the diagram

Spec A
p
/ [Spec A/G] π / Spec AG

q

BG,

extend Exercise 4.1.27 by providing descriptions of the functors p∗ , p∗ , π∗ , π ∗ , q∗

and q ∗ on quasi-coherent sheaves.
(c) If U is a k-scheme with an action of G, then a line bundle with a G-action is
∼
a line bundle L on U together with an isomorphism α : σ ∗ L → p∗2 L satisfying
a cocycle condition p23 α ◦ (idG ×σ) α = (µ × idU ) α; see B.1.28. Show that a
∗ ∗ ∗

line bundle with a G-action is the same as a line bundle on the quotient stack
[U/G].

Example 6.1.12. Two non-isomorphism affine algebraic groups G and H over a field
k can have isomorphic BG ∼ = BH classifying stacks and thus equivalent categories of
representations. For example, if O(q) and O(q ′ ) are orthogonal groups with respect
to non-degenerate quadratic forms q and q ′ on an n-dimensional k-vector space V ,
then BO(q) ∼ = BO(q ′ ) (see Exercise 3.1.13), and thus O(q) and O(q ′ ) have equivalent
categories of representations.

We’ve already seen examples of quasi-coherent sheaves on a Deligne–Mumford

stack where the pluricanonical line bundles λ = det π∗ (ωU⊗k/M ) on Mg played a
g g

prominent role in the projectivity of M g (see §5.9). Determinantal line bundles play
a similar role in the study of the moduli stack of vector bundles.

Example 6.1.13 (Determinantal line bundles). Consider the stack M := Bun C,r,d
of vector bundles on a smooth, connected, and projective curve C over k. Consider
the diagram
C ×M
p1 p2

{ $
C M.
The projection p2 : C × M → M is representable, projective, and smooth of relative
dimension 1. For every vector bundle F on C × M, the cohomology Ri p2,∗ F (as
defined below) is computed as a 2-term complex [K 0 → K 1 ] of vector bundles and
that the line bundle

det Rp2,∗ F := det(K0 ) ⊗ det(K1 )∨

307
is well-defined on M. Note that if rk K0 = rk K1 , i.e., rk Rp2,∗ F = 0, then we
have a map det K 0 → det K 1 of line bundles and the corresponding map OM →
det(K 0 )∨ ⊗ det(K1 ) defines a section of the dual (det Rp2,∗ F )∨ .
Let Euniv be the universal vector bundle on C × M. For every vector bundle V
on C, we define the determinantal line bundle

∨
LV := det Rp2,∗ (Euniv ⊗ p∗1 V ) .
associated to V .
Example 6.1.14. If X is an algebraic stack of finite presentation over a scheme
S, then the relative sheaf of differentials ΩX /S on Xlis-ét , defined on a smooth
X -scheme U by ΩX /S (U ) = ΩU/S , is not quasi-coherent. This is because for a
non-étale map f : U → V of smooth X -schemes, f ∗ ΩV /S → ΩU/S is not necessary
an isomorphism. This differs from the Deligne–Mumford case where the sheaf ΩX /S
on Xét is quasi-coherent (Example 4.1.19). When X is Deligne–Mumford, ΩX /S
extends to a quasi-coherent sheaf on Xlis-ét by defining ΩX /S (U ), for a smooth map
f : U → X from a scheme, to be the global sections of the sheaf f ∗ ΩX /S on Ulis-ét .
Exercises 4.1.28 and 4.1.45 generalize to algebraic stacks.
Proposition 6.1.15 (Flat Base Change). Consider a cartesian diagram
g′
X′ /X

f′ □ f
 
Y′
g
/Y
of algebraic stacks, and let F be a quasi-coherent sheaf on X. If g : Y ′ → Y is flat
and f : X → Y is quasi-compact and quasi-separated, the natural adjunction map
g ∗ f∗ F → f∗′ g ′∗ F
is an isomorphism.

Proposition 6.1.16. Let X be a noetherian algebraic stack. Every quasi-coherent

sheaf on X is a directed colimit of its coherent subsheaves. If U ⊆ X is an open
substack, then every coherent sheaf on U extends to a coherent sheaf on X .

Exercise 6.1.17. Let X → Y be a smooth affine morphism of noetherian algebraic

stacks with affine diagonal.
(1) Show that there is a vector bundle ΩX /Y on X with the property that if
V → Y is a morphism from a scheme, the pullback of ΩX /Y to XV := X ×Y V
is ΩXV /V .
(2) Given a commutative diagram

Spec A
f0
/X
_ 0 ;

 
Spec A /Y

where A ↠ A0 is a surjection of noetherian rings with square-zero kernel

J, show that the set of liftings is a torsor under HomA0 (f0∗ ΩX /Y , J) and in
particular is non-empty.
(3) Can you weaken the hypotheses?

308
6.1.3 Quasi-coherent constructions
Extending the constructions of §4.1.5 on a Deligne–Mumford stack to an algebraic
stack X , a quasi-coherent OX -algebra is a quasi-coherent OX -module A with a
compatible structure as a ring object. The relative spectrum SpecX A, defined
as the stack of pairs (f, α) where f : S → X is a morphism from a scheme and
α : f ∗ A → OS is a map of OS -algebras, is an algebraic stack affine over X . On a
noetherian algebraic stack, every quasi-coherent OX -algebra is a directed colimit of
finite type subalgebras.
The reduction of X is Xred := SpecX OX red
where OXred
is the sheaf of OX -
algebras defined by OX (U ) = Γ(U, OU )red for a smooth X -scheme U . If X is
red

integral, the normalization of X is defined as Xe := SpecX A, where A is the OX -

algebra whose ring of sections over a smooth X -scheme U is the normalization
of Γ(U, OU ); this is well-defined since normalization commutes with smooth base
change (Proposition A.7.4). For a quasi-compact and quasi-separated morphism
f : X → Y of algebraic stacks, there is a factorization f : X → Spec f∗ OX → Y.
The morphism f is affine if and only if X → Spec f∗ OX is an isomorphism, and
quasi-affine if and only if X → Spec f∗ OX is an open immersion. The proof of
Zariski’s Main Theorem (4.5.9) for Deligne–Mumford stacks extends to algebraic
stacks.
Theorem 6.1.18 (Zariski’s Main Theorem). A representable, quasi-finite, and sepa-
rated morphism f : X → Y of noetherian algebraic stacks factors as the composition
of a dense open immersion X ,→ Ye and a finite morphism Ye → X .

6.1.4 Picard groups

If X is an algebraic stack, we let Pic(X ) denote the set of isomorphism classes of
line bundles on X . It is an abelian group under tensor product.
Example 6.1.19. If G is an affine algebraic group over a field k, then Pic(BG) is
identified with the group X∗ (G) of characters of G. For example, Pic(BGm ) = Z,
Pic(BGLn ) = Z, and Pic(PGLn ) = {0}. More generally, if U is a scheme with an
action of G, then Pic([U/G]) is identified with the group PicG (U ) of line bundles on
U with a G-action (Exercise 6.1.11(c)).
Exercise 6.1.20. Let X be a smooth and irreducible algebraic stack over a field
k. If D ⊆ X is a reduced substack of codimension 1 with complement U, show that
there is a naturally defined line bundle O(D) (generalizing the usual construction
for schemes) such that O(D)|U ∼ = OU .
Exercise 6.1.21. Let Gm acts on An over a field k with weights d1 , . . . , dn . Let O(1)
be the line bundle on [An /Gm ] corresponding to the projection [An /Gm ] → BGm .
(a) Show that Pic([An /Gm ]) ∼= Z generated by O(1).
(b) Show that the restriction Pic([An /Gm ]) → Pic(P(d1 , . . . , dn )) is an isomor-
phism, where P(d1 , . . . , dn ) is the weighted projective stack (see Example 3.9.14).
(c) If f ∈ Γ(An , OAn ) is a homogenous polynomial of degree d such that V (f ) ⊆ An
is reduced, show that O(V (f )) ∼= O(d) on [An /Gm ].
The following offers an elementary approach to Mumford’s calculation that
Pic(M1,1 ) = Z/12, first offered as a difficult exercise in Exercise 4.1.35.
Exercise 6.1.22. Let k be a field with char(k) ̸= 2, 3.

309
 (a) Show that Pic(M1,1 ) = Z.
Hint: Use the description M1,1 = [(A2 ∖0)/Gm ] of Exercise 3.1.19(c) where Gm
acts with weights 4 and 6. Show that the restriction Pic([A2 /Gm ]) → Pic(M1,1 )
is an equivalence.
(b) Show that Pic(M1,1 ) = Z/12.
Show that the restriction Pic(M1,1 ) → Pic(M1,1 ) is surjective and that the
image of O(∆) = O(12) is trivial. Show that the images of O(4) and O(6) are
non-trivial by considering their restrictions to the residual gerbes of the unique
elliptic curves with Z/4 and Z/6 automorphism groups. See also [Mum65]
(and [FO10] for a generalization over an arbitrary base).

6.1.5 Sheaf cohomology

Sheaf cohomology for algebraic stacks can be developed using essentially the same
approach as we used in §4.1.7 for Deligne–Mumford stacks.

Lemma 6.1.23. If X is an algebraic stack, the categories Ab(Xlis-ét ) and QCoh(X )

have enough injectives.

Proof. The category of abelian sheaves on any site has enough injectives [SP,
Tag 01DP] and QCoh(X ) is a Grothendieck abelian category, which implies that it
has enough injectives [SP, Tag 079H].

Definition 6.1.24 (Cohomology). Let X be an algebraic stack and F be a sheaf of

abelian groups on Xlis-ét . The cohomology group Hi (Xlis-ét , F ) is defined as the ith
right derived functor of the global sections functor Γ : Ab(Xlis-ét ) → Ab.

Čech cohomology is defined just as in the Deligne–Mumford case (Definition 4.1.39).

Definition 6.1.25 (Čech cohomology). Given a smooth covering U = {Ui → X }i∈I

of algebraic stacks and an abelian sheaf F on Xlis-ét , the Čech complex of F with
respect to U is Cˇ• (U, F ) where
Y
Cˇn (U, F ) = F (Ui0 ×X · · · ×X Uin )
(i0 ,...,in )∈I n+1

with differential
 n+1
X 
d : Cˇn (U, F ) → Cˇn+1 (U, F ),
n
(si0 ,...,in ) 7→ k ∗
(−1) pbk si0 ,...,ibk ,...,in
k=0 (i0 ,...,in+1 )

where pbk : Ui0 ×X · · · ×X Uin → Ui0 ×X · · · ×X U d ik ×X · · · ×X Uin is the map forgetting

the kth component (with indexing starting at 0). The Čech cohomology of F with
respect to U is
Ȟi (U, F ) := Hi (Cˇ• (U, F )).

The arguments of Lemma 4.1.41, Theorem 4.1.42, Proposition 4.1.43, and Propo-
sition 4.1.44 extend to algebraic stacks.

Theorem 6.1.26. For a quasi-coherent OXlis-ét -module F on an affine scheme X,

Hi (Xlis-ét , F ) = 0 for all i > 0.

310
Lemma 6.1.27. Let X be an algebraic stack with affine diagonal and F be a quasi-
coherent sheaf. If U = {Ui → X } is an étale covering with each Ui affine, then
Hi (Xlis-ét , F ) = Ȟi (U, F ).

Proposition 6.1.28. If X is a Deligne–Mumford stack (e.g., a scheme) with affine

diagonal and F is a quasi-coherent sheaf, then Hi (Xét , F ) = Hi (Xlis-ét , F ) for all
i.

A similar argument shows that the cohomology of a quasi-coherent sheaf F can

also be computed in the big fppf topology. For this reason, it is customary to drop
the subscript and simply write Hi (X , F ).
Proposition 6.1.29. If X is an algebraic stack with affine diagonal of F ∈ QCoh(X ),
then the cohomology Hi (Xlis-ét , F ) of F as an abelian sheaf agrees with the ith right
derived functor of Γ : QCoh(X ) → Ab.
If X does not have affine diagonal, then the sheaf cohomology Hi (X , F ) of a quasi-
coherent sheaf may differ from the ith right derived functor of Γ(X , −) : QCoh(X ) →
Ab.
Exercise 6.1.30. Let E be an elliptic curve over a field and BE be its classifying
stack. Show that H2i (BE, OBE ) ̸= 0 for each i, but that the cohomology is zero if
it is computed in QCoh(BE).
Proposition 6.1.31. If X is an algebraic stack and Fi is a directed system of
abelian sheaves in Xlis-ét , then colimi Hi (Xlis-ét , Fi ) → Hi (Xlis-ét , colimi Fi ) is an
isomorphism.

Example 6.1.32 (Group Cohomology). Group cohomology for an affine algebraic

group G over a field k can be interpreted as sheaf cohomology on BG just as in
the case of a finite group (Example 4.1.47). Indeed, a G-representation V can
viewed as a quasi-coherent OBG -module, and by Proposition 6.1.29, the sheaf
cohomology Hi (BG, V ) agrees with the group cohomology Hi (G, V ), defined as
the ith derived functor of Rep(G) → Vectk , given by V 7→ V G . Moreover, by
Lemma 6.1.27, we may compute the cohomology using the Čech complex Č • (U, V )
of V with respect to the smooth cover U = {Spec k → BG}. This complex has terms
Č n (U, V ) = Γ(G, OG )⊗n ⊗ V with differentials given by (4.1.48), and corresponds
to the standard or bar resolution in group cohomology.
If G is linearly reductive, i.e., V → V G is exact, then Hi (G, V ) = 0 for every
G-representation V and i > 0. If G acts on an affine scheme Spec A over k, then it
follows that Hi ([Spec A/G], F ) = 0 for every quasi-coherent sheaf F and i > 0.
Exercise 6.1.33.
(a) If k is characteristic 0 and k is the trivial Ga -representation, show that
Hi (BGa , k) = k for i = 0, 1 and zero otherwise.
(b) If G is an affine algebraic group over a field k, show that Hi (BG, V ) is finite
dimensional for all i and zero for i ≫ 0.
Hint: Choose an embedding G ,→ GLn , use that GLn is linearly reductive
in characteristic 0 (Theorem B.1.43), and consider the composition BG →
BGLn → Spec k.

311
 (c) Show that H1 (BGa , k) is infinite dimensional.
In positive characteristic, Hi (G, V ) may be nonzero for arbitrary large i; this
already occurs in the case of finite groups such as Z/pZ (Exercise 4.1.49).

6.1.6 Chow groups

Following [Tot99] and [EG98], we introduce the Chow groups of a quotient stack.
Let G be a smooth affine algebraic group over an algebraically closed field k of
dimension g, and let X be an n-dimensional scheme of finite type over k. For each i,
choose an r-dimensional G-representation V such that there is a nonempty open
subscheme U ⊆ A(V ) such that (a) G acts freely on U , (b) the quotient U/G is
a scheme, and (c) codim A(V ) ∖ U > n − i − g. It is not too hard to prove that
such representations exist. We define the (i − g)th equivariant Chow group of X or
equivariantly the ith Chow group of [X/G] as

CHG G
i−g (X) = CHi ([X/G]) := CHi+r (X × U ).

Exercise 6.1.34.
(a) Show that there exists a representation V with properties (a)–(c).
(b) Check that the definition of CHG
i−g (X) is independent of the choices.

While this definition may at first appear mysterious, it is actually forced upon
us if we require the Chow groups to be invariant under and open immersions of high
codimension and under vector bundles:

X ×G U  / [(X × A(V ))/G] CHi+r (X ×G U ) o
open ∼
CHi+r ([(X × A(V ))/G])
O
vect bdl ∼

[X/G] CHi ([X/G]).

If [X/G] is smooth of pure dimension d = n − g, then we define

CHiG (X) = CHi ([X/G]) := CHd−i ([X/G])

M
CH∗G (X) = CH∗ ([X/G]) := CHi ((X/G]).
i

The intersection product gives a ring structure, and we call CH∗G (X) the equivariant
Chow ring of X and CH∗ ([X/G]) the Chow ring of [X/G].
Example 6.1.35 (CH∗ (BGm )). Let V be the r-dimensional Gm -representation over
k with equal weights 1. Then Gm acts freely on An ∖ 0, and for −1 ≥ i > −r − 1, we
have that CHi (BGm ) = CHi+r (Pr−1 ) = Z. It follows that CHi (BGm ) = Z for i ≤ −1
and is 0 otherwise. Therefore CHj (BGm ) = Z for j ≥ 0, and CH∗ (BGm ) = Z[x].
More generally, if T ∼ = Grm is a rank r torus, then CH∗ (BT ) is isomorphic to the
character ring Z[x1 , . . . , xr ] of T .
We summarize some of the important properties of equivariant Chow groups.
Properties 6.1.36.
(1) (Independence of quotient presentation) If [X/G] ∼
= [X ′ /G′ ], then CHG ∼
i (X) =
G′ ′
CHi (X ), and in particular the definition of CHi ([X/G]) is independent of
the quotient presentation.

312
 (2) (Vector bundle invariance) If Y → X is G-equivariant and a Zariski-local
affine fibration of relative dimension r (e.g., the total space of a rank r vector
bundle), then CHG ∼ G
∗ (X) = CH∗+r (Y ).
(3) (Excision sequence) If Z ⊆ X = [X/G] is a closed substack with complement
U, then there is a right exact sequence

CH∗ (Z) → CH∗ (X ) → CH∗ (U) → 0.

(4) (Comparison with coarse moduli space) If X ∼= [X/G] is a separated Deligne–

Mumford stack with coarse moduli space X, then CH∗ (X ) ⊗ Q ∼
= CH∗ (X) ⊗ Q.
(5) (Functoriality) Flat morphisms induce pullback maps on Chow groups while
proper morphisms induce pushforward maps.
(6) (Self-intersection) If X = [X/G] is smooth and i : Z ,→ X is a smooth substack
of pure codimension d, then there is pullback i∗ : CH∗ (X ) → CH∗ (Z) given by
intersection with Z such that i∗ i∗ α = cd (NZ/X ) ∩ α for α ∈ CH∗ (Z), where
cd is the top Chern class of the normal bundle.
Exercise 6.1.37. Prove Properties 6.1.36.
Hint: Reduce to the usual properties of Chow groups of schemes, c.f., [Ful98].
Kresch has extended the theory of Chow groups to (not necessarily quotient)
algebraic stacks; see [Kre99].
Exercise 6.1.38.
(a) If T is a torus acting trivially on a smooth scheme X, show that CHT∗ (X) ∼=
CH∗ (X) ⊗ CH∗ (BT ). is a product of two tori with T2 acting trivially. Then
CH∗T (X) ∼= CH∗T1 (X) ⊗ CH∗ (BT2 ).
(b) If G is a connected reductive group with maximal torus T , and X is a smooth
scheme with a G-action, then the Weyl group W = NG (T )/T acts on CH∗T (X).
Show that CH∗G (X)Q = CH∗T (X)W Q and that if in addition G is special (Exam-
ple B.1.59), then CHG (X) = CH∗T (X)W .
∗

Exercise 6.1.39.
(a) Let P(d0 , . . . , dn ) be the weighted projective stack of Example 3.9.14. Show
that CH∗ (P(d0 , . . . , dn )) ∼
= Z[x]/(d1 · · · dn xn+1 ).
(b) If char(k) ̸= 2, 3, show that CH∗ (M1,1 ) ∼ = Z[x]/(12x) and CH∗ (M1,1 ) ∼
=
Z[x]/(24x ). (Compare to the Picard group from Exercise 6.1.22).
2

(c) Let Gm act on Pn with weights Pn d0 ,i . . . , dn . Show that A ([P /Gm ]) =

∗ n

Z[h, t]/p(h, t), where p(h, t) = i=0 h ei (a0 t, . . . , an t) and ei is the ith sym-
metric polynomial.

6.1.7 de Rham and singular cohomology

We quickly discuss the de Rham and singular cohomology of an algebraic stack
following [Beh04].
Analyticification. If X is a smooth algebraic stack over C with affine diagonal,
there is an analyticification X an , analogous to the analyticification of a finite type
C-scheme, such that X an is a differentiable stack. If U0 → X is a smooth presentation
by a scheme so that X is the quotient of the smooth groupoid U1 := U0 ×X U0 ⇒ U0 ,
then U0an → X an is a smooth presentation and X an is the quotient of the Lie groupoid
U1an ⇒ U0an .

313
De Rham cohomology of a differential stack. Given a differentiable stack X
with a smooth presentation U0 → X , we can define a simplicial manifold
// // U / U0 , where Up := U0 ×X · · · ×X U0 (6.1.40)
U• : · · · U3 / U2 1
| {z }
p + 1 times

with differential maps ∂0 , . . . , ∂p : Up → Up−1 forgetting the ith term and degen-
eracy maps si : Up−1 → Up inserting an identity morphism in the ith term. This
defines a double complex Ωq (Up )P with differentials given by exterior differentiation
p
d : Ωq−1 (Up ) → Ωq (Up ) and ∂ := i=0 (−1)i ∂i∗ : Ωq (Up−1 ) → Ωq (Up ). We define the
de Rham complex CdR (X ) as the total complex
•

M
k
CdR (X ) := Ωq (Up ),
p+q=k

with differential δ : CdR

k
(X ) → k+1
CdR defined by δ(ω) = ∂(ω) + (−1)p d(ω) for
(X )
ω ∈ Ωp (Uq ). The de Rham cohomology is
•
HndR (X ) := Hn (CdR (X )),
and is independent of the choice of presentation.
Singular homology/cohomology of a topological stack. For a topological
stack X , one can replicate the constructions of singular homology and cohomology.
Let U0 → X be a presentation and U• be the simplicial topological space as in
(6.1.40). For each p, we have the singular chain complex C• (Up ) with differentials
q (Up ) → Cq−1 (Up ). This defines a double complex Cq (Up ) using the differential
d : CP
p
∂ = i=0 (−1)i ∂j : Cq (Up ) → Cq (Up−1 ) induced by the maps ∂i : Up → Up−1 . The
singular chain complex C• (X ) of X is defined as the total complex
M
Ck (X ) := Cq (Up )
p+q=k

with the differential δ : Ck (X ) → Ck−1 (X ) given by δ(γ) = (−1)p+q ∂(γ) + (−1)q d(γ)
for γ ∈ Cq (Up ). For an abelian group A, we can therefore define the singular
homology groups of X with coefficients in A as
Hn (X , A) := Hn (C• (X ) ⊗Z A).
Dualizing, we define the singular cochain complex C • (X ) by C n (X ) := Hom(Cn (X ), Z)
and the singular cohomology groups of X with coefficients in A as
Hn (X , A) := Hn (C • (X ) ⊗Z A).

Comparisons.
– There are pairings Hk (X , Z) ⊗ Hk (X , Z) → Z which after tensoring with Q
gives identifications Hk (X , Q) ∼
= Hk (X , Q)∨ .
– If G is a topological group acting on a space U , then the equivariant cohomology
is defined as H∗G (U, A) := H∗ (EG ×G U, A), where EG is a contractible space
with a free action of G, and there is an identification H∗ ([U/G], A) = H∗G (U, A).
– For a differential stack X , there is an identification H∗dR (X ) with the cohomol-
ogy H∗ (X , R) of the constant sheaf R computed in the big site of manifolds
over X where coverings are usual open coverings.
– If X is a topological Deligne–Mumford stack (e.g., the topological stack associ-
ated to a separated Deligne–Mumford stack of finite type over C) with coarse
moduli space X → X, then H∗ (X , Q) = H∗ (X, Q).

314
6.2 Global quotient stacks and the resolution prop-
erty
Did you ever stop to think, and forget to start again?

Winnie the Pooh

Quotient stacks of the form [U/G] form an important class of algebraic stacks.
Their geometry is readily accessible as the G-equivariant geometry of U (see Ta-
ble 0.6.15). It is an interesting question on when algebraic stacks are quotient stacks,
related to both geometric properties such as the resolution property and to arithmetic
properties such as the surjectivity of the Brauer map. In the Local Structure of
Deligne–Mumford Stacks (4.3.1), we showed that every Deligne–Mumford stack has
an étale stabilizer-preserving neighborhood which a quotient stack [Spec A/G] by
a finite group, and we will shortly prove an analogous result for algebraic stacks
around points with linearly reductive stabilizer (see Theorem 6.7.1).

6.2.1 Global quotient stacks

Definition 6.2.1. An algebraic stack X is a global quotient stack if there exists an
isomorphism X ∼= [U/ GLn ] where U is an algebraic space.
In other words, X is a global quotient stack if and only if there is a principal GLn -
bundle U → X from an algebraic space, or equivalently a representable morphism
X → BGLn .

Exercise 6.2.2 (important). Show that a noetherian algebraic stack X is a global

quotient stack if and only if there exists a vector bundle E on X such that for every
geometric point x : Spec k → X with closed image, the stabilizer Gx acts faithfully
on the fiber E ⊗ k.
Hint: Use the correspondence between principal GLn -bundles and vector bundles
from Exercise B.1.56.

Exercise 6.2.3. Show that a smooth Deligne–Mumford stack X over a noetherian

scheme with generically trivial stabilizer is a global quotient stack.
Hint: Consider the tangent bundle and the higher jet bundles.

Exercise 6.2.4. If X is a reduced noetherian algebraic stack with affine stabilizers,

show that there is a stratification X = ⨿i Xi , where each Xi ,→ X is a locally closed
substack and a quotient stack. If X is normal, show that Xi be written as a quotient
stack [Ui / GLn ] where Ui is a quasi-projective scheme Ui .
Hint: Construct a coherent sheaf with the property such that every stabilizer acts
faithfully on the fiber, and consider the open locus where it is a vector bundle. For
the final statement, apply Sumihiro’s Theorem on Linearizations (B.1.30).

In Exercise 4.3.2, we’ve seen that a Deligne–Mumford stack can be written as a

quotient stack [U/G] with G a finite group and U an affine scheme (resp., scheme,
algebraic space) if and only if it admits a finite étale cover by an affine scheme (resp.,
scheme, algebraic space). In a similar spirit, we have:

Exercise 6.2.5. Let X → Y be a surjective, flat, and projective morphism of

noetherian algebraic stacks. If X is a quotient stack, show that Y is a quotient stack.

315
Remark 6.2.6 (Quasi-projective stacks). Let X be a separated Deligne–Mumford of
finite type over a field k with coarse moduli space π : X → X. It is a theorem of
Kresch [Kre09, Thm. 5.3] that X is quasi-projective and X is a global quotient stack
if and only if X ∼= [U/ GLn ] with U quasi-projective. Moreover, if k is characteristic 0,
this if further equivalent to (a) X is quasi-projective and X has a generating coherent
sheaf V, i.e., for every quasi-coherent sheaf F on X, π∗ π ∗ H omOX (V, F )⊗OX V → F
is surjective, or to (b) there exists a locally closed immersion X ,→ W, where W
is a smooth proper Deligne–Mumford stack over k with projective coarse moduli
space. See also the generalization in [BOW24].
Remark 6.2.7 (Non-quotient stacks). It is not easy to give examples of algebraic
stacks with affine stabilizers that are not global quotient stacks. For instance, it is not
known whether every separated Deligne–Mumford stack is a global quotient stack. A
necessary condition for a noetherian algebraic stack to be a global quotient is that it
have quasi-affine diagonal since every quasi-separated algebraic space has quasi-affine
diagonal (Corollary 4.5.8). In [Kre13, Prop. 5.2], Kresch showed that the moduli
stack M≤m 0 parameterizing prestable curves of genus 0 with at most m ≥ 2 does not
have quasi-affine diagonal. Thus, M≤m 0 is not a global quotient. Another necessary
condition is that X has sufficiently many vector bundles (Exercise 6.2.2), and this
can also be used to construct counterexamples (see Exercise 6.2.8). Non-quotient
stacks can also be constructed using the non-surjectivity of the Brauer map (see
Exercise 6.4.38). Give two forward references to examples using Brauer map
Exercise 6.2.8. Consider a split torus T = Grm,P1 of rank at least r ≥ 2 over P1 = P1k
for a field k. Show that there is a non-trivial involution α : T0 = Grm,k → Grm,k = T∞
which glues to a non-split torus T ′ over the projective nodal cubic C = P1 /(0 ∼ ∞)
such that BT ′ has no vector bundles. Conclude that BT ′ is not a global quotient
stack.

6.2.2 The resolution property and Totaro’s Theorem

Being a quotient stack is closely related to the resolution property.
Definition 6.2.9. A noetherian algebraic stack has the resolution property if every
coherent sheaf is the quotient of a vector bundle.
A smooth or quasi-projective scheme over a field has the resolution property. More
generally, a scheme admitting an “ample family” of line bundles has the resolution
property, and this implies that every noetherian normal Q-factorial scheme with
affine diagonal has the resolution property [BS03].
Proposition 6.2.10. Let G be an affine algebraic group over a field k acting on a
quasi-projective k-scheme U . Assume that there is an ample line bundle L with an
action of G (e.g., U is quasi-affine and L = OU ). Then [U/G] has the resolution
property.
Proof. The line bundle L corresponds to a line bundle L on [U/G] which is relatively
ample with respect to the morphism p : [U/G] → BG. For a coherent sheaf F on
[U/G], the natural map
L−N ⊗ p∗ p∗ (LN ⊗ F ) ↠ F
is surjective for N ≫ 0. The pushforward p∗ (LN ⊗ F ) is a quasi-coherent sheaf on
BG, i.e., a G-representation, which we can write as a union of finite dimensional
G-representations Vi (B.1.17(1)). We therefore obtain a surjection colimi (L−N ⊗
p∗ Vi ) ↠ F. Since F is coherent, L−N ⊗ p∗ Vi ↠ F is surjective for i ≫ 0.

316
Exercise 6.2.11. Show that [U/ GLn ] has the representation property if U is a
quasi-projective scheme over Z with an action of GLn such that U has an ample
line bundle with a GLn -action. (See [Tho87, §2] for further generalizations.)
Totaro’s Theorem gives a converse.
Theorem 6.2.12 (Totaro’s Theorem). Let X be a normal noetherian algebraic stack
whose stabilizer groups at closed points are affine. Then the following are equivalent
(1) X has the resolution property,
(2) X ∼
= [U/ GLn ] where U is quasi-affine, and
if in addition X is of finite type over a field k, then the above are equivalent to
(3) X ∼= [Spec A/G] where G is a smooth affine algebraic group over k.
In particular, X has affine diagonal.
Remark 6.2.13. This has been extended to non-normal stacks by Gross [Gro17].
Proof. Since X is normal, Sumihiro’s Theorem on Linearizations (B.1.30) implies
that OU admits a GLn -action (resp., OSpec A admits a G-action), and therefore
Exercise 6.2.11 (resp., Proposition 6.2.10) yields the implication (2) ⇒ (1) (resp.,
(3) ⇒ (1)).
To see (3) ⇒ (2), it suffices to find a faithful representation G ,→ GLN over k
such that GLN /G is quasi-affine. Indeed, in this case, [Spec A/G] ∼ = [(Spec A ×G
GLN )/ GLN ] (Exercise 3.4.19) and Spec A × GLN is affine over GLN /G. We
G

begin by choosing any faithful representation G ⊆ GLn . By B.1.16(7), GLn /G is

quasi-projective, and so there is a GLn -representation V and a k-point x ∈ P(V )
with stabilizer G. Under the action of GLn ×Gm on A(V ) (where Gm acts via
scaling), the stabilizer of a lift xe ∈ A(V ) of x is G. The map (GLn ×Gm )/G ,→
A(V ), defined by g 7→ ge x, is a locally closed immersion and thus (GLn ×Gm )/G
is quasi-affine. Under the natural inclusion GLn ×Gm ,→ GLn+1 , the quotient
GLn+1 /(GLn ×Gm ) is affine (and is sometimes called the “Steifel manifold”). The
composition BG → B(GLn ×Gm ) → BGLn+1 is quasi-affine and therefore so is
GLn+1 /G and U := Spec A ×G GLN .
Conversely for (2) ⇒ (3), we first choose a GLn -equivariant open immersion
U ,→ Spec A into an affine scheme of finite type over k: indeed, since the morphism
p : [U/ GLn ] → BGLn is quasi-affine, [U/ GLn ] → SpecBGLn p∗ O[U/ GLn ] is an open
immersion, and thus by writing p∗ O[U/ GLn ] = colimλ Aλ as a colimit of finite
type OBGLn -algebras, Limit Methods (B.3) imply that [U/ GLn ] → SpecBGLn Aλ
is an open immersion for λ ≫ 0. Letting Z ⊆ Spec A be the reduced complement
of U , choose a GLn -equivariant morphism f : Spec A → Ar such that f −1 (0) = Z
(Proposition B.1.18(2)). This gives an affine morphism U → Ar ∖ 0. The complement
Ar ∖0 can be realized as the quotient GLr /H where H ∼ = Gr−1a ⋊GLr−1 ⊆ GLr is the
subgroup consisting of matrices whose last row is (0, . . . , 0, 1). In the GLn -equivariant
cartesian diagram
P / GLr

□
 
U / Ar ∖ 0,
∼
P is affine over GLr , thus affine. We conclude using the equivalence [U/ GLn ] =
[P/(GLn ×H)].
It remains to show (1) ⇒ (2). We first show that there is a vector bundle on
E such that every stabilizer group Gx acts faithfully on the fiber E ⊗ κ(x). By

317
Exercise 6.2.2, this implies that X = [U/ GLn ] where n = rk E and U = Fr(E) is
the frame bundle. For every closed point x ∈ X , let ix : Gx ,→ X be the inclusion of
the residual gerbe (Proposition 3.5.16). Choosing a finite type point x : Spec k → X
representing x, there is an induced finite morphism BGx → Gx . Since Gxe is an affine
algebraic group, we can choose a faithful representation W (Proposition B.1.18(2)).
Using the resolution property on X , there is a vector bundle E and a surjection
E ↠ (ix ◦ p)∗ W . The kernel subgroup SE ⊆ IX of E, i.e., the subgroup stack of
the inertia stack parameterizing elements acting trivially on E, is trivial over x.
If F is another vector bundle, then SE⊕F ⊆ SE is a closed subgroup. Since IX is
noetherian, we can use noetherian induction to choose a vector bundle E with trivial
subgroup SE .
Since X is normal, so is U , and we may apply Exercise 4.6.4 to arrange that
U is the coarse moduli space of the action of a finite group H acting on a normal
scheme U ′ . Let g : U ′ → U be the quotient morphism, and let {Ui′ } be L a finite
affine covering with reduced complementsL Zi′ = U ′ ∖ Ui′ . Then F := g∗ ( i IZi′ ) is
a coherent sheaf on U such that g ∗ F ↠ i IZi′ is surjective. Since q : U → X is
quasi-affine, q ∗ q∗ F ↠ F is surjective, and by writing q∗ F as a colimit of coherent
sheaves, we may find a coherent sheaf G on X and a surjection q ∗ G → F . Since X
has the resolution property, weL can arrange that G is a vector bundle. This gives
a surjection g ∗ q ∗ G ↠ g ∗ F ↠ i IZi . Letting V = Fr(G) be the frame bundle,
′

consider the cartesian diagram

UV′ / UV /V

β □ □
  
U′
g
/U q
/ X,

where g is a principal H-bundle, q is a principal GLn -bundle, and V → X is a

principal GLL m -bundle (with m = rk G). Since the pullback of G to V is trivial, the
pullback β ∗ ( i IZi′ ) is globally generated. This implies that β −1 (Zi′ ) is defined by
global functions on UV′ and that the complement β −1 (Ui′ ) is covered by affine opens
of the form {f ̸= 0} for f ∈ Γ(UV′ , OUV′ ). In other words, OUV′ is ample and UV′
is a quasi-affine scheme. Since β : UV′ → UV is the quotient by the finite group H,
UV is also quasi-affine (Exercise 4.2.14). We can write X ∼ = [Fr(E ⊕ F )/ GLn+m ],
where the frame bundle Fr(E ⊕ F ) is identified with UV ×(GLn × GLm ) GLn+m , which
is quasi-affine since UV is quasi-affine and the quotient GLn+m /(GLn × GLm ) is
affine.

6.3 The fppf topology

Waste not thy time in windy argument but let the matter drop.

William Shakespeare

While not essential for the two main theorems of this book, we cover the important
facts that algebraic spaces/stacks are sheaves/stacks in the fppf topology and that
quotients by fppf groupoids/equivalence relations are algebraic. One upshot is that
BG is an algebraic stack for any (non-necessarily smooth) algebraic group, e.g., µ p
in characteristic p. The fppf topology will allow us to define gerbes in a suitable
generality in the next section.

318
6.3.1 Fppf criterion for algebraicity
Theorem 6.3.1 (Fppf Criterion for Algebraicity).
(1) If X is a sheaf on Schfppf such that there exists an fppf representable morphism
U → X from an algebraic space, then X is an algebraic space.
(2) If X is a stack over Schfppf such that there exists an fppf representable morphism
U → X from an algebraic space, then X is an algebraic stack.
Proof. Part (1) follows from (2) since an algebraic stack that is equivalent to a sheaf
is an algebraic space (Theorem 4.5.10). The proof uses a technique analogous to the
general Keel–Mori Theorem (see Remark 4.4.11). Since U → X is representable, the
relative Hilbert stack Hilbd over X , whose objects over S → X consist of a closed
subscheme Z ,→ U ×X S finite and flat of degree d over S, is an algebraic stack.
Restricting to the open substack` Hilbd,lci parameterizing local complete intersections
Z ,→ U ×X S, one shows that d Hilbd → X is smooth and surjective using the
Infinitesimal Lifting Criterion (3.7.1). For details, see [Art74b, Thm. 6.1], [LMB00,
Thm. 10.1], and [SP, Tag 06DC].
Remark 6.3.2. The result is not true if you replace ‘fppf’ with ‘fpqc’; see [MO10].
Algebraic spaces are by definition sheaves in the big étale topology but also
sheaves in the big fppf topology.
Corollary 6.3.3 (Algebraic Stacks are Fppf Stacks).
(1) An algebraic space X over a scheme S is a sheaf on (Sch/S)fppf .
(2) An algebraic stack X over a scheme S is a stack over (Sch/S)fppf .
Proof. It suffices to show (2). Let U → X be a smooth presentation and X ′ be the
stackification of X in the fppf topology. Since U → X ′ is fppf and representable,
X ′ is an algebraic stack by Theorem 6.3.1. As U ×X U ∼ = U ×X ′ U , the natural
map X → X ′ is fully faithful. On the other hand, if T → X ′ is any map from
a scheme, the fiber product U ×X ′ T is a smooth algebraic space over T . Since
smooth morphisms have sections étale locally on the target, there exists an étale
cover T ′ → T together with a map T ′ → U lifting T ′ → T → X . The composition
T ′ → U → X shows that objects of X ′ étale locally lift to X , which implies that
X → X ′ is essentially surjective. See [LMB00, Cor. 10.7(a)] and [SP, Tag 076V]
Remark 6.3.4. In the Stacks Project [SP], an algebraic stack is by definition a stack
in the fppf topology, and the above result ensures that our definition of an algebraic
stack is equivalent to theirs. It is also true that algebraic spaces are sheaves in
the fpqc topology, a result attributed to Gabber; see [LMB00, Thm. A.4] and [SP,
Tag 0APL]. It is also known that algebraic stacks with quasi-affine diagonal are
stacks in the fpqc topology [LMB00, Cor. 10.7(b)].
Exercise 6.3.5. Extend Proposition 3.3.6 by showing that the following properties of
morphisms of algebraic stacks are fppf local on the target: representable, isomorphism,
open immersion, closed immersion, locally closed immersion, affine, or quasi-affine.

6.3.2 Fppf groupoids and quotient stacks

If R ⇒ U is an fppf equivalence relation of algebraic spaces, we define U/R as the
sheafification in big fppf topology Schfppf of the presheaf T 7→ U (T )/R(T ). Likewise,
if s, t : R ⇒ U is an fppf groupoid of algebraic spaces, we define [U/R] as the
stackification in Schfppf of the prestack [U/R]pre , whose fiber category [U (T )/R(T )]

319
over a scheme T is the category of T -points of U where a morphism from a ∈ U (T )
to b ∈ U (T ) is an element r ∈ R(R) such that s(r) = a and t(r) = b. The definitions
of U/R and [U/R] are consistent with the quotient of a smooth equivalence relation
or groupoid (Definition 3.4.10) as a consequence of Corollary 6.3.3.

Corollary 6.3.6.
(1) If R ⇒ U is an fppf equivalence relation of algebraic spaces, then the quotient
U/R is an algebraic space.
(2) If R ⇒ U is an fppf groupoid of algebraic spaces, then the quotient [U/R] is
an algebraic stack.

Proof. This follows from the Fppf Criterion for Algebraicity (6.3.1) since U → [U/R]
is fppf and representable.

As a result, we can construct quotient stacks by fppf group algebraic spaces

generalizing the quotient stacks by smooth affine group schemes from Theorem 3.1.10.
We first need some definitions.

Definition 6.3.7 (Principal G-bundles). If G → S is an fppf group algebraic space,

then a principal G-bundle over an S-scheme T is an algebraic space P with an action
of G via σ : G ×S P → P such that P → X is a G-invariant fppf morphism and
(σ, p2 ) : G ×S P → P ×T P is an isomorphism. Morphisms of principal G-bundles
are G-equivariant morphisms of schemes. We say that a principal G-bundle P → T
is trivial if there is a G-equivariant isomorphism P ∼
= G × T.

In the above definition, it is important to require that P is an algebraic space as

will want principal G-bundles to satisfy descent in étale topology. Raynaud provides
an example of an abelian variety G and a principal G-bundle that is not scheme
[Ray70, XIII 3.2]. When G → S is smooth and affine, this definition agrees with
Definition B.1.47: indeed, P is a scheme since it is affine over T and since every
principal G-bundle P → T is trivialized by the smooth cover P → T , we may use
the fact that smooth morphisms étale locally have sections to construct an étale
cover T ′ → T such that PT ′ is trivial.

Definition 6.3.8 (Quotient and classifying stacks). Let G → S be an fppf group

algebraic space acting on an algebraic space U over S. We define the quotient stack
[U/G] as the category over Sch/S whose objects over an S-scheme T are diagrams

T ←P →U (6.3.9)

where P → T is a principal G-bundle and P → U is a G-equivariant morphism of

schemes. A morphism (T ′ ← P ′ → U ) → (T ← P → U ) consists of a morphism
T ′ → T and a G-equivariant morphism P ′ → P of schemes such that the diagram

/P '/
P′ U
□
 
T′ /T

is commutative and the left square is cartesian.

The classifying stack BG of G, defined as the quotient stack [S/G], classifies
principal G-bundles P → T .

320
Proposition 6.3.10. If G → S is an fppf group algebraic space acting on an
algebraic space U over S, then the quotient stack [U/G] is an algebraic stack. In
particular, the classifying stack BG is algebraic.
Proof. Given a map T → [U/G] corresponding to an object (6.3.9), there is a
cartesian diagram
P /U

□
 
T / [U/G]

of stacks over Schfppf (see Exercise 2.4.37). As P → T is an fppf morphism of

algebraic spaces, U → [U/G] is an fppf representable morphism, the Fppf Criterion
for Algebraicity (6.3.1) implies that [U/G] is an algebraic stack.
Exercise 6.3.11. Let G → S be an fppf group algebraic space acting on an algebraic
space U over S.
(a) Show that the stackification of the prestack [U/G]pre in the fppf topology is
[U/G]. (This generalizes Exercise 2.5.21.)
(b) (hard) Provide an example where the stackification of [U/G]pre in the étale
topology is not isomorphic to [U/G].
Recalling that µ n,Z is the subgroup of Gm,Z defined by Spec Z[x]/(xn − 1), we
can now deduce that Bµ µn,Z is an algebraic stack. If k is a field of characteristic p,
then µ n := µ n,k is smooth and Bµ µn is Deligne–Mumford if and only if p doesn’t
divide n.
Exercise 6.3.12. Let k be a field.
(a) Exhibit an explicit smooth presentation of Bµ µn .
(b) Show that Bµµn is equivalent to the stack over (Sch/k)ét whose objects over a
scheme T are pairs (L, α) consisting of a line bundle L on T and a trivialization
∼
α : OT → L⊗n .
(c) Show that Bµµn is a smooth and proper algebraic stack of dimension 0.
(d) If x : Spec k → Bµ
µn denotes the canonical presentation, compute the tangent
space TBµµn ,x .
Hint: The answer depends on the characteristic.
(e) (hard) Show that the stack parameterizing µ n -bundles that are trivializable in
the étale topology is not algebraic.

6.4 Gerbes, rigidification, and Picard schemes

Most hyperelliptic curves are pointless.

Don Zagier

Gerbes play a central role in the theory of stacks. For the purposes of this
book, we want to know that residual gerbes are gerbes (Proposition 6.4.26) thereby
justifying the terminology and that the map Pic X → PicX from the Picard stack
to Picard scheme and Bun sr,d (C) → Mr,d
s
(C) from the stack of stable vector bundles
to its coarse moduli space is a banded Gm -gerbe (Corollary 8.2.23).

321
6.4.1 Torsors
For a sheaf G of groups, a G-torsor is a sheaf of sets locally isomorphic to G. When
G is representable by an fppf group scheme, this notion is equivalent to that of a
principal G-bundle.

Definition 6.4.1 (Torsors). Let S be a site and G a sheaf of (not necessarily

abelian) groups on S. A G-torsor on S is a sheaf P of sets on S with a left action
σ : G × P → P of G such that
(1) for every object T ∈ S, there exists a covering {Ti → T } such that P (Ti ) ̸= 0
for each i, and
(2) the action map (σ, p2 ) : G × P → P × P is an isomorphism.
If T ∈ S is an object and G is a sheaf of groups on the restricted site S/T
(Example 2.2.12), then a G-torsor over T is by definition a G-torsor on the S/T .

Morphisms of G-torsors are G-equivariant morphisms of sheaves. We say that a

G-torsor P is trivial if P is G-equivalently isomorphic to G.

Exercise 6.4.2. Show that any morphism of G-torsors is an isomorphism.

Example 6.4.3. Let X be a stack over a site S, and let a, b ∈ X be objects over
S ∈ S. The sheaf IsomS (a, b) of isomorphisms is a torsor for Aut(a) under the action
given by precomposition.

Given a morphism f : T ′ → T and a G-torsor P over T , the restriction P |T ′ ,

which is the sheaf on S/T ′ whose whose sections over a T ′ -scheme S are P (S), is
naturally a G-torsor over T ′ .

Exercise 6.4.4. Let S be a site with a final object S and G be a sheaf of groups
on S.
(a) Show that Axiom (1) is equivalent to P → S being an epimorphism of sheaves.
(b) If P is a G-torsor, show that S is isomorphic to the quotient sheaf P/G.
(c) Show that a G-torsor P is trivial if and only if there exists a section s : S → P
of the structure morphism P → S.
(d) Show that a sheaf P of sets on S with a left action by G is a G-torsor if and
only if there exists a covering {Si → S} and isomorphisms P |Si ∼ = G|Si of
G|Si -torsors.

Example 6.4.5 (Principal G-bundles). For an fppf group scheme G → S, there is

an equivalence of categories between G-torsors in the fppf topology and principal
G-bundles (as defined in Definition 6.3.7). To see this, observe that if P → T is a
principal G-bundle over an S-scheme T , then since algebraic spaces are sheaves in
the fppf topology (Corollary 6.3.3), both G ×S T and P are sheaves on (Sch/T )fppf .
Since P → T is trivialized by the fppf map P → T , Exercise 6.4.4(d) implies that
P → T is a G-torsor. Conversely, given a G-torsor P on (Sch/T )fppf and an fppf
cover T ′ → T trivializing P , i.e., P ×T T ′ ∼
= G ×T T ′ , it follows that P ×T T ′ is
an algebraic space and P ×T T → T is an fppf and representable morphism. The
′ ′

Fppf Criterion for Algebraicity (6.3.1) implies that P is an algebraic space, and
thus P → T is a principal G-bundle. If in addition G → S is smooth, then there is
an equivalence of categories between G-torsors in the étale topology and principal
G-bundles. This holds because smooth morphisms have sections étale locally and
thus every principal G-bundle P → T is étale locally trivial.

322
6.4.2 Gerbes
Gerbes are a 2-categorical generalization of torsors. While torsors are locally
isomorphic to a sheaf of groups G, gerbes are locally isomorphic to classifying stacks
BG.
Definition 6.4.6 (Gerbes). A stack X over a site S is called a gerbe if
(1) for every object T ∈ S, there exists a covering {Ti → T } in S such that each
fiber category X (Ti ) is non-empty; and
(2) for objects x, y ∈ X over T ∈ S, there exists a covering {Ti → T } and
∼
isomorphisms x|Ti → y|Ti for each i.
We say that a gerbe X is trivial if there is a section S → X of X → S. When S
has a final object S, triviality is equivalent to the non-emptiness of X (S).
Example 6.4.7. For a sheaf of groups G on a site S, we extend Definition 6.3.8
by defining the classifying stack of G as the stack BG over S consisting of pairs
(P, T ) where T ∈ S and P is G-torsor over S/T (Definition 6.4.1). A morphism
(P ′ , T ′ ) → (P, T ) is the data of a morphism T ′ → T in S and an isomorphism
P ′ → P |T ′ of G-torsors, where P |T ′ denotes the restriction of P along T ′ → T .
The classifying stack BG is a gerbe over S because every G-torsor over T is locally
isomorphic to the trivial G-torsor G × T .
Exercise 6.4.8 (Gerbes are locally classifying stacks). Let S be a site with a final
object S ∈ S, and let X be a stack over S. Show that X is a gerbe if and only if
there exists a covering {Si → S} and sheaves of groups Gi on the restricted site
S/Si (Example 2.2.12) such that there is an isomorphism X ×S Si ∼= BGi over S/Si .
Exercise 6.4.9. Let S be a scheme and let X be a gerbe over (Sch/S)fppf . If the
diagonal X → X ×S X is representable, show that X is an algebraic stack.
A banding of a gerbe X by a sheaf of groups G is the additional data of a natural
isomorphism G(T ) → AutT (x) for every object x ∈ X (t).
Definition 6.4.10 (Banded G-gerbes). Let G be an abelian sheaf on a site S. A
stack X over S is a gerbe banded by G (or a banded G-gerbe or simply a G-gerbe) is
a gerbe together with the data of isomorphisms ψx : G|T → AutT (x) of sheaves for
∼
each object x ∈ X (T ). We require that for each isomorphism α : x → y over T , the
diagram

G|T
ψx ψy
(6.4.11)
z $
AutT (x)
Innα
/ AutT (y).

commutes, where Innα (τ ) = ατ α−1 . The data of the isomorphisms ψx is called the
band of X . A morphism of banded G-gerbes is a morphism of stacks compatible with
the bands.
Remark 6.4.12. For another way to think about a band of a gerbe, let S/X be the
restricted site whose underlying category is X and where a covering of a ∈ X (S)
is a covering of S. Then the inertia stack IX = X ×X ×X X is a sheaf of groups
on S/X : for a ∈ X (S), we have IX (a) = IsomS (a). The compatibility condition
(6.4.11) ensures that there is an isomorphism ψ : G|X → IX of sheaves on S/X .

323
Example 6.4.13 (The trivial banded gerbe). If G is an abelian sheaf on a site S
with a final object S, then the classifying stack BG of Example 6.4.7 is a banded
G-gerbe and is trivial. A banded G-gerbe X over S is trivial if and only if X ∼
= BG.
Exercise 6.4.14 (Band associated to a gerbe). Let S be a site with a final object
S. Let X be an abelian gerbe over S, i.e., a gerbe X such that AutT (a) is abelian
for every object a ∈ X (T ). Show that there is a sheaf of groups G on S such that X
is banded by G.
Hint: Use Axiom (1) of a gerbe to find a covering {Xi → X} and elements ai ∈
X (Xi ). Use Axiom (2) to glue the sheaves Gi := AutXi (ai ) to a sheaf G.

6.4.3 Algebraic gerbes

Attached to any algebraic stack Y is the big fppf site (Sch/Y)fppf of schemes over Y:
the underlying category of (Sch/Y)fppf is Y and a covering of an object y ∈ Y(T ) is
a covering of T . If X → Y is a morphism of algebraic stacks, then X is a stack over
(Sch/Y)fppf thanks to Corollary 6.3.3.
Definition 6.4.15 (Gerbes and banded G-gerbes). A morphism X → Y of algebraic
stacks is a gerbe if X is a gerbe over the big fppf site (Sch/Y)fppf . If G → S is a
commutative fppf group scheme, a morphism X → Y of algebraic stacks over S is
called a banded G-gerbe (or simply G-gerbe) if X is a gerbe over (Sch/Y)fppf and is
banded by the sheaf of groups G ×S Y. We say that an algebraic stack X is a gerbe
(resp., banded G-gerbe) if there exists a morphism X → X to an algebraic space
which is a gerbe (resp., banded G-gerbe).
A banded G-gerbe X → X over an algebraic space X is trivial if and only if
∼ BG ×S X.
X =
Proposition 6.4.16. Let X → Y be a morphism of algebraic stacks.
(1) The morphism X → Y is a gerbe if and only if there exists an fppf morphism
S → Y from a scheme and an fppf group algebraic space G → S such that
X ×Y S ∼
= BG.
(2) If X → Y is a gerbe, then X → Y is a smooth morphism.
(3) If G → S is an fppf group scheme and X → X is a banded G-gerbe over
an algebraic space X, then there exists an étale cover X ′ → X such that
X ×X X ′ ∼
= X ′ ×S BG.
Proof. For the (⇒) of (1), by Exercise 6.4.8, there is an fppf morphism S → Y and
a sheaf of groups G on S such that X ×Y S ∼ = BG. Since X ×Y S is an algebraic
stack, its diagonal is representable, and thus G is an algebraic space. Conversely, if
X ×Y S ∼ = BG, then Axiom (1) of a gerbe holds: if a ∈ (Sch/Y) is an object over a
scheme T , then since ST := S ×Y T → T is an fppf covering and X ×Y ST ∼ = BGST ,
there is an object of X over ST . Similarly for Axiom (2), if x1 , x2 ∈ X are objects
over y ∈ Y(T ), then pullbacks of x1 and x2 become isomorphic under the fppf
covering ST → T . Part (2) follows from (1) since BG → S is a smooth morphism
(even if G → S is not smooth!): this follows from the considering the composition
S → BG → S and using that smoothness is an fppf local property on the source
(Proposition 2.1.27). For part (3), X → X has a section after base changing by the
smooth and surjective morphism X → X. Choosing a smooth presentation U → X ,
then the composition U → X → X is smooth and the base change X ×X U → U
has a section. The statement now follows from the fact that smooth morphisms
étale locally have sections (Corollary A.3.5).

324
Exercise 6.4.17. Show that a morphism X → Y of algebraic stacks is a gerbe if
and only if X → Y and X → X ×Y X are fppf.
In fact, any algebraic stack with flat inertia is a gerbe. The following result will
be applied to show that the moduli stack of simple bundles is a Gm -gerbe over an
algebraic space (Proposition 8.2.20).
Proposition 6.4.18. Let X be an algebraic stack such that IX → X is fppf. Let X
be the sheaf on Schfppf defined by the sheafification of the functor assigning a scheme
S to the set of isomorphism classes X (S)/∼ of objects. Then X is an algebraic
space and X → X is a gerbe.
Proof. To show that X is an algebraic space, it suffices to show that X → X is a
smooth representable morphism. Indeed, a smooth presentation U → X induces
a surjective, smooth, and representable morphism U → X → X, and it follows
from Corollary 4.5.12 (or Theorem 6.3.1) that X is an algebraic space. As gerbes
are smooth morphisms, it suffices to show that for every morphism S → X from
a scheme, the fiber product X ×X S → S is a gerbe. By construction, there is an
fppf cover S ′ → S and a morphism a′ : S ′ → X lifting the composition S ′ → S → X.
Since the property of being a gerbe is fppf local on the target, after replacing S with
S ′ , we may assume that S → X lifts to a map a : S → X . In this case, there is an
isomorphism
Ψ : X ×X S → BAutS (a).
The map Ψ is defined by sending an object (b, f ) ∈ X ×X S(T ) to the principal
AutS (a)-bundle IsomT (f ∗ a, b). The reader is encouraged to check that Ψ is an
isomorphism. Since AutS (a) → S is an fppf group scheme, BAutS (a) → S is a
gerbe.
Alternatively, the algebraic space X can be constructed directly. Choosing a
smooth presentation U → X inducing a smooth groupoid R ⇒ U , the stabilizer
groupoid scheme SU = R ×U ×U U = IX ×X U is fppf over U . There is an fppf
equivalence relation SU ×U R ⇒ R where one arrow is given by composition and the
other is projection. By applying the algebraicity of quotients by fppf equivalence
relations (Corollary 6.3.6) twice, the fppf quotient R′ := R/(SU ×U R) is an algebraic
space, which induces an fppf equivalence relation R′ ⇒ U , and X is isomorphic
to the fppf quotient U/R′ . See also [LMB00, Cor. 10.8] and [SP, Tags 06QD and
06QJ].
Exercise 6.4.19. Show that every gerbe X → X, where X is an algebraic stack
and X is an algebraic space, is a coarse moduli space (Definition 4.2.1).
Exercise 6.4.20. If X is a reduced noetherian algebraic stack, show that there is a
dense open substack U which is a gerbe.
Hint: Use Generic Flatness (3.3.32).
Exercise 6.4.21. Show that there is a non-trivial isomorphism

α : B(Z/2) × (A1 ∖ 0) → B(Z/2) × (A1 ∖ 0)

of trivial banded Z/2-gerbes over A1 which glues to a non-trivial banded Z/2-gerbe

over P1 .
Exercise 6.4.22. Let 1 → K → G → Q → 1 be a short exact sequence of affine
algebraic groups over k such that K is commutative. Show that BG → BQ is a
banded K-gerbe which is trivial if and only if the sequence splits.

325
Exercise 6.4.23. Assume that char(k) ̸= 2, 3. Recall from Exercise 3.1.19(c) that
the moduli stack of stable elliptic curves has a quotient description M1,1 = P(4, 6) :=
[(A2 ∖ 0)/Gm ] where Gm acts with weights 4 and 6.
(a) Show that the j-line π : M1,1 → A1 is a trivial banded Z/2-gerbe over A1 ∖
{0, 1728}.
Hint: Construct a family of elliptic curves over A1k ∖ {0, 1728} via the Weier-
strass equation

36 1
y 2 z + xyz = x3 − xz 2 − z3,
t − 1728 t − 1728

where t is the coordinate on A1 , where the discriminant ∆ = t2 /(t − 1728)3 .

See [Sil09, Prop. III.1.4(c)].
(b) Consider the map M1,1 = P(4, 6) → P(2, 3) on weighted projective stacks
induced the homomorphism Gm → Gm defined by t 7→ t2 (which we can
view shortly as an example of rigidification). Note that the restriction along
P(2, 3) ∖ {0, 1728, ∞} is the gerbe from (a). Show that M1,1 → P(2, 3) is
non-trivial.
Hint: If it is trivial, show that there are torsion line bundles. This contradicts
Pic(M1,1 ) = Z from Exercise 6.1.22(a).
(c) Show that M1,1 → P(2, 3) ∖ ∞ is also non-trivial.
Hint: If it is trivial, show that M1,1 has three 2-torsion line bundles contra-
dicting that Pic(M1,1 ) = Z/12 from Exercise 6.1.22(b).

Exercise 6.4.24.
(a) Show that every gerbe X over an algebraic space that is étale locally isomorphic
to BZ/2 is, in fact, banded by Z/2.
(b) Give an example of a gerbe over an algebraic space that is étale locally
isomorphic to BGm but not banded by Gm .
Hint: Consider the classifying stack of a form of Gm .

Exercise 6.4.25 (Root gerbes and root stacks revisited). Root gerbes and stacks
were introduced in Examples 3.9.21 and 3.9.22. Since we now know how to construct
quotient stacks by actions of µ r , it follows that Exercise 3.9.23 extends to the case
where r may be√ non-invertible in Γ(S, OS ). Given a scheme X and a line bundle L,
show that X( r L) → X is a banded µ r -gerbe, which is trivial if and only if L has
an rth root.

6.4.4 Residual gerbes revisited

Given an algebraic stack X and x ∈ |X |, recall from Definition 3.5.12 that the
residual gerbe at x (if it exists) is a reduced, locally noetherian algebraic stack Gx
with a monomorphism Gx ,→ X such that |Gx | is a point mapping to x. We have
already shown that the residual gerbe at a finite type point exists and is unique
(Proposition 3.5.16).

Proposition 6.4.26 (Existence of Residual Gerbes II). If X is a noetherian algebraic

stack and x ∈ |X | is a point, then the residual gerbe Gx exists and is a gerbe over a
field κ(x), called the residue field of x.

326
Proof. After replacing X with the closure {x}, we may assume that X is reduced and
that x ∈ |X | is dense. By Exercise 6.4.20 and Theorem 4.5.1, after replacing X with
an open substack, we may assume that X is a gerbe over a scheme X. Letting y ∈ X
be the image of x, then Gx := X ×X κ(y) is a gerbe over κ(x) := κ(y). The inclusion
Gx → X is a monomorphism with image x and Gx is reduced noetherian substack
with |Gx | a singleton. Therefore, Gx is a residual gerbe. The uniqueness follows from
the same argument as in Proposition 3.5.16 See also [LMB00, Thm. 11.3], [Ryd11,
Thm. B.2], or [SP, Tag 06UH].

If X is a quasi-separated algebraic stack of finite type over a field k and x ∈ X (k),

then Gx = BGx (Proposition 3.5.16). More generally, we have:

Exercise 6.4.27. Let X be a noetherian algebraic stack and x ∈ |X | be a finite

type point.
(1) For any representative x : Spec k → X of x, there is a cartesian diagram

BGx / Gx   /X

□
 
Spec k / Spec κ(x).

(2) If the stabilizer of x is smooth, show that there is a finite separable extension
κ(x) → k and a representative of x over k.

Exercise 6.4.28. Let X be a separated Deligne–Mumford stack with coarse moduli

space π : X → X. Show that the residual gerbe of x ∈ |X | is reduction of the fiber
Xπ(x) .

Exercise 6.4.29. Let C ⊆ P2k be a non-split quadric over a field k, and let k → k′
be a quadratic extension such that C ×k k′ ∼ = P1k′ . Let D ⊆ C be a divisor of
degree 6 and let X → Pk′ be the double cover ramified over D ×k k′ . Show that the
1

residual gerbe of [X] ∈ M2 is non-trivial and has residue field k. (In fact, every
Deligne–Mumford gerbe over a field can be realized as the residual gerbe of some
curve [C] ∈ Mg for some g; see [BL24].)

6.4.5 Cohomological characterization

The following exercises provide cohomological characterizations of torsors and gerbes
for an abelian sheaf G on the small fppf site Sfppf of a scheme S. If G → S is a smooth,
commutative, and quasi-projective group scheme, then Hp (Sét , G) = Hp (Sfppf , G)
[Mil80, Thm. 3.9]. For Gm , there are identifications Pic(S) = H1 (SZar , OS∗ ) =
H1 (Sét , Gm ) = H1 (Sfppf , Gm ) (Hilbert’s Theorem 90, [Mil80, Prop. 4.9]). We also
note that if S is a smooth scheme over C and G is a finite abelian group, then
the classical complex cohomology Hp (S(C), G) agrees with the étale cohomology
Hp (Sét , G) of the constant sheaf [Mil80, Thm. 3.12].
For an extra challenge, the reader is encouraged to prove the statements below
more generally for abelian sheaves over any site. The reader may consult [Gir71]
and [Ols16, §12] for detailed proofs.

Exercise 6.4.30 (Torsors). Let S be a scheme.

(a) If G is an abelian sheaf on Sfppf , show that H1 (Sfppf , G) is in bijective corre-
spondence with isomorphism classes of G-torsors.

327
 ∗
Hint: Imitate the proof using Čech cohomology that H1 (X, OX ) = Pic(X) for
a scheme X.
(b) Let 0 → G′ → G → G′′ → 0 be an exact sequence of abelian sheaves on Sfppf ,
and let
δ
0 → H0 (Sfppf , G′ ) → H0 (Sfppf , G) → H0 (Sfppf , G′′ ) −
→
α β
→ H1 (Sfppf , G′ ) − → H1 (Sfppf , G′′ ) → · · ·
→ H1 (Sfppf , G) −

be the corresponding long exact sequence. Show that under the bijection in
(a), the boundary map δ assigns a section S → G′′ to the G′ -torsor defined
by the fiber product G ×G′′ S. Show also that α assigns a G′ -torsor P ′ to the
′
quotient P ′ ×G G := (P ′ × G)/G′ while β assigns a G-torsor P to P ×G G′′ .
Exercise 6.4.31 (Gerbes). Let S be a scheme.
(a) If G is an abelian sheaf on Sfppf , show that H2 (Sfppf , G) is in bijective corre-
spondence with isomorphism classes of G-banded gerbes.
d0 d1 d2
Hint: Let 0 → G → I 0 −→ I 1 −→ I 2 −→ · · · be an injective resolution. For a
cohomology class α ∈ H2 (Sfppf , G), define a stack Gα over Sfppf as follows. Choose
τ ∈ Γ(Sfppf , I 2 ) with d2 (τ ) = 0 representing α. Define Gα as the category of pairs
(T, σ) consisting of an object T ∈ Sfppf and a section σ ∈ Γ(T, I 1 ) with d1 (σ) = τ |T .
A morphism (T ′ , σ ′ ) → (T, σ) is the data of a map f : T ′ → T and an element
ρ ∈ Γ(T ′ , I 0 ) such that d0 (ρ) = σ ′ − f ∗ (σ). Show that Gα is a G-banded gerbe and
that the assignment α 7→ Gα gives the stated bijection. See also [dJ03, §2.5].
Let 0 → G′ → G → G′′ → 0 be an exact sequence of abelian sheaves on Sfppf , and
let
δ α β
· · · → H1 (Sfppf , G′′ ) −
→ H2 (Sfppf , G′ ) − → H2 (Sfppf , G′′ ) → · · ·
→ H2 (Sfppf , G) −

be the corresponding long exact sequence.

(b) Show that under the bijection in (a), the boundary map δ assigns a G′′ -
torsor P ′′ → S to the gerbe of trivializations GP ′′ , where an object over an
S-scheme T is a pair (P, α) consisting of a G-torsor P → T and a trivialization
α : P ×G G′′ ∼
= P ′′ ×S T of G′′ -torsors.
(c) Suppose that G′ , G, and G′′ are representable by commutative and affine
algebraic groups over a field k. Show that if P ′′ → S is a G′′ -torsor, then GP ′′
is identified with both the quotient stack [P ′′ /G] and with the fiber product
of the map S → BG′′ classifying the G′′ -torsor P ′′ and BG → BG′′ .
(d) In the special case that the exact sequence is 1 → µ r → Gm → Gm → 1 and
P ′′ corresponds
√ to a line bundle L′′ , show that GP ′′ is isomorphic to the root
stack X( L).
r

Exercise 6.4.32 (Group structure). Show that the group laws of H1 (Sfppf , G) and
H2 (Sfppf , G) can be described geometrically as follows:
(a) The product of two G-torsors P1 and P2 is the contracted product P1 ∧G P2
defined as the sheaf quotient (P1 × P2 )/G where h · (p1 , p2 ) = (h−1 p1 , hp2 )
with the G-action specified by g · (p1 , p2 ) = (gp1 , p2 ) = (p1 , gp2 ). The inverse
of a G-torsor P is the sheaf P with the inverted G-action: g · p = g −1 p.
(b) The product of two banded G-gerbes (X1 , ψ1,x ) and (X2 , ψ2,x ) is the contracted
product X1 ∧G P2 , which is defined as the rigidification (X1 ×X2 )( G (see Propo-
sition 6.4.18) of the product X1 × X2 along the subgroup (ψ1 , ψ2 ) : G|X1 ×X2 →

328
 IX1 ×X2 defined by the bands ψ1 and ψ2 . The inverse (X , ψx )−1 = (X , ψx−1 )
inverts the band.

Remark 6.4.33 (Banded µn -gerbes over P1 ). Over an algebraically closed field k,

isomorphism classes of banded µn -gerbes over P1 are in bijection with Z/nZ. To see
n
this, observe that the exact sequence 1 → µn → Gm − → Gm → 1 induces an exact
sequence on cohomology
n δ
H1 (Pét
1
→ H1 (Pét
, Gm ) − 1
→ H2 (Pét
, Gm ) − 1
, µn ) → H2 (Pét
1
, Gm ).

Since H1 (Pét
1 1
, Gm ) = Pic(Pét ) = Z, we can use the fact that H2 (Pét
1
, Gm ) = 0 to
conclude that H (Pét , µn ) = Z/nZ. The image ofpa line bundle O(d) under the
2 1

boundary map δ is equivalent to the root stack P1 ( n O(d)), and this gerbe is trivial
if and only if n divides d.

Exercise 6.4.34 (Forms of group schemes). Let G be an algebraic group over a

field k. We say that a group scheme H → Spec k is a form of G if there is an
isomorphism Gk ∼= Hk . We call G the trivial form of G.
(a) Show the algebraic group H = Spec R[x, y]/(x2 + y 2 − 1) over R, with the
group structure induced from the embedding H ⊆ SL2 given by
 
x y
(x, y) 7→ ,
−y x

is a non-trivial form of Gm,R .

(b) Assume that char(k) ̸= 2. Recall the orthogonal groups O(q) defined in
Exercise B.1.58 for a non-degenerate quadratic form q on an n-dimensional
vector space V . Show that every O(q) is a form of the subgroup On ⊆ GLn of
orthogonal matrices.
(c) If G is smooth and commutative, show that forms of G are classified by
H1 ((Sch/k)ét , Aut(G)).

Exercise 6.4.35 (Gm -gerbes and twisted sheaves). Let X → X be a Gm -gerbe over
an algebraic space X. We say that a coherent sheaf F on X is 1-twisted if for every
field-valued point Spec k → X , the Gm -representation corresponding to the pullback
of E under BGm = BGx → X decomposes as a direct sum of one dimensional
representations of weight one. Show that the Gm -gerbe X → X is trivial if and only
if there exists a 1-twisted line bundle on X .

Exercise 6.4.36 (Azumaya algebras). An Azumaya algebra of rank r2 over a

noetherian scheme X is a (possibly non-commutative) associative OX -algebra A
which is coherent as an OX -module and such that there is an étale covering X ′ → X
where A ⊗OX OX ′ is isomorphic to the matrix algebra Matr×r (OX ). We say that A
is trivial if it is isomorphic to Matr×r (OX ). By Exercise B.1.68, Azumaya algebras
are in bijection with principal PGLr -bundles and with Brauer–Severi schemes of
relative dimension r. Let A be an Azumaya algebra over a noetherian scheme X of
rank r2 .
(a) Define the gerbe of trivializations of A as the stack GA over (Sch/X)ét where
an object over a X-scheme T is a pair (E, α) consisting of a vector bundle E
∼
on T of rank r and a trivialization α : EndOX (E) → A ⊗OX OT . Morphisms in
GA (T ) are isomorphisms of vector bundles compatible with the trivializations.
Show that GA → X is a banded Gm -gerbe.

329
 (b) Let PA be the principal PGLr -bundle corresponding to A. Identify GA with
the gerbe of trivializations GPA defined in Exercise 6.4.31(b) with respect to
the PGLr -torsor PA and the surjection GLr → PGLr .
(c) Use the quarternions to construct a non-trivial Gm -gerbe over Spec R.

Remark 6.4.37 (Brauer groups). Two Azumaya algebras A and A′ on a noetherian

scheme X are similar if there exists vector bundles E and E ′ on X such that
A ⊗OX EndOX (E) ∼ = A ⊗OX EndOX (A′ ). This defines an equivalence relation, and
the Brauer group of X is the set Br(X) of Azumaya algebras up to similarity. The set
Br(X) becomes a group under the operators [A] · [A′ ] = [A ⊗ A′ ] and [A]−1 = [Aop ]
(where Aop is the opposite algebra with same elements and addition as A but with
op
multiplication reversed: a ·A b = b ·A a).
The exact sequence 1 → Gm,X → GLr,X → PGLr,X → 1 of sheaves on Xét
induces a boundary map

δ : H1 (Xét , PGLr ) → H2 (Xét , Gm ), PA 7→ GA ,

where PA is the principal PGLr -bundle corresponding to A and GA is the gerbe of

trivializations of A. It is not hard to check that [GA ] is an r-torsion element and
that A is trivial if and only if GA is trivial. It follows that there is an injective map

Br(X) ,→ Br′ (X) := H2 (Xét , Gm )tors , A 7→ GA ,

into the cohomological Brauer group Br′ (X). See [Gro68] and [Mil80, §IV.2] for
additional background. Grothendieck asked whether Br(X) ,→ Br′ (X) is surjective.
This is known in some cases. The strongest result is due to Gabber: Br(X) = Br′ (X)
if X admits an ample line bundle (see [dJ03]). It is however open in general, even
for smooth separated schemes over a field.

Exercise 6.4.38. Let X be a noetherian scheme and X → X be a banded Gm -gerbe

corresponding to a cohomology class [X ] ∈ H2 (Xét , Gm ).
(a) Show that the following are equivalent:
(i) There exists an Azumaya algebra A on X such that X ∼
= GA , i.e., [X ] is
in the image of Br′ (X) → Br(X),
(ii) X is a global quotient stack, and
(iii) there exists a 1-twisted vector bundle E on X (see Exercise 6.4.35).
(b) Let X be a normal separated surface over C such that H2 (X, Gm ) contains a
non-torsion element α; for an example, see [Gro68, II.1.11.b]. Conclude that
the banded Gm -gerbe corresponding to α is not a global quotient stack.
(c) Let Y = Spec C[x, y, z]/(xy − z 2 ). Show that there is a non-trivial involution α
of (Y ∖0)×B(Z/2) such that the stack X , obtained by gluing the trivial banded
S over Y along α, is a banded Z/2-gerbe over the non-separated
Z/2-gerbes
union Y Y ∖0 Y and that X is not a global quotient stack.
See also [EHKV01].

6.4.6 Rigidification
Given an algebraic stack X , the inertia stack IX can be viewed as a group algebraic
space over the big étale site (Sch/X )ét of X . As a group functor, IX assigns an
object a ∈ X (S) to the group AutS (a), and a morphism α : a′ → a over S ′ → S to

330
the natural pullback map α∗ : AutS (a) → AutS ′ (a′ ) (see (3.2.11)). Given a : S → X ,
there is a canonical isomorphism IX ×X S ∼ = AutS (a) of group algebraic spaces over
S.
Suppose that H ⊆ IX is a closed subgroup space over X such that H → X
is fppf. This is equivalent to requiring that for every a ∈ X (S), there is a closed
fppf subgroup scheme Ha ⊆ AutS (a) over S such that if a′ → a is a morphism
over S ′ → S, the canonical isomorphism AutS ′ (a′ ) ∼
= AutS (a) ×S S ′ restricts to an
∼ ∼
isomorphism Ha′ = Ha ×S S . If α : a → a is an automorphism over the identity,
′

then the canonical isomorphism α∗ : AutS (a) → AutS (a) is conjugation by α. In

particular, Ha ⊆ AutS (a) is a normal group scheme.
Definition 6.4.39 (Rigidification). Let X be an algebraic stack and H ⊆ IX be
an fppf closed subgroup scheme over X . The rigidification X( H is defined as the
stackification in Schfppf of the prestack with the same objects as X and where the
set of morphisms between a ∈ X (S) and b ∈ X (T ) over f : S → T is defined as
Mor(a, b) = MorX (S) (a, f ∗ b)/H(T ).
One can think of the subgroup H as giving an action of BH on X and the
rigidification X( H as the quotient X /BH. Frequently in applications, the closed
subgroup H ⊆ IX is obtained by the pullback of a fppf group scheme H → S, i.e.,
H = H ×S X . In this case, it is customary to write X( H := X( H.
Example 6.4.40. If IX → X is fppf, then we can take H = IX and the rigidification
X( IX is the algebraic space X constructed in Proposition 6.4.18.
Proposition 6.4.41. Let X be an algebraic stack and H ⊆ IX be an fppf closed
subgroup scheme over X . The rigidification X( H is an algebraic stack such that
(1) the natural morphism π : X → X( H is a gerbe;
(2) for every object a ∈ X (S), the natural map AutS (a) → AutS (π(a)) is surjective
with kernel H(S);
(3) a morphism f : X → Y factors uniquely through X( H if and only if for every
object a ∈ X (S), the composition H(S) ⊆ ker(AutX (S) (a) → AutY(S) (f (a)));
and
(4) if H is a commutative group scheme, then H descends to an fppf group scheme
H → X such that X → X is banded H-gerbe. If in addition H = H ×S X is
the pullback of a commutative fppf group scheme H → S, then X → X is a
banded H-gerbe.
Proof. To show that X is algebraic, it suffices to show that π : X → X ( H is a
smooth representable morphism. Given g : S → X( H, the definition of X( H as the
stackification implies that there is an fppf cover S ′ → S such that S ′ → S → X( H
lifts to a map a′ : S ′ → X . By replacing S with S ′ , we may assume that g : S → X( H
lifts to a morphism a : S → X . As in the proof of Proposition 6.4.18, one shows that
there is an isomorphism
Ψ : X ×X( H S → BHa .
The details are left to the reader. See also [ACV03, Thm. 5.1.5], [AGV08, §C],
[Rom05, §5], and [AOV08, §A].
Example 6.4.42. Every family of stable elliptic curve has a canonical hyperelliptic
involution. This yields a subgroup µ2 := µ2,M1,1 of the inertia stack IM1,1 of the
stack of stable elliptic curves M1,1 ∼ = P(4, 6) over a field (assumed as usual to
have char(k) ̸= 2, 3). The rigidification M1,1 → M1,1( µ2 is identified with the
non-trivial µ2 -gerbe P(4, 6) → P(2, 3) discussed in Exercise 6.4.23.

331
Example 6.4.43 (Ridigification of Bun(C)). Moduli stacks of sheaves provide
interesting examples of rigidification since there is a canonical scaling Gm -action
on sheaves. For instance, consider the moduli stack Bun r,d (C) of vector bundles
over a fixed smooth, connected, and projective curve C over an algebraically closed
field k. For any vector bundle E on C × S where S is a k-scheme, there is a
canonical closed immersion iE : Gm,S → Aut(E) of group schemes over S. Thus,
Gm := Gm,Bun r,d ⊆ IBun r,d is a closed fppf group scheme of the inertia stack, and
we can construct the rigidification

Bun(C)( Gm .

Over the open substack Bun simple (C) of simple bundles (i.e., vector bundles E with
Aut(E) = k∗ ), the rigidification Bun simple (C)( Gm is an algebraic space so that
Bun simple (C) is Gm -gerbe over its coarse moduli space.

Exercise 6.4.44.
(a) If H is a commutative fppf group scheme over S, show that BH( H ∼ = S.
More generally, show that if X → X is a banded H-gerbe, then X ∼
= X( H.
(b) Let G → S be an fppf group scheme acting on a S-scheme U . Suppose that
H ⊆ G is a central commutative fppf subgroup scheme acting trivially on U .
Show that [U/G]( H ∼ = [U/(G/H)].

Exercise 6.4.45. Let X be a smooth, integral, and separated Deligne–Mumford

stack over a scheme S. Let Spec K → X be a representative of the generic point.
Show that the closure H ⊆ IX of generic fiber IX ×X K of the inertia is a closed
étale subgroup scheme and that the rigidification X( H is a smooth, integral, and
separated Deligne–Mumford stack over S with generically trivial inertia.

Exercise 6.4.46. Let X be an algebraic stack over a scheme S, H → S be an

fppf group scheme, and H ×S X ⊆ IX a closed subgroup scheme. Show that the
rigidification X( H can be given the moduli interpretation where an object over a
scheme S is a pair (G, f ), where G → S is a banded H-gerbe and f : G → X is an
H-equivariant morphism.

Exercise 6.4.47. For g ≥ 2, let Hg ⊆ Mg be the closed substack classifying

hyperelliptic curves. Show that the rigidification Hg → Y along the hyperelliptic
involution is a non-trivial banded Z/2-gerbe.

6.4.7 Picard stacks and spaces

If X → S is a proper flat morphism of noetherian schemes, define the Picard stack

Pic X/S

as the stack over (Sch/S)ét whose objects over an S-scheme T are line bundles on
XT = X ×S T and whose morphisms are isomorphisms of line bundles. This is an
open substack of Coh(X/S), which is an algebraic stack locally of finite type over S;
this follows from an explicit presentation using the Quot scheme (Exercise 3.1.23) in
the case that X → S is strongly projective and as a consequence of Artin’s Axioms
(Theorem C.7.7) in general. Therefore, Pic X/S is also algebraic and locally of finite
type over S.
On the other hand, there are several candidates for Picard functors:

332
 (1) The naive Picard functor (or absolute Picard functor ) is

Picnaive
X/S : Sch/S → Gps, (T → S) 7→ Pic(XT ).

(2) The Picard functor

PicX/S : Sch/S → Gps,
is the fppf sheafification of Picnaive
X/S .
(3) The relative Picard functor is

Picrel
X/S : Sch/S → Gps, (T → S) 7→ Pic(XT )/ Pic(T ),

where an object over an S-scheme T is a line bundle L on XT , and two line

bundles L and L′ on XT are identified if there exists M ∈ Pic(T ) such that
L∼= L′ ⊗ fT∗ M .
(4) We can define the rigidification of the Picard stack

Pic X/S( Gm
∼
under the hypothesis that OT → fT,∗ OXT is an isomorphism for any map
T → S. This hypothesis implies that for a line bundle L on XT , there is a
canonical isomorphism
∼
Gm (T ) = Γ(T, OT )∗ → Γ(XT , OXT )∗ = Aut(L),

which further implies that the inertia stack IPic X/S is isomorphic to fppf group
scheme Gm := Gm,Pic X/S over Pic X/S .
While the Picard functor PicX/S and the rigidification Pic X/S( Gm are sheaves
in the big fppf topology by definition, it may seem surprising that Picrel
X/S is also a
sheaf under relatively mild hypotheses.
Proposition 6.4.48. Let f : X → S be a proper flat morphism of noetherian
∼
schemes such that OS → f∗ OX is an isomorphism and this holds after base change,
∼
i.e., for every map g : T → S, the map OT → fT,∗ OXT is an isomorphism. Then
(1) PicX/S is representable by an algebraic space locally of finite type over S,
(2) the map Pic X/S → PicX/S from the Picard stack is a banded Gm -gerbe and
PicX/S ∼= Pic X/S( Gm , and
(3) if in addition there is a section s : S → X, then Pic ∼
= Picrel .
X/S X/S
If in addition the geometric fibers of f are integral, then PicX/S is separated over S.
Remark 6.4.49. If f : X → S is a proper flat proper morphism with geometrically
∼
connected and reduced fibers, then OS → f∗ OX is an isomorphism (and remains so
after base chagne) by Lemma A.6.12. In [FGAV, Thm. 3.1], Grothendieck proved
that PicX/S is a scheme in the case that X → S is projective. See [Mum66a, §20-21],
[AK80], and [Kle05, §9.4] for alternative expositions and various generalizations.
The representability as an algebraic space above was first established by Artin
[Art69b, Thm. 7.3], and this holds with the slightly weaker hypothesis that f is
cohomologically flat in dimension 0, i.e., the formation of f∗ OX commutes with base
change.
∼
Proof. As pointed out above, the condition that OS → f∗ OX holds after base change
implies that the inertia stack of Pic X/S is isomorphic to the fppf group scheme Gm :=

333
Gm,Pic X/S . Therefore Proposition 6.4.18 implies that the rigidification Pic X/S( Gm
is an algebraic space locally of finite type over S. Moreover, Pic X/S( Gm is identified
with the Picard functor PicX/S by the definition of rigidification. This gives both
(1) and (2). When the fibers are geometrically integral, the separatedness of PicX/S
over S follows from Proposition A.6.17 and Remark A.6.19.
To identify Picrel
X/S with PicX/S , it suffices to prove that PicX/S is a sheaf in the
rel

big fppf topology. To this end, it will be convenient to identify Picrel X/S with the
prestack P s−rig
, called the rigidification by the section s, whose fiber category over
an S-scheme T is
∼
P s−rig (T ) = (L, α) | L ∈ Pic(XT ) and α : OT → s∗T L ,


where a morphism (L, α) ∼ (L′ , α′ ) is the data of an isomorphism β : L → L′ of

line bundles such that α′ = s∗T β ◦ α. The advantage of considering P s−rig is that it
is a straightforward application of fppf descent of quasi-coherent sheaves to check
that P s−rig is a stack over the big fppf topology (Sch/S)fppf . It therefore suffices to
verify that the natural map

P s−rig → Picrel
X/S , (L, α) 7→ L (6.4.50)

is an equivalence.
We first check that (6.4.50) is faithful and in particular that P s−rig is equivalent
∼
to a functor. We must show that if β : L → L is an automorphism with s∗T β =
ids∗T L , then β = idL . Since OT → fT,∗ OXT is an isomorphism, the pullback map
fT∗ : H0 (T, OT ) → H0 (XT , OXT ) is an isomorphism; as sT is a section, s∗T is the
inverse of fT∗ . The composition

s∗
s∗T : HomOXT (L, L) ∼
= H0 (XT , OXT ) −−→ H0 (T, OT ) ∼
T
= HomOT (s∗T L, s∗T L)

is an isomorphism of groups, and thus β = idL . To see that (6.4.50) is full (i.e.,
the induced map on the functors of isomorphism classes is injective), let (L, α) ∈
∼
P s−rig (T ) be such that there is a line bundle M on T and an isomorphism β : L →
fT∗ M . The isomorphism

f∗ α f ∗ s∗ β
γ : OXT = fT∗ OT −−
T
→ fT∗ s∗T L −−−−→ fT∗ s∗T fT∗ M = fT∗ M
T T

satisfies s∗T γ = s∗T β ◦ α and shows that (L, α) is isomorphic to (OXT , id) ∈ P s−rig (T ).
Finally, to see that the functor (6.4.50) is essentially surjective, let L ∈ Pic(XT ) and
define
L′ = L ⊗ (fT∗ s∗T L)∨ .
∗ ′ ∼ ∗ ∨ ∼
The images of L and L′ are equal in Picrel
X/S (T ), and sT L = sT L⊗(sT fT sT L) = OT
∗ ∗ ∗
∼ ∗ ′
defines an isomorphism α : OT → sT L such that L ∈ PicX/S (T ) is the image of
′ rel

(L′ , α′ ) ∈ P s−rig (T ).

Over an algebraically closed field k, it is remarkably easy to verify that the

Picard functor PicX := PicX/k is a scheme.

Theorem 6.4.51. Let X be a proper integral scheme over an algebraically closed

field k.
(1) PicX is a group scheme locally of finite type over k, and in particular the
connected component Pic0X of the identity is quasi-projective.

334
 (2) PicX ∼
= Picrel
X and Pic X → PicX is a banded Gm -gerbe,
(3) If X is smooth, Pic0X is projective.
(4) If char(k) = 0, then PicX is smooth of dimension h0 (X, OX ). In particular,
Pic0X is an abelian variety.

Proof. As k is algebraically closed and X is integral, the structure map f : X →

∼
Spec k has a section and OT → fT,∗ OXT is an isomorphism for any k-scheme T .
Proposition 6.4.48 implies that PicX is an algebraic space locally of finite type
over k, and that (2) holds. The Picard stack Pic X is quasi-separated (it even has
affine diagonal) and it follows that PicX is also quasi-separated. This is enough
to show that PicX is a separated scheme and that Pic0X is quasi-projective. This
is a direct consequence of Theorem 4.5.28, but it is worth recalling the argument:
PicX has a dense open subspace which is a scheme (Theorem 4.5.1), the group
structure PicX allows us to translate this open to show that PicX is an algebraic
group, which is automatically separated with quasi-projective connected components
(Proposition B.1.16). This gives (1).
For (3), it suffices to show that Pic0X is proper. As we already know it is separated,
we only need to verify the existence part of the valuative criterion for properness: let
R be a DVR over k with fraction field K and L be a line bundle on XK . As XR is
regular, the line bundle L extends to a line bundle L e on XR (for instance, if L = O(D)
for a divisor D ⊆ XK , then take L = O(D)). The smoothness in (4) follows from
e
the fact that algebraic groups are smooth in characteristic 0 (Proposition B.1.16).
If L is a line bundle on X, then the Zariski tangent spaces of the Picard stack
and Picard scheme agree, and deformation theory (Proposition C.1.18) implies that
TPicX ,L ∼
= H1 (X, OX ).

Remark 6.4.52. As a consequence of the representability of PicX ∼ X , there is a

= Picrel
universal family (or Poincaré family) P on X × PicX that satisfies the following: for
any k-scheme T and any line bundle L on XT , there is a unique morphism T → PicX
such that
L∼ = P|X×T ⊗ p∗2 M
for some line bundle M on T . The connected component of the identity Pic0X has
the functorial description of parameterizing line bundles L algebraically equivalent
to OX (i.e., there is a connected k-scheme T with points t0 , t1 ∈ T (k) and a family
of line bundles L on XT such that Lt0 ∼ = L and Lt1 ∼= OX ). When X is a smooth
curve, Pic0X parameterizes degree 0 line bundles.
Remark 6.4.53. In characteristic p, Igusa showed that Pic(X) may fail to be reduced
[Igu55]. We also note that when X is not normal (e.g., a nodal or cuspidal curve),
then Pic0 (X) is not proper. Altman and Kleiman [AK80] provide a compactification
of Pic0 (X) by classifying rank 1 torsion free sheaves. Picard functors and schemes
have a fascinating history as they were one of the first examples of moduli spaces
constructed in algebraic geometry. See Kleiman’s article [Kle05] for a beautiful
account of the history and a broader discussion of the properties of Picard schemes.

6.5 Affine GIT and good moduli spaces

Das ist nicht Mathematik, das ist Theologie.

Paul Gordan

335
 The main theorem of affine GIT (Corollary 6.5.8) asserts that if G is a linearly
reductive group acting on an affine scheme Spec A, then the map Spec A → Spec AG
is a quotient where closed points of Spec AG are identified with closed G-orbits in
Spec A. Recall that an affine algebraic group G over a field k is linearly reductive if
the functor Rep(G) → Vectk , taking a G-representation V to its G-invariants V G ,
is exact. Examples include:

• finite discrete groups G whose order is not divisible by char(k) (Maschke’s

Theorem (B.1.38));
• tori Gnm and diagonalizable group schemes (Proposition B.1.15); and
• reductive groups (e.g., GLn , SLn and PGLn ) in char(k) = 0 (Theorem B.1.43).

See §B.1.6 for further equivalences, properties, and a discussion of linearly reductive
groups.
Given an action of G on an affine k-scheme Spec A, the inclusion AG ,→ A
induces a commutative diagram

Spec A
π
e

 &
[Spec A/G]
π / Spec AG .

Let us observe that π : [Spec A/G] → Spec AG enjoys the following two properties:
(1) Γ([Spec A/G], O[Spec A/G] ) = AG ; this follows from the definition of global
sections.
(2) The functor π∗ : QCoh([Spec A/G]) → QCoh(Spec AG ) is exact. This holds
because functor π∗ takes a quasi-coherent OX -module M f, corresponding to an
A-module M with a G-action, to M g G (Exercise 6.1.11) and is therefore exact

by the defining property of linear reductivity.

These are the defining properties of a good moduli space (Definition 6.5.1), and it
turn out most of the arguments showing for affine GIT quotients extend easily to
good moduli spaces. We discuss the projective case of GIT in Chapter 7.

6.5.1 Good moduli spaces

The definition of a good moduli space is inspired by the properties of GIT quotients
and specifically by the properties
L of the morphisms π : [Spec A/G] → Spec AG and
π : [X /G] → X //G := Proj d≥0 Γ(X, OX (d)) , where G is linearly reductive
ss ss G

and X ⊆ P(V ) is a G-invariant closed subscheme of a projectivized G-representation.

Definition 6.5.1 (Good moduli spaces). A quasi-compact and quasi-separated

morphism π : X → X from an algebraic stack X to an algebraic space X is a good
moduli space if
(1) OX → π∗ OX is an isomorphism, and
(2) π∗ : QCoh(X ) → QCoh(X) is exact.

Example 6.5.2 (Basic example: affine GIT). If G is a linearly reductive group over
a field k acting on an affine k-scheme Spec A, then [Spec A/G] → Spec AG is a good
moduli space. We will prove later (Corollary 6.7.4) that every good moduli space is
étale locally on X of this form.

336
Example 6.5.3 (Concrete examples). If Gm acts on An over a field k via t ·
(x1 , . . . , xn ) = (tx1 , . . . , txn ), then [An /Gm ] → Spec k is a good moduli space. While
0 ∈ [An /Gm ](k) is a closed point, every other k-point is not closed and contains
0 in its closure (which is simply the stack-theoretic translation of every Gm -orbit
containing 0 in its closure. Note also that [An /Gm ] ∖ 0 = Pn−1 .
If Gm acts on A2 via t · (x, y) = (tx, t−1 y), then [A2 /Gm ] → Spec k[xy] = A1
is a good moduli space. The fiber over a ̸= 0 ∈ A1 under the good quotient
A2 → A1 is the hyperbola xy = a in A2 and the fiber under the good moduli space
[A2 /Gm ] → A1 is the point Spec k ∼ = [V (xy − a)/Gm ). The fiber over the origin is
the union of the three orbits {(x, 0)|x ̸= 0} ∪ {(0, y)|y ̸= 0} ∪ {0, 0} in A2 . Note that
[A /Gm ] ∖ 0 = A A1 ∖0 A is the non-separated affine line. See Example 3.9.28 for
2 1 1
S
an illustration.
Example 6.5.4 (Conjugation action of n×n matrices). Under the conjugation action
of GLn on Matn,n , the trace and determinant are obvious invariant functions. More
generally, if X = (Xij ) is a generic n × n matrix, then the coefficients a0 , . . . , an−1
of the characteristic polynomial det(X − λI) are invariant functions, and it is a
classical fact that they freely generate the ring of invariants. Therefore, the map
[Matn,n / GLn ] → Matn,n // GLn = Spec k[a0 , . . . , an−1 ],
taking a matrix to its characteristic polynomial, is a good moduli space. Two
matrices are identified if they have the same Jordan block decomposition, and the
unique closed point
 in the a fiber
 corresponds to a diagonal matrix. For instance,
λ 1 λ 0
the matrix has in its GL2 -orbit closure, and these matrices are
0 λ 0 λ
identified in Mat 2, 2// GL2 .
Example 6.5.5 (Tame coarse moduli spaces). If X is a separated Deligne–Mumford
stack of finite type over a noetherian scheme S, then the Keel–Mori Theorem (4.4.6)
implies that there exists a coarse moduli space π : X → X. We say that the coarse
moduli space X → X is tame if every automorphism group has order prime to the
characteristic, i.e., invertible in Γ(S, OS ). A tame coarse moduli space is a good
moduli space (Lemma 4.4.22). If X has quasi-finite stabilizers, then in fact every
good moduli space π : X → X is a tame coarse moduli space with π separated; see
Proposition 6.5.33.
The goal of this section is to establish the following theorem.
Theorem 6.5.6. If π : X → X is a good moduli space with X quasi-separated, then
(1) π is surjective and universally closed;
(2) For closed substacks Z1 , Z2 ⊆ X , im(Z1 ∩ Z2 ) = im(Z1 ) ∩ im(Z2 ). For
geometric points x1 , x2 ∈ X (k), π(x1 ) = π(x2 ) ∈ X(k) if and only if {x1 } ∩
{x2 } =
̸ ∅ in |X ×S k|. In particular, π induces a bijection between closed points
in X and closed points in X;
(3) If X is noetherian, so is X. If X is of finite type over a noetherian algebraic
space S, then X is of finite type over S and π∗ preserves coherence, i.e., for
F ∈ Coh(X ), π∗ F ∈ Coh(X); and
(4) If X is noetherian, then π is universal for maps to algebraic spaces.
Remark 6.5.7. In (2), the images and intersections are taken scheme-theoretically.
Note that since π is closed, the set-theoretic image of a closed substack Z is identified
with the topological space of its scheme-theoretic image im(Z). If I ⊆ OX is the
sheaf of ideals defining Z, the image im(Z) is defined by π∗ I ⊆ π∗ OX = OX .

337
 In the case of affine GIT where we have a good moduli space π : [Spec A/G] →
Spec AG and a good quotient π e : Spec A → Spec AG , this theorem translates to:
Corollary 6.5.8 (Affine GIT). Let G be a linearly reductive group over an alge-
e : U = Spec A → U//G := Spec AG satisfies:
braically closed field k. Then π
e is surjective and for every G-invariant closed subscheme Z ⊆ U , im(Z) ⊆
(1) π
U//G is closed. The same holds for the base change T → U//G by a morphism
from a scheme;
(2) For closed G-invariant closed subschemes Z1 , Z2 ⊆ U , im(Z1 ∩ Z2 ) = im(Z1 ) ∩
im(Z2 ). For x1 , x2 ∈ X(k), π e(x2 ) if and only if Gx1 ∩ Gx2 ̸= ∅. In
e(x1 ) = π
particular, π
e induces a bijection between closed G-orbits of k-points in U and
k-points of U//G.
(3) If A is noetherian, so is AG . If A is finitely generated over k, then AG is also
finitely generated over k and for every finitely generated A-module M with a
G-action, M G is a finitely generated AG -module; and
(4) If A is noetherian, then π
e is universal for G-invariant maps to algebraic
spaces.

Remark 6.5.9. If Z ⊆ U = Spec A is defined by a G-invariant ideal I, then (1)

implies that π(Z) is defined by I G ⊆ AG . If Z1 , Z2 are defined by G-invariant ideals
I1 , I2 ⊆ A, then (2) implies that (I1 + I2 )G = I1G + I2G . In particular, if Z1 and
Z2 are disjoint, then so are im(Z1 ) and im(Z2 ) and we can write 1 = f1 + f2 with
f1 ∈ I1G and f2 ∈ I2G ; the function f1 restricts to 0 on Z1 and 1 on Z2 . We see that
G-invariant functions separate disjoint G-invariant closed subschemes.
Remark 6.5.10 (Hilbert’s 14th problem). Hilbert’s 14th problem asks when the
invariant ring AG is finitely generated. While this is not always true, Hilbert showed
it is true when G is linearly reductive, which is precisely the assertion of (3) above.
Hilbert’s original argument in [Hil90] is so elegant and played such an important
role in the development of modern algebra that we reproduce it here. Our proof of
Theorem 6.5.6(3), while similar in spirit, will not be as explicit.
Let f1 , . . . , fn be k-algebra generators of A and let V ⊆ A be a finite dimensional
G-invariant subspace containing each fi (Proposition B.1.17(1)). Then we have a
surjection Sym∗ V = k[x1 , . . . , xm ] ↠ A of k-algebras with G-actions and we set I =
ker(k[x1 , . . . , xm ] → A). Since G is linearly reductive, AG = (k[x1 , . . . , xm ]/I)G =
k[x1 , . . . , xm ]G /I G and we can assume that A = k[x1 , . . . , xm ] is the polynomial
ring so that AG is a graded k-algebra P whose degree 0 component is k. It therefore
suffices to show that the ideal J+ := d>0 AG d ⊆ A is finitely generated, as it is
G

not hard to see that ideal generators of J+ are also k-algebra generators of AG .
Hilbert first showed that every ideal in A = k[x1 , . . . , xn ] is finitely generated—
this is what is referred to today as Hilbert’s Basis Theorem, which was proved
by Hilbert precisely to make this argument. It follows that J+ A ⊆ A is finitely
generated by homogenous invariants f1 , . . . , fn ∈ AG . We will show that they also
generate J+ as an ideal in AG . For f ∈ AG d , we can write

n
X
f= fi gi (6.5.11)
i=1

with gi ∈ A a homogeneous (not necessarily invariant) function of degree d − deg fi

(with gi = 0 if deg fi > d). Since G is linearly reductive, there is a k-linear map
R : A → AG called the Reynolds operator (see Remark B.1.42), which is the identity

338
on AG , respects the grading, and satisfies R(xy)P = xR(y) for x ∈ AG and y ∈ A.
Applying R to (6.5.11) shows that f = R(f ) = i fi R(gi ) with R(gi ) ∈ AG and
thus f lies in the ideal in AG generated by the fi .2
Hilbert gave a constructive proof of this theorem in [Hil93], where in the course
of the argument, he established the Syzygy Theorem, the Nullstellensatz, a version of
Noether normalization, and a version of the Hilbert–Mumford criterion. These results
spurred the development of modern commutative algebra. We strongly encourage
you to read [Hil90] and [Hil93] (or Hilbert’s translated lecture notes [Hil93]).

Remark 6.5.12 (Reductivity in positive characteristic). In characteristic p, every

smooth linearly reductive group is an extension of a torus by a finite étale group
scheme prime to the characteristic. In particular, GLn is not linearly reductive (see
Example B.1.44). In characteristic p, there are the following variants for an affine
algebraic group G over an algebraically closed field k:
(1) G is reductive if G is smooth and every smooth, connected, unipotent, and
normal subgroup of G is trivial, and
(2) G is geometrically reductive if for every surjection V → W of G-representations
n n
and wn ∈ W G , there exists n > 0 such that wp is in the image of Symp V →
Symp W .
Mumford’s conjecture [GIT, Preface]—now Haboush’s theorem [Hab75]—that these
notions are equivalent when G is smooth (see Theorem B.1.46). See also §B.1.6 for
a further discussion.
Affine GIT was originally developed for linearly reductive groups, but the defining
property of geometrically reductive groups allows one to prove that Spec A →
Spec AG enjoys similar properties to affine GIT quotients by linearly reductive groups.
Namely, Corollary 6.5.8(1)–(4) hold (with the exception that the noetherianness of
A does not necessarily imply the noetherianness of AG ). While the arguments are
not substantially more complicated than the linearly reductive case, we only prove
them for linearly reductive groups for the sake of simplicity. We refer the reader to
[Nag64], [MFK94, App. 1.C], [New78, §3], [Dol03, §3.4], [Spr77, §2] and [DC71, §2].
Likewise, the notion of a good moduli space can be extended to characterize
quotients by geometrically reductive groups: in [Alp14], a quasi-compact and quasi-
separated morphism π : X → X, from an algebraic stack to an algebraic space,
is called an adequate moduli space if (1) OX → π∗ OX is an isomorphism and (2)
for every surjection A → B of quasi-coherent OX -algebras, then every section s of
π∗ (B) over a smooth morphism Spec A → Y has a positive power that lifts to a
section of π∗ (A)). An adequate moduli space satisfies Theorem 6.5.6(1)–(4) (except
again for the noetherian implication). Moreover, if G is geometrically reductive,
then π : [Spec A/G] → Spec AG is an adequate moduli space. In characteristic 0, an
adequate moduli space is necessarily good. Again, it is not substantially harder to
establish the properties of adequate moduli spaces, but we choose to only develop
the theory of good moduli spaces as it is more streamlined and probably best seen
first by a student.

2 Foran alternative argument that AG is noetherian, linear reductivity can be used to show
that JA ∩ AG = J for every ideal J ⊆ AG (see Lemma 6.5.24(5)). If J1 ⊆ J2 ⊆ · · · ⊆ AG is an
ascending chain of ideals, then the ascending chain J1 A ⊆ J2 A ⊆ · · · ⊆ A terminates, which implies
that the original sequence J1 = J1 A ∩ AG ⊆ J2 = J2 A ∩ AG ⊆ · · · ⊆ AG also terminates.

339
6.5.2 Cohomologically affine morphisms
The exactness condition on the pushforward π∗ in the definition of a good moduli
space (Definition 6.5.1(2)) is a non-representable analog of affineness.
Definition 6.5.13 (Cohomologically affine). A quasi-compact and quasi-separated
morphism f : X → Y of algebraic stacks is cohomologically affine if

f∗ : QCoh(X ) → QCoh(Y)

is exact. A quasi-compact and quasi-separated algebraic stack X is cohomologically

affine if X → Spec Z is.
Example 6.5.14. An affine algebraic group G over a field k is linearly reductive
(Definition B.1.34) if and only if BG is cohomologically affine. If G is linearly
reductive and acts on an affine k-scheme Spec A, then [Spec A/G] is cohomologically
affine as the structure map [Spec A/G] → Spec k is the composition of the affine
morphism [Spec A/G] → BG and the cohomologically affine morphism BG →
Spec k.
Remark 6.5.15. By Serre’s Criterion for Affineness (4.5.16), an algebraic space is
cohomologically affine if and only if it is an affine scheme. An algebraic stack X
with affine diagonal is cohomologically affine if and only if Hi (X , F ) = 0 for all i > 0
and every quasi-coherent sheaf F ; this follows because the cohomology Hi (X , F ) can
be computed in QCoh(X ) for such stacks X by Proposition 6.1.29. This is not true
for algebraic stacks with non-affine diagonal, e.g., BE for an elliptic curve E.
Likewise, a morphism f : X → Y of algebraic stacks, each affine diagonal, is
cohomologically affine if and only if Ri f∗ (F ) = 0 for all i > 0 and every quasi-
coherent sheaf F . If in addition f is representable, then f is cohomologically affine
if and only if it is affine (see Corollary 6.5.18 below).
Remark 6.5.16 (Noetherian case). If X is noetherian, then a quasi-compact, quasi-
separated morphism f : X → Y is cohomologically affine if and only if f∗ : Coh(X ) →
QCoh(Y) is exact. This holds because every quasi-coherent sheaf is a colimit
of coherent sheaves (Proposition 6.1.16) and f∗ commutes with colimits. Since
cohomology also commutes with colimits (Proposition 6.1.31), a morphism f : X → Y
of noetherian algebraic stacks, both with affine diagonal, is cohomologically affine if
and only if Ri f∗ (F ) = 0 for all i > 0 and every coherent sheaf F .
Lemma 6.5.17. Consider a cartesian diagram
g′
X′ /X

π′ □ π
 
Y′
g
/ Y.

of algebraic stacks.
(1) If g is faithfully flat and π ′ is cohomologically affine, then π is cohomologically
affine.
(2) If Y has quasi-affine diagonal (e.g., a quasi-separated algebraic space) and π
is cohomologically affine, then π ′ is cohomologically affine.
Proof. For (1), by Flat Base Change (6.1.15) there is an equivalence g ∗ π∗ ≃ π∗′ g ′∗ of
functors defined on categories of quasi-coherent sheaves. Since π∗′ and g ′∗ are exact
and g ∗ is faithfully exact, π∗ is exact.

340
 For (2), we first show that if g is quasi-affine and π is cohomologically affine, then
π ′ is also cohomologically affine. It suffices to handle the cases that g is an open
immersion and g is affine. If g is an open immersion and F ′ ↠ G′ is a surjection
in QCoh(X ′ ), we define G = im(g∗′ F ′ → g∗′ G′ ). Note that g ′∗ G ∼ = G′ . Since π∗ is
exact, π∗ g∗ F ↠ π∗ G is surjective. If we apply g and use the identities g ∗ π∗ ≃ π∗′ g ′∗
′ ∗

and g ′∗ g∗′ ≃ id, we obtain a surjection π∗′ F ′ ↠ π∗′ g ′∗ G ∼

= π∗′ G′ . On the other hand,
if g is affine then g∗ is faithfully exact. Since π∗ and g∗′ are are exact, the identity
g∗ π∗′ ≃ π∗ g∗′ implies that π∗′ is also exact. To show (2), we may assume that Y
and Y ′ are quasi-compact. Choose smooth presentations Y = Spec A → Y and
Y ′ = Spec A′ → Y ′ ×Y Y . Since Y has quasi-affine diagonal, Y → Y is quasi-affine,
and thus XY → Y is cohomologically affine. As Y ′ → Y is affine, XY ′ → Y ′ is also
cohomologically affine. By (1), we conclude that π ′ : X ′ → Y ′ is cohomologically
affine.
Corollary 6.5.18. Let f : X → Y be a representable morphism of algebraic stacks
where Y has quasi-affine diagonal. Then f is affine if and only if f is cohomologically
affine.
Proof. Under the hypotheses, both affine and cohomologically affine morphisms are
local properties in the fppf topology. The statement therefore reduces to the case
where Y is an affine scheme and X is an algebraic space, which is precisely the
content of Serre’s Criterion for Affineness (4.5.16).

6.5.3 Properties of linearly reductive groups

Recall that an affine algebraic group G over a field k is linearly reductive if the
functor Rep(G) → Vectk , defined by V 7→ V G , is exact (Definition B.1.34), or
equivalently if BG → Spec k is cohomologically affine.
Proposition 6.5.19. Let 1 → K → G → Q → 1 be an exact sequence of affine
algebraic groups over a field k. Then G is linearly reductive if and only if both K
and Q are.
Proof. We will use the cartesian diagram
Q / BK / Spec k

□ □
  
Spec k / BG / BQ

of Exercise 2.4.40(c). For (⇐), fppf descent (Lemma 6.5.17(2)) along the morphism
Spec k → BQ implies that BG → BQ is cohomologically affine, and thus so is
the composition BG → BQ → Spec k. Conversely, since Q is affine, fppf descent
along Spec k → BG implies that BK → BG is affine. Therefore, the composition
BK → BG → Spec k is cohomologically affine, and K is linearly reductive. To
see that Q is linearly reductive, we claim that for every Q-representation V , the
adjunction V → q∗ q ∗ V is an isomorphism. Indeed, q ∗ V is a G-representation in
which K acts trivially, and since the pushforward of a G-representation W under
q : BG → BQ is the Q-representation W K , we see that V = q∗ q ∗ V . It follows that
Γ(BQ, −) = Γ(BG, q ∗ −) is exact.
Proposition 6.5.20. Let H be a linearly reductive group over an algebraically closed
field k. If H acts freely on an affine scheme U over k, then the algebraic space
quotient U/H is affine.

341
Proof. The algebraic space U/H and the good quotient Spec AH are both universal
for maps to algebraic spaces (Theorem 6.5.6(4)). Alternatively, the composition
U/H → BH → Spec k is an affine morphism followed by a cohomologically affine
morphism. It follows from Serre’s Criterion for Affineness (4.5.16) that U/H is
affine.

In particular, if H is a linearly reductive subgroup of an affine algebraic group

G, then the quotient G/H is affine. Matsushima’s Theorem provides a converse.

Proposition 6.5.21 (Matsushima’s Theorem). Let G be a linearly reductive group

over an algebraically closed field k.
(1) A subgroup H of G is linearly reductive if and only if G/H is affine.
(2) Given an action of G on an algebraic space U of finite type over k and a
k-point u ∈ U with stabilizer Gu , then Gu is linearly reductive if and only if
the orbit Gu is affine. In particular, if U is affine, then the stabilizer of a
point with closed orbit is necessarily linearly reductive.

Proof. Part (2) follows from (1) since Gu = G/Gu . For (1), the (⇒) implication
follows from Proposition 6.5.20. For the converse, consider the cartesian diagram

G/H / Spec k

□
 
BH / BG.

If G/H is affine, then BH → BG is affine is affine by fppf descent along Spec k →

BG. Therefore, BH → BG → Spec k is cohomologically affine, i.e., H is linearly
reductive.

6.5.4 First properties of good moduli spaces

Lemma 6.5.22. Consider a cartesian diagram

g′
X′ /X

π′ □ π
 
X′
g
/X

of algebraic stacks where X and X ′ are quasi-separated algebraic spaces.

(1) If g is faithfully flat and π ′ is a good moduli space, then π is a good moduli
space.
For the remaining statements, assume in addition that π is a good moduli space.
(2) The morphism π ′ is a good moduli space.
(3) For F ∈ QCoh(X ) and G ∈ QCoh(X), the adjunction map π∗ F ⊗ G →
∼
π∗ (F ⊗π ∗ G) is an isomorphism. In particular, G → π∗ π ∗ G is an isomorphism.
∼
(4) For F ∈ QCoh(X ), then the adjunction map g ∗ π∗ F → π∗′ g ′∗ F is an isomor-
phism.
(5) For a quasi-coherent sheaf of ideals J ⊆ OX , the natural map J → π∗ (π −1 J ·
OX ) is an isomorphism.

342
Proof. If g : X ′ → X is flat, then the pullback of the natural map OX → π∗ OX
under g is the map OX ′ → π∗′ OX ′ . Thus, Lemma 6.5.17 and fppf descent imply both
(1) and the case of (2) when g is flat. Also note that since X ′ is quasi-separated,
it has quasi-affine diagonal (Corollary 4.5.8). Before proving the general case of
(2), we prove (3). Choose an étale presentation U → X with U the disjoint union
of affine schemes. The base change πU : XU → U is also a good moduli space
by the flat case of (2). Moreover, by Flat Base Change (6.1.15). the adjunction
map π∗ F ⊗ G → π∗ (F ⊗ π ∗ G) pulls back under U → X to the adjunction map
πU,∗ FU ⊗ GU → πU,∗ (FU ⊗ πU ∗
GU ), where FU and GU are the pullbacks to XU
and XU respectively. We may therefore assume that X = Spec A is affine. If
G2 → G1 → G → 0 is a free presentation, then the maps π∗ F ⊗ Gi → π∗ (F ⊗ π ∗ Gi )
are isomorphisms for i = 1, 1. Since π∗ F ⊗ − and π∗ (F ⊗ π ∗ −) are right exact, we
have a commutative diagram

π∗ F ⊗ G2 / π∗ F ⊗ G1 / π∗ F ⊗ G /0

  
π∗ (F ⊗ π ∗ G2 ) / π∗ (F ⊗ π ∗ G1 ) / π∗ (F ⊗ π ∗ G) /0

of right exact sequences. Since the left two vertical maps are isomorphisms, so is
the right one.
For (2), it remains show that OX ′ → π∗′ OX ′ is an isomorphism as Lemma 6.5.17(2)
already established that π∗′ is exact. We can assume that X and X ′ are affine. In
this case, g∗ is faithfully exact so it suffices to show that g∗ OX ′ → g∗ π∗′ OX ′ is an
isomorphism. We may use the identity g∗′ π ′∗ OX ′ ∼ = π ∗ g∗ OX ′ following from the
affineness of g to identify this map with

= π∗ g∗′ OX ′ ∼
g∗ OX ′ → g∗ π∗′ OX ′ ∼ = π∗ π ∗ g∗ OX ′ .

The composition is an isomorphism as it is identified with the adjunction isomorphism

of (3) applied to F = g∗ OX ′ .
For (4), we know by Flat Base Change (6.1.15) that (4) is fppf local on X and
X ′ and that it holds when g is flat. We may therefore reduce to when X ′ → X
is a morphism of affine schemes. By factoring X ′ → X as a closed immersion
followed by a flat morphism, we can further reduce to the case that X ′ ,→ X is a
closed immersion defined by a quasi-coherent sheaf of ideals J ⊆ OX . We aim to
show that π∗ F/Jπ∗ F ∼ = π∗ (F/(π −1 J · OX )F ). Using the exactness of π∗ , this is
equivalent to the inclusion Jπ∗ F ,→ π∗ ((π −1 J · OX )F ) being surjective. The sheaf
(π −1 J · OX )F is the image of π ∗ J ⊗ F → F . By the exactness of π∗ , the pushforward
π∗ ((π −1 J · OX )F ) is the image of π∗ (π ∗ J ⊗ F ) → π∗ F , but by (3) this is identified
with the image of J ⊗ π∗ F → π∗ F .
For (5), if Z ⊆ X is the closed subspace defined by J, then the preimage ideal
sheaf π −1 J · OX defines the preimage π −1 (Z). The exactness of π∗ implies that
there is a commutative diagram of short exact sequences

0 /J / OX / OZ /0

  
0 / π∗ (π −1 J · OX ) / π∗ OX / π∗ Oπ−1 (Z) / 0.

As X → X and π −1 (Z) → Z are good moduli spaces, the right two vertical arrows
are isomorphisms, and thus so is the left arrow.

343
Remark 6.5.23. The isomorphism π∗ F ⊗ G → π∗ (F ⊗ π ∗ G) in (3) is similar to the
projection formula, except that it holds even if G is not locally free!
Lemma 6.5.24. Let π : X → X be a good moduli space with X quasi-separated.
(1) If A is a quasi-coherent sheaf of OX -algebras, then SpecX A → SpecX π∗ A is
a good moduli space.
(2) If Z ⊆ X is a closed substack defined by a sheaf of ideals I and im Z ⊆ X
is the scheme-theoretic image, i.e., the closed subspace defined by π∗ I ⊆ OX ,
then Z → im Z is a good moduli space.
Proof. For (1), since X ×X SpecX π∗ A → SpecX π∗ A is cohomologically affine
by Lemma 6.5.17(2) and SpecX A → X ×X SpecX π∗ A is affine, it follows that
SpecX A → SpecX π∗ A is cohomologically affine. Since the pushforward of OSpecX A
is OSpecX π∗ A by construction, SpecX A → SpecX π∗ A is a good moduli space.
Applying (1) to Z = SpecX (OX /I) and using the identity π∗ (OX /I) = OX /π∗ I.
recovers (2).
The above lemmas allow for quick proofs of the first two parts of Theorem 6.5.6.
Proof of Theorem 6.5.6(1). For every field-valued point x ∈ X(k), the base change
Xx := X ×X Spec k → Spec k is a good moduli space (Lemma 6.5.22(2)). In
particular, Γ(Xx , OXx ) = k, and it follows that Xx is non-empty and that π : X → X
is surjective. For a closed substack Z ⊆ X , Lemma 6.5.24(2) implies that Z → im Z
is a good moduli space and therefore also surjective. Thus, the set-theoretic image
π(Z) is identified with the scheme-theoretic image im Z and is therefore closed. Since
good moduli spaces are stable under base change, they are universally closed.
Proof of Theorem 6.5.6(2). For two substacks Z1 , Z2 ⊆ X defined by ideal sheaves
I1 , I2 ⊆ OX , we apply the exact functor π∗ to the short exact sequence 0 → I1 →
I1 + I2 → I2 /I1 ∩ I2 → 0 and surjection I2 → I2 /I1 ∩ I2 to obtain a commutative
diagram

π∗ I 2

 ((
0 / π∗ I 1 / π∗ (I1 + I2 ) / / π∗ I2 /π∗ (I1 ∩ I2 ) / 0.

It follows that the natural inclusion π∗ I1 + π∗ I2 → π∗ (I1 + I2 ) is surjective.

Exercise 6.5.25. Let X be an algebraic stack with affine diagonal and X → X be
a good moduli space. Show that for every field valued point x ∈ X (k) whose image
in |X | is a closed point, the stabilizer Gx is linearly reductive.

6.5.5 Finite typeness of good moduli spaces

To show that a good moduli space X → X preserves finite typeness, i.e., we will use
that X → X is universally submersive, and prove that finite typeness descends under
universally submersive maps. Recall from §A.4.2 that a morphism f : X → Y of
schemes is universally submersive if f is surjective and Y has the quotient topology,
and that these properties are stable under base change. Both fppf morphisms and
universally closed morphisms of noetherian schemes are universally submersive. This
notion extends to morphisms of algebraic stacks and since good moduli spaces are
universally closed (Theorem 6.5.6(1)), they are universally submersive.

344
Proposition 6.5.26 (Universally Submersive Descent for Finite Typeness). Let
X ′ → X be a universally submersive morphism of noetherian schemes. If X → Y
is a morphism of noetherian schemes and X ′ → X → S is of finite type, then so is
X → S.

Proof. We can assume that S = Spec A and X = Spec B are affine. Since a
noetherian ring B is of finite type over A if and only if the reductions of the
irreducible components of Spec B are of finite type over A, we can assume that B is
an integral domain. By Generic Flatness (A.2.13) and Raynaud-Gruson Flatification
(A.2.18), there is a commutative diagram

Z′ / X′

f
 
Z
g
/X /S

where g : Z = BlI X → X is the blowup along an ideal I ⊆ B, Z ′ is the strict

transform of X ′ , i.e., the closure of (Z ∖ g −1 (V (I))) ×X X ′ in the base change
Z ×X X ′ , and Z ′ → Z is flat. We claim the universal submersiveness of X ′ → X
implies that Z ′ → Z is surjective. As g : Z → X is an isomorphism over U = X ∖V (I)
and f : X ′ → X is surjective, we know that g −1 (U ) ⊆ Z is contained in the image. If
z ∈ Z is a point, we can choose a map Spec R → Z from a DVR whose generic point
maps to g −1 (U ) and whose special point maps to z. Since X ′ → X is universally
submersive, there exists an extension of DVRs R → R′ and a lift Spec R′ → X ′ (see
Exercise A.4.9). The induced map Spec R′ → Z ×X X ′ factors through Z ′ , and we
see that z is thus in the image of Z ′ .
Since X ′ is of finite type over S, so is Z ′ . By faithfully flat descent (Propo-
sition 2.1.27(1)), Z → S is also of finite type. To show that XS→ S is of finite
type, we may choose generators f1 , . . . , fn ∈ I so that Z = i Spec Bi where
Bi = B⟨fj /fi ⟩ ⊆ K = Frac(B)S is the subalgebra generated by B and the elements
fj /fi for j ̸= i. Write B = λ Bλ as a union of its finitely generated A-subalgebras.
For λ ≫ 0, each fi ∈ Bλ and we set Iλ = (f1 , . . . , fn ) ⊆ Bλ . Since Z is finite type
over S, each Bi is finitely generated over B, and thus for λ ≫ 0, we see that in the
diagram

Bλ,i
_ Bλ ⟨fj /fi ⟩  / Frac(Bλ )
_

  
Bi B⟨fj /fi ⟩  / Frac(B)

the inclusion Bλ,i ,→ Bi is surjective. It follows that Z = BlI Spec B = BlIλ Spec Bλ
for λ ≫ 0. Considering the composition
g pλ
gλ : Z −
→ X = Spec B −→ Spec Bλ ,

the pushforward of the injection OX ,→ g∗ OZ along pλ yields an inclusion pλ,∗ OX ,→

gλ,∗ OZ . But gλ,∗ OZ is coherent, hence so is pλ,∗ OX . This shows that B is a finite
Bλ -module and thus finitely generated as an A-algebra.

Proof of Theorem 6.5.6(3). If X is noetherian and I1 ⊆ I2 ⊆ · · · is an ascending

chain of ideal sheaves of OX , then π −1 I1 · OX ⊆ π −1 I2 · OX ⊆ is an ascending chain
of ideal sheaves of OX which terminates. By Lemma 6.5.24(5), In = π∗ (π −1 In · OX )
and therefore the chain I1 ⊆ I2 ⊆ · · · terminates and X is noetherian.

345
 Assume now that S is noetherian and X is of finite type over S. As X → X is
universally closed (Theorem 6.5.6(1)), it is also universally submersive. Choose a
smooth presentation U → X from a scheme. Since U → X is universally submersive,
so is the composition U → X → X. Since U → S is of finite type and X is
noetherian, Proposition 6.5.26 implies that X → S is also of finite type.
Given a coherent sheaf F on X , to show that the pushforward π∗ F is coherent,
we may assume that X = Spec A is affine and that X is irreducible. We claim we
can further reduce to the case that X is reduced. Let I ⊆ OX be the ideal sheaf
defining Xred ,→ X . Then for some N > 0, we have that I N = 0. By examining the
exact sequences 0 → π∗ (I k+1 F ) → π∗ (I k F ) → π∗ (I k F/I k+1 F ) → 0 and using that
π∗ (I k F/I k+1 F ) is coherent (since I k F/I k+1 F is supported on Xred ), we conclude
by induction that π∗ F is coherent.
By noetherian induction, we can assume that π∗ F is coherent if Supp(F ) ⊊ X .
The maximal torsion subsheaf Ftors ⊆ F has support strictly contained in X . Using
the exact sequence 0 → Ftors → F → F/Ftors → 0 and the exactness of π∗ , we
see the coherence of π∗ (F/Ftors ) implies the coherence of π∗ F . In other words,
we can assume that F is torsion free. In this case, every section s : OX → F is
injective. We now argue by induction on the dimension of the vector space ξ ∗ F ,
where ξ : Spec K → X is a field-valued point whose image is the generic point. If F
has no sections, then π∗ F = 0 is coherent. Otherwise, a section induces a short exact
sequence 0 → OX → F → F/OX → 0 and ξ ∗ (F/OX ) has strictly smaller dimension.
By again appealing to the exactness of π∗ , we see that the coherence of π∗ (F/OX )
implies the coherence of π∗ F .

6.5.6 Universality of good moduli spaces

We now complete the proof of Theorem 6.5.6 by showing that π : X → X is universal
for maps to algebraic spaces. Our argument follows the same logic as for coarse
moduli spaces in Theorem 4.2.3.
Proof of Theorem 6.5.6(4). We need to show that every diagram

X
f
π (6.5.27)

X /Y

has a unique filling, or in other words that the natural map Mor(X, Y ) → Mor(X , Y )
is bijective. The uniqueness follows as in the proof of Theorem 4.2.3 and uses only
that π : X → X is universally closed, schematically dominant, and surjective: if
h1 , h2 : X → Y are two fillings of (6.5.27), then π : X → X factors through the
equalizer E → X of h1 and h2 . Since E → X is universally closed, locally of finite
type, surjective, and a monomorphism, it is an isomorphism.
For existence, the case when Y is affine is easy:

Mor(X, Y ) = Hom(Γ(Y, OY ), Γ(X, OX )) =

Hom(Γ(Y, OY ), Γ(X , OX )) = Mor(X , Y ).

(Although unnecessary for the argument below, the case when Y is a scheme is
also straightforward: if {Yi } is an affine cover of Y and we set Xi := f −1 (Yi ) ⊆ X
with complement Zi , then X ∖ π −1 (π(Zi )) → X ∖ π(Zi ) is a good moduli space
and X ∖ π −1 (π(Zi )) ⊆ Xi . By the affine case, we have unique factorizations

346
X ∖ π(Zi ) → Yi and since i π(Zi ) = ∅, these maps glue to the desired map X → Y ;
T
see also [GIT, §0.6].)
For the general case, since X is quasi-compact, the map X → Y factors through
a quasi-compact subspace, so we can further assume that Y is quasi-compact. We
can also use étale descent and limit methods to reduce to the case that X = Spec A
where A is a strictly henselian local ring. This reduction works just as in the case
of coarse moduli spaces (Theorem 4.2.3). Since A is local, there is a unique closed
point x ∈ |X |, and we let Gx ,→ X be the closed immersion of the residual gerbe
(Proposition 3.5.16). Let (Y ′ = Spec B, y ′ ) → (Y, f (x)) be an étale presentation. The
base change X ′ := X ×Y Y ′ → X is an étale, separated, surjective, and representable
morphism. Let x′ ∈ |X ′ | be a preimage of x ∈ |X | and U ′ ⊆ X ′ be a quasi-compact
open substack containing x′ . In the diagram,

U′ / X′ /X
π
X f

Y′ /Y |

U ′ → X is a representable, quasi-finite, and separated morphism, and Zariski’s Main

Theorem (6.1.18) implies that there is a factorization U ′ → Xe → X with U ′ ,→ Xe an
open immersion and Xe → X a finite morphism. Writing Xe = SpecX A for a coherent
sheaf of algebras A, Lemma 6.5.24(1) implies that π e : Xe → Xe := SpecX π∗ A is a
good moduli space and we know from Theorem 6.5.6(3) that π∗ A is coherent. Thus,
Xe → X = Spec A is finite with A henselian, and we can apply Proposition B.5.9
to write X e = ` Spec Ai with each Ai a henselian local ring. Replace Xe with the
i
copy of Xei := πe−1 (Spec Ai ) containing x′ and replace U ′ with Xei ∩ U ′ . Then Xe has
a unique closed point (i.e., x′ ∈ |U ′ |) and thus the complement Xe ∖ U ′ is empty
(i.e., U ′ = Xe). We conclude that U ′ → X is a finite étale morphism, and since
it induces an isomorphism of residual gerbes at x′ , the map has degree one, i.e.,
U ′ → X is an isomorphism. Finaly, since Y ′ is affine, the morphism X ∼ = U′ → Y ′
factors through a map X → Y , and thus f : X → Y factors through the composition
′

X →Y′ →Y.

6.5.7 Luna’s Fundamental Lemma

We will apply the following result in our construction of good moduli spaces (Theo-
rem 6.10.1), in the Local Structure Theorem for Good Moduli Spaces (6.7.3), and
in the proof of Luna’s Étale Slice Theorem (6.7.5), but it also used in many other
arguments as well.

Theorem 6.5.28 (Luna’s Fundamental Lemma). Consider a commutative diagram

X′
f
/X

π′ π (6.5.29)
 
X′
g
/ X,

where f : X ′ → X is a separated and representable morphism of noetherian algebraic

stacks, each with affine diagonal, and where π and π ′ are good moduli spaces. Let
x′ ∈ |X ′ | be a point such that

347
 (a) f is étale at x′ ,
(b) f induces an isomorphism of stabilizer groups at x′ , and
(c) x′ ∈ |X ′ | and x = f (x′ ) ∈ |X | are closed points.
Then there is an open neighborhood U ′ ⊆ X ′ of π ′ (x′ ) such that U ′ → X is étale
and such that U ′ ×X X ∼
= π ′−1 (U ′ ).
Remark 6.5.30. This result is really saying two things: first, g is étale at π ′ (x′ )
and second, after replacing X ′ with an open neighborhood of π ′ (x′ ) the diagram
(6.5.29) is cartesian. In the case of quotients by finite groups, this was established
in Proposition 4.4.1. Luna’s original formulation [Lun73, p. 94] was the case when
X′ ∼= [Spec A′ /G] and X ∼ = [Spec A/G] with G linearly reductive and where X ′ → X
is induced by a G-equivariant map Spec A′ → Spec A.
Proof. We will adapt the argument of Theorem 6.5.6(4). Since the question is étale
local on X, limit methods (see the proof of Proposition 4.4.1) allow us to assume
that X = Spec A with A a strictly henselian local ring. If U ′ ⊆ X ′ is the étale locus
of f , then X ′ ∖ π ′−1 (π ′ (X ′ ∖ U ′ )) contains x′ as Theorem 6.5.6(2) implies that π ′ (x′ )
and π ′ (X ′ ∖ U ′ ) are disjoint. We can therefore replace X ′ with X ′ ∖ π ′−1 (π ′ (X ′ ∖ U ′ ))
and assume that f is étale. By Zariski’s Main Theorem (6.1.18), we may choose
a factorization X ′ → Xe = SpecX A → X with X ′ ,→ Xe an open immersion and
Xe → X a finite morphism. Then Xe → X e := SpecX π∗ A is a good moduli space and
Xe → X is finite. As A is henselian, we can write X e = ` Spec Ai with each Ai a
i
henselian local ring. If U ′ = Spec Ai denotes the connected component containing
the image of x′ , then π e−1 (U ′ ) ⊆ Xe is an open substack containing a unique closed
point, which is necessarily x′ ; it follows that X ′ = π −1 (U ′ ). Since X ′ → X is a finite
étale morphism of degree one (as it preserves residual gerbes at x′ ), we see that
f : X ′ → X is an isomorphism and thus so is g : X ′ → X.
Corollary 6.5.31. With the same hypotheses as Theorem 6.5.28, suppose that f is
étale and that for all closed points x′ ∈ |X ′ |
(a) f (x′ ) ∈ |X | is closed, and
(b) f induces an isomorphism of stabilizer groups at x′ .
Then g : X ′ → X is étale and (6.5.29) is cartesian.

6.5.8 Finite covers of good moduli spaces

Proposition 6.5.32. Consider a commutative diagram

X′
f
/X

π′ π
 
X′
g
/X

where X and X ′ are noetherian algebraic stacks with affine diagonal, and π and π ′
are good moduli spaces. Assume that
(a) f : X ′ → X is representable, quasi-finite, and separated
(b) f maps closed points to closed points, and
(c) g is finite.
Then f is finite.

348
Proof. By Zariski’s Main Theorem (6.1.18), there is a factorization X ′ → Xe =
SpecX A → X with X ′ ,→ Xe an open immersion and Xe → X a finite morphism.
Then X e = SpecX π∗ A is a finite over X and Xe → X e is a good moduli space. By
replacing X → X with X → X, we can assume that f is an open immersion. By
e e
replacing X with the fiber product X ′ ×X X , we can further reduce to the case that
X ′ = X. For every closed point x ∈ X, let x′ ∈ |X ′ | be the unique closed point over
x. By (b), f (x′ ) ∈ |X | is the unique closed point over x. Since X ′ contains all the
closed points of X , f : X ′ → X is an isomorphism.

Proposition 6.5.33. Suppose that X is a noetherian algebraic stack with affine

diagonal admitting a good moduli space π : X → X. If the diagonal X → X × X is
quasi-finite, then it is finite (i.e., π : X → X is separated).

Proof. We claim that X ×X X → X is a good moduli space. By Lemma 6.5.17,

the projection p1 : X ×X X → X is cohomologically affine and therefore so is the
p1 π
composition X ×X X −→ X − → X. On the other hand, if U → X is a smooth
presentation, then p1 : U ×X X → U is a good moduli space (Lemma 6.5.22) and in
∼ ∼
particular OU → p1,∗ OU ×X X . It follows from descent that OX → p1,∗ OX ×X X and
∼
thus OX → (π ◦p1 )∗ OX ×X X . This establishes the claim. The diagonal X → X ×X X
is a representable, quasi-finite, and separated morphism that sends closed points to
closed points and induces an isomorphism on good moduli spaces. Proposition 6.5.32
implies that X → X ×X X is finite. Note that since X has affine diagonal, the
finiteness of the diagonal is equivalent to its properness.

6.5.9 Descending vector bundles

Proposition 6.5.34. Let X be a noetherian algebraic stack and π : X → X be a
good moduli space. A vector bundle F on X descends to a vector bundle on X if
and only if for every field-valued point x : Spec k → X with closed image, the action
of Gx on the fiber F ⊗ k is trivial. In this case, π∗ F is a vector bundle and the
adjunction map π ∗ π∗ F → F is an isomorphism.

Proof. We follow the argument in the case of a tame coarse moduli space (Proposi-
tion 4.4.27). The condition is clearly necessary. To see that the condition is sufficient,
consider the commutative diagram

Gx  /X

p π
 
Spec κ(x)  / X.

We first claim that π ∗ π∗ F → F is surjective. For every closed point x ∈ |X |,

the hypotheses imply that p∗ p∗ (F |Gx ) ∼ = F |Gx . Applying π ∗ π∗ (−)|Gx to the sur-
jection F → F |Gx and using the exactness of π∗ , we obtain that (π ∗ π∗ F )|Gx →
= p∗ p∗ (F |Gx ) ∼
π ∗ (π∗ (F |Gx ))|Gx ∼ = F |Gx is surjective. The claim now follows from
Lemma 6.5.35(2).
To show that π∗ F is a vector bundle, we may assumeLthat X = Spec A is
affine and that the rank r of F is constant. The surjection s∈Γ(X,π∗ F ) A → π∗ F
pulls back to a surjection s∈Γ(X ,F ) OX → π ∗ π∗ F , and by the above claim, the
L

composition s∈Γ(X ,F ) OX → π ∗ π∗ F → F is surjective. As F |Gx ∼ = OGr x is trivial,

L
⊕r
for each closed point x ∈ |X |, we can find r sections ϕ : OX → F such that

349
ϕ|Gx is an isomorphism. By Lemma 6.5.35(2), there exists an open neighborhood
⊕r
U ⊆ X of π(x) such that ϕ|π−1 (U ) is an isomorphism. Thus π∗ ϕ : OX → π∗ F is an
isomorphism over U and we conclude that π∗ F is a vector bundle of the same rank
as F . Finally, since π ∗ π∗ F → F is a surjection of vector bundles of the same rank,
it is an isomorphism. See also [Alp13, Thm. 10.3] and [Ryd20, Thm. B]. The case
of a good quotient is due to Kempf; see [KKV89, Prop. 4.2].

Lemma 6.5.35. Let X be a noetherian algebraic stack and π : X → X be a good

moduli space. Let x ∈ |X | be a closed point.
(1) If F is a coherent sheaf on X such that F |Gx = 0, then there exists an open
neighborhood U ⊆ X of π(x) such that F |π−1 (U ) = 0.
(2) If ϕ : F → G is a morphism of coherent sheaves (resp., vector bundles of
the same rank) on X such that ϕ|Gx is surjective, then there exists an open
neighborhood U ⊆ X of π(x) such that ϕ|π−1 (U ) is surjective (resp., an isomor-
phism).

Proof. The same (elementary) argument of Lemma 4.4.26 applies.

6.5.10 Finiteness of cohomology

Proposition 6.5.36. Let X be an algebraic stack proper over a noetherian affine
scheme Spec A and let π : X → X be a good moduli space. There exists an integer
N such that for every coherent sheaf F on X , Hi (X , F ) = 0 for i > N and such
that Hi (X , F ) is a finitely generated A-module for every i.

Proof. Since π∗ is exact, the Grothendieck spectral sequence implies that Hi (X , F ) =

Hi (X, π∗ F ). Since π∗ F is coherent (Theorem 6.5.6(3)), the statement follows from
the finiteness of cohomology on X (Theorem 4.6.8) and Exercise 4.6.10.

6.5.11 Singularities of good moduli spaces

If X → X is a tame coarse moduli space of a separated noetherian algebraic stack
and X is smooth, then X has tame quotient singularities (see Remark 4.4.25). Here
we consider analogous properties of good moduli spaces, focusing first on the case
of an action of a linearly reductive group G over a field k on a noetherian affine
k-scheme X = Spec A. First, it is a classical theorem of Hochster and Roberts that
if A is regular, then AG is Cohen–Macaulay [HR74]. Interestingly, it is not true
that if A is Cohen–Macaulay, then so is AG [HR74, Ex. 2.1]. It is a theorem of
Boutot that, assuming X is of finite type over k and char(k) = 0, then if X has
rational singularities, so does X//G = Spec AG [Bou87]. Since smooth varieties are
rational and rational varieties are Cohen–Macualay, Boutot’s theorem implies the
Hochster–Robert’s theorem in characteristic zero.
These results also hold for a good moduli space X → X of an algebraic stack X
of finite type over a field k of characteristic zero and with affine diagonal. This is a
consequence of the Local Structure for Good Moduli Spaces (6.7.3). We also mention
the recent generalization: if X has klt singularities, then so does X [BGLM24,
Thm. 5]. To summarize these implications:

X smooth =⇒ X Cohen–Macaulay
X rational =⇒ X rational
X klt =⇒ X klt.

350
6.6 Coherent Tannaka duality and coherent com-
pleteness
Perhaps the purpose of categorical algebra is to show that which is trivial
is trivially trivial.

Peter Freyd

We prove a version of Tannaka duality for noetherian algebraic stacks with affine
diagonal (Theorem 6.6.1). After introducing the notion of coherent completeness of
an algebraic stack X (Definition 6.6.5), we show that certain quotient stacks with a
unique closed point are coherently complete (Theorem 6.6.13). This includes the
important examples of ΘR and ϕR defined in §6.9.2 for a complete DVR R.
The combined power of Coherent Tannaka Duality and coherent completeness
allows us to extend compatible maps Xn → Y from the nth nilpotent thickenings of
X0 to a morphism X → Y (Corollary 6.6.9). This technique is used in an essential
way in the proof of the Local Structure Theorem for Algebraic Stacks (6.7.1), and
also appears in many other arguments.

6.6.1 Coherent Tannaka Duality

A classical theorem of Gabriel [Gab62] states that two noetherian schemes X and
Y are isomorphic if and only if their abstract categories Coh(X) and Coh(Y ) of
coherent sheaves are equivalent, or in other words that a scheme X can be recovered
from the category Coh(X). In representation theory, classical Tannaka duality by
Saavedra Rivano [SR72] (see also Deligne and Milne’s article [DMOS82, Ch. II])
asserts that an affine group scheme G over a field k can be recovered from the tensor
category Repfd (G) of finite dimensional representations and its forgetful functor
Repfd (G) → Vectk .
Combining these two facts, one might hope that an algebraic stack X is recovered
by the tensor category Coh(X ).3 Following a brilliant observation of Lurie [Lur04],
we will not only confirm this expectation, but we will show that a tensor functor
Coh(Y) → Coh(X ) is enough to recover a morphism X → Y of algebraic stacks.
Theorem 6.6.1 (Coherent Tannaka Duality). For noetherian algebraic stacks X
and Y with affine diagonal, the functor

Mor(X , Y) → Mor⊗ (Coh(Y), Coh(X )), f 7→ f ∗ (6.6.2)

is an equivalence of categories, where Mor⊗ (Coh(Y), Coh(X )) denotes the category

of right exact additive tensor functors Coh(Y) → Coh(X ) of symmetric monoidal
abelian categories where morphisms are tensor natural transformations.
Remark 6.6.3. A symmetric monoidal category is a category A endowed with a
bifunctor ⊗ : A × A → A and a unit 1 ∈ A together with associativity isomorphisms
∼ ∼
αA,B,C : A ⊗ (B ⊗ C) → (A ⊗ B) ⊗ C, left and right unit isomorphisms lA : 1 ⊗ A →
∼ ∼
A → A and rA : A ⊗ 1 → A, and commutativity isomorphisms sA,B : A ⊗ B ∼ = B ⊗A
(with sA,B ◦ sB,A = id) satisfying certain coherence conditions [Mac71, §XI.1]. A
tensor functor F : A → B between symmetric monoidal abelian categories is a functor
∼ ∼
equipped with isomorphisms ΦA,B : F (A) ⊗ F (B) → F (A ⊗ B) and φ : 1B → F (1A )
compatible with the isomorphisms αA,B,C , lA , rA and sA,B [Mac71, §XI.2]. A
3 The structure as an abelian category is not enough, e.g., Coh(B(Z/2)) ∼
`
= Coh(Spec k Spec k).

351
tensor natural transformation between tensor functors is a natural transformation of
functors compatible with the isomorphisms ΦA,B and φ [Mac71, §XI.2]. A symmetric
monoidal abelian category (resp., symmetric monoidal R-linear abelian category for
a ring R) is a symmetric monoidal (resp., R-linear) abelian category A such that
⊗ : A × A → A is additive (resp., R-linear) in each variable. A tensor functor is
additive or R-linear if the underlying functor is. When X and Y are defined over a
noetherian ring R, then Theorem 6.6.1 induces an equivalence
∼
MorR (X , Y) → Mor⊗
R (Coh(Y), Coh(X ))

between morphisms over R and right exact R-linear tensor functors.

Proof. Since every quasi-coherent sheaf on a noetherian algebraic stack is a colimit

of its coherent subsheaves (Proposition 6.1.16), every right exact tensor functor
F : Coh(Y) → Coh(X ) extends to a tensor functor F : QCoh(Y) → QCoh(X )
preserving colimits. Likewise every tensor natural transformation between functors
of coherent sheaves extends uniquely to one defined on quasi-coherent sheaves.
Fully faithfulness: Let f, g : X → Y. Choose a smooth presentation p : U → Y where
U is an affine scheme. Since the question is smooth local on X , after replacing X
fe p
with X ×f,Y,p U , we may assume there is a factorization f : X − → Y. Likewise,
→U −
g q
we may assume there is a factorization g : X −→V − → Y where V is an affine scheme.
e

Since Y has affine diagonal, p : U → Y is affine and we have identifications

MorY (X , U ) ∼
= HomOY −alg (p∗ OU , f∗ OX ) ∼
= HomOX −alg (f ∗ p∗ OU , OX )

Therefore fe and ge correspond to sections sfe : f ∗ p∗ OU → OX and sge : g ∗ q∗ OV → OX .

A 2-isomorphism α : f → g is identified with a factorization

(fe,e
g ,α)
X / U ×X V

π
id
# 
X,

which is the same data as a section sα of OX → f ∗ π∗ OU ×X V . Letting α∗ : f ∗ → g ∗

be the image of α under (6.6.2), i.e., the pullback tensor natural transformation, the
section sα can be written as
id ⊗α∗
q∗ OV s ⊗s g
f ∗ π∗ OU ×X V ∼
= f ∗ (p∗ OU ) ⊗ f ∗ (q∗ OV ) −−−−−−−→ f ∗ (p∗ OU ) ⊗ g ∗ (q∗ OV ) −−−−→ OS .
fe e

To see the faithfulness of (6.6.2), if α, α′ : f → g are 2-isomorphisms with

α = α′∗ , then αq∗∗ OV = αq′∗∗ OV . Therefore, the two sections sα and sα′ are equal
∗

and α = α′ . For the fullness of (6.6.2), let β : f ∗ → g ∗ be a tensor natural

transformation. Then id ⊗βq∗ OV defines a section f ∗ π∗ OU ×X V → OS and thus
a 2-isomorphism α : f → g such that βq∗ OV = αq∗∗ OV . To see that βE = αE ∗
for
every E ∈ QCoh(Y), note that the factorization g = q ◦ ge yields a splitting of
g ∗ E → g ∗ (q∗ q ∗ E). Since f ∗ and g ∗ commute with direct sums, it suffices to assume
that E = q∗ G for G ∈ QCoh(V ). Write G = colim(OV⊕I → OV⊕J ) as a colimit of free
OV -modules. Since f ∗ and g ∗ commute with colimits and q∗ is exact, we conclude
that βq∗ G = αq∗∗ G .

352
Essential surjectivity (affine case): Let F : QCoh(Y) → QCoh(X ) be a tensor
functor preserving colimits. Assuming that X = Spec A and Y = Spec B are
noetherian affine schemes, there is a map
F
ϕ: B ∼
= End(OY ) −−−→ End(OX ) ∼
= A.

We claim that ϕ is a ring homomorphism and that there is a functorial isomorphism

F (N ) = N ⊗B A for N ∈ ModB . For b, b′ ∈ B, consider the commutative diagrams

OY ⊗ OY / OY A⊗A /A
F
b⊗b′ bb′ 7−−−→ ϕ(b)⊗ϕ(b′ ) ϕ(bb′ )
   
OY ⊗ OY / OY A⊗A / A,

where the horizontal maps correspond to multiplication. The commutativity of

the right square is implied by the fact that F preserves tensor products. This
shows that ϕ(b)ϕ(b′ ) = ϕ(bb′ ). For a B-module N , choose a free presentation
B ⊕J → B ⊕I → N → 0. Since both F and −⊗B A are right exact and preserve direct
sums, applying them to the free presentation yields an identification F (N ) ∼
= N ⊗B A
as both are cokernels of A⊕J → A⊕I . One checks similarly that this identification is
functorial.
Reduction to the case that X is affine: Choose a smooth presentation g : U → X
from an affine scheme and consider the diagram

U ×X U
p2 p1
 
U
g ∗ ◦F
g
 F #/
X Y

where the dashed arrow X / Y is denoting the tensor functor F : QCoh(Y) →

QCoh(X ) in the other direction. Assuming that the result holds when U is affine,
∼
there is a morphism h : U → Y and an isomorphism h → g ∗ ◦ F of functors. By full
∼
faithfulness, there is an isomorphism p1 ◦ h → p2 ◦ h satisfying the cocycle condition,
and thus smooth descent implies that there is a unique morphism f : X → Y with
F ≃ f ∗.
Reduction to the case that Y is affine: Let X = Spec A and choose a smooth
presentation q : V = Spec C → Y. Since Y has affine diagonal, q is an affine
morphism. Define B := F (q∗ OV ) which is an A-algebra since q∗ OV is an OY -
algebra. Consider the diagram

F′ / V = Spec C
Spec B
q
 
Spec A = X
F /Y

where F ′ : ModC → ModB is the right exact tensor functor sending M to F (q∗ M
f)
(which is a module over B = F (q∗ OV ) because q∗ M
f is a q∗ OV -module). By the

353
affine case, F ′ is induced by a morphism f ′ : Spec B → Spec C. We can extend the
above diagram to:
f ′′
Spec B ⊗A B / V ×Y V

  f′  
Spec B / V = Spec C

q
 
Spec A = X
F / Y.

Since q is affine, V ×Y V is affine and the top square (under either set of projections)
is cartesian.
If we could show that A → B is faithfully flat, we would be done as the full
faithfulness in the affine case would imply that f ′ descends to our desired morphism
f : X → Y. This seems hard to directly check, but we do know already that the maps
B ⇒ B ⊗A B are faithfully flat as they correspond to base changes of the smooth
maps V ×Y V ⇒ V . We will show instead that A → B is universally injective. Since
faithful flatness descends under universal injectivity maps (Proposition A.2.24(4)),
the faithful flatness of A → B follows from the universal injectivity.
Universal injectivity of A → B: Recall from Definition A.2.22 that an injective map
of A-modules is called universally injective if it remains injective after tensoring
by every A-module. By Proposition A.2.24(3), this notion is local under faithfully
flat morphisms and thus extends to morphisms F → G of quasi-coherent sheaves
on an algebraic stack. Since q : V → Y is faithfully flat, OY → q∗ OV is universally
injective (Proposition A.2.24(1)). We write q∗ OV = colim Qi as a colimit of coherent
subsheaves (Proposition 6.1.16), and we may assume that each Qi contains the
image of OY → q∗ OV . Then OY → Qi is also universally injective. Since Qi is
coherent, OY → Qi is a split injection smooth locally on Y (Proposition A.2.24(2)).
Applying F to OY → q∗ OV = colim Qi and using that F preserves colimits, we have
a factorization

4 F (Qi )


A = F (OY ) / B = F (q∗ OV ) = colim F (Qi ).

It suffices to show that A → F (Qi ) is universally injective. We will show in fact

that it is a split injection. As OY → Qi is smooth locally split, the map on duals
Q∨i → OY = OY is surjective. Applying F , we have a surjection F (Qi ) → F (OY ) =
∨ ∨

A (using right exactness) and we can choose an element λ ∈ F (Q∨ i ) mapping to

1. Under the natural map F (Q∨ i ) → F (Qi ) , the element λ is sent to a map
∨

F (Qi ) → A, which is a section of the given map A → F (Qi ). See also [Lur04],
[HR19b], [BHL17] and [SP, Tag 0GRR].

Remark 6.6.4 (Relation to classical Tannaka duality). If G is an affine group scheme

over a field k, then the category C = Repfd (G) of finite dimensional representations is
a symmetric monoidal k-linear category and there is a tensor functor ω : Repfd (G) →
Vectk . For k-algebra R, define ωR as the composition
−⊗ R
ωR : Repfd (G) → Vectk −−−−
k
→ ModR ,

354
and let Aut⊗ (ωR ) denote the group of tensor natural isomorphisms of ωR . Then G
is recovered as the functor Aut⊗ (ω) on affine k-schemes assigning R to Aut⊗ (ωR )
[DMOS82, II.2.8]. On the other hand, Coherent Tannaka Duality (6.6.1) asserts
that for every noetherian k-algebra R, there is an equivalence of categories
∼
Mork (Spec R, BG) → Mor⊗ (Rep(G)fd , ModR ).

In this way, we see that Rep(G)fd determines BG. To recover G, the fiber functor
ω : Repfd (G) → Vectk corresponds to a morphism p : Spec k → BG and G =
Autk (p). For instance, consider orthogonal groups O(q) and O(q ′ ) with respect to
non-degenerate quadratic forms q and q ′ of the same dimension. Then Rep(O(q)) ∼ =
Rep(O(q ′ )) even though O(q) and O(q ′ ) may not be isomorphic; in this case the two
maps Spec k → BO(q) and Spec k → BO(q ′ ) define two different fiber functors on
the same category.
The classical version also provides conditions when the data of (C, ω) is isomorphic
to the category of representations of a group scheme. Namely, C is called rigid if
for every object of X ∈ C, there is a dual X ∨ ∈ C, i.e., an object X ∨ such that
X ∨ ⊗ − : C → C is right adjoint to X ⊗ − : C → C. If C is a rigid symmetric monoidal
k-linear abelian category with End(1) = k and ω : C → Vectk is an exact faithful
k-linear tensor functor, then Aut⊗ (ω) is representable by an affine group scheme G
over k and there is a tensor equivalence C ∼ = Repfd (G) under which ω corresponds
to the forgetful functor [DMOS82, II.2.11]. Moreover, G is of finite type over k if
and only if C has a tensor generator.

6.6.2 Coherent completeness

The combination of Coherent Tannaka Duality and coherent completeness (Corol-
lary 6.6.9) will become a valuable new tool at our disposal.
Definition 6.6.5. A noetherian algebraic stack X is coherently complete along a
closed substack X0 if the natural functor

Coh(X ) → lim Coh(Xn ), F 7→ (Fn )

←−
is an equivalence of categories, where Xn denotes the nth nilpotent thickening of X0
and Fn is the pullback of F to Xn .
Remark 6.6.6. If I ⊆ OX is the coherent sheaf of ideals defining X0 , then Xn
is defined by I n+1 . Letting in : Xn ,→ Xn+1 denote the natural inclusion, an
object in lim Coh(Xn ) corresponds to a sequence Fn ∈ Coh(Xn ) of coherent sheaves
←−
together with maps αn : in,∗ Fn → Fn+1 inducing isomorphism Fn → i∗n Fn+1 . A
morphism (Fn , αn ) → (Fn′ , αn′ ) is a sequence of maps ϕn : Fn → Fn′ such that
′
ϕn+1 ◦ αn = αn+1 ◦ in,∗ ϕn .
Example 6.6.7. If (R, m) is a complete noetherian local ring, then the Artin–Rees
Lemma (B.5.4) implies that Spec R is coherently complete along Spec R/m. The
same is true if R = lim R/I n is a noetherian I-adically complete ring.
Example 6.6.8. Grothendieck’s Existence Theorem (C.5.3) asserts that if X is
a proper scheme over a complete local ring (R, m) and X0 = X ×R R/m, then X
is coherently complete along X0 . If X is a proper Deligne–Mumford stack over
Spec R, the same is true (Exercise 4.6.13(b)) The result also holds if X is a proper
algebraic stack over any I-adically complete noetherian ring. See [Ols05, Thm. 1.4]
or [Con05a, Thm. 4.1].

355
Corollary 6.6.9 (Coherent Tannaka Duality). Let X and Y be noetherian algebraic
stacks with affine diagonal. Suppose that X is coherently complete along X0 . Then
there is an equivalence of categories

Mor(X , Y) → lim Mor(Xn , Y), f 7→ (fn ),

←−
where fn : Xn → Y denotes the restriction of f to the nth nilpotent thickening Xn of
X0 .

Proof. This follows from the equivalences

Mor(X , Y) ≃ Mor⊗ Coh(Y), Coh(X ) (Coherent Tannaka Duality)

⊗
(coherent completeness)



≃ Mor Coh(Y), lim Coh Xn
←−
≃ lim Mor⊗ Coh(Y), Coh Xn



←−
(Coherent Tannaka Duality).


≃ lim Mor Xn , Y
←−

Remark 6.6.10. If X and Y are defined over a noetherian ring R, then there is an
equivalence MorR (X , Y) → lim MorR (Xn Y). This follows in the same way using the
←−
Tannaka duality equivalence between the category of morphisms X → Y over R and
the category of right exact R-linear tensor functors (Remark 6.6.3).

Exercise 6.6.11. Give a direct proof Mor(Spec R, Y) → lim Mor(Spec R/mn , Y) is

←−
an equivalence when (R, m) is a complete noetherian local ring.
Hint: Use Proposition 4.3.4 to lift a map Spec R/m → Y to a smooth presentation
and the Infinitesimal Lifting Criterion (3.7.1) to reduce to the case that Y is a
scheme.

Exercise 6.6.12. Let G be an affine algebraic group acting on a separated noetherian

algebraic space W over k. Let W0 ⊆ W be a G-invariant closed subspace and let
Wn be its nth nilpotent thickenings. Suppose that [W/G] is coherently complete
along [W0 /G]. For every noetherian algebraic space X over k with affine diagonal
equipped with an action of G, the natural map on equivariant maps

MorG (W, X) → lim MorG Wn , X

←−
n

is bijective.
Hint: reduce to Corollary 6.6.9 by using that a G-equivariant map W → X corre-
sponds to a morphism [W/G] → [X/G] over BG.

6.6.3 Coherent completeness of quotient stacks

The coherent completeness result that we will apply, e..g, in the proof of the Local
Structure Theorem for Algebraic Stacks (6.7.1), is the following:

Theorem 6.6.13. Let k be an algebraically closed field and R be a complete noethe-

rian local k-algebra with residue field k. Let G be a linearly reductive group over k
acting on an affine scheme Spec A of finite type over R. Suppose that AG = R and
that there is a G-fixed k-point x ∈ Spec A. Then [Spec A/G] is coherently complete
along the closed substack BG defined by x.

356
Example 6.6.14. If Gm acts diagonally on Ar , then [Ar /Gm ] is coherently complete
along the origin BGm . In other words a Gm -equivariant module over k[x1 , . . . , xr ] is
equivalent to a compatible family of Gm -equivariant modules over k[x1 , . . . , xr ]/(x1 , . . . , xr )n+1 .
Remark 6.6.15. Consider the diagram
 / [Spec A/G] ×AG k   / [Spec A/G]
BG

□
  
Spec k  / Spec AG .

A formal consequence of the above theorem is that [Spec A/G] is also coherently
complete with respect to the fiber [Spec A/G] ×AG k. This version is analogous to
Grothendieck’s Existence Theorem (6.6.8), but the coherent completeness along
BG is a substantially stronger statement, e.g., for [An /Gm ] where the fiber of
[An /Gm ] → Spec k is everything.
Proof of Theorem 6.6.13. We need to show that Coh(X ) → lim Coh(Xn ) is an equiv-
←−
alence of categories, where X = [Spec A/G] and Xn is the nth nilpotent thickening
of BG ,→ X of the inclusion of the residual gerbe at x.
Full faithfulness: Suppose that F and F ′ are coherent OX -modules, and let Fn and
Fn′ denote the restrictions to Xn , respectively. We need to show that

HomOX (F, F ′ ) → lim HomOX (Fn , Fn′ )

←−
is bijective. Since X has the resolution property (Proposition 6.2.10), we can find a
resolution F2 → F1 → F → 0 by vector bundles. This induces a diagram

0 / HomO (F, F ′ ) / HomO (F1 , F ′ ) / HomO (F2 , F ′ )

X X X

  
0 / lim HomO (Fn , Fn′ ) / lim HomO (F1,n , Fn′ ) / lim HomO (F2,n , Fn′ )
←− X
←− X
←− X

with exact rows, and we see that it suffices to prove full faithfulness in the case that
F is a vector bundle. In this case,

HomOX (F, F ′ ) = HomOX (OX , F ∨ ⊗ F ′ )

HomOX (Fn , Fn′ ) = HomOX OXn , (Fn∨ ⊗ Fn′ ) .

Therefore, we can also assume that F = OX and we are reduced to showing that

Γ(X , F ′ ) → lim Γ Xn , Fn′ (6.6.16)

←−
is an isomorphism. Writing F ′ = Mf where M is a finitely generated A-module with
an action of G and letting m ⊆ A be the maximal ideal for x, then Γ(Xn , Fn′ ) =
M G /(mn M )G since G is linearly reductive. We must therefore verify that

M G → lim M G /(mn M )G (6.6.17)

←−
is an isomorphism. To this end, we first show that
\
G
mn M = 0, (6.6.18)
n≥0

357
or in other words that (6.6.17) is injective. Let N := n≥0 mn M . The Artin–Rees
T
Lemma (B.5.4) applied to N ⊆ M implies that there exists an integer c such that
mn M ∩ N = mn−c (mc M ∩ N ) for all n ≥ c. Taking n = c + 1, we see that N = mN
so N ⊗A A/m = 0. Since the support of N is a closed G-invariant subscheme of
Spec A which does not contain x, it follows that N = 0.
Note also that since G is linearly reductive, M G is a finitely generated AG -module
(Corollary 6.5.8(3)). We next establish that (6.6.17) is an isomorphism if AG is
artinian. In this case, ((mn M )G ) automatically satisfies the Mittag–Leffler condition
(it is a sequence of artinian AG -modules). Taking the inverse limit of the exact
sequences 0 → (mn M )G → M G → M G /(mn M )G → 0 and applying (6.6.18) yields
an exact sequence
0 → 0 → M G → lim M G /(mn M )G → 0,
←−
and implies that (6.6.17) is an isomorphism. To establish (6.6.17) in the general
case, let J = mG A ⊆ A and observe that

n
G
MG ∼ = lim M G / mG M G ∼ = lim M/J n M , (6.6.19)
←− ←−
since G is linearly reductive. For each n, we know that

G
G
M/J n M ∼
= lim M G / (J n + ml )M (6.6.20)
←−
l

by the artinian case. Finally, combining (6.6.19) and (6.6.20) together with the
observation that J n ⊆ ml for n ≥ l, we conclude that

G
MG ∼ = lim M/J n M
←−
n

G
∼
= lim lim M G / (J n + ml )M
←− ←−
n l

G
∼
= lim M G / ml M .
←−
l

Essential surjectivity (explicit construction): The linear reductivity of G implies

that every coherent sheaf F = M f on [Spec A/G] decomposes as a direct sum
M
M= M (ρ) , (6.6.21)
ρ∈Γ

where Γ denotes the set of isomorphism classes of irreducible representations of G

and M (ρ) is the isotypic component corresponding to ρ; explicitly, if Wρ denotes the
irreducible representation corresponding to ρ, then M (ρ) = HomG k (Wρ , M ) ⊗ Wρ .
Moreover, the decomposition (6.6.21) is compatible with the A-module structure of
M and the decomposition A = ρ∈Γ A(ρ) .
L

Let us also note that if F = M

f ∈ Coh(X ) with restrictions Mn = M/mn+1 M ,
(ρ)
then applying (6.6.17) to M ⊗Wρ∨ shows that M (ρ) = lim Mn . By Theorem 6.5.6(3),
←−
we also know that M (ρ) is a finitely generated AG -module. In particular, A(ρ) =
lim(A/mn+1 )(ρ) is a finitely generated AG -module.
←−
This suggests that if Fn = M fn is a compatible system of coherent OX -modules
n
(ρ)
with Mn = ρ Mn , we define
L

M
M (ρ) := lim Mn(ρ) and M := M (ρ) . (6.6.22)
←−
ρ∈Γ

358
To see that M is an A-module with a G-action, let ρ, γ ∈ Γ be irreducible represen-
tations and let Λ ⊆ Γ denote the finite set of nonzero irreducible representations
(ρ) (γ) (λ)
appearing Wρ ⊗Wγ . Taking limits of the maps An ⊗(A/mn+1 )G Mn → λ∈Λ Mn ,
L
defines multiplication

Mn(λ) ∼
M
M (λ)
A(ρ) ⊗A M (γ) → lim A(ρ) ⊗(A/mn+1 )G Mn(γ) → lim


= M .
←− n ←−
λ∈Λ λ∈Λ

Note that M/mn+1 M ∼ = Mn by construction.

It remains to show that the A-module M of (6.6.22) is finitely generated. The
coherent sheaf F0 = Mf0 on X0 = BG is a finite dimensional G-representation and
we can consider the coherent OX -module F0 ⊗OX0 OX or equivalently the A-module
M0 ⊗k A with its natural G-action. Since X is cohomologically affine, the functor

HomOX (F0 ⊗OX0 OX , −) = Γ(X , (F0∨ ⊗OX0 OX ) ⊗OX −)

is exact. Apply the functor to the surjection M ↠ M0 induces a map

M 0 ⊗k A → M (6.6.23)

which we would like to show is surjective. We already know that the restrictions
M0 ⊗k (A/mn+1 ) → Mn are surjective as its cokernel is a coherent module on Xn
not supported at the unique closed point.
∼
As above, we first handle the case that AG is artinian. Since A(ρ) → lim(A/mn )(ρ)
←−
is a finitely generated AG -module, it follows that (A/mn )(ρ) stabilizes to A(ρ) for
n ≫ 0. Since (6.6.23) induces surjections M0 ⊗k (A/mn+1 ) → Mn , the modules
(ρ) (ρ) (ρ)
Mn stabilize to M∞ for n ≫ 0 and M = ρ M∞ is finitely generated. In the
L

general case, let Xm = Spec AG /(m ∩ AG )m+1 and consider the cartesian diagram

 / X ×X Xm+1   / · · · 
X ×X Xm  /X
im

πm □ πm+1 π
     
Xm  / Xm+1  / · · · / X.
jm

For each m, we may consider the nth nilpotent thickenings Zm,n of X0 ,→ X ×X Xm

which is a closed substack of Xn . Since Xm is the spectrum of artinian ring, the
restrictions Fn |Zm,n extend to a coherent sheaf Hm = N
em on X ×X Xm . Moreover,
there is a canonical isomorphism between Hm and the restriction of Hm+1 to
∼
X ×X Xm . By Lemma 6.5.22(4), the adjunction morphism jm ∗
πm+1,∗ → πm,∗ i∗m
(ρ)
is an isomorphism on quasi-coherent sheaves. This implies that Nm+1 = Γ(X ×X
(ρ) (ρ)
Xm+1 , Hm+1 ⊗Wρ∨ ) restricts to Nm and that M (ρ) = lim Nm is a finitely generated
←−
AG -module. The map (6.6.23) is surjective as it surjective on each ρ-isotypical
component by Nakayama’s lemma.
Essential surjectivity (indirect construction): For an alternative argument, use the
resolution property of X (Proposition 6.2.10) to choose a surjection E ↠ F0 from
a vector bundle E on X . Since each Fn+1 → Fn is surjective and HomOX (E, −) =
Γ(X , E ∨ ⊗OX −) is exact, we can lift E ↠ F0 to compatible maps E → Fn , each
which is surjective (Nakayama’s lemma). The sequence (ker(En → Fn )) is not
necessarily an adic sytem of coherent sheaves as the restriction ker(En+1 → Fn+1 ) to
Xn may not be ker(En → Fn ). But we can modify it as follows: for each l ≥ m ≥ n,

359
the images of ker(El → Fl ) in Em stabilize to Km′
for l ≫ m and Km ′ ′
/mn+1 Km
stabilize to Kn for m ≫ n (see also [SP, Tag 087X]). Then (Kn ) ∈ lim Coh(Xn ) is
←−
an adic sequence. Repeating the construction, we can find a vector bundle E ′ on
X and compatible surjections E ′ → Kn . By full faithfulness, there is a morphism
E ′ → E extending the maps En′ → En . Then coker(E ′ → E) is a coherent OX
extending (Fn ). See also [AHR20, Thm. 1.3] and [AHR19, Thm. 1.6].

Exercise 6.6.24. If S is a noetherian affine scheme, show that [A1 /Gm ]S is coherently
complete along BGm,S .

6.7 Local structure of algebraic stacks

All human knowledge thus begins with intuitions, proceeds thence to
concepts, and ends with ideas.

Emmanual Kant

We prove the Local Structure Theorem for Algebraic Stacks (6.7.1) around points
with linearly reductive stabilizer. Philosophically, this means that quotient stacks
of the form [Spec A/G], where G is linearly reductive, are the building blocks of
algebraic stacks near points with linearly reductive stabilizers in a similar way to
how affine schemes are the building blocks of schemes and algebraic spaces. This
allows one to reduce many properties of algebraic stacks to the well-understood case
of [Spec A/G]. This theorem is a generalization of the Local Structure Theorem for
Deligne–Mumford Stacks (4.3.1) and will be applied in a similar way. For example,
we will apply it to construct good moduli spaces (Theorem 6.10.1) in an analogous
way to how the Deligne–Mumford case was applied to construct coarse moduli spaces
in the proof of the Keel–Mori Theorem (4.4.6).

Theorem 6.7.1 (Local Structure Theorem for Algebraic Stacks). Let X be an

algebraic stack of finite type over an algebraically closed field k with affine diagonal.
For every point x ∈ X (k) with linearly reductive stabilizer Gx , there exists an affine
étale morphism
f : ([Spec A/Gx ], w) → (X , x)

which induces an isomorphism of stabilizer groups at w.

Remark 6.7.2. In the case that x ∈ |X | is a smooth point, then one can say more:
there is also an étale morphism

([Spec A/Gx ], w) → ([TX ,x /Gx ], 0)

where TX ,x is the Zariski tangent space equipped as a Gx -representation. This

addendum follows from the proof but also follows from applying Luna’s Étale Slice
Theorem (6.7.5) to [Spec A/Gx ]. The upshot is that we can reduce étale local
properties of X to Gx -equivariant properties of TX ,x ; for moduli problems, this
translates into studying the first-order deformation space as a representation under
the automorphism group.
By combining this theorem with Luna’s Fundamental Lemma (6.5.28), we obtain
the following result.

360
Corollary 6.7.3 (Local Structure for Good Moduli Spaces). Let X be an algebraic
stack of finite type over an algebraically closed field k with affine diagonal. Suppose
that there exists a good moduli space π : X → X. Then for every closed point x ∈ |X |,
there exists an étale neighborhood W → X of π(x) and a cartesian diagram

[Spec A/Gx ] /X

□ π
 
W = Spec AGx / X.

A nice upshot is that we can provide an equivalent and more intuitive definition
of a good moduli space.

Corollary 6.7.4 (Characterization of Good Moduli Spaces). Let X be an algebraic

stack of finite type over an algebraically closed field k with affine diagonal. A
morphism π : X → X to an algebraic space is a good moduli space if and only if
every point x ∈ X has an étale neighborhood Spec B → X such that X ×X Spec B ∼ =
[Spec A/G] with G linearly reductive and B = AG .

Section outline: We first discuss Luna’s Étale Slice Theorem (6.7.5), a beautiful
argument providing an explicit construction of an étale neighborhood in the case
that X is already known to have the form [Spec B/G] with G linearly reductive.
The proof of the Local Structure Theorem (6.7.1) is far less explicit requiring: (1)
deformation theory, (2) coherent completeness, (3) Coherent Tannaka Duality and
(4) Artin Approximation or Equivariant Artin Algebraization.
Letting T = [TX ,x /Gx ], deformation theory produces an embedding Xn ,→ Tn of
the nth nilpotent thickenings of x and 0. The key step in the proof is to show that the
system of closed morphisms (Xn → Tn ) algebraizes. The first step is effectivization:
the fiber product Tb := T ×T Spec O bT,π(0) , where π : T → T := TX ,x //Gx , is
coherently complete (Theorem 6.6.13). We can thus construct a closed substack
Xb ,→ Tb extending Xn ,→ T and then apply Coherent Tannaka Duality (6.6.9) to
construct a morphism Xb → X extending Xn → X . If x ∈ |X | is smooth, Artin
Approximation over the GIT quotient TX ,x //Gx produces an étale neighborhood
U → TX ,x //Gx such that π −1 (U ) → X algebraizes Tb → X . The general case is more
involved, requiring an equivariant version of Artin Algebraization (Theorem 6.7.18).

6.7.1 Luna’s Étale Slice Theorem

The Local Structure Theorem (6.7.1) is inspired by Luna’s Étale Slice Theorem in
equivariant geometry.

Theorem 6.7.5 (Luna’s Étale Slice Theorem). Let G be a linearly reductive group
over an algebraically closed field k and let X be an affine scheme of finite type over
k with an action of G. If x ∈ X(k) has linearly reductive stabilizer, then there exists
a Gx -invariant, locally closed, and affine subscheme W ⊆ X such that the induced
map
[W/Gx ] → [X/G] (6.7.6)

361
is affine and étale. If in addition the orbit Gx ⊆ X is closed, then there is a cartesian
diagram
[W/Gx ] / [X/G]

□
 
W//Gx / X//G

where W//Gx → X//G is also étale.

Moreover, if x ∈ X is a smooth point and if we denote by Nx = TX,x /TGx,x the
normal space to the orbit, then we can arrange that there is a cartesian diagram

[Nx /Gx ] o / [X/G]

g
[W/Gx ]

□ □
  
Nx //Gx o W//Gx / X//Gx

where all horizontal arrows are étale and g is induced from a Gx -invariant étale
morphism W → Nx .
Remark 6.7.7. If the orbit Gx is closed, then Matsushima’s Theorem (6.5.21) implies
that the stabilizer Gx is linearly reductive.
Remark 6.7.8. One can also formulate the statement G-equivariantly: G acts
naturally on the quotient G ×Gx W := (G × W )/Gx and there is an identifica-
tion [W/Gx ] ∼ = [(G ×Gx W )/G] and likewise W//Gx ∼ = (G ×Gx W )//G (see Exer-
cise 3.4.19). The morphism (6.7.6) corresponds to an étale G-equivariant morphism
G ×Gx W → X.
The proof will rely on the existence of a Gx -invariant morphism X → TX,x , which
we refer to as the Luna map. The use of TX,x here is an abuse of notation—TX,x is
a vector space over k and we view it as a scheme via Spec(Sym∗ TX,x ∨
).
Lemma 6.7.9 (Luna map). Let G be a linearly reductive group over an algebraically
closed field k and let X be an affine scheme of finite type over k with an action of G.
If x ∈ X(k) has linearly reductive stabilizer, there exists a Gx -equivariant morphism

f : X → TX,x (6.7.10)

sending x to the origin. If X is smooth at x, then f is étale at x.

Proof. Letting X = Spec A and m ⊆ A be the maximal ideal of x, then m and
m/m2 are Gx -representations, and thus Gx acts naturally on the tangent space
TX,x := Spec(Sym∗ m/m2 ). Since Gx is linearly reductive, the surjection m → m/m2
of Gx -representations has a section m/m2 ,→ m. This induces a Gx -equivariant
ring map Sym∗ m/m2 → A and thus a Gx -equivariant morphism f : Spec A → TX,x
sending x to the origin. If x ∈ X is smooth, then since f induces an isomorphism of
tangent spaces at x, we conclude that f is étale at x (Theorem A.3.2).
Proof of Theorem 6.7.5. Since X is affine and of finite type, we can choose a finite
dimensional G-representation V and a G-equivariant closed immersion X ,→ A(V )
(Proposition B.1.18). If W ⊆ A(V ) is an affine Gx -invariant locally closed subscheme
such that [W/Gx ] → [A(V )/G] is étale, then the same is true for W ′ := W ∩ X ⊆ X
and [W ′ /Gx ] → [X/G]. We can therefore immediately reduce to the case that x ∈ X
is smooth. In this case, the Luna map f : X → TX,x is Gx -invariant, étale at x, and

362
with f (x) = 0 (Lemma 6.7.9). The subspace TGx,x ⊆ TX,x is Gx -invariant and using
again that Gx is linearly reductive, the surjection TX,x → Nx = TX,x /TGx,x has a
section Nx ,→ TX,x . We define W as the preimage of Nx under f :

W _ / Nx
_
□
 
X
f
/ TX,x .

Since the maps f : [W/Gx ] → [X/G] and g : [W/Gx ] → [Nx /Gx ] induce an isomor-
phism of tangent spaces and stabilizer groups at w, they are both étale at x ∈ W
(or equivalently the G-equivariant maps G ×Gx W → X and G ×Gx W → G ×Gx Nx
are étale at (id, x)). We have a commutative diagram

[Nx /Gx ] o / [X/G]

g f
[W/Gx ]

  
Nx //Gx o W//Gx / X//Gx

where both f and g are étale at x, preserve stabilizer groups at x, and map x
to closed points. We can therefore apply Luna’s Fundamental Lemma (6.5.28) to
replace W with a Gx -equivariant, open, and affine neighborhood of x so that the
above squares are cartesian.
When X is already known to be a quotient stack of a normal quasi-projective
scheme, the Local Structure Theorem follows from a direct argument. This case is
sufficient to handle many moduli problems, e.g., Bun ss
r,d (C) in characteristic 0.

Exercise 6.7.11. If G is a connected affine algebraic group over an algebraically

closed field k acting on a normal finite type k-scheme X, and x ∈ X(k) has linearly
reductive stabilizer, show that there is a Gx -invariant, locally closed, and affine
subscheme W ,→ X such that [W/Gx ] → [X/G] is étale.
Hint: Use Sumihiro’s Theorem on Linearizations (B.1.30) to reduce to the case that
X = P(V ). Choose a homogenous polynomial f not vanishing at x such that P(V )f
is Gx -invariant and then argue as in the proof of Luna’s Étale Slice Theorem by
considering the Gx -equivariant étale map P(V )f → Tx P(V ).

6.7.2 Deformation theory

In our proof of the Local Structure Theorem (6.7.1), we will need some deformation
theory of algebraic stacks.
Proposition 6.7.12. Consider a commutative diagram

W _
f
/X
>

 
W′ /Y

of noetherian algebraic stacks with affine diagonal, where X → Y is smooth and

affine and W ,→ W ′ is a closed immersion defined by a square-zero sheaf of ideals.
If W is cohomologically affine, there exists a lift in the above diagram.

363
Proof. When W is affine, the statement follows from the Infinitesimal Lifting Crite-
rion (A.3.1). To reduce to this case, let U ′ → W ′ be a smooth presentation with
U ′ an affine scheme, and define U := U ′ ×W ′ W. Since W has affine diagonal, each
n-fold fiber product (U/W)n := U ×W · · · ×W U is affine. There is a commutative
diagram
// U q1 / W f / X
(U/W)
_
2
_ _ 7?
′
fU f′
   
(U ′ /W ′ )2
p1
// U ′ / W′ / Y,
p2

where we have chosen a lift fU′ : U ′ → X . Define the coherent sheaf F = f ∗ (Ω∨X /Y )⊗J
on W, where J is the ideal sheaf defining W ,→ W . We know by Exercise 6.1.17
′

that the set of lifts U ′ → X is a torsor under Γ(U, q1∗ F ), i.e., any other lift differs
from fU′ by an element of Γ(U, q1∗ F ). Because X → Y is representable, to check that
fU′ descends to a morphism f ′ : W ′ → X , we need to arrange that fU′ ◦ p1 = fU′ ◦ p2 .
Let qn : (U/W)n → W. The difference fU′ ◦ p1 − fU′ ◦ p2 can be viewed an element
of Γ((U/W)2 , q2∗ F ). Since q1 : U → W is a surjective, smooth, and affine morphism,
there is an exact sequence of quasi-coherent sheaves

0 → F → q1,∗ q1∗ F → q2,∗ q2∗ F → q3,∗ q3∗ F → · · · ;

see Exercise 2.1.3. Since W is cohomologically affine, taking global sections yields
an exact sequence

Γ(U, q1∗ F )
d0
/ Γ((U/W)2 , q2∗ F ) d1
/ Γ((U/W)3 , q3∗ F )
 / p∗1 s − p∗2 s,  / p∗12 t − p∗13 t + p∗23 t.
s t

One checks that d1 (fU′ ◦ p1 − fU′ ◦ p2 ) = 0 so there exists an element s ∈ Γ(U, q1∗ F )
with d0 (s) = fU′ ◦ p1 − fU′ ◦ p2 . After modifying the lift fU′ by s, we see that
fU′ ◦ p1 − fU′ ◦ p2 = 0 so that fU′ descends to f ′ : W ′ → X .

Remark 6.7.13. Alternatively, one can show that the obstruction to this deformation
problem lies in Ext1OW (f ∗ ΩX /Y , J) = H1 (W, f ∗ (Ω∨
X /Y ) ⊗ J), which vanishes since W
is cohomologically affine. The above result holds more generally [Ols06, Thm. 1.5].

Exercise 6.7.14. Let W ,→ W ′ be a closed immersion of algebraic stacks of finite

type over k with affine diagonal defined by a square-zero sheaf of ideals. Let G be
an affine algebraic group over k. If W is cohomologically affine, show that every
principal G-bundle P → W extends to a principal G-bundle P ′ → W ′ .
Hint: Use smooth descent and the deformation theory of principal G-bundles over
schemes (Exercise C.2.6).

Remark 6.7.15. The deformation question in Exercise 6.7.14 is equivalent to de-

forming the morphism f : W → BG to a morphism W ′ → BG, which is analogous
to Proposition 6.7.12 except that X = BG → Y = Spec k is not affine. Since
maps to BG correspond to principal G-bundles, this question is equivalent to de-
formation a principal G-bundle P → W to P ′ → W ′ , and just like the case when
W is a scheme (Exercise C.2.6), the obstruction lies in the group H2 (W, g ⊗ J),
where J is the ideal sheaf defining W ,→ W ′ . Indeed, the general theory of the
cotangent complex (see Remarks C.3.7 and C.7.6) implies that the obstruction

364
 p
lies in Ext1OW (Lf ∗ LBG/k , J). Under the composition Spec k −
→ BG → Spec k, we
have an exact triangle p∗ LBG/k → Lk/k → Lk/BG . Since Lk/k = 0, we obtain
that p∗ LBG/k = Lk/BG [−1] ∼ = g∨ [−1] and LBG/k ∼
= g∨ [−1], where the Lie alge-
bra g is equipped with the adjoint representation. Thus Ext1OW (Lf ∗ LBG/k , J) =
H1 (W, f ∗ g[1] ⊗ J) = H2 (W, g ⊗ J). Since W is cohomologically affine with affine
diagonal, this cohomology group is 0 and the obstruction vanishes.
We will also need the following criteria for morphisms to be closed immersions
or isomorphisms.
Lemma 6.7.16. Let f : X → Y be a representable morphism of algebraic stacks of
finite type over an algebraically closed field k, each with affine diagonal. Assume that
|X | = {x} and |Y| = {y} consist of a single point and that f induces an isomorphism
between X0 := BGx and Y0 := BGy . Let mx ⊆ OX and my ⊆ OY be the ideal
sheaves defining X0 and Y0 , and let f1 : X1 → Y1 be the induced morphism between
the first nilpotent thickenings of X0 and Y0 .
(1) If f1 is a closed immersion, then so is f .
(2) If f1 is a closed immersion and there is an isomorphism n≥0 mny /mn+1
L ∼
=
y
n n+1
L
n≥0 mx /mx of graded OX0 -modules, then f is an isomorphism.
Proof. Choose a smooth presentation V = Spec B → Y from an affine scheme such
that V ×Y Y0 ∼ = Spec k (Theorem 3.6.1). Then B is a local artinian k-algebra
as Y consists of only one point. The base change U = V ×Y X is an algebraic
space and since Ured = Vred is a point, it follows from Proposition 4.5.19 that
U = Spec A with A a local artinian k-algebra. We can therefore assume that
f : Spec A → Spec B is a morphism of local artinian schemes. For (1), we need to
show that if B/m2B → A/m2A is surjective, so is B → A. We first claim that the
inclusion mB A ,→ mA is surjective. By Nakayama’s Lemma, it suffices to show that
mB A/mA mB A → mA /m2A is surjective, but this follows from the hypothesis that
the composition mA /m2A → mB A/mA mB A → mA /m2A is surjective. Since B/mB →
A/mB A = A/mA is surjective, another application of Nakayama’s Lemma shows
that B → A is surjective. See also [Har77, Lem. II.7.4] for a related criterion. For
(2), since dimk mnB /mn+1
B A , the surjections mB /mB
= dimk mnA /mn+1 n n+1
→ mnA /mn+1
A
are isomorphisms and it follows that f is an isomorphism.

6.7.3 Proof of the Local Structure Theorem—smooth case

In the case that x ∈ |X | is a smooth point with linearly reductive stabilizer, we will
show that there are étale morphisms

[TX ,x /Gx ] ← [Spec A/Gx ] → X .

The proof of the general case (which does not rely on the smooth case) is technically
much more involved than the smooth case but nevertheless follows the same logic.
For this reason, we recommend the reader to first digest the proof in the smooth
case.
Proof of Theorem 6.7.1—smooth case. Since the k-point x ∈ |X | is locally closed
(Proposition 3.5.16), by replacing X by an open substack we may assume that x ∈ |X |
is a closed point. Let I be the coherent sheaf of ideals defining X0 := BGx ,→ X
and set Xn to be the nth nilpotent thickening defined by I n+1 . The Zariski tangent
space TX ,x can be identified with the normal space (I/I 2 )∨ to the orbit, viewed as
a Gx -representation. (Note that when X = [X/G] with G a smooth affine algebraic

365
group, then TX ,x is identified with the normal space to the orbit TX,ex /TGx,ex for a
point x
e ∈ X(k) over x.)
Define the quotient stack T := [TX ,x /Gx ]. Let T0 = BGx be the closed substack
of T supported at the origin and Tn its nth nilpotent thickenings. We claim
that there are compatible isomorphisms Xn ∼ = Tn . Since Gx is linearly reductive,
X0 = BGx is cohomologically affine. By the deformation theory of principal Gx -
bundles (Exercise 6.7.14), we can inductively extending the principal Gx -bundle
Spec k → X0 to principal Gx -bundles Spec An → Xn . This yields isomorphisms
Xn ∼= [Spec An /Gx ] and affine morphisms Xn → BGx . We have a closed immersion
X0 ,→ T and we can inductively find lifts

Xn _ /T
;

 
Xn+1 / BGx

since T → BGx is smooth and affine (Proposition 6.7.12). The induced morphism
X1 → T1 is an isomorphism
Since X1 → T1 is a morphism between deformations of BGx by the coherent
sheaf I/I 2 , it is necessarily an isomorphism (see Lemma C.1.7). (In fact, both X1
and T1 are trivial deformations as they admit retractions to BGx .) Lemma 6.7.16(2)
now implies that the maps Xn → Tn are isomorphisms.
Let π : T → T = TX ,x //Gx be the morphism to the GIT quotient. The fiber
product Tb := Spec O bT,π(0) ×T T is a quotient stack of the form [Spec B/G] where B is
of finite type over the complete noetherian local k-algebra B G = O bT,π(0) . Therefore Tb
∼
is coherently complete along T0 (Theorem 6.6.13) and Mor(T , X ) → lim Mor(Tn , X )
←−
is an equivalence by Coherent Tannaka Duality (6.6.9). It follows that the morphisms
Xn ∼ = Tn ,→ X extend to a morphism Tb → X filling in the diagram

/ Tb /T ()
Xn ∼
= Tn X

□
 
Spec O
bT,π(0) / T.

The functor parameterizing isomorphism classes of morphisms

(T ′ → T ) 7→ T ′ ×T T → X /∼

F : Sch /T → Sets,

is limit preserving as X is of finite type over k (see Exercise 3.3.33). The morphism
Tb → X yields an element of F over Spec O bT,π(0) . By Artin Approximation (B.5.18),
there exists an étale morphism (U, u) → (T, 0) where U is an affine scheme with a k-
point u ∈ U and a morphism (U ×T T , (u, 0)) → (X , x) agreeing with (Tb , 0) → (X , x)
to first order. Observe that U ×T T ∼ = [Spec A/Gx ] for a finitely generated k-algebra
A such that U = Spec AGx . Since U ×T T is smooth at (u, 0) and X is smooth at
x, and since U ×T T → X induces an isomorphism of tangent spaces and stabilizer
groups at (u, 0), the morphism U ×T T → X is étale at (u, 0). After replacing U
with an affine open subscheme, we can arrange that U ×T T → X is affine (by
Proposition 6.7.17) and étale.
Proposition 6.7.17. Let X be an algebraic stack of finite type over an algebraically
closed field with affine diagonal. Let f : [Spec A/G] → X be a finite type morphism

366
with G linearly reductive. If w ∈ Spec A has closed G-orbit and f induces an
isomorphism of stabilizer groups at w, then there exists a G-invariant affine open
subscheme U ⊆ Spec A containing w such that f |[U/G] is affine.

Proof. Set W = [Spec A/G] with π : W → Spec AG . Since f : W → X induces an

isomorphism of stabilizer groups at w, after replacing Spec AG with an affine open
containing the image w, we can use upper semicontinuity to arrange that the inertia
stack IW/X → W is quasi-finite. Choose a smooth presentation V = Spec B → X
and consider the fiber product

WV / V = Spec B

□
 
[Spec A/G] = W
f
/ X.

Since X has affine diagonal, Spec B → X is affine and therefore WV is cohomo-

logically affine. As WV has quasi-finite diagonal, Proposition 6.5.33 implies that
WV → V is separated, and it follows from descent that W → X is also separated.
Therefore, the relative inertia IW/X → W is both proper and quasi-finite, hence
finite. Since the fiber over w ∈ |W| is trivial, there is an open neighborhood U over
which the relative inertia is trivial. Therefore, after replacing Spec AG with an affine
open, we can arrange that f is representable. Since f is also cohomologically affine,
Serre’s Criterion for Affineness (6.5.18) implies that f is affine.

6.7.4 Equivariant Artin Algebraization

The smoothness hypothesis of x ∈ |X | was used above to establish that Tn ∼ = Xn
and that U ×T T → X is étale. More critically, it implied that lim Γ(Xn , OXn ),
←−
which is identified with the Gx -invariants of a miniversal deformation space, is
the completion of a finitely generated k-algebra, namely O bT,0 . If x ∈ |X | is not
smooth, it seems difficult to directly establish that lim Γ(Xn , OXn ) is the completion
←−
of a finitely generated k-algebra. Recall that we encountered a similar issue when
discussing Artin Algebraization (C.6.8): when the complete local ring R is known to
be the completion of a finitely generated algebra, then Artin Algebraization is an
easy consequence of Artin Approximation (see Remark C.6.9). To circumvent this
issue in our general proof of Artin Algebraization, we wrote R = O bV,v /I where V is
a finite type k-scheme (e.g., V = Ak and v = 0) and used Artin Approximation to
n

simultaneously approximate both the given object over R and the equations defining
I. We follow a similar strategy but proceed G-equivariantly.
We will use the following extension of the notion of formal versality introduced in
Definition C.4.2: for an algebraic stack Tb with a unique closed point t, a morphism
ξb: Tb → X of prestacks over Sch is formally versal at t if every commutative diagram

Z _ / Tb
?
ξb
 
Z′ /X

has a lift, where Z ,→ Z ′ is a closed immersion of noetherian algebraic stacks with

affine diagonal, |Z| = |Z ′ | consists of a single point, and the image of Z → Tb is t.

367
Theorem 6.7.18 (Equivariant Artin Algebraization). Let k be an algebraically
closed field and R be a complete noetherian local k-algebra with residue field k. Let
Tb = [Spec B/G] be an algebraic stack of finite type over R = B G , where G is linearly
reductive. Assume that the unique closed point t ∈ |Tb | has stabilizer equal to G. If X
is a limit preserving prestack over Sch/k and η : Tb → X is a morphism of prestacks
formally versal at t, then there exists
(1) an algebraic stack W = [Spec A/G] of finite type over k and a closed point
w ∈ |W|;
(2) morphisms f : W → X and φ : Tb → W such that in the diagram

Tb
φ
η
(6.7.19)

W
f
/X

the induced morphisms φn : Tbn → Wn between the nth nilpotent thickenings

of t and w are isomorphisms, and there exists compatible 2-isomorphisms
∼
ηn → fn ◦ φn .
Moreover, if X is an algebraic stack of finite type over k with affine diagonal, then
it can be arranged that (6.7.19) is commutative and that φ induces an isomorphism
Tb → Wc := W ×W Spec O bW,π(w) , where π : W → W = Spec AG .

Remark 6.7.20. If one takes G to be the trivial group, one recovers the classical
version of Artin Algebraization (C.6.8).
As in the proof of Artin Algebraization, we will apply Artin Approximation to a
well-chosen integer N to construct W such that there are isomorphisms Wn ∼ = Tbn
for n ≤ N and such that the Artin–Rees Lemma (B.5.4) implies that there are
also isomorphisms Wn ∼ = Tbn for n > N . To get control over the constant in the
Artin–Rees Lemma, we need to generalize Definition C.6.3: for a noetherian algebraic
stack X with a closed point x defined by a a sheaf of ideals mx and an integer c ≥ 0,
we say that (AR)c holds at x for a map φ : E → F of coherent sheaves on X if

φ(E) ∩ mnx F ⊆ φ(mn−c

x E), ∀n ≥ c.

When X is a scheme, (AR)c holds for all sufficiently large c by the Artin–Rees
Lemma, even if {x} is replaced with an arbitrary closed subscheme. By smooth
descent, (AR)c also holds for algebraic stacks for c ≫ 0.

Proof. The morphism η : Tb → X and Tb → BG induce a morphism Tb → X × BG.

Let b
t = Spec R be the GIT quotient of Tb = [Spec B/G] with R = B G . Since R is
the colimit of its finitely generated k-subalgebras and X × BG is limit preserving,
limit methods (§B.3) imply that there is a commutative diagram
,
Tb /S / X × BG

□
  
T
b /S / Spec k

where S = Spec R′ is an affine scheme of finite type over k and S is an algebraic

stack of finite type over S with affine diagonal such that Tb = Tb ×S S. Moreover,

368
we can arrange that S → BG is affine. Let se ∈ |S| and s ∈ S be the images of t.
After possibly adding generators to R′ , we can arrange that R′ → R → R/m2R is
surjective. By Complete Nakayama’s Lemma (B.5.6(3)), O bS,s → R is surjective. In
particular, both Tb ,→ Sb := Spec O
bS,s and Tb ,→ S ×S Sb are closed immersions.
⊕r
By choosing a resolution OSb → OSb ↠ R and pulling it back to S ×S S, b we
obtain a resolution
α ⊕r β
ker(β) ,−−→ OS× −−→ OS×S Sb ↠ OTb . (6.7.21)
SS
b

Consider the functor F : Sch/S → Sets assigning an S-scheme U to the set of

isomorphism classes of complexes
α
⊕r β
L−
→ OS× SU
−
→ OS×S U

of finitely presented, quasi-coherent OS×S U -modules. By limit arguments, F is

limit preserving. The complex (6.7.21) defines an element (α, β) ∈ F (S) b such that
coker(β) = OTb . Let N be an integer such that (AR)N holds for α and β at (e s, s).
Artin Approximation (B.5.18) gives an étale neighborhood (S ′ , s′ ) → (S, s) and
an element (α′ , β ′ ) ∈ F (S ′ ) such that (α, β) = (α′ , β ′ ) in F (OS,s /mN
s
+1
). We let
W ,→ S ×S S be the closed substack defined by coker(β ) and set w = (e
′ ′
s, s′ ) ∈ |W|.
Letting Sn , Sn′ and Tbn be the nth nilpotent thickenings of S, S ′ and Tb at the images
of t ∈ |T |, we have that Tb ×Tb TbN and W ×S ′ SN ′
are equal as closed substacks of
S ×S SN . This gives (1)–(2) for n ≤ N . In particular, we have an isomorphism
φN : TbN → WN and we let ψN : WN → TbN be its inverse.
Using that η : Tb → X is formally versal, we can inductively find compatible lifts
for n ≥ N
Wn _
ψn
/7 Tb
ψn+1
η
 
Wn+1 / X.

On the other hand, applying Lemma C.6.4 (generalized to stacks by smooth descent)
on S ×S Sb with c = N to the complex (6.7.21) and the restriction of the complex
defined by (α′ , β ′ ), we obtain an isomorphism Grmt OTb ∼= Grmw OW of graded OBG -
modules. By Lemma 6.7.16, the induced morphisms ψn : Wn → Tbn are isomorphisms
for all n. As Tb is coherently complete (Theorem 6.6.13), Coherent Tannaka Duality
(6.6.9) implies that the inverses φn = ψn−1 : Tbn → Wn effectivize to a morphism
φ : Tb → W. This completes (1)–(2).
For the final statement, when X is algebraic, we again apply Coherent Tannaka
Duality, using the coherent completeness of both Tb and W. By applying Corol-
lary 6.6.9 to the inverses ψn = φ−1 n , we can construct an inverse ψ : W → T of φ.
c b
Thus φ : Tb → W c is an isomorphism. Using the fully faithfulness of Corollary 6.6.9,
∼
there is a 2-isomorphism η → f ◦ φ extending the given 2-isomorphisms ηn → fn ◦ φn
and thus Tb → W c is a morphism over X .

6.7.5 Proof of the Local Structure Theorem—general case

Proof of Theorem 6.7.1. We may assume that x ∈ |X | is a closed point. Let T :=
[TX ,x /Gx ], let π : T → T = TX ,x //Gx be the morphism to the GIT quotient, and let
Tb := Spec O bT,π(0) ×T T . Let T0 = BGx be the closed substack supported at the

369
origin and Tn its nth nilpotent thickenings. We will construct compatible closed
immersions Xn ,→ Tn . Since Gx is linearly reductive, X0 = BGx is cohomologically
affine. By deforming the principal Gx -bundle Spec k → X0 using Exercise 6.7.14, we
can inductively construct isomorphisms Xn ∼ = [Spec An /Gx ]. By the deformation
theory of the smooth and affine morphism Tb → BGx (Proposition 6.7.12), we can
inductively find lifts
Xn _ /T
;

 
Xn+1 / BGx .

As in the smooth case, X1 → T1 is an isomorphism. By Lemma 6.7.16(1), each

morphism Xn → Tn is a closed immersion.
If In denotes the ideal sheaf defining Xn ,→ Tn , then OTn /In is a system of
coherent OTn -modules. Since Tb is coherently complete (Theorem 6.6.13), there
exists a coherent sheaf of ideals I ⊆ OTb such that the surjection OTb → OTb /I
extends the surjections OTn → OXn . The closed immersion Xb ,→ Tb defined by I
extends the given closed immersions Xn ,→ Tn yielding a commutative diagram

 /( X
Xn _  / Xb η
_

  
Tn  / Tb /T

□
 
Spec O
bT,0 /T

of solid arrows. Since Xb is also coherently complete, Coherent Tannaka Duality

(6.6.9) gives a morphism η : Xb → X extending the above diagram. Since Xb has the
same nilpotent thickenings of Xb, the morphism η : Xb → X is formally versal at 0.
By Equivariant Artin Algebraization (6.7.18) with G = Gx , we obtain a morphism
f : W = [Spec B/Gx ] → X from an algebraic stack W of finite type over k with a
closed point w ∈ |W| and a morphism φ : Xb → W over X inducing an isomorphism
Xb → W ×W Spec O bW,π(w) where π : W → Spec B Gx . Since f : W → X induces
isomorphisms Wn → Xn , f is étale at w. After replacing W with an open substack,
we can arrange that f is étale everywhere. By Proposition 6.7.17, we can also
arrange that f is affine. See also [AHR20, AHR19, AHHLR24].

6.7.6 The coherent completion at a point

We say that (X , x) is a complete local stack if X is a noetherian algebraic stack
with affine stabilizers and with a unique closed point x such that X is coherently
complete along the residual gerbe Gx . Important examples are the quotient stack
([Spec A/G], x) of Theorem 6.6.13, e.g., ([An /Gm ], 0) is complete local. The coherent
completion of a noetherian algebraic stack X at a point x is a complete local stack
b) together with a morphism η : (Xbx , x
(Xbx , x b) → (X , x) inducing isomorphisms of
nth infinitesimal neighborhoods of x b and x. If X has affine diagonal, then the pair
(Xbx , η) is unique up to unique 2-isomorphism by Coherent Tannaka Duality (6.6.9).

370
Theorem 6.7.22. Let X be an algebraic stack of finite type over an algebraically
closed field k with affine diagonal. For every point x ∈ X (k) with linearly reductive
stabilizer Gx , the coherent completion Xbx exists. Moreover,
(1) The coherent completion is a quotient stack Xbx = [Spec B/Gx ] such that the
invariant ring B Gx is the completion of a finite type k-algebra and B Gx → B
is of finite type.
(2) If f : (W, w) → (X , x) is an étale morphism where W = [Spec A/Gx ], the
point w ∈ |W| is closed, and f induces an isomorphism of stabilizer groups
at w, then Xbx ∼
= W ×W Spec O bW,π(w) , where π : W → W = Spec AGx is the
morphism to the GIT quotient.
(3) If π : X → X is a good moduli space, then Xbx = X ×X Spec O
bX,π(x) .

Proof. The Local Structure Theorem (6.7.1) gives an étale morphism f : (W, w) →
(X , x), where W = [Spec A/Gx ] and f induces an isomorphism of stabilizer groups at
the closed point w. The main statement as well as (1) and (2) follow by taking Xbx =
W ×W Spec O bW,π(w) and B = A ⊗AGx A dGx because X bx = [Spec B/Gx ] is coherently
complete (Theorem 6.6.13). Part (3) follows from (2) using Corollary 6.7.3.
The fact that completions determine the étale local structure of finite type
schemes (Corollary B.5.21) generalizes to stacks.
Theorem 6.7.23. Let X and Y be algebraic stacks of finite type over an algebraically
closed field k with affine diagonal. Suppose x ∈ |X | and y ∈ |Y| are k-points
with linearly reductive stabilizer group schemes Gx and Gy , respectively. Then the
following are equivalent:
(1) There exist compatible isomorphisms Xn → Yn .
(2) There exists an isomorphism Xbx → Yby .
(3) There exist an affine scheme Spec A with an action of Gx , a point w ∈ Spec A
fixed by Gx , and a diagram of étale morphisms

[Spec A/Gx ]
f g

y %
X Y

such that f (w) = x and g(w) = y, and both f and g induce isomorphisms of
stabilizer groups at w.
If, in addition, the points x ∈ |X | and y ∈ |Y| are smooth, then the conditions above
are equivalent to the existence of an isomorphism Gx → Gy of group schemes and an
isomorphism TX ,x → TY,y of tangent spaces which is equivariant under Gx → Gy .
Proof. The implications (3)⇒(2)⇒(1) are immediate. We also have (1)⇒(2) by Co-
herent Tannaka Duality (6.6.9) To show that (2)⇒(3), let (W = [Spec A/Gx ], w) →
(X , x) be an étale neighborhood as given by the Local Structure Theorem (6.7.1).
Let π : W → W = Spec AGx denote the good moduli space. Then Xbx ∼ = W ×W
Spec O bW,π(w) . The functor

F : Sch/W → Sets, (T → W ) 7→ Mork (W ×W T, Y)

is locally of finite presentation. Artin Approximation (B.5.18) applied to F and

α ∈ F (Spec O bW,π(w) ) provides an étale morphism (W ′ , w′ ) → (W, w) and a morphism

371
φ : W ′ := W ×W W ′ → Y such that φ|W1′ : W1′ → Y1 is an isomorphism. Since
Wc′ w′ ∼= Xbx ∼
= Yby , it follows that φ induces an isomorphism W
c′ → Yb by Lemma 6.7.16.
After replacing W ′ with an open neighborhood, we obtain an étale morphism
(W ′ , w′ ) → (Y, y). The final statement follows from Luna’s Etale Slice Theorem
(6.7.5).

6.7.7 Applications to equivariant geometry

Sumihiro’s Theorem on Torus Action (B.1.31) asserts that for a normal scheme
of finite type over k with the action of a torus T , every k-point has a T -invariant
affine open neighborhood. If X is not normal, there are not necessarily T -invariant
affine open neighborhoods, e.g., the node in the projective nodal cubic C with its
Gm -action. However, there is always a T -equivariant affine étale neighborhood, e.g.,
there is a Gm -equivariant étale neighborhood Spec k[x, y]/(xy) → C of the node.

Theorem 6.7.24. Let X be an algebraic space locally of finite type over an alge-
braically closed field k with affine diagonal. Suppose that X has an action of an affine
algebraic group G. If x ∈ X(k) has linearly reductive stabilizer, then there exists a
G-equivariant étale neighborhood (Spec A, u) → (X, x) inducing an isomorphism of
stabilizer groups at u.
If G is a torus, then every point x ∈ X(k) has a G-invariant étale neighborhood
(Spec A, u) → (X, x) inducing an isomorphism of stabilizer groups at u.

Proof. By the Local Structure Theorem (6.7.1), there is an étale neighborhood

([Spec A/Gx ], u) → ([X/G], x) such that u is a closed point and f induces an isomor-
phism of stabilizer groups at u. By Proposition 6.7.17, after replacing Spec A with a
Gx -invariant open affine neighborhood of u, we can arrange that the composition
[Spec A/Gx ] → [X/G] → BG is affine. Therefore, W := [Spec A/Gx ] ×[X/G] X is
an affine scheme and W → X is a G-equivariant étale neighborhood of x. If G is a
torus, then any subgroup of G is linearly reductive.

6.8 Gm -actions and the Białynicki-Birula Stratifica-

tion
We show that the fixed locus of a linearly reductive group action on a smooth variety
is smooth (Theorem 6.8.2). We also prove the representability and properties of the
attractor locus with respect to a Gm -action (Theorem 6.8.9), which will be useful in
understanding the valuative criteria of Θ- and S-completeness in the next section.
Finally, after establishing a general version of the Białynicki-Birula Stratification
(Theorem 6.8.14), we discuss applications to computing cohomology (§6.8.20).

6.8.1 Fixed loci

Definition 6.8.1 (Fixed locus). If X is an algebraic space over a field k equipped
with an action of an affine algebraic group G, we define the fixed locus as the functor

X G := MorG
k (Spec k, X) : Sch/k → Sets

assigning a k-scheme S to the set MorG

k (S, X) of G-equivariant maps S → X, where
S is endowed with the trivial G action.

372
Theorem 6.8.2. Let X be a separated algebraic space of finite type over an al-
gebraically closed field k with affine diagonal equipped with an action of a linearly
reductive group G. Then the fixed locus X G is representable by a closed algebraic
subspace of X. If X is smooth, so is X G .

Proof. If G is connected and U → X is a G-invariant étale morphism, we claim that


UG /U

□ (6.8.3)
  
XG /X

is cartesian. Indeed, suppose S → U is a map such that S → U → X is G-invariant.

Let US → S be the base change of U → X by S → X. Since US → S is G-invariant,
it suffices to show that the section j : S → US is G-invariant. As U → X is étale,
j : S → US is an open immersion. Because G is connected, for each point s ∈ S, the
G-orbit Gj(s) ⊆ US is connected and thus contained in S.
Given a fixed point x ∈ X G (k), Theorem 6.7.24 produces a G-invariant étale
neighborhood (U, u) → (X, x) with U affine and u ∈ U G (k). If G is connected, then
U G → X G is étale and representable by (6.8.3), and it suffices to show that U G
is representable by a closed subscheme of U . Since U is affine, we can choose a
G-equivariant embedding U ,→ A(V ) into a finite dimensional G-representation. In
this case, A(V )G = A(V G ) and thus U G = U ∩ A(V )G ⊆ U is a closed subscheme is
affine. In general, let G0 ⊆ G be the connected component of the identity, and let
g1 , . . . , gn ∈ G(k) be representativesT
of the finitely many cosets G(k)/G0 (k). Then
G/G0 acts on X and then X = i (X G0 )gi is a closed subscheme of X.
G0 G

If in addition x is a smooth point of X, then u ∈ U is smooth and the Luna

map (6.7.9) gives a a G-invariant étale morphism U → TU,u . Since TU,u G
is a linear
subspace, it is smooth. Since U → X and U → TU,u are étale at u, then étale
G G G G

descent implies that x ∈ X G is smooth. See also [Ive72, Prop. 1.3] and [Mil17,
Thm. 13.1].

6.8.2 Limits under Gm -actions and attractor loci

Definition 6.8.4 (Limits). Given a Gm -action on an algebraic space U over a field
k and a point u ∈ U (k), we say that the limit limt→0 t · u exists if there exists an
extension of the diagram
t7→t·u/
Gm _ AU


A1 .

Valuative Criteria (3.8.7 imply that the limit is unique if X is separated and
that there exists a unique limit if X is proper.

Example 6.8.5. If X = P(V ) where V is a finite dimensional representation of G and

λ : Gm → G is a one-parameter subgroup, then we can choose a basis of V such that
λ(t) · (x1 , . . . , xn ) = (td1 x1 , . . . , tdn xn ) with d1 ≤ · · · ≤ dn . If d = min{di | xi ̸= 0},
then limt→0 λ(t) · [x0 : · · · : xn ] = [x′0 : · · · : x′n ] where x′i = xi for all i such that
di = d and is 0 otherwise.

373
Definition 6.8.6 (Attractor locus). Let X be a separated algebraic space of finite
type over k equipped with an action of Gm . Define the attractor locus as the functor
X + := MorG 1
k (A , X) : Sch/k → Sets
m

assigning a k-scheme S to the set Mork (S × A1 , X) of Gm -equivariant maps S × A1 →

X, where Gm acts trivially on S and with the usual scaling action on A1
Evaluation at 0 and 1 define morphism of functors
ev0 : X + → X Gm and ev1 : X → X.
On k-points, X + (k) is the set of points x ∈ X(k) such that limt→0 t · x exists, and
ev0 (x) is this limit. Since X is separated, the limit is unique if it exists. If X is
proper, the limit always exists and X + (k) = X(k). The functorial definition of X +
endows it with an interesting scheme-structure, e.g., when Gm acts on X = P1 via
t · [x : y] = [tx : y], then X = A
+ 1
`
{∞}.
Exercise 6.8.7 L(Affine case). If X = Spec A is affine, then the Gm -action induces a
Z-grading A = d∈Z Ad . Show that the functors X G P and X are
m +
Prepresentable
by the closed subschemes of X defined by the ideals d̸=0 Ad and d<0 Ad .
Example 6.8.8 (Centralizers and parabolics). Let G be an affine algebraic group
over an algebraically closed field k. A one-parameter subgroup λ : Gm → G induces
a Gm -action on G via conjugation t · g := λ(t)gλ(t)−1 . Under this action, the
fixed locus GGm = Cλ is identified with the centralizer of λ and the attractor locus
G+λ = Pλ is identified with the subgroup consisting of elements g ∈ G such that
limt→0 λ(t)gλ(t)−1 exists. The unipotent subgroup Uλ is identified with kernel of
ev0 : Pλ → Cλ . When G is reductive, Pλ ⊆ G is a parabolic subgroup or in other
words G/Pλ is projective. See §B.1.4 for more properties of these subgroups.
We say that a map X → Y is an affine fibration (resp., Zariski-local affine
fibration) if there exists an étale (resp., Zariski) cover {Yi → Y } such that X ×Y Yi ∼
=
AnYi over Yi . Since the transition functions are not required to be linear, this notion
is more general than a vector bundle.
Theorem 6.8.9. Let X be a separated algebraic space of finite type over an alge-
braically closed field k equipped with an action of Gm . The functor X + is representable
by an algebraic space of finite type over k, ev1 : X + → X is a monomorphism, and
ev0 : X + → X Gm is an affine morphism.
Assume in addition that X is smooth (resp., smooth scheme). Then X Gm is
also smooth and ev0 : X + → X Gm is an affine fibration (resp., Zariski-local affine
fibration), and in particular Xλ+ is smooth. If x ∈ X Gm and TX,x = T>0 ⊕ T0 ⊕ T<0
is the Gm -equivariant decomposition into nonnegative, zero, and nonpositive weights,
then TXi ,x = T0 ⊕ T>0 , TFi ,x = T0 , and Xi → Fi has relative dimension dim T>0 .
Proof. If X = Spec A is affine, then X Gm and X + are closed subschemes of X
(Exercise 6.8.7). In the special case that X = A(V ) where V is a finite dimensional
G-representation, then X Gm = A(V Gm ) and X + = A(V≥0 ) where V≥0 is the direct
sum of the nonnegative isotypic components, and moreover ev0 : X + → X Gm is a
relative affine space. We claim that if U → X is a Gm -invariant étale morphism,
then the diagram

U+
ev0
/ U Gm   /U

□ □ (6.8.10)
  
X+
ev0
/ X Gm   /X

374
is cartesian. The right square was verified in the proof of Theorem 6.8.2. For the
left square, we need to show that there exists a unique Gm -equivariant morphism
filling in a Gm -equivariant diagram

Spec k × S /: U

(6.8.11)
 
A1 × S /X

for every k-scheme S, where the vertical left arrow is the inclusion of the origin. By
uniqueness, we can assume that S is affine, and by Limit Methods (B.3), we can
further assume that S is of finite type over k. For each n ≥ 1, the Infinitesimal Lifting
Criterion for Étaleness (3.7.1) yields a unique Gm -equivariant map Spec k[x]/xn ×
S → U such that
Spec k × S 7/ U

 
Spec(k[x]/xn ) × S /X

commutes. As [A1 /Gm ] × S is coherently complete along BGm,S (Exercise 6.6.24),

Coherent Tannaka Duality in the form of Exercise 6.6.12 yields a unique Gm -
equivariant morphism A1 × S → U such that (6.8.11) commutes.
Choose a Gm -invariant étale surjective morphism U → X from an affine scheme
(Theorem 6.7.24). Then (6.8.10) implies that U + → X + is etale and representable,
and since U + is an affine scheme of finite type, it follows that X + is an algebraic
space of finite type. Since X is separated, ev1 : X + → X is a monomorphism. Since
U + → U Gm is affine, étale descent implies that ev0 : X + → X Gm is also affine. If
X is smooth, then X Gm is smooth by Theorem 6.8.2. As U is also smooth, for
each u ∈ U Gm (k), there is a Gm -equivariant morphism U → TU,u étale at u with
f (u) = 0 (Lemma 6.7.9). Then U + → TU,u +
is also étale at u. Let V ⊆ U + be the
open locus where U + → TU,u +
is étale. Since V is Gm -equivariant, if v ∈ V Gm , then
−1
ev0 (v) ⊆ V . Choosing an affine subscheme V ′ ⊆ V Gm containing u and replacing
U + with ev−1
0 (V ), we may assume that U
′ + +
→ TU,u is everywhere étale. By (6.8.10),
we have a cartesian diagram
+ o / X+
TU,u U+

□ □ (6.8.12)
  
Gm o / X Gm
TU,u U Gm

where the horizontal arrows are étale. With TX,x = T>0 ⊕ T0 ⊕ T<0 , there are
identifications TX,x
Gm
= T0 and TX,x
+
= T>0 ⊕ T0 . Since TU,u + Gm
→ TU,u a surjection
of vector spaces, U → U
+ Gm
is a Zariski-local affine fibration. By étale descent,
X + → X Gm is an affine fibration of relative dimension dim T>0 .
If X is a smooth`scheme, then by Sumihiro’s Theorem on Torus Actions (B.1.31),
we may choose U = i Ui → X such that {Ui } is a Gm -invariant affine open covering.
Then (6.8.12) implies that X + → X Gm is a Zariski-local affine fibration. See also
[Dri13, Prop. 1.2.2, Thm. 1.4.2] and [AHR20, Thm. 5.16].
Remark 6.8.13. For an alternative approach to the algebraicity of X + , consider
an algebraic stack X of finite type over k with affine diagonal and the stack

375
Mork ([A1 /Gm ], X ) of morphisms, whose objects over a k-scheme S are morphisms
[A1 /Gm ]S → X . It can be verified using Artin’s Axioms (C.7.4) that Mork ([A1 /Gm ], X )
is algebraic, where the crucial step is to verify the effectivity condition (AA5 ): this
follows from the coherent completeness of [A1 /Gm ]R , where R is a noetherian local
k-algebra, along the unique closed point (Theorem 6.6.13) together with Coherent
Tannaka Duality (6.6.9).
Taking X = [X/Gm ], a Gm -equivariant morphism A1 → X corresponds to a
morphism [A1 /Gm ] → [X/Gm ] over BGm (Exercise 3.1.16), and there is a cartesian
diagram
MorGm (A1 , X) / Mor ([A1 /Gm ], [X/Gm ])
k k

□
 
Spec k / Mor ([A1 /Gm ], BGm ).
k

Therefore, the algebraicity of the stacks of morphisms on the right imply that
k (A , X) is an algebraic space.
MorGm 1

6.8.3 The Białynicki-Birula Stratification

Theorem 6.8.14 (Białynicki-Birula Stratification4 ). Let X be a separated algebraic
space of`finite type over an algebraically closed field k with an action of Gm . Let
n
X Gm = i=1 Fi be the decomposition of the fixed locus into connected components
`i . Then there exists an affine morphism Xi → Fi for each i and a monomorphism
F
i Xi → X. Moreover,
`
(1) If X is proper, then i Xi → X is surjective.
(2) If X is smooth (resp., smooth scheme), then Fi is smooth and Xi → Fi is a
(resp., Zariski-local) affine fibration. If x ∈ Fi and TX,x = Tx,>0 ⊕ Tx,0 ⊕ Tx,<0
is the Gm -equivariant decomposition into nonnegative, zero, and nonpositive
weights, then TXi ,x = Tx,>0 ⊕ Tx,0 , TFi ,x = Tx,0 , and Xi → Fi has relative
dimension dim Tx,>0 .
(3) The map Xi ,→ X is a locally closed immersion under any of the following
conditions:
(a) X is affine,
(b) X is a smooth scheme, or
(c) there exists a Gm -equivariant locally closed immersion X ,→ P(V ) where
V is a Gm -representation (e.g., X is a normal quasi-projective variety).
+
` X is smooth, irreducible, and quasi-projective, then the stratification X =
(4) If
S i Xi is filterable, i.e., there is an ordering of the indices such that X≥i :=
j≥i Xj is closed for each i. If in addition there are finitely many fixed
points {x1 , . . . , xn }, then Txi ,0 = 0 and Xi = A(Txi ,>0 ) is an affine space; in
particular,
X + = X≥1 ⊇ X≥2 ⊇ · · · ⊇ X≥n ⊇ ∅
is a cell decomposition, i.e., each Xi = X≥i ∖ X≥i−1 is an affine space.
Proof. By Theorem 6.8.9, X + is representable by an algebraic space of finite type
over k, ev1 : X + → X is a monomorphism, and ev0 : X + → X Gm is affine. We define
4 This is frequently referred to as the ‘Białynicki-Birula Decomposition’ as some authors prefer

to reserve the term ‘stratification’ to a decomposition where each stratum has a neighborhood
which is topologically locally trivial.

376
Xi as the preimage ev−1 0 (Fi ). This gives the main statement, and the conclusion
of (2) also follows directly from Theorem 6.8.9. If X is proper, then X + → X is
surjective (i.e., (1) holds) as limt→0Lt · x exists for every x ∈ X(k).
For (3), if X = Spec A and A = d Ad is the Z-grading P induced by the Gm -action,
then X + is the closed subscheme defined by the ideal d<0 Ad (Exercise 6.8.7) and
in particular affine. If X is a smooth scheme, then there exists a Gm -invariant affine
open cover (Theorem B.1.30). For any point x ∈ X + , let x0 be the image of x under
ev0 : X + → X 0 , and choose a Gm -invariant affine open neighborhood U ⊆ X of x0 .
This induces a diagram
 / ev−1 (U ) / U
U + r 1 _ _
(6.8.15)
$  
X+
ev1
/ X.

Since U + → U is a closed immersion (as U is affine) and X + → X is separated (it is a

monomorphism), U + → ev−1 1 (U ) is a closed immersion. Since U
+
= X + ×X 0 U 0 (see
(6.8.10)), x ∈ U and U → X is an open immersion. In particular, U + ⊆ ev−1
+ + +
1 (U )
is an open and closed subscheme containing x. On the other hand, Xi is smooth
and connected (as Xi → Fi is an affine fibration), thus irreducible. It follows that
Xi ∩ U + = Xi ∩ ev−1 −1
1 (U ) and that Xi ∩ ev1 (U ) → U is a closed immersion which
in turn implies that Xi → X is a locally closed immersion. The final case (3)(c)
easily reduces to the case of X = P(V ) in which a direct calculation shows that each
Xi is of the form P(W ) ∖ P(W ′ ) for linear subspaces W ′ ⊆ W ⊆ V . See also [BB73,
Thm. 4.1], [Hes81, Thm. 4.5,p. 69], [Dri13, Thm. B.0.3], [AHR20, Thm. 5.27], and
[JS21, Thm. 1.5].
For (4), by Sumihiro’s Theorem on Linearizations (B.1.30), we can choose
a G-equivariant locally closed immersion X ,→ Pn , where Gm acts on Pn via
t · [x0 : · · · : xn ] = [td0 x0 : · · · : tdn xn ] with d0 ≤ · · · ≤ dn . Let D1 , . . . , Ds be the
distinct weights and set Ji = ` {j | dj = Di } so that J1 ∪ · · · ∪ Js is a partition of
{0, 1, . . . , n}. Then (Pn )Gm = i = 1s Fi where Fi = V (xj | j ∈ Ji ). The preimage
of Fi under the morphism ev0 : Pn → (Pn )Gm , given by p 7→ limt→0 t · p, is

xj = 0 for all j ∈ J1 ∪ · · · ∪ Ji−1

 
Pi := ev−1 (F ) = [x : · · · : x ] ,
0 i 0 n
xk ̸= 0 for some k ∈ Ji

and the union

[
P≥i := Pj = V (xk | k ∈ J1 ∪ · · · ∪ Ji−1 ) ⊆ Pn
j≥i

is closed. The fixed locus for X is X Gm = (Pn )Gm ∩ X = i Fi ∩ X. For each i, we

`
`li `li
write Fi ∩ X = j=1 Fij and Pi ∩ X = j=1 Xij as the irreducible decompositions.
Then ev0 : P → (P )
n n Gm
restricts to morphisms ev0 : Xij → Fij . For j ̸= k, the
strata Xij and Xik are disjoint, and thus X ij ∩ X ik ⊆ P≥i+1 ∩ X. It follows that

(P≥i+1 ∩ X) ∪ Xi1 ∪ · · · ∪ Xij ⊆ X

is closed for each j = 1, . . . , s. Ordering the strata as X11 , . . . , X1l1 , . . . , Xs1 , . . . , Xsl
establishes the claim. See also [BB76, Thm. 3].
Remark 6.8.16. It is not true in general that Xi ,→ X is a locally closed immersion.
Based on Hironaka’s example of a proper, non-projective, smooth 3-fold, Sommese

377
constructed a smooth algebraic space X such that Xi ,→ X is not a locally closed
immersion [Som82]. Konarski provided an example of a normal proper toric variety
X such that Xi ,→ X is not a locally closed immersion [Kon82].
Remark 6.8.17 (Morse stratifications). The Białynicki-Birula stratification of X can
be obtained as the Morse stratification corresponding to the non-degenerate Morse
function µ : X → Lie(S 1 )∨ = R: a point x ∈ X lies in Xi if only if the limit of its
forward trajectory under the gradient flow of µ lies in Fi . See [CS79].

Example 6.8.18. Suppose Gm acts on X = P2 via t · [x : y : z] = [x : ty : t2 ]. Then

X Gm = F1 ⨿ F2 ⨿ F3 where F1 = {[1 : 0 : 0]}, F2 = {[0 : 1 : 0]}, and F3 = {[0 : 0 : 1]},
and X1 = {x ̸= 0} = A2 , X2 = {[0 : y : z] | y ̸= 0} = A1 and X3 = F3 .
Let Xe be the blowup Blp X at the fixed point p = [0 : 1 : 0]. Then Gm acts on the
exceptional divisor E ∼= P1 via t · [u : v] = [u : t2 v] with fixed points q1 = [1 : 0] and
q2 = [0 : 1]. The fixed locus Xe Gm contains four points Fe1 = {[1 : 0 : 0]}, Fe2 = {q1 },
Fe3 = {q2 }, and Fe4 = {[0 : 0 : 1]}. We have that X e1 = X1 ∼ e2 = X2 ∼
= A2 , X = A1 ,
∼
X3 = E ∖ {q2 } = A , and X4 = X4 = F4 as illustrated in Figure 6.8.19. Observe
1 e e
that X 3 ∖ X3 = {q2 } is not the union of other strata.

Figure 6.8.19: Białynicki-Birula stratifications for P2 (left) and Blp P2 (right).

6.8.4 Applications of the Białynicki-Birula Stratification to

cohomology
Proposition 6.8.20. Let X be a smooth, irreducible, and quasi-projective scheme
over an algebraically closed field k with a Gm -action of with only finitely many fixed
points. Then the Chow group CHi (X) is a free Z-module generated by the closures
of the i-dimensional cells. If in addition k = C, then the cycle map CHi (X) →
HBM BM
2i (X, Z) to Borel–Moore homology is an isomorphism and H2i+1 (X, Z) = 0.

Remark 6.8.21. When X is compact (e.g., projective), then H2i

BM
(X, Z) is ordinary
integral singular homology.

Proof. The Białynicki-Birula Stratification (6.8.14(4)) implies that X has a cell

decomposition, and the statement follows from [Ful98, Ex. 19.1.11]. See also [Bri97,
§3.2].

Example 6.8.22 (Chow groups of Hilbn (A2 )). Let X = Hilbn (A2 ) be the Hilbert
scheme of n points; this is a smooth irreducible scheme (see 1.5.3). The natural
action of G2m induces a G2m -action on X. Under the Gm -action induced by Gm →
G2m , t 7→ (t, t), the evaluation map ev0 : X + → X is surjective, and since the Gm -fixed

378
points correspond to subschemes Z = V (I) ⊆ A2 supported at the origin where I is
a monomial ideal, there are only finitely many Gm -fixed points. We may therefore
use Proposition 6.8.20 to compute CH∗ (X).
For a monomial ideal I ⊆ R := k[x, y] and an integer i ≥ 0, define

ai := min{j | xi y j ∈ I},

and let r be the largest integer such that ar > 0. Then a0 ≥ · · · ≥ ar is a partition of
n and I = (y a0 , xy a1 , . . . , xr+1 ). We need to compute the dimension of the positive
weight space TI,>0 of the Gm -action on the tangent space

TI = HomR (I, R/I)

of X at the monomial ideal I, which is identified with HomR (I, R/I) by Exercise 1.5.6.
To accomplish this, we first argue that
aj −1
X X
TI = (χi−j−1
1 χa2 i −s−1 + χj−i
1 χ2
s−ai
), (6.8.23)
0≤i≤j≤r s=aj+1

as G2m representations, where χi : G2m → Gm denotes the one dimensional represen-

7 t−1
tation giving by (t1 , t2 ) → i . There are
X X
2(aj − aj+1 ) = 2 ai = 2n = dim TI
0≤i≤j≤r 0≤i≤r

one dimensional representations appearing on the right-hand side, and they are
linearly independent. It thus suffices to show that each of them occurs in TI . An
R-module map ϕ : I → R/I is given by the values ϕ(xi y ai ) subject to the relations

ϕ(xi+1 y ai ) = xϕ(xi ai ) and ϕ(xi y ai−1 ) = y ai−1 −ai ϕ(xi ai ).

Let 0 ≤ i ≤ j ≤ r and aj+1 ≤ s < aj . Defining

if l ≤ i
 l+j−i a +s−a
x y l i
ϕi,j,s : I → R/I, xl y al 7→
 l+i−j−1 a +s−a otherwise
0
l al x y l i
if l ≥ j + 1
ψi,j,s : I → R/I, x y 7→
0 otherwise,

one checks that ϕi,j,s and ψi,j,s are R-module maps that are eigenvectors for
χj−i
1 χ2
s−ai
and χi−j−1
1 χa2 i −s−1 . Thus (6.8.23) holds.
Choose λ = (λ1 , λ2 ) : Gm → G2m with λ1 ≫ λ2 . Under our sign conventions, a
character χa1 χb2 appearing in (6.8.23) has positive weight with respect to λ if a < 0,
or if a = 0 and b < 0. Thus
aj −1 r aj −1
X X X X
TI,>0 = χ1i−j−1 χa2 i −s−1 + χj−i
1 χ2
s−ai

0≤i≤j≤r s=aj+1 j=0 s=aj+1

and
r X
X r r
 X 
dim TI,>0 = (aj − aj+1 ) + (aj − aj+1 )
i=0 j=i j=0
r
X 
= ai + a0 = n + a0
i=0

379
Since there is a bijection between monomial ideals I ⊆ R = k[x, y] with dimk R/I = n
and partitions a0 ≥ · · · ≥ ar of n, for every d ≥ 0, the number of monomial ideals I
such that dim TI,>0 = d is equal to

P (2n − d, d − n) := # {partitions a1 ≥ · · · ≥ ar of 2n − d with each ai ≤ d − n}.

(6.8.24)
It follows from Proposition 6.8.20 that

dim CHd (Hilbn (A2 ))Q = P (2n − d, d − n).

See also [ES87, Thm. 1.1] and [Göt94, §2.2].

Exercise 6.8.25 (Chow groups of Hilbn (P2 )). Follow the above strategy to show
that the dth Betti number bd of Hilbn (P2 ) (or equivalently dim CHd (Hilbn (P2 ))) is
equal to
X X
bd = P (p, n0 − p)P (n1 )P (2n2 − r, r − n2 ),
n0 +n1 +n2 =n p+r=d−n1

where P (a) is the number of partitions of a and P (a, b) is defined by (6.8.24).

Remark 6.8.26. Göttsche used the Weil conjectures in [Göt90, Thm. 0.1] (see also
[Göt94, Thm. 2.3.10]) to show that for anyP smooth projective surface S over C or Fq
that the Poincaré polynomial p(S [n] , z) = i bi (S [n] )z n of S [n] := Hilbn (S) satisfies
∞ ∞
X
[n] n
Y (1 + z 2m−1 tm )b1 (S) (1 + z 2m+1 tm )b3 (S)
p(S , z)t =
n=0 m=1
(1 − z 2m−2 tm )b0 (S) (1 − z 2m tm )b2 (S) (1 − z 2m+2 tm )b4 (S)

In particular, the Betti numbers of S [n] only depend on the Betti numbers of S. While
each term p(S [n] , z) does not admit a particularly nice expression, the generating
function involving all n does. On the other hand, Nakajima constructed an action of
the Heisenberg algebra on H∗ (S [n] ), which can be used to recover the above formula
as well as additional properties of the cohomology ring [Nak97] (see also [Nak99b]).
The Białynicki-Birula Stratification can also be applied to compute equivariant
Chow rings CH∗G (X); see Exercise 7.7.12.

6.9 Valuative criteria for algebraic stacks: Θ- and

S-completeness
You might underestimate the importance of definitions in mathematics,
but you shouldn’t.

Claire Voisin

We discuss the valuative criteria of Θ- and S-completeness (see Definitions 6.9.10

and 6.9.14) which play a central role in the Existence Theorem of Good Moduli
Spaces (6.10.1) in the next section.

6.9.1 Maps from Θ = [A1 /Gm ]

We define the stack ‘Theta’ as

Θ := [A1 /Gm ]

380
over Spec Z.5 When we are working over a field k, we will abuse notation by also
using Θ to denote Θk = [A1k /Gm,k ]. While a map X → Θ from an algebraic stack
is classified by a line bundle and a section (Example 3.9.25), maps Θ → X from Θ
often also have geometric significance, e.g., in the case that X is a quotient stack,
a stack of coherent sheaves, and the stack of all curves. These descriptions will be
useful to interpret the valuative criteria of Θ- and S-completeness introduced in
§6.9.2.
Quotient stacks. Given a quotient stack [X/G], a one-parameter subgroup
λ : Gm → G, and a point x ∈ X(k) such that limt→0 λ(t) · x ∈ X exists, the
Gm -equivariant extension A1 → X induces a morphism [A1 /Gm ] → [X/G] of alge-
braic stacks. The next proposition asserts that the converse is also true, i.e., any map
[A1 /Gm ] → [X/G] is induced by a one-parameter subgroup λ and a point x ∈ X(k).
Recall that Cλ denotes the centralizer of a one-parameter subgroup λ and Pλ is
the subgroup of elements g ∈ G such that limt→0 λ(t)gλ(t)−1 exists; see §B.1.4.

Proposition 6.9.1. If G is a smooth affine algebraic group over an algebraically

closed field k acting on a separated algebraic space X of finite type over k, then there
is an equivalence of groupoids
∼ 
Mork (Θ, [X/G]) → (x ∈ X(k), λ : Gm → G) | lim λ(t) · x ∈ X(k) exists ,
t→0

where a morphism (x, λ) → (x′ , λ′ ) is an isomorphism class of a pair (g, h) with

g ∈ Pλ (k) and h ∈ G(k) such that x′ = hgx and λ′ = hλh−1 , with (g, h) ∼ (c−1 g, hc)
for c ∈ Cλ . Under this correspondence, the morphism Θ → [X/G] sends 1 to x and
0 to limt→0 λ(t) · x.

Remark 6.9.2. Observe that the automorphism group of (x, λ) is Pλ ∩ Gx .

Proof. Given (x, λ), the Gm -equivariant map mx,λ : Gm → X defined by t 7→ λ(t) · x
extends to a commutative diagram
mx,λ
Gm _ /X
>

 m
e x,λ
A1

The extension is Gm -equivariant and induces a morphism of quotient stacks fx,λ : Θ →

[X/G]. We first show that this defines a functor

(x, λ) | lim λ(t) · x exists → Mork (Θ, [X/G])


t→0
(6.9.3)
(x, λ) 7→ fx,λ .
∼
Given a morphism (g, h) : (x, λ) → (x′ , λ′ ), we need to define a 2-morphism fx,λ →
∼
fx′ ,λ′ . Since h determines a canonical isomorphism fx′ ,λ′ → fh−1 x′ ,λ , it suffices to
∼
define a 2-morphism fx,λ → fh−1 x′ ,λ . Since g ∈ Pλ (k), the map t 7→ λ(t)gλ(t)−1
extends to a map ge : A → G such that m
1
e x,λ (as h−1 x′ = gx).
e h−1 x′ ,λ = ge · m
∼
The element ge defines an isomorphism fx,λ → fh−1 x′ ,λ . For c ∈ Cλ , the pairs
(g, h) and (c−1 g, hc) define the same isomorphism: indeed this follows from the
observation that if c ∈ Gx , then (c−1 , c) defines the identify automorphism of fx,λ .
5 The symbol Θ is used as it resembles the picture of the two orbits of Gm on the complex plane.

381
 ∼
Conversely, any isomorphism fx,λ → fx′ ,λ is an induced by element ge ∈ G(A1 )
satisfying m e x,λ , and we conclude that (6.9.3) is a fully faithful functor.
e x′ ,λ = ge · m
For the essential surjectivity of (6.9.3), let f : Θ → [X/G] be a morphism. In the
fiber diagram
P /X

□
 
Θ
f
/ [X/G],

P → Θ is a principal G-bundle. The restriction P|BGm along the unique closed

point 0 ∈ Θ corresponds to a Gm -equivariant principal G-bundle P on Spec k. After
choosing an isomorphism P ∼ = G, P corresponds to a one-parameter subgroup
λ : Gm → G. This induces a Gm -action on the product A1 × G by t · (x, g) =
(tx, gλ(t)−1 ) and thus a principal G-bundle

Pλ := [(A1 × G)/Gm ] → Θ.

We claim that there is an isomorphism α : P → Pλ of principal G-bundles. By con-

struction, we have an isomorphism α0 : P|BGm → Pλ |BGm . Since IsomΘ (P, Pλ ) → Θ
is smooth (as it is a principal G-bundle and G is smooth), we may use deformation
theory (Proposition 6.7.12) to construct compatible isomorphisms αn : P|Xn → Pλ |Xn
over the nilpotent thickenings Xn of BGm ,→ Θ. Coherent Tannaka Duality (6.6.9)
coupled with the coherent completeness of Θ along BGm (Theorem 6.6.13) implies
that the isomorphisms αn extend to an isomorphism α : P → Pλ . Restricting the
composition
α−1
A1 × G → Pλ −−→ P → X
to the identity in G yields a Gm -equivariant morphism A1 → X, which in turn
induces a map f : Θ → [X/G] Letting x ∈ X(k) be the image of 1, one checks that
fx,λ is 2-isomorphic to f .
Exercise 6.9.4 (details). Letting R be a complete noetherian local k-algebra,
upgrade the proposition to an equivalence between Mork (ΘR , [X/G]) and pairs
(x ∈ X(R), λ : Gm → G).
Remark 6.9.5. Proposition 6.9.1 can be further upgraded to a description of the
stack of morphisms from [A1 /Gm ] to [X/G]. Namely, there is a decomposition

Mor([A1 /Gm ], [X/G]) ∼

a
= [Xλ+ /Pλ ]
λ

where λ varies over conjugacy classes of one-parameter subgroups. The algebraicity

of this stack was discussed already in Remark 6.8.13. See also [HL14, Thm. 1.37].

Stacks of coherent sheaves. Given a projective scheme X, recall that Coh(X)

denotes the algebraic stack of coherent sheaves on X (see Exercise 3.1.23).
Proposition 6.9.6. Let X be a projective scheme over an algebraically closed field
k. For a noetherian k-algebra R, Mork (ΘR , Coh(X)) is equivalent to the groupoid
of pairs (E, E• ) where E is a coherent sheaf on XR flat over R and

E• : 0 ⊆ · · · ⊆ Ei−1 ⊆ Ei ⊆ Ei+1 ⊆ · · · ⊆ E

is a Z-graded filtration such that Ei = 0 for i ≪ 0, Ei = E for i ≫ 0, and each factor

Ei /Ei−1 is flat over R. A morphism (E, E• ) → (E ′ , E•′ ) is an isomorphism E → E ′

382
of coherent sheaves compatible with the filtration. Under this correspondence, the
morphism
L ΘR → Coh(X) sends 1 to E and 0 to the associated graded gr E• :=
i Ei /Ei−1 , and factors through Bun(X) ⊆ Coh(X) if and only if E and each
factor Ei /Ei−1 is a vector bundle.

Proof. The reader may prefer to read the proof in the special (but important) case
that R = k. A morphism ΘR → Coh(X) corresponds to a coherent sheaf F on
X × ΘR flat over ΘR , or alternatively using smooth descent to a coherent sheaf on
X × A1R flat over A1R . Since f : X × ΘR → X × BGm,R is an affine morphism, F can
be recovered from the data of the quasi-coherent sheaf f∗ F on X × BGm,R together
with the structure of a f∗ OX×ΘR -module. Using that quasi-coherent sheaves on
X × BGm,R are Z-graded quasi-coherent sheaves on XR (see Exercise 6.1.10), it
follows
L that F corresponds to a Z-graded OXR [x]-module flat over R[x]. Writing
F = i∈Z Ei with each Ei a coherent sheaf on XR , then multiplication by x induces
maps x : Ei → Ei+1 which are necessarily injective as F is flat over R[x], hence
torsion free. Since F is finitely generated as a Z-graded R[x]-module, there exists
finitely many homogeneous generators with bounded degree. ThusL Ei = E for i ≫ 0.
On the other hand, considering the OXR [x]-submodule E≥d := i≥d Ei ⊆ F , the
ascending chain · · · ⊆ E≥d ⊆ E≥d−1 ⊆ · · · ⊆ F must terminate as F is noetherian.
It followsLthat Ei = 0 for i ≪ 0. Since F is flat as an R[x]-module, the quotient
F/xF = i Ei /Ei−1 is flat as an R-module and thus each factor Ei /Ei−1 is flat
over R.
Conversely, given E and a filtration
L E• satisfying the above conditions, consider
the graded OXR [x]-module F := i Ei , which is often
L referred to as the Rees
construction. We will show by induction that E≥d := i≥d Ei is flat and finitely
generated over R[x]; this implies that F is flat and finitely generated over R[x]
since Ei = 0 for i ≪ 0. For d ≫ 0, E≥d is isomorphic to the graded R[x]-module
(E ⊗R R[x])⟨d⟩, where ⟨d⟩ denotes the grading shift, and is thus flat and finitely
generated. For every d, we have an exact sequence

0 → (Ed ⊗R R[x])⟨d⟩ → E≥d → ((Ed+1 /Ed ) ⊗R R[x])⟨d + 1⟩ → 0.

The flatness of E and the quotients Ed+1 /Ed implies the flatness of each Ed . Thus
the left and right term above are flat and finitely generated as R[x]-modules, and
thus so is the middle term.

Stack of all curves.

Proposition 6.9.7. Let Mall g be the algebraic stack of all proper curves (Theo-
rem 5.4.6) over an algebraically closed field k. For every k-algebra R, Mork (ΘR , Mall
g )
is the groupoid whose objects are Gm -equivariant families of proper curves C → A1R ,
where Gm acts on A1R with the usual scaling action. Morphisms are Gm -equivariant
morphisms.

Proof. The statement follows from smooth descent applied to A1R → ΘR .

Remark 6.9.8. A similar description holds for other moduli stacks of varieties. These
Gm -equivariant families are often called test configurations in the literature.

6.9.2 The valuative criteria: Θ- and S-completeness

We maintain the notation that Θ := [A1 /Gm ] over Spec Z. Let R be a DVR with
fraction field K and residue field κ. Define ΘR := Θ × Spec R and set 0 ∈ ΘR to

383
be the unique closed point. Observe that ΘR is a local model of the quotient stack
[A2 /Gm ] with weights 0, 1 as it is identified with the base change of the good moduli
space [A2 /Gm ] → Spec k[x] along the map Spec R → Spec k[x] where x maps to a
uniformizer π in R. The following cartesian diagram gives a schematic picture of
ΘR (where x is the coordinate on A1 and π ∈ R is the uniformizer).

Spec
: q
R
M m m,Rd
BG
x̸=0 x=0

, # { 2R
Spec K r lL m,κ (6.9.9)
; ΘR c BG

% - π̸=0 π=0
1Q y
ΘK Θκ

where the maps to the left are open immersions

Sand to the right are closed immersions.
In particular, a morphism ΘR ∖ 0 = Spec R Spec K ΘK → X to an algebraic stack
is the data of morphisms Spec R → X and ΘK → X together with an isomorphism
of their restrictions to Spec K.

Definition 6.9.10. A noetherian algebraic stack X is Θ-complete 6 DVR R, every

commutative diagram

ΘR ∖ 0 /X
;
(6.9.11)

ΘR

of solid arrows can be uniquely filled in.

Remark 6.9.12. Equivalently, using the stack Mor(Θ, X ) classifying morphisms

Θ → X , evaluation at 1 gives a morphism

ev1 : Mor(Θ, X ) → X , f 7→ f (1),

X is Θ-complete if and only if ev1 satisfies the valuative criterion for properness.
If X is of finite type over an algebraically closed field k, then Mor(Θ, X ) is an
algebraic stack locally of finite type over k; see Remark 6.9.5 where an explicit
description is given when X is a quotient stack. The stack Mor(Θ, X ) is however
rarely quasi-compact, e.g., for X = BGm , and ev1 is thus rarely proper.
For a DVR R with fraction field K, residue field κ, and uniformizer π, we define

(6.9.13)


ϕR := [Spec R[s, t]/(st − π) /Gm ],

where s and t have Gm -weights 1 and −1 respectively.7 The quotient ϕR is a local

model of the quotient stack [A2 /Gm ] with weights 1, −1 as it is identified with
the base change of the good moduli space [A2 /Gm ] → Spec k[xy] along the map
Spec R → Spec k[xy] given by xy 7→ π.
6 This notion was introduced in [HL14], where it was called ‘Θ-reductive’.
7 The symbol ϕ is used because it looks like the non-separated affine line with an additional origin.
This stack was introduced in [Hei17, §2.B] using the S notation STR , as it is a compactification of
the ‘standard test’ scheme STR = STR ∖ 0 = Spec R Spec K Spec R for for separatedness.

384
 The locus where s ̸= 0 in ϕR is isomorphic to [Spec R[s, t]s /(t − π/s) /Gm ] ∼


=
[Spec(R[s]s )/Gm ] ∼
= Spec R and the locus where t ̸= 0 has a similar description. We
thus have cartesian diagrams analogous to (6.9.9)

Spec
9 q
R N n Θκ c
s̸=0 s=0

, # } 1Q
Spec K r ; ϕR a mM m,κ
BG

% - t̸=0 t=0
0P {
Spec R Θκ

where the maps to the left are open immersions S and to the right are closed immersions.
In particular, a morphism ϕR ∖ 0 = Spec R Spec K Spec R → X to an algebraic
stack is the data of two morphisms ξ, ξ ′ : Spec R → X together with an isomorphism
ξK ≃ ξK ′
over Spec K.

Definition 6.9.14. A noetherian algebraic stack X is S-complete if for every DVR

R, every commutative diagram

ϕR ∖ 0 /X
;
(6.9.15)

ϕR

of solid arrows can be uniquely filled in.8

The definition of Θ-completeness and S-completeness naturally extends to mor-

phisms f : X → Y, but we will not use this.

Exercise 6.9.16. Let X be a k-scheme and R be a DVR over k.

(1) Show that a quasi-coherent sheaf F on X × ϕR corresponds to a Z-graded
quasi-coherent sheaf n∈Z Fn on XR together with a diagram of maps
L

s
, s
*
s
, s
*
··· j Fn−1 l Fn j Fn+1 l ···
t t t t

such that st = ts = π.
(2) Show that F is coherent if each Fn is coherent, s : Fn−1 → Fn is an isomor-
phism for n ≫ 0 and t : Fn → Fn−1 is an isomorphism for n ≪ 0. Show that
the converse holds if the maps s and t are injective.
(3) Show that F is flat over ϕR if and only if the maps s and t are injective, and
the induced map t : Fn+1 /sFn → Fn /sFn−1 is injective.
(4) Assuming that F is coherent and flat over ϕR , characterize the restrictions of
F as follows:
s s s
(i) F|{s̸=0} ∼
= colim(· · · −
→ Fn − → · · · ) on XR ,
→ Fn+1 −
t t t
(ii) F|{t̸=0} ∼
= colim(· · · →
− Fn+1 → − · · · ) on XR
− Fn →
8 The ‘S’ stands for ‘Seshadri’ as S-completeness is a geometric property reminiscent of how the

S-equivalence relation on sheaves implies separatedness of the moduli space.

385
 (iii) F|{s=0} on Θκ corresponds under Proposition 6.9.6 to the Z-filtration
t t t
··· →
− Fn+1 /sFn+2 → − · · · , and
− Fn /sFn−1 →
s s
(iv) F|{t=0} on Θκ corresponds to the Z-filtration · · · −
→ Fn /tFn+1 −
→
s
→ ···.
Fn+1 /tFn+2 −

6.9.3 Properties of Θ- and S-completeness

Lemma 6.9.17. A noetherian algebraic stack with affine diagonal is Θ-complete
(resp., S-complete), if and only if every diagram (6.9.11) (resp., (6.9.15)), there
exists a lift after an extension of DVRs R ⊆ R′ . In particular, Θ- and S-completeness
can be verified on complete DVRs with algebraically closed residue fields.

Proof. We begin with the observation that if X → Y has affine diagonal and
j : U → T is an open immersion of algebraic stacks over Y with j∗ OU = OT ,
then two extensions f1 , f2 : T → X of a Y-morphism U → X are canonically 2-
isomorphic. Indeed, since IsomT (f1 , f2 ) → T is affine, the section over U induced
∼
by the 2-isomorphism f1 |U → f2 |U extends uniquely to a section of T .
Consider a diagram (6.9.11), an extension of DVRs R ⊆ R′ , and a lifting ΘR′ →
X . The open immersion j : ΘR ∖ 0 → ΘR satisfies j∗ OΘR ∖0 = OΘR and by Flat
Base Change (6.1.15) the same property holds for the morphisms obtained by base
changing j along ΘR′ → ΘR , ΘR′ ×ΘR ΘR′ → ΘR , and ΘR′ ×ΘR ΘR′ ×ΘR ΘR′ → ΘR .
By the above observation, there exists a canonical 2-isomorphism between the two
extensions ΘR′ ×ΘR ΘR′ ⇒ ΘR′ → X which necessarily satisfies the cocycle condition.
By fpqc descent, the lifting ΘR′ → X descends to a lifting ΘR → X . The same
argument works for S-completeness.

Remark 6.9.18. It is also true that when X is of finite type over k, these criteria
can be verified on DVRs essentially of finite type over k; see [AHLH23, §4]. We will
not use this fact.

Lemma 6.9.19. Let f : X → Y be an affine morphism of noetherian algebraic

stacks. If Y is Θ-complete (resp., S-complete), so is X .

Proof. Since ΘR is regular and 0 ∈ ΘR is codimension 2, the pushforward of the

structure sheaf along ΘR ∖ 0 → ΘR is the structure sheaf. We therefore have
canonical equivalences

MorY (ΘR ∖ 0, X ) ∼
= MorOY −alg (f∗ OX , (ΘR ∖ 0 → Y)∗ OΘR ∖0 )
∼
= MorO −alg (f∗ OX , (ΘR → Y)∗ OΘ )
Y R
∼
= MorY (ΘR , X ).

The case of S-completeness is identical.

Proposition 6.9.20. If G is a reductive group over an algebraically closed field k,

then every quotient stack [Spec A/G] is Θ-complete and S-complete.

Proof. We first show that BGLn is Θ-complete. A morphism ΘR ∖ 0 → X corre-

sponds to a vector bundle E on ΘR ∖ 0. The algebraic stack ΘR is regular and
0 ∈ ΘR is a codimension 2 point. If Ee is a coherent sheaf on ΘR extending E, then
the double dual Ee ∨∨ is a vector bundle extending E. (In fact, the pushforward of
E along ΘR ∖ 0 ,→ ΘR is a vector bundle.) This provides the desired extension
ΘR → X . As G is affine, we can choose a faithful representation G ⊆ GLn . As G is

386
reductive, the quotient GLn /G is affine by Matsushima’s Theorem (B.1.45). Using
the cartesian diagram
GLn /G / Spec k

□
 
BG / BGLn

and smooth descent, we see that BG → BGLn is affine. We conclude that BG and
[Spec A/G] are Θ-complete by Lemma 6.9.19.

For a ring R, the map ΘR → Spec R is a good moduli space, and thus every map
ΘA → X to an algebraic space factors through Spec R by the universality of good
moduli spaces (Theorem 6.5.6(4)). A similar result holds if X is a Deligne–Mumford
stack.

Lemma 6.9.21. Let X be a noetherian algebraic stack with affine and quasi-finite
diagonal. If R is a complete DVR, every map ΘR → X (resp., ϕR → X ) factors
through ΘR → Spec R (resp., ϕR → Spec R).

Proof. We will reduce to the case when X is affine, where the statement follow easily
from the fact that Γ(ΘR , OΘR ) = Γ(ϕR , OϕR ) = R. Let κ = R/m and x ∈ X (κ)
be the image of 0 ∈ ΘR . Since Gm has no non-trivial finite quotients, the induced
map Gm → Gx on stabilizers is trivial. By Proposition 4.3.4, we may find a smooth
presentation U → X from an affine scheme together with a lift u ∈ U (κ) of x.
The map BGm,κ → X factors through u : Spec κ → U and thus lifts to a map
u
BGm,κ → Spec κ − → U . Letting Tn be the nth nilpotent thickening of BGm,κ ,→ ΘR ,
deformation theory (Proposition 6.7.12) implies that we may find compatible lifts
Tn → U of Tn ,→ ΘR → X . By Coherent Tannaka Duality (6.6.9), there is an
extension ΘR → U . Since ΘR → U factors through Spec R, so does ΘR → X .

Proposition 6.9.22. Every noetherian algebraic stack X with affine and quasi-
finite diagonal (e.g., a Deligne–Mumford stack with affine diagonal) is Θ-complete.
Moreover, X is S-complete if and only if it is separated.

Proof. By Lemma 6.9.17, Θ-completeness and S-completeness can be tested on a

complete DVR R. Lemma 6.9.21 implies that X is Θ-complete and also implies that
X is S-complete if only if every diagram

/7 X
S
Spec R Spec K Spec R


Spec R

has a lift, which is the usual valuative criterion for separatedness.

Θ- and S-completeness are necessary for the existence of a separated good moduli
space.

Proposition 6.9.23. Let X be an algebraic stack of finite type over an algebraically

closed field k with affine diagonal. If π : X → X be a good moduli space, then X is
Θ-complete. Moreover, X is S-complete if and only if X is separated.

387
Proof. Since X is Θ-complete, to see that X is Θ-complete, it suffices to find a lift
of every commutative diagram

ΘR ∖ 0 /X
;
π
 
Spec R /X

of solid arrows. By the Local Structure for Good Moduli Spaces (6.7.3), there
exists an étale morphism Spec B → X containing the image of Spec R such that
X ×X Spec B ∼ = [Spec A/G] with G linearly reductive and B = AG . Since Spec R →
X lifts to Spec B after an extension of DVRs and since Θ-completeness can be
checked after an extension (Lemma 6.9.17), we are reduced to the case of [Spec A/G].
This is Proposition 6.9.20.
If X is separated, then X is S-complete as ϕR ∖ 0 = Spec R ∪Spec K Spec R → X
factors through Spec R by the valuative criterion for separatedness. The same
argument as above shows that X is S-complete. Conversely, suppose f, g : Spec R →
X are two maps such that f |K = g|K . After possibly an extension of R, we may
choose a lift Spec K → X of f |K = g|K . Since X → X is universally closed
(Theorem 6.5.6(1)), after possibly further extensions of R, we may choose lifts
fe, ge : Spec R → X of f, g such that fe|K ∼ = ge|K by the Valuative Criterion for
Universal Closedness (3.8.7). Since X is S-complete, we can extend fe and ge to a
morphism ϕR → X . As ϕR → Spec R is a good moduli space and hence universal
for maps to algebraic spaces, the morphism ϕR → X descends to a unique morphism
Spec R → X which necessarily must be equal to both f and g. We conclude that X
is separated by the Valuative Criterion for Separatedness.

6.9.4 Examples of Θ- and S-completeness

We discuss Θ- and S-completeness for quotient stacks, stacks of coherent sheaves,
and the stack of all curves.

Quotient stacks. By Proposition 6.9.1, a map Θ → [U/G] is classified by a point

u ∈ U and a one-parameter subgroup λ : Gm → G such that limt→0 λ(t) · u exists.
We apply this to provide a geometric characterization of Θ-completeness for quotient
stacks. Under the Gm -action induced by λ, the attractor locus Uλ+ is defined as the
algebraic space representing the functor MorG k (A , U ) (see Theorem 6.8.9). The
m 1

evaluation map ev1 : Uλ → U is defined by sending f : A1 → U to f (1).

+

Proposition 6.9.24. Let G be a smooth linearly reductive group over an algebraically

closed field k, and U be a separated algebraic space of finite type over k with an
action of G. Then

[U/G] is Θ-complete ⇐⇒ for every map u : Spec R → U from a complete

DVR over k with algebraically closed residue field
and one-parameter subgroup λ : Gm → G such that
limt→0 λ(t) · uK ∈ U (K) exists, then
limt→0 λ(t) · u ∈ U (R) also exists;
⇐⇒ for every one-parameter subgroup λ : Gm → G,
the morphism ev1 : Uλ+ → U is a closed immersion.

388
Proof. Since G is linearly reductive, BG is Θ-complete (Proposition 6.9.20). There-
fore Θ-completeness of [U/G] is equivalent to the existence of a lift in every diagram

_ 0
ΘR ∖ / [U/G]
:
(6.9.25)
 
ΘR / BG

where R is a complete DVR with algebraically closed residue field (Lemma 6.9.17).
By Proposition 6.9.1 and Exercise 6.9.4, the map ΘR → BG corresponds to a
one-parameter subgroup λ : Gm → G while ΘR ∖ 0 → [U/G] corresponds to a map
u : Spec R → U such that limt→0 λ(t) · uK ∈ U (K) exists. In other words, we have
a commutative diagram

Spec K / U+
; λ

ev1 (6.9.26)
 
Spec R
u /U

of solid arrows. A lift of (6.9.25) corresponds to the existence of limt→0 λ(t)·u ∈ U (R)
or equivalently to a lift of (6.9.26). Since ev1 : Uλ+ → U is a monomorphism of finite
type, it is closed immersion if and only if it is proper if and only if satisfies the
existence part of the valuative criterion.

Example 6.9.27. When L U = Spec A is affine, a one-parameter subgroup λ : Gm →

G induces a grading A = d∈Z Ad , and Uλ+ is representable by V ( d<0 Ad ) (Exer-
P

cise 6.8.7). We see thus that ev1 : Uλ+ ,→ U is a closed immersion; this recovers the
fact that [U/G] is Θ-complete (Proposition 6.9.20).

Example 6.9.28. Consider the ` Gm -action on P given 1by t · [x : y] = [tx : y].

1

Taking λ = id, then (P )λ = A

1 + 1
{∞}, and we see that [P /Gm ] is not Θ-complete.
Similarly, the quotient stack [C/Gm ] of the nodal cubic C is not Θ-complete: letting
P1 → C be the normalization where the fiber over the node is {0, ∞}, then Cλ+ =
P1 ∖ ∞ for λ = id. Finally, if Gm acts on X = A2 ∖ 0 via t · (x, y) = (tx, y) then
[X/Gm ] is not Θ-complete as Xλ+ = {y ̸= 0}.

Remark 6.9.29. Consider the algebraic stack Mor(Θ, [X/G]) of morphisms, which
decomposes as a disjoint union λ [Xλ+ /Pλ ], where λ varies over conjugation classes
`
of one-parameter subgroups λ : Gm → G (Remark 6.9.5). The evaluation morphism
ev0 : [Xλ+ /Pλ ] → [X/G] is induced by the inclusion Xλ+ → X. The Θ-completeness
of [X/G] is therefore equivalent to the properness of the maps [Xλ+ /Pλ ] → [X/G].
It is also possible to give a criterion for when [U/G] is S-complete in terms of one-
parameter subgroups λ : Gm → G and properties of the morphism MorG 2
A1 (A , U ×
m

A ) → U × U × A , where the maps to U are obtained by restricting along the two

1 1

maps A1 → A2 given by x 7→ (x, 1) and x 7→ (1, x). We resist giving this criterion
here as it is more involved and not needed for later.
Stacks of coherent sheaves. Recall that maps Θ → Coh(X) correspond to
filtrations (Proposition 6.9.6).

Proposition 6.9.30. For every projective scheme X over an algebraically closed

field k, the algebraic stack Coh(X) is Θ- and S-complete.

389
Proof. Given a DVR R, Proposition 6.9.6 implies that a map ΘR ∖ 0 → Coh(X)
corresponds to a coherent sheaf E on XR flat over R and a Z-graded filtration
F• : ⊆ · · · Fi−1 ⊆ Fi ⊆ · · · ⊆ EK such that Fi = EK for i ≫ 0, Fi = 0 for i ≪ 0,
and Fi /Fi−1 is flat over R. Viewing E as a subsheaf of EK , we define Ei := Fi ∩ E
as the intersection in EK . Since Ei /Ei−1 is a subsheaf of Fi /Fi−1 , it is torsion free,
hence flat as an R-module. The filtration E• defines an extension ΘR → Coh(X).
(Aside: this is exactly the argument for the valuative criterion of properness of the
Quot scheme (Proposition 1.4.2). Note also that if we let j : ΘR ∖ 0 ,→ ΘR , let
jx : Spec R ,→ ΘR and let jπ : ΘK → ΘR denote the open immersions, and let E
be the coherent sheaf on C × (ΘR ∖ 0) denoting the union of E and F• , then the
extension is given by (id ×j)∗ E = jx,∗ E ∩ jπ,∗ F• = E[x±1 ] ∩ F• = E• , where E[x±1 ]
is the Z-graded filtration given by placing E in every degree.)
For S-completeness, suppose we are given a map ϕR ∖0 → Coh(X) corresponding
to coherent sheaves E and F flat over R and an isomorphism α : EK → FK . Recalling
the quotient presentation ϕR = [Spec(R[s, t]/(st − π))/Gm ], we have several natural
open immersions: j : ϕR ∖ 0 ,→ ϕR , js , jt : Spec R ,→ ϕR (with s ̸= 0 and t = ̸ 0),
and jst : Spec K → ϕR (with st ̸= 0). We compute the pushforward as the equalizer

0 / (id ×j)∗ E / (id ×js )∗ E ⊕ (id ×jt )∗ F / (id ×jst )∗ FK

(a, b)  / a − α(b).

The pushforwards can be computed as graded modules over R[s, t]/(st − π):
M
(id ×jst )∗ FK = FK ⊗R R[t±1 ] = FK tn ,
n∈Z
M
±1
(id ×js )∗ E = E ⊗R R[t ]= Etn ,
n∈Z

]∼
M
±1
(id ×jt )∗ F = F ⊗R R[s = (π −n · F )tn ⊆ (id ×jst )∗ FK
n∈Z

where we have used that s = t−1 π. Thus

j∗ E ∼
M
E ∩ (π −n · F ) tn ⊆ (id ×jst )∗ FK .


=
n∈Z

Each R-module E ∩ (π −n · F ) ⊆ E is finitely generated since E is. Since the

ascending chain · · · ⊆ E ∩ (π −n · F ) ⊆ E ∩ (π −n−1 · F ) ⊆ · · · terminates to E, j∗ E
is coherent. To show that j∗ E is flat over ϕR , we only need to check that it is flat
over 0. By the Local Criterion for Flatness (Theorem A.2.5), we need to show that
TorA1 (A/m, j∗ E) = 0 where A = R[s, t]/(st − π) and m = (s, t). The Koszul complex
gives a resolution of the residue field κ = A/m = R/π:
(t,−s) (s,t)
0 → A −−−−→ A ⊕ A −−−→ A → κ → 0.

Tensoring with j∗ E yields a complex

(t,−s) (s,t)
0 → j∗ E −−−−→ j∗ E ⊕ j∗ E −−−→ j∗ E. (6.9.31)

The pushforward of the exact sequence

(t,−s) (s,t)
0 → OϕR ∖0 −−−−→ OϕR ∖0 ⊕ OϕR ∖0 −−−→ OϕR ∖0 → 0

390
along id ×j : C × ϕR ∖ 0 ,→ C × ϕR is a left exact sequence of vector bundles, and
tensoring with j∗ E yields a left exact sequence, which identified with (6.9.31). Thus
TorA
1 (A/m, j∗ E) = 0.

In Proposition 8.3.6, we provide a criterion for when an open substack of Coh(X)

is Θ- and S-complete. For X = C is a smooth curve, we apply this criterion in
Proposition 8.3.8 to establish the Θ- and S-completeness of Bun ss r,d , from which we
conclude the existence of a good moduli space.
Stack of all curves. The stacks Mg and Mg of smooth and stable curves are both
Θ- and S-complete as they are separated Deligne–Mumford stacks. While maps from
Θ to the stack of all curves correspond to test configurations (Proposition 6.9.7),
there is unfortunately no known simple criteria—similar to the above criteria for
quotient stacks and stacks of coherent sheaves—to verify whether a given substack
of the stack Mall
g of all curves is Θ-complete or S-complete.

6.10 Existence of good moduli spaces

I believe that GIT did not fulfill all the early expectations. The first
successes leading to the construction of the moduli spaces of curves,
Abelian varieties, vector bundles and sheaves were not carried much
further. For instance, GIT never came up with a good approach to
compactify the moduli space of surfaces of general type, and it did not
tackle the moduli problem for higher dimensional varieties.

János Kollár [Kol06, pp. 112-3]

We prove necessary and sufficient conditions for the existence of a separated

good moduli space in characteristic 0.
Theorem 6.10.1 (Existence Theorem of Good Moduli Spaces). Let X be an
algebraic stack, of finite type over an algebraic closed field k of characteristic 0, with
affine diagonal. There exists a good moduli space π : X → X with X a separated
algebraic space if and only if X is Θ-complete (Definition 6.9.10) and S-complete
(Definition 6.9.14). Moreover, X is proper if and only if X satisfies the existence
part of the valuative criterion for properness.

6.10.1 Strategy for constructing good moduli spaces

The Local Structure Theorem for Algebraic Stacks (6.7.1) gives us a natural strategy
to construct the good moduli space X. For each closed point x ∈ X , we have an
étale quotient presentation

W = [Spec A/Gx ]
f
/X


W = Spec AGx

where f is affine and étale, and there is a preimage w ∈ W of x such that f induces
an isomorphism of stabilizer groups at w. We want to show that the GIT quotients
W = Spec AGx , as x ranges over closed points, provide étale models that can be
glued to a good moduli space of X . To this end, we need to construct an étale

391
equivalence relation on W . Since f is affine, the fiber product R := W ×X W is
isomorphic to a quotient stack [Spec B/Gx ] and we have a diagram

R
p1
// W f
/X
p2

 
R
q1
// W
q2

where R = Spec B Gx . If q1 , q2 : R ⇒ W defines an étale equivalence relation, the

algebraic space quotient W/R gives a candidate for a good moduli space of f (W) ⊆ X .
Luna’s Fundamental Lemma (6.5.31) provides condition on when q1 , q2 : R ⇒ W are
étale: we need that for all closed points r ∈ R that
(a) p1 (r), p2 (r) ∈ W are closed points; and
(b) p1 and p2 induce isomorphisms of stabilizer groups at r.
On the other hand, we know that f (w) ∈ X is closed and f induces an isomorphism
of stabilizer groups at the given preimage w of x. We want to show that there
is an open neighborhood U of w such that the restriction f |U satisfies: (a) f |U
sends closed points map to closed points and (b) f |U induces isomorphisms of
stabilizer groups at closed points, and moreover that these conditions are stable
under base change.9 While property (b) is stable under base change, property
(a) is not, and we will introduce a stronger condition below—called Θ-surjectivity
(Definition 6.10.8)—which is stable under base change and implies (a).
In the construction, the role of Θ-completeness and S-completeness is the follow-
ing: the Θ-completeness of X implies that Θ-surjectivity holds (and thus condition
(a) and its base changes hold) in an open neighborhood of w (Proposition 6.10.13),
while S-completeness implies that condition (b) holds in an open neighborhood of w
(Proposition 6.10.20).

6.10.2 Counterexamples
The following examples do not admit good moduli spaces. As the proof strategy
above fails, these examples are instructive for understanding the role of Θ- and
S-completeness. We work over an algebraically closed field k.
The following three examples—[P1 /Gm ], [C/Gm ], and [(A2 ∖ 0)/Gm ]—violate
Θ-completeness as demonstrated in Example 6.9.28.

Example 6.10.2. The quotient stack [P1 /Gm ] does not admit a good moduli space,
where Gm acts on P1 via t·[x : y] = [tx : y]. For a stack admitting a good moduli space,
every k-point has a unique closed point in its closure (Theorem 6.5.6(2)). This fails
here as [1 : 1] specializes to two closed points [1 : 0] and [0 : 1]. Alternatively, if there
were a good moduli space, it would have to be [P1 /Gm ] → Spec k (which is universal
for maps to algebraic spaces), but then the composition P1 → [P1 /Gm ] → Spec k
would be affine by Serre’s Criterion for Affineness (4.5.16), a contradiction.
There are two open substacks U1 , U2 ⊆ [P1 /Gm ] isomorphic to [A1 /Gm ] each
which admits a good moduli space πi : Ui → Spec k but they do not glue to a good
moduli space of [P1 /Gm ]: the intersection U1 ∩ U2 is the open point in both U1 and
U2 and not the preimage of an open subscheme under πi . To see how the approach
9 By Luna’s Fundamental Lemma (6.5.28), we do know that there is an open neighborhood R
1
(resp., R2 ) of ρ = (w, w, id) ∈ R where p1 (resp., p2 ) satisfies (a) and (b), but it is not clear that
these loci are R-invariant, i.e., the preimage of an open substack of W.

392
above fails, observe that the étale presentation f : W := U1 U` ` /Gm ] satisfies
1
`
2 → [P
(a) and (b) but the base changes p1 , p2 : W ×[P1 /Gm ] W = U1 U2 U1 ∩ U2 → W
fails (b), i.e., the closed point in U1 ∩ U2 is mapped to a non-closed point under
either projection.
Example 6.10.3. For a related example, let C be the projective nodal cubic with
its Gm -action. The quotient [C/Gm ] has two points—one open and one closed—but
while there is no topological obstruction as above, [C/Gm ] again does not admit
a good moduli space because C is projective, not affine. Viewing the nodal cubic
as the quotient of nodal union X ′ of two P1 ’s along 0 and ∞ modulo the rotation
action of Z/2, we have a finite étale cover [X ′ /Gm ] → [X/Gm ]. Removing one of the
origins, we have an affine étale cover W = [Spec(k[x, y]/xy)/Gm ] → [C/Gm ] where
Gm acts via t · (x, y) = (tx, t−1 y). Again, this map sends closed points to closed
points, but the projections W ×[C/Gm ] W ⇒ W do not.
Example 6.10.4. Let Gm act on A2 via t · (x, y) = (tx, y) and set X = [A2 /Gm ] ∖ 0.
The point p = (1, 0) ∈ X is closed with trivial stabilizer, and the open immersion
f : A1 ,→ X , sending z to (z, 1), is an étale quotient presentation. Note that while
f (0) is closed, the image f (z) is not closed for z ̸= 0. The map X → A1 defined by
(x, y) 7→ y is not a good moduli space as A2 ∖ 0 is not affine.
Examples violating Θ-completeness naturally occur in moduli, e.g., by removing
a single polystable but not stable vector bundle from Bun r,d (C)ss . The next three
examples violate S-completeness.
Example 6.10.5.S Suppose char(k) ̸= 2 and let G = Z/2 act on the non-separated
union U = A1 x̸=0 A1 by exchanging the copies of A1 . The quotient stack [U/G]
has a Z/2 stabilizer everywhere except at the origin. This is a Deligne–Mumford
stack with quasi-finite but not finite inertia, and is not S-complete. In fact, we have
seen this before in Exercise 4.4.16, when we illustrated the necessity of finite inertia
hypothesis in the Keel–Mori Theorem (4.4.6). By precomposing by the inclusion of
one of the A1 ’s, we have an affine étale morphism A1 → [U/G] which is stabilizer
preserving at 0 but not in any open neighborhood of 0.
Example 6.10.6. The Deligne–Mumford locus X DM in the moduli stack X =
[Sym4 P1 / PGL2 ] of four unordered points in P1 is not separated (see Example 4.4.17)
and thus not S-complete. Note however that the stable locus X s consisting of four
distinct points is separated and the semistable locus X ss = X DM ∪ {[0 : 0 : ∞ : ∞]}
has a projective good moduli space.
Example 6.10.7. Consider the action of G = Gm ⋊ Z/2 on X = A2 ∖ 0 via
t · (a, b) = (ta, t−1 b) and −1 · (a, b) = (b, a). Note that every point (a, b) ∈ X with
ab ̸= 0 is fixed by the order 2 element (a/b, −1) ∈ G. The quotient stack [X/G]
is a non-separated Deligne–Mumford stack that is not S-complete and that does
not admit a good moduli space; note however that [A2 /G] → Spec k[xy] is a good
moduli space.

6.10.3 Θ-completeness and Θ-surjectivity

The property that a morphism X → Y sends closed points to closed points is not
stable under base change (see Examples 6.10.2 and 6.10.3). We introduce a stronger
and better behaved property called Θ-surjectivity. The main result of this section
is that an étale quotient presentation ([Spec A/Gx ], w) → (X , x) is Θ-surjective in
an open neighborhood of w as long as X is Θ-complete (Proposition 6.10.13). As

393
motivated in §6.10.1, this result will be crucial in proving the main existence theorem
(Theorem 6.10.1) of this chapter.
Definition 6.10.8. Let f : X → Y be a morphism of algebraic stacks and x ∈ X (k)
be a geometric point. We say that f is Θ-surjective at x if every diagram

Spec _ k
x /X
<
1 f (6.10.9)
 
Θk /Y

has a lift. We say that f is Θ-surjective if it is Θ-surjective at every geometric point.

This notion is clearly stable under base change. Every morphism f : X → Y
of noetherian algebraic stacks, where Y has affine and quasi-finite diagonal, is Θ-
surjective since in this case every map Θk → Y factors through Spec k (Lemma 6.9.21).
The next lemma gives conditions for when the lift is unique and when the definition
is independent of the choice of geometric point.
Lemma 6.10.10. Let f : X → Y be a separated, representable, and finite type
morphism of noetherian algebraic stacks.
(1) Every lift of (6.10.9) is unique.
(2) If f is Θ-surjective at a geometric point x ∈ X (k), then f is Θ-surjective at
every other geometric point x′ ∈ X (k′ ) representing the same point in |X | as
x.
Proof. Part (1) follows from descent and the valuative criterion for separatedness.
To show (2), it suffices to show that given an extension k → k′ of algebraically
S closed
fields, a lift Θk′ → X implies the existence of a lift Θk → X . We write k′ = λ Aλ as
a union of finitely generated k-subalgebras. By Limit Methods (§B.3), there exists a
lift ΘAλ → X of Spec Aλ → X . Restricting along a closed point of Spec Aλ provides
a lift over k.
Proposition 6.10.11. Let f : X → Y be a morphism of algebraic stacks, each of
finite type over an algebraically closed field k with affine diagonal. Suppose that
the closed points of Y have linearly reductive stabilizer. If f is Θ-surjective, then f
sends closed points to closed points.
Proof. Let x ∈ |X | be a closed point. Let f (x) ⇝ y0 be a specialization to a closed
point. By the Destabilization Theorem (7.3.8), this specialization can be realized by
a map Θ → Y. Since f is Θ-surjective, this can be lifted to a map g : Θ → X with
g(1) = x. But x ∈ |X | is a closed point, so this lift must correspond to the trivial
specialization x ⇝ x. It follows that f (x) = y0 is a closed point.
Remark 6.10.12. The converse is not true. In Example 6.10.3, where C is the nodal
cubic with Gm -action, the étale morphism [Spec(k[x, y]/(xy))/Gm ] → [C/Gm ] sends
closed points to closed points but is not Θ-surjective.
Proposition 6.10.13. Let X be an algebraic stack of finite type over an algebraically
closed field k with affine diagonal such that the closed points of X have linearly
reductive stabilizers. Let x ∈ |X | be a closed point, let f : ([Spec A/Gx ], w) → (X , x)
be an affine étale morphism inducing an isomorphism of stabilizer groups at w,
and let π : [Spec A/Gx ] → Spec AGx . If X is Θ-complete, there exists an open
affine neighborhood U ⊆ Spec AGx of π(w) such that f |π−1 (U ) : π −1 (U ) → X is
Θ-surjective.

394
Proof. Let W = [Spec A/Gx ] and define Σf ⊆ |W| as the set of points y ∈ |W|
such that f is Θ-surjective at y. We first show that Σf ⊆ W is open in the case
that X ∼
= [Spec B/G] with G linearly reductive. Zariski’s Main Theorem (6.1.18)
provides a factorization
j ν
f : W ,→ Xe −
→X
where j is an open immersion and ν is a finite morphism. By Lemma 6.9.19,
[Spec B/G] is Θ-complete. By Proposition 6.9.20, since ν is affine, Xe is also Θ-
complete. The finiteness of ν implies that Σj = Σf , and we may thus assume
that f is an open immersion. Let Z ⊆ X be the reduced complement of W and
let π : X → Spec B G denote the good moduli space. We claim that |W| ∖ Σf =
π −1 (π(|Z|)) ∩ |W|. The inclusion “⊆” is clear: the morphism X ∖ π −1 (π(|Z|)) ,→ X
is the base change of the Θ-surjective morphism X ∖ π(|Z|) ,→ X of algebraic
spaces. Conversely, let y ∈ π −1 (π(|Z|)) ∩ |W| represented by a geometric point
Spec F → X . Let z ∈ |ZF | be the unique closed point in the closure of y ∈ |ΘF | and
let ΘF → XF be a morphism representing the specialization y ⇝ z (Corollary 7.3.8).
Since ΘF → X does not lift to W, y ∈ / Σf .
We now claim that Σf ⊆ W is constructible. Use the Local Structure Theorem
(6.7.1) to choose an affine, étale, and surjective morphism g : X ′ = [Spec B/G] → X
with G linearly reductive. Let W ′ = W ×X X ′ with projections g ′ : W ′ → W and
f ′ : W ′ → X ′ . Since we already know that Σf ′ is open, the claim follows from
Chevalley’s Theorem (3.3.31) once we show that W ∖ Σf = g ′ (W ′ ∖ Σf ′ ). To see this,
it suffices to show that for an algebraically closed field F, every map h : ΘF → X lifts
to a map h′ : ΘF → X ′ . Let x′ ∈ X ′ (F) be a preimage of h(0) ∈ X (F). Since g is
representable and étale, the induced map Gx′ → Gh(0) on stabilizers is injective with
finite cokernel. Thus the map Gm,F → Gh(0) on stabilizers induced by h : ΘF → X
factors through Gx′ . We may therefore lift the map h|BGm,F to a map BGm,F → X ′ .
Letting Xn be the nth nilpotent thickening of BGm,F ,→ ΘF , there are compatible
lifts Xn → X ′ of Xn → X by deformation theory (Proposition 6.7.12) which extends
to a lift ΘF → X ′ by Coherent Tannaka Duality (6.6.9).
Since Σf ⊆ W is constructible and w ∈ Σf , to show that Σf is open, it suffices
to show that every generization ξ ⇝ w of w is contained in Σf . Let h : Spec R → W
be a morphism from a complete DVR representing the specialization ξ ⇝ w. Letting
K and κ be the fraction and residue field of R, we claim that there exists a lift
(necessarily unique as f is separated)

Spec _ K
hF
/W
;
g
e
f (6.10.14)
 
ΘK
g
/ X.

This claim implies that f is Θ-surjective at ξ, i.e., ξ ∈ Σf . To show the claim, we

first apply the Θ-completeness of X to construct a lift

(f ◦h)∪g
_ 0
ΘR ∖ /X
;
q

ΘR .

Since W → X is stabilizer preserving at w, we have a lift BGm,κ → W of q|BGm,κ .

Since ΘR is coherently complete along BGm,κ (6.6.13), we may apply deformation

395
theory (Proposition 6.7.12) and Coherent Tannaka Duality (6.6.9) to construct a lift

BGm,κ /W
_ ;
qe
f
 
ΘR
q
/X

The restriction qe|Spec R is 2-isomorphic to h since it agrees at the closed point and f
is étale. It follows that ge := qe|ΘK is a lift of (6.10.14).

The topology of k-points of Θ-complete stacks is analogous to the topology of

quotient stacks arising from GIT.

Exercise 6.10.15. Let X be an algebraic stack of finite type over an algebraically

closed field k with affine diagonal. Assume that X is Θ-complete and that the closed
points of X have linearly reductive stabilizers. Show that the closure of every k-point
contains a unique closed point.

Exercise 6.10.16. With the hypotheses of Exercise 6.10.15, show that if in addition
X has a unique closed point, then X ∼ = [Spec(A)/Gx ] such that AGx is an artinian
local k-algebra with residue field k.

6.10.4 Unpunctured inertia

We introduce the concept of unpunctured inertia for an algebraic stack X , and show
that it implies that every étale quotient presentation f : ([Spec A/Gx ], w) → (X , x)
is stabilizer preserving in an open neighborhood of w (Proposition 6.10.20).

Definition 6.10.17. A noetherian algebraic stack has unpunctured inertia if for

every closed point x ∈ |X | and for every formally versal morphism p : (T, t) → (X , x),
where T is the spectrum of a local ring with closed point t, every connected component
of the inertia group scheme AutX (p) → T has non-empty intersection with the fiber
over t.

Remark 6.10.18. Here (T, t) → (X , x) is formally versal if the map Tb → X from the
completion is is formally versal at t as in Definition C.4.2.
Remark 6.10.19. Unpuncturedness is related to the purity of the morphism AutX (p) →
T as defined in [GR71, §3.3] (see also [SP, Tag 0CV5]). If T is the spectrum of a
strictly henselian local ring, then purity requires that if s ∈ T is an arbitrary point
and γ is an associated point in the fiber AutX (p)s , then the closure of γ in AutX (p)
has non-empty intersection with the fiber over the closed point t of T .

Proposition 6.10.20. Let X be an algebraic stack of finite type over an algebraically

closed field k with affine diagonal. Let x ∈ |X | be a closed point with linearly reductive
stabilizer Gx . Let f : ([Spec A/Gx ], w) → (X , x) be an affine étale morphism inducing
an isomorphism of stabilizer groups at w, and let π : [Spec A/Gx ] → Spec AGx . If X
has unpunctured inertia, there exists an open affine neighborhood U ⊆ Spec AGx of
π(w) such that f |π−1 (U ) : π −1 (U ) → X induces isomorphisms of stabilizer groups at
all points.

Proof. Set W = [Spec A/Gx ]. It suffices to find an open neighborhood U ⊆ W of

w such that f |U : U → X induces an isomorphism IU → U ×X IX . Consider the

396
cartesian diagram
IW / W ×X IX

□
 
W / W ×X W;

see Exercise 3.2.15. Since f is separated and étale, the morphism IW → W ×X IX

is finite and étale. We set Z ⊆ W ×X IX to be the open and closed substack over
which IW → W ×X IX is not an isomorphism. Since f is stabilizer preserving at w,
the point w is not contained in the image of Z under p1 : W ×X IX → W.
Consider a formally smooth morphism (T, t) → (X , x) from the spectrum of a
local ring with closed point t. Since X has unpunctured inertia, the preimage of Z
in W ×X IX ×X T is empty; indeed, if there were a non-empty connected component
of this preimage, it must intersect the fiber over t non-trivially contradicting that
w∈ / p1 (Z). This in turn implies that w ∈
/ p1 (Z). Therefore, if we set U = W ∖p1 (Z),
the induced morphism IU → U ×X IX is an isomorphism.

Proposition 6.10.21. Let X be a noetherian algebraic stack.

(1) If X has quasi-finite inertia, then X has unpunctured inertia if and only if X
has finite inertia.
(2) If X has connected stabilizer groups, then X has unpunctured inertia.

Proof. If X has finite inertia, then AutX (p) → T is finite, so clearly the image of
each connected component contains the unique closed point t ∈ T . For the converse,
we may assume ` that T is the spectrum of a Henselian local ring, in which case
AutX (p) = G H where G → T finite and the fiber of H → T over t is empty
(Proposition B.5.9). If T is nonempty (i.e., AutX (p) → T is not finite), then any
connected component of T doesn’t meet the central fiber and thus X does not have
unpunctured inertia.
For (2), by definition, all fibers of AutX (p) → U are connected, so every connected
component of AutX (p) intersects the component containing the identity section.

Remark 6.10.22. For algebraic stacks with connected stabilizer groups (e.g., the mod-
uli stack Bun ss
r,d (C) of semistable vector bundles on a curve), Proposition 6.10.21(2)
implies unpunctured inertia. The deeper result below (Theorem 6.10.23) is therefore
unneeded in the proof of the existence of a good moduli space of Bun ss r,d (C).

6.10.5 S-completeness implies unpunctured inertia

Theorem 6.10.23. Let X be an algebraic stack of finite type over an algebraically
closed field k with affine diagonal. Assume that the closed points have linearly
reductive stabilizers. If X is S-complete, then X has unpunctured inertia.

Proof. Let x ∈ |X | be a closed point, let p : (U, u) → (X , x) be a formally smooth

morphism from the spectrum of a local ring, and let H ⊆ AutX (p) be a connected
component. The image of the projection H → U is a constructible set whose closure
contains u. Choose a DVR R with residue field k and a map Spec R → U whose
special point maps to u and whose generic point lies in the image of H → U . Let
p
ξ : Spec R → U −→ X denote the composition. After a residually-trivial extension of
DVRs, we may assume that the generic point Spec K → U lifts to H. This gives a

397
commutative diagram

Spec K / Spec R
ξ

  #
H /U p
/ X.

Let HK be the base change of H → U along Spec K → U . We claim we can choose

a finite type point g ∈ HK of finite order. If g ∈ HK is any finite type point, then
after replacing K with a finite field extension, we can decompose g = gs gu under the
Jordan decomposition, where gs is semisimple and gu is unipotent (see §B.1.3). Now
consider the reduced closed K-subgroup H ′ ⊆ AutX (p)K generated by gs . Because
gs is semisimple, H ′ is a diagonalizable group scheme over K, and we may replace
gs with a finite order element in H ′ which still commutes with gu . If char(K) > 0,
then gu has finite order and we are finished. If char(K) = 0, then gu lies in the
identity component of G, so g lies on the same component as the finite order element
gs . This gives the desired element.
We claim that after replacing R with a residually-trivial extension, there is a map
ξ ′ : Spec R → X such that ξK ′
≃ ξK and g ∈ HK extends to an automorphism of ξ ′ .
This would finish the proof: since the closure of g meets the fiber of AutX (p) → U
over u, the component H must also meet the central fiber. If X ∼ = [Spec A/ GLn ],
then this claim is precisely the content of Proposition 6.10.24 below. We use the Local
Structure Theorem (6.7.1) to reduce to this case: let f : ([Spec A/Gx ], w) → (X , x) be
an étale quotient presentation. After replacing R with a residually-trivial extension,
we may lift ξ to a map ξe: Spec R → [Spec A/Gx ] such that ξ(0) e = w. To show that
g lifts to an element ge ∈ Aut(ξeK ), we use S-completeness. We may glue ξ to itself
along g to define a morphism
[
Spec R Spec R = ϕR ∖ 0 → X .
Spec K

Since X is S-complete, this map extends to a morphism h : ϕR → X . Since ξ(0) = x

and x is a closed point, the image h(0) of 0 ∈ ϕR is also x. Since f is stabilizer
preserving at w, we may lift h|BGm to a map e h0 : BGm → [Spec A/Gx ] with image
w. By Deformation Theory (6.7.12), we may find compatible lifts to [Spec A/Gx ] of
the restrictions of h to the nilpotent thickenings of ϕR along 0, and by Coherent
Tannaka Duality (6.6.9), we may find construct a lift eh below

h0
/ [Spec A/Gx ]
e
BG m
_ 8
h
e
f
 
ϕR
h / X.

Since f is affine and étale, both restrictions e h|s̸=0 and eh|t̸=0 to Spec R are iso-
morphic to ξe and thus e h|ϕR ∖0 gives a lift ge ∈ Aut(ξeK ) of g. Finally, we apply
Proposition 6.10.24 to construct a map ξe′ : Spec R → [Spec A/Gx ] with ξe′ (0) = w
such that ξeK ≃ ξeK ′
and ge extends to an automorphism of ξe′ . The composition
ξ ′′ := f ◦ ξe′′ : Spec R → X then satisfies the claim. See also [AHLH23, Thm. 5.2].

Our proof used the following valuative criterion for a quotient stack.

398
Proposition 6.10.24. Let X = [Spec A/G] where Spec A is an affine scheme of
finite type over an algebraically closed field k equipped with an action by a linearly
reductive group G. Let x ∈ |X | be a closed point. Then X satisfies the following
property:
(⋆) For every DVR R with residue field k and fraction field K, for every morphism
ξ : Spec R → X with ξ(0) ≃ x, and for every K-pint g ∈ AutX (ξK ) of finite
order, there is an extension R → R′ of DVRs (with K ′ = Frac(R′ )) and a
morphism ξ ′ : Spec R′ → X such that ξ ′ (0) ≃ x, ξK
′
′ ≃ ξK ′ and g|K ′ extends
′
to an automorphism of ξ .

Remark 6.10.25. In other words, for every map ξ : Spec R → Spec A and element
g ∈ GξK ⊆ G(K) of finite order, there exists after an extension R ⊆ R′ of DVRs
and an element h ∈ G(K ′ ) such that h · ξK ′ extends to a map ξ ′ : Spec R′ → Spec A
with ξ ′ (0) ∈ Gx and such that h−1 g|K ′ h extends to an R′ -point of G. To illustrate
this criterion, consider the action of G = Gm ⋊ Z/2 on A2 via t · (a, b) = (ta, t−1 b)
and −1 · (a, b) = (b, a). Note that every point (a, b) ∈ A2 with ab ̸= 0 is fixed by the
order 2 element (a/b, −1) ∈ G. Consider ξ : Spec R = k[[z]] → A2 via z 7→ (z 2 , z).
The element g = (z −1 , −1) ∈ G(k((z))) stabilizes ξK but does not extend √ to G(k[[z]]).
However, we √ may take the
√ degree√ 2 ramified
√ extension k[[z]] → k[[ z]] and define
ξ ′ : Spec k[[ z]] → A2 by z 7→ (( z)3 , ( z)3√ ). Over the generic
√ point, there is an
isomorphism ξk(( ′ √
z))
≃ ξ √
k(( z)) given by h = ( z, −1) ∈ G(k(( z))) and the element
√ √ √
g|k(( z)) = ( z, −1) · g|K ′ · ( z, −1) = (1, −1) ∈ G(k(( z))) extends to an element
√ −1
√
of G(k[[ z]])-point.

Proof. After choosing an embedding G ,→ GLn and replacing [Spec A/G] with
[(Spec A ×Gx GLn )/ GLn ], we may assume that G = GLn . We first verify (⋆) for
quotient stacks [Spec A/G] = Spec A × BG with a trivial action. As R is local and
G = GLn , the composition Spec R → [Spec A/G] → BG corresponds to the trivial
G-bundle. We need to prove that every finite order element g ∈ G(K) is conjugate to
an element of G(R) after passing to an extension of the DVR R. We can conjugate
g to its Jordan canonical form (after an extension of R). Since g has finite order, the
diagonal entries of the resulting matrix are rth roots of unity for some r. Because
the group µ r of rth roots of unity is a finite group scheme over Spec R, the entries
of the Jordan canonical form must lie in R.
If X = [X/G] with X a proper scheme over k, we show that (⋆) holds except that
ξ ′ (0) may not be isomorphic to x. Since p : X → BG is proper and representable,
for every morphism ξ : Spec R → X from a DVR, we have a closed immersion
AutX (ξ) ,→ AutBG (p ◦ ξ) of group schemes over Spec R. Moreover, any lift of the
generic point of a morphism Spec R → BG to [X/G] extends to a unique morphism
Spec R → BG. Therefore, given an element g ∈ AutX (ξK ), we use that (⋆) holds
for BG to find (after replacing R with an extension) a morphism η : Spec R → BG
such that ηK ≃ (p ◦ ξ)K and g|K extends to a R-point of AutBG (η). If we lift η to
a morphism ξ ′ : Spec R → [X/G] such that ξK ′
≃ ξK , then the element g|K extends
to an automorphism of ξ . ′

In verifying (⋆) for [Spec A/G], we may assume that A is reduced. Viewing
[Spec A/G] as an algebraic stack which is affine and of finite type over Spec AG ×BG,
we can choose a vector bundle E on Spec AG × BG and a G-equivariant embedding
Spec A ,→ A(E|Spec AG ) over AG . Viewing A(E|Spec AG ) as an open subscheme of
P(E|Spec AG ⊕OSpec AG ), we let X be the closure of Spec A in P(E|Spec AG ⊕OSpec AG ).

399
This gives a G-equivariant diagram

Spec A  /X (6.10.26)

% 
Spec AG

where X is a reduced projective scheme and the complement X∖Spec A is the support
of an ample G-invariant Cartier divisor E. We also claim that Spec A is precisely the
semistable locus of X with respect to OX (E) in the sense of Exercise 7.2.11. Indeed
the tautological invariant section s : OX → OX (E) restricts to an isomorphism over
Spec A and thus Spec A ⊆ X ss . Conversely, sn defines an isomorphism
∼
AG → Γ(X, OX (nE))G

for all n ≥ 0. Under this isomorphism, for every invariant global section f ∈
Γ(X, OX (nE))G , the restriction f |Spec A agrees with a section of the form gsn , where
g is the pullback of a function under the map X → Spec AG . It follows that f = g ·sn
because X is reduced. This shows that X ss ⊆ Spec A.
We now verify that (⋆) holds for [Spec A/G]. Let ξ : Spec R → [Spec A/G] be a
map with ξ(0) ≃ x, and let g ∈ AutX (ξK ) be a finite order K-point. By applying
the above result to [X/G], there exists (after an extension of R) a map ξ ′ : Spec R →
[X/G] such that ξK ′
≃ ξK and g extends to an element of AutX (ξ ′ ) but where ξ ′ (0)
may not be isomorphic to x. The stabilizer group scheme StabG (X) ⊆ X × G is
a closed subscheme equivariant with respect to the product action of G on X × G
where G acts on itself via conjugation. The pair (ξ ′ , g) defines a morphism

η : Spec R → [StabG (X)/G].

We will show that after an extension of R, there is a map η ′ : Spec R → [StabG (Spec A)/G]
with ηK
′
≃ ηK . Similar to (6.10.26), we have a G-equivariant diagram

 / StabG (X)
StabG (Spec A)

( 
Spec AG × G

with StabG (X) projective over Spec AG × G. We claim that the semistable locus
of StabG (X) for the action of G with respect to the pullback of OX (E) is precisely
StabG (Spec A). The invariant section s ∈ Γ(X, OX (E))G pulls back to an invariant
section on StabG (X) and thus StabG (Spec A) ⊆ StabG (X)ss . To see the converse,
suppose that (y, h) ∈ StabG (X) with y ∈ / X ss = Spec A. Applying Kempf’s Optimal
Destabilizing Theorem (7.6.6) to a lift yb of y to the affine cone X b → Spec AG of X
yields a one-parameter subgroup λ : Gm → G such that limt→0 λ(t) · yb ∈ X b exists and
is contained in the zero section Spec A . Moreover, since Gy ⊆ Pλ (Exercise 7.6.13),
G

y , h) also exists and is contained in the zero section of the affine cone
limt→0 λ(t) · (b
over of StabG (X); thus (y, h) is not semistable.
The induced morphism StabG (Spec A)//G → (Spec AG × G)//G of GIT quotients
is proper, and the good moduli space [StabG (Spec A)/G] → StabG (Spec A)//G is
universally closed. By the valuative criterion, after an extension of R, there exists a

400
lift

[StabG (Spec A)/G] / StabG (Spec A)//G

3
χ


Spec R
η
/ StabG (X) / Spec AG × G / (Spec AG × G)//G

χ
such that χK ≃ ξK . Under the composition ξ ′ : Spec R −
→ [StabG (Spec A)/G] →
[Spec A/G], ξK ′
≃ ξK and g extends to an element of AutX (ξ ′ ). To arrange that
ξ ′ (0) ≃ x, we apply Lemma 6.10.27 below.

Lemma 6.10.27. Let X = [Spec A/G] where Spec A is an affine scheme of finite
type over an algebraically closed field k equipped with an action of a reductive group
G. Let ξ, ξ ′ : Spec R → X be morphisms from a DVR with residue field k such that
′
ξK ≃ ξK and ξ(0) ∈ |X | is a closed point. For every element g ∈ AutX (ξ ′ ), there
exists (after replacing R with an extension) a morphism ξ ′′ : Spec R → X such that
′′ ′
ξK ≃ ξK , g|K extends to an automorphism of ξ ′′ , and ξ ′′ (0) ≃ ξ(0).

Proof. Since ξ(0) and ξ ′ (0) lie in the same fiber of X → Spec AG , the closure of
ξ ′ (0) in |X | must contain ξ(0). Kempf’s Criterion (7.6.5) yields a canonical map
f : Θ → [Spec A/G] with f (1) ≃ ξ ′ (0) and f (0) ≃ ξ(0). Since f is canonical, every
automorphism of f (1) extends to an automorphism of the map f . In particular the
restriction of g ∈ AutX (ξ ′ ) to f (1) = ξ ′ (0) extends uniquely to an automorphism gf
of f .
We now apply the Strange Gluing Lemma (6.10.28), which after replacing R
with R[π 1/N ] and precomposing f with the map Θ → Θ defined by x 7→ xN for
N ≫ 0, yields a unique map γ : ϕR → X , such that γ|s=0 ≃ f and γ|t̸=0 ≃ ξ ′ .
The uniqueness γ guarantees that the automorphism g of ξ ′ and gf of f extends
uniquely to an automorphism gγ of γ. Finally, we construct the desired map ξ ′′ as
the composition
√ q γ
ξ ′′ : Spec(R[ π]) − → ϕR − → X,
√ √
where in (s, t, π) coordinates √ the first map q is defined by ( π, π, π). Under q,
the special point of Spec(R[ π]) maps to the point 0 ∈ ϕR . By construction,
ξ ′′ (0) ≃ ξ(0) and the automorphism g of γ restricts to an automorphism of ξ ′′
extending g|K(√π) .

Lemma 6.10.28 (Strange Gluing Lemma). Let X be an algebraic stack of finite type
over an algebraically closed field k with affine diagonal. Let R be a DVR with residue
field k. Let f : Θ → X and ξ : Spec R → X be morphisms with an isomorphism
f (1) ≃ ξ(0). For n ≫ 0, after replacing R with R[π 1/n ] and f with the composition
n f
Θ− →Θ− → X , there is a unique morphism γ : ϕR → X such that γ|s=0 ≃ f and
γ|t̸=0 ≃ ξ.

Proof. For n > 0, define

ϕn,1 n
R = [Spec(R[s, t]/(st − π))/Gm ]

where the Gm -acts with weight n on s and −1 on t. We have a closed immersion

Θ ,→ ϕn,1 n,1
R defined by s = 0 and an open immersion Spec R ,→ ϕR defined by
n,1
t ̸= 0. Note that any morphism ϕR → X restricts to morphisms f : Θ → X and
ξ : Spec R → X along with an isomorphism ξ(0) ≃ f (1). We will show conversely

401
that for n ≫ 0, any f : Θ → X and ξ : Spec R → X with ξ(0) ≃ f (1) extends
canonically to a map ϕn,1
R → X.
Letting C = R[t, π/t, π/t2 , . . .] ⊆ R[t]t , the diagram

Spec k[t]t / Spec k[t]

⌟
 
Spec R[t]t / Spec C

is a pushout in the category of schemes (Theorem B.4.1). This diagram is Gm -

equivariant, and the diagram obtained by taking the fiber product with Gm is also
a pushout. By Properties of Pushouts (B.4.8), taking quotients by Gm yields a
pushout square

Spec k /Θ

⌟
  f

Spec R / [Spec(C)/Gm ] (6.10.29)

Ψ

&/ 
ξ X
in the category of algebraic stacks with affine S diagonal. This induces the dotted
arrow Ψ. We can write C as a union C = Cn where Cn := R[t, π/tn ] ⊆ R[t]t .
Note that Cn ∼ = R[s, t]/(stn − π) so in particular [Spec(Cn )/Gm ] ∼
n,1
= ϕR . As X → S
is locally of finite presentation, for n ≫ 0 the morphism Ψ factors uniquely as
[Spec(C)/Gm ] → ϕn,1 R → X (see Exercise 3.3.33).
To finish the proof, compose the uniquely defined map ϕn,1 R → X with the
canonical map ϕR[π1/n ] → X induced by the map of graded algebras R[s, t]/(stn −
π) → R[π 1/n ][s1/n , t]/(s1/n t − π), where s1/n has weight 1.

6.10.6 S-completeness and reductivity

We have already seen that S-completeness characterizes separatedness (Proposi-
tion 6.9.22 and Proposition 6.9.23). We have also seen that it implies unpunctured
inertia (Theorem 6.10.23) and therefore implies the existence of stabilizer preserving
local quotient presentations (Proposition 6.10.20). We now prove a third remarkable
property of S-completeness: it characterizes reductivity. More precisely, a smooth
affine algebraic group G is reductive if and only if BG is S-complete if and only if G
has Cartan Decompositions. The existence of Cartan Decompositions is discussed
further in §7.3, where we use it to prove the Hilbert–Mumford Criterion (7.4.4)
Proposition 6.10.30. Let G be a smooth affine algebraic group over an algebraically
closed field k. The following are equivalent:
(1) G is reductive,
(2) BG is S-complete, and
(3) G satisfies the Cartan Decomposition: for every complete DVR R over k with
residue field k and fraction field K and for every element g ∈ G(K), there
exists elements h1 , h2 ∈ G(R) and a one-parameter subgroup λ : Gm → G such
that
g = h1 λ|K h2 .

402
Proof. We have already seen that S(1) ⇒ (2) in Proposition 6.9.20. For (2) ⇒ (3),
observe that since ϕR ∖0 = Spec R Spec K Spec R, an element g ∈ G(K) determines
a morphism
ρg : ϕR ∖ 0 → BG
by gluing two trivial G-torsors over Spec R via the isomorphism of their restrictions
to Spec K determined by g. Since BG is S-complete, we have a lift
ρg
ϕR ∖ 0 / BG
;
h
(6.10.31)

ϕR .

The restriction hBGm : BGm → BG corresponds to a one-parameter subgroup

λ : Gm → G (up to conjugation), which is our candidate for the decomposition
g = h1 λ|K h2 . We make two observations:
• If g, g ′ ∈ G(K) are elements, the morphisms ρg , ρg′ : ϕR ∖ 0 → BG are
isomorphic if and only if there are elements h, h′ ∈ G(R) such that hg = g ′ h′ .
• If λ : Gm → G is a one-parameter subgroup and λ|ϕR ∖0 denotes the composition
λ
→ BG, then λ|ϕR ∖0 and ρg′ , where g ′ = λ|K , are
ϕR ∖ 0 ,→ ϕR → BGm −
isomorphic.
It therefore suffices to show that the extension h in (6.10.31) is isomorphic to
λ
λ|ϕR : ϕR → BGm − → BG. To see this, let P and P ′ denote the principal G-bundles
over ϕR classifying h and λ|ϕR . Since G is smooth and affine, IsomϕR (P, P ′ ) →
ϕR is smooth and affine. We have a section over the inclusion X0 := BGm ,→
ϕR of 0. Letting Xn denote the nth nilpotent thickening, deformation theory
(Proposition 6.7.12) and the cohomological affineness of Xn imply that there are
compatible sections over Xn . Coherent Tannaka Duality (6.6.9) and the coherent
completeness of ϕR along BGm (Theorem 6.6.13) imply that

MorϕR (ϕR , IsomϕR (P, P ′ )) → lim MorϕR (Xn , IsomϕR (P, P ′ ))

←−
is an equivalence. We thus obtain a section of IsomϕR (P, P ′ ) → ϕR , i.e., an
isomorphism between P and P ′ . (The logic in this implication is reversible.)
For (3) ⇒ (1), if G is not reductive, there is a normal subgroup Ga ⊴ Ru (G) of
the unipotent radical. As G/Ru (G) and Ru (G)/Ga are both affine, the composition
BGa → BRu G → BG is affine. By Lemma 6.9.19, this would imply that BGa is
S-complete but this is a contradiction: taking R = k[[x]] and K = k((x)) the element
x ∈ Ga (K) cannot be written as h1 λ|K h2 . See also [AHHL21, Thm. A].
In particular, if X is an S-complete algebraic stack and x ∈ |X | is a closed point
with smooth affine stabilizer Gx , then Gx is reductive. With the stronger hypothesis
that X admits a good moduli space, this was established in Exercise 6.5.25.

6.10.7 Proof of the Existence Theorem of Good Moduli

Spaces
The necessity of Θ- and S-completeness for the existence of a good moduli space
was established in Proposition 6.9.23. We now prove the sufficiency following the
strategy outlined in §6.10.1.

403
Proof of Theorem 6.10.1. Since X is S-complete and char(k) = 0, the stabilizer Gx
of every closed point x ∈ |X | is linearly reductive (Proposition 6.10.30). By the Local
Structure Theorem (6.7.1), there exists an affine étale morphism f : ([Spec A/Gx ], w) →
(X , x) inducing an isomorphism of stabilizer groups at x. Since X is Θ- and
S-complete, we may assume that f is Θ-surjective and stabilizer preserving at
all points after replacing [Spec A/Gx ] with an open neighborhood of x (Proposi-
tions 6.10.13 and 6.10.20). Since X is quasi-compact, there exists finitely many
closed points xi ∈ |X | and morphisms fi : [Spec Ai /Gxi ] → X as above whose images
cover X . Choosing embeddings Gxi ,→ GLn for some n, Q there are equivalence
[Spec Ai /Gxi ] ∼
= [(Spec Ai ×Gxi GLn )/ GLn ]. Setting A = i (Ai ×Gxi GLn ), there
is an surjective, affine, and étale morphism

f : X1 := [Spec A/ GLn ] → X

which is Θ-surjective and stabilizer preserving at all points. Since char(k) = 0,

there is a good moduli space X1 → X1 := Spec AGLN . Set X2 = X1 ×X X1 .
The projections p1 , p2 : X2 → X1 are also affine, étale, Θ-surjective, and stabilizer
preserving. Since f is affine, X2 ∼
= [Spec B/ GLn ] and there is a good moduli space
X2 → X2 := Spec B GLn . This provides a commutative diagram

X2
p1
// X f
/X
1
p2
(6.10.32)
  
X2
q1
// X /X
1
q2

which each square on the left is cartesian by Luna’s Fundamental Lemma (6.5.31).
Moreover, by the universality of good moduli spaces (Theorem 6.5.6(4)), the étale
groupoid structure on X2 ⇒ X1 induces a étale groupoid structure on X2 ⇒ X1 .
We claim that X2 ⇒ X1 is an étale equivalence relation, i.e., that the quotient
stack [X1 /X2 ] is an algebraic space. By the Characterization of Algebraic Spaces
(3.6.6), it suffices to show that if x1 ∈ X1 is a k-point, then (x1 , x1 ) has a unique
preimage under (q1 , q2 ) : X2 → X1 × X1 . Let x2 , x′2 ∈ X2 be two points mapping
to (x1 , x1 ) ∈ X1 × X1 , and let x e′2 ∈ |X2 | be the unique closed points in their
e2 , x
preimages. Since f is Θ-surjective, the images p1 (e x2 ), p2 (e x′2 ), and p2 (e
x2 ), p1 (e x′2 )
are all closed points of X1 over x1 , and therefore they are all identified with the
unique closed point x e1 over x1 . On the other hand, since f is stabilizer preserving,
the stabilizer groups of x e2 and x e′2 are the same as the stabilizer groups of x e1 and
of its image in X . Let us denote this stabilizer group by G. It follows that the
fiber product of (p1 , p2 ) : X2 → X1 × X1 along the inclusion of the residual gerbe
G(ex1 ,ex1 ) = BG × BG → X1 × X1 is isomorphic to BG and thus identified with the
residual gerbe of a unique closed point. Therefore x2 = x′2 . Since X2 ⇒ X1 is an
étale equivalence relation, the quotient X = X1 /X2 is an algebraic space. From
étale descent, there is a morphism X → X which pulls back under X1 → X to the
good moduli space X1 → X1 . By descent of good moduli spaces (Lemma 6.5.22(2)),
X → X is a good moduli space. Finally, we use that X is S-complete to conclude
that X is separated (Proposition 6.9.23).

404
Chapter 7

Geometric Invariant Theory

It is sometimes hard to appreciate how transformative some definitions

and theorems were 60 years ago. Several of Mumford’s ideas seem to
have been quite alien to his predecessors, but once Mumford introduced
them, they quickly became viewed as the “obvious approach.”

János Kollár [Kol21, p. 623]

Geometric Invariant Theory (GIT) was developed by Mumford in [GIT] as a

means to construct quotients and moduli spaces in algebraic geometry. For other
expository accounts, we recommend [New78], [Kra84], [Dol03], [Muk03], and [Stu08].

7.1 Good quotients

The basic problem that I wish to discuss is this: if V is a variety, or
scheme, parameterizing the set, or functor, of all structures of some type
in projective n-space Pn , then the group PGL(n) of automorphisms of
Pn acts on V . Then under what conditions does there exist a quotient
or orbit space V / PGL(n), i.e. when can we construct enough “projective
invariants” for these structures?

David Mumford [Mum63, p. 526]

Let G be an affine algebraic group over an algebraically closed field k acting on

an algebraic space U of finite type over k. For a finite abstract group G viewed as
a group scheme over k, we have already established the existence of a geometric
quotient U/G (Definition 4.2.1) in the following cases:
• If G acts freely on U (i.e., the action map G × U → U × U is a monomorphism),
then U/G := [U/G] exists as an algebraic space of finite type over k (Corol-
lary 3.1.14). This also holds in the non-finite case: if G is an algebraic group
and the action is free, then [U/G] is an algebraic stack (Proposition 6.3.10) such
that [U/G] → [U/G] × [U/G] is a monomorphism and therefore U/G := [U/G]
is an algebraic space (Theorem 4.5.10).
• If U = Spec A is affine, then U/G := Spec AG is a geometric quotient (Theo-
rem 4.2.3). Moreover, if U is projective (resp., quasi-projective, quasi-affine),
then the quotient U/G exists as a projective (resp., quasi-projective, quasi-
affine) k-scheme (Exercise 4.2.14).

405
 • If U is separated, then U/G exists as a separated algebraic space as a conse-
quence of the Keel–Mori Theorem (4.4.6). This also holds in the non-finite
case: if G is an affine algebraic group, the stabilizers of the action are finite
and reduced, and the action map G × U → U × U is proper, then [U/G] is
a separated Deligne–Mumford stack (Theorem 3.6.4) and the existence of a
geometric quotient follows from the Keel–Mori Theorem.

GIT studies the case where G is linearly reductive1 but not necessarily finite.
GIT allows for the possibility of points u ∈ U where the stabilizer Gu may not
be finite and the orbit Gu may not be closed, e.g., Gm acting on A1 . We have
already covered Affine GIT (6.5.8): if G on an affine k-scheme Spec A, then there is
a commutative diagram

Spec A
π
e

 )
[Spec A/G]
π / (Spec A)//G := Spec AG

where π : [Spec A/G] → Spec AG is a good moduli space. The map π e : Spec A →
Spec AG is a categorical quotient and identifies two k-points if and only if their orbit
closures intersect; we will call this map a good quotient.

7.1.1 The definition and first properties

Definition 7.1.1 (Good quotients). Given an action of a linearly reductive group
G over an algebraically closed field k on an algebraic space U over k, a G-invariant
map π
e : U → X is a good quotient 2 if
(1) OX → (π∗ OU )G is an isomorphism (where (π∗ OU )G (V ) = Γ(UV , OUV )G for
an étale X-scheme V ) and
(2) π
e is affine.
The good quotient of U by G is often denoted as U//G = X.

Remark 7.1.2. The map π e : U → X is a good quotient if and only if π : [U/G] → X

is a good moduli space. To see the equivalence, we may assume that X = Spec B is
affine since both properties are étale local (Lemma 6.5.22(1)). For (⇒), U = Spec A
is also affine and B = AG , and thus [Spec A/G] → Spec AG is a good moduli space.
To see (⇐), observe that since U → [U/G] is affine and π∗ is exact on quasi-coherent
sheaves, the pushforward π e∗ is exact on quasi-coherent sheaves and thus π e is affine
by Serre’s Criterion for Affineness (4.5.16).

Proposition 7.1.3. Let G be a linearly reductive group over an algebraically closed

e : U → X is a good quotient, then
field k acting on an algebraic space U over k. If π
(1) π
e is surjective and the image of a closed G-invariant subscheme is closed. The
same holds for the base change T → X by a morphism from a scheme;
(2) for closed G-invariant closed subschemes Z1 , Z2 ⊆ U , im(Z1 ∩ Z2 ) = im(Z1 ) ∩
im(Z2 ). In particular, for u1 , u2 ∈ U (k), π e(u2 ) if and only if Gu1 ∩
e(u1 ) = π
1 GIT can be developed in the more general setting of actions by reductive algebraic groups; see

Remark 6.5.12.
2 In [Ses72, Def. 1.5], for any affine algebraic group G, a good quotient by G is defined as an affine
∼
G-invariant morphism π e : U → X such that OX → (π∗ OU )G and properties Proposition 7.1.3(1)–(2)
holds. When G is linearly reductive, these notions coincide.

406
 Gu2 ̸= ∅, and π
e induces a bijection between closed G-orbits in U and k-points
of X;
(3) if U is noetherian, so is X. If U is finite type over k, then so is X, and for
every coherent OU -module F with a G-action, (π∗ F )G is coherent; and
(4) π
e is universal for G-invariant maps to algebraic spaces.

Proof. This follows from Theorem 6.5.6 as [U/G] → X is a good moduli space.

Remark 7.1.4 (Semistable reduction in GIT). Since [U/G] → X is universally

closed (Theorem 6.5.6(1)), it satisfies the valuative criterion for universal closedness
(Theorem 3.8.7). This translates into the following: for every DVR R over k with
fraction field K and every map Spec R → X with a lift η : Spec K → U , there exists
an extension R → R′ of DVRs, an element g ′ ∈ G(K ′ ) over the fraction field of R′ ,
and a lift in the commutative diagram

g ′ ·η|K ′

&
5/ U
η
Spec K ′ Spec K
π
e
  
Spec R′ / Spec R / X.

In fact, if R = k[[x]], it can be arranged that R → R′ is finite; see [Mum77, Lem. 5.3]
and [AHLH23, Thm. A.8].

7.2 Projective GIT

Ich habe mich überzeugt, dass auch die Theologie ihre Vorzüge hat.

Paul Gordan

We cover the projective case of GIT: if G acts on a G-invariant closed subscheme

U ⊆ P(V ), then the semistable locus (Definition 7.2.1) has a projective good quotient
U ss //G (Theorem 7.2.6).

7.2.1 Semistability and stability

Definition 7.2.1. Let G be a linearly reductive group acting on a projective k-
scheme U and let OU (1) be an ample line bundle with a G-action. The semistable
and stable loci are defined as

U ss := {u ∈ U | there exists f ∈ Γ(U, OU (d))G with d > 0 such that f (u) ̸= 0},
there exists f ∈ Γ(U, OU (d))G with d > 0 such that
 

 

−f (u) ̸= 0,

 

 
s
U := u ∈ U −the orbit Gu ⊆ Uf is closed, and .
−the function U → Z, x →
7 dim G is constant in an open
 
x

 

neighborhood of u3

 


3 Since the function x 7→ dim Gx is upper semicontinuous, this condition is automatic if dim Gu =
0.

407
 A point u ∈ U is called semistable (resp., stable) if u ∈ U ss (resp., u ∈ U s ).4
Remark 7.2.2 (G-equivariant projective embeddings). If U is proper over k, the data
an ample line bundle OU (1) with a G-action, i.e., a very ample G-linearization (see
§B.1.28), is equivalent to the data of a G-equivariant embedding U ,→ P(V ) where V
is a finite dimensional G-representation. We stress that the stable and semistable loci
depend on the choice of G-equivariant embedding U ,→ P(V ): different embeddings
give different semistable locus U ss and different quotients U ss //G. In fact, the same
very ample line bundle on U may have different G-actions and embeddings. When
G is semisimple, then G has no non-trivial characters G → Gm , and a line bundle
on U has a unique G-action if it has any G-action. On the other hand, when U is
a normal projective variety, then every line bundle L has a positive tensor power
L⊗n that has a G-action by Sumihiro’s Theorem on Linearizations (B.1.30). For
example, O(1) on Pn does not have a PGLn+1 -action, but O(n + 1) does.
Remark 7.2.3 (Non-projective case). If U is an arbitrary k-scheme (or k-algebraic
space) and L is a line bundle with G-action, then the definition of semistability and
stability, with the additional requirement that each complement Uf is affine, for
each invariant section f in Definition 7.2.1 extends. If U is projective and OU (1) is
ample, the Uf is automatically affine.
Remark 7.2.4 (Nullcone). The nullcone N b ⊆ A(V ) is by definition the affine cone
over U ∖ U : it is the closed G-invariant locus consisting of points u
ss
b∈U b in the
affine cone such that f (b
u) = 0 for every non-constant G-invariant polynomial on
A(V ).

7.2.2 The first fundamental theorem of GIT

Given
La G-equivariant closed subscheme U ⊆ P(V ), we consider the graded ring
R = d≥0 Γ(U, OU (d)) and the morphism

e : U ss → U ss //G := Proj RG .
π (7.2.5)

Note that U ss may be empty in which case Proj RG is the empty scheme. If U ss is
non-empty, it is precisely the locus where the rational map Proj R 99K Proj RG is
defined.
Theorem 7.2.6. Let G be a linearly reductive group over an algebraically closed
field k. Let U ⊆ P(V ) be a G-equivariant closed subscheme where V is a finite
dimensional G-representation. Then there is a cartesian diagram
 
Us / U ss  /U

□ π
 
e

U s /G  / U ss //G

where U s /G ⊆ U ss //G is an open subscheme, the map π e of (7.2.5) is a good quotient,

and the restriction π e|U s : U s → U s /G is a geometric quotient. Moreover, U ss //G is
projective with an ample line bundle L such that π e∗ L ∼
= OU (N ) for some N .
If in addition the action of G on U has generically finite stabilizers, then the
action of G on U s is proper (i.e., the action map G × U s → U s × U s is proper) or
in other words [U s /G] is separated.
4 In the literature, a non-semistable point u ∈ U ∖ U ss is sometimes called ‘unstable’ if it is not

semistable; we avoid this potentially misleading terminology.

408
Proof. Since U is projective, R = d≥0 Γ(U, OU (d)) is finitely generated over k.
L

Thus by Corollary 6.5.8(3), RG is also finitely generated over k and U ss //G = Proj RG
is projective. As localization commutes with taking invariants, (RG )(f ) = (R(f ) )G
for every homogeneous element f ∈ RG of positive degree. We thus have a cartesian
diagram
 / U ss  
Uf = Spec R(f )  /U

□ π
e
  
Uf //G = (U ss //G)f  / U ss //G.

Since the property of being a good quotient is Zariski local and since the loci
(U ss //G)f cover U ss //G, we conclude that π
e : U ss → U ss //G is a good quotient. By
construction, there is an integer N such that L := OU ss //G (N ) is an ample line
bundle which pulls back to OU (N )|U ss .
To show that U s → U s /G is a geometric quotient, it suffices to show that every
G-orbit in U s is closed. Since the dimension of the stabilizer increases under orbit
degeneration, it in fact suffices to show that the dimension of the stabilizers in U s is
locally constant. Every point u ∈ U s has by definition an open neighborhood V ⊆ U
such that dim Gv = dim Gu for all v ∈ V . Since dim G = dim Gv + dim Gv, we see
that the dimension of the orbit is constant on V . Finally, if there is a dense open
subset of U which has dimension 0 stabilizers, then it follows from the definition
of stability that every u ∈ U s has a finite (possibly non-reduced) stabilizer. Since
[U s /G] → U s /G is also a good moduli space and [U s /G] has quasi-finite diagonal,
Proposition 6.5.33 implies that [U s /G] is separated.

7.2.3 First examples

Example 7.2.7. Given Gm acting on P2 via t · [x : y : z] = [tx : t−1 y : z], the
semistable locus is the complement of V (xy, z) = {[0 : 1 : 0], [1 : 0 : 0]} and the good
quotient is (P2 )ss → Proj k[xy, z] = P1 . The fiber over xy = 0 is the union of three
orbits and its complement in (P2 )ss is the stable locus. Observe that the restriction
to z =
̸ 0 is the good quotient A2 → A1 , given by (x, y) 7→ xy, while the fiber over
z = 0 is the line at infinity with [0 : 1 : 0] and [1 : 0 : 0] removed.

Example 7.2.8. Consider the diagonal action of SL2 on X = (P1 )4 and the
SL2 -equivariant Segre embedding

(P1 )4 → P15 , ([x1 : y1 ], . . . , [x4 : y4 ]) 7→ [x1 x2 x3 x4 : · · · : y1 y2 y3 y4 ].

This L
corresponds to the SL2 -linearization of L := O(1) ⊠ · · · ⊠ O(1). The invariant
ring d≥0 Γ(X, L⊗d ) is generated in degree 1 by the generalized cross ratios

I1 = (x1 y2 − x2 y1 )(x3 y4 − x4 y3 )
I2 = (x1 y3 − x3 y1 )(x2 y4 − x4 y2 )
I3 = (x1 y4 − x4 y1 )(x2 y3 − x3 y2 )

with the linear relation I1 − I2 + I3 = 0. The invariant ring is k[I1 , I2 ] and the
quotient X ss // SL2 = P1 . The semistable locus X ss consists of tuples where at most
two points are equal, while the stable locus consists of tuples of distinct points.

409
 Figure 7.2.9: 4 unordered points up to projective equivalence

An ordered tuple (p1 , . . . , p4 ) of distinct points is mapped to the cross ratio

(p1 − p2 )(p3 − p4 )
.
(p1 − p3 )(p2 − p4 )

In particular, two stable tuples are projectively equivalent (i.e., in the same SL2
orbit) if and only if they have the same cross ratio. The complement X ss ∖ X s
contains 3 closed orbits: the SL2 -orbits of (0, 0, ∞, ∞), (0, ∞, 0, ∞), and (0, ∞, ∞, 0).
Tuples such as (0, 0, 1, ∞) or (1, ∞, 0, 0) have non-closed SL2 -orbits in X ss with
SL2 ·(0, 0, ∞, ∞) in the orbit closure. See Example 7.5.1 for the more general case
of n ordered points in P1 .

7.2.4 Other GIT setups

Exercise 7.2.10 (Affine GIT with respect to a character). Let U = Spec A be a finite
type scheme over an algebraically closed field k with an action of a linearly reductive
group G specified by a coaction σ : A → Γ(G, OG ) ⊗ A. Let χ : G → Gm = Spec k[t]t
be a character and let OU (χ) be the G-action on the trivial line bundle OU determined
by χ. The complement Uf of any invariant section is affine, and thus the definitions
of semistability and stability (Definition 7.2.1 and Remark 7.2.3) with respect to
OU (χ) translate to:

there exists f ∈ A such that f (u) ̸= 0

 
U ss := u ∈ U
and σ(f ) = χ∗ (t)d ⊗ f for d > 0
there exists f ∈ A such that f (u) ̸= 0, σ(f ) = χ∗ (t)d ⊗ f 
 

U s := u ∈ U for d > 0, the orbit Gu ⊆ Uf is closed, and the function .
x 7→ dim Gx is constant in an open neighborhood of u
 

Defining U ss //G := Proj d≥0 Ad where Ad = {f ∈ A | σ(f ) = χ∗ (t)d ⊗ f }, show

L
that the conclusion of Theorem 7.2.6 holds except that U ss //G is projective over
AG = A0 (rather than k).

For example, under the scaling Gm -action on U = An and with respect to the
identity character χ = id, then U ss = U s = An ∖ 0 and the quotient is Pn−1 .

Exercise 7.2.11 (Projective GIT over an affine). Let U be a projective scheme

over an affine finite type k-scheme Spec B, where k is an algebraically closed field.
Let G be a linearly reductive group acting compatibly on U and Spec B. Suppose
that there is a G-equivariant embedding U ,→ P(E), where E is a vector bundle on
Spec B with a G-action. Show that the conclusion of Theorem 7.2.6 holds except
that U ss //G is projective over B G (rather than k).

410
Remark 7.2.12 (Symplectic reduction). There is an interesting connection between
GIT and symplectic geometry. Let G be a reductive group over C acting on a smooth
projective variety U ⊆ P(V ), where V is an n + 1 dimensional G-representation. Let
ω be a symplectic form on U , and let K ⊆ G be a maximal compact subgroup K
and k its Lie algebra. There is a moment map

µ : U → k∨

which is K-equivariant with respect to the coadjoint action on k∨ and satisfies

dµ(x)(ξ) · a = ωx (ξ, vx ) for u ∈ U , ξ ∈ Tx U , and a ∈ k, where vx is the vector field
on U obtained by the infinitesimal action of K on U . Then

u ∈ U is semistable ⇐⇒ Gu ∩ µ−1 (0) ̸= ∅

and the inclusion µ−1 (0) ,→ U induces a homeomorphism µ−1 (0)/K → U ss //G. See
[MFK94, §8].

7.3 The Cartan Decomposition and the Destabiliza-

tion Theorem
The Cartan Decomposition (7.3.1) is the key algebraic input to prove the Destabi-
lization Theorem (7.3.6), which in turn will be applied in the next section to prove
the Hilbert–Mumford Criterion (7.4.4).

7.3.1 The Cartan Decomposition

The Cartan Decomposition–sometimes referred to as the Iwahori Decomposition or
the Cartan–Iwahori–Matsumoto Decomposition—states than an element g ∈ G(K)
over the fraction field K of a DVR R can be multiplied on the left and right by
elements of G(R) such that it is induced from a one-parameter subgroup. For a
one-parameter subgroup λ : Gm → G, we denote by λ|K ∈ G(K) the image of the
composition
λ
Spec K → Gm −→G
where the first map is defined by the k-algebra map k[t]t → K taking t to a
uniformizer in R.

Theorem 7.3.1 (Cartan Decomposition). Let G be a reductive group over an

algebraically closed field k. Let R be a complete DVR over k with residue field k and
fraction field K. Then for every element g ∈ G(K), there exists h1 , h2 ∈ G(R) and
a one-parameter subgroup λ : Gm → G such that

g = h1 λ|K h2 .

Proof. We have already proved this in Proposition 6.10.30, where we also showed
that the existence of Cartan Decompositions characterizes reductivity. For classical
proofs, see [IM65, Cor. 2.17], [Ses72, Thm. 2.1], and [BT72, §4].

Remark 7.3.2 (Equivalent formulation). Let T ⊆ G be a maximal torus. The above

theorem is equivalent to the identity

G(K) = G(R)T (K)G(R).

411
To see how the decomposition implies the above identity, choose h ∈ G(R) such that
hλ|K h−1 ∈ T (K). Then

g = h1 λ|K h2 = (h1 h−1 ) (hλ|K h−1 ) (hh2 ) .

| {z } | {z } | {z }
∈G(R) ∈T (K) ∈G(R)

∼ Gr
Conversely, suppose g = h1 th2 for h1 , h2 ∈ G(R) and t ∈ T (K). If we write T = m
and π ∈ R as the uniformizing parameter, then t = (u1 π , . . . , ur π ) for units
d1 dr

ui ∈ R× and integers di ∈ Z. After replacing h1 with h1 · (u1 , . . . , ur ), we can

write g = h1 λ|K h2 where λ : Gm → T ⊆ G is the one-parameter subgroup given by
t 7→ (td1 , . . . , tdr ).
Remark 7.3.3 (Case of GLn ). The Cartan Decomposition for GLn can be estab-
lished by an elementary linear algebra argument. Let g = (gij ) ∈ GLn (K). After
performing row and column operations, we can assume that g1,1 = π d has minimal
valuation among the gij , where π ∈ R is a uniformizer. For each k ≥ 2, we write
gk,1 = uπ e . Now perform the row operations where the nth row rn is exchanged for
rn − uπ e−d r1 . In this way, we can arrange that gk,1 = 0 for k ≥ 2. By performing
analogous column operations, we can also arrange that g1,k = 0 for k ≥ 2. The
statement is thus established by induction.
Exercise 7.3.4.
(a) Let X ⊆ P(V ) be a Gm -equivariant locally closed subscheme where V is a
finite dimensional Gm -representation. Show that [X/Gm ] is separated if and
only if X has no Gm -fixed points. In other words, the diagonal [X/Gm ] →
[X/Gm ] × [X/Gm ] is finite if and only if it is quasi-finite.
(b) Let G be a reductive group acting on an algebraic space X over k. Show that
[X/G] is separated if and only if for every one-parameter subgroup λ : Gm → G,
the induced quotient stack [X/Gm ] is separated.
Hint: Verify the valuative criterion by applying the Cartan Decomposition.
Remark 7.3.5. Exercise [Link].4 does not holf if Gm is replaced with a general
linearly reductive group; a counterexample is given by a free action of SL2 on a
quasi-affine variety (see Exercise 3.9.4(d)).

7.3.2 The Destabilization Theorem

Theorem 7.3.6 (Destabilization Theorem). Let G be a reductive group over an
algebraically closed field k acting on an affine scheme X of finite type over k.
Given x ∈ X(k), there exists a one-parameter subgroup λ : Gm → G such that
x0 := limt→0 λ(t) · x exists and has closed G-orbit.
Proof. Let R = k[[t]] with fraction field K = k((t)). We can choose an element
g ∈ G(K) and a commutative diagram

Spec _ K / Gx
_

 
Spec R
g
e
/X

g
where the top map is given by the composition Spec K −
→ G → Gx and such that
y := ge(0) has closed G-orbit. By the Cartan Decomposition (7.3.1), there exists

412
h1 , h2 ∈ G(R) and a one-parameter subgroup λ : Gm → G such that h1 g = λ|K h2 .
By applying the general fact that for a ∈ G(R) and b ∈ X(R), (a · b)(0) = a(0) · b(0)
to h1 ∈ G(R) and ge ∈ X(R), we obtain that
lim λ(t)h2 (t) · x = lim h1 (t)g(t) · x = h1 (0) · ge(0) = h1 (0) · y ∈ Gy.
t→0 t→0

We claim that the related but possibly different limit limt→0 λ(t)h2 (0) · x exists and
is also contained in the closed orbit Gy. Once this is established, the theorem would
be established by using the one-parameter subgroup h2 (0)−1 λh2 (0):
lim (h2 (0)−1 λh2 (0))(t) · x = h−1
2 (0) · lim λ(t)h2 (0) · x ∈ Gy.
t→0 t→0

First, to see that limt→0 λ(t)h2 (0) · x exists, we may apply Proposition B.1.18(1)
below to reduce to the case that X = A(V ) is a G-representation. We may choose
a basis of V ∼ = kn such that the λ-action has weights λ1 , . . . , λn . We may also
write h2 · x = (a1 , . . . , an ) ∈ X(R) with each ai ∈ k[[t]] and further decompose
ai = ai (0) + a′i with a′i ∈ (t). Since
lim λ(t)h2 (t) · x = lim tλ1 (a1 (0) + a′1 ), . . . , tλn (an (0) + a′n ) (7.3.7)


t→0 t→0

exists, we see that for each i with λi < 0, we must have that ai (0) = 0, which in
turn implies that limt→0 λ(t)h2 (0) · x exists.
Finally, to see that this limit lies in Gy, we may apply Proposition B.1.18(2) to
obtain a G-equivariant map f : X → A(W ) such that f −1 (0) = Gy. We are thus
reduced to showing that limt→0 λ(t)h2 (0) · f (x) = 0. By computing the limit in
(7.3.7), the same argument as above shows that that since limt→0 λ(t)h2 (t) · f (x) = 0
implies limt→0 λ(t)h2 (0) · f (x) = 0. See also [GIT, p. 53] and [Kem78, Thm. 1.4].
Corollary 7.3.8 (Destabilization Theorem II). Let X be an algebraic stack of
finite type over an algebraically closed field k with affine diagonal. Let x ⇝ x0 be
a specialization of k-points such that the stabilizer Gx0 is linearly reductive. Then
there exists a morphism [A1 /Gm ] → X representing the specialization x ⇝ x0 .
Proof. The Local Structure Theorem (6.7.1) yields an étale morphism [Spec A/Gx0 ] →
X and a point w0 mapping to x0 . After possibly replacing Spec A with a Gx0 -invariant
affine subscheme, we can assume that w0 is a closed point. The specialization x ⇝ x0
lifts a specialization w ⇝ w0 in [Spec A/Gx0 ], and we can choose a representative
we ∈ Spec A of the orbit corresponding to w. The Destabilization Theorem (7.3.6)
gives a one-parameter subgroup λ : Gm → G such that w e exists
e0 = limt→0 λ(t) · w
and has closed orbit. By Affine GIT (6.5.8), there is a unique closed orbit in
Gw e and thus w e0 ∈ Spec A maps to w0 . The Gm -equivariant extension A1 → X
of t 7→ λ(t) · we defines a morphism of algebraic stacks [A1 /Gm ] → [Spec A/Gx0 ]
such that the image of the specialization 1 ⇝ 0 is w ⇝ w0 . The composition
[A1 /Gm ] → [Spec A/Gx0 ] → X yields the desired map.

7.4 The Hilbert–Mumford Criterion

A word of warning - and apology. There are several thousand formulas
in this paper which allow one or more ‘sign-like ambiguities’: i.e., alter-
nate and symmetric but non-equivalent reformulations. These occur in
definitions and theorems. I have made a superhuman effort5 to achieve
consistency and even to make correct statements: but I still cannot guar-
antee the result.
David Mumford [Mum66b, p. 288]

413
7.4.1 The Hilbert–Mumford index
The stable and semistable locus can often be effectively computed using the Hilbert–
Mumford Criterion (7.4.4). To set up the formulation, let U ⊆ P(V ) be a G-
equivariant closed subscheme where V is a finite dimensional G-representation, and let
u ∈ U be a k-point with a lift u
e ∈ A(V ). Given a one-parameter subgroup λ : Gm →
G, we can choose a basis V ∼ = kn such that λ(t) · (v1 , . . . , vn ) = (td1 v1 , . . . , tdn vn ).
Definition 7.4.1 (Hilbert–Mumford index). The Hilbert–Mumford index of u with
respect to λ is
µ(u, λ) := max −di . (7.4.2)
ui ̸=0
i,e

This definition depends on the ample G-line bundle L defining the projective
embedding U ⊆ P(V ). To emphasize this dependence, we sometimes write the
Hilbert–Mumford index as µL (u, λ). The Hilbert–Mumford index can be equivalently
defined as follows: if u0 = limt→0 λ(t) · u ∈ P(V ) (which exists since P(V ) is proper),
then Gm fixes u0 and µ(u, λ) is the opposite of the weight of the induced Gm -action
on the line Lu0 ⊆ V classified by u0 .
Remark 7.4.3. From the definition of the Hilbert–Mumford index, we see that
(a) limt→0 λ(t) · u
e exists if and only if µ(u, λ) ≤ 0,
(b) limt→0 t · u
e = 0 if and only if µ(u, λ) < 0, and
(c) µ(gx, gλg −1 ) = µ(x, λ).

7.4.2 The second fundamental theorem in GIT

Theorem 7.4.4 (Hilbert–Mumford Criterion). Let G be a linearly reductive group
over an algebraically closed field k acting on a G-equivariant closed subscheme
U ⊆ P(V ), where V is a finite dimensional G-representation. Let u ∈ P(V ) be a
e ∈ A(V ). Then
k-point with a lift u

u ∈ U ss ⇐⇒ 0 ∈
/ Ge
u
⇐⇒ lim λ(t) · u
e ̸= 0 for all λ : Gm → G
t→0
⇐⇒ µ(u, λ) ≥ 0 for all λ : Gm → G.

If in addition the action of G on U has generically finite stabilizers, then

u ∈ U s ⇐⇒ Ge
u ⊆ A(V ) is closed
⇐⇒ µ(u, λ) > 0 for all non-trivial λ : Gm → G.

Remark 7.4.5. The criterion that is now referred to as the Hilbert–Mumford Criterion
was first developed by Hilbert in [Hil93, § 15-16] and then adapted by Mumford in
[GIT, p. 53]. It holds more generally when G is reductive.
Proof. For semistability, the first (⇒) implication is clear: if 0 ∈ Ge u, then for
every non-constant invariant function, we have that f (e u) = f (0) = 0; hence u ∈ /
U ss . For the converse, if 0 ∈ u, then 0 and Ge
/ Ge u are disjoint closed G-invariant
subschemes of A(V ). Therefore their images in A(V )//G = Spec(Sym∗ V ∨ )G are
disjoint (Corollary 6.5.8(2)). We may thus find an invariant function f ∈ (Sym∗ V ∨ )G
with f (0) = 0 and f (eu) ̸= 0 which we may assume to be homogeneous of positive
degree, i.e f ∈ Symd V ∨ = Γ(P(V ), O(d)) for d > 0. In the second equivalence, (⇒)
5I have not made such a superhuman effort and also don’t guarantee the correctness of all signs.

414
is again clear: if there is a λ such that limt→0 λ(t) · u
e = 0, then 0 ∈ Ge u. Conversely,
if 0 ∈ Geu, the Destabilization Theorem (7.3.6) provides a one-parameter subgroup
λ such that the limit of u under λ is 0. The third equivalence follows from the
definition of the Hilbert–Mumford index (see Remark 7.4.3).
For stability, we may assume that u ∈ U ss ; otherwise 0 is in the closure of Ge
u and
thus Ge u is not closed. By definition, there is an invariant section f ∈ Γ(U, O(d))G
of positive degree not vanishing at u. After possibly increasing d, we can arrange
that f extends to an invariant section f ∈ Γ(P(V ), O(d))G : this follows from the
exact sequence 0 → IU → OP(V ) → OU → 0 using the vanishing of H1 (P(V ), IU (N ))
for N ≫ 0 and the exactness of taking invariants (i.e., the linear reductivity of G).
We may thus view f as a homogeneous polynomial of degree d on A(V ). Letting
α = f (eu), we have a commutative diagram

G
Ψu
e
/ V (f − α)   / A(V )

Ψu
# 
P(V )f

where Ψu (g) = g · u and Ψue (g) = g · u e. By assumption, we have that dim Gu =

dim Gue = 0 so both stabilizers are finite, thus proper. By Exercise 3.8.6(b), Gu ⊆
P(V )f (resp., Ge u ⊆ A(V )) is closed if and only if Ψu (resp., Ψue ) is proper. On
the other hand, V (f − α) → P(V )f is proper, and thus Ψu is proper if and only
if Ψue is. Thus Gu ⊆ Uf is closed if and only if Ge u ⊆ A(V ) is closed giving the
first equivalence. For the second equivalence, if Ge u is not closed, then there exists
a one-parameter subgroup λ : Gm → G such that limt→0 λ(t) · u e exists and is not
contained in Ge u. This gives a non-trivial λ with µ(u, λ) ≤ 0. Conversely, if Ge u is
closed, then Ψue is proper and therefore for every non-trivial λ, the map Gm → A(V ),
defined by t 7→ λ(t) · u
e, is also proper, which implies that limt→0 λ(t)e
u does not exist
and thus µ(u, λ) > 0.

7.4.3 Variants of the Hilbert–Mumford Criterion

The Hilbert–Mumford Criterion (7.4.4) translates to a stack-theoretic criterion for a
point u ∈ [U/G] to be semistable, i.e., u is contained in the open substack [U ss /G].
The data of a G-equivariant embedding U ⊆ P(V ) corresponds to a line bundle L
with on [U/G] such that the pullback of L under U → [U/G] is very ample. Since
the stable and semistable locus are G-invariant, they define open substacks of [U/G].
The data of a point u ∈ U (k) and a one-parameter subgroup λ : Gm → G up to
conjugation is classified by a map fu,λ : [A1 /Gm ] → [U/G] such that the induced
map
0 fu,λ
BGm ,→ [A1 /Gm ] −−−→ [U/G] → BG
corresponds λ (see Proposition 6.9.1). The Hilbert–Mumford index is µ(u, λ) =
∗
− wt(fu,λ L)|BGm .

Corollary 7.4.6 (Hilbert–Mumford Criterion). Let G be a linearly reductive group

over an algebraically closed field k acting on a projective k-scheme U . Let L be a line
bundle on [U/G] whose pullback to U is very ample. Then u ∈ [U/G] is semistable
if and only if wt((f ∗ L)|BGm ) ≥ 0 for all maps

f : [A1 /Gm ] → [U/G], with f (1) ≃ u.

415
If in addition the action of G on U has generically finite stabilizers, then u is stable if
and only if wt((f ∗ L)|BGm ) > 0 for all maps f : [A1 /Gm ] → [U/G] such that f (1) ≃ u
and the induced map Gm → Gf (0) on stabilizers is non-trivial.

Exercise 7.4.7 (Affine Hilbert-Mumford Criterion). Let G be a linearly reductive

group over an algebraically closed field k acting on an affine scheme U = Spec A of
finite type. Let χ : G → Gm be a character, and let U ss and U s be the semistable
and stable locus with respect to χ as defined in Exercise 7.2.10. For u ∈ U (k), show
that
u ∈ U ss ⇐⇒ for all one-parameter subgroups λ : Gm → G such that
lim λ(t) · u exists, ⟨χ, λ⟩ ≥ 0 ,
t→0

where ⟨−, −⟩ is the natural pairing of characters and one-parameter subgroups. If

in addition the action of G on U has generically finite stabilizers, show that u ∈ U s
if and only if the same condition holds with strict inequality ⟨χ, λ⟩ > 0.
Hint: Consider the action of G on U × A1 induced by χ defined by g · (u, z) =
(g · u, χ(g)−1 · z), and show that u ∈
/ U ss if and only if G · (u, 1) ∩ (U × {0}) ̸= ∅. Use
the Destabilization Theorem (7.3.6) to show that this is equivalent to the existence
of a one-parameter subgroup λ such that

lim λ(t) · (u, 1) = lim (λ(t) · u, t−⟨χ,λ⟩ ) ∈ U × {0}.

t→0 t→0

7.5 Examples in GIT

A good stock of examples, as large as possible, is indispensable for a thor-
ough understanding of any concept, and when I want to learn something
new, I make it my first job to build one.

Paul Halmos

See Example 6.5.3–6.5.4 for examples of Affine GIT, and Example 7.2.8–7.2.7 for
elementary examples of Projective GIT.

7.5.1 Ordered points in projective space

Example 7.5.1. Consider the diagonal action of SL2 on X = (P1 )n , and consider
n
the SL2 -equivariant Segre embedding (P1 )n → P2 −1 defined by the line bundle
O(1) ⊠ · · · ⊠ O(1) with SL2 -action. We claim that

X s = {(p1 , . . . , pn ) | for all q ∈ P1 , #{i | pi = q} < n/2},

X ss = {(p1 , . . . , pn ) | for all q ∈ P1 , #{i | pi = q} ≤ n/2}.

To see this, let (p1 , . . . , pn ) ∈ X(k) and λ : Gm → SL2 be a one-parameter sub-

group. There
 −1 exists  g ∈ SL2 (k) such that gλg
−1
= λd0 for some d ∈ Z where
t 0
λ0 (t) = . We can assume d ≥ 0 as the case d < 0 is handled sim-
0 t
ilarly. Since µ(x, λ) = µ(gx, λd0 ) = dµ(gx, λ0 ), it suffices to compute µ(gx, λ0 ).
Since µ(−, λ0 ) is symmetric with respect to the Sn -action, we can assume that
, pn ) withQpk , . . . , pn ̸= 0. A coordinate of the Segre embed-
gx = (0, . . . , 0, pk , . . .Q
ding is of the form ( i∈Σ xi )( i∈Σ / yi ) for a subset Σ ⊆ {1, . . . , n}, and its weight

416
is n − 2(#Σ). The coordinate, where gx is nonzero, that has largest weight is
y1 · · · yk xk+1 · · · xn with weight 2k − n. Thus µ(gx, λ0 ) = n − 2k. Therefore, if no
more than (resp., less than) n/2 of the points pi are the same, then x is semistable
(resp., stable) if and only if n ≥ 2k (resp., n > 2k). Conversely, if more than (resp.,
at least) n/2 of the same, then after translating by an element of SL2 and using
the symmetry of the Sn -action, we can write u = (0, . . . , 0, pk , . . . , pn ) with k > n/2
(resp., k ≥ n/2) and λ0 = diag(t−1 , t) destabilizes u.
If n is odd, then X ss = X s and X ss → X ss // SL2 is a geometric quotient. If n
is even, the map X ss → X ss // SL2 identifies (p1 , . . . , pn ) and (q1 , . . . , qn ) if there is
a subset Σ ⊆ {1, . . . , n} of size n/2 such that pi = pj and qi = qj for all i, j ∈ Σ;
in this case, the unique closed orbit in fiber is the orbit of the n-tuple with 0’s in

positions in Σ and ∞’s elsewhere. The complement X ss ∖ X s has precisely 12 n/2 n

closed orbits. Since Aut(P1 ) = PGL2 , the quotient X ss // SL2 = X ss // PGL2 can be
viewed as a compactification of the moduli of n ordered points in P1 up to projective
equivalence.
Exercise 7.5.2.
(a) Under the action of SL2 on the projectivization P(Γ(P1 , O(n))) ∼= Pn of binary
forms of degree n, show that the semistable (resp., stable) locus consists of
binary forms f (x, y) such that every linear factor has multiplicity less than or
equal to (resp., less than) n/2.
(b) Under the SL2 -linearization O(a1 ) ⊠ · · · ⊠ O(an ) on (P1 )n with each ai > 0,
show that the semistable (resp., stable) locus consists of tuples (p1 , . . . , pn )
such that for all q ∈ P1 (k),
X Xn
ai ≤ ( ai )/2 (resp., < ).
pi =q i=1

(c) Under the SLr+1 action on (Pr )n and the SLr+1 -linearization O(a1 ) ⊠ · · · ⊠
O(an ) with each ai > 0, show that the semistable (resp., stable) locus consists
of tuples (p1 , . . . , pn ) such that for every linear subspace W ⊊ Pr
n
X dim W + 1 X
ai ≤ ai ) (resp., < ).
r+1
pi ∈W i=1

7.5.2 Stability of hypersurfaces

Exercise 7.5.3 (Cubic curves). Consider the action of SL3 on the projective space
P(H0 (P2 , O(3))) of cubic curves in P2 . Show that the stable (resp., semistable) locus
consists of smooth (resp., at worst nodal) curves. See also [Mum77, §1.11], and
[New78, §4.4], and [HM98, pp.202-6]
Exercise 7.5.4 (Quartic curves, hard). A more involved calculation shows that
under the SL3 action on P(H0 (P2 , O(4))), a quartic curve is semistable if and only
if it doesn’t contain a triple point and is not the union of a cubic curve and an
inflection tangent line, and is stable if and only if it has at worst nodal and cuspidal
singularities. See also [Mum77, §1.12-13]. See also [HM98, §4A]
Remark 7.5.5 (Cubic surfaces). Under the action of SL4 on P(H0 (P3 , O(3))), a cubic
surface is stable (resp., semistable) if and only if it has finitely many singular points
and the singularities are ordinary double points (resp., ordinary double points or
rank two double points whose axes are not contained in the surface). See [Muk03,
Thm. 7.14].

417
7.5.3 Quiver and toric GIT
Exercise 7.5.6 (Quiver GIT). A quiver Q = (Q0 , Q1 ) is a directed graph, where Q0
is a finite set of vertices and Q1 is a finite set of arrows; there are source and target
maps s, t : Q1 → Q0 . A quiver representation of Q consists of a vector space Vi for
every i ∈ Q0 together with linear maps Lα : Vi → Vj for every arrow α : i → j. If
each Vi is finite dimensional with di = dim Vi , we say that d = (di ) is the dimension
vector of V .
Fix a dimension vector d = (di ) ∈ ZQ0 and consider the space
Y
R(Q, d) = Hom(kds(α) , kdt(α) )
α∈Q1

of representations with dimension vector d. This inherits an Q action of i GLdi via

Q
−1
(gi ) · (Lα ) = (gt(α) Lα gs(α) ). The diagonal subgroup Gm ⊆ i GLdi consisting of
tuples (t idkdi ) of scalar
Q matrices for t ∈ Gm is normal and acts trivially. Therefore
the quotient G := ( i GLdi )/Gm also acts on R(Q, d). P
For any tuple a = (ai )i∈Q0 of integers such that i ai di = 0, consider the
character Y
χa : G → Gm , (gi ) 7→ det(gi )ai .
i

Use the Affine Hilbert–Mumford Criterion (7.4.7) to show that a representation

V ∈ R(Q, d) is semistable (resp., stable) with respect to χ if and only if for every
subrepresentation W ⊆ V (i.e., subspaces Wi ⊆ Vi such that Lα (Ws(α) ) ⊆ Wt(α) ),
X
ai dim Wi ≥ 0 (resp., > ).
i

See also [Kin94, Prop. 3.1].

Remark 7.5.7 (Cox construction of toric varieties). Let X = X(Σ) be a proper toric
variety with fan Σ ⊆ NR and torus TN , where N is a lattice with dual M . Letting
Σ(1) denote the rays of the fan, the divisors Dρ associated to ρ ∈ Σ(1) generate the
class group. There is a short exact sequence

0 → M → ZΣ(1) → Cl(X) → 0.

The algebraic group G := Hom(Cl(X), Gm ) is diagonalizable (hence linearly reduc-

tive) and sits in a short exact sequence

1 → G → GΣ(1)
m → TN → 1

obtained by applying Hom(−, Gm ) to the above sequence. The group G acts naturally
on AΣ(1) .
For a cone σ ∈ Σ, let xσ := ρ∈σ(1) xρ . Define the closed G-invariant subscheme
Q

Z ⊆ AΣ(1) by the vanishing of the ideal generated by the monomials xσ as σ

varies over maximal dimensional cones; this set can also be described as the union
ρ | ρ ∈ C) where the union runs over primitive collections C ⊆ Σ(1), i.e.,
S
C V (x
subsets C such that C is not contained in σ(1) for any σ ∈ Σ and such that for any
C ′ ⊊ C, there exists σ ∈ Σ with C ′ ⊆ σ(1).
The main theorem here is that X is isomorphic to the good quotient (AΣ(1) ∖
Z)//G. This is the so-called Cox construction of X, and it gives X homogeneous
coordinates in a similar fashion to how An+1 gives homogeneous coordinates for

418
Pn = (An+1 ∖ 0)/Gm . When Σ is a simplicial fan, X is a geometric quotient
(AΣ(1) ∖ Z)/G. Moreover, the class group Cl(X) is identified with group of character
X∗ (G), and if L is an ample line bundle on X corresponding to a character χ, then
AΣ(1) ∖ Z is the semistable locus for the action of G on AΣ(1) with respect to the
character χ. See [Cox95] and [CLS11, §5].

7.5.4 GIT and birational geometry

Example 7.5.8 (Variation of GIT for Gm -actions). Consider a Gm -action on an
affine scheme X = Spec A of finite type over k. In this example, we will consider
how the GIT quotients vary as we vary the character of Gm ; see Exercise 7.2.10 for
∼
affine GIT with respect to a character. There is a bijection Hom(Gm , Gm )L
= Z and
we write χd (t) = t as the character corresponding to d ∈ Z. Let A = n∈Z An
d

be the induced grading. There are three cases for the semistable locus Xχssd with
respect to the character χd :
(i) d = 0: X ss (0) := Xχss0 = X and Xχss0 //Gm = Spec A0 .
(ii) d > 0: X ss (+) := Xχssd = X ∖ V ( n>0 An ) and Xχssd = Proj n≥0 And is
P L

independent of d; moreover X ss (0) is identified with Xχ+d with respect to the

one-parameter subgroup χd (Exercise 6.8.7).
(iii) d < 0: X ss (−) := Xχssd = X∖V ( n<0 An ) = Xχ+d and Xχssd = Proj n≥0 A−nd
P L
is independent of d.
There is a commutative diagram
 /Xo ? _ X ss (−)
X ss (+)

  
X ss (+)//Gm / X//Gm o X ss (−)//Gm

where the vertical maps are good quotients, the top maps are open immersions,
and the bottom maps are projective. The Affine Hilbert–Mumford Criterion (7.4.7)
implies that there are identifications of the stable loci with respect to χ0 , χ1 ,
and χ−1 : X s (0) = X ∖ (X ss (+) ∩ X ss (−)), X s (+) = X ss (+) = X ∖ X ss (−),
and X s (−) = X ss (−) = X ∖ X ss (+). Therefore, we see that if both X ss (+)
and X ss (−) are nonempty, then X ss (+)/Gm → X//Gm and X ss (−)/Gm → X//Gm
are isomorphisms over X s (0)/Gm , and in particular birational. We also see that
if the complements of X ss (+) and X ss (−) in X each have codimension at least
two, then the birational map X ss (+)//Gm 99K X ss (−)//Gm is an isomorphism in
codimension 2 such that the divisor O(1) (which is relatively ample over X//Gm )
pushes forward to a divisor on X ss (−)//Gm whose dual is relatively ample, i.e.,
X ss (+)//Gm 99K X ss (−)//Gm is a flip with respect to O(1).

Remark 7.5.9 (Variation of GIT). The previous example can be generalized to actions
of a linearly reductive group G. Extending the notion of algebraic independence, two
line bundles with G-actions L1 and L2 on a projective scheme X are G-algebraically
equivalent if there is a connected variety T , points t1 , t2 ∈ T (k), and a line bundle
with a G-action L on X × T such that Li ∼ = L|X×{ti } . The G-equivariant Neron–
Severi group NSG (X) of line bundles with a G-action on X up to G-algebraic
equivalence is finitely generated. The kernel of NSG (X)R → NS(X) is identified
with the rational character group X∗ (G)R . We let Eff G (X) ⊆ NSG (X)R be the cone
of G-effective linearizations, i.e., G-linearizations L such that there is a nonzero

419
invariant section of L⊗d for some d > 0, or in other words such that XLss ̸= ∅. We
also let AmpG (X) ⊆ NSG (X)R be the cone of ample G-linearizations.
The main results of Variation of GIT can be formulated as follows. The semistable
locus XLss only depends on the G-algebraic equivalence class of L. There is a
polyhedral decomposition of the cone AmpG (X) ∩ Eff G (X) defined by codimension
1 walls such that the semistable locus is constant in any open chamber. If L0 is
on a wall while L+ and L− are on opposite adjacent chambers, then there is a
commutative diagram

XLss+  / X ss o
L0
? _ X ss
L−

  
XLss+ //G / X ss //G o XLss− //G
L0

where the vertical maps are good quotients, the top maps are open immersions,
and the bottom maps are projective. If XLss+ and XLss− are non-empty, the bottom
maps are birational; when the bottom maps are isomorphisms in codimension 2,
then XLss+ //G 99K XLss− //G is a flip with respect to the line bundle O(1) on XLss+ //G,
which is relatively ample over XLss0 //G. See [Tha96] and [DH98].
Remark 7.5.10 (Mori Dream Spaces). There is an interesting connection between
the Mori program and variation of GIT. A normal Q-factorial projective variety X
is a Mori dream space if (1) Pic(X)Q = NS(X)Q , (2) the cone Nef(X) of nef line
bundles is the affine hull of finitely many semiample line bundles, and (3) there are
finitely many birational maps fi : X 99K Xi , which are isomorphisms in codimension
1, to a Q-factorial normal projective variety Xi such that the movable cone Mov(X)
is the union of fi−1 (Amp(Xi )Q ) (where a line bundle is called movable if its stable
base locus has codimension at least 2). In other words, X is a Mori dream space if
Mov(X) has a finite wall and chamber decomposition such that the projective variety
determined by the line bundle is constant within an open chamber. Equivalently, X
is a Mori dream space if Pic(X)Q = NS(X)Q and the Cox ring
M
Cox(X) := Γ(X, Ld11 ⊗ · · · ⊗ Ldnn )
(d1 ,...,dn )∈Nn

is finitely generated, where L1 , . . . , Ln is a basis for Pic(X)Q such that their affine
hull contains Eff(X)Q . If X is a Mori dream space, then X along with each birational
model Xi is a GIT quotient of the semistable locus of Spec(Cox(X)) by the torus
Gnm with respect to some character. Moreover, there is an identification of the Mori
chambers of Mov(X) with the variation of GIT chambers for the action of Gnm on
Spec(Cox(X)). See [HK00].

Example 7.5.11 (Partial desingularization). If U is a smooth variety and U → X

is a geometric quotient by a linearly reductive group, then X necessarily has finite
quotient singularities; this is a consequence of the Local Structure Theorem (4.4.13).
On the other hand, if U → X is a good quotient, then X can have worse singularities
(Section 6.5.11). Nevertheless, in the case that there is an open subset X ′ ⊆ X such
that π0 (X ′ ) → X ′ is a geometric quotient (e.g., U = V ss is the semistable locus with
respect to a G-equivariant embedding V ,→ Pn and the stable locus V s is nonempty),
there is a canonical procedure to partially resolve the singularities of X so that they

420
become finite quotient singularities. Namely, there exists a commutative diagram

Un / Un−1 / ··· / U = U0

πn πn−1 π0
  
Xn / Xn−1 / ··· / X = X0

such that:

– Each Ui+1 is a G-invariant open subscheme of a blowup BlZi Ui , where Zi

is a G-invariant smooth closed subscheme whose stabilizers are of maximal
dimension, and Ui+1 ⊆ BlZi Ui is the complement of the strict transform of
πi−1 (πi (Zi )). If U = V ss is the semistable locus of a projective variety with
respect to a G-linearization L, then Ui+1 is the semistable locus with respect
to (q ∗ L)⊗n ⊗ O(−E) for n ≫ 0, where q : BlZi Ui → Ui and E denotes the
exceptional divisor.
– The maps Xi+1 → Xi are projective and birational.
– The maps πi : Ui → Xi are good quotients by G, and the map πn : Un → Xn
is a geometric quotient. In particular, Xn has finite quotient singularities.

See [Kir85], [Rei89], and [ER21].

For a simple example of this procedure in action, consider the Gm -action on A2
with weights 1 and −1. In this case, the quotient A2 //Gm ∼ = A1 is smooth but not a
geometric quotient. The procedure tells us to take the blowup Bl0 A2 at the origin
and remove of the strict transform of V (xy). Then Gm -acts acts on the complement
U1 with finite stabilizers, and U1 → A2 is Gm -invariant birational (but neither proper
nor surjective) map inducing an isomorphism U1 /Gm → A2 //Gm on quotients.

7.6 Kempf’s Optimal Destabilization Theorem

By the Hilbert–Mumford Criterion (7.4.4), for every non-semistable point x ∈ X,
there exists a one-parameter subgroup λ such that the Hilbert–Mumford index
µ(x, λ) < 0. Kempf’s Optimal Destabilization Theorem (7.6.6) asserts that there is
a canonical one-parameter subgroup λ minimizing the normalized Hilbert–Mumford
index µ(x, λ)/ ∥λ∥. In other words, there is a ‘worst’ one-parameter subgroup that
is most responsible for the instability of x.

7.6.1 Measuring instability

Given an algebraic group G over an algebraically closed field k, we denote by X∗ (G)
the set of one-parameter subgroups Gm → G. Recall that for a torus T ∼ = Gnm ,
∼
X∗ (T ) = Z (see Example B.1.22).
n

Definition 7.6.1. A length ∥−∥ on X∗ (G) is a nonnegative real-valued function

on X∗ (G) which is conjugation invariant, i.e., gλg −1 = ∥λ∥ for λ ∈ X∗ (G) and
g ∈ G(k), and such that for every maximal torus T ⊆ G, there is a positive definite
2
integral-valued bilinear form (−, −) on X∗ (T ) with (λ, λ) = ∥λ∥ for λ ∈ X∗ (T ).

Example 7.6.2. If G = GLn , then any one-parameter subgroup λ is conjugate

to a one-parameter subgroup of the form t 7→ diag(td1 , . . . , tdn ) and we can define
∥λ∥ = d1 + · · · + d2n .
p
2

421
Example 7.6.3. For every reductive group G, there exists a length ∥−∥ on X∗ (G).
To see this, let T ⊆ G be a maximal torus and choose a positive definite integral-
valued bilinear form (−, −) on X∗ (T ), which is invariant under the conjugation
action of the Weyl group W := N (T )/T . There is a bijection X∗ (G)/G ∼ = X∗ (T )/W
between conjugacy classes of X∗ (G) under G and conjugacy classes of X∗ (T ) under W .
In other words, for every λ ∈ X∗ (G) there exists g ∈ G(k) such that gλg −1 ∈ X∗ (T ),
and moreover for any other element g ′ ∈ G(k) such that g ′ λg ′−1 ∈ X∗ (T ), then
2
gλg −1 and g ′ λg ′−1 are conjugate under W . It follows that ∥λ∥ := (gλg −1 , gλg −1 )
is well-defined.

Let X = Spec A be an affine k-scheme with the action of G and let x0 ∈ X(k)
be a point with closed orbit. For every point x ∈ X(k) with Gx0 ⊆ Gx and a
one-parameter subgroup λ : Gm → G such that limt→0 λ(t) · x exists, we define the
Hilbert–Mumford index of x with respect to λ as
−1
µ(x, λ) = − deg fx,λ (Gx0 ). (7.6.4)

where fx,λ : A1 → X is the map extending Gm → X, t 7→ λ(t) · x. Note that if

limt→0 λ(t) · x ∈ / Gx0 , then µ(x, λ) = 0.
In the projective case of a G-equivariant embedding X ,→ P(V ), we have
already defined the Hilbert–Mumford index µ(x, λ) in (7.4.2) as follows: choosing
a basis of V such that Gm acts on A(V ) = An with weights d1 , . . . , dn and a lift
b = (u1 , . . . , un ) ∈ A(V ) of x, then −µ(x, λ) is defined as the smallest di with
x
ui ̸= 0. If limt→0 λ(t) · x b exists, then this agrees with the definition in (7.6.4). To
see this, observe that the extension fxb,λ : A1 → An of the map t 7→ λ(t) · x b is the
map t 7→ (tdi ui ) and fxb−1,λ (0) = Spec k[t]/(t d
) where d is the smallest d i with ui ̸= 0.
Since µ(x, λ ) = n·µ(x, λ), it natural to consider the normalized Hilbert–Mumford
n

index
µ(x, λ)
∥λ∥
as a measure of how quickly λ(t) · x approaches the closed orbit Gx0 . The more
negative the normalized Hilbert–Mumford index is, the faster λ(t) · x approaches
Gx0 . Kempf proved that there is a one-parameter subgroup minimizing this index
and that it is unique up to conjugation.

7.6.2 Statements of Kempf ’s theorem

Given a one-parameter subgroup λ : Gm → G, recall from §B.1.4 that the centralizer
Cλ is the subgroup of elements commuting with λ, the parabolic subgroup is Pλ =
{g ∈ G | limt→0 λ0 (t)gλ0 (t)−1 exists} and the unipotent subgroup is Uλ0 = {g ∈
G | limt→0 λ0 (t)gλ0 (t)−1 = 1}. There is a short exact sequence 1 → Uλ → Pλ →
Cλ → 1, identifying Uλ as the unipotent radical of Pλ .

Theorem 7.6.5 (Kempf’s Optimal Destabilization Theorem—affine version). Let G

be a reductive group over an algebraically closed field k with a length ∥−∥ on X∗ (G).
Let X = Spec A be an affine scheme of finite type over k with an action of G. Let
x0 ∈ X(k) be a point with a closed orbit. For every point x ∈ X(k) with Gx0 ⊆ Gx,
there exists a one-parameter subgroup λ0 : Gm → G such that µ(x, λ0 )/ ∥λ0 ∥ achieves
the minimal value
 
µ(x, λ)
M (x) := inf lim λ(t) · x ∈ Gx0
λ∈X∗ (G) ∥λ∥ t→0

422
If λ′0 is another such one-parameter subgroup, then Pλ0 = Pλ′0 and λ′0 = uλ0 u−1
for a unique element u ∈ Uλ0 . Every maximal torus T ⊆ Pλ0 contains a unique
indivisible element achieving this minimum value.

The projective version below follows from applying the affine version (Theo-
rem 7.6.5) to a lift x
b ∈ A(V ) of a non-semistable point x ∈ P(V ), in which case
the closed orbit in Gbx is the fixed point 0. The projective version also holds for
reductive groups, but we restrict to linearly reductive groups as we’ve only discussed
semistability in that context.

Theorem 7.6.6 (Kempf’s Optimal Destabilization Theorem—projective version).

Let G be a reductive group over an algebraically closed field k with a length ∥−∥
on X∗ (G). Let X ⊆ P(V ) be a G-equivariant closed subscheme where V is a finite
dimensional G-representation. For every non-semistable point x ∈ X(k), there
exists a one-parameter subgroup λ0 : Gm → G such that µ(x, λ0 )/ ∥λ0 ∥ achieves the
minimal value  
µ(x, λ)
M (x) := inf
λ∈X∗ (G) ∥λ∥
If λ′0 is another such one-parameter subgroup, then Pλ0 = Pλ′0 and λ′0 = uλ0 u−1
for a unique element u ∈ Uλ0 . Every maximal torus T ⊆ Pλ0 contains a unique
indivisible element achieving this minimum value.

Definition 7.6.7. We call any λ0 satisfying Theorem 7.6.5 or Theorem 7.6.6 an

optimal destabilizing one-parameter subgroup for x, and we call M (x) < 0 the optimal
normalized Hilbert–Mumford index for x.

7.6.3 Proof of Kempf ’s theorem

The proof below is simpler when x0 ∈ Gx is a fixed point, such as in the projective
version when the closed orbit is 0 ∈ A(V ). The reader may want to keep this case in
mind.

Proof of Theorem 7.6.5. By Proposition B.1.18, we may choose finite dimensional

G-representations V and W along with G-equivariant maps

i 6 A(V )
)
X f
(7.6.8)
(
A(W ),

where i : X ,→ A(V ) is a closed immersion with i(x0 ) = 0 and f : X → A(W ) is a

morphism with f −1 (0) = Gx0 . (When x0 is a fixed point, we can take f = i in
(7.6.8).) A one-parameter
L subgroup λ : Gm L→ G induces Gm -actions on V and W ,
and thus gradings V = d∈Z Vd and W = d∈Z Wd . We define

m(i(x), λ) = min{d | the projection of i(x) to Vd is nonzero},

m(f (x), λ) = min{d | the projection of f (x) to Wd is nonzero}.

For any g ∈ G, we have the identities m(i(x), λ) = m(i(g·x), gλg −1 ) and m(f (x), λ) =
m(f (g · x), gλg −1 ). It is easy to see that if limt→0 λ(t) · x exists, then µ(x, λ) =

423
−m(f (x), λ), and that

lim λ(t) · x exists ⇐⇒ m(i(x), λ) ≥ 0,

t→0
lim λ(t) · x ∈ Gx0 ⇐⇒ m(i(x), λ) ≥ 0 and m(f (x), λ) > 0.
t→0

By the Destabilization Theorem (7.3.6), there exists λx ∈ X∗ (G) such that m(i(x), λx ) ≥
0 and m(f (x), λx ) > 0.
Case Lof a torus: Let T ⊆ G be a maximal torus containing λx . We can decompose
V = χ∈X∗ (T ) Vχ as a T -representation, where X∗ (T ) denotes the set of characters
of T . We define the state of i(x) ∈ V with respect to T as the set

StateT (i(x)) := {χ ∈ X∗ (T ) | the projection of i(x) to Vχ is nonzero}.

Likewise, we have the state StateT (f (x)) ⊆ X∗ (T ) of f (x) ∈ W with respect to

T . Let ⟨−, −⟩ be the natural pairing X∗ (T ) × X∗ (T ) → Z. For a one-parameter
subgroup λ ∈ X∗ (T ), we have identifications

m(i(x), λ) = min ⟨χ, λ⟩ and m(f (x), λ) = min ⟨χ, λ⟩.

χ∈StateT (i(x)) χ∈StateT (f (x))

We claim that the function λ 7→ m(f (x), λ)/ ∥λ∥ achieves a maximum value on the
set {λ ̸= 0 ∈ X∗ (T ) | m(i(x), λT ) ≥ 0} at a one-parameter subgroup λT , and that
any other one-parameter subgroup achieving this minimum is a positive multiple
of λT . This is precisely the conclusion of Lemma 7.6.9 below applied to the lattice
L = X∗ (T ) ∼
= Zr and the subsets of X∗ (T ) ∼= HomZ (L, Z) given by F := StateT (i(x))
and G := StateT (f (x)).
General case: For each maximal torus T ⊆ G and each g ∈ G(k), there is an
identification X∗ (T ) ∼
= X∗ (gT g −1 ), given by identifying χ ↔ gχg −1 , under which
StateT (i(x)) = StategT g−1 (i(gx)). Given a one-parameter subgroup λ ∈ X∗ (G),
we have seen that m(f (x), λ) = m(f (gx), gλg −1 ) for g ∈ G(k). We claim that,
in fact, m(f (x), λ) = m(f (x), pλp−1 ) for p ∈ Pλ . By symmetry, it suffices to
show that m(f (x), λ) ≤ m(f (x), pλp−1 ). Interpreting −m(f (x), λ) as the smallest
integer d such that limt→0 td λ(t) · f (x) ∈ A(W ) exists, we need to show that
limt→0 td pλ(t)p−1 · f (x) ∈ A(W ) exists. This follows from the computation
 
lim td pλ(t)p−1 · f (x) = lim p · λ(t)p−1 λ(t)−1 · td λ(t)f (x)




t→0 t→0

= p · lim λ(t)p−1 λ(t)−1 · lim td λ(t)f (x) .

t→0 t→0

We now show that the function λ 7→ m(f (x), λ)/ ∥λ∥ achieves a minimum value
on
Λ := {λ ∈ X∗ (G) | m(i(x), λ) ≥ 0}.
By the torus case, we know that for every g ∈ G(k) there is a minimum value on each
non-empty set X∗ (gT g −1 ) ∩ Λ, and that the minimum is determined by the subsets
of X∗ (T ) given by StategT g−1 (i(x)) ∼
= StateT (i(g −1 u)) and StategT g−1 (f (x)) ∼
=
StateT (f (g u)). Since these subsets are contained in the finite set of characters
−1

χ with Vχ ̸= 0 (resp., Wχ ̸= 0), there are only finitely many minimum values as
g ranges over G(k). Since the image of any λ ∈ X∗ (G) is contained in gT g −1 for
some g ∈ G(k), it follows that there is a global minimum value achieved by a
one-parameter subgroup λ0 ∈ Λ. We may assume that λ0 is indivisible, i.e., λ0
cannot be written as a positive multiple of another one-parameter subgroup.

424
 To establish the uniqueness, choose a maximal torus T0 ⊆ G containing λ0 . By
the torus case, λ0 ∈ X∗ (T0 ) ∩ Λ is the unique indivisible one-parameter subgroup
achieving the minimal value. For p ∈ Pλ0 , the conjugate one-parameter subgroup
pλ0 p−1 also achieves this minimal value. Since any other maximal torus T ′ ⊆ Pλ0 is
pT0 p−1 for some p ∈ Pλ0 , we see that X∗ (T ′ ) ∩ Λ also contains a unique indivisible
element achieving the minimum value. Finally, let λ1 ∈ X∗ (G) be another indivisible
element achieving the minimum value. The intersection Pλ0 ∩Pλ1 contains a maximal
torus T of G (Proposition B.1.26(d)), and we can write λT = p0 λ0 p−1 −1
0 = p1 λ1 p1
for p0 , p1 ∈ PλT . It follows that Pλ0 = PλT = Pλ1 , and that λ0 and λ1 are conjugate
by a unique element element of UλT (Proposition B.1.26(c)). See also [Kem78,
Thm. 3.4].
The argument above used the following lemma in convex geometry.
Lemma 7.6.9. Let Λ be a finite dimensional lattice, and let F and G be non-empty
finite subsets of Λ∨ = HomZ (Λ, Z). Assume that ΛR = Λ ⊗Z R has a positive definite
inner product which is integral valued on Λ. Define

fmin : ΛR → R, λ 7→ min f (λ) and gmin : ΛR → R, λ 7→ min g(λ).

f ∈F g∈G

Suppose that there exists λ ∈ ΛR such that fmin (λ) ≥ 0 and gmin (λ) > 0. Then the
function
CF := {λ ̸= 0 ∈ ΛR | fmin (λ) ≥ 0} → R
λ 7→ gmin (λ)/ ∥λ∥ ,
obtains a maximum value M . There exists a unique element λ0 ∈ CF ∩ Λ such
that M = gmin (λ0 )/ ∥λ0 ∥ and such that any other element λ ∈ CF ∩ Λ with M =
gmin (λ)/ ∥λ∥ is an integral multiple of λ0 .
Proof. The set {λ ∈ CF | gmin (λ) ≥ 1} is closed and convex, and therefore contains
a unique point λ′ closest to the origin. Since gmin (αλ′ ) = αgmin (λ′ ) for α ∈ R, we
must have that gmin (λ′ ) = 1 and that λ′ ∈ CF is the unique point with gmin (λ′ ) = 1
and gmin (λ′ )/ ∥λ′ ∥ = M .
We now argue that the ray spanned by λ′ contains an integral point. If λ′ is
in the interior of {λ ∈ CF | gmin (λ) ≥ 1}, i.e., f (λ′ ) > 0 for all f ∈ F and there
is a unique g ∈ G with g(λ′ ) = 1, then λ′ is the closest point to the origin on the
affine plane defined by g = 1. We claim that λ′ = g ∗ /⟨g ∗ , g ∗ ⟩ where g ∗ ∈ ΛR is
the unique point such that ⟨g ∗ , λ⟩ = g(λ) for all λ ∈ ΛR . Indeed, the point λ′ is
contained in the plane g = 1, and for any other point λ on this plane, we have
that ⟨λ′ , λ⟩ = 1/⟨g ∗ , g ∗ ⟩ = ⟨λ′ , λ′ ⟩. The Cauchy–Schwarz inequality implies that
⟨λ′ , λ′ ⟩2 = ⟨λ′ , λ⟩2 ≤ ⟨λ′ , λ′ ⟩⟨λ, λ⟩ so that ⟨λ′ , λ′ ⟩ ≤ ⟨λ, λ⟩. Since the inner product
and g take integral values, g ∗ ∈ Λ. We then take λ0 to be the unique indivisible
element in the ray spanned by g ∗ .
To reduce to this case, let f1 , . . . , ft ∈ F be the functions satisfying fi (λ′ ) = 0,
and let g1 , . . . , gs ∈ G be the functions satisfying gi (λ′ ) = gmin (λ′ ). Since each fi
and gj take integral values, we may restrict to the subspace
 
f1 (λ) = · · · = ft (λ) = 0
W := λ ∈ ΛR ,
g1 (λ) = · · · = gs (λ)

and the lattice W ∩ Λ. Then λ′ is in the interior of {λ ∈ CF ∩ W | gmin (λ) ≥ 1}

and thus is the closest point to the origin contained in the affine plane define by
g1 = 1.

425
Corollary 7.6.10. In the setting of Theorem 7.6.5 or Theorem 7.6.6, there is a
unique morphism f : [A1 /Gm ] → [X/G] with f (1) ≃ x and f (0) ≃ x0 .

Proof. By Proposition 6.9.1, a morphism [A1 /Gm ] → [X/G] is determined by a

one-parameter subgroup λ such that limt→0 λ(t)x ∈ Gx0 , and that λ is unique up
to conjugation by Pλ . Since any two optimal destabilizing one-parameter subgroups
are conjugate under Uλ (and thus Pλ ), the statement follows interpretation of maps
from [A1 /Gm ] into quotient stacks (Proposition 6.9.1).

7.6.4 Examples and extensions

Example 7.6.11. We revisit the SL2 -action on (P1 )n with the linearization given
n
by the Segre embedding (P1 )n ,→ P2 −1 (Example 7.5.1). The non-semistable
locus consists of tuples x = (p1 , . . . , pn ) where more than n/2 points are equal.
Suppose that precisely k > n/2 points are equal. Since the Hilbert–Mumford index
is symmetric, we can assume that the first k are equal. If λ : Gm → SL2 is a one-
parameter subgroup,
 −1  we can choose g ∈ SL2 (k) with gλg = λ0 where d ∈ Z and
−1 d

t 0
λ0 (t) = . After rescaling the norm, we can assume that ∥λ0 ∥ = 1. We
0 t
also assume that d ≥ 0 as the d < 0 case can be handled similarly. Then

µ(x, λ) µ(gx, gλg −1 )

= = µ(gx, λ0 )
∥λ∥ ∥gλg −1 ∥

This index is negative if and only if gx = {0, . . . , 0, pk+1 , . . . , pn } in which case

µ(gx, λ0 ) = n − 2k. It follows that λ0 (resp., g −1 λ0 g) is an optimal destabilizing
one parameter subgroup for gx (resp x). Observe that the parabolic Pλ0 ⊆ SL2 of
lower triangular matrices is also the stabilizer of 0 ∈ P1 , and thus Ggx ⊆ Pλ0 .

Exercise 7.6.12. Under the action of SL3 on the space P(Γ(P2 , O(3))) on ternary
cubics, the cuspidal cubic y 2 z − x3 is not semistable (see Exercise 7.5.3). Find the
optimal destabilizing one-parameter subgroup.

Exercise 7.6.13. Let G be a reductive group over an algebraically closed field k

with a length ∥−∥ on X∗ (G). Let X = Spec A be an affine scheme of finite type over
k with an action of G. Let x0 ∈ X(k) have closed G-orbit. Let x ∈ X(k) be a point
such that Gx0 ⊆ Gx, and let Px be the parabolic determined by Kempf’s Optimal
Destabilization Theorem (7.6.5).
(a) Show that for all g ∈ G(k) that gPx g −1 = Pgx .
Hint: Show that if Px = Pλ for a one-parameter subgroup λ, then Pgx = Pgλg−1 .
(b) Show that Gx ⊆ Px .
Hint: Use that for a parabolic P , NG (P ) = P (Proposition B.1.26).

The following criterion can sometimes be used to check stability/semistability by

computing Hilbert–Mumford indices only for one-parameter subgroups in a fixed
maximal torus.

Exercise 7.6.14 (Kempf–Morrison Criterion). Let G = GL(W ) or SL(W ), where W

is finite dimensional vector space over an algebraically closed field k of characteristic
0. Let X ⊆ P(V ) be a G-invariant closed subscheme, where V is a finite dimensional
G-representation. Let x ∈ X(k). Assume that there is a linearly reductive subgroup

426
H ⊆ Gx such that W decomposes as a direct sum of distinct H-representations. Let
T ⊆ G be a maximal torus compatible with this decomposition. Show that
x ∈ X ss ⇐⇒ µ(x, λ) ≤ 0 for all λ : Gm → T,
x ∈ X s ⇐⇒ µ(x, λ) < 0 for all λ : Gm → T.

Hint: If x ∈/ X ss , let λ0 : Gm → G be an optimal destabilizing one-parameter subgroup

and 0 = V0 ⊆ V1 ⊆ · · · ⊆ Vk = V be the filtration induced by the parabolic Pλ0 . Use
Exercise 7.6.13 to conclude that each Vi is H-invariant, and use the hypothesis on
the H-representation V to show that each Vi is T -invariant; thus T ⊆ Pλ0 . Apply
Kempf ’s Optimal Destabilization Theorem again to find λ in T with µ(x, λ) < 0.
/ X s , letting x
If x ∈ b ∈ A(V ) be a lift of x and x
b0 ∈ Gb
x be a point with closed orbit,
repeat the above argument using the affine version of Kempf ’s Optimal Destabilization
Theorem.
Exercise 7.6.15 (Existence of destabilizing one-parameter subgroups over a perfect
field). Let X be an affine scheme of finite type over a perfect field k, and let G be a
reductive group over k acting on X. This exercise will show that for every point
x ∈ X(k), there exists a one-parameter subgroup λ : Gm → G defined over k such
that limt→0 λ(t) · x has closed G-orbit. See also [Kem78, §4].
(1) Show that if Gal := Gal(k/k) is the geometric Galois group, then Gal acts on
the set X∗ (Gk ) of one-parameters subgroups such that X∗ (G) = X∗ (Gk )Gal .
(2) Show that there exists a length ∥−∥ on X∗ (Gk ) which is invariant under the
action of Gal.
(3) Show that the subsets {λ ∈ X∗ (Gk ) | limt→0 λ(t) · x ∈ X(k) exists} and
{λ ∈ X∗ (Gk ) | limt→0 λ(t) · x ∈ Gk x0 } are Gal-invariant where Gk x0 is
the unique closed orbit in Gk x. Moreover, show that if V and W are G-
representations as in (7.6.8), then the functions m(i(x), λ) and m(f (x), λ) are
Gal-invariant.
(4) Generalize Theorem 7.6.5 and Theorem 7.6.6 to the case when k is a perfect
field and x ∈ X(k).
In particular, if G has no non-trivial one-parameter subgroups defined over k, then
the G-orbit of any k-point is closed.
Finally, we record the following consequence of the proof of Kempf’s Optimal
Destabilization Theorem (7.6.5). This will play a key role in the proof of the HKKN
Stratification (7.7.1).
Proposition 7.6.16. Let G be a reductive group over an algebraically closed field
k with a length ∥−∥ on X∗ (G). Let X = Spec A be an affine scheme of finite type
over k with an action of G with a unique closed orbit Gx0 . Fix a maximal torus
T ⊆ G. There are finitely many one-parameter subgroups λ1 , . . . , λn ∈ X∗ (T ) and
numbers M1 , . . . , Mn ∈ R<0 such that for every point x ∈ X(k), there exists a unique
i = 1, . . . , n such that λi is an optimal destabilizing one-parameter subgroup for gx
for some g ∈ G, and such that Mi = µ(x, λi )/ ∥λi ∥.
Proof. We will use the notation of the proof of Theorem 7.6.5. For x ∈ X(k),
the unique parabolic subgroup of a optimal destabilization one-parameter sub-
group is determined by the subsets StategT g−1 (i(x)) ∼ = StateT (i(gx)) ⊆ X∗ (T ) and
StategT g−1 (f (x)) ∼
= StateT (f (gx)) ⊆ X∗ (T ) as g ranges over G(k). These subsets
are contained in the finite subset of characters χ ∈ X∗ (T ) with Vχ ̸= 0 or Wχ ̸= 0.
Thus there are only finitely many possibilities for an optimal destabilizing subgroup
of T .

427
7.7 The HKKN stratification of the unstable locus
For an action of a linearly reductive group G on a projective variety X ⊆ Pn ,
we show that the non-semistable locus X ∖ X ss admits a stratification, called the
Hesselink–Kempf–Kirwan–Ness Stratification or simply the HKKN stratification,
into locally closed subschemes according to the function

µ(x, λ)
M (x) := inf ∈ R<0 .
λ∈X∗ (G) ∥λ∥

By Kempf’s Optimal Destabilization Theorem (7.6.6), for each x ∈ X ∖ X ss , the

infimum is obtained by an optimal destabilizing one-parameter subgroup λ. The
more negative the index M (x) is, the more non-semistable or ‘unstable’ the point x
is. The strata will be indexed by pairs (λ, M ) where λ ∈ X∗ (G) and M ∈ R<0 . Like
the Białynicki-Birula Stratification (6.8.14), the HKKN stratification is a powerful
tool to compute equivariant cohomology.

7.7.1 The Hesselink–Kempf–Kirwan–Ness Stratification

For a one-parameter subgroup λ : Gm → G, recall that X λ := MorGm (Spec k, X) is
the λ-fixed locus (Definition 6.8.1) and that Xλ+ := MorGm (A1 , X) is the the attractor
locus parameterizing points x such that limt→0 λ(t) · x exists (Definition 6.8.6). We
denote by ev0 : Xλ+ → X λ the evaluation map at 0 defined by x 7→ limt→0 λ(t) · x.
If X is projective, the Białynicki-Birula Stratification (6.8.14) implies that Xλ+ is a
disjoint union of locally closed subschemes. Recall also that Cλ and Pλ denote the
centralizer and parabolic subgroups associated to λ (Definition B.1.24). The HKKN
Stratification also holds for reductive groups, but as usual we restrict to linearly
reductive groups as we’ve only developed GIT in that setting.

Theorem 7.7.1 (The HKKN Statification). Let G be a linearly reductive group over
an algebraically closed field k and a length ∥−∥ on X∗ (G). Let X ⊆ P(V ) be a G-
equivariant closed subscheme where V is a finite dimensional G-representation. For
a maximal torus T ⊆ G, there is a finite subset Σ ⊆ X∗ (T ) × R<0 and a stratification
of the non-semistable locus into G-invariant locally closed subschemes
a
X ∖ X ss = Sλ,M
(λ,M )∈Σ

such that for each (λ, M ) ∈ Σ,

(1) The locus
Yλ,M := x ∈ Xλ+ M = M (x) = µ(x, λ)/ ∥λ∥ ,


is a Pλ -invariant locally closed subscheme of X and Sλ,M := G · Yλ,M is a

G-invariant locally closed subscheme of X;
(2) a point x ∈ Yλ,M if and only if ev0 (x) = limt→0 λ(t) · x ∈ Yλ,M ∩ X λ ; thus

Zλ,M := x ∈ X λ M (x) = M = µ(x, λ)/ ∥λ∥



is a Cλ -invariant closed subscheme of Yλ,M such that Yλ,M = ev−1

0 (Zλ,M ).
(3) the natural map G ×Pλ Yλ,M → Sλ,M is finite, surjective, and universally
injective; if char(k) = 0, then G ×Pλ Yλ,M → Sλ,M is an isomorphism.

428
 (4) the locus [
Sλ′ ,M ′
(λ′ ,M ′ )∈Σ, M ′ ≤M

is closed and in particular contains S λ,M ;

(5) if X is smooth, then so is each Yλ,M ; in char(k) = 0, the strata Sλ,M are also
smooth.

Remark 7.7.2 (Strata, blades, and centers). The locus Sλ,M is called a stratum, Yλ
a blade, and Zλ,M a center (or sometimes a limit set). Stack-theoretically, there is a
stratification a
[X/G] = [X ss /G] ⨿ [Sλ,M /G],
(λ,M )∈Σ

and for each (λ, M ), there is a diagram

ev0
u
[Zλ,M /Cλ ] / [Yλ,M /Pλ ] / [X/G] (7.7.3)
i

such that ev0 ◦i = id. In characteristic 0, there is an equivalence [Yλ,M /Pλ ] ∼

=
[Sλ,M /G] and the map [Yλ,M /Pλ ] → [X/G] is a locally closed immersion.

Proof. Let X b ⊆ A(V ) be the affine cone of X, and let N b ⊆X b be the nullcone, i.e.,
the affine cone of X ∖ X , which has 0 ∈ N as the unique closed G-orbit. Applying
ss b
Proposition 7.6.16 to the nullcone N b ⊆ A(V ), there is a finite subset Σ ⊆ X∗ (T )×R<0
such that for every point x b ∈ N ∖ 0, there is a unique (λ, M ) ∈ Σ such that λ is an
b
optimal destabilizing one-parameter subgroup for x b with M = µ(b x, λ)/ ∥λ∥. Since N
b
is affine, the locus Nλ ⊆ N is a closed subscheme for each (λ, M ) ∈ Σ (Exercise 6.8.7).
b + b
As N b + is invariant under scaling, it corresponds to a closed subscheme Yλ ⊆ X. We
λ
consider the loci
Yλ,M ⊆ Yeλ,M ⊆ Yλ ,
where Yeλ = {x ∈ Yλ | M = µ(x, λ)/ ∥λ∥} and Yλ,M := {x ∈ Yλ | M = M (x) =
µ(x, λ)/ ∥λ∥}. Since the function x ∈ Yλ 7→ µ(x, λ) is locally constant, Yeλ ⊆ Yλ is
open and closed. Using the identification
[
Yλ,M = Yeλ,M ∖ Yeλ′ ,M ′ ,
(λ′ ,M ′ )∈Σ, ,M ′ <M

we conclude that Yλ,M ⊆ Yeλ,M is open. Since G is reductive, Pλ ⊆ G is parabolic

and
[Yλ /Pλ ] ∼
= [G ×Pλ Yλ /G] → [X/G]
is projective, and thus the image Seλ,M = G · Yeλ,m ⊆ X of Yeλ,M is closed, and the
image Sλ,M = G · Seλ,M ⊆ Seλ,M of Yλ,M is open. This implies both (1) and (4).
For (2), if x ∈ X ⊆ P(V ), then the limit x0 = limt→0 λ(t)·x is the projection onto
the subspace W = ⊕Vχ ranging over characters χ ∈ X∗ (T ) such that the projection
projχ (x) of x to Vχ is nonzero and ⟨χ, λ⟩ = −µ(x, λ). By Lemma 7.6.9, λ lies on
the ray spanned by the unique point closest to the origin in the closed convex set of
Cx = {λ ∈ X∗ (T )R | ⟨χ, λ⟩ ≥ 1, projχ (x) ̸= 0}. It follows that λ is also the closest
point to the origin in the analogously defined set Cx0 . Alternatively, one can check
that if λ0 ∈ X∗ (T ) is a optimal destabilizing one-parameter subgroup for x0 , then

429
µ(x0 , λ0 )/ ∥λ0 ∥ ≤ µ(x, λ)/ ∥λ∥ (giving the implication x0 ∈ Yλ,M ⇒ x ∈ Yλ,M ) and
µ(x, λN λ0 )/ λN λ0 ≤ µ(x, λ)/ ∥λ∥ for N ≫ 0 (giving the implication x ∈ Yλ,M ⇒
x0 ∈ Yλ,M ).
For (3), for x ∈ Yλ,M we claim that
Pλ = {g ∈ G(k) | gx ∈ Yλ,M }. (7.7.4)
Since Yλ,M is Pλ -invariant, we have the inclusion ‘⊆’. Conversely, if gx ∈ Yλ,M ,
then both λ and gλg −1 are optimal destabilization one-parameter subgroups for
x. By Kempf’s Optimal Destabilization Theorem (7.6.5), the parabolics Pλ and
Pgλg−1 = gPλ g −1 are equal. Since NG (Pλ ) = Pλ (Proposition B.1.26), we conclude
that g ∈ Pλ . Since [Xλ+ /Pλ ] → [X/G] is proper, so is [Yλ,M /Pλ ] → [Sλ,M /G]. The
map [Yλ,M /Pλ ] → [Sλ,M /G] is surjective by construction, and injective on k-points
by (7.7.4); it is thus finite, surjective, and universally injective. In char(k) = 0, it is
an isomorphism.
For (5), if X is smooth, by Theorem 6.8.9, so is each Yλ . Since Yλ,M ⊆ Yλ is
open, Yλ,M is also smooth. In char(k) = 0, by (3), Sλ,M = G ×Pλ Yλ,M and thus also
smooth. See also [Hes81, §3], [Hes79, §4], and [Kir84, Thms. 12.26 and 13.5].

7.7.2 Examples and related stratifications

Example 7.7.5. Let Gm act linearly on X = P2 with weights −1, 2, 3. Letting λ = id
be the identity one-parameter subgroup, the non-semistable locus is V (x2 y, x3 z) has
the stratification Sλ−1 ,−1 ∪ Sλ,−2 ∪ Sλ,−3 where Sλ−1 ,−1 = {[1 : 0 : 0]}, Sλ,−2 = {[0 :
y : z] | y ̸= 0}, and Sλ,−3 = {[0 : 0 : 1]}.
Example 7.7.6. Revisiting the action of SL2 on X = (P1 )n with the Segre lineariza-
tion (Example 7.6.11), let λ0 : Gm → SL2 be the one-parameter subgroup defined
by λ0 (t) = diag(t−1 , t). The strata are indexed by (λ0 , −1), (λ0 , −3), . . . , (λ0 , −n)
if n is odd and by (λ0 , −2), (λ0 , −4), . . . , (λ0 , −n) if n is even. The strata Sλ0 ,n−2k
consists of tuples with precisely k > n/2 points in common and has codimension
k − 1. The blade Yλ+0 ,n−2k consists of tuples where precisely k points are 0 while the
center Zλ0 ,n−2k is the set of Gm -fixed points where k points are 0 and n − k points
are ∞.
Remark 7.7.7 (Morse stratifications). When X is a smooth projective variety over
C, the HKKN stratification coincides with the Morse stratification of the square-
2
norm of the moment map ∥−∥ : X → R. Given x ∈ X, the optimal destabilizing
one-parameter subgroup corresponds to the path of steepest descent starting from x.
2
The centers Zλ,M correspond to the set of critical values of ∥−∥ , while the strata
Sλ,M are the locally closed submanifolds consisting of points which flow to Zλ,M .
See [Kir84, §6] and [Nes84].
Remark 7.7.8 (Θ-stratifications). As indicated in Remark 6.9.5, there is an identifi-
cation a
Mor([A1 /Gm ], [X/G]) = [Xλ+ /Pλ ],
λ∈X∗ (G)/∼

where X∗ (G)/ ∼ represents the set of one-parameter subgroups up to conjugation.

A Θ-stratification of an algebraic stack X locally of finite type over k is the data
of a totally ordered set Σ with a minimal element 0 ∈ Σ and a stratification into
locally closed substacks a
X = Sλ
λ∈Σ
such that:

430
 (1) for each λ ∈ Σ, X≤λ := Sρ is an open substack of X ,
S
ρ≤λ
(2) for each λ ∈ Σ, there is a union of connected components (called a Θ-stratum
of X≤λ )
Sλ′ ⊆ Mor([A1 /Gm ], X≤λ )
such that ev1 : Sλ′ → X≤λ is a closed immersion mapping isomorphically onto
Sλ , and
(3) for every x ∈ |X |, the set {λ ∈ Σ | x ∈ |X≤λ |} has a minimal element.
See [HL14]. The semistable locus X ss is by definition the open substack X≤0 = S0 .
Let Zλ′ be the preimage of Sλ′ under the map

i : Mor(BGm , X ) → Mor([A1 /Gm ], X ).

The map ev0 : Mor([A1 /Gm ], X ) → Mor(BGm , X ) obtained by restricting to 0 is a

section of i, and there is a diagram analogous to (7.7.3)
ev0
z
Zλ′
i / S′   / X.
λ

In characteristic 0, the HKKN stratification is an example of a Θ-stratification,

where one orders the indices (λ, M ) first by −M and then arbitrarily by λ. In the
next chapter, we will see that the moduli stack Bun r,d (C) has a Θ-stratification
called the Harder–Narasimhan–Shatz stratification.

7.7.3 Applications to cohomology: Kirwan surjectivity

Kirwan Surjectivity (7.7.13) provides a formula for the equivariant Chow groups
of X in terms of the equivariant Chow groups of X ss and the centers Zλ,M of the
HKKN stratification. To prove this, we will need some preparatory lemmas.
Lemma 7.7.9. Let X be a smooth irreducible scheme over an algebraically closed
k with an action of a smooth affine algebraic group G. Let S1 , . . . , Sr ⊆ X be
` smooth, irreducible, andSlocally closed G-invariant subschemes
nonempty, disjoint,
such that X = i Si and such that S≥i := j≥i Sj is closed for each i. Let di be
∗
the codimension of Si in X. If the top Chern class cG di (NSi /X ) ∈ CHG (Si )Q is a
nonzerodivisor for each i, then
r
X
dim CHkG (X)Q = dim CHk−d
G
i
(Si )Q
i=1

for each k.
Proof. By assumption, S≤i = j≤i Sj is open for each i, and Si ⊆ S≤i is a closed
S
subscheme with open complement S<i . We have a commutative diagram

CHk−di
(Si ) / CHkG (S≤i ) / CHkG (S<i ) /0
G

& 
CHkG (Si )

where the top row is the right exact excision sequence (6.1.36(3)) and the vertical
downward arrow is given by intersecting with Si . By the self-intersection formula

431
(6.1.36(6)), the composition CHk−d
G
i
(Si ) → CHkG (Si ) is multiplication by cG
di (NSi /X ).
By hypothesis, this map is injective after tensoring with Q. It follows that the top
row is an exact sequence after tensoring with Q, and that

dim CHkG (S≤i )Q = dim CHk−d

G
i
(Si )Q + dim CHkG (S<i )Q .

The formula follows from induction. See also [AB83, Prop. 1.9].
Remark 7.7.10. If [Si /G] is Deligne–Mumford, then CHkG (Si ) vanishes for k ≫ 0
and cG
di (NSi /X ) is a zero divisor.
There is a convenient criterion for the top Chern class to be a nonzerodivisor.
Lemma 7.7.11. Let X be a smooth irreducible scheme over an algebraically closed
k with an action of a connected, smooth, and affine algebraic group G, and let N
be a G-equivariant vector bundle of rank d on X. Suppose that there is a subgroup
Gm ⊆ G acting trivially on X and a point x ∈ X(k) such that N ⊗ κ(x) contains no
∗
Gm -invariant vectors. Then cG
d (N ) ∈ CHG (X)Q is a nonzerodivisor.

Proof. Choose a maximal torus T containing Gm and a character T → Gm such

that the composition Gm ,→ T → Gm is given by t 7→ td for d > 0. By (6.1.38(b)),
CH∗G (X)Q = CH∗T (X)W ∗
Q where W is the Weyl group. Since CHG (X)Q is a subring
∗ ∗
of CHT (X)Q , we are reduced to show that cd (N ) ∈ CHT (X)Q is a nonzerodivisor.
T

If we write T as the product of the given Gm and a subtorus T ′ , then

CH∗T (X) ∼
= CH∗T ′ (X) ⊗ CH∗ (BGm ) ∼
= CH∗T ′ (X)[z]

by (6.1.38(a)). For x ∈ X(k), we can write

X ′
cTd (N ) = cTi (N ) ⊗ cG
d−i (N ⊗ κ(x))
m

d (N ⊗ κ(x)) + higher degree terms.

= 1 ⊗ cG m

If a1 , . . . , ad denote the Gm -weights of N ⊗ κ(x), then by hypothesis each ai ̸= 0 and

ai z d ∈ CH∗ (BGm )Q ∼
Y

cG
d (N ⊗ κ(x)) =
m
= Q[z]
i

is a nonzerodivisor, and therefore cTd (N ) is also a nonzerodivisor. See also [AB83,

Prop. 13.4] and [Bri97, §3.2].
We define the G-equivariant Chow–Poincaré polynomial of a G-equivariant scheme
X as
∞
X
dim CHdG (X)Q td .


pG (X, t) =
d=0
P∞
We also denote p(X, t) = dim CHd (X)Q td as the (non-equivariant) Chow–


d=0
Poincaré polynomial.
Exercise 7.7.12. Let X be a smooth, irreducible, and quasi-projective scheme
over an algebraically closed field k with an ` action of Gm such that `rX → X is
+
r
surjective (i.e., X is projective). Let X = i=1 Xi and X Gm
= i=1 Fi be the
Białynicki-Birula Stratification (6.8.14), and let di be the codimension of Xi in X.
Show that
Xr
pGm (X, t) = p(Fi , t) · tdi (1 − t)−1 .
i=1

432
Proposition 7.7.13 (Kirwan Surjectivity). Under the hypotheses of the HKKN
Stratification (7.7.1), assume further assume that X is smooth and irreducible, and
that char(k) = 0. Suppose that for all (λ, M ), the stratum Sλ,M is equidimensional
of codimension dλ,M . Then
k−d
X
dim CHkG (X)Q = dim CHkG (X ss )Q + dim CHCλ λ,M (Zλ,M )Q
(λ,M )

and X
pG (X, t) = pG (X ss , t) + pCλ (Zλ,M , t)tdλ,M .
(λ,M )

Proof. From the HKKN Stratification (7.7.1), [Sλ,M /G] ∼ = [Yλ,M /Pλ ], and from the
Białynicki-Birula Stratification (6.8.14), ev0 : Yλ,M → Zλ,M is a Zariski-local affine
fibration and equivariant with respect to Pλ → Cλ . We claim that [Yλ,M /Pλ ] →
[Zλ,M /Cλ ] induces an isomorphism

CH∗Cλ (Zλ,M ) → CH∗Pλ (Yλ,M ). (7.7.14)

By the definition of the equivariant Chow groups (§6.1.6), CHiPλ (Yλ,M ) is identified
by CHi (Yλ,M ×Pλ V ), where V is an open subspace A(W ) of a Pλ -representation
W such that Pλ acts freely on V and A(W ) ∖ V has sufficiently high codimension.
Similarly, CHiCλ (Zλ,M ) is identified with CHi (Zλ,M ×Cλ V ). The map (7.7.14)
corresponds to the pullback map on Chow induced from the composition

Yλ,M ×Pλ V → Zλ,M ×Pλ V → Zλ,M ×Cλ V.

The first map is a Zariski local affine fibration and the second map is a principal
bundle under Uλ = ker(Pλ → Cλ ). Since Uλ is unipotent, Uλ is isomorphic to affine
space (see Example B.1.20) and principal Uλ -bundles are locally trivial in the Zariski
topology (see Example B.1.59). From Properties 6.1.36(2), we conclude that (7.7.14)
is an isomorphism.
We also claim that cdλ,M (NSλ,M /X ) ∈ CH∗G (Sλ,M ) is a nonzerodivisor. Since
(NSλ,M /X )|Zλ,M is identified with NSλ,M /X under CH∗Cλ (Zλ,M ) ∼ = CH∗G (Sλ,M ), it
∗
suffices to show that cdλ,M ((NSλ,M /X )|Zλ,M ) ∈ CHCλ (Zλ,M )Q is a nonzerodivisor.
By the Białynicki-Birula Stratification (6.8.14), λ acts on a fiber of the normal bundle
with nonzero weights. Thus Lemma 7.7.11 implies that cdλ,M ((NSλ,M /X )|Zλ,M ) is a
nonzerodivisor.
We therefore can apply Lemma 7.7.9 with the strata Sλ,M ordered first by −M
and then with any ordering of the λ’s; the semistable locus U ss is viewed as a stratum
with the smallest index. This yields
k−d
X
dim CHkG (X)Q = dim CHkG (X ss )Q + dim CHG λ,M (Sλ,M )Q
(λ,M )
k−d
X
= dim CHkG (X ss )Q + dim CHCλ,Mλ,M (Zλ,M )Q ,
(λ,M )

and the formula for the Chow–Poincaré polynomial also follows.

Remark 7.7.15. This formula was established for de Rham cohomology in [Kir84,
Thm. 5.4]. Instead of the excision sequence
k−dλ,M
CHG (Sλ,M ) → CHkG (S≤(λ,M ) ) → CHkG (S<(λ,M ) ) → 0,

433
one uses the Thom–Gysin long exact sequence
k−dλ,M
· · · → HG (Sλ,M ) → HkG (S≤(λ,M ) ) → HkG (S<(λ,M ) ) → · · · .

Kirwan proved that the long exact sequence splits into short exact sequences by
showing that each map on the right is surjective, which explains why the result is
referred to Kirwan Surjectivity. To see this, observe that the surjectivity of the right
map for all (λ, M ) is equivalent to the injectivity of the left map for all (λ, M ), and
that the latter condition is verified by arguing as above that the top Chern class of
the normal bundle is a nonzerodivisor.

Example 7.7.16. We compute the dimension of the rational Chow groups of

[(P1 )n,ss / SL2 ] using the HKKN Stratification, as determined in Example 7.7.6.
When n is odd, this also gives the dimension of the rational Chow groups of the
GIT quotient (P1 )n,ss / SL2 by Properties 6.1.36(4).
Since [(P1 )n / SL2 ] → B SL2 is an iterated P1 -bundle and CH∗ (SL2 ) ∼
= Z[T ]
generated in degree 2,

CH∗ ([(P1 )n / SL2 ]) ∼

= CH∗ ((P1 )n ) ⊗ CH∗ (B SL2 )
∼
= Z[H1 , . . . , Hn ]/(H1 , . . . , Hn )2 ⊗ Z[T ],

and the Chow–Poincaré polynomial is pSL2 ((P1 )n , t) = (1 + t)n (1 − t2 )−1 . On

the other hand, the strata Sλ,n−2k , where precisely k points are the same, has
codimension k − 1 and its center consists of n
-fixed points. Thus


Z λ,n−2k k G m
pGm (Zλ,n−2k , t) = nk (1 − t)−1 and

X n
pG ((P1 )ss , t) = (1 + t)n (1 − t)−1 − tk−1 (1 − t)−1
k
k>n/2
    
n−1 n−1
= 1 + nt + · · · + 1 + (n − 1) + + ··· + td
2 min(d, n − 3 − d)
+ · · · + ntn−4 + tn−3 .

See also [Kir84, §16.1].

434
Chapter 8

Moduli of semistable vector

bundles

Vector bundles and their associated moduli spaces are of fundamental

importance in algebraic geometry. In recent decades the richness of the
subject has been enhanced by its relationship to other areas of mathematics,
including differential geometry (which has always played a fundamental
role), topology (where the moduli spaces not only present interesting
topological features but turn out to be useful in settling topo- logical
problems) and, perhaps most surprising of all, theoretical physics, in
particular gauge theory, quantum field theory and string theory.

Editors [BBPGPR09, preface]

This chapter proves Theorem B: for a smooth, connected, and projective curve C
of genus g over an algebraically closed field k of characteristic 0, the stack Bun ss
r,d (C)
parameterizing semistable vector bundles of rank r and degree d over is a smooth,
universally closed, and irreducible algebraic stack of dimension r2 (g − 1) which
ps
admits a projective good moduli space Mr,d (C). We follow the six-step strategy to
construct projective moduli spaces outlined in §0.7.2:

① (Algebraicity) We express Bun ssr,d (C) as a substack of Coh(C), parameterizing

all coherent sheaves on C, which is an algebraic stack locally of finite type
(Theorem 3.1.21).
② (Openness of Semistability) Bun ss r,d (C) ⊆ Coh(C) is an open substack (Theo-
rem 8.2.4).
③ (Boundedness of Semistability) Bun ss r,d (C) is of finite type (Theorem 8.2.8).
④ (Semistable reduction) Bun r,d (C) is universally closed by Langton’s Theorem
ss

(8.3.1).
⑤ (Existence of a moduli space) Since Bun ss r,d (C) is Θ- and S-complete (Proposi-
tion 8.3.8), the Existence Theorem for Good Moduli Spaces (6.10.1) implies
ps ps
that exists a good moduli space Bun ss r,d (C) → Mr,d (C), with Mr,d (C) a proper
ps
algebraic space of dim Mr,d (C) = r (g − 1) + 1 (Theorem 8.3.10).
2
ps
⑥ (Projectivity) We follow Faltings’s approach to verify that Mr,d (C) is projective.
ps
We also provide a GIT construction of Mr,d (C). In the rank r = 1 case, Bun 1,d (C)
is identified with the Picard stack Pic C of degree d line bundles on C. We showed
d

435
in Theorem 6.4.51 that there is a banded Gm -gerbe Pic dC → Picd (C) (which in
particular is a coarse moduli space), where Picd (C) is a projective scheme. If in
addition d = 0, then Pic0 (C) is an abelian variety with the group law given by tensor
product.
Notation: Throughout this chapter, C denotes a smooth, geometrically connected,
and projective curve over a field k. We will often further impose that k is algebraically
closed and/or characteristic 0.1 As C is fixed, we will drop it from the notation of
ps ps
the moduli stacks and spaces, e.g., Bun ss
r,d := Bun r,d (C) and Mr,d := Mr,d (C).
ss

8.1 Semistable vector bundles

We introduce the fundamental notion of semistability for vector bundles on a smooth
curve. At first glance, the definition may appear unmotivated, but we will show that
it enjoys several desirable properties:
– Every vector bundle has a unique filtration—the Harder–Narasimhan filtration—
with semistable factors (Theorem 8.1.21), while every semistable vector bundle
has a (non-canonical) filtration—the Jordan–Hölder filtration—with stable
factors (Proposition 8.1.32).
– Semistability is an open and bounded condition (Theorems 8.2.4 and 8.2.8).
– The algebraic stack of semistable vector bundles Bun ss
r,d is universally closed
(Theorem 8.3.1).
ps
– There is a projective good moduli space Bun ss
r,d → Mr,d identifying S-equivalent
vector bundles and inducing a bijection between the polystable vector bundles
ps
(which are identified with the closed points of Bun ss
r,d ) and Mr,d (k) (Theorem
B and Proposition 8.1.35).
– Semistable vector bundles exist for every r > 0 and d (Theorem 8.1.36). In
fact, stable vector bundles also exist (Theorem 8.2.14).
In §8.1.6, we provide a case study of vector bundles on elliptic curves.

8.1.1 Review of coherent sheaves on a smooth curve

Let C be a smooth, geometrically connected, and projective curve C over an
arbitrary field k. Any coherent sheaf F on C has a torsion subsheaf Ftors whose
sections on an open subset U ⊆ C is the subset F (U )tors ⊆ F (U ) of torsion
elements. The quotient F/Ftors is a vector bundle and since Ext1OC (F/Ftors , Ftors ) =
H1 (C, Ftors ⊗ (F/Ftors )∨ ) = 0 (as dim(Supp Ftors ) = 0), the short exact sequence
0 → Ftors → F → F/Ftors → 0 splits and F ∼ = Ftors ⊕ F/Ftors .
If F is a vector
Vr bundle of rank rk F = r, we define its determinant as the line
bundle det F := F and its degree as deg F := deg(det F ). For a coherent sheaf
F ∼= Ftors ⊕ F/Ftors , the degree of F is deg F := h0 (C, Ftors ) + deg(F/Ftors ). We
define the rank of a coherent sheaf F as rk F := dimκ(ξ) F ⊗ κ(ξ) where ξ ∈ C is the
generic point. Note that the rank of a torsion sheaf is 0. If 0 → F ′ → F → F ′′ → 0
is a short exact sequence, then deg F = deg F ′ + deg F ′′ and rk F = rk F ′ + rk F ′′ .
In other words, the degree and rank extend to group homomorphisms
deg, rk : K(C) → Z
1 Most results in this chapter hold (with the same proof) when k is an arbitrary field of

characteristic 0. The only place where the characteristic 0 hypothesis is essential is in the
application of the Existence Theorem for Good Moduli Spaces (Theorem 6.10.1).

436
from the Grothendieck group.

Exercise 8.1.1 (easy). If F1 and F2 are vector bundles on C, show that deg(F1 ⊗
F2 ) = deg F1 rk F2 + rk F1 deg F2 .

Riemann–Roch (5.1.2) for line bundles extends to coherent sheaves.

Theorem 8.1.2 (Riemann–Roch for Coherent Sheaves). Let C be a smooth, con-

nected, and projective curve over an algebraically closed field k. If F is a coherent
sheaf on C, then
χ(C, F ) = deg(F ) + rk(F )(1 − g).

Proof. We will prove this by induction on the rank using the observation that the
Euler characteristic, rank, and degree are each additive in short exact sequences.
If F is torsion, then χ(C, F ) = deg F by the definition of degree. If rk(F ) > 0,
then there is a line bundle subsheaf L ⊆ F with rk(F/L) = rk(F ) − 1. Since the
statement holds for F/L by induction and for L by Riemann–Roch (5.1.2), it also
holds for F .

Exercise 8.1.3 (easy). Show that a vector bundle F on C of rank r and degree d
has Hilbert polynomial P (t) := rt + (d + r(1 − g)).

Exercise 8.1.4 (easy). If F1 and F2 are vector bundles on C of degree di and rank
ri , show that

dim Ext1OC (F1 , F2 ) − dim Ext0OC (F1 , F2 ) = r1 r2 (g − 1) + (d1 r2 − d2 r1 ).

In particular, if F is a vector bundle of r and degree d, then dim Ext1OC (F, F ) −

dim Ext0OC (F, F ) = r2 (g − 1).

We also recall Serre–Duality (5.1.3): for a coherent sheaf F on C, there is an

natural isomorphism
H1 (C, F ) ∼
= HomOC (F, ΩC )∨ .

Example: vector bundles on P1 . It is a classical theorem that vector bundles on

P1 split as direct sums of line bundles.

Theorem 8.1.5 (Birkhoff–Grothendieck Theorem). If E is a vector bundle on

P1 over
Lr an arbitrary field k, there exists unique integers d1 ≥ · · · ≥ dr such that
E∼= i=1 OP1 (di ).

Proof. We will argue by induction on the rank r. If r = 1, every line bundle is

isomorphic to O(d) [Har77, Prop. II.6.4]. If E is a vector bundle of rank r > 0, we
let d1 be the largest integer such that H0 (P1 , E(−d1 )) ̸= 0. A nonzero section of
E(−d1 ) defines an injection O(d1 ) → E and a short exact sequence

0 → O(d1 ) → E → E ′ → 0. (8.1.6)

We claim that E ′ is a vector bundle: otherwise, the kernel K := ker(E → E ′ /Etors

′
)
is a line bundle satisfying O(d1 ) ⊊ K so that K ∼= O(d)
Lr with d > d1 , contradicting
the choice of d1 . By induction, we can write E ′ = i=2 O(di ) with d2 ≥ · · · ≥
dr . Twisting the Labove sequence gives a short exact sequence 0 → O(−1) →
r
i − d1 − 1) → 0. Since H (P , E(−d1 − 1)) = 0 and
0 1
E(−d1 − 1) → i=2 O(d
H (P , O(−1)) = H (P , O(−1)) = 0, we have that H (P , O(di − d1 − 1)) = 0 for
0 1 1 1 0 1

437
all i ≥ 2. Therefore, di − d1 − 1 < 0 or equivalently di ≤ d1 . To show that (8.1.6)
splits, we apply HomOP1 (−, O(d1 )) and compute that

M
Ext1OP1 (E ′ , O(d1 )) = H1 (P1 , O(d1 − di )) = 0
i≥2

because d1 − di ≥ 0. We leave the uniqueness to the reader. See also [LP97,

Lem. 4.4.1] and [HL10, Thm. 1.3.1].
Alternatively, this can be proved using matrix factorizations, which is a similar
spirit to Birkhoff’s argument [Bir09]. As every vector bundle on A1 is trivial, a vector
bundle E on P1 can be specified by a transition function g ∈ GLr (k[x]x ) which is
unique up to automorphisms of E|P1 ∖0 and E|P1 ∖∞ , i.e., there is a bijection

vector bundles on P1 /∼ ←→ GLr (k[x])\ GLr (k[x]x )/ GLr (k[1/x]).



It suffices to show that for any g ∈ GLr (k[x]x ), there exist A ∈ GLr (k[x]) and
B ∈ GLr (k[1/x]) such that AgB = diag(xd1 , . . . , xdr ) for integers d1 ≥ · · · ≥ dr . We
leave this linear algebra fact as an exercise.

Remark 8.1.7. Grothendieck proved an analogous statement for G-bundles on P1 for

a reductive group G [Gro57a], which recovers the above statement with G = GLr .

8.1.2 Definition of semistability and first properties

Definition 8.1.8 (Slope). The slope of a coherent sheaf F on a smooth, geometrically

connected, and projective curve C over an arbitrary field k is defined as

deg F
µ(F ) := ∈ Q ∪ {∞}
rk F

with the convention that µ(F ) = ∞ if F is torsion.

Remark 8.1.9 (Visualizing the slope). We may plot a coherent sheaf F in the complex
plane as the point Z(F ) := − deg(F ) + rk(F )i.2 With this convention, E has larger
slope than F if and only if the complex argument of Z(E) is larger than Z(F ). The
function Z(F ), sometimes called the charge, is additive in short exact sequences.
This perspective can be useful to visualize basic properties, e.g., for a subsheaf
E ⊆ F , µ(E) ≤ µ(F ) if and only if µ(F/E) ≥ µ(F ).

2 The reader may find it more natural to plot instead the point (rk F, deg F ). Our choice

of writing the charge as Z(F ) = − deg(F ) + rk(F )i ∈ C is motivated by Bridgeland stability:

(Coh(C), Z) defines a stability condition on Db (C).

438
 rk

ZINE ZF
ZLE

deg

Figure 8.1.10: Visualizing the slope: for a subsheaf E ⊆ F , Z(F ) = Z(E) + Z(F/E).

Definition 8.1.11 (Semistability/stability/polystability). A nonzero coherent sheaf

F on a smooth, geometrically connected, and projective curve C over a field k is
defined to be
(1) semistable if µ(E) ≤ µ(F ) for every nonzero subsheaf 0 ̸= E ⊆ F ,
(2) stable if µ(E) < µ(F ) for every nonzero proper subsheaf 0 ̸= E ⊊ F , and
(3) polystable if F ∼ Fi where each Fi is stable with µ(Fi ) = µ(F ).
L
= i
If F is not semistable, a subsheaf E ⊆ F with µ(E) > µ(F ) is called a destabilizing
subsheaf. If F is semistable but not stable, we say that F is striclty semistable.
Example 8.1.12 (C = P1 ). The only stable vector bundles on P1 are line bundles.
A vector bundle E ∼ = O(di ) is semistable if and only if it is polystable if and only if
all the di are equal.
Lemma 8.1.13.
(1) If F is a semistable vector bundle of rank r and degree d with gcd(r, d) = 1,
then F is stable.
(2) A semistable coherent sheaf is a vector bundle.
(3) A vector bundle F is semistable (resp., stable) if and only if µ(F ′ ) ≤ µ(F )
(resp., <) for every nonzero proper subbundle F ′ ⊊ F (i.e., F ′ is a subsheaf
such that F/F ′ is a vector bundle).
(4) For a line bundle L on C, a vector bundle F is semistable if and only if F ⊗ L
is.
Proof. For (1), if 0 ̸= E ⊆ F is a subbundle with µ(E) = µ(F ), then there exists
an integer k ≥ 1 with rk F = k rk E and deg F = k deg E. Since gcd(r, d) = 1,
k = 1 and F = E. For (2), if F is not a vector bundle, then the torsion subsheaf
Ftors ⊆ F is a destabilizing subsheaf. For (3), if E ⊆ F is a destabilizing subsheaf
with quotient Q = F/E, then E ′ := ker(F → Q → Q/Qtors ) is a subbundle of F
containing E such that rk(E ′ ) = rk(E) and deg(E ′ ) ≥ deg(E). Since µ(E ′ ) ≥ µ(E),
we see that E ′ ⊆ F is a destabilizing subbundle. For (4), observe that deg(F ⊗ L) =
deg(F ) + rk(F ) · deg(L) and µ(F ⊗ L) = µ(F ) + deg(L). Hence, for any subsheaf
E ⊆ F , we have that
µ(F ) − µ(E) = µ(F ⊗ L) − µ(E ⊗ L),
which implies that F is semistable if and only if F ⊗ L is.

439
Exercise 8.1.14 (easy).
(a) Show that a coherent sheaf F is semistable if and only if µ(Q) ≥ µ(F ) for
every quotient coherent sheaf F ↠ Q.
(b) Show that a vector bundle F is semistable if and only if F ∨ is.
Lemma 8.1.15 (easy).
(1) If E and F are semistable vector bundles with µ(E) > µ(F ), then HomOC (E, F ) =
0.
(2) If E and F are stable vector bundles with µ(E) = µ(F ), then a nonzero
homomorphism E → F is an isomorphism.
Proof. For (1), suppose that E → F is a nonzero map, and let K and I = E/K
denote its kernel and image respectively. Since E is semistable, µ(K) ≤ µ(E).
Hence,
deg(I) = µ(I) rk(I) = µ(E) rk(E) − µ(K) rk(K)
≥ µ(E) rk(E) − µ(E) rk(K)
= µ(E) rk(I)
so µ(I) ≥ µ(E) > µ(F ). This contradicts the semistability of F unless I = 0. For
(2), if E → F is a nonzero map of stables sheaves of slope α = µ(E) = µ(F ), the
kernel K also has slope α. Since E is stable and K ⊊ E, K = 0. Similarly the image
I has slope α and since F is stable, I = F . Thus E → F is an isomorphism.
Corollary 8.1.16. Assume k is algebraically L closed. If F is a stable vector bundle,
then Aut(F ) ∼
= Gm . More generally, if F = Fi⊕ri is a polystable vector bundle with
each Fi stable of slope µ(F ) such that Fi ≇ Fj for i ̸= j, then Aut(F ) ∼
Q
= i GLri .
Proof. If a stable bundle F , Aut(F ) is a division algebra by part (2). Because k
is algebraically closed, Aut(F ) ∼
= k∗ . The isomorphism Aut(F ) ∼ = Gm of group
schemes follows from a simple formal argument (see Proposition 8.2.17(2)). For
the ∼
Q final statement,
⊕ri Q since HomOC (Fi , Fj ) = 0 for i ̸= j, we have that Aut(F ) =
i Aut(Fi ) = i GLri .
Exercise 8.1.17 (good practice, easy). Show that category of semistable vector
bundles of fixed slope is an abelian category closed under extension.
A polystable vector bundle has a reductive automorphism group (Corollary 8.1.16),
but this does not hold for every semistable vector bundle.
Exercise 8.1.18 (moderate). Show that a strictly semistable vector bundle may
have a non-reductive automorphism group.
Remark 8.1.19 (Narasimhan–Seshadri’s Theorem). Let C be a smooth, connected,
and projective curve over C, viewed as a Riemann surface. Let C e → C be the uni-
versal cover. If π : π1 (C) → GLn (C) is a representation, then π1 (C) acts diagonally
on Ce × Cn and the quotient (C e × Cn )/π1 (C) is the total space of a vector bundle
on C. Narasimhan and Seshadri proved that this correspondence induces a bijection
between irreducible unitary representations of π1 (C) and stable vector bundles on C
[NS65].
Remark 8.1.20 (Semistability under tensor products). Narasimhan and Seshadri
applied their correspondence between stability and irreducible unitary representations
to prove that over C the tensor product of two semistable vector bundles is semistable.
This property fails in positive characteristic.

440
8.1.3 Harder–Narasimhan filtrations
Every vector bundle is built from semistable vector bundles in a unique way: this is
called the Harder–Narasimhan filtration, sometimes abbreviated as the HN filtration.
Theorem 8.1.21 (Harder–Narasimhan Filtration). A vector bundle on a smooth,
geometrically connected, and projective curve C over a field k admits a unique
filtration rk
F• : 0 = F0 ⊆ F1 ⊆ F2 ⊆ · · · ⊆ Fn = F
such that each factor Fi /Fi−1 is semistable for each i and

µ(F1 /F0 ) > µ(F2 /F1 ) > · · · > µ(Fn /Fn−1 ).

ZIFE
If k′ /k is a field extension, then 0 ⊆ F1 ⊗k k′ ⊆ · · · ⊆ Fn ⊗k k′zisFthe Harder–
Narasimhan filtration of F ⊗k k′ .
ZLE
Remark 8.1.22. The subsheaf F1 ⊆ F is called the maximal destabilizing subsheaf :
it is semistable and characterized uniquely by the property that for every subsheaf
E ⊆ F , µ(F1 ) ≥ µ(E) with equality if and only if E ⊆ F1 .
Ln
Remark 8.1.23. If C = P1 and F = i=1 O(di )⊕r L with d1 ⊕r
i
> · · · > dn , then the
Harder–Narasimhan filtration is given by Fj = i≤j O(di ) i such that the jth
factor Fj /Fj−1 ∼= O(dj )⊕rj has slope dj .
deg
Remark 8.1.24. Using the correspondence between maps [A1 /Gm ] → Bun r,d and
filtrations of vector bundles (Proposition 6.9.1), the Harder–Narasimhan filtration
of a vector bundle F corresponds to a unique map [A1 /Gm ] → Bun r,d such that 1
maps to F .

ZF
ZLE

ZLE

E
ZIF
deg
Figure 8.1.25: Using Z(F ) = − deg(F ) + rk(F )i to visualize the slope, the Harder–
Narasimhan filtration is determined by the Harder–Narasimhan polygon.

Proof. As in Figure 8.1.25, consider the Harder–Narasimhan polygon of F

HN P (F ) := convex hull of {Z(E) | E ⊆ F with µ(E) ≥ µ(F )}.

441
 ZF
ZLE

ZLEthere exists an integer D such that deg(E) ≤ D for all E ⊆ F ,

By Lemma 8.1.27,
and thus HN P (F ) is bounded from below, i.e., appears to the right of the vertical
line x = −D. It is also bounded above by the horizontal line y = rk(E). Since
Z(E) ∈ Z + EEiZ, there are only finitely many possibilities for Z(E). Thus, there are
finitely many extremal points zi ∈ HN P (F ), each that can be written as zi = Z(Fi )
for a subsheaf Fi ⊆ F . WeZIFmay order the Fi so that µ(F1 ) > · · · > µ(Fk ) = µ(F ),
and we set F0 = 0. We claim that deg
(a) Fi−1 ⊆ Fi ,
(b) if Fi ⊆ F is the unique subsheaf with zi = Z(Fi ), and
(c) Fi /Fi−1 is semistable with decreasing slopes µ(F1 /F0 ) > · · · > µ(Fk /Fk−1 ).
To see (a), consider the subsheaves Fi−1 ∩ Fi and Fi−1 + Fi of F . The short exact
sequence
Z Fitfin
I
0 → Fi−1 ∩ Fi → Fi−1 ⊕ Fi → Fi−1 + Fi → 0,
implies that Z(Fi−1 ∩ Fi ) + Z(Fi−1 + Fi ) = Z(Fi−1 ) + Z(Fi ).

zFo Z Etf
Z FinnFL
ZFi
[Link]

Since Fi−1 and Fi are extremal points, Z(Fi−1 ∩ Fi ) lies to the right and below
Z(Fi−1 ), while Z(Fi−1 + Fi ) lies to the right and above Z(Fi ). Since Fi−1 and Fi
are adjacent extremal points, this can only happen if Z(Fi−1 ∩ Fi ) = Z(Fi−1 ) and
Z(Fi−1 + Fi ) = Z(Fi ). This implies that Fi−1 ∩ Fi = Fi−1 , i.e., Fi−1 ⊆ Fi . The same
argument implies (b): if Fi′ ⊆ F is a subsheaf with Z(Fi′ ) = Z(Fi ), then Fi′ = Fi .
For (c), the semistability of Fi /Fi−1 follows from the convexity of HN P (F ): if
E ⊆ Fi /Fi−1 is a destabilizing subsheaf, then its preimage E e ⊆ Fi would appear
left to the line segment connecting Z(Fi−1 ) and Z(Fi ). Because µ(Fi−1 ) > µ(Fi ),
µ(Fi /Fi−1 ) > µ(Fi+1 /Fi ).
The uniqueness follows formally from the uniqueness of the extremal points of the
convex hull, and this implies that filtration is stable under field extensions. See also
[HN75, Lem. 1.3.7], [LP97, Prop. 5.4.2], [HL10, Thm. 1.3.4], and [Bay11, §2].

Corollary 8.1.26 (of the proof). If E ⊆ F is a subsheaf with µ(E) ≥ µ(F ), then
Z(E) ∈ HN P (F ).

Lemma 8.1.27. If F is a vector bundle on C, there exists an integer D such that

deg E ≤ D for all subsheaves E ⊆ F .

Proof. If C = P1 , this follows from the Birkhoff–Grothendieck Theorem (8.1.5)

classifying vector bundles. Otherwise, we choose a finite morphism f : C → P1 .
Since f is flat, f∗ F is a vector bundle. If E ⊆ F is a subsheaf, then f∗ E ⊆ f∗ F is
a subsheaf. By Riemann–Roch (8.1.2), χ(E) = deg(E) + rk(E)(1 − g) and it thus
suffices to show that χ(E) is bounded above, but this follows from the case of P1 as
χ(C, E) = χ(P1 , f∗ E) = deg(f∗ E) + rk(f∗ E).

442
Exercise 8.1.28 (easy, needed for later). Establish the following relative version of
the above lemma: if S is a noetherian k-scheme and F is a family of vector bundles
on C × S of rank r and degree d, show that there exists an integer D such that for
every s ∈ S and every subsheaf E ⊆ Fs , deg E ≤ D.

Since the Harder–Narasimhan filtration is unique and stable under field extensions,
we obtain:

Corollary 8.1.29. For a field extension k′ /k, a vector bundle F on C is semistable

if and only if F ⊗k k′ is semistable on Ck′ .

Exercise 8.1.30 (moderate). Show that the analogous fact is not true for stability.

Exercise 8.1.31. For a vector bundle E on C, denote by µmax (E) and µmin (E) the
maximum and minimum slope of the Harder–Narasimhan factors of E. Extending
Lemma 8.1.15(1), show that if E and F be vector bundles on C with µmin (E) >
µmax (F ), then HomOC (E, F ) = 0.

8.1.4 Jordan–Hölder filtrations and S-equivalence

Every semistable vector bundle F on C admits a Jordan–Hölder filtration F• where
the factors gri := Fi /Fi−1 are stable vector bundles with the same slope as F . While
this filtration is not unique, the factors are unique up to permutation. By combining
this with the Harder–Narasimhan filtration, we can filter every vector bundle by
stable vector bundles.

Proposition 8.1.32 (Jordan–Hölder Filtration). Let F be a semistable vector

bundle on C. Suppose that

0 = F0 ⊆ F1 ⊆ · · · ⊆ Fk = F and 0 = F0′ ⊆ F1′ ⊆ · · · ⊆ Fℓ′ = F

are filtrations such that the factors gri = Fi /Fi−1 and gr′i = Fi′ /Fi−1
′
are stable vector
′
bundles with µ(F ) = µ(gri ) = µ(gri ). Then ℓ = k and there exists a permutation
σ ∈ Symk such that gri ∼ = gr′σ(i) .

Proof. Semistable vector bundles of a given slope form an abelian category (Exer-
cise 8.1.17), whose simple objects are the stable vector bundles. This proposition
holds for any artinian and noetherian object in an abelian category [SP, Tag 0FCK].
The proof is an elementary induction argument analogous to the existence of com-
positions series for finite groups.

As a result, the following definitions are well-defined.

Definition 8.1.33 (Associated graded). The associated graded of a semistable

bundle F is
gr F ∼
M
= gri ,
i

where gri = Fi /Fi−1 are the factors with respect to any Jordan–Hölder filtration
0 = F0 ⊆ F1 ⊆ · · · ⊆ Fk .

Definition 8.1.34 (S-equivalence). Two semistable vector bundles E and F ′ of the

same rank and degree are called S-equivalent 3 if gr E ∼
= gr F .
3 The ‘S’ in ‘S-equivalence’ refers to Seshadri.

443
 By the Jordan–Hölder Filtration (8.1.32), every semistable vector bundle is S-
equivalent to its associated graded. This filtration corresponds via Proposition 6.9.1
to
L a map [A /Gm ] → Bun r,d taking 1 to [F ] and 0 to the associated graded gr F =
1 ss

i Fi /Fi−1 , which is polystable. While the polystable limit gr F is unique, the map
[A1 /Gm ] → Bun ssr,d is not canonical.
For the following statement and proof, we will appeal to the fact (proven later
but independently) that Mss r,d is an algebraic stack of finite type (Theorem 8.2.8).

Proposition 8.1.35. Let k be an algebraically closed field of characteristic 0.

(1) A vector bundle F on C of rank r and degree d is polystable if and only if
[F ] ∈ Bun ss
r,d is a closed point.
(2) Two semistable vector bundles F1 , F2 on C of rank r and degree d are S-
̸ ∅ in |Bun ss
equivalent if and only if {[F1 ]} ∩ {[F2 ]} = r,d |.

Proof. For (1), if F is not polystable, then the Jordan–Hölder filtration F• defines
a map [A1 /Gm ] → Bun ss r,d taking 1 to F and 0 to gr F , which shows that [F ] is
not a closed point. On the other hand, if F is polystable and [F ] is not a closed
point, there exists a closed point [G] ∈ Mss r,d in the closure of [F ] (by the quasi-
compactness of Mss r,d ) with G ≇ F . By above, G is necessarily also polystable and, in
particular, Aut(G) is reductive by Corollary 8.1.16, hence linearly reductive (because
char(k) = 0!). By the Destabilization Theorem (7.3.8), there exists a morphism
[A1 /Gm ] → Bun r,d taking 1 to F and 0 to G. By Proposition 6.9.6, this corresponds
to filtration 0 = F0 ⊆ · · · ⊆ Fn = F with associated graded G. Since F = i∈I Si
L
is direct sum of stable vector bundles of the same slope of F , each Fi must be a
direct sum of a subset of the Si . This implies that the associated graded bundle G
is isomorphic to F , a contradiction.
For (2), if F1 and F2 are S-equivalent, then there exist Jordan–Hölder filtrations
F1,• and F2,• with isomorphic associated graded bundles. These correspond to
maps f1 , f2 : [A1 /Gm ] → Bun ss r,d with fi (1) ≃ Fi and f1 (0) ≃ f2 (0). Hence, the
closures of [F1 ] and [F2 ] in Bun ss r,d have nonempty intersection. Conversely, if
{[F1 ]} ∩ {[F2 ]} is nonempty, then it contains a closed point [G] ∈ |Bun ss r,d | in the
closure of both [F1 ] and [F2 ]. By the Destabilization Theorem (7.3.8), there exists
maps f1 , f2 : [A1 /Gm ] → Bun ss with fi (1) ≃ Fi and f1 (0) ≃ f2 (0) ≃ G. Since G is
a closed point, G is polystable and the filtrations corresponding to f1 and f2 are
Jordan–Hölder filtrations with isomorphic associated graded bundles. Hence, F1 and
F2 are S-equivalent.
ps
Once we have established the existence of a good moduli space π : Bun ss
r,d → Mr,d
(Theorem 8.3.10), part (2) above implies that two vector bundles F1 and F2 are
S-equivalent if and only if π([F1 ]) = π([F2 ]).

8.1.5 Existence of semistable bundles

Theorem 8.1.36 (Existence of Semistable Bundles). Let C be a smooth, connected,
and projective curve over an algebraically closed field k of genus g ≥ 2. Then for
every degree d and rank r > 0, there exists a semistable vector bundle on C of rank
r and degree d.

Remark 8.1.37. When gcd(r, d) = 1, this theorem implies the existence of a stable
vector bundle. In a moment, we will show that there exists a stable vector bundle of
every rank and degree (Theorem 8.2.14).

444
 deg rd

ride
Proof. We will use induction on the rank r. When r = 1, the statement is clear as
there exist line bundles of every degree and every line bundle is stable. If r > 1,
choose (r1 , d1 ) such that d1 /r1 > d/rrk and the triangle with vertices (0, 0), (r1 , d1 ),
(r, d) has no integral points in the interior nor integral points on the boundary line
segments (0, 0) ↔ (r1 , d1 ) and (r1 , d1 ) ↔ (r, d). This can be arranged by choosing
the closest integral point to the left of the line segment (0, 0) ↔ (r, d).

deg rd

ad
rank

Set r2 = r − r1 and d2 = d − d2 . By induction, we may choose semistable vector

bundles F1 and F2 with rkchoose
Fi = ri and deg Fi = di . Since gcd(r1 , d1 ) = gcd(r2 , d2 ) =
r
1, both F1 and F2 are stable. By Riemann–Roch Fr
d (Exercise 8.1.4),
dim Ext1OC (F1 , F2 ) ≥ r1 r2 (g − 1) + (d1 r2 − d2 r1 ) > 0
is strictly positive as g > 1 and d1 /r1 > d/r > d2 /r2 . Choose a non-trivial extension
deg F
0 → F2 → F → F1 → 0. (8.1.39)
We will show that any destabilizing subsheaf of F is necessarily isomorphic to F1 and
splits the sequence. Namely, if 0 ̸= E ⊆ F is a subsheaf with µ(E) > µ(F ) = d/r.
Let E2 := E ∩F2 ⊆ F2 and E1 = E/E2 ⊆ F1 . As F1 and F2 are stable, µ(Ei ) ≤ di /ri
as long as Ei ̸= 0. It follows that (rk E, deg E) is contained in the triangle with
vertices (0, 0), (r1 , d1 ), (r, d). Since the line segments (0, 0) ↔ (r1 , d1 ) ↔ (r, d) have
no integral points, we must have that rk E = r1 and deg E = d1 . This implies that
∼ ∼
E2 = 0, E1 = F1 , and E → E1 is an isomorphism. Thus F1 = E1 → E ,→ F splits
rank
the exact sequence (8.1.39), contradicting that F is a non-trivial extension. See also
[Ari15].
Exercise 8.1.40.
(a) If k not algebraically closed, show that the above argument is valid as long as
C(k) ̸= ∅.
(b) Show that semistable bundles exist if k is a finite field.
Example 8.1.41 (Examples from geometry). Let C ⊂ Pg−1 be a general canonical
curve of genus g ≥ 3. The restricted tangent bundle TPg−1 |C is always stable, and if
g ̸= 4, 6 (resp., g ≡ 1 or 3 (mod 6)), then the normal bundle NC/Pg−1 is semistable
(resp., stable); see [CLV23].

8.1.6 Case study: elliptic curves

∼ OC ,
Let (C, p) be an elliptic curve over an algebraically closed field k. Then ωC =
and for a vector bundle F , Riemann–Roch (8.1.2) asserts that χ(C, F ) = deg F ,
while Serre–Duality states that H1 (C, F ) = H0 (C, F ∨ )∨ .

445
Exercise 8.1.42 (Line bundles).
(a) For a line bundle L on C of degree 1, show that h0 (C, L) = χ(C, L) = 1 and a
nonzero section vanishes at a unique point x ∈ C(k).
∼
(b) Conclude the map x 7→ [OC (x)] defines an isomorphism C → Pic1 (C) from
the Picard scheme (see Theorem 6.4.51) and extend this to an isomorphism
∼
C → Picd (C) for all d.

Atiyah’s classification of indecomposable bundles. A vector bundle F is

indecomposable if it cannot be written as F1 ⊕ F2 for nonzero vector bundles F1
and F2 . By the Krull–SchmidtLTheorem (e.g., [Ati56]), every vector bundle F can
be uniquely written as F = i Fi for indecomposable vector bundles Fi . This
holds in any genus, but what is unique to elliptic curves is the relationship between
indecomposability and stability (Exercise 8.1.46). Let Bunind r,d denote the set of
isomorphism classes of indecomposable bundles of rank r and degree d.
Exercise 8.1.43 (moderate).
(a) Show that there is an isomorphism of stacks
∼
Bun r,d → Bun r,d+r , [F ] 7→ [F ⊗ OC (p)],

which restricts to a bijection Bunind

r,d → Bunr,d+r .
ind

(b) If F is a vector bundle of rank r and degree 0 ≤ d < r, show that there is
bijection Ext1OC (F, H0 (C, F ) ⊗ OC ) ∼
= EndOC (F ), under which the identity
map idF corresponds to an extension

0 → H0 (C, F ) ⊗ OC → E → F → 0

such that the boundary map H0 (C, F ) → H1 (C, H0 (C, F ) ⊗ OC ) is an isomor-

phism. Show that F is indecomposable if and only if E is, and that the map
[F ] 7→ [E] defines a bijection Bunind
r,d → Bunr+d,d .
ind

∼
(c) Letting h = gcd(r, d), show that there is a bijection αr,d : Bunind ind
h,0 → Bunr,d
such that det αr,d (F ) = det(F ) ⊗ OC (dp).
See also [Ati57, Lem. 10, Thm. 6].
Exercise 8.1.44 (hard). Let r > 0.
(a) Show that there exists a unique indecomposable vector bundle Fr of rank r and
degree 0 with H0 (C, Fr ) ̸= 0. Moreover, show that H0 (C, Fr ) = 1, Fr∨ ∼
= Fr ,
and that Fr is an extension 0 → OC → Fr → Fr−1 → 0.
(b) Show that every [E] ∈ Bunind ∼
r,0 can be expressed as E = Fr ⊗ L for a unique
line bundle L of degree 0. Conclude that there is a bijection C(k) ∼
= Bunind .
r,0

(c) For every d, show that there is a canonical vector bundle [Fr,d ] ∈ Bunind r,d
of rank r and degree d such that every [E] ∈ Bunind r,d can be expressed as
E∼ = Fr,d ⊗ L for a line bundle L of degree 0. If Fr,d ⊗ L ∼
= Fr,d ⊗ L′ , show
that (L′ ⊗ L∨ )r/ gcd(r,d) ∼
= OC .
∼
(d) Conclude that C(k) = Bunind for every r > 0 and d.
r,d
See also [Ati57, Thm. 10].

Semistable vector bundles and their moduli.

Exercise 8.1.45 (Unstable bundles, easy).

446
 (a) If F1 and F2 are semistable vector bundles on C with µ(F1 ) < µ(F2 ), show
that Ext1OC (F1 , F2 ) = 0.
(b) Conclude that an unstable vector bundle is a direct sum of semistable vector
bundles, not all of the same slope.
Exercise 8.1.46 (Stable vs. indecomposable bundles).
(a) (easy) Prove that a stable vector bundle is indecomposable. (This holds in any
genus.)
(b) (hard) Show that an indecomposable vector bundle F on C is semistable.
(c) (easy) Conclude that if gcd(r, d) = 1, a vector bundle of rank r and degree d is
stable if and only if it is indecomposable.
(d) (moderate) If gcd(r, d) ̸= 1, show that there are no stable vector bundles of
rank r and degree d.
See also [Tu93, App. A].
Exercise 8.1.47 (hard). For r > 0 and d, set h = gcd(r, d), r′ = r/h, and d′ = d/h.
(a) Show that every semistable vector bundle F of rank r and degree d can be
written as
h
F ∼
M
= Fr′ ,d′ ⊗ Li ,
i=1
where Fr′ ,d′ is the vector bundle introduced in Exercise 8.1.44 and each Li
is a line bundle of degree 0. Show that the line bundles Li are unique up to
permutation and multiplication by an r′ -torsion element of Pic0 (C).
(b) Assume the existence of a proper good moduli space Bun ss
r,d → Mr,d (Theo-
ss
∼
rem 8.3.10). Using the isomorphism Pic0 (C) → C (see Exercise 8.1.42), show
that there is an isomorphism
h h
∼
ps
M X
Mr,d → Symh C, Fr′ ,d′ ⊗ Li 7→ Li (p).
i=1 i=1

See also [Tu93, Thms. 1, 2, and 16].

8.2 Moduli stack of semistable vector bundles

It then becomes apparent that the successes arising from the theory of
“twisted sheaves” are primarily psychological: by viewing the moduli stack
as its own object (rather than something to be compressed into a scheme),
one more easily retains information that can be flowed back to the land
of varieties.

Max Lieblich [Lie18]

We cover properties of the moduli stack Bun ss r,d of semistable vector bundles.
Highlights include Openness of Semistability (8.2.6), Boundedness of Semistability
(8.2.8), and the non-emptiness of the stable locus Bun sr,d (Theorem 8.2.14). This
latter property is proved inductively using stacks of extensions (§8.2.4). We also
demonstrate that Bun sr,d is a banded Gm -gerbe over an algebraic space Mr,d s
(Corol-
lary 8.2.23). Finally, we introduce the stack of vector bundles with fixed determinant
(§8.2.7).

447
8.2.1 Review of the stack of vector bundles over a curve
Let C be a smooth, connected, and projective curve over an algebraically closed
field k of genus g ≥ 2. We have already established a handful of properties of the
stack Coh := Coh(C) parameterizing coherent sheaves on C, which we recall now.
First, Coh is an algebraic stack (Theorem 3.1.21), which is locally of finite type
(Example 3.3.3) and has affine diagonal (Example 3.3.15). Theorem 3.1.21 also
implies that the substack Cohr,d ⊆ Coh, parameterizing coherent sheaves of rank r
and degree d, is open and closed, and furthermore that the substacks Bun ⊆ Coh
and Bun r,d ⊆ Cohr,d of vector bundles are open substacks. Every quasi-compact
open substack of Cohr,d is isomorphic to a quotient stack [Q′N / GLP (N ) ] for some N ,
where Q′N ⊂ QuotP (OC (−N )⊕P (N ) /C/k) is an open subscheme of the projective
Quot scheme and P is the Hilbert polynomial of any coherent sheaf of rank r and
degree d.
Using the Deformation Theory (C.2.11), we showed in Proposition 3.7.9 that
Cohr,d is smooth over Spec k of dimension r2 (g − 1). Indeed, if [F ] ∈ Cohr,d (k),
then TCohr,d ,[F ] ∼
= Ext1 (F, F ), which is computed by Riemann-Roch to have di-
mension r2 (g − 1) (see Exercise 8.1.4). The stacks Cohr,d and Bun r,d satisfy the
valuative criterion for universally closed (Exercise 3.8.13) but are not quasi-compact
(Exercise 8.2.3).
The stacks Cohr,d and Bun r,d are not Deligne–Mumford as Aut(F ) always
contains a copy of Gm (as long as (r, d) ̸= (0, 0)). It follows that Cohr,d and Bun r,d
are not separated: if they were separated, then each automorphism group would
be proper and affine, hence finite. On the other hand, we know that Cohr,d is both
Θ- and S-complete (Proposition 6.9.30), which will be useful in establishing the
existence of a good moduli space in §8.3.2.
Exercise 8.2.1. Let F be a coherent sheaf of rank r and degree d on C, and let N be
an integer such that H1 (C, F (N )) = 0 and F (N ) is globally generated, i.e., such that
F is expressed as a quotient [OC (−N ) ⊗ V ↠ F ] ∈ Q′N , where V ∼ = H0 (C, F (N )) is
a vector space of dimension P (N ). Show that there is an exact sequence where each
term has the displayed identification

0 → HomOC (F, F ) → HomOC (OC (−N ) ⊗ V, F ) → HomOC (K, F )

| {z } | {z } | {z }
End(F ) End(V ) TQ ′ ,[OC (−N )⊗V →F ]
N

→ Ext1OC (F, F ) → Ext1OC (OC (−N ) ⊗ V, F ) .

| {z } | {z }
TCohr,d ,[F ] 0

Remark 8.2.2 (The stack Bun r,d (P1 )). Despite that the Birkhoff–Grothendieck
Theorem (8.1.5) provides a simple classification of isomorphism classes of vector
bundles on P1 , the stack Bun r,d (P1 ) of vector bundles is more complicated. It is
not true that Bun r,d (P1 ) is a disjoint union over the isomorphism classes of vector
bundles. For instance, consider the short exact sequence 0 → O(−1) → O⊕2 →
O(1) → 0 defined by a non-zero element of Ext1OP1 (O(1), O(−1)) ∼ = k. The universal
extension defines a Gm -equivariant map A1 → Bun 2,0 (P1 ) taking 1 to [O⊕2 ] and 0
to [O(−1) ⊕ O(1)]. Equivalently, we can view this as a map [A1 /Gm ] → Bun 2,0 (P1 ),
whose existence also follows from the characterization of maps from [A1 /Gm ] in
Proposition 6.9.6. This implies that [O(−1) ⊕ O(1)] ∈ |Bun 2,0 (P1 )| is in the closure
of [O⊕2 ]. Moreover, since Bun 2,0 (P1 ) contains O(d) ⊕ O(−d) for every d > 0, it
follows that Bun 2,0 (P1 ) is not bounded, i.e., not of finite type.
Exercise 8.2.3. Show that the algebraic stack Bun r,d (C) is not bounded.

448
8.2.2 Openness of semistability
We prove that stability and semistability are open properties, and conclude that the
stacks Bun sr,d and Bun ss
r,d are open substacks of Bun r,d .

Theorem 8.2.4 (Openness of Stablity/Semistability). Let C be a smooth, connected,

and projective curve over an algebraically closed field k of genus g ≥ 2. Let S be a
k-scheme and F be a family of vector bundles of rank r and degree d on CS . The
locus of points s ∈ S such that Fs := F |Cs is semistable is open. Likewise, the locus
of points s ∈ S such that Fs := F |C×Spec κ(s) is stable is open.

Remark 8.2.5. The difference in the two formulations above is due to the fact
that unlike semistability, stability is not always preserved under field extensions
(Exercise 8.1.30).

Proof. By the Limit Methods of §B.3, we can assume that S is of finite type over k.
We may also assume that S is connected. We will express the locus of points s ∈ S
such that Fs is not semistable as a finite union of images of proper morphisms. Let
P (t) := rt + (d + r(1 − g)) be the Hilbert polynomial of a rank r and degree d vector
bundle. By Exercise 8.1.28, there exists an integer D such that every subsheaf E ⊆ Fs
of a fiber has deg E ≤ D. By definition, Fs is not semistable if and only if there exists
a subsheaf E ⊆ Fs with µ(F ) > d/r. Since rk E ≤ r and deg E ≤ D, there are only
finitely many possibilities of (rk E, deg E) for a destabilizing subsheaf E of Fs . For
each possible rank and degree (r′ , d′ ), let P ′ (t) = r′ t + (d′ + r′ (1 − g)) be the Hilbert
polynomial of a rank r′ and degree d′ vector bundle. Let Q(t) = P (t) − P ′ (t) be the
Hilbert polynomial of a quotient Fs /E. By Projectivity of the Quot scheme (1.1.3),
there is a scheme QuotQ (F/CS /S), parameterizing quotients of F with Hilbert
polynomial Q, which is projective over S. The image of QuotQ (F/CS /S) → S is
closed and identified with the locus of points s ∈ S such that Fs has a destabilizing
subsheaf of rank r′ and degree d′ . Therefore, the non-semistable locus of S is closed.
Similarly, if Fs is semistable but not stable, then letting h = gcd(r, d), there
exists a subsheaf 0 ̸= E ⊊ Fs of rank kr/h and degree kd/h for some integer
0 < k < gcd(r, d). The above argument extends to show that the locus of points
s ∈ S such that Fs is not stable is closed.

Corollary 8.2.6. The stacks Bun sr,d and Bun ss r,d are open substacks of Bun r,d and,
in particular, algebraic stacks locally of finite type over k.

8.2.3 Boundedness of semistability

The crucial input to the boundedness of semistable vector bundles on a curve is the
following lemma.

Lemma 8.2.7. If E is a semistable vector bundle on C and µ(E) > 2g − 1, then E

is globally generated and H1 (C, E) = 0.

Proof. By Serre Duality (5.1.3), H1 (C, E) ∼= HomC (E, ΩC )∨ . The line bundle
ΩC is semistable of slope 2g − 2. Since E is also semistable and µ(E) > 2g − 2,
HomC (E, ΩC ) = 0 by Lemma 8.1.15(1). Thus H1 (C, E) = 0. The same argument
shows that H1 (C, E ⊗ OC (−p)) = 0 for any point p ∈ C(k) because E ⊗ OC (−p) is
a vector bundle which is semistable (Lemma 8.1.13(4)) of slope µ(E) − 1 > 2g − 2.
Applying this vanishing to the short exact sequence 0 → E ⊗ OC (−p) → E →
E ⊗ κ(p) → 0, we obtain that H0 (C, E) → E ⊗ κ(p) is surjective for every p ∈ C.

449
Theorem 8.2.8 (Boundedness of Semistability). Let C be a smooth, connected, and
projective curve over an algebraically closed field k of genus g ≥ 2. For every pair
of integers r > 0 and d, the algebraic stack Bun ss
r,d is of finite type over k.

Proof. Choose a point p ∈ C(k) and an integer N such that d/r + N > 2g − 1.
For every semistable vector bundle E of rank r and degree d, the vector bundle
E ⊗ OC (N p) is semistable by (Lemma 8.1.13(4)) and has slope µ(E ⊗ OC (N p)) =
µ(E) + N > 2g − 1. By Lemma 8.2.7, E ⊗ OC (N p) is globally generated and
H1 (C, E⊗OC (N p)) = 0. Thus, E⊗OC (N p) can be expressed as a quotient of V ⊗OC
where V is a vector space of fixed dimension h0 (C, E ⊗OC (N p)) = χ(C, E ⊗OC (N p))
(which is independent of E), and thus E can be expressed as a quotient

V ⊗ OC (−N p) → E.

Letting P (t) = rt + (d + r(1 − g)), the Quot scheme QuotP (V ⊗ OC (−N p)/C) is of
finite type over k by Theorem 1.1.3. By Openness of Semistability (8.2.4), there is an
open subscheme Q′ ⊆ QuotP (V ⊗OC (−N p)/C) parameterizing semistable quotients.
The universal quotient on Q′ defines a surjective morphism Q′ ↠ Bun ss r,d (and in
fact expresses Bun ss
r,d as a quotient stack [Q / GLP (N ) ]). Since Q is quasi-compact,
′ ′

so is Bun ss
r,d .

8.2.4 Stacks of extensions

We discuss the algebraicity of moduli functors and stacks parameterizing extensions
of two given vector bundles. As a consequence, we will be able construct a stack
of extensions over Bun r1 ,d1 × Bun r2 ,d2 which admits a morphism to Bun r1 +r2 ,d1 +d2 .
This can be viewed as a analogous construction to the Gluing Morphisms (5.6.16)
Mg1 ,n1 × Mg2 ,n2 → Mg1 +g2 ,n1 +n2 −2 . As in the case of stable curves, this provides
an inductive strategy to establish properties of Bun r,d . We will use this strategy in
the next subsection to show the nonemptiness of the stable locus Bun sr,d .
Extensions. If F1 and F2 are vector bundles on a scheme X, let EXT(F1 , F2 )
a b
denote the groupoid of extensions 0 → F2 − →E→ − F1 → 0, where a morphism of
extensions is an isomorphism α : E → E fitting in a commutative diagram

0 / F2 a /E b / F1 /0

id α id
  
0 / F2 a / E′ b / F1 / 0.

Exercise 8.2.9 (easy). Show that AutEXT(F1 ,F2 ) ([E]) ∼= HomOX (F1 , F2 ), where
γ ∈ HomOX (F1 , F2 ) corresponds to the automorphism id +a ◦ γ ◦ b of the extension
E.

Functor of extensions. Before discussing the stack of extensions, we first describe

properties of the functor of extensions. Recall that the group Ext1OX (F1 , F2 ) classifies
isomorphism classes of EXT(F1 , F2 ). If X is defined over a field k, then the functor
T 7→ Ext1OX×T (p∗1 F1 , p∗1 F2 ) from the category of affine k-schemes, is representable
by A(Ext1OX (F1 , F2 )∨ ) (Exercise 0.3.21). The group Aut(F1 ) × Aut(F2 ) acts on
A(Ext1OX (F1 , F2 )∨ ) via

a b 2 a◦β −1 β1 ◦b
(β1 , β2 ) · [0 → F2 −
→E→
− F1 → 0] = [0 → F2 −−−− → E −−−→ F1 → 0].

450
The anti-diagonal subgroup Gm = {(t−1 , t)} ⊆ Aut(F1 ) × Aut(F2 ) corresponds to
the scaling automorphisms on E and acts trivially on A(Ext1OX (F1 , F2 )∨ ). On the
other hand, Gm = {(t, 1)} ⊆ Aut(F1 ) × Aut(F2 ) acts non-trivially via the scaling
action on A(Ext1OX (F1 , F2 )∨ ). Under this latter action, the universal extension on
X × A(Ext1OX (F1 , F2 )∨ ) is Gm -equivariant and defines a morphism

[A(Ext1OX (F1 , F2 )∨ )/Gm ] → Bun(X),

restricting to a morphism P(Ext1OX (F1 , F2 )∨ ) → Bun(X) on the locus of non-split

extensions.
In the relative setting, we have:
Exercise 8.2.10. Let f : X → S be a proper flat morphism of noetherian schemes,
e.g., X = C × S → S for a curve C. Let F1 and F2 be vector bundles on X such
that f∗ E xt1OX (F1 , F2 ) is a vector bundle on S whose construction commutes with
base change. Show that the functor
Sch/S → Sets
(T → S) 7→ H0 (XT , E xt1OX (F1,T , F2,T )),
T

is representable by A((f∗ E xt1OX (F1 , F2 ))∨ ) → S. See also [Lan83].

If T is affine, observe that H0 (XT , E xt1OX (F1,T , F2,T )) = Ext1OX (F1,T , F2,T ).
T T
In particular, A((p∗ E xt1OX (F1 , F2 ))∨ ) → S represents the functor (T → S) 7→
Ext1OX (F1,T , F2,T ) on the category of affine schemes over S. For non-affine
T
schemes T , these groups are connected via the local-to-global spectral sequence
Hp (T, fT,∗ E xtqOX (F1,T , F2,T )) ⇒ Extp+q
OXT (F1,T , F2,T ), which on low degree terms
T
yields an exact sequence

0 → H1 (T, fT,∗ H omOXT (F1,T , F2,T )) → Ext1OX (F1,T , F2,T ) →

T

H 0
(T, fT,∗ E xt1OX (F1,T , F2,T )) → H (T, fT,∗ H omOXT (F1,T , F2,T )).
1
T

When T is affine, the left and right terms are 0, and the two middle terms are
identified.
Stack of extensions. Due to the presence of automorphisms of extensions, it is
more natural to describe the moduli of extensions as a stack. This allows us to
conveniently handle the case that f∗ E xt1OX (F1 , F2 ) is not a vector bundle, e.g., when
the dimension of Ext1OXs (F1,s , F2,s ) depends on the point s ∈ S.
Define the stack
EXT(F1 , F2 )
over (Sch/S)ét , whose category over T → S is the groupoid EXT(F1,T , F2,T ) of
extensions of the base changes F1,T and F2,T to XT .
Proposition 8.2.11. If f : X → S is a family of curves and F1 and F2 are
vector bundles on X, then EXT(F1 , F2 ) is an algebraic stack isomorphic to the
vector bundle stack C(Rf∗ H omOX (F1 , F2 )) over S defined in Example 3.9.32. In
particular, EXT(F1 , F2 ) → S is smooth and quasi-compact, whose relative dimension
is computed as

dim Ext1OXs (F1,s , F2,s ) − dim Ext0OXs (F1,s , F2,s )

for any point s ∈ S.

451
 d
Proof. Let K • : K 0 −
→ K 1 be a complex of vector bundles computing Rf∗ H omOX (F1 , F2 ).
Then
C(Rf∗ H omOX (F1 , F2 )) ∼
= C(K • ) ∼
= [A(K 1,∨ )/A(K 0,∨ )],
where the additive group A(K 0,∨ ) acts on A(K 1,∨ ) via v0 · v1 = v1 + d(v0 ). If
g : T → S is a morphism from an affine scheme, then all principal A(K 0,∨ )-bundles
are trivial, so that

C(K • )(T ) ∼
= [A(K 1,∨ )(T )/A(K 0,∨ )(T )]
∼
= [H0 (T, g ∗ K 1 )/H0 (T, g ∗ K 0 )] (see (3.9.33))
∼
= EXT(F1,T , F2,T ), (as ExtiOX ∼
(F1,T , F2,T ) = Hi (g ∗ K • )).
T

Remark 8.2.12. If Hom(F1,s , F2,s ) = 0 for every s ∈ S, then f∗ E xt1OX (F1 , F2 ) is a

vector bundle on S by Cohomology and Base Change (A.6.8). In this case, we recover
Exercise 8.2.10: EXT(F1 , F2 ) is representable over S by A(f∗ E xt1OX (F1 , F2 )∨ ). More
generally, if f∗ E xtiOX (F1 , F2 ) are vector bundles for i = 0, 1, then EXT(F1 , F2 ) → S
is a gerbe banded by f∗ H omOX (F1 , F2 ).
Remark 8.2.13 (The universal extension). For two rank-degree pairs (r1 , d1 ) and
(r2 , d2 ), we can apply this construction to the morphism

C × Bun r1 ,d1 × Bun r2 ,d2 → Bun r1 ,d1 × Bun r2 ,d2

and the vector bundles p∗12 F1,univ and p∗13 F2,univ , where Fi,univ is the universal
vector bundle on C × Bun ri ,di . We have the stack of extensions

q : EXT(p∗12 F1,univ , p∗13 F2,univ ) → Bun r1 ,d1 × Bun r2 ,d2 ,

where an object over T is that data of vector bundles F1 and F2 on CT with

rk Fi = ri and deg Fi = di together with an extension 0 → F2 → E → F1 → 0, while
a morphism over the identity on T is the data of a commutative diagram

0 / F2 a /E b / F1 /0

β2 α β1
 ′  ′ 
0 / F2′ a / E′ b / F1′ / 0,

where α, β1 , and β2 are isomorphisms.

There is a universal extension Euniv of rank r = r1 + r2 and degree d = d1 + d2

0 → (id ×q)∗ p∗13 F2,univ → Euniv → (id ×q)∗ p∗12 F1,univ → 0

on C×EXT(p∗12 F1,univ , p∗13 F2,univ ). This defines a representable morphism EXT(p∗12 F1,univ , p∗13 F2,univ ) →
Bun r,d fitting in a diagram

[0 → F2 → E → F1 → 0] 
[Euniv ]
EXT(p∗12 F1,univ , p∗13 F2,univ ) / Bun r,d / [E]
_
q
 
Bun r1 ,d1 × Bun r2 ,d2 ([F1 ], [F2 ]).

452
8.2.5 Existence of stable bundles
We prove the Existence of Stable Bundles (8.2.14), extending a prior result on the
Existence of Semistable Bundles (8.1.36). Our argument is moduli-theoretic relying
on the algebraicity of Bun sr,d , Bun ss
r,d , and the stack of extensions.

Theorem 8.2.14 (Existence of Stable Bundles). Let C be a smooth, connected, and

projective curve over an algebraically closed field k of genus g ≥ 2. Then Bun sr,d is
nonempty and dense in Bun ss r,d .

Proof. Theorem 8.1.36 gives the existence of a semistable vector bundle, and hence
a stable vector bundle if gcd(r, d) = 1. We will prove the theorem by induction on
the rank by showing that dim Bun ss r,d ∖ Bun r,d < dim Bun r,d . If gcd(r, d) > 1, then
s ss

a vector bundle F of rank r and degree d is strictly semistable if and only if if it can
be written as an extension 0 → F2 → F → F1 → 0 of semistable vector bundles with
rk(Fi ) = ri < r and µ(Fi ) = µ(F ). Letting EXT(p∗12 F1,univ , p∗13 F2,univ ) be the stack
of extensions over Bun ss
r1 ,d1 × Bun r2 ,d2 (Proposition 8.2.11), the universal extension
ss

Euniv (Remark 8.2.13) defines a representable morphism

EXT(p∗12 F1,univ , p∗13 F2,univ ) → Bun ss

r,d ,

whose image consists of strictly semistable vector bundles E containing a subbundle

of rank r2 and slope d/r. The theorem follows from the calculation

dim EXT(p∗12 F1,univ , p∗13 F2,univ ) = dim Bun ss ss

r1 ,d1 + dim Bun r2 ,d2

+ dim Ext1OC (F1 , F2 ) − dim Ext0OC (F1 , F2 ) (for any Fi ∈ Bun ss

ri ,di )

= (r12 + r22 + r1 r2 )(g − 1) (by Riemann-Roch (8.1.4))

2
< (r1 + r2 ) (g − 1) = dim Bun ss
r,d .

Exercise 8.2.15. Provide an alternative argument for the Existence of Semistable

Vector Bundles (8.2.14).
Hint: Follow the proof above using the stack of extensions and the observation that
there are only finitely many ranks and degrees of destabilizing subbundles. Instead of
using the stack of extensions, one can also argue using dimensions of Quot schemes
as in [NR69, Lemma 4.3] and [LP97, §8.6].

8.2.6 Simple vector bundles and their rigidification

Definition 8.2.16 (Simple vector bundles). A vector bundle E on C is simple if
Aut(C) ∼ = k∗ . For a k-scheme S, a family of vector bundles E on CS over S is simple
if for every s ∈ S, the geometric fiber Es := E|C×Spec κ(s) is simple.

Every stable vector bundle is simple by Lemma 8.1.15.

Proposition 8.2.17. Let E be a family of vector bundles on CS over a k-scheme S.

(1) There is a closed immersion iE : Gm,S ,→ Aut(E) of group schemes which is
compatible under base change: if S ′ → S is a map of schemes and E ′ denotes
the pullback of E to CS ′ , then iE ′ is identified with the base changes of iE under
S ′ → S under the natural isomorphism Aut(E) ×S S ′ ∼ = Aut(E ′ ).
∼
(2) If E is simple, then iE : Gm,S → Aut(E) is an isomorphism.

453
Proof. We define iE functorially: for an S-scheme S ′ , let iE (S ′ ) : Gm,S (S ′ ) → Aut(E ′ )
be the map taking a unit f ∈ Γ(S ′ , OS ′ )∗ to the isomorphism E ′ → E ′ given by
multiplication by the unit p∗2 f ∈ Γ(CS ′ , OCS′ )∗ . Since iE (S ′ ) is injective, iE : Gm,S →
Aut(E) is a monomorphism of group schemes. To see that iE is a closed immersion,
we first claim that the algebraic space quotient X := Aut(E)/Gm,S is affine over S.
Indeed, since p : Aut(E) → S is affine and Gm is linearly reductive, the natural map
X → SpecS (p∗ OX )Gm,S is a good moduli space. By Theorem 6.5.6, this map is
universal for maps to algebraic spaces and therefore an isomorphism. The inclusion
Gm,S → Aut(E) is obtained by base changing X → X × X by Aut(E) → X × X. As
X is separated, Gm,S → Aut(E) is a closed immersion.
For (2), we first observe that the Fibral Flatness Criterion (A.2.10) implies that
iE is flat. As iE is a flat surjective monomorphism of finite type, it is an isomorphism
(Proposition A.2.15).

Exercise 8.2.18. For a vector bundle E on C, show that the following are equivalent:

E is simple ⇐⇒ Aut(E) ∼
= Gm ⇐⇒ dim Aut(E) = 1.

Definition 8.2.19. The stack of simple vector bundles of rank r and degree d is the
substack
Bun simple
r,d ⊆ Bun r,d
whose objects over a k-scheme S are families of simple vector bundles of rank r and
degree d.

Proposition 8.2.20. The substack Bun simple

r,d ⊆ Bun r,d is open and is a banded
simple
Gm -gerbe over the algebraic space Mr,d := Bun simple
r,d ( Gm .

Proof. For openness, we need to show that if E is a family of vector bundles over a
k-scheme S, then the locus

{s ∈ S | the geometric fiber Es is simple}

is an open subset of S. Indeed, by Exercise 8.2.18 this locus is the set of points s ∈ S
where the group scheme Aut(E) → S has fiber dimension 1, and this is open by the
upper semicontinuity of the fiber dimension of a group scheme (Proposition B.1.8).
∼
The canonical isomorphism iE : Gm,S → Aut(E) from Proposition 8.2.17(2) induces
an isomorphism between the inertial stack of Bun simple
r,d and Gm × Bun simple
r,d . The
statement therefore follows by rigidifying the Gm using Proposition 6.4.18 (or using
Proposition 6.4.41).

Remark 8.2.21 (Ridigification of the stack of vector bundles). We may apply Propo-
sition 6.4.41 to construction the rigidification Bun r,d( Gm of the stack of all vector
bundles using the fppf closed subgroup scheme Gm := Gm,Bun r,d of the inertia stack
IBun r,d defined by the closed immersions
S iE : Gm,S ,→ Aut(E) of Proposition 8.2.17(1).
Alternatively, by viewing Bun r,d = N [Q′N / GLP (N ) ] as a union of quotient stacks
(Theorem 3.1.21), where ′ P
S QN′ ⊆ Quot (OC (−N )
⊕P (N )
/C) are open subschemes,
then Bun r,d ( Gm = N [QN / PGLP (N ) ]. The stack Bun r,d ( Gm contains the
simple
algebraic space Mr,d as an open substack.

Exercise 8.2.22 (moderate). For each pair of integers r > 0 and d, show that
simple
Mr,d is not separated and not of finite type over k.

454
Application to stable bundles. Since Bun sr,d ⊆ Bun simple
r,d is an open substack,
we may apply Proposition 8.2.20 to conclude:

Corollary 8.2.23. The stack Bun sr,d of stable bundles is a banded Gm -gerbe over
s
the algebraic space Mr,d := Bun sr,d( Gm .

8.2.7 Vector bundles with fixed determinant

Definition 8.2.24. For a line bundle ξ ∈ Pic(C), define the stack Bun r,ξ over
(Sch/k)ét whose objects over a k-scheme S are pairs (E, φ) where E is a vector
∼
bundle on CS and φ : det E → p∗1 ξ is an isomorphism. A morphism (E, φ) → (E ′ , φ′ )
∼
is an isomorphism α : E → E ′ compatible with the isomorphisms φ and φ′ .

The stack Bun r,ξ of vector bundles with determinant ξ fits in a cartesian diagram

Bun r,ξ / Spec k

□ [ξ] (8.2.25)
 
Bun r,d
det / Pic d ,

where Picd = Bun 1,d is the Picard stack of line bundles of degree d (see also the
discussion in §6.4.7), and det : Bun r,d → Pic d is the morphism defined by E 7→ det E.
There are open substacks

Bun sr,ξ ⊆ Bun ss

r,ξ ⊆ Bun r,ξ

consisting of stable and semistable vector bundles with determinant ξ. For a

stable bundle E of rank r, there is a canonical isomorphism Aut(E) ∼ = µ r . In
particular, Bun sr,ξ is a Deligne–Mumford stack. Moreover, the same argument as
Proposition 8.2.20 shows that Bun sr,ξ is a µ r -gerbe over Mr,ξ
s
:= Bun sr,ξ( µ r .

Proposition 8.2.26. For every line bundle ξ ∈ Pic(C) and rank r > 0, there exists
a stable vector bundle E on C of rank r with det(E) ∼
= ξ.
Proof. By the Existence of Stable Bundles (8.2.14), there exists a stable bundle E of
rank r and degree d = deg ξ. Then ξ ⊗ (det E)−1 is a line bundle of degree 0. Since
Pic0 (C) is an abelian variety, it is a divisible group, and hence there is a line bundle
L of degree 0 with L⊗r ∼ = ξ ⊗ (det E)−1 . The vector bundle E ⊗ L is semistable
(Lemma 8.1.13(4)) of rank r with det(E ⊗ L) ∼ = (det E) ⊗ L⊗r ∼= ξ.

Exercise 8.2.27 (hard). If (C, p) is an elliptic curve and ξ is a line bundle of degree
ps ∼ gcd(r,d)−1
d, show that Mr,ξ =P . (This extends the classification of Exercise 8.1.47
to bundles with fixed determinant. See also [Tu93, Thm. 3].)

Remark 8.2.28. Instead of using the fiber diagram (8.2.25) to define a stack of
vector bundles with fixed determinant, one could use the residual gerbe instead:
if BGm ,→ Pic d is the inclusion of the residual gerbe of ξ, then the fiber product
X := Bun r,d ×Pic d BGm parameterizes vector bundles E of rank r such that there
∼
exists an isomorphism det E → ξ. The automorphism of a stable vector bundle
E ∈ X (k) is Gm and the stable locus X s := Bun sr,d ×Pic d BGm is a Gm -gerbe over

X s( Gm ∼ s ∼
= Mr,ξ = Bun sr,ξ( µ r .

455
8.3 Semistable reduction, moduli spaces, and strati-
fications
Algebra is the offer made by the devil to the mathematician. The devil
says: “I will give you this powerful machine; it will answer any question
you like. All you need to do is give me your soul: give up geometry and
you will have this marvelous machine.”

Michael Atiyah [Ati04, Commentary]

We prove further properties of the moduli stack of vector bundles over a curve:
the semistable locus Bun ss
r,d is universally closed (Theorem 8.3.1) and admits a proper
moduli space (Theorem 8.3.10), while the unstable locus Bun r,d ∖ Bun ss r,d admits a
stratification indexed by the Harder–Narasimhan polygon (Theorem 8.3.13).

8.3.1 Semistable reduction: Langton’s Theorem

Theorem 8.3.1 (Langton’s Theorem). Let C be a smooth, connected, and projective
curve of genus g ≥ 2 over an algebraically closed field k. Let R be a DVR over k
with fraction field K. If E ∗ is a semistable vector bundle on CK , then there exists a
family E of semistable vector bundles on CR such that EK ∼ = E∗.

Remark 8.3.2. Unlike the general case of algebraic stacks where base changes of
R are necessary for the existence part of the valuative criterion (see the examples
3.8.16–3.8.18 and 5.5.11), no base change is necessary here.

Proof. We may assume that R/π ∼ = k,4 where π is a uniformizer. Let r = rk(E ∗ )
and d = deg(E ). By the valuative criterion for Bun r,d (Exercise 3.8.13), there
∗

exists a vector bundle E on CR with EK ∼ = E ∗ . Our strategy will be to modify E

by ‘elementary modifications’ so that Ek becomes semistable. Namely, if Ek is not
semistable, then we may choose a short exact sequence

0 → D → Ek → F → 0 (⋆)

such that F is maximal among semistable quotients of F of smallest slope, i.e., D is

the last term of the Harder–Narasimhan filtration of Ek ; see Figure [Link] Define
the elementary modification

E ′ := ker(E ↠ Ek ↠ F ) ⊆ E.
′ ∼
By construction, EK = EK ∼
= E ∗ and E/E ′ ∼ = F . The containments πE ⊆ E ′ ⊆ E
∼
imply that E /πE = D by using the identifications in the short exact sequence (⋆):
′

0 → E ′ /πE → E/πE → E/E ′ → 0.

| {z } | {z } | {z }
D Ek F

We have a short exact sequence

0 → F → Ek′ → D → 0 (⋆⋆)
4 In fact, the theorem holds with the same proof when k is an arbitrary field.
5 As in Langton’s proof, one can modify the argument to take instead D ⊆ Ek to be the maximal
destabilizing subsheaf, i.e., the first term of the Harder–Narasimhan filtration of Ek .

456
 zp
NPLE

x
where the roles of D and F are flipped in comparison to (⋆). This follows from the
surjection Ek′ ∼
= E ′ /πE ′ ↠ E ′ /πE ∼
= D with kernel F ∼= πE/πE ′ .6 deg
Let HN P (Ek ) ⊆ C be the Harder–Narasimhan polygon of Ek , i.e., the convex
hull of the points Z(E ′ ) = − deg(E ′ ) + rk(E ′ )i ∈ C as E ′ ⊆ Ek ranges over all
subsheaves with µ(E ′ ) > d/r.

rk ᵗ
p Eptic
NPLE
A B
DYEND C

A
deg

Figure 8.3.3: The Harder–Narasimhan polygon HN P (Ek ) of Ek is featured in grey.

The red (resp., blue) vectors are drawn to help illustrate the proof of the inclusion
HN P (Ek′ ) ⊆ HN P (Ek ) (resp., the splitting of (⋆⋆) in the case of equality).

Mirroring the construction of (⋆) for Ek , let 0 → D′ → Ek′ → F ′ → 0 be the short

exact sequence obtained by taking D′ to be the last term in the Harder–Narasimhan
filtration of Ek′ . We claim that:
(i) HN P (Ek′ ) ⊆ HN P (Ek ), and
(ii) if HN P (Ek′ ) = HN P (Ek ), then the compositions F → Ek′ → F ′ and D′ →
Ek′ → D are isomorphisms. In particular, (⋆⋆) is split.
For (i), a subsheaf B ⊆ Ek′ induces a commutative diagram of short exact sequences

0 /A /B /C /0
_ _ _

  
0 /F / E′ /D / 0.
k

6We provide another way to think about this setup. Recall from §6.9.2 that ϕR :=
[Spec R[s, t]/(st − π) /Gm ] and that ϕR ∖ 0 ∼

S
= Spec R Spec K Spec R is the non-separated DVR.
Since E E ∗ E ′ defines family over C ×(ϕR ∖0) and since Coh(C) is S-complete (Proposition 6.9.30),
S
we know that this family extends to a family F over ϕR , but in this case
L we can describe it explicitly
using the description from Exercise 6.9.16 as a Z-graded sheaf F = n∈Z Fn on CR with maps s
and t in a diagram:
s
+ s=π
+ s=1
*
s=1
)
s
*
··· j F−2 k F−1 k F0 i F1 i ··· .
t t=1 t=π t=π t
E′ E′ E E
The exact sequences (⋆) and (⋆⋆) then correspond to the filtrations obtained by restricting to Θk
along t = 0 and s = 0, respectively.

457
Since C ⊆ D ⊆ Ek , we have that µ(C) ≤ d/r or Z(C) ∈ HN P (Ek ). Since F is
semistable, µ(A) ≤ µ(F ) (if A ̸= 0). Using that Z(B) = Z(A) + Z(C), these two
facts implies that either µ(B) ≤ d/r or that Z(B) ∈ HN P (Ek ).
For (ii), we have a short exact sequence 0 → D′ → Ek′ → F ′ → 0 with
Z(F ′ ) = Z(F ). The kernel of the composition F → Ek′ → F ′ is D′ ∩ F which sits in
a short exact sequence

0 → F ∩ D′ → D′ → D′ /F ∩ D′ → 0.

Assuming that F ∩D′ = ̸ 0, the semistability of F and the constructions of F = Ek /D

and D′ ⊆ Ek′ yield inequalities

µ(F ∩ D′ ) ≤ µ(F ) < µ(Ek ) = µ(Ek′ ) < µ(D′ ),

which in turn implies that µ(D′ /F ∩ D′ ) > µ(D′ ). Since D′ /F ∩ D′ ⊆ D ⊆ Ek ,

Z(D′ /F ∩D′ ) ∈ HN P (Ek ) lies to the left of Z(D′ ). But this implies that µ(F ∩D′ ) >
µ(F ), contradicting the semistability of F . Thus F ∩ D′ = 0 and F → Ek′ → F ′ is
injective. Since F and F ′ have the same degree and rank, F → F ′ is an isomorphism
splitting (⋆⋆).
Since there are only finitely many Harder–Narasimhan polygons contained in
HN P (Ek ), by inductively replacing E with E ′ and applying claim (i), we either ob-
tain a family of semistable vector bundles extending E ∗ or we obtain an infinite family
· · · ⊆ E (2) ⊆ E (1) ⊆ E (0) such that each E (n) has the same Harder–Narasimhan
polygon. We will show that the latter case leads to a contradiction. By claim
(ii), we know that each Ek ∼
(n)
= D ⊕ F is split. After replacing E with E (0) , the
reader is left to check that the sheaf E/E (n) on CR/mn is flat over R/mn with
(E/E (n+1) ) ⊗R/mn+1 R/mn ∼ = E/E (n) (Exercise 8.3.4). This implies that we have
compatible lifts

[Ek ↠F ]
Spec _ k / QuotPF (E/CR /R)
4
[ER/mn ↠E/E (n) ]

  
Spec R/mn  / Spec R,

where PF is the Hilbert polynomial of F . By properness of the Quot scheme

(Proposition 1.4.2), the map QuotPF (E/CR /R) → Spec R has closed image and
is thus surjective. Therefore, there exists an extension R → R′ and a section
[ER′ ↠ Fe] : Spec R′ → QuotPF (E/CR /R). (If R were complete, Grothendieck’s
Existence Theorem (C.5.3) would imply the existence of a section without an
extension.) Restricting to the fraction field K ′ of R′ , we have a quotient EK ′ ↠ FeK ′
with µ(FeK ′ ) > µ(EK ′ ). This implies that EK ′ is not semistable. Since semistability
is insensitive to field extension (Corollary 8.1.29), this contradicts the semistability
of EK . See also [Lan75], [HL10, Thm. 2.B.1], and [Mar78, Thm. 5.7].

Exercise 8.3.4. Verify the missing details in the proof above: (E/E (n+1) ) ⊗R/mn+1
R/mn ∼
= E/E (n) and E/E (n) is flat over R/mn .
Hint: First show that (E/E (n) )k ∼
= F . Then compare the short exact sequence
0 → E (n) /E (n+1) → E/E (n+1) → E/E (n) → 0 with the right exact sequence

E/E (n+1) ⊗R/mn+1 mn → E/E (n+1) → E/E (n+1) ⊗R/mn+1 R/mn → 0

458
using the identifications E (n) /E (n+1) ∼
=F ∼= (E/E (n+1) )k ∼
= E/E (n+1) ⊗R/mn+1 mn .
(n+1) n+1
Establish the flatness of E/E over R/m by using induction on n and the
Local Criterion for Flatness (A.2.6) with respect to R/mn+1 → R/mn .

Corollary 8.3.5. The algebraic stack Bun ss

r,d (C) is universally closed over k.

Proof. By Boundedness of Semistability (8.2.8), Bun ss

r,d (C) is of finite type and,
in particular, quasi-compact. We may therefore apply the Valuative Criteria for
Universal Closedness (3.8.7).

8.3.2 Existence of a proper moduli space

Recall that Θ- and S-completeness are valuative criteria (Definitions 6.9.10 and 6.9.14)
which guarantee the existence of a separated good moduli space (Theorem 6.10.1).
We will verify that Bun ss
r,d is both Θ- and S-complete by interpreting maps from Θ
as filtrations of coherent sheaves (Proposition 6.9.6) and exploiting that the stack
Coh(X) parameterizing coherent sheaves is Θ- and S-complete (Proposition 6.9.30),
a fact that holds for any projective scheme X. To this end, we provide a general
criterion for when an open substack of Coh(X) is Θ- and S-complete. To formulate
the S-completeness criterion, we will need the following notion: two Z-graded
filtrations
E• : 0 ⊆ · · · ⊆ Ei−1 ⊆ Ei ⊆ Ei+1 ⊆ · · · ⊆ E
and
F • : F ⊇ · · · ⊇ F i−1 ⊇ F i ⊇ F i+1 ⊇ · · · ⊇ 0
are called opposite if Ei /Ei−1 =∼ F i /F i+1 for all i. Extending the correspondence
between filtrations and maps from Θ, the data of opposite filtrations E• and F • such
that Ei = 0 and F i = F for i ≪ 0, and Ei = E and F i = 0 for i ≫ 0 is equivalent
to giving a map
[(Spec k[x, y]/xy)/Gm ] → Coh(X),
where t · (x, y) = (tx, t−1 y). In this case, this map takes (1, 0) 7→ E, (0, 1) 7→ F , and
(0, 0) 7→ gr E• ∼
= gr F • .
Proposition 8.3.6. Let X be a projective scheme over an algebraically closed field
k, and let U ⊆ Coh(X) be an open substack.
(1) The substack U is Θ-complete if and only if for every DVR R (with fraction
field K and residue field κ), for every coherent sheaf E on XR flat over R,
and for every Z-graded filtration E• satisfying Ei = 0 for i ≪ 0, Ei = E for
i ≫ 0, and with each Ei /Ei−1 flat over R, then if E and gr(E• |K ) are in U,
then gr(E• |κ ) must be in U as well.
(2) The substack is S-complete if for every pair of opposite filtrations E• , F • of
E, F ∈ U(κ) for a field extension κ/k, the associated graded gr E• is in U(κ).

Proof. Since we already know that Coh(X) is Θ- and S-complete (Proposition 6.9.30),
the valuative criteria for U are equivalent to the existence of lifts for all commutative
diagrams

_ 0
ΘR ∖ /9 U _ 0
ϕR ∖ /9 U
_ _
and
   
ΘR / Coh(X) ϕR / Coh(X),

459
where R is a DVR. In other words, we need to show that the images of 0 under
the unique fillings ΘR → Coh(X) and ϕR → Coh(X) are contained in U. By
Proposition 6.9.6, the map ΘR → Coh(X) is classified by a coherent sheaf E on CR
flat over R and a Z-graded filtration E• satisfying Ei = 0 for i ≪ 0, Ei = E for i ≫ 0,
and with each Ei /Ei−1 flat over R. Moreover, the restriction ΘR ∖ 0 → Coh(X)
factors through U if and only if E and gr(E• |K ) are in U. As the image of 0 under
ΘR → Coh(X) is gr(E• |κ ), we see that part (1) holds.
For (2), the restriction of ϕR → Coh(X) along π = 0 yields a map [Spec(κ[x, y]/xy)/Gm ] →
Coh(X) corresponding to opposite filtrations E• and F • . If gr E• is in U, then the
image of ϕR → Coh(X) is contained in U.

While not needed for later, the following exercise is conceptually helpful.

Exercise 8.3.7. Show that the converse of Proposition 8.3.6(2) holds when X = C
is a curve.
Hint: Letting X0 := [Spec(k[x, y]/xy)/Gm ] ⊂ ϕR and Xn be its nilpotent thickenings,
use deformation theory to extend a map X0 → Coh(C) to compatible maps Xn →
Coh(C) and use Coherent Tannaka Duality (6.6.9) to extend further to a map
ϕR → Coh(C).

Proposition 8.3.8. For a smooth, connected, and projective curve C over an

algebraically closed field k, the algebraic stack Bun ss
r,d is Θ- and S-complete.

Proof. We will apply Proposition 8.3.6 to establish each property. For Θ-completeness,
let E be a coherent sheaf on CR flat over R and E• be a Z-graded filtration satisfying
Ei = 0 for i ≪ 0, Ei = E for i ≫ 0, and with each Ei /Ei−1 flat over R. Suppose that
E and gr(E• |K ) are in Bun ss
r,d . Since the degree and rank are constant in flat families,
gr(E• |κ ) ∈ Cohr,d . To show that gr(E• |κ ) is semistable, it suffices to show that each
factor (Ei /Ei−1 )κ is semistable of slope d/r. Since (Ei )κ ⊆ Eκ is a subsheaf of a
semistable sheaf of the same slope, each (Ei )κ is semistable. As semistable sheaves of
a fixed slope form an abelian category (Exercise 8.1.17), (Ei /Ei−1 )κ = (Ei )κ /(Ei−1 )κ
is semistable of slope d/r.
For S-completeness, suppose that

E• : · · · ⊆ Ei−1 ⊆ Ei ⊆ · · · ⊆ E and F • : F ⊇ · · · ⊇ F i ⊇ F i+1 ⊇ · · ·

are opposite filtrations of semistable vector bundles E and F on Cκ for a field ex-
tension κ/k. This means that Ei /Ei−1 ∼ = F i /F i+1 . We will prove by induction that
each Ei andLF is semistable of slope d/r. This establishes the desired semistability
i

of gr E• = Ei /Ei−1 by Exercise 8.1.17. Since Ei = 0 and Fi = F for i ≪ 0 and

F is semistable, Ej and Fj are semistable for i ≪ 0. Let’s suppose that Ej , Fj are
semistable for j ≤ i. Since Fi+1 ∼ = ker(Fi → Ei /Ei−1 ) is the kernel of semistable
vector bundles of the same slope, Fi+1 is semistable of slope d/r. On the other hand,
we can express Ei+1 /Ei ∼ = Fi+1 /Fi+2 both as a subsheaf of the semistable vector
bundle E/Ei (using the semistability of E and the inductive hypothesis) and as a
quotient of the semistable vector bundle Fi+1 . It follows that µ(Ei+1 /Ei ) = d/r,
and hence Ei+1 is semistable of slope d/r.

The stack Bun r,d of vector bundles, on the other hand, is not Θ-complete nor
S-complete.

Example 8.3.9. Suppose that g ≥ 2. Let p ∈ C(k) be a point defined by the

vanishing of a section s ∈ Γ(C, O(p)), and let I ⊆ OCR be the ideal sheaf of

460
(p, 0) ∈ C × Spec R. The injection (s, −π) : OCR (−p) ,→ OCR ⊕ OCR (−p) has
quotient I, which is torsion free, hence flat over R, but is not a vector bundle. By
Proposition 8.3.6, Bun 2,−1 is not Θ-complete.
Let L and M be line bundles on C, and let p ∈ C(k) be a point such that
Ext1OC (M, L(p)) and Ext1OC (L, M (p)) are nonzero; if L and M have the same
degree, then a Riemann–Roch calculation shows that both Ext1 groups are nonzero.
Let Q (resp., Q′ ) be a non-trivial extension of M by L(p) (resp., L by M (p)). Then

E• : 0 ⊆ L ⊆ L(p) ⊆ Q and F • : Q′ ⊇ M (p) ⊇ M ⊇ 0

define opposite filtrations where E0 = L and F 0 = Q′ . The associated graded

gr E• = L ⊕ κ(p) ⊕ M is not a vector bundle. By Proposition 8.3.6, Bun 1,d is not
S-complete where d = deg L + deg M + 1.
Theorem 8.3.10. Let C be a smooth, connected, and projective curve over an
algebraically closed field k of characteristic 0. For each r > 0 and d, there exists a
cartesian diagram 
Bun sr,d  / Bun ss
r,d

□
  
s  / M ps ,
Mr,d r,d
ps ps
where Mr,d is a proper algebraic space over k, Bun ss
r,d → Mr,d is a good moduli space,
s s
and Bun r,d → Mr,d is a banded Gm -gerbe. Moreover,
ps
dim Mr,d = dim Bun ss 2
r,d + 1 = r (g − 1) + 1.

ps
Proof. The existence of a separated algebraic space π : Bun ssr,d → Mr,d follows directly
ps
from the Existence Theorem of Good Moduli Spaces (6.10.1) using that Mr,d is of
finite type by Boundedness of Semistability (8.2.8) and both Θ- and S-complete
(Proposition 8.3.8). Since Bun ss
r,d is universally closed by Langton’s Theorem (8.3.5),
ps
Mr,d is proper.
Since every stable bundle is polystable, hence a closed point of Bun ss r,d by Propo-
sition 8.1.35, it follows that π (π(Bun r,d )) = Bun r,d . Since π is universally closed,
−1 s s
ps
π(Bun sr,d ) ⊆ Mr,d is open. On the hand, Corollary 8.2.23 gives the existence of
a banded Gm -gerbe Bun sr,d → Mr,d s
. Since both good moduli spaces and banded
s ∼
Gm -gerbes are universal for maps to algebraic spaces, Mr,d = π(Bun sr,d ).
ps
To compute the dimension, we will use that Mr,d ⊆ Mr,d is open and dense; this
s

follows from the fact Bun sr,d ⊆ Bun r,d is open and dense (Theorem 8.2.14). Since
Bun s → Mr,d s
is a banded Gm -gerbe and dim BGm = −1, we conclude that
ps s
dim Mr,d = dim Mr,d = dim Bun sr,d + 1 = r2 (g − 1) + 1,

using the calculation of the dimension of Bun r,d from Proposition 3.7.9.
Exercise 8.3.11. For every ξ ∈ Pic(C), denote by Bun ss r,ξ (resp., Bun r,ξ ) the
s

substack of the moduli stack Bun r,ξ parameterizing semistable (resp., stable) vector
bundles with determinant ξ (Definition 8.2.24). Show that Bun ss r,ξ is Θ- and S-
complete. Conclude that in characteristic 0, there exists a proper good moduli space
ps ps
r,ξ → Mr,ξ restricting to a µ r -gerbe Bun r,ξ → Mr,ξ with Mr,ξ ⊆ Mr,ξ an open
Bun ss s s s

subspace.
Hint: Use that Bun ss ss
r,ξ → Bun r,deg L is affine.

461
 Note that every closed point of Bun ss
r,d (C) (and similarly for Bun r,ξ (C)) has reduc-
ss

tive automorphism group; this follows either from the S-completeness of Bun ss r,d (C)
using Proposition 6.10.30, or in characteristic 0 by the existence of a good moduli
space using Exercise 6.5.25. Since the closed points of Bun ss r,d (C) correspond to
polystable vector bundles (a fact that holds generally but that we only proved
in Proposition 8.1.35 in characteristic 0), we recover the fact that a polystable
vector bundle has a reductive automorphism group, which we established directly in
Corollary 8.1.16. A strictly semistable but non polystable vector bundle, by contrast,
may have a non-reductive automorphism group.
Exercise 8.3.12. Construct a semistable vector bundle with non-reductive auto-
morphism group.
Hint: Consider a non-split extension of stable vector bundles.

8.3.3 A stratification of the unstable locus

The moduli stack Bun r,d of vector bundles admits a stratification into locally closed
substacks indexed by the Harder–Narasimhan polygon. This stratification is called
the Harder–Narasimhan or Shatz stratification. It was first proven in [Sha77], and as
with most other results in this chapter (e.g., the Harder–Narasimhan Filtration and
Langton’s Theorem), it generalizes to torsion free sheaves on projective varieties of
arbitrary dimension.
Recall that if F is a vector bundle on C, the Harder–Narasimhan polygon
HN P (F ) ⊆ C is the convex hull of Z(E) = − deg(E) + rk(E)i as E ⊆ F ranges over
all destabilizing subsheaves, i.e., µ(E) ≥ µ(F ) or equivalently arg Z(E) ≥ arg Z(F );
see Figure 8.1.25. If 0 = F0 ⊆ F1 ⊆ · · · ⊆ Fl = F is the Harder–Narasimhan
Filtration (8.1.21), then Z(Fi ) are the extremal points of HN P (F ) (Corollary 8.1.26).
We will use the partial order on Harder–Narasimhan polygons given by containment.
We also note that if F is semistable, then HN P (F ) is the straight line from 0 to
Z(F ).
Theorem 8.3.13 (The Harder–Narasimhan/Shatz Stratification of Bun r,d ). Let C
be a smooth, connected, and projective curve of genus g ≥ 2 over an algebraically
closed field k. There is a stratification
a
Bun r,d = MP ,
convex polygons P

where MP ⊆ Bun r,d is a locally closed substack such that for a vector bundle E on
C of rank r and degree d, then
E ∈ MP ⇐⇒ HN P (E) = P.
Moreover, for each convex polygon P:
(1) Let S be a k-scheme and F be a family of vector bundle on CS of rank r and
degree d.
(a) The map S → Bun r,d classifying F factors through MP if and only if
there exists a filtration
0 = F0 ⊆ F1 ⊆ · · · ⊆ Fl = F
of vector bundles with vector bundle factors Fi /Fi−1 such that for every
s ∈ S, HN P (Fs ) = P and 0 = F0,s ⊆ F1,s ⊆ · · · ⊆ Fl,s = Fs is the
Harder–Narasimhan filtration of Fs .

462
 (b) If s ⇝ s0 is a specialization, then HN P (Fs ) ⊆ HN P (Fs0 ).
(2) the union [
M≤P := MP ′
P ′ ⊆P

is an open substack of Bun r,d , while MP ⊆ M≤P is a closed substack.

(3) Letting e1 , . . . , el ∈ C denote the extremal points of P with increasing imaginary
parts and P1 , . . . , Pl denote the Hilbert polynomials corresponding to vector
bundles Gi with Z(Gi ) = ei − ei−1 , there is an identification

MP ∼
= FlagP1 ,...,Pl (Euniv /C × M≤P /M≤P )

with the Flag scheme (Exercise 1.4.7) over M≤P parameterizing filtrations F•
of the universal bundle Euniv whose ith factor Fi /Fi−1 has Hilbert polynomial
Pi .

Remark 8.3.14. When P corresponds to the straight line from 0 to Z(F ), MP =

r,d is the semistable locus.
Bun ss
If S is a noetherian k-scheme and F is a family of vector bundles on CS , there is
a finite stratification S = SP with each SP ⊆ S locally closed and stable under
`
base change (i.e., for T → S, TP = T ×S SP ).

Proof. We begin by showing that if F is a vector bundle ` on CS over a noethe-

rian scheme S, then there is a finite stratification S = P SP into locally closed
subschemes. First, there are only finitely many polygons P that can arise as the
Harder–Narasimhan polygon of a fiber Fs ; this is because the degree of a subsheaf
of a fiber is bounded from above (Exercise 8.1.28). Second, for each such polygon P
whose extremal vertices have Hilbert polynomials P1 , . . . , Pl (following the notation
of (3)), there is a Flag scheme FlagP1 ,...,Pl (F/CS /S) proper over S. Denote the
closed image as
S≥P := im(FlagP1 ,...,Pl (F/CS /S) → S).
A point s ∈ S lies in S≥P if and only if Fs admits a filtration whose associated
polygon is P. Then [
SP := S≥P ∖ S≥P ′
P ′ ⊋P

is an open substack of S≥P . Since the Harder–Narasimhan polygon contains every

destabilizing subsheaf (Corollary 8.1.26), it follows that s ∈ S lies in SP if and only
if HN P (Fs ) = P.
To see (1)(a) let Spec R → S be a map from a DVR representing the specialization.
Since the Harder–Narasimhan filtration is stable under field extensions, it suffices to
show that HN P (FK ) ⊆ HN P (Fκ ), where K and κ are the fraction and residue field
of R. By the valuative criterion for Quot schemes (Proposition 1.4.2), the Harder–
Narasimhan filtration of FK extends to a filtration of FR , which in turn restricts to
a filtration of Fκ . Since each term of this filtration of Fκ lies in HN
S P (Fκ ), we have
that HN P (FK ) ⊆ HN P (Fκ ). Part (1)(a) implies part S≤P := P ′ ⊆P SP ′ is open:
it is constructible and stable under generization.
It remains to show that the scheme structures SP ⊆ S are functorial in S, which
will follow if we can establish that SP ∼ = FlagP1 ,...,Pl (F/C × S≤P /S≤P ). In other
words, we need to show that the proper morphism

FlagP1 ,...,Pl (F/C × S≤P /S≤P ) → S≤P

463
is a closed immersion mapping isomorphically onto SP . To see this, it suffices to
show that it is injective on tangent spaces or, in other words, that the relative
tangent spaces are zero. We can factor the above map as a composition

FlagP1 ,...,Pl (F/C × S≤P /S≤P ) → FlagP2 ,...,Pl (F/C × S≤P /S≤P ) → · · ·
→ QuotPl (F/C × S≤P /S≤P ) → S≤P ,

where each morphism is identified by a relative Quot scheme (see Exercise 1.4.7). By
analyzing the relative tangent space of each map above, we can reduce the injectivity
claim to the following statement: if 0 = Fs,0 ⊆ Fs,1 ⊆ · · · ⊆ Fs,l = Fs is the
Harder–Narasimhan filtration of a fiber Fs over s ∈ SP and q : Fs,i ↠ Fs,i /Fs,i−1
denotes the ith quotient, then Tq QuotPi (Fs,i /Cs /κ(s)) = 0. By Exercise 1.5.6,
this tangent space is identified with HomOCs (Fs,i , Fs,i /Fs,i−1 ). Since the slope of
every factor in the Harder–Narasimhan filtration of Fs,i is strictly greater than the
slope of the semistable sheaf Fs,i /Fs,i−1 , there are no nontrivial homomorphisms
(Exercise 8.1.31). See also [Sha77] and [Nit11].
Remark 8.3.15 (Relative Harder–Narasimhan Filtrations). As a consequence of the
theorem, for a vector bundle F on CS , the restriction FSP of a vector bundle F
admits a filtration 0 = F0 ⊆ F1 ⊆ · · · ⊆ Fl = FSP of vector bundles with vector
bundle factors Fi /Fi−1 . This filtration restricts to the Harder–Narasimhan filtration
of Fs on Cs for each s ∈ SP , and for this reason is called the relative Harder–
Narasimhan filtration. By taking P to be minimal among Harder–Narasimhan
polygons of the fibers Fs , we see that there is an open subscheme U := SP ⊆ S is
an open subscheme over which there exists a relative Harder–Narasimhan filtration.
While this filtration may not extend over S in general, it does extend after a blow-up
[HL10, Thm. 2.3.2].
Remark 8.3.16 (Θ-stratification). The Harder–Narasimhan/Shatz stratification of
Bun r,d is an example of a Θ-stratification as defined in Remark 7.7.8. For each
convex polygon P, M≤P := P ′ ≤P MP ′ ⊆ Bun r,d is an open substack and there is
S
an identification
′
Mor([A1 /Gm ], M≤P ) ∼
a
= FlagP (Euniv /C × M≤P /M≤P ),
P′

′ ′ ′
where FlagP is shorthand notation for FlagP1 ,...,Pl as in part (3) above; this fact
is an enhancement of the correspondence between maps from [A1 /Gm ] and fil-
trations (Proposition 6.9.6). Taking P ′ = P, the Flag scheme FlagP (Euniv /C ×
M≤P /M≤P ) ⊆ Mor([A1 /Gm ], M≤P ) is a union of connected components and
by part (3) defines a Θ-stratum, i.e., that the induced map FlagP (Euniv /C ×
M≤P /M≤P ) → M≤P is a closed immersion mapping isomorphically onto MP .

464
Appendix A

Morphisms of schemes

The theory of schemes is widely regarded as a horribly abstract algebraic

tool that hides the appeal of geometry to promote and overwhelming and
often unnecessary generality. By contrast, experts know that schemes
make things simpler.

David Eisenbud and Joe Harris [EH00]

We recall definitions and properties of morphisms of schemes—locally of finite

presentation, flat, smooth, étale, unramified, quasi-finite, and proper. As we intend
to highlight results important in moduli theory, we pay close attention to functorial
properties, i.e., properties of schemes and their morphisms characterized by their
functors.
While first courses in algebraic geometry (e.g., [Har77]) often impose noetherian
hypotheses, we try to state results in the non-noetherian setting when possible.
Since we define moduli functors and stacks on the entire category of schemes, it is
essential to work with non-noetherian schemes. Limit Methods (B.3) allow one to
reduce properties of general schemes to noetherian schemes.

A.1 Morphisms locally of finite presentation

A morphism of schemes f : X → Y is locally of finite type (resp., locally of fi-
nite presentation) if for all affine open subschemes Spec B ⊆ Y and Spec A ⊆
f −1 (Spec B), there is a surjection B[x1 , . . . , xn ] → A of B-algebras (resp., a sur-
jection B[x1 , . . . , xn ] → A with finitely generated kernel). If in addition f is
quasi-compact (resp., quasi-compact and quasi-separated), we say that f is of finite
type (resp., of finite presentation). When Y is locally noetherian, being locally of fi-
nite type (resp., finite type) is equivalent to being locally of finite presentation (resp.,
finite presentation). In the non-noetherian setting, even closed immersions may not
be locally of finite presentation, e.g., Spec C ,→ Spec C[x1 , x2 , . . .]. Morphisms of
finite presentation are better behaved than morphisms of finite type. Many standard
results (e.g., Semicontinuity (A.6.4)) for proper flat morphisms do not hold in the
non-noetherian setting without a finite presentation condition; see [Vak17, §28.2.11]
and [SP, Tag 05LB]. Therefore, in this text, when we define for instance a family of
stable curves π : C → S (Definition 5.3.19), we require not only that π is proper and
flat but also of finite presentation.

465
 The functorial characterization of locally of finite presentation morphisms uses
the notion of an inverse system in a category C: a partially ordered set (I, ≥) which
is directed, i.e., for every i, j ∈ I there exists k ∈ I such that k ≥ i and k ≥ j,
together with a contravariant functor I → C.

Proposition A.1.1. A morphism f : X → Y of schemes is locally of finite presen-

tation if and only if for every inverse system {Spec Aλ }λ∈I of affine schemes over
Y , the natural map

colimλ MorY (Spec Aλ , X) → MorY (Spec(colimλ Aλ ), X) (A.1.2)

is bijective.

Proof. See [EGA, IV.8.14.2] or [SP, Tag 01ZC], where (A.1.2) is also shown to be
bijective for inverse systems of quasi-compact and quasi-separated schemes with
affine transition maps.

This is not a deep result: requiring that every map Spec A → X over Y factors
through some Spec Aλ → Spec A is essentially the condition that Spec A → X
depends on only finite data, i.e., there are only finitely many generators and relations
for the ring maps locally defining X → Y .

Exercise A.1.3. Verify Proposition A.1.1 in the case of a morphism Spec A →

Spec B of affine schemes.

Proposition A.1.1 is a functorial condition as it only depends on the functor

MorY (−, X), and therefore we can extend the definition of locally of finite presenta-
tion to functors.

Definition A.1.4. Let Y be a scheme. A contravariant F : Sch /Y → Sets is locally

of finite presentation (or limit preserving) if for every inverse system {Spec Aλ }λ∈I
of affine schemes over Y , the natural map

colimλ F (Aλ ) → F (colimλ Aλ )

is bijective.

By Proposition A.1.1, a scheme X is locally of finite presentation over Y if and

only if the functor MorY (−, X) is locally of finite presentation.

A.2 Flatness
Art is fire plus algebra.

Jorge Luis Borges

You cannot get very far in moduli theory without flatness. While its definition is
seemingly abstract and algebraic, it is a magical geometric property of a morphism
X → Y that ensures that fibers Xy ‘vary nicely’ as y ∈ Y varies. This principle is
nicely evidenced by Flatness via the Hilbert Polynomial (A.2.4). It is the reason
why we define objects of our moduli stacks as flat families.

466
A.2.1 Flatness criteria
A module M over a ring A is flat if the functor

− ⊗A M : Mod(A) → Mod(A)

is exact. We recall the following criteria:

(1) (Stalk Criterion) M is flat over A if and only if Mp is flat over Ap for every
prime (equivalently maximal) ideal p. More generally, if A → B is a ring map,
a B-module N is flat if and only if for every prime q ⊆ B with preimage p ⊆ A,
Nq is flat over Ap .
(2) (Ideal Criterion) M is flat if and only if for every finitely generated ideal I ⊆ A,
the map I ⊗A M → M is injective [Eis95, Prop. 6.1]. (When A is a PID, this
implies that M is flat if and only if M is torsion free.)
(3) (Tor Criterion) M is flat if and only if TorA
1 (A/I, M ) = 0 for all finitely
generated ideals I ⊆ A [Eis95, Prop. 6.1].
(4) (Finitely Presented Criterion) M is finitely presented and flat over A if and
only if M is finite and projective if and only if M is finite and locally free (i.e.,
Mp is finite and free for all prime—or equivalently maximal—ideals p); see
[SP, Tag 00NX]. (Without the finitely presented hypothesis, Lazard’s Theorem
states that M is flat over A if and only if M can be written as a directed colimit
colimi∈I Mi of free finite A-modules Mi ; see [Eis95, A6.6] or [SP, Tag 058G].)
Pn
(5) (Equational Criterion) M is flat if and only if for every relation i=1 ai mi = 0
with ai ∈ A and mi ∈ M , there exists m′j ∈ M for j = 1, . . . , r and a′ij ∈ A
Pr Pn
such that j=1 aij mj = mi for all i and
′ ′
i=1 aij ai = 0 for all j [Eis95,
′

Cor. 6.5].
If f : X → Y is a morphism of schemes, then a quasi-coherent OX -module F
is flat if for all affine opens Spec B ⊆ Y and Spec A ⊆ f −1 (Spec B), the B-module
Γ(Spec A, F) is a flat.

Proposition A.2.1 (Flat Equivalences). Let f : X → Y be a morphism of schemes

and F be a quasi-coherent OX -module. The following are equivalent:
(1) F is flat over Y ;
(2) There exists a Zariski-cover {Spec Bi } of Y and {Spec Aij } of f −1 (Spec Bi )
such that Γ(Spec Aij , F) is flat as an Bi -module under the ring map Bi → Aij ;
(3) For all x ∈ X, the OX,x -module Fx is flat as an OY,y -module.
(4) The functor

QCoh(Y ) → QCoh(X), G 7→ f ∗ G ⊗OX F

is exact.

Proof. See [Har77, §III.9] or [SP, Tag 01U2].

We say that a morphism f : X → Y of schemes is flat at x ∈ X (resp., a quasi-

coherent OX -module F is flat at x ∈ X) if there exists a Zariski open neighborhood
U ⊆ X containing x such that f |U (resp., F|U ) is flat over Y . This is equivalent to
the flatness of OX,x (resp., Fx ) as an OY,y -module.

Proposition A.2.2 (Flatness Criterion over Smooth Curves). Let C be an integral

and regular scheme of dimension 1 (e.g., the spectrum of a DVR or a smooth

467
connected curve over a field), and let X → C be a morphism of schemes. A quasi-
coherent OX -module F is flat over C if and only if every associated point of F maps
to the generic point of C.

Proof. A short argument shows that this follows from the fact that a module over a
DVR is flat if and only if it is torsion free; see [Har77, III.9.7].

Over higher dimensional bases, it is sometimes possible to check flatness by

reducing to the above criterion over a smooth curve. This is called the valuative
criterion for flatness: if f : X → S is a finite type morphism of noetherian schemes
over a reduced scheme S and F is a coherent OX -module, then F is flat at x ∈ X if
and only if for every map (Spec R, 0) → (S, f (x)) from a DVR, the restriction F|XR
is flat over R at all points in XR := X ×S Spec R over 0 and x [EGA, IV.11.8.1].
Despite providing a conceptual geometric criterion for flatness, it is surprisingly
rarely used in moduli theory.

Proposition A.2.3 (Flatness Criterion over Artinian Rings). A module over an

artinian ring is flat if and only if it is free if and only if it is projective.

Proof. See [SP, Tag 051E].

Recall that if X ⊆ PnK is a subscheme and F is a quasi-coherent OX -module, the

Hilbert polynomial of F is PF (z) = χ(X, F(z)) ∈ Q[z].

Proposition A.2.4 (Flatness via the Hilbert Polynomial). Let S be a reduced,

connected, and noetherian scheme, and let X ⊆ PnS be a closed subscheme. A coherent
OX -module F is flat over S if and only if the function

S → Q[z], s 7→ PF |Xs

assigning a point s ∈ S to the Hilbert polynomial of the restriction F|Xs to the fiber
Xs ⊆ Pnκ(s) is constant.

Proof. See [Har77, Thm. 9.9].

Theorem A.2.5 (Local and Infinitesimal Criteria for Flatness). Let A → B be a

local homomorphism of noetherian local rings, and let M be a finite B-module. The
following are equivalent:
(1) M is flat over A,
(2) (Local Criterion) TorA
1 (A/mA , M ) = 0, and
(3) (Infinitesimal Criterion) M/mnA M is flat over A/mnA for every n ≥ 1.

Proof. See [Eis95, Thm. 6.8, Exc. 6.5] or [SP, Tag 00MK].

The following consequence of the Local Criterion for Flatness is particularly

useful in deformation theory.

Corollary A.2.6. Let A ↠ A0 be a surjective homomorphism of noetherian rings

with kernel I such that I 2 = 0. An A-module M is flat over A if and only if
(1) M0 := M ⊗A A0 is flat over A0 , and
(2) the map M0 ⊗A0 I → M is injective.

468
Proof. For (⇒), condition (1) holds by base change and condition (2) holds by
tensoring the exact sequence 0 → I → A → A0 → 0 with M and using the
identification M ⊗A I ∼ = M0 ⊗A0 I. For (⇐), by the Local Criterion for Flatness
(A.2.5) it suffices to show that TorA
1 (A/p, M ) = 0 for all prime ideals p ⊆ A. Let
p0 := p/I ⊆ A. Consider the following diagram which is obtained by tensoring the
exact sequences 0 → I → p → p0 → 0 and 0 → I → A → A0 → 0 with M :

TorA
1 (M, A/p)
/ TorA0 (M0 , A0 /p0 )
1

 
0 / M ⊗A I / M ⊗A p / M0 ⊗A0 p0 /0

 
0 / M ⊗A I /M / M0 /0

 
M ⊗A A/p / M0 ⊗A A0 /p0
0

Condition (2) implies that the second row is exact, and it follows that the first row is
also exact, where we have used the identification M ⊗A p0 ∼ = M0 ⊗A0 p0 . Condition
(1) implies that TorA
1
0
(M 0 , A0 /p0 ) = 0 and it follows from the snake lemma that
TorA1 (M, A/p) = 0. See also [Har10, Prop. 2.2].

Remark A.2.7. Applying this with A = k[ϵ]/(ϵ2 ) being the dual numbers and A′ = k,
ϵ
we recover the fact that an A-module M is flat if and only if M ⊗k[ϵ]/(ϵ2 ) k →
− M is
injective. This also follows from the fact that a module N over a ring B is flat if
and only if for every ideal I ⊆ B, the map I ⊗B M → M is injective, and using that
the only ideal in k[ϵ]/(ϵ2 ) is (ϵ).
The following convenient facts are closely related to the Local Criterion of
Flatness (A.2.5).

Lemma A.2.8. Let (A, mA ) → (B, mB ) be a local ring homomorphism of noetherian

local rings.
(1) Let M be a flat A-module and N be a finitely generated B-module. If ϕ : N →
M is a morphism of R-modules such that N/mN → M/mM is injective, then
ϕ : N → M is injective and M/ϕ(N ) is flat over A.
(2) If in addition A → B is flat and f ∈ mB is a nonzerodivisor in B ⊗A A/mA ,
then A → B/(f ) is flat.

Proof. The proofs are elementary (see [Mat89, Thm. 22.5] or [SP, Tag 00ME]). Note
that Part (2) follows directly from (1).

Part (2) can be viewed as a ‘slicing criterion for flatness’ and is often applied
inductively to regular sequences. It has the following geometric interpretation.

Corollary A.2.9 (Slicing Criterion for Flatness). Let f : X → S be a morphism

locally of finite presentation, and let x ∈ X be a point with image s ∈ S. If f is
flat at x and the image of h ∈ mx ⊆ OX,x in the local ring OXs ,x of the fiber is
a nonzerodivisor, then there exists an open neighborhood U ⊆ X of x such that h
extends to a global function on U and the composition V (h) ,→ U → S is locally of
finite presentation and flat at x.

469
Proof. The noetherian case is a direct consequence of Lemma A.2.8(2), and the
general case can be reduced to the noetherian case using the limit methods of §B.3.
See also [SP, Tag 056X].
Theorem A.2.10 (Fibral Flatness Criterion). Consider a commutative diagram

X /Y

 
S
of schemes, and let F be a quasi-coherent OX -module of finite presentation. Assume
that X → S is locally of finite presentation and Y → S is locally of finite type. Let
x ∈ X with images y ∈ Y and s ∈ S. If the stalk Fx is nonzero, then the following
are equivalent:
(1) F is flat over S at x, and Fs := F|Xs is flat over Ys at x, and
(2) Y is flat over S at y and F is flat over Y at x.
Proof. See [SP, Tag 039A].
If A → B is a local ring map of noetherian local rings, then dim B = dim A +
dim B/mA B. The following is a partial converse.
Theorem A.2.11 (Miracle Flatness). Let A → B be a local homomorphism of
noetherian local rings. Assume that
1. A is regular,
2. B is Cohen–Macaulay, and
3. dim B = dim A + dim B/mA B.
Then A → B is flat.
Proof. See [Nag62, Thm. 25.16] or [SP, Tag 00R4].

A.2.2 Properties of flatness

Proposition A.2.12 (Flat Base Change). Consider a cartesian diagram

g′
X′ /X

f′ □ f
 
Y′
g
/Y

of schemes, and let F be a quasi-coherent sheaf on X. If g : Y ′ → Y is flat and

f : X → Y is quasi-compact and quasi-separated, the natural adjunction map

g ∗ Ri f∗ F → Ri f∗′ g ′∗ F

is an isomorphism for all i ≥ 0.

Proof. See [Har77, Prop. III.9.3] or [SP, Tag 02KH].
Theorem A.2.13 (Generic Flatness). Let f : X → S be a finite type morphism of
schemes and F be a finite type quasi-coherent OX -module. If S is reduced, there
exists a dense open subscheme U ⊆ S such that XU → U is flat and of presentation
and such that F|XU is flat over U and of finite presentation as on OXU -module.

470
Proof. See [SP, Tag 052B].
Proposition A.2.14 (Fppf Morphisms are Open). Let f : X → Y be a morphism
of schemes. If f is flat and locally of finite presentation, then for every open subset
U ⊆ X, the image f (U ) ⊆ Y is open.
Proof. See [SP, Tag 01UA].
Proposition A.2.15. A flat monomorphism locally of finite presentation (e.g., an
étale monomorphism) is an open immersion.
Proof. This is a nice application of descent theory. By the previous result, it suffices
to assume that f : X → Y is surjective, hence fppf. Since X → Y is a monomorphism,
the base change X ×Y X → Y is an isomorphism. Since being an isomorphism is an
Fpqc Local Property on the Target (2.1.26), X → Y is an isomorphism.
The following theorem is the an essential ingredient in Projectivity of the Quot
Scheme (1.1.3).
Theorem A.2.16 (Existence of Flattening Stratifications). Let X → S be a
projective morphism of noetherian schemes, OX (1) be a relatively ample line bundle,
and F be a coherent sheaf on X. For each polynomial P ∈ Q[z], there exists a locally
closed subscheme SP ,→ S such that a morphism T → S factors through SP if and
only if the pullback FT of F to XT is flat over T and for every t ∈ T , the pullback
Fκ(t) to Xκ(t) has Hilbert polynomial P .
` Moreover, there exists a finite indexing set I of polynomials such that S =
P ∈I SP set-theoretically.
S The closure of SP in S is contained set-theoretically in
the union P ≤Q SQ , where P ≤ Q if and only if P (z) ≤ Q(z) for z ≫ 0.
Proof. See [FGAIV, Lem. 3.3] or [Mum66a, §8]. The central idea is to reduce to
X = S, in which case the closed loci SP are defined by fitting ideals of F [SP,
Tag 05P9].
Remark A.2.17. When X → S is only proper, there is a universal flattening, i.e.,
an algebraic space S ′ and a morphism S ′ → S such that a map T → S factors
through S ′ → S if and only if the pullback F|XT to XT := X ×S T is flat over T [SP,
Tag 05UG]. In general, S ′ may not be a disjoint union of locally closed subschemes
of S; see [Kre13].
Theorem A.2.18 (Raynaud-Gruson Flatification). Let Y be a quasi-compact and
quasi-separated scheme and X → Y be a finitely presented morphism which is flat
over a quasi-compact open subscheme U ⊆ Y . Then there is a commutative diagram

X
e /X

f
 
Y′
p
/Y

where p : Y ′ → Y is a blowup of a finitely presented closed subscheme Z ⊆ Y disjoint

from U and the strict transform X e of X is flat over Y ′ .

The strict transform X

e above is by definition the closure of (Y ′ ∖ p−1 (Z)) ×Y X
in the base change Y ×Y X.
′

Proof. See [GR71, Thm. I.5.2.2] or [SP, Tag 0815].

471
A.2.3 Faithful flatness
A module M over a ring A is faithfully flat if the functor −⊗A M : Mod(A) → Mod(A)
is faithfully exact, i.e., a sequence N ′ → N → N ′′ of A-modules is exact if and only
if N ′ ⊗A M → N ⊗A M → N ′′ ⊗A M is exact
for every nonzero map ϕ : N → N ′ of A-modules, the induced map ϕ⊗A M : N ⊗A
M → N ′ ⊗A M is also nonzero.
Proposition A.2.19 (Faithfully Flat Equivalences). Let A be a ring and M be a
flat A-module. The following are equivalent:
(1) M is faithfully flat;
(2) for every nonzero map ϕ : N → N ′ of A-modules, the induced map ϕ ⊗A
M : N ⊗A M → N ′ ⊗A M is also nonzero;
(3) for every nonzero A-module N , the tensor product N ⊗A M is nonzero;
(4) for every prime ideal p ⊆ A, the tensor product M ⊗A κ(p) is nonzero; and
(5) for every maximal ideal m ⊆ A, the tensor product M ⊗A κ(m) ∼ = M/mM is
nonzero.
Proof. See [SP, Tag 00H9].
When M = B is an A-algebra, then by (4) a flat ring map A → B is faithfully
flat if Spec B → Spec A is surjective, or equivalently by (5) every maximal ideal of
A is in the image of Spec B → Spec A. The latter equivalence implies that any flat
local ring map is faithfully flat.
A morphism f : X → Y of schemes is faithfully flat if f is flat and surjective.
This is equivalent to the condition that f ∗ : QCoh(Y ) → QCoh(X) is faithfully
exact. It is also equivalent to the condition that a quasi-coherent OY -module (resp.,
a morphism of quasi-coherent OY -modules) is zero if and only if its pullback is.
Faithfully flat morphisms play an important role in descent theory; see §2.1.

A.2.4 Fppf and fpqc morphisms

Fppf and fpqc morphisms are acronyms for ‘fidèlement plate de présentation finie’
and ‘fidèlement plat et quasi-compact,’ respectively. Despite this terminology, an
fpqc morphism is more general than a faithfully flat and quasi-compact map.
Definition A.2.20. A morphism f : X → Y of schemes is:
(1) fppf if f is faithfully flat and locally of finite presentation, and
(2) fpqc if f is faithfully flat and every quasi-compact open subset of Y is the
image of a quasi-compact open subset of X.
Remark A.2.21. A quasi-compact and faithfully flat morphism is fpqc. In addition,
an open and faithfully flat
S morphism is fpqc: for a quasi-compact open subset V ⊆ Y ,
we can write f −1 (V ) = i Ui as a union of affines, and since each f (Ui ) ⊆ V is open
and V is quasi-compact, we see that V is the image of finitely many of the Ui ’s.
Fppf Morphisms are Open (A.2.14) gives the implication: fppf ⇒ fpqc.
An fpqc morphism f : X → Y can be equivalently characterized by either
requiring that there exists an affine covering {Yi } of Y such that each Yi is the
image of quasi-compact open subset of X, or by requiring that every point x ∈ X
has an open (resp., quasi-compact open) neighborhood U such that f (U ) is open
and U → f (U ) is quasi-compact; see [Nit05, Prop. 2.33].
` An fppf (resp., fpqc) cover {Xi → X} is a collection of morphisms such that
i Xi → X is fppf (resp., fpqc).

472
A.2.5 Universally injective homomorphisms
The defining characteristic of a flat module is that it preserves every injection under
tensoring. The dual notion of an injection of modules, which is preserved under
tensoring by every module, is also a very useful property.

Definition A.2.22. A homomorphism M → N of A-modules is universally in-

jective if for every A-module P , the map M ⊗A P → N ⊗A P is injective. A ring
homomorphism A → B is universally injective if it is as a map of A-modules.

Remark A.2.23. This should not be confused with a universally injective or radiciel
morphism of schemes X → Y , i.e., an injective map that remains injective after any
base change; see [SP, Tag 01S2].
We will use this notion in a fundamental way in our proof of Coherent Tannaka
Duality (Theorem 6.6.1). To this end, the following properties will be used:

Proposition A.2.24.
(1) A faithfully flat ring homomorphism A → B is universally injective.
(2) A split injective M → N of A-modules is universally injective. The converse
is true if N/M is finitely presented.
(3) If A → A′ is faithfully flat, then a map M → N of A-modules is universally
injective if and only if M ⊗A A′ → N ⊗A A′ is.
(4) If A → B is universally injective and B → B ⊗A B, defined by b 7→ b ⊗ 1, is
faithfully flat, then A → B is faithfully flat.

Proof. For (1), (2), and (4), see [SP, Tags 08WP, 058L, and 08XD]. Part (3) follows
directly from the faithful exactness of − ⊗A A′ . See also [Laz69] or [Lam99, §4J].

Remarkably universally injective ring maps are precisely those maps that satisfy
effective descent for modules; see Remark 2.1.6.

A.3 Étale, smooth, and unramified morphisms

A morphism f : X → Y of schemes is

– smooth if f is locally of finite presentation, flat, and for every y ∈ Y , the

geometric fiber Xκ(y) = X ×Y Spec κ(y) is regular,
– étale if f is smooth of relative dimension 0, i.e., dim Xy = 0 for all y ∈ Y , and
– unramified if f is locally of finite type1 and every geometric fiber is discrete
and reduced, i.e., for all y ∈ Y , the Xy ∼ = i Spec Ki where each Ki is a
`
separable field extension of κ(y).

We say that a morphism f : X → Y of schemes is smooth (resp., étale, unramified)

at x ∈ X if there exists an open neighborhood U ⊆ X of x such that f |U : U → Y is
smooth (resp., étale, unramified). In §A.3.5, we discuss local complete intersections
and syntomic morphisms. There are the following implications:

unramified ⇐= étale ⇒ smooth ⇒ syntomic ⇒ fppf ⇒ fpqc.

1We are following the conventions of [GR71] and [SP] rather than [EGA] as we only require
that f is locally of finite type rather than locally of finite presentation.

473
A.3.1 Equivalences
Smooth, étale, and unramified morphisms have many equivalent characterizations.
These equivalences take considerable work to establish, and we recommend [Mil80,
Ch. 1] and [Liu02, §4.3] for accessible accounts.

Theorem A.3.1 (Smooth Equivalences). Let f : X → Y be a locally of finite

presentation morphism of schemes (resp., locally noetherian schemes). The following
are equivalent:
(1) f is smooth;
(2) (Differential Criterion) for every point x ∈ X, f is flat at x and the OX,x -
module ΩX/S,x can be generated by at most dimx Xf (x) elements (equivalently
is free of rank dimx Xf (x) );
(3) (Infinitesimal Lifting Criterion) for every surjection A → A0 of rings with
nilpotent kernel (resp., surjection of local artinian rings with ker(A → A0 ) ∼
=
A/mA ) and every commutative diagram

Spec A /X
_ 0 ;
f
 
Spec A /Y

of solid arrows, there exists a dotted arrow filling in the diagram;

(4) (Jacobian Criterion) for every point x ∈ X, there exists affine open neigh-
borhoods Spec B of f (x) and Spec A ⊆ f −1 (Spec B) of x and an A-algebra
isomorphism
B∼= A[x1 , . . . , xn ]/(h1 , . . . , hr )
for some h1 , . . . , hr ∈ A[x1 , . . . , xn ] with r ≤ n such that the determinant
δh
det( δxji )1≤i,j≤r ∈ B of the Jacobi matrix, defined by the partial derivatives
with respect to the first r xi ’s, is a unit. (The map Spec A → Spec B is called
a standard smooth morphism.).
If in addition X and Y are locally noetherian and x ∈ X has image y ∈ Y with
κ(x) = κ(y), then f : X → Y is smooth at x if and only if
bX,x ∼
(4) there is an isomorphism O =O bY,y [[x1 , . . . , xr ]] of O
bY,y -algebras.
If in addition X and Y are smooth over an algebraically closed field k, then f is
smooth at x ∈ X(k) if and only if
(5) the induced map TX,x → TY,y on tangent spaces is surjective.

Proof. See [Har77, Exc. II.8.6, Prop. III.10.4], [EGA, [Link], IV4 .17.5,14], and [SP,
Tags 01V9, 02H6, and 02HX].

We say that f : X → Y is smooth of relative dimension n if f is smooth and

every fiber is equidimensional of dimension n, or equivalently if f is fppf, all fibers
are equidimensional of dimension n, and ΩX/S is locally free of rank n. If f is only
fppf and ΩX/S is locally free of dimension d, it is not necessarily true that f is
smooth of relative dimension d, e.g., Spec Fp [x] → Spec Fp [xp ].

Theorem A.3.2 (Étale Equivalences). Let f : X → Y be a locally of finite presen-

tation morphism of schemes (resp., locally noetherian schemes). The following are
equivalent:

474
 (1) f is étale;
(2) f is smooth and ΩX/Y = 0;
(3) f is smooth and unramified;
(4) f is flat and unramified;
(5) (Infinitesimal Lifting Criterion) for every surjection A → A0 of rings with
nilpotent kernel (resp., surjection of local artinian rings with ker(A → A0 ) ∼
=
A/mA ) and every commutative diagram

Spec A /X
_ 0 ;
f
 
Spec A /Y

of solid arrows, there exists a unique dotted arrow filling in the diagram; and
(6) (Jacobi Criterion) for every point x ∈ X, there exists affine open neighborhoods
Spec B of f (x) and Spec A ⊂ f −1 (Spec B) of x and an A-algebra isomorphism

B∼
= A[x1 , . . . , xn ]/(h1 , . . . , hn )
δh
for some h1 , . . . , hn ∈ A[x1 , . . . , xn ] such that the determinant det( δxji )1≤i,j≤n ∈
B is a unit. (The map Spec A → Spec B is called a standard étale morphism.)
If in addition X and Y are locally noetherian and x ∈ X has image y ∈ Y with
κ(x) = κ(y), then f : X → Y is smooth at x if and only if
(6) ObX,x ∼
=O bY,y .
If in addition X and Y are smooth over an algebraically closed field k, then f is
étale at x ∈ X(k) if and only if
(7) the induced map TX,x → TY,y on tangent spaces is an isomorphism.

Proof. See [Har77, Exc. III.10.3], [EGA, IV4 .17.14.1-2, IV4 .17.6.3], and [SP, Tags 02GH
and 02HF].

Theorem A.3.3 (Unramified Equivalences). Let f : X → Y be morphism of schemes

locally of finite type. The following are equivalent:
(1) f is unramified;
(2) ΩX/Y = 0;
(3) the diagonal ∆f : X → X ×Y X is an open immersion;
(4) (Infinitesimal Lifting Criterion for Unramifiedness) for every surjection A → A0
of rings with nilpotent kernel (resp., surjection of local artinian rings with
ker(A → A0 ) ∼ = A/mA ) and every commutative diagram

Spec A /X
_ 0 ;
f
 
Spec A /Y

of solid arrows, there exists at most one dotted arrow filling in the diagram.
If in addition X and Y are locally noetherian and x ∈ X has image y ∈ Y with
κ(x) = κ(y), then f : X → Y is smooth at x if and only if

475
 (4) O
bY,y → O
bX,x is surjective.

Proof. See [EGA, IV4 .17.14.1-2, IV4 .17.6.3], and [SP, Tags 02G3, 02H7, and 02GE].

A.3.2 Étale-local structure of smooth, étale, and unramified

morphisms
Every smooth morphism is étale-locally relative affine space.

Proposition A.3.4 (Local Structure of Smooth Morphisms). A morphism X → Y

of schemes is smooth at x ∈ X if and only if there exists affine open subschemes
Spec A ⊆ X and Spec B ⊆ Y with x ∈ Spec A, and a commutative diagram

Xo
op
? _ Spec A ét / An
B

  {
Y o
op
? _ Spec B

where Spec A → AnB is étale.

Proof. See [SP, Tag 039P] and [EGA, IV4 .17.11.4].

An important consequence is that smooth morphisms have sections étale locally.

Corollary A.3.5. Let f : X → Y be a morphism of schemes smooth at x ∈ X.

Then there exists an étale neighborhood Y ′ → Y of f (x) such that X ×Y Y ′ → Y ′
has a section.

Proof. After applying the proposition, observe that the morphism AnB → Spec B
admits the zero section Spec B → AnB . The scheme Y ′ := Spec B ×AnB Spec A is étale
over Y , and the composition Y ′ → Spec A ,→ X defines a section Y ′ → X ×Y Y ′ of
X ×Y Y ′ → Y ′ .

Every étale (resp., unramified) morphism is étale-locally an isomorphism (resp.,

closed immersion).

Proposition A.3.6. Let f : X → S be a separated morphism of schemes étale (resp.,

unramified) at x ∈ X. Then there exists an étale neighborhood (U, u) → (S, f (x))
and a finite disjoint union decomposition
a
XU = W ⨿ Vi
i

such that each Vi → U is an isomorphism (resp., closed immersion) and the fiber
Wu contains no point over x.

Proof. See [SP, Tags 04HM and 04HG].

It is sometimes convenient to know that étale and smooth morphisms of affine

schemes can be lifted along closed immersions. It is also holds for syntomic morphisms
(see Definition A.3.17).

476
Proposition A.3.7. Consider a diagram

Spec A0  / Spec A

□
  
Spec B0  / Spec B

of solid arrows where Spec B ,→ Spec B0 is a closed immersion. If Spec A0 →

Spec B0 is étale (resp., smooth, syntomic), then there exists an étale (resp., smooth,
syntomic) morphism Spec A → Spec B making the above diagram cartesian.

Proof. See [SP, Tags 04D1 and 07M8].

A.3.3 Further properties

Proposition A.3.8 (Fibral Étaleness/Smoothness/Unramifiedness Criteria). Con-
sider a diagram

X /Y

 
S
of schemes where X → S and Y → S are locally of finite presentation. Let x ∈ X
with image s ∈ S. Then
(1) X → Y is unramified at x if and only if Xs → Ys is unramified at x, and
(2) if X → S is flat at x, then X → Y is étale (resp., smooth) at x if and only if
Xs → Ys is étale (resp., smooth) at x.

Proof. Let y ∈ Y be the image of x ∈ X. Part (1) holds since unramified is defined
as a condition on the fiber and the fiber Xy is identified with the fiber of Xs → Ys
over y ∈ Ys . For the non-trivial direction (⇐) of (2), the Fibral Flatness Criterion
(A.2.10) implies that X → Y is flat at x. Therefore, the smoothness (resp., étaleness)
of X → Y at x is equivalent to the smoothness (resp., étaleness) of Xy → Spec κ(y)
at x. The latter condition holds by the smoothness (resp., étaleness) of Xs → Ys at
x.

While smoothness is clearly an open condition on the source, it is also an open

condition on the target when the morphism is proper.

Corollary A.3.9. If f : X → Y is a proper, flat, and locally of finite presentation

morphism, then the set of points y ∈ Y where Xy → Spec κ(y) is smooth is open.

Proof. If y ∈ Y is a point such that Xy → Spec κ(y) is smooth, then f : X → Y

is smooth in an open neighborhood of Xy . If Z ⊂ X is the closed locus where
f : X → Y is not smooth, then f (Z) ⊆ Y is precisely the locus where the fibers of f
are not smooth. Since f is proper, f (Z) is closed.

Proposition A.3.10. Let X → Y be a smooth morphism of noetherian schemes.

For every point x ∈ X with image y ∈ Y ,

dimx (X) = dimy (Y ) + dimx (Xy ).

Proof. See [SP, Tag 0AFF].

477
Proposition A.3.11. If X → Y is a finite étale morphism, there exists a finite
étale cover Y ′ →`Y such that X ×Y Y ′ → Y ′ is a trivial covering, i.e., X ×Y Y ′ is
isomorphism to i Y ′ over Y ′ .
Proof. We may assume that the degree d of X → Y is constant. The scheme
(X/Y )d = X ×Y · · · ×Y X
| {z }
d

represents the functor Sch/Y → Sets assigning a Y -scheme T to the set of d sections
of X ×Y T → T . Each pairwise diagonal (X/Y )d−1 → (X/Y )d is an open and closed
immersion, and we set (X/Y )d0 ⊆ (X/Y )d to be the complement of all pairwise
diagonals. The projection morphism (X/Y )d0 → Y is finite étale and the functorial
description gives d disjoint sections of X ×Y (X/Y )d0 → (X/Y )d0 .
Proposition A.3.12. A dominant unramified morphism X → Y of schemes with
Y normal and X connected is étale.
Proof. See [SGA1, Cor. I.9.11].
The following result is often called ‘Nagata–Zariski Purity’.
Proposition A.3.13 (Purity of the Branch Locus). Let f : X → Y be a quasi-finite
morphism of integral noetherian schemes such that X is normal and Y is regular.
Then the locus of points in X where f is not étale is either empty or codimension 1.
Proof. See [Zar58], [Nag59], [SGA1, Thm. X.3.1], and [SP, Tag 0BMB].

A.3.4 Fitting ideals and the singular locus

Fitting ideals allows for a scheme-theoretic description of the singular locus of
a scheme. We use fitting ideals in the Characterization of Nodes (5.2.4). For
background references on fitting ideals, we recommend [SP, Tag 07Z6] and [Eis95,
§20].
If R is a ring and M is a finitely generated R-module, the kth fitting ideal Fitk (M )
of M is the ideal generated by the (n − k) × (n − k) minors of a matrix A defining a
presentation M A
R− → Rn → M → 0.
i∈I
Of course, when M is finitely presented (e.g., R is noetherian), then the left-hand
term can be assumed to be a finite free module Rm , in which case A is an m × n
matrix and Fitk (M ) is a finitely generated ideal. The fitting ideal is independent of
the choice of presentation, and defines an increasing sequence of ideals
0 = Fit−1 (M ) ⊆ Fit0 (M ) ⊆ Fit1 (M ) ⊆ · · · R
such that Fitk (M ) = R if M can be generated by k elements. The R-module M
is locally free of rank r if and only if Fitr−1 (M ) = 0 and Fitr (M ) = R, and in
this case Fitk (M ) = 0 for all k < r. There is an identification Fitk (M ⊗R S) =
Fitk (M )S for a ring map R → S. In particular, Fitk (Mf ) = Fitk (M )f for f ∈ R,
Fitk (Mp ) = Fitk (M )p for a prime ideal p ⊆ R, and Fitk (M ) ⊗R R c) if R
b = Fitk (M
is a noetherian local ring.
If X is a scheme and F is a finite type quasi-coherent sheaf on X, the kth
fitting ideal sheaf of F is the quasi-coherent sheaf of ideals Fitk (F ) ⊆ OX defined by
Γ(U, Fitk (F )) = Fitk (Γ(F, U )) for affine open subsets U ⊆ X. Fitting ideal sheaves
give a scheme structure to the singular locus.

478
Definition A.3.14. If X is a noetherian scheme of pure dimension d over a field
k , we define the singular locus of X as the subscheme Sing(X) := V (Fitd (ΩX/k ))
defined by the dth fitting ideal of of ΩX/k . More generally, if X → S is an fppf
morphism such that every fiber has pure dimension d, we define the relative singular
locus as the subscheme Sing(X/S) := V (Fitd (ΩX/S )).

For example, if X = Spec k[x1 , . . . , xn ]/I with I = (f1 , . . . , fm ), the exact

sequence I/I 2 → ΩAn /k |X → ΩX/k → 0 induces a resolution
 
m J ∂fj
OX −
→ n
OX → ΩX/k → 0 with J = ,
∂xi

and Sing(X) is defined by all (n − d) × (n − d) minors of J.

A.3.5 Local complete intersections and syntomic morphisms

Definition A.3.15. A scheme X locally of finite type over a field k is a local
complete intersection at p ∈ X (or lci at p) if there exists an affine open neighborhood
p ∈ Spec A ⊆ X such that A is a global complete intersection over k, i.e., A ∼ =
k[x1 , . . . , xn ]/(f1 , . . . , fc ) with dim A = n − c. The scheme X is a local complete
intersection if it is at every point.

Proposition A.3.16. For a scheme X locally of finite type over a field k and a
point p ∈ X, the following are equivalent:
(1) X is a local complete intersection at p,
(2) the local ring OX,x ∼ = R/(f1 , . . . , fc ) where R is a regular local ring and
f1 , . . . , fc ∈ R is a regular sequence, and
(3) the completion O bX,x ∼= R/(f1 , . . . , fc ) where R is a regular complete local ring
and f1 , . . . , fc ∈ R is a regular sequence.

Proof. See [SP, Tags 00S8 and 09PY].

For a scheme locally of finite type over a field k, there are implications:

smooth ⇒ local complete intersection ⇒ Gorenstein ⇒ Cohen–Macaulay.

Here is the relative notion:

Definition A.3.17. A morphism of schemes f : X → S is syntomic (or a flat local

complete intersection morphism) if f is fppf and every fiber is a local complete
intersection. We say that f : X → S is syntomic at x ∈ X if there is an open
neighborhood U of x such that f |U is syntomic.

Syntomic morphisms have a local structure analogous to local complete intersec-

tions.

Proposition A.3.18. A morphism f : X → S is syntomic at x ∈ X if and only

if there are affine open neighborhood x ∈ Spec A ⊆ X and Spec B ⊆ Y with
f (Spec A) ⊆ Spec B such that A ∼
= B[x1 , . . . , xn ]/(f1 , . . . , fc ) and every nonempty
fiber of Spec A → Spec B has dimension n − c.

Proof. See [EGA, IV4 §19.3], [SGA6, VIII §1] and [SP, Tag 01UB].

479
A.4 Properness and the Valuative Criterion
Each day learn something new, and just as important, relearn something
old
Robert Brault

Properness, separatedness, and universal closedness can be verified using the

Valuative Criteria (3.8.7). While the importance of valuative criteria may not be
apparent after a first course in algebraic geometry, they becomes indispensable in
moduli theory, as it provides a geometric strategy to verify separated, properness,
and universal closedness. In this text, we apply the Valuative Criteria to show that
Mg is proper (Theorem 5.5.23) and Bun ss r,d (C) is universally closed.

A.4.1 The Valuative Criteria

As we generalize the criteria to algebraic stacks in Theorem 3.8.7, we quickly recap
how the Valuative Criteria (A.4.5) are established for schemes. The starting point
of the proof of is the following lifting criterion for quasi-compact morphisms to be
closed.
Lemma A.4.1. A quasi-compact morphism f : X → Y of schemes is closed if
and only if for every point x ∈ X, every specialization f (x) ⇝ y0 in Y lifts to a
specialization x ⇝ x0 in X:

X x /o /o /o / x0
_ _
f
 
Y f (x) / y0 .

Proof. The implication (⇒) is clear as f ({x}) ⊆ Y is closed. For the converse, after
replacing X with a closed subscheme, it suffices to show that f (X) is closed. We
can assume that X = Spec A and Y = Spec B are affine (since f is quasi-compact)
and reduced (since the question is topological). The scheme-theoretic image of
Spec A → Spec B is defined by I := ker(B → A). By replacing B with B/I, we
can assume that B → A is injective. For every minimal prime p ∈ Spec B, the
localization Bp is a field and the map Bp → Ap is injective. Thus, Ap ̸= 0 and the
fiber f −1 (p) = Spec Ap is non-empty. Since f (X) contains all the minimal primes
and is closed under specialization, f (X) = Y is closed.
The noetherian valuative criterion depends on the following algebraic fact:
Proposition A.4.2. Let (A, mA ) be a noetherian local domain with fraction field K
such that A is not a field. If K → L is a finitely generated field extension, then there
exists a DVR R with fraction field L dominating A (i.e., A ⊆ R and mA ∩ K = mR ).
Proof. We reduce to the case that K → L is a finite field extension by choosing
a transcendence basis x1 , . . . , xn ∈ L over K and replacing A with A[x1 , . . . , xn ]n
where n = mA A[x1 , . . . , xn ] + (x1 , . . . , xn ). Let B be the blow of Spec A at mA and
let E ⊆ B be the exceptional divisor. If ξ ∈ E is a generic point, then OB,ξ is a
noetherian domain of dimension 1 (by Krull’s Hauptidealsatz) with fraction field K.
We now let R ⊆ L be the integral closure of OB,ξ in L. By Krull–Akizuki (A.4.3),
R is noetherian. Since R is also normal of dimension 1, it is a DVR.

480
Proposition A.4.3 (Krull–Akizuki). Let R be a noetherian domain of dimension
1 with fraction field K. If K → L is a finite extension of fields, then every ring A
with R ⊆ A ⊆ L is noetherian.
Proof. See [Nag62, p. 115] or [SP, Tag 00PG].
Proposition A.4.2 and Krull–Akizuki have the following geometric implication.
Proposition A.4.4. If f : X → Y is a finite type morphism of noetherian schemes,
x ∈ X, and f (x) ⇝ y0 is a specialization, there exists a commutative diagram

Spec K /X x
_
f
  
Spec R /Y f (x) / y0 .

where R is a DVR with fraction field K, the image of Spec K → X is x and

Spec R → Y realizes the specialization f (x) ⇝ y0 . In particular, every specialization
x ⇝ x0 in a noetherian scheme is realized by a map Spec R → X from a DVR.
Proof. After replacing X with {f (x)} and Y with {x}, we may assume that X
and Y are integral with generic points x and f (x). Then OY,y0 is a noetherian
local domain with fraction field κ(f (x)). By applying Proposition A.4.2 to the field
extension κ(f (x)) → κ(x), we obtain a DVR R with fraction field κ(x) dominating
OY,y0 , yielding the desired diagram.
We only state a noetherian version of the Valuative Criteria.
Theorem A.4.5 (Valuative Criteria for Proper/Separated/Universally Closed Mor-
phisms). Let f : X → Y be a quasi-compact morphism of noetherian schemes.
Consider a commutative diagram

Spec K /X
;
f (A.4.6)
 
Spec R /Y

of solid arrows where R is a DVR with fraction field K. Then

(1) f is proper if and only if f is of finite type and for every diagram (A.4.6),
there exists a unique lift,
(2) f is separated if and only if for every diagram (A.4.6), any two lifts are equal,
and
(3) f is universally closed if and only if for every diagram (A.4.6), there exists a
lift.
Proof. We first claim that it suffices to handle the universally closed case. Indeed, a
morphism X → Y is separated if and only if the diagonal X → X ×Y X is universally
closed, and the equality of two lifts in the valuative criterion for X → Y corresponds
to the existence of a lift in the valuative criterion for X → X ×Y X.
Suppose that X → Y satisfies the valuative criterion for universal closedness. To
show that X → Y is universally closed, we claim that it suffices to check that the base
change XT → T is closed for every finite type morphism T → Y . Indeed, suppose
that fT : XT → T is not closed for some map T → Y . By Lemma A.4.1, there exists

481
x ∈ XT and a specialization fT (x) ⇝ t0 which doesn’t lift to a specialization x ⇝ x0 .
This implies that Z = {x} ⊆ XT has trivial intersection with the fiber (XT )t0 .
Applying Lemma A.4.8 shows that, after replacing T with an open neighborhood of
t0 , there is a commutative diagram

x XT / XT ′ /X
_
fT fT ′ f
   
fT (x) / t0 T
g
/ T′ /Y

where T ′ → Y is finite type and a closed subscheme Z ′ ⊆ XT ′ such that fT ′ (Z ′ )

contains g(fT (x)) but not g(t0 ). This shows that fT ′ : XT ′ → T ′ is not closed.
Since the valuative criterion holds for X → Y , it also holds for the morphism
XT → T of noetherian schemes. It therefore suffices to show that X → Y is closed.
By Lemma A.4.1, it suffices to show that given x ∈ X, every specialization f (x) ⇝ y0
lifts to a specialization x ⇝ x0 . By Proposition A.4.4, there exists a diagram (A.4.6)
such that Spec R → Y realizes f (x) ⇝ y0 with a lift Spec K → X whose image is
x. The valuative criterion implies the existence of a lift Spec R → X, which in turn
yields a specialization x ⇝ x0 lifting f (x) ⇝ y0 .
Conversely, assume that f : X → Y is universally closed and that we are given a
diagram (A.4.6). By replacing Y with Spec R and X with X ×Y Spec R, we may
assume that Y = Spec R and that we have a diagram

Spec K
x /X

 {
Spec R

By replacing X with {x}, we may assume that X is integral with generic point
x. Since X → Spec R is closed, there exists a specialization x ⇝ x0 in X over
the specialization of the generic point to the closed point in Spec R. This gives an
inclusion of local rings R ,→ OX,x0 in K. Since R is a valuation ring with fraction
field K (i.e., is maximal among local rings properly contained in K), we see that
R = OX,x0 and the inclusion Spec OX,x0 → X gives the desired lift.
See also [Har77, Thm. 4.7, Exc. II.4.11], [EGA, §II.7], and [SP, Tags 0BX4 and
0CM1]
Remark A.4.7. The quasi-compactness of f (resp., ∆f , ∆∆f ) are essential in the
valuative criterion of universal closedness (resp., separatedness, separated diagonal).
In fact, universally closed morphisms are necessarily quasi-compact [SP, Tag 04XU].
The lemma below was used in the proof and is also used in the proof of the
Valuative Criteria (3.8.7) for algebraic stacks.
Lemma A.4.8. Let f : X → Y be a quasi-compact morphism of schemes. Let T → Y
be a morphism of schemes, t0 ∈ T be a point, and Z ⊆ XT a closed subscheme
such that Z ∩ (XT )t0 = ∅. Then after replacing T with an open neighborhood of
t0 , there exists a finite type morphism T ′ → Y of schemes with a factorization
g
T − → T ′ → Y and a closed subscheme Z ′ ⊆ XT ′ such that Z ′ ∩ (XT ′ )g(t0 ) = ∅ and
im(Z ,→ XT → XT ′ ) ⊆ Z ′ .
Proof. After reducing to the affine case T = Spec B → Y = Spec A, we write B
as a colimit of finite type A-algebras Bλ . Using techniques analogous to Limits of

482
Schemes (§B.3), one shows that for λ ≫ 0 there exists a subscheme Zλ ⊆ XBλ with
the desired properties. The details are not hard, but also not inspiring. See [SP,
Tag 05BD].

A.4.2 Universally submersive morphisms

A morphism f : X → Y of schemes is submersive if f is surjective and Y has the
quotient topology, i.e., a subset U ⊆ Y is open if and only if f −1 (U ) is open, and
f : X → Y is universally submersive if for every map Y ′ → Y , the base change
X ×Y Y ′ → Y ′ is submersive.

Exercise A.4.9.
(1) Show that a morphism f : X → Y of noetherian schemes is universally sub-
mersive if and only if every map Spec R → Y from a DVR has a lift

Spec R′ /X

f
 
Spec R / Y,

where R → R′ is a local homomorphism of DVRs.

(2) Show that universally closed morphism of noetherian schemes is universally
submersive.
(3) Show that every fpqc morphism of schemes is universally submersive.

A.5 Dévissage and finiteness of cohomology

MacPherson told me that my theorem can be viewed as blah blah blah
Grothendieck blah blah blah, which makes it much more respectable.

Jim Propp

Dévissage, or ‘unscrewing’ in French, is a specific type of noetherian induction

designed to verify properties of coherent sheaves. We apply it to extend the theorem
of Finiteness of Cohomology (A.5.3) from projective morphisms to proper morphisms.

A.5.1 Dévissage
Proposition A.5.1 (Dévissage). Let X be a noetherian scheme. Let P be a property
of coherent sheaves on X satisfying
(a) if 0 → F ′ → F → F ′′ → 0 is a short exact sequence of coherent sheaves on X
and two out of the three satisfy P, then the third satisfies P, and
(b) for every integral closed subscheme Z ⊆ X with generic point ξ, there exists
a coherent sheaf G satisfying P with mξ Gξ = 0 and dimκ(ξ) Gξ = 1 (so that
Supp(G) = Z).
Then every coherent sheaf on X satisfies P.

Proof. We say that the property PY holds for a closed subset Y ⊆ X if every
coherent sheaf F on X with Supp(F ) ⊆ Y satisfies P. We prove the proposition
by noetherian induction: we need to show that if PY ′ holds for every closed subset

483
Y ′ ⊊ Y , then PY also holds. Let F be a coherent sheaf on X with Supp(F ) ⊆ Y .
Giving Y ⊆ X the reduced scheme structure, let I ⊆ OX be its ideal sheaf. To
show that F satisfies P, we first claim that it suffices to assume that IF = 0. Since
I n F = 0 for some n > 0, we have a filtration 0 = I n F ⊆ I n−1 F ⊆ · · · ⊆ F and short
exact sequences

0 → I k−1 F/I k F → F/I k F → F/I k−1 F → 0.

By induction and property (a), it suffices to show that P holds for I k−1 F/I k F ,
which is annihilated by I. Second, we claim that we can assume that Y is irreducible,
hence integral. Supposing that Y = Y1 ∪ Y2 with F1 = F |Y1 and F2 = F |Y2 , the
map ϕ : F → F1 ⊕ F2 has kernel and cokernel supported on Y1 ∩ Y2 . Applying
property (a) to the exact sequence 0 → im ϕ → F1 ⊕ F2 → coker ϕ → 0 and then
to 0 → ker ϕ → F → im ϕ → 0, shows that P also holds for F . Finally, letting
ξ ∈ Y be the generic point, property (b) gives a coherent sheaf G satisfying P with
Supp(G) ⊆ Y and dimκ(ξ) Gξ = 1. Setting d = dimκ(ξ) Fξ , since Fξ and G⊕d ξ are
isomorphic, there is an open subscheme U ⊆ Y and an isomorphism F |U → G⊕d |U .
Let H be the image of the graph F |U → F |U ⊕ G⊕d |U , and H e ⊆ F ⊕ G⊕d be
a subsheaf on Y extending H [Har77, Exc. II.5.15]. The projections H → G⊕d
and H → F induce isomorphisms over U and hence have kernels and cokernels
supported on a closed subscheme Y ′ ⊊ Y . Since P holds for G, P also holds for
G⊕d by property (a). It follows that P holds for H and thus also F . See also [EGA,
III1 .3.1.2] and [SP, Tag 01YI].

Remark A.5.2. The same proof also establishes some useful variants. First, if we
assumed that F ⊕n ∈ P implies that F ∈ P, then condition (b) can be weakened
to exhibiting a coherent sheaf G satisfying P with Supp(G) ⊆ Z and Gξ ̸= 0.
Alternatively, (b) can be replaced with the condition that for every integral closed
subscheme Z ⊆ X with ideal sheaf IZ and every coherent sheaf F on X with
IZ F = 0, there exists a coherent sheaf G on X with IZ G = 0 satisfying P and a
morphism F → G which is an isomorphism on a non-empty open subset of Z.

A.5.2 Dévissage
Theorem A.5.3 (Finiteness of Cohomology). Let f : X → Y be a proper morphism
of noetherian schemes. For any coherent sheaf F on X and any i ≥ 0, Ri f∗ F is
coherent.

Proof. By Flat Base Change (A.2.12), we can assume that Y = Spec A is the
spectrum of a noetherian ring. We need to show that for every coherent sheaf
F on X and any i ≥ 0, Hi (X, F ) is a finite A-module. When X → Spec A is
projective, this is [Har77, Thm. III.5.2]—this is a really nice argument exhibiting
the power of cohomology, even if you are only interested in the H0 case. In the
proof, one quickly reduces to the case that X = PnA . Choosing an exact sequence
0 → K → OPnA (−m)⊕d → F → 0, the statement holds for the middle term by a
Céch cohomology computation, and the vanishing of cohomology in sufficiently high
degree and a descending induction argument shows that it holds for F .
To apply Dévissage (A.5.1), we let P be the property of a coherent sheaf F on
X that Hi (X, F ) is a finite A-module for all i. This satisfies the two-out-of-three
condition (a) of (A.5.1). To see that (b) holds, let Z ⊆ X be an integral closed
subscheme with generic point ξ. By Chow’s Lemma [Har77, Exc. II.4.10], there
exists a projective birational morphism g : Z ′ → Z such that Z ′ → Z → Y is also

484
projective. The ample sheaf OZ ′ (1) is relatively ample over Z and thus for d ≫ 0 and
i > 0, Ri g∗ (OZ ′ (d)) = 0 and Hi (Z ′ , OZ ′ (d)) = 0. Taking G := g∗ OZ ′ (d) for d ≫ 0,
dimκ(ξ) Gξ = 1. Using the vanishing of Ri g∗ (OZ ′ (d)), the Leray spectral sequence
Hp (Z, Rq g∗ OZ ′ (d)) ⇒ Hp+q (Z ′ , OZ ′ (d)) implies that Hi (Z, G) = Hi (Z ′ , OZ ′ (d)) is
a finite A-module (and in fact 0 for i > 0). See also [EGA, III1 .3.2.1] and [SP,
Tag 02O5].
This argument can also be formulated using derived categories. Chow’s Lemma
gives a projective birational morphism g : X ′ → X with X ′ projective over Y =
Spec A, and consider the exact triangle F → Rg∗ g ∗ F → C. Since g is projective,
Rg∗ g ∗ F ∈ DCoh
b
(X) and thus C ∈ DCoh b
(X). Since g is birational F → Rg∗ g ∗ F is
an isomorphism over a dense open. Using the exact triangles C n → C → τ>n C
arising from truncation and the fact that C n = 0 for n ≪ 0, an induction argument
shows that RΓ(X, C) ∈ DCoh b
(Y ). Since RΓ(X, g∗ g ∗ F ) ∈ DCoh b
(Y ), we conclude
that RΓ(X, F ) ∈ DCoh (Y ). Formalizing this argument leads to version of dévissage
b

for derived categories (e.g., [LMB00, Lem. 15.7]).

The following version of Formal Functions is often applied over a complete local
ring (A, m), but the non-local case is sometimes useful.

Theorem A.5.4 (Formal Functions). Let X be a scheme proper over a noetherian

ring A which is complete with respect to an ideal I ⊆ A. Let Xn = X ×A A/I n+1 .
If F is a coherent sheaf on X, there is a natural isomorphism
∼
Hi (X, F ) → lim Hi (Xn , F |Xn )
← −
n

for every i ≥ 0.

Proof. See [Har77, Thm. III.11.1] (projective over complete local), [Vak17, Thm. 30.8.1],
[Ill05, Cor. 8.2.4], [EGA, III1 .4.1.7], and [SP, Tag 02OC].

Exercise A.5.5. Show more generally that for coherent sheaves F and G on X,
there is a natural isomorphism
∼
ExtiOX (F, G) → lim ExtiOXn (F |Xn , G|Xn )
←−
n

for every i ≥ 0.

Exercise A.5.6. Use dévissage to reduce the proper case of Formal Functions
(A.5.4) to the projective case.
Hint: Reduce to showing that there is a universal d such that Ri g∗ OZn′ (d) = 0
for all n, i ≥ 0, where Z ′ is as in the proof of Finiteness of Cohomology (A.5.3)
and Zn′ = Z ′ ×A A/mn+1 . To show this, apply Serre’s Vanishing Theorem [Har77,
Thm. II.5.2] to the projective morphism SpecZ ′ i≥0 mi OZ ′ → Spec i≥0 mi .
L L

A.6 Cohomology and Base Change

There is hardly any theory which is more elementary [than linear algebra],
in spite of the fact that generations of professors and textbook writers
have obscured its simplicity by preposterous calculations with matrices.

Jean Dieudonné

485
 If f : X → Y is a proper morphism of noetherian schemes and F is a coherent
sheaf on X, then Finiteness of Cohomology (A.5.3) implies that Ri f∗ F is coherent.
We often want to know more:
(a) When is Ri f∗ F a vector bundle on Y ?
(b) When does the construction of Ri f∗ F commute with base change, i.e., for a
map g : Y ′ → Y of schemes inducing a cartesian diagram
g′
XY ′ /X

f′ f
 
Y′
g
/ Y,

when is the comparison map

ϕiY ′ : g ∗ Ri f∗ F → Ri f∗′ g ′∗ F (A.6.1)
an isomorphism?
When f : X → Y is flat, Flat Base Change (A.2.12) tells us that (A.6.1) is always
an isomorphism. Cohomology and Base Change (A.6.8) provides an answer when F
is flat over Y .
Cohomology and Base Change is an essential tool in moduli theory. It can
be applied to verify properties of families of objects and construct vector bundles
on moduli spaces. For instance, for a family π : C → S of smooth curves, we can
verify that π∗ (Ω⊗k
C/S ) is a vector bundle for k > 0 whose construction commutes
with base change on S, and show that C embeds canonically into P(π∗ (Ω⊗k C/S )) for
k ≥ 3 (Proposition 5.1.16). We apply Cohomology and Base Change in our study
of Mg . It is used to verify that it is an algebraic stack (Theorem 3.1.17) and to
establish various geometric properties. Applied to the universal family π : Ug → Mg ,
Cohomology and Base Change shows that π∗ (ΩUg /Mg ) is a vector bundle of rank g
on Mg , called the Hodge bundle (Example 4.1.4).

A.6.1 Formulations of Cohomology and Base Change

We begin with the key algebraic version of Cohomology and Base Change, which is
used to establish the other versions.
Theorem A.6.2 (Cohomology and Base Change I). Let X → Spec A be a proper
morphism of noetherian schemes and F be a coherent sheaf on X which is flat over
A. There is a complex
K• : 0 → K0 → K1 → · · · → Kn → 0
of finite and locally free A-modules such that Hi (X, F ) = Hi (K • ) for all i. Moreover,
for every A-module M , Hi (X, F ⊗A M ) = H i (K • ⊗A M ). In particular, for a
morphism Spec B → Spec A of schemes, Hi (XB , FB ) = Hi (K • ⊗A B) where XB :=
X ×Spec A Spec B and FB is the pullback of F to XB .
Proof. This is established by choosing a finite affine cover {Ui } of X and considering
the corresponding alternating Čech complex C • on {Ui } with coefficients in F .
Then C • is a finite complex of flat (but not finitely generated) A-modules and
Hi (X, F ) = Hi (C • ). The result is then obtained by inductively refining C • to build
a finite complex K • of finite and flat A-modules which is quasi-isomorphic to C • .
See [Mum70a, Thm. p.46], [SP, Tag 07VK], and [Vak17, 28.2.1].

486
Remark A.6.3 (Perfect complexes). A bounded complex K • of S coherent sheaves on
a noetherian scheme X is perfect if there is an affine cover X = i Ui such that each
K • |Ui is quasi-isomorphic to a bounded complex of vector bundles on Ui . If X is
affine (or more generally has the resolution property, i.e., every coherent sheaf is the
quotient of a vector bundle), then K • is perfect if and only if it is quasi-isomorphic to
a bounded complex of vector bundles on X [SP, Tag 0F8F]. Moreover, the compact
objects in DQCoh (X) are precisely the perfect complexes [SP, Tag 09M8].
With this terminology in place, Theorem A.6.2 has the following translation:
Rf∗ F ∈ DCohb
(Spec A) is perfect. More generally, if F • is a perfect complex on X,
then Rf∗ F is also perfect [SP, Tag 0A1H].
•

Theorem A.6.2 tells us that the cohomology Hi (X, F ) can be computed as the
cohomology of a bounded complex K • of vector bundles on Spec A, and thus reduces
cohomological questions to linear algebra.
Theorem A.6.4 (Semicontinuity Theorem). Let X → Y be a proper morphism of
noetherian schemes and F be a coherent sheaf on X which is flat over Y .
(1) For each i ≥ 0, the function

Y → Z, y 7→ hi (Xy , Fy )

is upper semicontinuous.
(2) The function
∞
X
Y → Z, y 7→ χ(Xy , Fy ) = (−1)i hi (Xy , Fy )
i=0

is locally constant.
Proof. We may assume that Y = Spec A so that Theorem A.6.2 applies: there is
di−1 di
bounded complex K • : · · · → K i−1 −−−→ K i −→ K i+1 → · · · of finite and locally
free A-modules such that Hi (Xy , Fy ) = H i (K • ⊗A κ(y)) for all y ∈ Y . Using that
im(di ⊗ κ(y)) = (K i ⊗A κ(y))/ ker(di ⊗ κ(y)), we have

hi (Xy , Fy ) = dimκ(y) ker(di ⊗ κ(y)) − dimκ(y) im(di−1 ⊗ κ(y))

= dimκ(y) K i ⊗ κ(y) − dimκ(y) im(di ⊗ κ(y)) − dimκ(y) im(di−1 ⊗ κ(y)).
(A.6.5)
The statement follows as both dimκ(y) im(di ⊗ κ(y)) and dimκ(y) im(di ⊗ κ(y)) are
lower semicontinuous. See also [Mum70a, p. 47], [Har77, Thm. 12.8], or [Vak17,
Thm. 25.1.1].
To show more powerful results, we will need more sophisticated linear algebra.
We follow an argument of Eric Larson, as also described in [Vak17, §25.2].
Proposition A.6.6. Let X be a scheme and ϕ : E → F be a map of vector bundles
on X of rank e and f . For every point x ∈ X and an integer r ≤ min(e, f ), the
following are equivalent:
(1) coker(ϕ) is a vector bundle of rank f − r in an open neighborhood of x,
(2) there is an open neighborhood U of x and identifications E|U ∼ = OU ⊕e
and
∼
F |U = O ⊕f
such that ϕ|U corresponds to the composition of the projection
U
⊕e ⊕r ⊕r ⊕f
OU ↠ OU and inclusion OU ,→ OU of the first r summands, and
(3) ker(ϕ) ⊗ κ(x) → ker(ϕ ⊗ κ(x)) is surjective.

487
If in addition X is reduced, the above are equivalent to
(4) there is an open neighborhood U of x such that ϕ ⊗ κ(u) has rank r for all
u ∈ U.
If these hold, then coker(ϕ), ker(ϕ), and im(ϕ) are vector bundles, and their con-
struction commutes with base change.

Proof. This is an exercise in linear algebra.

If the equivalent conditions above hold, we say that ϕ : E → F is a map of vector

bundles (of rank r). We can now provide a quick proof of Grauert’s Theorem.

Theorem A.6.7 (Grauert’s Theorem). Let f : X → Y be a proper morphism of

noetherian schemes such that Y is reduced and connected. Let F be a coherent sheaf
on X flat over Y . For each integer i, the following are equivalent:
(1) the function y 7→ hi (Xy , Fy ) is constant; and
(2) Ri f∗ F is a vector bundle and the comparison map

ϕiy : Ri f∗ F ⊗ κ(y) → Hi (Xy , Fy )

is an isomorphism for all y ∈ Y .

If these hold, then the construction of Ri f∗ F commutes with base change by an
arbitrary map T → Y .

Proof. The direction (1) ⇒ (2) is clear. For the converse, we can reduce to the case
that Y = Spec A as the question is Zariski local. Let

di−1 di
K • : · · · → K i−1 −−−→ K i −→ K i+1 → · · ·

be the complex of vector bundles on Y produced by Theorem A.6.2. As y 7→

hi (Xy , Fy ) is constant, the identity (A.6.5) implies that y 7→ dimκ(y) im(di−1 ⊗ κ(y))
and y 7→ dimκ(y) im(di ⊗ κ(y)) are also constant. As Y is reduced, Proposition A.6.6
implies that di−1 and di are maps of vector bundles, that im(di ) and coker(di−1 )
are vector bundles, and that ker(di ) commutes with base change. The cohomology
Hi (K • ) = im(di−1 )/ ker(di ) sits in a short exact sequence

0 → Hi (K • ) → K i / ker(di ) → coker(di−1 ) → 0,
| {z }
im(di )

and thus Hi (K • ) is also a vector bundle. As cokernels always compute with base
change, so does Hi (K • ) = coker(K i−1 → ker(di )). See also [Mum70a, Cor. 2, p.48],
[Har77, Cor. 12.9], or [Vak17, 28.1.5].

The reducedness hypothesis in Grauert’s Theorem is quite restrictive in applica-

tions to moduli theory, where we often need to establish properties of families of
objects over an arbitrary base. Fortunately, with a little more linear algebra, we can
establish the following criterion which holds over any base.

Theorem A.6.8 (Cohomology and Base Change II). Let f : X → Y be a proper

and finitely presented morphism of schemes, and let F be a finitely presented quasi-
coherent sheaf on X flat over Y . Suppose that for a point y ∈ Y and integer i, the
comparison map ϕiy : Ri f∗ F ⊗ κ(y) → Hi (Xy , Fy ) is surjective. Then the following
hold:

488
 (a) There is an open neighborhood V ⊆ Y of y such that for every morphism
Y ′ → V of schemes, the comparison map ϕiY ′ : g ∗ Ri f∗ F → Ri f∗′ g ′∗ F is an
isomorphism. In particular, ϕiy is an isomorphism.
(b) ϕi−1
y is surjective if and only if Ri f∗ F is a vector bundle in an open neighbor-
hood of y.
Proof. Assuming that Y is noetherian, we reduce to the case that Y = Spec A is
affine. Theorem A.6.2 constructs a complex K • such that for each y ∈ Y , there is a
morphism of complexes

K i−1
di−1 / Ki di / K i+1 computing Hi (X, F )

 di−1 ⊗κ(y)
 di ⊗κ(y)

K i−1 ⊗A κ(y) / K i ⊗A κ(y) / K i+1 ⊗A κ(y) computing Hi (Xy , Fy )

The following claim will be used twice in the proof: the surjectivity of Hi (X, F ) →
Hi (Xy , Fy ) is equivalent to the surjectivity of ker(di ) → ker(di ⊗ κ(y)). Since
K i−1 → K i−1 ⊗A κ(y) is surjective, so is im(di−1 ) → im(di−1 ⊗ κ(y)). The snake
lemma applied to

0 / im(di−1 ) / ker(di ) / Hi (X, F ) /0

  
0 / im(di−1 ⊗ κ(y)) / ker(di ⊗ κ(y)) / Hi (Xy , Fy ) /0

establishes the claim.

For (a), our hypothesis is that Hi (X, F ) → Hi (Xy , Fy ) is surjective, which by the
claim implies that ker(di ) → ker(di ⊗ κ(y)) is surjective. Proposition A.6.6 implies
that di : K i → K i+1 is a map of vector bundles, and that after replacing Y with an
open neighborhood of y, ker(di ) is a vector bundle whose construction commutes
with base change. Thus Hi (X, F ) = coker(K i−1 → ker(di )) also commutes with
base change. For (b), we use the equivalences:
A.6.6
Hi (X, F ) is a vector bundle ⇐⇒ K i−1 → ker(di ) is a map of vector bundles
⇐⇒ K i−1 → K i is a map of vector bundles
A.6.6
⇐⇒ ker(di−1 ) ⊗ κ(y) → ker(di−1 ⊗ κ(y)) is surjective
claim
⇐⇒ Hi−1 (X, F ) → Hi−1 (Xy , Fy ) is surjective.

The first equivalence follows from Proposition A.6.6 as Hi (X, F ) is the cokernel of
K i−1 → ker(di ), the second follows from the observation that since di is a map of
vector bundles, ker(di ) and im(di ) are vector bundles, and the map ker(di ) → K i
(whose cokernel is im(di )) is also a map of vector bundles, the third also follows
from Proposition A.6.6, and the fourth is the claim at the beginning of the proof.
Using the methods of Limits of Schemes (§B.3), it is not hard to see how the
general statement follows from the noetherian version. Assuming Y is affine, write
Y = limλ∈Λ Yλ as a limit of affine schemes of finite type over Z. Since X → Y is
finitely presented, there exists an index 0 ∈ Λ and a finitely presented morphism
X0 → Y0 such that X ∼ = X0 ×Y0 Y (Proposition B.3.3). For each λ > 0, we can
define Xλ = X0 ×Y0 Yλ and we have X ∼ = Xλ ×Yλ Y . By Proposition B.3.7, Xλ → Yλ

489
is proper for λ ≫ 0. By Proposition B.3.4(1), there exists an index µ ∈ Λ and a
coherent sheaf Fµ on Xµ that pulls back to F under X → Xµ . For λ > µ, set Fλ
to be the pullback of Fµ under Xλ → Xµ . By Proposition B.3.4(3), Fλ is flat over
Yλ for λ ≫ 0. We may now apply noetherian Cohomology and Base Change to the
data of Xλ → Yλ and Fλ for λ ≫ 0, and we may deduce the same properties for
X → Y and F under the base change Y → Yλ . See also [EGA, III2 .7.7.5, III2 .7.7.10,
III2 .7.8.4], [Har77, Thm. 12.11], and [Vak17, Thm. 25.1.6].

The following exercise will give you some practice applying Cohomology and
Base Change.

Exercise A.6.9. Let f : X → Y be a proper morphism of noetherian schemes. For

a coherent sheaf F flat over Y , the following are equivalent:
(1) Hi (Xy , Fy ) = 0 for all y ∈ Y and i > 0; and
(2) Ri f∗ F = 0 for all i > 0, and f∗ F is a vector bundle whose construction
commutes with base change on Y .

A.6.2 Applications to moduli theory

Here is a typical moduli-theoretic application of Cohomology and Base Change
establishing properties of smooth families of curves (Proposition 5.1.16), which is
applied for instance in the Algebraicity of Mg (3.1.17). The argument below applies
equally to families of stable curves (Proposition 5.3.23).

Proposition A.6.10. Let π : C → S be a family of smooth curves of genus g ≥ 2

(i.e., C → S is a smooth, proper morphism of schemes such that every geometric
fiber is a connected curve of genus g). Then
(1) π∗ OC = OS ,
(2) π∗ (ΩC/S ) is a vector bundle of rank g whose construction commutes with base
change on S and R1 π∗ (ΩC/S ) ∼ = OS while Ri π∗ (ΩC/S ) = 0 for i ≥ 2, and
(3) for k > 1, the pushforward π∗ (Ω⊗k
C/S ) is a vector bundle of rank (2k − 1)(g − 1)
whose construction commutes with base change on S and Ri π∗ (Ω⊗k
C/S ) = 0 for
i > 0.

Proof. To see (1), observe that H0 (Cs , OCs ) = κ(s) for all s ∈ S since Cs is proper
and geometrically connected. It follows that ϕ0s : π∗ OC ⊗ κ(s) → H0 (Cs , OCs ) is
surjective. Cohomology and Base Change (A.6.8(a)–(b)) with i = 0 implies that ϕ0s
is an isomorphism and that π∗ OC is a line bundle. On a fiber over s ∈ S, the natural
map OS → π∗ OC induces a surjective map κ(s) → π∗ OC ⊗ κ(s) (as post-composing
with ϕ0s : π∗ OC ⊗ κ(s) → H0 (Cs , OCs ) = κ(s) is the identity). Thus OS → π∗ OC is a
surjective morphism of line bundles, hence an isomorphism.
For (2), since ΩC/S is a relative dualizing sheaf (see [Liu02, §6.4]), Grothendieck–
Serre Duality implies that R1 π∗ ΩC/S ∼ = π∗ OC and this is identified with OS by
(1). For i ≥ 2, Hi (Cs , ΩC/S ⊗ κ(s)) = 0 (as dim Cs = 1), and A.6.8(a) implies that
Ri π∗ ΩC/S = 0. Applying A.6.8(b) with i = 2 yields that ϕ1s : R1 π∗ ΩC/S ⊗ κ(s) →
H1 (Cs , ΩCs /κ(s) ) is surjective for every s ∈ S and applying A.6.8(a) with i = 1 shows
that ϕ1s is an isomorphism. Since R1 π∗ ΩC/S is a line bundle, applying A.6.8(b) with
i = 1 shows that ϕ0s is surjective, and applying A.6.8(a)–(b) with i = 0 implies that
π∗ ΩC/S is a vector bundle of rank h0 (Cs , ΩCs /κ(s) ) = g whose construction commutes
with base change.

490
 ⊗(1−k)
For (3), since k > 1, we have that deg(ΩCs /κ(s) ) < 0 for each s ∈ S, and Serre
⊗(1−k)
Duality (5.1.3) implies that H1 (Cs , Ω⊗k
Cs /κ(s) ) = H (Cs , ΩCs /κ(s) ) = 0. Observe that
0

Hi (Cs , Ω⊗k
Cs /κ(s) ) = 0 for i ≥ 2 since dim Cs = 1. Cohomology and Base Change
(A.6.8(a)) gives Ri π∗ (Ω⊗k ⊗k
C/S ) = 0 for i > 0. On the other hand, h (Cs , ΩCs /κ(s) ) =
0

deg(Ω⊗k
Cs /κ(s) ) + 1 − g = (2k − 1)(g − 1) by Riemann–Roch (5.1.2). Applying
Cohomology and Base (A.6.8(b)) with i = 0 yields that π∗ (Ω⊗k
C/S ) is a vector bundle
of rank (2k − 1)(g − 1).

Similarly, we can apply Cohomology and Base Change to establish properties of

families of coherent sheaves, which we will need for instance for the Algebraicity of
Bun(C) (3.1.21).

Proposition A.6.11. Let p : X → S be a proper morphism of schemes and F be a

finitely presented quasi-coherent sheaf on X flat over S. Suppose that dim Xs ≤ d
for all s ∈ S. The subset S ′ of points s ∈ S such that Hj (Xs , Fs ) = 0 for all j > 0
is open. Denoting X ′ = p−1 (S ′ ), p′ := p|X ′ : X ′ → S, and F ′ = F |X ′ , we have
that Rj p′∗ F ′ = 0 for all j > 0 and that p′∗ F ′ is a vector bundle whose construction
commutes with base change.

Proof. For each j = 1, . . . , d, A.6.8(a) implies that the locus of points s ∈ S such that
Hj (Xs , Fs ) = 0 is open and the comparison map ϕjs : Rj p∗ F ⊗ κ(s) → Hj (Xs , Fs )
is an isomorphism. It follows that Rj p′∗ F = 0 which allows us to apply A.6.8(b)
with i = 1 to conclude that ϕ0s : p′∗ F ′ ⊗ κ(s) → H0 (Xs , Fs ) is surjective. Applying
A.6.8(a)–(b) with i = 0 gives the final statement.

A.6.3 Applications to line bundles

Given a proper flat morphism f : X → Y , when is a line bundle L on X the pullback
of a line bundle on Y ? More generally, is there a largest subscheme Z ⊆ Y where
LZ on XZ = X ×Y Z is the pullback of a line bundle on Z? In this section, we
provide three answers in increasing complexity. As the results depend on properties
of the fibers Xy , we first discuss relationships between various conditions.

Lemma A.6.12. Let f : X → Y be a proper flat morphism of noetherian schemes.

Consider the following conditions:
(1) the geometric fibers of f : X → Y are non-empty, connected, and reduced;
(2) h0 (Xy , OXy ) = 1 for all y ∈ Y ; and
(3) OY = f∗ OX and this holds after arbitrary base change (i.e., OT = fT,∗ OXT
for a morphism T → Y of schemes).
Then (1) ⇒ (2) ⇐⇒ (3).

Proof. If (1) holds, then H0 (Xy , OXy )⊗κ(y) κ(y) = H0 (X ×Y κ(y), OX×Y κ(y) ) by Flat
Base Change (A.2.12, and since a connected, reduced, and proper scheme over an
algebraically closed field has only constant functions, we conclude that h0 (Xy , OXy ) =
1. If (2) holds, then the comparison map ϕ0y : f∗ OX ⊗ κ(y) → H0 (Xy , OXy ) = κ(y)
is necessarily surjective as there is a global section 1 ∈ H0 (Y, f∗ OX ). Applying
Theorem A.6.8 with i = 0 shows that f∗ OX is a line bundle. As OY → f∗ OX is a
surjection of line bundles, it is an isomorphism. Since the same argument applies to
the base change XT → T , this gives (3). The converse (3) ⇒ (2) by considering the
map T = Spec κ(y) → Y .

491
 When the base is reduced, Grauert’s Theorem provides a complete answer to
when a line bundle is a pullback.
Proposition A.6.13 (Version 1). Let f : X → Y be a proper flat morphism of
noetherian schemes such that h0 (Xy , OXy ) = 1 for all y ∈ Y . Let L be a line bundle
on X. If Y is reduced, then L = f ∗ M for a line bundle M on Y if and only if Ly
is trivial for all y ∈ Y . Moreover, if these conditions hold, then M = f∗ L and the
adjunction morphism f ∗ f∗ L → L is an isomorphism.
Proof. The condition on geometric fibers implies that h0 (Xy , Ly ) = 1 and Grauert’s
∼
Theorem (A.6.7) implies that f∗ L is a line bundle and that f∗ L ⊗ κ(y) → H0 (Xy , Ly )
is an isomorphism. We claim that f f∗ L → L is surjective. It suffices to show
∗

that (f ∗ f∗ L)|Xy → L|Xy is surjective. Denoting fy : Xy → Spec κ(y), we have

identifications (f ∗ f∗ L)|Xy = fy∗ (f∗ L ⊗ κ(y)) = fy∗ (OSpec κ(y) ) = OXy and the claim
follows. Since f ∗ f∗ L → L is a surjection of line bundles, it is an isomorphism.
Exercise A.6.14. Show that if Y is a connected and reduced noetherian scheme
and E is a vector bundle on Y , then Pic(P(E)) = Pic(Y ) × Z. See also [Har77,
Exc. III.12.5].
Proposition A.6.15 (Version 2). Let f : X → Y be a proper flat morphism of
noetherian schemes with integral geometric fibers. For a line bundle L on X, the
locus {y ∈ Y | Ly is trivial on Xy } is a closed subset of Y .
Proof. The important observation here is that for a geometrically integral and proper
scheme Z over field k, a line bundle M is trivial if and only if h0 (Z, M ) > 0 and
h0 (Z, M ∨ ) > 0. To see that the latter condition is sufficient, observe that we have
nonzero homomorphisms OZ → M and OZ → M ∨ , the latter of which dualizes to
a nonzero map M → OZ . Since Z is integral, the composition OZ → M → OZ
is also nonzero and thus an isomorphism as it corresponds to a nonzero constant
in H0 (Z, OZ ) = k. It follows that M → OZ is a surjective map of line bundles,
hence an isomorphism. By the Semicontinuity Theorem (A.6.4) the condition that
h0 (Xy , Ly ) > 0 and h0 (Xy , L∨
y ) > 0 is closed, and the statement follows. See also
[Mum70a, Cor. 6, p. 54].
Remark A.6.16. If the geometric fibers are only connected and reduced, the locus
may fail to be closed. For example, giving a smooth family f : X → Y of curves
over a smooth curve, and consider the blowup Blx X → X at a closed point x ∈ X
with exceptional divisor E. Then Blx X → Y is a proper flat morphism, and the
fiber over f (x) ∈ Y is connected and reduced but reducible. Setting L = OBlx X (E),
the fiber Ly is trivial if and only if y ̸= f (x).
For moduli-theoretic applications, it is essential that we allow for the base Y
to be non-reduced, and provide the locus Z ⊆ Y with a functorial description. For
many applications (e.g., to families of stable curves), it is also necessary to allow
for reducible fibers Xy . Our final and strongest version incorporates both versions
above and is proved using the algebraic formulation of Cohomology and Base Change
(A.6.2). This result will be applied in the proof of Algebraicity of Mg (3.1.17) to
exhibit a locally closed subscheme of the Hilbert scheme parameterizing smooth
curves that are tricanonically embedded.
Proposition A.6.17 (Version 3). Let f : X → Y be a proper flat morphism of
noetherian schemes such that h0 (Xy , OXy ) = 1 for all y ∈ Y (resp., the geometric
fibers are integral). For a line bundle L on X, there is a unique locally closed (resp.,
closed) subscheme Z ⊆ Y such that

492
 (1) LZ on XZ = X ×Y Z is the pullback of a line bundle on Z, and
(2) if T → Y is a morphism of schemes such that LT on XT is the pullback of a
line bundle on T , then T → Y factors through Z.
In other words, the functor

Sch/Y → Sets,
if LT is the pullback of a line bundle on T

{∗}
(T → Y ) 7→
∅ otherwise

is representable by a locally closed (resp., closed) subscheme of Y .

Proof. We begin with the observation that L is the pullback of a line bundle if and
only if f∗ L is a line bundle and the adjunction map f∗ f ∗ L → L is an isomorphism.
Indeed, if L = f ∗ M for a line bundle M on Y , then the projection formula and the
equality OY = f∗ OX (Lemma A.6.12) shows that

f∗ L ∼
= f∗ f ∗ M ∼
= f∗ OX ⊗ M ∼
=M

is a line bundle and that f ∗ f∗ L → L is an isomorphism. As the same holds for the
base change XT → T , we see that the question is Zariski-local on Y and T . We will
show that every point y ∈ Y has an open neighborhood where the proposition holds.
By the Semicontinuity Theorem (A.6.4), the locus V = {y ∈ Y | h0 (Xy , Ly ) ≤ 1}
is open. Since Ly is not trivial for points y ∈ / V , we may replace Y with V and
assume that h0 (Xy , Ly ) ≤ 1 for all y ∈ Y . By Cohomology and Base Change (A.6.2)
and after replacing Y with an open affine neighborhood of y, we may assume that
d
there is a homomorphism d : Ar0 − → Ar1 of finitely generated and free A-modules
such that for every ring map A → B, H0 (XB , LB ) = ker(d ⊗ B). Using the dual d∨
of d, we define M as the cokernel in the sequence
d∨
Ar1 −−→ Ar0 → M → 0.

Tensoring over A → B yields a right exact sequence

d∨ ⊗B
B r1 −−−−→ B r0 → M ⊗A B → 0,

which after applying the contravariant left-exact functor HomB (−, B) becomes
d⊗ B
0 → HomB (M ⊗A B, B) → B r0 −−−A−→ B r1 .

We conclude that

H0 (XB , LB ) = HomB (M ⊗A B, B) = HomA (M, B). (A.6.18)

Applying this to A → κ(y) for y ∈ Y , we obtain that H0 (Xy , Ly ) =

HomA (M, κ(y)) = (M ⊗A κ(y))∨ .
If h0 (Xy , Ly ) = 0, then Ly is not trivial and since M ⊗A κ(y) = 0, there is
an open neighborhood U of y such that M f|U = 0. The proposition holds over
U . If h0 (Xy , Ly ) = 1, then M ⊗A κ(y) = κ(y) and by Nakayama’s lemma, after
replacing Y with an open affine neighborhood of y, there is a surjection A → M .
Write M = A/I for an ideal I and define the closed subscheme Z = V (I) ⊆ Y .
Observe that H0 (Z, LZ ) = HomA (A/I, A/I) = A/I so that fZ,∗ LZ is the trivial line
bundle. To see that the construction of fZ,∗ LZ commutes with base change, let B

493
be an A/I-algebra and observe that H0 (XB , LB ) = HomA (A/I, B) = B and that
H0 (XZ , LZ ) ⊗A/I B → H0 (XB , LB ) is an isomorphism.
We claim that T → Y factors through Z if and only if fT,∗ LT is a line bundle. The
(⇒) implication is clear: fZ,∗ LZ is a line bundle and its construction commutes with
base change. The converse is Zariski-local on T and we may assume that T = Spec B
is affine and fT,∗ LT = OT . Then (A.6.18) implies that B = HomA (A/I, B). Thus,
I ⊆ ker(A → B) or, in other words, A → B factors as A → A/I → B. Finally,
considering the adjunction morphism λ : fZ∗ fZ,∗ LZ → LZ on XZ , we claim that
for y ∈ Z, Ly is trivial if and only if λ|Xy is surjective. If λ|Xy is surjective, then
using that fZ,∗ LZ = OZ , we have a surjection OXy → Ly of line bundles, hence an
isomorphism. For converse, since fZ,∗ LZ commutes with base change, the comparison
map fZ,∗ LZ ⊗ κ(y) = H0 (Xy , Ly ) is an isomorphism. Denoting fy : Xy → Spec κ(y),
we have identifications (fZ∗ fZ,∗ LZ )|Xy = fy∗ (fZ,∗ LZ ⊗ κ(y)) = fy∗ fy,∗ Ly and λ|Xy
corresponds to the adjunction map fy∗ fy,∗ Ly → Ly , which is an isomorphism (as Ly
is trivial). Replacing Z with Z ∖ Supp(coker(λ)) establishes the proposition. If, in
addition, the fibers Xy are geometrically integral, then Proposition A.6.15 implies
that Z is closed. See also [Mum70a, p. 90], [Vie95, Lem. 1.19], and [SP, Tags 0BEZ
and 0BF0].
Remark A.6.19. For a proper flat morphism X → S, the relative Picard functor is
defined as
Picrel
X/S : Sch/S → Sets, T 7→ Pic(XT )/ Pic(T );
see §6.4.7 for an exposition of Picard functors. If f : X → S has geometrically
reduced (resp., integral) fibers, then the existence of a locally closed (resp., closed)
subscheme Z ⊆ Y characterized by Proposition A.6.17 is equivalent to the diagonal
of Picrel
X/S being representable by locally closed immersions (resp., closed immersions).
In the case of geometrically integral fibers, this translates to the separatedness of
X/S → S. In this language, the above result was established in [FGAV, Thm. 3.1].
Picrel

A.7 Quasi-finite morphisms and Zariski’s Main The-

orem
A locally of finite type morphism f : X → Y of schemes is locally quasi-finite at
x ∈ X if x is isolated in the fiber Xf (x) = X ×Y Spec κ(f (x)). When f : X → Y is
also quasi-compact, then this is equivalent to the finiteness of the set f −1 (f (x)),
and we say that f : X → Y is quasi-finite.

A.7.1 Étale Localization of Quasi-Finite Morphisms

Theorem A.7.1 (Étale Localization of Quasi-Finite Morphisms). Let f : X → S be
a separated and finite type morphism of schemes. Suppose that f is quasi-finite at
every preimage of s ∈ S. There exists an étale neighborhood (S ′ , s′ ) → (S, s) with
κ(s′ ) = κ(s) and a decomposition X ×S S ′ = Z ⨿ W into open and closed subschemes
such that Z → S ′ is finite and the fiber Ws′ is empty. Moreover, it can be arranged
that Z factors as Z1 ⨿ · · · ⨿ Zn where each Zi contains precisely one point zi over s
with κ(zi )/κ(s) purely inseparable.
Proof. See [EGA, IV.8.12.3] or [SP, Tag 04HF].
The statement implies that any quasi-finite algebra A over a henselian local ring
R is a product A ∼
= B × C with B finite over R and C ⊗R R/mR = 0, a property

494
that, in fact, characterizes henselian local rings (Proposition B.5.9). It also provides
the key technical input in factoring quasi-finite morphisms.

A.7.2 Factorizations of quasi-finite morphisms

Proposition A.7.2. A quasi-finite and separated morphism f : X → Y of schemes
factors as
f : X → SpecY f∗ OX → Y
where X ,→ SpecY f∗ OX is an open immersion and SpecY f∗ OX → Y is affine.
Proof. As f∗ OX commutes with étale (even flat) base changes on Y , so does the
factorization. Therefore, it suffices to show that every point y ∈ Y has an étale
neighborhood where the proposition holds. By Theorem A.7.1 we may assume
that X = X1 ⨿ X2 with X1 finite over Y and (X2 )y = ∅. After replacing Y with
SpecY f∗ OX , we may also assume that f∗ OX = OY . As OX = A1 ×A2 is the product
of quasi-coherent OX -algebras, OY = f∗ OX = f∗ A1 × f∗ A2 and thus Y decomposes
as Y1 ⨿ Y2 such that y ∈ Y1 and f (Xi ) ⊆ Yi for i = 1, 2. After replacing Y with Y1 ,
we see that X → Y is finite. Thus X is affine and X = Y = SpecY f∗ OX .
In the above factorization, f∗ OY may not be a finite type OY -algebra; even if
Y is a noetherian affine scheme, then Γ(X, OX ) may not be a noetherian ring (see
[Ols16, Ex. 7.2.15]). However, we may modify the factorization to arrange that
X → Y factors as an open immersion followed by a finite morphism.
Theorem A.7.3 (Zariski’s Main Theorem).
Proof. If A ⊆ f∗ OX denotes the integral closure of OY → f∗ OX , f factors as the
composition of
j
f: X −→ SpecY A → Y.
We claim that j is an open immersion. It suffices to show that for every point x ∈ X,
there is an open neighborhood V ⊆ SpecY A of j(x) such that j −1 (V ) → V is an iso-
morphism. Since normalization commutes with étale base change (Proposition A.7.4)
and since being an open immersion is an fpqc local property (Proposition 2.1.26),
we are free to replace Y by an étale neighborhood of f (x). By Theorem A.7.1, we
can assume that X = F ⨿ W with F finite over Y and x ∈ F . In this case, the
normalization SpecY A of Y in X is F ⨿ W f where W f is the normalization of Y in
∼
W . As j −1 (F ) → F , the claim follows. By construction, SpecY A → Y is integral.
We can write A = colim Aλ , where each Aλ is a finite type OY -algebra. Since open
immersions descent under limits (Proposition B.3.7), X → SpecY Aλ is an open
immersion for λ ≫ 0. Since SpecY Aλ → Y is integral and of finite type, it is finite.
See also [EGA, IV.8.12.6] or [SP, Tag 05K0].
The following algebra result was used above and will also be used in the general-
izations of Zariski’s Main Theorem to algebraic spaces (Theorem 4.5.9) and stacks
(Theorem 6.1.18).

Proposition A.7.4. Let Y be a scheme, B be a quasi-coherent OY -algebra and Be

be the integral closure of OY in B. If f : X → Y is a smooth morphism, then f ∗ Be is
identified with the integral closure of OX in f ∗ B.
Proof. See [SP, Tag 03GG] or [LMB00, Prop. 16.2].
Zariski’s Main Theorem has some useful corollaries.

495
Corollary A.7.5. A quasi-finite and proper morphism (resp., proper monomor-
phism) of schemes is finite (resp., a closed immersion).
Proof. If f : X → Y is a quasi-finite and proper, Zariski’s Main Theorem (A.7.3)
gives a factorization f : X ,→ X e → Y and the dense open immersion X ,→ X e is
also closed, thus an isomorphism. On the other hand, if f : X → Y is a proper
monomorphism, then it is also quasi-finite, thus finite. The statement reduces to the
algebraic fact that a finite epimorphism of rings is surjective (c.f.,[SP, Tag 04VT]).
Remark A.7.6. As universally closed morphisms are necessarily quasi-compact [SP,
Tag 04XU], every universally closed and locally of finite type monomorphism is a
closed immersion; see also [SP, Tag 04XV].

496
Appendix B

Further topics in scheme theory

Algebraic geometry seems to have acquired the reputation of being esoteric,

exclusive, and very abstract, with adherents who are secretly plotting to
take over all the rest of mathematics. In one respect this last point is
accurate.
David Mumford

B.1 Algebraic groups

We provide a crash course in group schemes, algebraic groups, and their actions. For
a more detailed exposition for algebraic groups, we recommend [Bor91], [Hum75],
[Spr98], [Wat79], and [Mil17], while for group schemes over a general base, we
recommend [SGA3I ],[SGA3II ], [SGA3III ], [DG70], and [Con14].

B.1.1 Group schemes and their actions

Definition B.1.1. A group scheme over a scheme S is a morphism π : G → S
of schemes together with a multiplication morphism m : G ×S G → G, an inverse
morphism ι : G → G, and an identity morphism e : S → G (with each morphism
over S) such that the following diagrams commute:

G ×S G ×S G
id ×m
/ G ×S G G
(id,ι)
/ G ×S G G
(e◦π,id)
/ G ×S G
e◦π id
m×id m (ι,id) m (id,e◦π) m
   %/   %/ 
G ×S G
m /G G ×S G
m
G G ×S G
m
G
Associativity Law of inverse Law of identity

For group schemes H and G over S, a morphism of group schemes is a morphism

ϕ : H → G of schemes over S such that mG ◦ (ϕ × ϕ) = ϕ ◦ mH . A (closed) subgroup
mG
of G is a nonempty (closed) subscheme H ⊆ G such that H × H ,→ G × G −−→ G
ι
and H ,→ G → − G factor through H. We say that a subgroup H ⊆ G is normal if
for every S-scheme T , the subgroup H(T ) ⊆ G(T ) is normal.
Remark B.1.2. If G and S are affine, then by reversing the arrows above gives
Γ(G, OG ) the structure of a Hopf algebra over Γ(S, OS ).

497
Exercise B.1.3. Show that a group scheme over S is equivalently defined as a
scheme G over S together with a factorization

Sch/S / Gps

MorS (−,G)
# 
Sets
where Gps → Sets is the forgetful functor.
(We are not requiring that there exists a factorization; the factorization is part of
the data! Indeed, the same scheme can have multiple structures as a group scheme,
e.g., Z/4 and Z/2 × Z/2 over C.)
Example B.1.4 (Important examples). Let S = Spec R.
(1) The multiplicative group scheme over R is Gm,R = Spec R[t]t with comulti-
plication m∗ : R[t]t → R[t]t ⊗R R[t′ ]t′ given by t 7→ tt′ . The product Gnm,R is
called the split torus of rank n.
(2) The group scheme of nth roots of unity is µn,R = Spec R[t]/(tn − 1) with
comultiplication given by t 7→ tt′ .
(3) The additive group scheme over R is Ga,R = Spec R[t] with comultiplication
m∗ : R[t] → R[t] ⊗R R[t′ ] given by t 7→ t + t′ .
Let V be a free R-module of finite rank.
(3) The general linear group on V is

GL(V ) = Spec(Sym∗ (End(V ))det ),

where det denotes the determinant polynomial and where the comultiplication
m∗ : Sym∗ (End(V )) → Sym∗ (End(V )) ⊗R Sym∗ (End(V )) is defined as follow-
ing: for a basis v1 , . . . , vn of V , then for i, j = 1, . . . , n, the endomorphisms
xij : V → V defined by vi 7→ vj and vk 7→ 0 if k ̸= i define a basis of End(V ),
and we define m∗ (xij ) = xi1 ⊗ x1j + · · · + xin ⊗ xnj .
(4) The special linear group on V is the closed subgroup SL(V ) ⊆ GL(V ) defined
by det = 1.
(5) The projective linear group PGL(V ) is the affine group scheme

Proj(Sym∗ (End(V )))det

with the comultiplication analogously to GL(V ).

We write GLn,R = GL(Rn ), SLn,R = GL(Rn ), and PGLn,R = PGL(Rn ). We often
simply write Gm , GLn , SLn , and PGLn when the base ring is understood.
Exercise B.1.5.
(a) Provide functorial descriptions of each of example above.
(b) Show that every abstract group G can be given the structure of a group scheme
⨿g∈G S over a base scheme S. Provide both explicit and functorial descriptions.
By abuse of notation, this group scheme is sometimes denoted as G → S.
(c) Show that if n is invertible in Γ(S, OS ), then µn,S is isomorphic to the group
scheme induced by the finite group Z/nZ.
Example B.1.6 (Diagonalizable group schemes). Let R be a ring and A be a
finitely generated abelian group. If we define R[A] as the free R-module generated

498
by elements of A, then R[A] inherits an R-algebra structure with multiplication
on generators induced from multiplication in A. The comultiplication R[A] →
R[A] ⊗R R[A] defined by a 7→ a ⊗ a defines a group scheme DR (A) = Spec R[A] over
Spec R. A group scheme G over Spec R is diagonalizable if G ∼ = DR (A) for some A.
The group scheme DR (Zr ) = Grm,R is the r-dimensional split torus while
DR (Z/n) = µ n,R = Spec R[t]/(tn − 1) is the group of nth roots of unity; this
holds for any ring R and integer n even if n is not invertible in R. The classification
of finitely generated abelian groups implies that every diagonalizable group scheme
is a product of Grm × µ n1 × · · · × µ nk .
A group scheme G → S is of multiplicative type if it becomes diagonalizable after
étale cover of S.
Exercise B.1.7.
(1) Describe DR (A) as a functor Sch/R → Gps.
(2) Show that a homomorphism DR (A) → DR (B) of group schemes over R is
equivalent to a group homomorphism B → A.
We recall the following general properties of group schemes.
Proposition B.1.8. Let G → S be a locally of finite type group scheme.
(1) If S = Spec k, then dim G = dime G, where e ∈ G(k) denotes the identity.
(2) The function
S → Z, s 7→ dim Gs
is upper semicontinuous.
(3) G → S is trivial if and only if the fiber Gs is trivial for each s ∈ S.
(4) G → S is unramified (resp., separated, quasi-separated) if and only if the
identity section e : S → G is an open immersion (resp., a closed immersion,
quasi-compact).
Proof. For (1), we may assume that k is algebraically closed. In this case, an
element g ∈ G(k) defines an isomorphism g : G → G, so that dime G = dimg G.
For (2), for any locally of finite type morphism π : G → S, the function G → Z,
defined by g 7→ dim Gπ(g) , is upper semicontinuous [EGA, IV.13.1.3]. As G → S is
a group scheme, there is an identity section S → G and therefore the composition
S → G → Z, defined by s 7→ dime(s) Gs = dim Gs , is upper semicontinuous. For
(3), G → S is unramified since every fiber is. Therefore ΩG/S = 0 and the diagonal
G → G ×S G is an open immersion. It follows that the identity section S → G is
a surjective open immersion, thus an isomorphism. For (4), G → S is unramified
(resp., separated, quasi-separated) if and only if ∆G/S : G → G ×S G is an open
immersion (resp., a closed immersion, quasi-compact), and the cartesian diagram

S
e /G π /S

e □ ∆G/S □ e
  
G / G ×S G /G
∆G/S (g,h)7→g −1 h

implies that this is equivalent to e : S → G being an open immersion (resp., a closed

immersion, quasi-compact).

Actions.

499
Definition B.1.9 (Actions). Let G → S be a group scheme with multiplication m
and identity e. An action of G on a scheme p : X → S is a morphism a : G×S X → X
over S such that the following diagrams commute:

G ×S G ×S X
id ×a
/ G ×S X X
(e◦p,id)
/ G ×S X
m×id a a
  id
& 
G ×S X
a /G X
If X and Y are S-schemes with actions of G, a morphism f : X → Y of S-schemes
is G-equivariant if aY ◦ (id ×f ) = f ◦ aX , and is G-invariant if G-equivariant and Y
has the trivial G-action.
For example, a group scheme G → S acts on itself via left multiplication g·g ′ = gg ′
or via right multiplication g · g ′ = g ′ g −1 .
Exercise B.1.10. Show that giving a group action of G → S on X → S is the
same as giving an action of the functor MorS (−, G) : Sch/S → Gps on the functor
MorS (−, X) : Sch/S → Sets.
(This requires first spelling out what it means for a functor to groups to act on a
functor to sets.)
Definition B.1.11 (Stabilizers and Orbits). Given an action σ of a group scheme
G → S on X → S, the stabilizer of f of an S-morphism f : T → X is the group
scheme Gf (sometimes written as StabG (f ) or Stab(f )) over T defined as the fiber
product
Gf /T

□ f
 
G ×S T / G ×S X σ / X,
σ
while the orbit of f is defined set-theoretically as the image of G×S T → G×S X −
→ X.
The stabilizer group scheme SX → X is defined as the stabilizer of the identity
id : X → X and is identified with the fiber product
SX /X

∆
 (σ,p2 ) 
G ×S X / X ×S X.

When S = Spec k for a field k and x ∈ X(k), then the stabilizer Gx = StabG (x)
is preimage of x under the map σx : G → X, given by g 7→ x, and is identified with
the fiber of the stabilizer group scheme SX → X. On the other hand, the orbit
Gx (sometimes written as O(x)) is the image of σx . When G and X are of finite
type, the image Gx ⊆ X is locally closed (Proposition B.1.16(5)), and thus Gx has
a natural scheme structure inherited from the scheme-theoretic image of σx ; note
that when G is smooth (e.g., char(k) = 0), the orbit Gx has the reduced scheme
structure.
Quotients. Constructing quotients of subgroups or group actions is a subtle business
in algebraic geometry. If G → S is an fppf group scheme acting freely on a scheme
X → S, the quotient functor is defined as the fppf sheafification of the functor
(X/G)pre : Sch/S → Sets, T 7→ X(T )/G(T ).

500
While the quotient sheaf X/G is not always representable by a scheme, it is a
theorem that X/G is an algebraic space (Theorem 6.3.1). It is also a theorem that
if G → S is smooth, then X/G is identified with the étale sheafification of (X/G)pre
(Corollary 6.3.3).
In special situation however, the quotient is known to exist as a scheme. Here
are three examples, each of which plays a prominent role in this text:
– If G is an algebraic group over k and H ⊆ G is a subgroup, then G/H is a
quasi-projective scheme (Proposition B.1.16(7)).
– If G is a finite group acting freely on an affine scheme X, then X/G is affine
(Proposition 4.3.3). In fact, if G is a finite group acting (not necessarily freely)
on an affine (resp. projective, quasi-projective) scheme X, there exists a
geometric quotient (see Definition 4.2.2); see Exercise 4.2.14. The quotient is
also denoted as X/G but it does not necessarily represent the sheafification of
(X/G)pre .
– If G is a linearly reductive group over k acting on an affine scheme X = Spec A,
there exists a good quotient X//G = Spec AG which has desirable geometric
properties (Definition 7.1.1 and Proposition 7.1.3). Given an action of G on a
projective scheme X, Geometric Invariant Theory addresses how to identify
open subschemes U ⊆ X which admit projective good quotients U//G; see
Chapter 7.
Under quite general conditions, a geometric quotient X/G exists as an algebraic
space, e.g., if G is an affine algebraic group and the action map G × X → X × X is
proper (Corollary 4.4.12).
Inducing actions and the Borel construction. If H ⊆ G is an inclusion of
algebraic groups over k and X is a scheme with an H-action, then H acts freely on
G × X via h · (g, x) = (gh−1 , hx), and we define

G ×H X := (G × X)/H

as the quotient algebraic space, which inherits an action of G via g · (g ′ , x) = (gg ′ , x).
This is sometimes referred to as the Borel construction. In this case, the quotient
stacks [X/H] and [G ×H X/G] are isomorphic (Exercise 3.4.19).
Representations.
Definition B.1.12 (Representations). Let S = Spec R be an affine scheme, and let
G → S be a group scheme with multiplication m and identity e. A representation
(or comodule) of G is an R-module V together with a homomorphism σ : V →
Γ(G, OG ) ⊗R V of R-modules (referred to as a coaction) such that the following
diagrams commute:

V
σ / Γ(G, OG ) ⊗R V V
σ / Γ(G, OG ) ⊗R V
σ id ⊗σ e∗ ⊗id
 m∗ ⊗id
 id
& 
Γ(G, OG ) ⊗R V / Γ(G, OG ) ⊗R Γ(G, OG ) ⊗R V V

Morphisms of representations and subrepresentations are defined in the natural

way. If V is a G-representation, the invariant subspace is defined as V G = {v ∈ V |
σ(v) = 1 ⊗ v}.
Example B.1.13.

501
 (1) Given an R-module V , the trivial representation on V is defined using the
coaction σ(v) = 1 ⊗ v.
(2) The regular representation on Γ(G, OG ) is defined using the comultiplication
m∗ : Γ(G, OG ) → Γ(G, OG ) ⊗R Γ(G, OG ).
(3) The standard representation of GLn,R = Spec R[xij ]det (or a subgroup scheme
of GLn,R ) on
PV = Rn is given by the coaction σ : V → Γ(GLn,R ) ⊗R V defined
n
by σ(ei ) = j=1 xij ⊗ ej where (e1 , . . . , en ) is the standard basis of V .

A representation V of G induces an action of G on A(V ) = Spec(Sym∗ V ), which

we refer to as a linear action.

Exercise B.1.14. If G is a group scheme over a field k, show that a G-representation

of a finite dimensional vector space V is equivalent to a homomorphism G → GL(V )
of group schemes.

Proposition B.1.15. Let G = Dk (A) be a diagonalizable group scheme over a ring

k. Every representation of G is a direct sum of one dimensional representations.

Proof. Let V be a representation of G with coaction σ : V → k[A] ⊗k V . Each

a ∈ A defines a one dimensional representation Wa = k of G defined by the coaction
Wa → k[A] ⊗k Wa defined by 1 7→ a ⊗ 1. For a ∈ A, the subspace

Va := {v ∈ V | σ(v) = a ⊗ v}

is isomorphic to Wa ⊗ Va as G-representations, where Va is viewed as the trivial

representation; if Va is finite dimensional, then Va ∼
= Wadim Va . Note that when a = 0,
Wa is the trivial one
Ldimensional representation and V = V0 . We leave the reader
G
∼
to check that V = a∈A Va as G-representations. See also [Mil17, Thm 12.30] and
[SGA3I , Thm. 5.3.3].

B.1.2 Algebraic groups

An algebraic group over a field k is a group scheme G of finite type over k.

Proposition B.1.16. Let G be a group scheme locally of finite type over field k
(e.g., an algebraic group).
(1) G is separated.
(2) (Cartier’s Theorem) If char(k) = 0, then G is smooth.
(3) If k is perfect, then G is smooth if and only if G is reduced if and only if G is
geometrically reduced, and moreover Gred ⊆ G is a subgroup scheme.
(4) The connected component G0 ⊆ G containing the identity element is an open
and closed irreducible subgroup scheme of finite type over k. Moreover, the
construction of G0 commutes with field extensions of k, and G/G0 is an etale
algebraic group over k.
(5) If G acts on a finite type k-scheme X and x ∈ X is a closed point, the orbit
Gu, defined set-theoretically as the image of G → X, g 7→ g · x, is open in its
closure Gx. In particular, orbits of minimal dimension are clsoed. Moreover,

dim G = dim Gx + dim Gx ,

and the function x 7→ dim Gx is upper semicontinuous while x 7→ dim Gx is

lower semicontinuous.

502
 (6) Every subgroup H ⊆ G is closed.
(7) If G is of finite type and H ⊆ G is a subgroup, then G/H is quasi-projective.
In particular, every algebraic group is quasi-projective.
(8) (Barsotti–Chevalley’s Structure Theorem) If G is smooth and connected, then
there is a unique connected, affine, and normal subgroup H ⊴ G, which is
smooth if k is perfect, such that G/H is an abelian variety.

In (8), an abelian variety by definition is a smooth proper algebraic group over

a field. It is necessarily projective and has a commutative group law [Mum70a,
pp. 39, 59]. An elliptic curve is an abelian variety of dimension 1.

Proof. Proposition B.1.8(4) implies (1) since any k-point of a locally of finite type
k-scheme is closed. For (2), see [Car62, §15], [Oor66], [Mum66a, p.167], [Wat79,
§11.4], [Mil17, Thm. 3.23 and Cor. 8.39], and [SP, Tag 047N]. For (3), see [Mil17,
Prop. 1.26 and Cor. 1.39] and [SP, Tags 047P and 047R]. For (4), see [Wat79, §6.7],
[Hum75, §7.3], [Spr98, Prop. 2.2.1], [Mil17, Prop. 1.34], and [SP, Tag 0B7R]. What
may seem surprising here is that G0 is automatically quasi-compact.1 This follows
from a simple argument: reduce to the case that k is algebraically closed and choose
a nonempty open affine subscheme U ⊆ G. After shrinking, we may assume that
U is closed under taking inverses. The quasi-compactness of G follows from the
surjectivity of the multiplication map U × U → G is surjective. If g ∈ G(k), then
since U is dense, the intersection U ∩ gU contains an element h. If we write h = gu,
then g = hu−1 .
For the first part of (5), see [Bor91, §I.1.8], [Hum75, §8.3], [Spr98, Lem. 2.3.3],
and [Mil17, Prop. 1.68]. The identity dim G = dim Gx + dim Gx follows from
the identification Gx ∼ = G/Gx , while the semicontinuity statements follow from
Proposition B.1.8(2) applied to the stabilizer group scheme SX → X. Part (6)
follows from (5) by considering the action of H on G. For (7), see [Cho57, p.128],
[Bor91, Thm. 6.8], [Ray70, Cor VI.2.6], [Bri17, Thm. 5.2.2], [Hum75, §12], [Spr98,
Thm. 5.5.5], and [Mil17, Thm. 8.4.4]. Chevalley announced a proof of (8) in 1953,
but a proof did not appear until [Che60]. In the meantime, Barsotti provided
an independent proof [Bar55a], [Bar55b]. Rosenlicht provided a more elementary
argument in [Ros56]. See also [Con02], [Bri17, Thms. 1 and 2], and [Mil17, Thm. 8.27].

B.1.3 Affine algebraic groups.

We are particularly interested in affine algebraic groups, which are sometimes also
called linear algebraic groups, as justified by (2) below.

Proposition B.1.17. Let G be an affine algebraic group over a field k.

(1) Every representation V of G is a union of its finite dimensional subrepresen-
tations.
(2) There exists a finite dimensional representation V and a closed immersion
G ,→ GL(V ) of group schemes.

Proof. For (1), let σ : V → Γ(G, OG ) ⊗k V be the coaction. It suffices to show that
every finite dimensional subspace W ⊆ V is contained in a finite dimensional sub-G-
representation W ′ ⊆ V . If w1 , . . . , wn is a basis of W and σ(wi ) = j fij ⊗ vij , then
P
one checks that the subspace generated by vij is G-invariant and contains W . For (2),
1We use this in the text to show the boundedness of Pic0X ; see Theorem 6.4.51.

503
we consider the regular representation Γ(G, OG ) of G and apply (1) to construct a
finite dimensional subrepresentation V containing k-algebra generators. One checks
that this gives a closed immersion G ,→ GL(V ). See [Bor91, §I.1.9-10], [Hum75,
§8.6], [Spr98, Prop. 2.3.6 and Thm. 2.3.7], and [Mil17, Prop. 4.7, Cor. 4.10].

We will repeatedly use the following simple consequence of Proposition B.1.17(1).

Proposition B.1.18. Let G be an affine algebraic group over a field k. Let X be

an affine scheme of finite type over k with an action of G.
(1) There exists a G-equivariant closed immersion X ,→ A(V ) where V is a finite
dimensional G-representation.
(2) For every G-invariant closed subscheme Z ⊆ X, there exists a G-equivariant
morphism f : X → A(W ), where W is a finite dimensional G-representation,
such that f −1 (0) = Z.

Proof. Write X = Spec A and let f1 , . . . , fn be k-algebra generators. By B.1.17(1)

there is a finite dimensional G-invariant subspace V ⊆ A containing each fi . The
surjection Sym∗ V → A induces a G-equivariant embedding X ,→ A(V ). For (2),
let Z = Spec A/I and let g1 , . . . , gm ∈ I be generators. Letting W ⊆ I be a finite
dimensional G-invariant subspace containing each gi , the G-invariant morphism
f : X → A(W ) has the desired property that f −1 (0) = Z.

Tori. A subgroup T ⊆ G of an affine algebraic group over a field k is called a torus

(resp., split torus) of rank n if Tk ∼
= Gnm,k (resp., T ∼
= Gnm,k over k ), and a maximal
torus if it T is not contained in a larger subtorus of G. For example, the set of
diagonal matrices in GLn is a split maximal torus of rank n.

Proposition B.1.19. Let G be an affine algebraic group over a field k.

(1) G contains a maximal torus T such that Tk′ ⊆ Gk′ is a maximal torus for
every field extension k → k′ .
(2) If k is algebraically closed, all maximal tori are conjugate.

Proof. See [Bor91, §III.8], [Hum75, §34.3-5], [Spr98, Thm. 13.3.6], and [Mil17,
Thms. 17.82 and 17.105].

There is of course much more to the theory of affine algebraic groups. We quickly
mention a few facts that we use.

Jordan decompositions. Recall that an element g ∈ GLn (k) is semisimple if it

becomes diagonalizable after an extension of k and unipotent if g − 1 is nilpotent,
i.e., (1 − g)n = 0 for some n. If G is an affine algebraic group over a perfect field
k, then for every element g ∈ G(k), there are unique elements gs , gu ∈ G(k), called
the semisimple and unipotent parts of g, such that g = gs gu = gu gs and such that
the images of gs and gu under any representation G → GLn are semisimple and
unipotent, respectively. See [Bor91, §4], [Hum75, §15.3], [Spr98, §2.4], and [Mil17,
Thm. 9.17].

Example B.1.20 (Unipotent groups). An affine algebraic group G over a field k is

unipotent if there is a faithful representation V and a basis V ∼
= kn such that the
image of the induced map G ,→ GL(V ) ∼ = GLn is contained in the subgroup Un of

504
upper triangle matrices with 1’s along the diagonal. For example, Ga is unipotent.
We have the following equivalences:
G has a filtration 1 = G0 ⊴ G1 ⊴ · · · ⊴ Gn = G
G is unipotent ⇐⇒
of normal subgroups with Gi /Gi−1 ∼ = Ga
⇐⇒ V G ̸= 0 for every nonzero representation V
⇐⇒ every element g ∈ G is unipotent, i.e., g = gu .

Every unipotent group is isomorphic to affine space as a scheme. See [Bor91, §4.8],
[Hum75, §17.5], [Spr98, §2.4], and [Mil17, §14].

B.1.4 One-parameter subgroups, centralizers, and parabolics

Definition B.1.21 (One-parameter subgroups and characters). If G is an algebraic
group over a field k, a one-parameter subgroup (also called a cocharacter ) is a
homomorphism λ : Gm → G of algebraic groups (which is not required to be a
subgroup). A character is a homomorphism χ : G → Gm .
Despite the terminology, we do not require that a one-parameter subgroup is
injective, e.g., Gm → Gm , t 7→ t2 is a one-parameter subgroup. We let X∗ (G) be the
set of one-parameter subgroups and X∗ (G) be the group of characters. Since any
character Gm → Gm is given by t 7→ td for some d ∈ Z, there is a pairing

⟨−, −⟩ : X∗ (G) × X∗ (G) → X∗ (Gm ) ∼

= Z, (λ, χ) 7→ χ ◦ λ.

Example B.1.22 (Tori). If T ∼ = Gnm is an n-dimensional torus, then any one-

parameter subgroup λ : Gm → T is given by t 7→ (tλ1 , · · · , tλn ) for integers λi while
a character of T is given by (t1 , . . . , tn ) 7→ tχ1 1 · · · tχnn for integers χi . We thus have
bijections X∗ (T ) ∼
= Zn and X∗ (T ) ∼= Zn such that ⟨−, −⟩ : X∗ (T ) × X∗ (T ) → Z is the
standard inner product.
Example B.1.23 (GLn ). Every one-parameter subgroup λ is contained in a maximal
torus, and since maximal tori are conjugate (Proposition B.1.19), there exists
g ∈ G(k) such that gλg −1 is contained in the maximal torus consisting of diagonal
matrices.
Definition B.1.24 (Centralizers, parabolics, and unipotents). Given a one-parameter
subgroup λ : Gm → G of an algebraic group, we define the subgroups:
Cλ = {g ∈ G | λ(t)g = gλ(t) for all t} (centralizer)
Pλ = {g ∈ G | limt→0 λ(t)gλ(t)−1 exists} (parabolic)
Uλ = {g ∈ G | limt→0 λ(t)gλ(t)−1 = 1} (unipotent).

More precisely, there is a subgroup Cλ (resp., Pλ , Uλ ) of G which represent the

functor assigning a k-algebra R to the subgroup of elements g ∈ G(R) such that
λR = g −1 λR g (resp., limt→0 λR (t)gλ(t)−1 exists, limt→0 λ(t)gλR (t)−1 = 1). Note
that, by definition, the limit of λR (t)gλR (t)−1 exists as t → 0 if the natural map
Gm,R → G, t 7→ λR (t)gλR (t)−1 extends to A1R → G, and the limit is the composition
0
Spec R ,→ A1R → G.
Under the conjugation action of λ on G, Cλ is precisely the fixed locus, while Pλ
is the attractor locus G+
λ as defined in §6.8.1. There is a homomorphism Pλ → Cλ
defined by g 7→ limt→0 λ(t)gλ(t)−1 which is the identity on Cλ . This yields a split
short exact sequence
1 → Uλ → Pλ → Cλ → 1.

505
Example B.1.25 (GLn ). Let λ : Gm → GLn be a one-parameter subgroup. After
a change of basis, we can assume that λ(t) = diag(tλ1 , · · · , tλn ) with λ1 ≤ · · · ≤ λn .
Given (gij ) ∈ GLn , λ(t)(gij )λ(t)−1 = (tλi −λj gij ). If n1 , . . . , ns are integers with
i ni = n such that
P

λ1 = · · · = λn1 < λn1 +1 = · · · = λn1 +n2 < · · · < λn−ns +1 = · · · = λn ,

then Cλ = GLn1 × · · · × GLns is the subgroup of block diagonal matrices while
Pλ is the subgroup of block upper triangular matrices. For example, if λ(t) =
(t−1 , t2 , t2 , t7 ), then
     
1 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 0 0 0
0 1 0 ∗ 0 ∗ ∗ ∗ 0 ∗ ∗ 0
Uλ =  0 0 1 ∗ , Pλ = 0 ∗ ∗ ∗ , and Cλ = 0 ∗ ∗ 0 .
    

0 0 0 1 0 0 0 ∗ 0 0 0 ∗
We record the following properties of parabolic subgroups. Reductive groups are
defined and discussed in §B.1.6.
Proposition B.1.26. Let G be a connected reductive algebraic group over an
algebraically closed field k, and let λ : Gm → G be a one-parameter subgroup.
(a) The centralizer Cλ is connected and reductive.
(b) The subgroup Pλ is connected and parabolic, i.e., G/Pλ is projective, and
NG (Pλ ) = Pλ .
(c) The subgroup Uλ is the unipotent radical of Pλ , and it acts freely and transitively
on the set of one-parameter subgroups of Pλ which are conjugate (under Pλ )
to λ.
(d) If λ, λ′ : Gm → G are one-parameter subgroups, the intersection Pλ ∩ Pλ′
contains a maximal torus of G.
Proof. For (a)–(c), see [Spr98, §13.4], [Con14, Thm. 4.1.7 and Cor. 5.2.8]. For (d),
see [Bor91, Prop. 20.7].
Remark B.1.27 (Spherical buildings). The set of one-parameter subgroups of a
reductive group can be given the structure of an “ungainly but remarkable metric
space” (as described by Mumford in [GIT, p.55]): first introduced by J. Tits, the
spherical building is the quotient of X∗ (G) by the equivalence relation where λ ∼ ρ
if there exists g ∈ Pλ (k) such that ρ(tm ) = g −1 λ(tm )g for integers n, m.

B.1.5 Line bundles with G-actions

Definition B.1.28. If G is an algebraic group over a field k acting on a k-scheme
U via σ : G × U → U , a line bundle with a G-action (also called a G-linearization)
∼
is a line bundle L on U together with an isomorphism α : σ ∗ L → p∗2 L satisfying the
cocycle condition p∗23 α ◦ (id ×σ)∗ α = (m × id)∗ α, i.e., the diagram
(id ×σ)∗ α
(σ ◦ (id ×σ))∗ L / (p2 ◦ (id ×σ))∗ L

(σ ◦ (m × id))∗ L (σ ◦ p23 )∗ L
(m×id)∗ α
p∗
( w 23 α

(p2 ◦ (m × id))∗ L ∗
(p2 ◦ p23 ) L

506
commutes.
When U is projective, a G-action on a very ample line bundle L corresponds
to a finite dimensional G-representation V = H0 (U, L) and a G-equivariant closed
immersion U ,→ P(V ). The cocycle condition in the diagram above is analogous to
the cocycle condition in Fpqc Descent for Quasi-Coherent Sheaves (2.1.4). As line
bundles on algebraic stacks are defined in terms of fpqc descent, a line bundle on
[U/G] is precisely a line bundle on U with a G-action.
Example B.1.29. Under the PGLn+1 and SLn+1 action on Pn , the line bundle
O(1) admits an action by SLn+1 but not by PGLn+1 . However, O(n + 1) does admit
an action by PGLn+1 .
Theorem B.1.30 (Sumihiro’s Theorem on Linearizations). Let G be a smooth,
connected, and affine algebraic group over an algebraically closed field k. Let U be a
normal scheme of finite type over k with an action of G.
(1) If L is a line bundle on U , there exists an integer n > 0 such that L⊗n admits
a G-action.
(2) If U is quasi-projective, there exists a locally closed embedding U ,→ P(V )
where V is a finite dimensional G-representation.
(3) Every point u ∈ U has a G-invariant quasi-projective open neighborhood.
Moreover,(1) holds more generally if G → S is a smooth affine group scheme with
connected fibers (e.g., G = GLn,S ) and U is a normal noetherian scheme.
Proof. For (1), see [Sum74, Thm. 1], [Sum75, Lem. 1.2], and [KKLV89, Prop. 2.4].
Part (2) is a direct consequence of (1). For (3), see [Sum74, Lem. 8] and [Sum75,
Thm. 3.8].
When G is a torus, there is a G-invariant affine cover.
Theorem B.1.31 (Sumihiro’s Theorem on Torus Actions). Let U be a normal
scheme of finite type over an algebraically closed field k with an action of a torus T .
Then any point u ∈ U has a T -invariant affine open neighborhood.
Proof. See [Sum74, Cor. 2] and [Sum75, Cor. 3.11].
Remark B.1.32. Theorems B.1.30 and B.1.31 can fail if U is not normal, e.g., the
plane nodal cubic curve has a Gm -action and no Gm -invariant neighborhood of the
origin can be embedded Gm -equivariantly into projective space. There is nevertheless
a Gm -equivariant étale affine neighborhood Spec k[x, y]/(xy) → U (where x and y
have weights 1 and −1). In fact, every non-normal scheme (and even algebraic space)
with a Gm -action admits such an étale neighborhood (see Theorem 6.7.24).

B.1.6 Reductivity
Linearly reductive groups are used in the development of Geometric Invariant Theory
(GIT) in Chapter 7. In characteristic p, there are three distinct properties—linear
reductive, reductive, and geometrically reductive—of algebraic groups:
char=0 G smooth
wo &.
linearly reductive +3 geometrically reductive ks reductive. (B.1.33)

Linear reductive groups are very restrictive in characteristic p: it is a theorem

of Nagata [Nag62] that a smooth algebraic group G in characteristic p is linearly

507
reductive if and only if the connected component G0 is a torus and the order of G/G0
is prime to p. While it is not much more difficult to develop GIT for geometrically
reductive groups (see Remark 6.5.12), it is easier for students to first learn the theory
in the context of linear reductive groups.
Linear reductive groups. We denote by Rep(G) the category of representations of
an algebraic group G. If V is a G-representation with coaction σ : V → Γ(G, OG )⊗V ,
then the invariants are V G := {v ∈ V | σ(v) = 1 ⊗ v}. A representation V of G is
irreducible if every subrepresentation W ⊆ V is either 0 or V .

Definition B.1.34. An affine algebraic group G over a field k is linearly reductive

if the functor Rep(G) → Vectk , taking a G-representation V to its G-invariants V G ,
is exact.

Proposition B.1.35. Let G be an affine algebraic group over a field k. The following
are equivalent:
(1) G is linearly reductive;
(1′ ) The functor Repfd (G) → Vectk , V 7→ V G , on the category of finite dimensional
representations is exact;
(2) Every G-representation (resp., finite dimensional G-representation) is a direct
sum of irreducible representations.
(3) Given a G-representation (resp., finite dimensional G-representation) V and
a G-invariant subspace W ⊆ V , there exists a G-invariant subspace W ′ ⊆ V
such that V = W ⊕ W ′ .
(4) For every finite dimensional representation V and fixed k-point x ∈ P(V )G ,
there exists a G-invariant linear function f ∈ Γ(P(V ), O(1))G such that f (x) ̸=
0.

Proof. Condition (4) translates to: for every surjection V → k onto the trivial
representation, there exists f ∈ V G mapping to a nonzero element. To see that
this implies (1), let π : V → W be a surjection of G-representations and w ∈ W G .
By apply Proposition B.1.17 to π −1 (⟨w⟩), there is a nonzero finite dimensional
G-representation V ′ ⊆ V surjecting onto ⟨w⟩ and (4) implies that there is an element
v ∈ V ′G ⊆ V G mapping to w. We conclude that (1) ⇔ (1′ ) ⇔ (4).
Denote the finite dimensional conditions in (2) and (3) as (2′ ) and (3′ ). The
implications (2) ⇒ (2′ ) ⇒ (4) and (3) ⇒ (3′ ) ⇒ (1′ ) are easy. As (3) ⇒ (2) is also
clear, it suffices to show that (1) ⇒ (3). To this end, applying the exact functor

MorRep(G) (V /W, −) = MorRep(G) (k, (V /W )∨ ⊗ −) = ((V /W )∨ ⊗ −)G

to the exact sequence 0 → W → V → V /W → 0 implies that V → V /W has a

G-invariant section.

Remark B.1.36. With the terminology introduced in §6.5, G is linearly reductive if

and only if BG → Spec k is cohomologically affine or, equivalently, a good moduli
space.
For a field extension k → k′ , G is linearly reductive if and only if Gk′ is;
this is easy to see directly but also follows from the more general statement of
Lemma 6.5.17. Linear reductive groups are closed under quotients and extensions
(Proposition 6.5.19).

508
Example B.1.37 (Diagonalizable groups). Since every representation of a diago-
nalizable group scheme is a direct sum of one dimensional representations (Proposi-
tion B.1.15), every diagonalizable group scheme is linearly reductive.

Proposition B.1.38 (Maschke’s Theorem). Let G be a finite abstract group viewed

as a finite group scheme over a field k. If the order of G is prime to char(k), then
G is linearly reductive.

Proof. If V is a G-representation, averaging over translates gives a G-equivariant

k-linear map
1 X
RV : V → V G , v 7→ g · v, (B.1.39)
|G|
g∈G

which is the identity on V G . These maps are functorial with respect to maps
f : V → W of G-representations, i.e., RW ◦ f = f ◦ RV . It follows that a surjection
V → W of G-representations induces a surjection V G → W G on invariants.

Example B.1.40 (Z/pZ). In characteristic p, G = Z/pZ is not linearly reductive.

To see this, let V bethe two-dimensional
 representation of G where a generator
1 1
acts via the matrix . The surjection V → k onto the first component is a
0 1
surjection of G-representations, but the induced map V G → k on invariants is the
zero map. Geometrically, this corresponds to an action of G on A2 = A(V ) such that
the G-fixed point (1, 0) is contained in every invariant hyperplane. Note however
that taking invariants of the pth power map Symp V → Symp k = k is surjective, or,
in other words, there is a G-invariant hypersurface not containing (1, 0).

Example B.1.41 (Ga ). Over any field k, the additive group Ga is not linearly
reductive
 a . This time, let V = k be the two-dimensional representation given by
G 2

1 t
t 7→ . The projection V → k, defined by (x, y) 7→ x, is a surjection of Ga -
0 1
representations with no complement. In this case, not only is there no Ga -invariant
hyperplane avoiding (1, 0) ∈ A(V ), there is no such Ga -invariant hypersurface.

Remark B.1.42 (Reynolds operator). The map (B.1.39) is called a Reynolds operator
for the action of G on V . If G is linearly reductive, the canonical projections
RV : V → V G are Reynold operators, i.e., k-linear maps which are the identity
on V G and compatible with maps of G-representations. For an action of G on
a k-scheme Spec A with dual action A → Γ(G, OG ) ⊗ A, there is a projection
RA : A → AG . This is not a ring map, but since multiplication AG ⊗ A → A is a
map of G-representations commuting with the Reynold operators, we have that

RA (xy) = xRA (y) for x ∈ AG , y ∈ A.

This is called the Reynolds identity and implies that RA : A → AG is an AG -module

homomorphism.
In Remark 6.5.10, the Reynolds operator was applied to show that AG is finitely
generated whenever A is. While we use the exactness of the invariant functor to
prove the properties of affine GIT quotients in Corollary 6.5.8, the Reynolds operator
can also be used; see [GIT, §1.2]. Moreover, an effective method to establish linearly
reductivity is to construct a Reynolds operator. This was the proof technique in
Maschke’s Theorem (B.1.38) and it will be used again in Theorem B.1.43.

509
Reductive groups. A smooth affine algebraic group G over an algebraically closed
field k is called reductive if every smooth, connected, unipotent, and normal subgroup
of G is trivial.2 Over C, an G is reductive if and only if it is the complexification
of any maximal compact subgroup [Hoc65, XVII.5]. Over an arbitrary field k,
G is called reductive if Gk is. Reductive groups are a particularly nice class of
algebraic groups appearing in many branches of mathematics. They admit an
explicit classification in terms of their root data. See [Bor91, Hum75, Spr98, Mil17].
For a smooth affine algebraic group G, there are subgroups R(G) and Ru (G) of
G, called the radical and unipotent radical, which are maximal among connected,
normal, and solvable (resp., connected, normal, and unipotent) subgroups, which
commute with separable field extensions. Over a perfect field k, G is reductive if
and only if Ru (G) is trivial, and the quotient G/Ru (G) is reductive. On the other
hand, G is defined to be semisimple if R(G) is trivial. For a reductive group G, the
center Z(G) is diagonalizable and contains R(G) as its largest subtorus, and the
quotient G/R(G) is semisimple.
The classical algebraic groups GLn , PGLn , SLn , and SP2n are reductive in every
characteristic. As we develop GIT for actions by linearly reductive groups, it is
imperative to know that these groups are linearly reductive in characteristic 0.
Theorem B.1.43. In characteristic 0, a reductive algebraic group is linearly reduc-
tive. The converse is true in every characteristic for smooth algebraic groups.
Proof. In [Hil90], Hilbert established the linearly reductivity for SLn and GLn
over C using a explicit differential operator well-known to 19th century invariant
theorists: the Ω-process. We will sketch the argument for G = GLn over C. Write
Γ(GLn , OGLn ) = C[Xij ]det . Let V be a finite dimensional GLn -representation such
that the scalar matrices act with weight k, and let σ : Sym∗ V → C[Xij ]det ⊗ Sym∗ V
be the dual action on A(V ) = Spec Sym∗ V . The differential operator
 
∂
Ω := det
∂Xij
acts linearly on C[Xij ]det and C[Xij ]det ⊗ Sym∗ V . One checks that the map
1
V → V GLn ,
Ωk det(Xij )k σ(f )


f 7→
Ωk det(Xij )k

defines a Reynolds operator, which implies that GLn is linearly reductive. The
argument is algebraic and works over every field of characteristic 0. See [Stu08, §4.3],
[Dol03, §2.1] and [DK15, §4.5.3].
Extending an integral procedure developed by Hurwitz, Schur, and Cartan, Weyl
proved that every reductive algebraic group over C is linearly reductive [Wey26,
Wey25]. The technique is now referred to as ‘Weyl’s unitarian trick’. A compact Lie
group K has a left K-invariant finite measure µ, called the left Haar measure. For a
finite dimensional K-representation V , averaging gives a k-linear map
Z
1
V → V K , v 7→ R (g · v)dµ(g)
K
dµ(g) K

constant on V K and compatible with maps of K-representations. This is a Reynolds

operator (Remark B.1.42) just as in Maschke’s Theorem (B.1.38), and implies that
2 Sometimes G is also assumed to be connected. For a reductive group scheme G → S, there is

no such ambiguity in the literature: G is smooth and affine over S with connected and reductive
geometric fibers [SGA3III , Exp. XIX, Defn. 2.7].

510
V 7→ V K is exact. For a reductive algebraic group G over C, there is a real compact
Lie subgroup K ⊆ G(C) which is dense in the Zariski topology. For example, for
GLn , K = Un is the subgroup of unitary matrices (hence the name ‘unitarian trick’).
For a finite dimensional G-representation V , there is an identification V K = V G , and
since the functor taking K-invariant is exact, so is the functor taking G-invariants.
See also [Dol03, §3.2] and [Bum13, Thm. 14.3].3
There is also an algebraic argument using the Casimir operator. First, one reduces
to the case that G is semisimple because every reductive group is an extension of a
torus by a semisimple group. Given a Lie algebra representation ρ : g → gl(V ) of G,
there is a symmetric bilinear form on g defined by ⟨x, y⟩ = Tr(ρ(x) ◦ ρ(y)). Letting
{ei } be a basis of g and {e′i } be a dualPbasis with respect to ⟨−, −⟩, the Casimir
n
operator is the g-endomorphism cV := i=1 ρ(ei ) ◦ ρ(e′i ) on V . To show that G is
linearly reductive, it suffices to find a complement of any codimension one irreducible
subspace W ⊆ V . As G is semisimple, G acts trivial on V /W and therefore so does
g. It follows that g takes V into W and therefore so does cV , i.e., cV (V ) ⊆ W . On
the other hand, since W is irreducible, cV acts on W by multiplication by a scalar
(Schur’s lemma). It follows that ker(cV ) ⊆ V is a complement of W . See also [Mil17,
Thm. 22.42], [Muk03, §4.3], [Hum78, §6.2], and [DK15, §4.5.2].
For the converse, we need to show that the unipotent radical Ru (G) of a linearly
reductive group G is trivial. Since G/Ru (G) is affine, Matsushima’s Theorem (6.5.21)
for linearly reductive groups implies that Ru (G) is linearly reductive. However, a non-
trivial unipotent group is not linearly reductive. Indeed, by the structure of unipotent
groups, it suffices to show that Ga is not linearly reductive (Example B.1.41). See
also [NM64].

Example B.1.44. The algebraic groups such as GLn , PGLn , SLn , or SP2n are not
linearly reductive in characteristic p. For example, in characteristic 2, consider the
action of SL2 acts on the space V = Sym2 (k2 ) = {Ax2 + Bxy + Cy 2 } of degree
2 binary forms. The subspace W consisting of squares L2 of linear forms is a
GL2 -invariant subspace with no complement; the quotient V → V /W = C is given
by (A, B, C) 7→ B. While there is no invariant linear function not vanishing at
(0, 1, 0), the discriminant ∆ = B 2 ∈ Sym2 V ∨ is an invariant function nonzero at
(0, 1, 0) (verifying the geometric reductivity condition given below).

Theorem B.1.45 (Matsushima’s Theorem). Let G be a reductive group over a field

k. Then a subgroup H ⊆ G is reductive if and only if G/H is affine.

Proof. See Proposition 6.5.21 for a proof when G is linearly reductive. The general
case can be proven in a similar way relying on a generalization of Serre’s Criterion
for Affineness (4.5.16): an algebraic space U is affine if for every surjection A → B
of OU -algebras, every global section of B has a positive power that lifts to a global
section of A. See also [Mat60], [BB63], [Ric77], [FS82], and [Alp14, Thm 9.4.1].

Geometrically reductive groups. An affine algebraic group G over a field

k of characteristic p ≥ 0 is called geometrically reductive if for every surjection
n
V → W of G-representations andn w ∈ W G , there exists n > 0 such that wp is
pn
in the image of Sym V → Sym W . This condition translates to the geometric
p

3 By the limit methods of §B.3, this analytic argument suffices to show the linear reductivity of

a reductive group G over every characteristic 0 field: by limit methods (§B.3), there is a subfield
k′ ⊆ k of finite transcendence degree over Q and a group scheme G′ → Spec k′ such that G′k = G.
Choosing an embedding k′ ,→ C and using that reductivity and linear reductivity are insensitive to
separable field extensions, we see that if G′C is linearly reductive, so is G.

511
property analogous to Proposition B.1.35(4): for a fixed k-point x ∈ P(V )G , there
is an invariant homogenous polynomial f ∈ Γ(P(V ), O(pn ))G for n > 0 such that
f (x) ̸= 0. In characteristic 0, linear reductivity is equivalent to geometric reductivity.
In an effort to extend GIT to actions by reductive groups such as SLn and GLn
in positive characteristic, Mumford conjectured in [GIT, preface] a reductive group
is geometrically reductive. This conjecture was resolved by Haboush.

Theorem B.1.46 (Haboush’s Theorem). A reductive group G over a field k is

geometrically reductive.

Proof. See [Hab75]. See also [SS11].

The converse is true when G is smooth. In fact, an affine algebraic group G is

geometrically reductive if and only if Gred is reductive. A smooth algebraic group
G in characteristic p is linearly reductive if and only if the connected component
G0 is a torus and the order of G/G0 is prime to p [Nag62]. Every finite (possibly
non-reduced) group scheme G is geometrically reductive, and is linearly reductive if
and only if G0 is diagonalizable and G/G0 has order prime to p [HR15, Thm. 1.2].
A commutative algebraic group G is reductive if and only if it is diagonalizable. We
also point out that reductivity of a smooth algebraic group G is characterized by
the condition that the ring of invariants is finitely generated for every coaction on a
finitely generated k-algebra.

B.1.7 Principal G-bundles

A principal G-bundle is an algebraic version of a topological fiber bundle P → T
where G acts freely and transitively on P with quotient T = P/G, e.g., A2 ∖ 0 → P1
is a principal Gm -bundle. Principal G-bundles and their properties are essential
in the development of the theory of algebraic stacks. For instance, we define an
object of a quotient stack [U/G] over a scheme T as a principal G-bundle P → T
together with a G-equivariant map P → U . The reader may consult [Poo17, §5.12]
and [Bal09] for additional background on principal G-bundles.

Definition B.1.47. Let G → S be an fppf affine group scheme. A principal

G-bundle over an S-scheme X is a scheme P over X with an action of G via
σ : G ×S P → P such that P → X is a G-invariant fppf morphism (where X has
the trivial action) and

(σ, p2 ) : G ×S P → P ×X P, (g, p) 7→ (gp, p)

is an isomorphism.

Observe that a principal G-bundle over X is the same data as a principal G ×S X-

bundle over X. Morphisms of principal G-bundles are G-equivariant morphisms
of schemes. A principal G-bundle P → X is trivial if there is an G-equivariant
isomorphism P ∼ = G ×S X, where G acts on G ×S X via multiplication on the first
factor.
A principal G-bundle P → X can also be viewed as a G-torsor,(Definition 6.4.1),
which is a general concept for a sheaf P of sets on a site with a free and transitive
action of a sheaf G of groups. When G → S is an fppf affine group scheme, there
is an equivalence of categories between principal G-bundles and G-torsors; see
Example 6.4.5. In these notes, we will always distinguish between these two notions,
but in conversation or the literature, they are often conflated.

512
Exercise B.1.48. Show that a morphism of principal G-bundles is necessarily an
isomorphism.

Principal G-bundles can be trivialized fppf locally, and even étale locally if G is
smooth.

Proposition B.1.49. Let G → S be an fppf affine group scheme and P → X

be a G-equivariant morphism of S-schemes where X has the trivial action. Then
P → X is a principal G-bundle if and only if there exists an fppf morphism X ′ → X
such that P ×X X ′ is isomorphic to the trivial principal G-bundle G ×S X ′ over X ′ .
Moreover, if G → S is smooth, we can arrange that X ′ → X is surjective and étale.

Proof. The (⇒) direction follows from the definition by taking X ′ = P → X. For
(⇐), after base changing G → S by X → S, we may assume that G is defined over
X. Let GX ′ and PX ′ be the base changes of G and P along X ′ → X. The base
change of the action map (σ, p2 ) : G ×X P → P ×X P along X ′ → X is the action
map GX ′ ×X ′ PX ′ → PX ′ ×X ′ PX ′ of GX ′ acting on PX ′ over X ′ , which is trivial
as PX ′ is trivial. Since the property of being an isomorphism is an Fpqc Local
Property on the Target (2.1.26), we conclude that (σ, p2 ) : G ×X P → P ×X P is an
isomorphism. If G is smooth, then X ′ = P → S is a surjective smooth morphism
such that PX ′ is trivial. Since there is a section of X ′ → S after a surjective étale
morphism S ′ → S (Corollary A.3.5), PS ′ is also trivial.

Examples of principal G-bundles.

Exercise B.1.50. Let L/K be a finite Galois extension and G = Gal(L/K) be its
Galois group viewed as a finite group scheme over Spec K. Show that Spec L →
Spec K is a principal G-bundle.

Exercise B.1.51 (Line bundles vs Gm -bundles). If X is a scheme, show that there

is a covariant equivalence of categories
∼
{line bundles on X} → {principal Gm -bundles on X}
L 7→ A(L∨ ) ∖ X = Spec(Sym∗ L∨ ) ∖ zero section

between the groupoids of line bundles on X (where the only morphisms allowed are
isomorphisms) and principal Gm -bundles on X.

Exercise B.1.52 (Sd -bundles). If X is a scheme and d ≥ 1, show that there there
is an equivalence of groupoids
∼
{finite, étale, and degree d covers of X } → {principal Sd -bundles over X}
(Y → X) 7→ (Y ×X · · · ×X Y ∖∆ → X)
| {z }
d times
(P/Sd−1 → X) ←[ (P → X).

For the rightward map, the symmetric group Sd acts on the d-fold fiber product
Y ×X · · · ×X Y by permutation, and ∆ denotes the big diagonal, i.e., the Sd -
equivariant closed locus of d-tuples where at least two points coincide. Alternatively,
Y ×X · · · ×X Y ∖ ∆ can be identified with the scheme IsomX (X × {1, . . . , d}, Y )
parameterizing isomorphisms between the trivial degree d cover and Y . For the
leftward map, P/Sd−1 denotes the quotient of the free action by the subgroup
Sd−1 ⊆ Sd fixing the dth index.

513
Exercise B.1.53.
(a) Show that the standard projection An+1 ∖ 0 → Pn is a principal Gm -bundle.
(b) For each line bundle O(d) on Pn , explicitly determine the corresponding
principal Gm -bundle. In particular, which O(d) correspond to the principal
Gm -bundle of (a)?
Exercise B.1.54. Let G → S be an fppf affine group scheme.
(a) For principal G-bundles P and Q over an S-scheme X, show that the functor

IsomX (P, Q) : Sch/X → Sets,

assigning a X-scheme T to the set of isomorphisms of the principal G-bundles

P ×X T and Q ×X T , is representable by a principal G-bundle over X.
(b) For a principal G-bundle P → X, show that AutX (P ) := IsomX (P, P ) is
isomorphic to G ×G P := (G × P )/G, where h · (g, p) = (h−1 gh, h · p).
Exercise B.1.55 (Frame bundles). Let T be a scheme and E be a vector bundle
over X of rank n.
⊕n
(a) The frame bundle FrE is the functor IsomOX (OX , E) on Sch/X, i.e

FrE : Sch/X → Sets

∼
(T → X) 7→ {trivializations OT⊕n → ET }.
⊕n
Recalling from Exercise 0.3.20that the functor HomOX (OX , E) is repre-
⊕n
sentable by the scheme H := A(H omOX (OX , E) ) = A((E ∨ )⊕n ), show
∨

that FrE is representable by the open subscheme of H defined by det(u) ̸= 0,

⊕n
where u : OH → EH is the universal homomorphism. Moreover, show that
FrE → X is a principal GLn -bundle.
(b) The projectivized frame bundle PFrE is the functor

PFrE : Sch/X → Sets

L is a line bundle on T and
 
(T → X) 7→ (L, α) ∼ ∼,
α : OT⊕n → ET ⊗ L is an isomorphism

where (L, α) ∼ (L′ , α′ ) if there is an isomorphism β : L → L′ with α′ =

(id ⊗β) ◦ α. Show that PFrE is representable by the open subscheme of
⊕n
P := P(H omOX (OX , E)∨ ) = P((E ∨ )⊕n ) defined by det(u) ̸= 0, where

u : OP⊕n → EP ⊗ OP (1)

is the homomorphism corresponding to the universal quotient H omOP (OP⊕n , EP )∨ ↠

OP (1). Moreover, show that PFrE → X is a principal PGLn -bundle.
Exercise B.1.56 (Vector bundles vs GL-bundles). For a scheme X, show that the
assignment of a vector bundle E to the frame bundle FrE defines an equivalence of
groupoids
∼
{vector bundles over X} → {principal GLn -bundles over X}
E 7→ FrW

with the inverse defined by assigning P → X to the vector bundle whose total space
is the quotient P ×GLn An := (P × An )/ GLn of the diagonal GLn -action on P × An .

514
Exercise B.1.57 (SL-bundles). Show that the groupoid of principal SLn -bundles
over a scheme X is equivalent to the groupoid of pairs (V, α) where V is a vector
∼
bundle on X of rank n and α : OX → det V is a trivialization. A morphism
(V ′ , α′ ) → (V, α) of pairs is an isomorphism ϕ : V ′ → V such that α′ = α ◦ det ϕ.
Exercise B.1.58 (Orthogonal group). Let k be a field with char(k) ̸= 2, and let
V be an n dimensional vector space with a non-degenerate quadratic form q. Let
O(q) ⊆ GL(V ) be the subgroup of invertible matrices preserving the quadratic form.
If q = x21 + · · · + x2n is the diagonalized quadratic form, O(q) = On is the set of
orthogonal matrices A (i.e., AA⊤ = I). Show that there is a bijection between
principal O(q)-bundles over a k-scheme X and vector bundles of rank n on X with
a non-degenerate quadratic form.
Example B.1.59 (Special groups). An affine algebraic group G over a field k is
special if every principal G-bundle P → T is Zariski locally trivial. For example,
GLn (e.g., Gm = GLn ) is a special group as principal GLn -bundles correspond to
vector bundles (Exercise B.1.56). It is also true that Ga is special. One way to
see this is to use that quasi-coherent cohomology can be computed either in the
Zariski or small étale site (Proposition 4.1.43). Viewing the structure sheaf OT as
Ga , this implies that H1 (T, Ga ) = H1 (Tét , Ga ) and it follows that there is a bijective
correspondence between Ga -torsors in the Zariski and small étale topology.
If 1 → K → G → Q → 1 is an exact sequence of affine algebraic groups and
both K and Q are special, it is not hard to see that G is also special. It follows from
the characterization of unipotent groups as extensions of Ga that every unipotent
group is special.

Inducing G-bundles and reduction of structure group.

Definition B.1.60 (Induced G-bundles). Let H → G be a homomorphism of fppf
affine groups schemes over a scheme S. If P → X is a principal H-bundle, the
principal G-bundle induced by P via H → G is

G ×H P → X,

where G ×H P is the quotient (G × P )/H of the action h · (g, p) = (gh−1 , hp), and
where G acts on G ×H P via g ′ · (g, p) = (g ′ g, p).
Exercise B.1.61. Verify that G ×H P → X is a principal G-bundle.
Definition B.1.62 (Reduction of structure group). Let H → G be a homomorphism
of fppf affine groups schemes over a scheme S. If Q → X is a principal G-bundle, a
reduction of structure group of Q by H → G is a principal H-bundle P → X and an
∼
isomorphism Q → G ×H P of principal G-bundles.
Lemma B.1.63. Let H → G be a monomorphism of fppf affine groups schemes
over a scheme S, and let Q → X be a principal G-bundle. A reduction of structure
group of Q by H → G is equivalent to giving a section of Q/H over X.
Proof. A section s : X → Q/H induces a principal H-bundle P → X via pullback

P /Q

□
  
X / Q/H,
s

515
and the induced map G ×H P → Q, defined by (g, p) 7→ g · p, is an isomorphism of
∼
principal G-bundles. Conversely, by precomposing an isomorphism G ×H P → Q
with the H-equivariant inclusion P → G ×H P , given by p 7→ (id, p), defines
an H-equivariant map P → Q which descends under the H-action to a section
X = P/H → Q/H.

Isomorphism classes of principal G-bundles over a scheme X is classified by the

étale cohomology group H1 (Xét , G); see Exercise 6.4.30.

Exercise B.1.64. Let 1 → K → G → Q → 1 be an exact sequence of abstract

abelian groups, which induces a short exact sequence

H1 (Xét , K) → H1 (Xét , G) → H1 (Xét , Q)

of étale cohomology groups over a scheme X. Show that a principal G-bundle

P → X admits a reduction of structure group to K if and only if the class of
[P ] ∈ P ∈ H1 (Xét , G) maps to 0 ∈ H1 (Xét , Q).

Exercise B.1.65. Let G → S be an fppf affine group scheme, and let P1 and P2
be principal G-bundles over an S-scheme X. Show that a reduction of structure
of P1 × P2 → X × X by the diagonal G → G × G corresponds to an isomorphism
P1 → P2 of principal G-bundles.

Brauer–Severi schemes and Azumaya algebras.

Exercise B.1.66 (Brauer–Severi schemes). A morphism P → X of schemes is a

Brauer–Severi scheme of relative dimension r if there exists an étale cover X ′ → X
and an isomorphism P ×X X ′ ∼ = PrX ′ . An example of a non-trivial Brauer-Severi
scheme is Proj R[x, y, z]/(x + y 2 + z 2 ) → Spec R. Show that
2

{Brauer–Severi schemes of rel. dim. r over X} → {principal PGLr -bundles over X}

X 7→ IsomX (PrX , X)
r
(P × P )/ PGLr ←[ P

defines an equivalence of groupoids.

Exercise B.1.67. Let P → X be a proper, flat, and finitely presented morphism of

schemes. Assume that for every geometric point Spec k → X, the geometric fiber
P ×X k is isomorphic to P1k . Show that P → X is a Brauer–Severi scheme of relative
dimension 1.
Approach 1 (local-to-global): Show that for every point x ∈ X, there is a finite and
separable field extension κ(s) → K such that P ×X K ∼ = P1K . Then show that there
an étale neighborhood (X ′ , x′ ) → (X, x) such that K ∼= κ(x′ ) over κ(x). Assuming
now that P ×s κ(s) ∼ P 1
= κ(s) , use deformation theory (Proposition C.2.4) to show
n ∼ 1
that there are compatible isomorphisms P ×X OX,x /mx = POX,x /mn for n > 0. Use
x

Grothendieck’s Existence Theorem (C.5.3) to show that P ×X O bX,x ∼

= P1 . Finally,
O
bX,x
apply Artin Approximation (B.5.18) to show that there is an étale neighborhood
(X ′ , x′ ) → (X, x) such that P ×X X ′ ∼
= P1X ′ .
Approach 2 (direct): Assuming that there is a section σ : X → P of π : P → X, show
that every point x ∈ X has an open neighborhood U ⊆ X such that P ×X U ∼ = P1U .
Letting L be the line bundle on P corresponding to the Cartier divisor σ, use

516
Cohomology and Base Change (A.6.8) to show that E := π∗ L is a rank 2 vector
bundle on X, that π ∗ E ↠ L is surjective, and that P ∼ = P(E) over X. Conclude by
choosing an open neighborhood of x ∈ X where E is trivial. Returning to the general
case, show that there is an effective divisor D associated to Ω∨
P/X such that D → X
is étale. Reduce to the case where P → X has a section by base changing by D → X.
See also [Har77, Prop. 25.3 and Exc. 25.2].

Exercise B.1.68 (Azumaya algebras). An Azumaya algebra of rank r2 over a scheme

X is a (possibly non-commutative) associative OX -algebra A, which is coherent
as an OX -module, such that there is an étale covering X ′ → X with A ⊗OX OX ′
isomorphic to the matrix algebra Mr (OX ); see [Mil80, §IV.2]. An Azumaya algebra
over a field k is a central simple algebra (i.e., a finite dimensional associative k-
algebra which is simple and whose center is k); the quaternions defines a central
simple algebra over R. Show that the assignment

A 7→ IsomX (Mr (OX ), A)

defines a bijection between Azumaya algebras of rank r2 over X and PGLn -torsors
over X.

Remark B.1.69. Exercises B.1.66 and B.1.68 provide bijections

{Azumaya algebras of rank r2 } ≃ {principal PGLn -torsors}

≃ {Brauer–Severi schemes of relative dimension r}

on sets of isomorphism classes of objects over a scheme X. The composition of these

bijections can be interpreted as the map taking an Azumaya algebra A over X to the
Brauer–Severi scheme defined as the closed subscheme X ⊆ GrX (r, A) classifying
rank r right ideals. The Brauer group Br(X) is defined in terms of Azumaya algebras
and closely related to Gm -gerbes; see Remark 6.4.37 and Exercise 6.4.38.
In this appendix, we cover several important topics needed in the development
of moduli theory that are not always covered in a first course in algebraic geometry.

B.2 Birational geometry and positivity

We summarize basic facts of birational geometry needed in our development of moduli
theory. In Stable Reduction (5.5.1) for M g , the birational geometry of surfaces in
§B.2.1 is used crucially . To prove Kollár’s Criterion for Ampleness (5.9.2), which we
apply to prove that M g is projective, we appeal to the Nakai–Moishezon Criterion
for Ampleness (B.2.28) for ampleness and properties of nef vector bundles in §B.2.4.

B.2.1 Birational geometry of surfaces

For an integral noetherian scheme X, a resolution of singularities is a proper
birational morphism X ′ → X from an integral regular scheme X ′ .

Theorem B.2.1 (Existence of Resolutions). Every two dimensional integral noethe-

rian scheme X has a resolution of singularities.

Proof. This was shown by Zariski in characteristic 0 [Zar39], by Abhyankar in

characteristic p [Abh56], and by Lipman in mixed characteristic [Lip78]. See also
[Kol07, §2] and [SP, Tag 0BGP].

517
Theorem B.2.2 (Existence of Minimal Resolutions). Let X be a two dimensional
e →X
integral noetherian scheme. There exists a resolution of singularities π : X
such that every other resolution of singularities Y → X factors as Y → X → X.
e
Moreover, KXe · E ≥ 0 for every π-exceptional curve E.

Proof. See [Kol07, Thm. 2.16].

Theorem B.2.3 (Existence of Embedded Resolutions). Let X be a regular scheme

of dimension 2 and Y ⊆ X be a subscheme of pure dimension one. Assume that for
every irreducible component Z ⊆ Y , the normalization Ze → Z is finite. Then there
is a finite sequence of blowups

Xn → · · · → X1 → X

at reduced closed points such that the preimage Yn ⊆ Xn of Y is an effective Cartier

divisor supported on a normal crossings divisor, i.e., (Yn )red is nodal.

Proof. See [Har77, Thm V.3.9], [Kol07, Thm. 1.47], and [SP, Tag 0BIC].

Theorem B.2.4 (Factorization of Birational Maps). Let X and Y be regular,

integral, and noetherian schemes of dimension two. Every proper birational morphism
f : X → Y is the composition of blowups at reduced closed points.

Proof. See [Har77, Thm V.5.5], [Kol07, Thm 2.13], and [SP, Tag 0C5R].

Theorem B.2.5 (Hodge Index Theorem for Exceptional Curves). Let f : X → Y be

a projective and generically finite morphism of noetherian schemes of dimension 2,
where X is regular and Y is quasi-projective over a field or DVR. Let E1 , . . . , En be
the exceptional curves. Then the intersection form matrix (Ei ·Ej ) is negative-definite.
In particular, Ei2 < 0 for each i.

Proof. See [Kol07, Thm 2.12].

Theorem B.2.6 (Castelnuovo’s Contraction Theorem). Let X be a regular scheme

of dimension 2 which is projective over either a field k or a DVR R with residue
field k, and let E = P1k ⊆ X be a smooth rational curve with E 2 < 0. Then there
is a projective morphism X → Y to a projective surface and a point y ∈ Y such
that f −1 (y) = E and X ∖ E → Y ∖ {y} is an isomorphism. If E 2 = −1, then Y is
smooth.

Proof. See [Har77, Thm. V.5.7, Exc. V.5.2] and [Kol07, Thm. 2.14, Rmk. 2.15].

One can show that the process of repeatedly contracting smooth rational −1
curves in a smooth projective surface terminates (see [Har77, Thm 5.8]). Thus
by applying Castelnuovo’s Contractibility Criterion a finite number of times, one
obtains:

Corollary B.2.7 (Existence of Minimal Models). A smooth surface X admits

a projective birational morphism X → Xmin to a smooth surface such that every
projective birational morphism Xmin → Y to a smooth surface is an isomorphism.
In particular, Xmin has no smooth rational −1 curves.

518
B.2.2 Positivity
We discuss positivity properties of line bundles, some of which are extended to
algebraic spaces in §5.9.2. An excellent reference for this material is [Laz04a, Laz04b].
Ample line bundles. A line bundle L on a scheme X is ample if X is quasi-compact
and for every x ∈ X, there exists a section s ∈ Γ(X, L) such that Xs = {s ̸= 0} is
affine and contains x.
Proposition B.2.8 (Characterizations of Ampleness). For a line bundle L on a
noetherian scheme X, the following are equivalent:
(1) L is ample,
Γ(X, L⊗d ) is well-defined and an open im-
L
(2) the natural map X → Proj d≥0
mersion, and
(3) for every coherent sheaf F , the tensor product F ⊗ L⊗m is base point free for
m ≫ 0.
If in addition X is proper over a noetherian ring R, then the above are also equivalent
to:
(4) for some m > 0, L⊗m is very ample, i.e., defines a closed embedding |L⊗m | : X ,→
PNR into projective space, and
(5) for every coherent sheaf F on X, Hi (X, F ⊗ L⊗m ) = 0 for m ≫ 0 and i > 0.
Proof. See [Har77, §II.7 and III.5.3], [EGA, II.4.5 and III.2.6], or [SP, Tags 01PR
and 0B5U].
Proposition B.2.9 (Properities of Ampleness). Let X be a proper scheme over a
field k and L be a line bundle on X.
(1) If f : X ′ → X is a finite surjective morphism, L is ample if and only if f ∗ L is.
(2) For a field extension k → k′ , L is ample on X if and only if Lk′ is ample on
Xk′ .
Proof. Both follow from the cohomological characterization of ampleness. For (1),
see [Har77, Exer III.5.7], [EGA, III.2.6.2], and [SP, Tag 0B5V]. Part (2) also follows
directly from Fpqc Descent for Ampleness (B.2.12).
Part (1) implies that a line bundle L on X is ample if and only if its restriction
L|(Xi )red to the reduced subscheme of each irreducible component Xi is ample.
Proposition B.2.10 (Openness of Ampleness). Let f : X → S be a proper and
finitely presented morphism of schemes, and L be a line bundle on X. If for some
s ∈ S, the restriction Ls of L to the fiber Xs is ample (resp., very ample and
Hi (Xs , Ls ) = 0 for i > 0), then there exists an open neighborhood U ⊆ S of s such
that the restriction LU on XU is relatively ample (resp., relatively very ample) over
U . In particular, for all u ∈ U , Lu is ample (resp., very ample) on Xu .
Proof. We present a proof under the condition that X → S is flat (which suffices for
many applications). If Ls is ample on Xs , then for n ≫ 0, L⊗n s is very ample and
Hi (Xs , L⊗n
s ) = 0 for i > 0. It therefore suffices to handle the very ample case. By
Cohomology and Base Change (A.6.8), after replacing S with an open neighborhood
of s, f∗ L is a vector bundle and the comparison map f∗ L ⊗ κ(t) → Hi (Xt , Lt ) is an
isomorphism for t ∈ S. By further replacing S with an affine open neighborhood,
we can arrange that H0 (X, L) is freely generated by sections t0 , . . . , tn that restrict
to a basis in H0 (Xs , Ls ). The vanishing locus V := V (t0 , . . . , tn ) ⊆ X is closed and

519
disjoint from Xs . By replacing S with an affine open neighborhood of s contained
in S ∖ f (V ), we may assume that the sections ti generate L and that they define a
morphism g : X → PnS over S that restricts to a closed immersion gs : Xs ,→ Pnκ(s) .
By upper semicontinuity of fiber dimension, there is a closed locus Z ⊆ PnS consisting
of points z such that dim g −1 (z) > 0. Since Z is disjoint from Pnκ(s) , we may shrink
S further so that g : X → PnS is quasi-finite, and hence finite (as g is proper). The
cokernel of OPnS → g∗ OX is coherent and its support is a closed subscheme of Pn
disjoint from Pnκ(s) . By shrinking S further, we may arrange that g : X → PnS is
a closed immersion, and hence L = g ∗ OPnS (1) is very ample. See also [Laz04a,
Thm. 1.2.17, Thm. 1.7.8], [EGA, III1 .4.7.1, IV3 .9.6.4], [KM98, Prop. 1.41] and [SP,
Tag 0D3D].
Example B.2.11 (suggested by Brian Nugent). It is not true that very ampleness
is an open condition. If C is a non-hyperelliptic curve of genus 3, then KC is very
ample and we can write KC = OC (p1 + · · · + p4 ). Considering the constant family
C × C → C with the constant sections p1 , p2 , and p3 and the diagonal section ∆,
then the fiber of OC×C (p1 + p2 + p3 + ∆) is very ample over p4 ∈ C but the fiber
over a point s ∈ C near to p4 is not very ample.
We will also need the fact that ampleness is an fpqc local property on the target.
Proposition B.2.12 (Fpqc Descent of Ampleness). Let f : X → S be a morphism
of schemes and L be a line bundle on X. If S ′ → S is an fpqc morphism of schemes,
then L is relatively ample over S if and only if the pullback of L to X ×S S ′ is
relatively ample over S ′ .
Proof. We know that relative ampleness is stable under base change. If the pullback
L′ to X ′ := X ×S S ′ is relatively ample, then X ′ → S ′ is quasi-compact and
quasi-separated. Since these are Fpqc Local Properties on the Target (2.1.26),
f : X → S is also quasi-compact and quasi-separated. We may therefore form
the quasi-coherent graded OS -algebra A = ⊗d
and the corresponding
L
d≥0 f∗ L
morphism X → ProjS A. We appeal to the fact that L is relatively ample if and
only if X → ProjS A is well-defined and an open
L immersion. Since X → ProjS A
base changes to X ′ → ProjS ′ A′ with A′ = d≥0 f ′ ∗ L′⊗d , the statement follows
from an open immersion being an Fpqc Local Properties on the Target (2.1.26). See
also [EGA, IV2 .2.7.2] and [SP, Tag 0D2P].

Nef line bundles. A line bundle L on a proper scheme X over a field k is nef (or
numerically effective) if Z
c1 (L) ≥ 0
C

for every integral closed curve C ⊆ X. Here C c1 (L) denote the same quantity as
R

c1 (L) · C, L · C, or deg L|C . We say that L is strictly nef if c1 (L) · C > 0 for every
integral closed curve.
Theorem B.2.13 (Kleiman’s Theorem). If L is a line bundle on a proper scheme
X over a field k, then L is nef if and only if for every integral subscheme Z ⊆ X of
dimension k, Z
c1 (L)k ≥ 0.
Z

Proof. See [Laz04a, Thm. 1.4.9], [Kol96, Thm. 2.17], or the original source [Kle66].

520
 It is often convenient to write line bundles in additive notation, so that mL + H
corresponds to L⊗m ⊗ H.

Corollary B.2.14 (Characterization of Nefness). Let X be a projective scheme

over a field k and H be an ample line bundle. A line bundle L on X is nef if and
only if mL + H is ample for m ≫ 0.

Proof. See [Laz04a, Cor. 1.4.10].

Proposition B.2.15 (Properties of Nefness). Let X be a proper scheme over a

field k and L be a line bundle on X.
(1) If f : X ′ → X is a surjective proper morphism, then L is nef if and only if
f ∗ L is.
(2) For a field extension k → k′ , L is nef on X if and only if Lk′ is nef on Xk′ .

Proof. For (1), if C ′ ⊆ X ′ is an integral curve and d is the degree of the induced
map C ′ → f (C ′ ), then
f ∗ L · C ′ = d(L · f (C ′ )) (B.2.16)
by the projection formula. This gives the (⇒) implication. Conversely, if C ⊆ X is
an integral curve, we may choose an integral curve C ′ ⊆ X ′ with C = f (C ′ ), and thus
(B.2.16) also implies the (⇐) implication. For (2), by Chow’s Lemma and (1), we may
assume that X is projective. In this case, the Characterization of Nefness (B.2.14)
reduces us to the corresponding statement for ampleness (Proposition B.2.9(2)).

Proposition B.2.17 (Nefness is Stable under Generization). Let X be a proper

flat scheme over a DVR R and L be a line bundle on X. If the restriction L0 of L
to the central fiber X0 is nef, then so is the restriction Lη to the generic fiber Xη .

Proof. By Chow’s Lemma and Proposition B.2.15(1), we may assume that X is

projective with an ample line bundle H. By the Characterization of Nefness (B.2.14),
mL0 + H0 is ample for m ≫ 0. By Openness of Ampleness (B.2.10), mL + H is
ample, and thus so is mLη + Hη . By applying again the Characterization of Nefness,
we conclude that Lη is nef.

Remark B.2.18. For a surjective proper morphism X → S of varieties and a line

bundle L on X whose fiber Ls over s ∈ S is nef, there exists a countable union
B ⊆ S of proper subschemes not containing s such that Lt is nef for every t ∈ S ∖ B
[Laz04a, Prop 1.4.14]. It is not true that nefness is open in general; see [Lan13,
Ex. 5.3] and [Les14, Thm. 1.2].
Remark B.2.19 (Ample and nef cones). The ample and nef line bundles generate
cones Amp(X), Nef(X) ⊆ N 1 (X)R , called the ample cone and nef cone. For a
projective variety, Kleiman’s Theorem (B.2.13) implies that the nef cone is the
closure of the ample cone and the ample cone is the interior of the nef cone; see
[Laz04a, Thm. 1.4.23].

Effective, base point free, and semiample line bundles. We have the following
notions for a line bundle L on a proper scheme X over a field k:

• L is effective if Γ(X, L) ̸= 0,
• L is base point free (or globally generated ) if for every x ∈ X, there exists
s ∈ Γ(X, L) with s(x) ̸= 0, or equivalent the complete linear series |L| defines
a morphism X → P(H0 (X, L)), and

521
 • L is semiample if for some m > 0, L⊗m is base point free.

A semiample line bundle L is necessarily nef; indeed if for some m > 0, L⊗m
defines a morphism f : X → PN with f ∗ O(1) ∼ = c1 (L⊗m ), then the projection
formula implies that C · L⊗m
= f (C) · c1 (O(1)) ≥ 0. We thus have the implications

base point free ⇒ semiample ⇒ nef.

Big line bundles. A line bundle L on a proper scheme X over a field k is big if
there exists a constant C > 0 such that h0 (X, L⊗m ) > C · mdim(X) for m ≫ 0.

Proposition B.2.20 (Kodaira’s Lemma). Let X be a projective scheme over a field

k and L be a big line bundle on X. If E is an effective line bundle, then mL − E is
effective for m sufficiently divisible.

Proof. See [Laz04a, Prop. 2.2.6].

Proposition B.2.21 (Characterizations of Bigness). For a projective scheme X

over a field k and a line bundle L on X, the following are equivalent:
(1) L is big,
(2) for every ample divisor A on X, there exists a positive integer m > 0 and an
effective divisor N on X such that mL = A + N (linear equivalence), and
(3) there exists an ample divisor A on X, a positive integer m > 0, and an effective
divisor N on X such that mL ≡ A + N (numerical equivalence).
If in addition X is normal, then the above are also equivalent to:
(4) for some m > 0, |L⊗m | defines a rational map X 99K P(H0 (X, L⊗m )) which
is birational onto its image.

Proof. See [Laz04a, Cor. 2.2.7].

As a consequence, we see that up to scaling (i.e., taking positive tensor powers),

a big line bundle is the same as the sum of an ample and effective line bundle. This
implies that the sum of a big and effective line bundle is also big. To summarize,
up to scaling
big ⇐=======⇒ ample + effective

big + effective ⇒ big.

Theorem B.2.22 (Asymptotic Riemann–Roch). Let X be a proper scheme over a

field k of dimension n, and let L be a line bundle on X. Then the Euler characteristic

(c1 (L)n ) n
χ(X, L⊗m ) = m + O(mn−1 )
n!

is a polynomial of degree ≤ n in m. If in addition L is nef, then hi (X, L⊗m ) =

O(mn−1 ) and
(c1 (L)n ) n
h0 (X, L⊗m ) = m + O(mn−1 ).
n!
Proof. See [Laz04a, Cor. 1.4.41] (projective case), [Kol96, Thm. VI.2.14-15], and
[SP, Tag 0BJ8].

522
 Accepting that χ(X, L⊗m ) is a polynomial, one can define the intersection
number c1 (L)n as the normalized leading coefficient. It can also be defined in several
other ways, c.f., [Kol96, Thm. VI.2]. Asymptotic Riemann–Roch provides a useful
characterization of bigness for nef line bundles, implying for instance that ample
line bundles are big.
Corollary B.2.23 (Characterization of Bigness II). Let X be a proper scheme
over a field k of dimension n. A nef line bundle L on X is big if and only if
c1 (L)n > 0.

Proposition B.2.24 (Properties of Bigness). Let X be an integral proper scheme

over a field k and L be a line bundle on X.
(1) Let f : X ′ → X be a generically quasi-finite and proper morphism of schemes.
Then L is big if and only if f ∗ L is big.
(2) For a field extension k → k′ , L is big on X if and only if Lk′ is big on Xk′ .
Proof. For (1), the projection formula implies that

H0 (X ′ , f ∗ L⊗m ) = H0 (X, f∗ f ∗ L⊗m ) = H0 (X, L⊗m ⊗ f∗ OX ′ ).

As OX ,→ f∗ OX ′ is injective, we have an inclusion H0 (X, L⊗m ) ,→ H0 (X, L⊗m ⊗

f∗ OX ′ ). Since dim X ′ = dim X, the bigness of L implies the bigness of f ∗ L.
Note that if X ′ → X is birational and X is normal, then OX = f∗ OX ′ and
H0 (X ′ , f ∗ L⊗m ) = H0 (X, L⊗m ), which gives the converse. In general, the projection
′
formula for intersection numbers implies that c1 (f ∗ L)dim X = deg(f )c1 (L)dim X .
Since L is nef if and only if f ∗ L is nef (Proposition B.2.15), the characterization of
Bigness II (B.2.24) shows that L is big if and only if f ∗ L is.
Part (2) follows from the identification H0 (X ′ , f ∗ L⊗m ) = H0 (X, L⊗m ) ⊗k k′ .
Remark B.2.25 (Big and pseudo-effective cones). Big and effective divisors generate
the big cone Big(X) and effective cone Eff(X) in N 1 (X)R , and the closure Eff(X)
is called the pseudo-effective cone. The big cone Big(X) is contained in the interior
of Eff(X), and Eff(X) = Big(X) [Laz04a, Thm. 2.2.6].

B.2.3 Ampleness criteria

We review techniques to verify ampleness of a line bundle on a proper scheme. The
first strategy to keep in mind is: semiample and strictly nef ⇒ ample.
Lemma B.2.26. On a proper scheme over a field k, a line bundle L is ample if
and only if it is strictly nef and semiample.
Proof. For the non-trivial direction, for some m > 0, L⊗m defines a morphism
f : X → PN which does not contract any curves. It follows that f : X → PN is a
proper and quasi-finite, thus finite. Therefore, L⊗m = f ∗ O(1) is ample.
Remark B.2.27. The semiampleness condition can be very challenging to verify in
practice. There are powerful base point free theorems in birational geometry that
can sometimes be applied to reduce semiampleness to bigness and nefness. For
instance, Kawamata’s base point freeness theorem states that if (X, ∆) is a proper
klt pair with ∆ effective and D is a nef Cartier divisor such that aD − KX − ∆ is nef
and big for some a > 0, then D is semiample [KM98, Thm. 3.3]. One can contrast
this result with the Abundance Conjecture that states that if (X, ∆) is a proper log

523
canonical pair with ∆ effective, then the nefness of KX + ∆ implies semiampleness
[KM98, Conj. 3.12]. Alternatively, it is a classical result of Zariski and Wilson that if
X is a normal projective variety and DL is a nef and big divisor, then D is semiample
if and only if its graded section ring n Γ(X, OX (nD)) is finitely generated; see
[Laz04a, Thm. 2.3.15]. While [BCHM10] can sometimes be applied to verify the
finite generation, this result already presumes the projectivity of X. Nevertheless,
it is sometimes useful. In positive characteristic, Keel’s theorem [Kee99] provides
another technique: on a projective variety X over a field k of charaacteristic p, a nef
line bundle L is semiample if and only if the restriction of L to the exceptional locus
E is semiample, where the exceptional locus E is defined as the union of irreducible
subvarieties Z ⊆ X satisfying Ldim Z · Z = 0.

Numerical criteria for ampleness. The Nakai–Moishezon Criterion for Ample-

ness4 for ampleness provides a convenient method to establish projectivity. We will
extend this criterion to algebraic spaces in Theorem 5.9.10, as it enters into the
proof of Kollár’s Criterion (5.9.2), which in turn is used for the projectivity of M g .
Theorem B.2.28 (Nakai–Moishezon Criterion for Ampleness). Let X be a proper
scheme over an algebraically closed field k, and let L be a line bundle on X. The
following are equivalent:
(1) L is ample;
(2) for every integral closed subscheme Z ⊆ X, c1 (L)dim Z · Z > 0;
(3) L is nef and for every integral closed subscheme Z ⊆ X, L|Z is big; and
(4) L is strictly nef and for every integral closed subscheme Z ⊆ X, L|⊗m
Z is
effective for some m > 0.
Proof. A nef line bundle L is big if and only if the top intersection is positive
(Corollary B.2.23). This gives the equivalence between (2) and (3). We therefore
have: (1) ⇒ (2) ⇐⇒ (3) ⇒ (4). For (4) ⇒ (1), since L is strictly nef, it suffices by
Lemma B.2.26 to verify that L is semiample. Write L = OX (D) for a divisor D.
Since D is big on X, some positive multiple mD is effective. After replacing D by
mD, let s ∈ H0 (X, OX (D)) be a nonzero section. Then OX (D) has no base points
on X ∖ D. We will show that for m ≫ 0, OX (mD) also has no base points on D.
By induction on dim X, we can assume that OX (D)|D is ample. Consider the exact
sequence
0 → OX ((m − 1)D) → OX (mD) → OX (mD)|D → 0.
For m ≫ 0, OX (mD)|D is base point free and H1 (X, OX (mD)|D ) = 0. It follows that
H1 (X, OX ((m − 1)D)) ↠ H1 (X, OX (mD)) is surjective, but since each vector space
is finite dimensional, we see that these surjections eventually become isomorphisms
for m ≫ 0. Thus, for m ≫ 0, H0 (X, OX (mD)) → H0 (D, OX (mD)|D ) is surjective,
and OX (mD) has no base points on D. See also [Har77, Thm. V.1.10] (surface case),
[Laz04a, Thm. 1.2.23] (projective case), [Kol96, Thm VI.2.18], [KM98, Thm. 1.37],
and [Kle66, §III.1].
While we will not apply the following criteria in this text, they are often useful
in other contexts.
Theorem B.2.29 (Kleiman’s Criterion). If X is a projective scheme or a Q-factorial
(e.g., smooth) proper scheme over an algebraically closed field k, a line bundle L on
X is ample if and only if for all C ∈ Eff(X), c1 (L) · C > 0.
4 This is also known as the Nakai Criterion or the Nakai–Moishezon–Kleiman Criterion. See

[Laz04a, §1.2.B] for a historical account and further references.

524
Proof. See [Kle66, §III.1], [Kol96, Thm. VI.2.19], [KM98, Thm. 1.18], and [Laz04a,
Thm. 1.4.23].
Note that it is not enough to check that c1 (L) · C > 0 for only integral curves
C ⊆ X; one must check it on curve classes in the closure Eff(X) of the effective
cone of curves. See [Har70, p.50-56] for a counterexample due to Mumford.
Theorem B.2.30 (Sesahdri’s criterion). If X is a proper scheme over an alge-
braically closed field k, a line bundle L on X is ample if and only if there ex-
ists an ϵ > 0 such that for every point x ∈ X and every integral curve C ⊆ X,
c1 (L) · C > ϵ multx (C), where multx (C) denotes the multiplicity of C at x.
Proof. See [Laz04a, Thm. 1.4.13] and [Kol96, Thm. 2.18].

B.2.4 Nef vector bundles

In Kollár’s Criterion for Ampleness (5.9.2), nefness of vector bundles plays an
essential role.
Definition B.2.31. A vector bundle E on a scheme X is called nef (or semipositive)
if OP(E) (1) is nef on P(E).
There is a related notion of an ample vector bundle, which we will not need,
defined by requiring OP(E) (1) to be ample on P(E); see [Har66a] and [Laz04b, §6].
Proposition B.2.32 (Characterization of Nefness for Bundles). Let E be a vector
bundle on a proper scheme X over an algebraically closed field k. Then the following
are equivalent:
(1) E is nef,
(2) for every map f : C → X from a smooth proper curve, every quotient line
bundle of f ∗ E ↠ L has nonnegative degree, and
(3) for every map f : C → X from a smooth proper curve, every quotient vector
bundle f ∗ E ↠ W has nonnegative degree.
Proof. See [Bar71, p.437], [Laz04b, Prop. 6.1.18], and [Kol90, Def.-Prop. 3.3]
Proposition B.2.33 (Properties of Nefness for Bundles). Let X be a proper scheme
over a field k and E be a vector bundle on X.
(1) If f : X ′ → X is a surjective proper morphism, then E is nef if and only if
f ∗ E is.
(2) For a field extension k → k′ , E is nef on X if and only if Ek′ is nef on Xk′ .
(3) Quotients, extensions, and tensor products of nef vector bundles are nef. If E
Vk
is nef, so is E, Symk E, and Symk (E ∨ )∨ for k ≥ 0.
Proof. Parts (1) and (2) follow from the analogous properties of nef line bundles
(Proposition B.2.15). Part (3) requires some work. By (2), we can assume that k is
algebraically closed. From the Characterization of Nefness for Bundles (B.2.32), it
suffices to assume that X is a smooth curve of genus g. This characterization makes
it clear that quotients of nef bundles are nef, and it is not hard to show that an
extension of nef bundles is nef.
Before we prove the remaining parts, we claim that if E is a nef vector bundle on X
and L is a line bundle with deg L ≥ 2g −1 (resp., deg L ≥ 2g), then H1 (X, E ⊗L) = 0
(resp., E ⊗ L is globally generated). By Serre–Duality (5.1.3), H1 (X, E ⊗ L) =

525
HomOX (E ⊗ L, ωX ). If E ⊗ L → ωX is a nonzero map, the image I ⊆ ωX is a
line bundle with deg I ≤ deg ωX = 2g − 2. If deg L ≥ 2g − 1, the induced quotient
E ↠ I ⊗ L∨ would have negative degree, contradicting the nefness of E, hence
H1 (X, E ⊗ L) = 0. If deg L ≥ 2g, then for every p ∈ X(base), H1 (X, E ⊗ L(−p)) = 0
and H0 (X, E ⊗ L) → H0 (X, E ⊗ L ⊗ κ(p)) is surjective.
Assume that E and F be nef bundles on X of rank e and f . Let L be a line
bundle of degree d ≥ 2g. Since E ⊗ L is globally generated, we may choose e global
sections of E ⊗ L that restrict to a basis of E ⊗ L ⊗ κ(ξ) over the generic point
ξ ∈ X. This gives a generically surjective map (L∨ )⊕e → E. Similarly, there is a
generically surjective map (L∨ )⊕f → F . Taking the tensor product of these maps
gives a generically surjective map

(L∨ )⊕ef → E ⊗ F.

If Q is a quotient line bundle of E ⊗ F , then the image of Q′ ⊆ Q of (L∨ )⊕ef satisfies

deg Q ≥ deg Q′ ≥ −d. This shows that every quotient line bundle Q of the tensor
product E ⊗ F of any two nef bundles E and F of rank e and f satisfies deg Q ≥ −d.
Assume that char(k) = p. Suppose that there is a line bundle quotient E ⊗F ↠ Q
with deg Q < 0. Denote FrN : X → X be the N th power of the absolute Frobenius,
the quotient (FrN )∗ E ⊗ (FrN )∗ F ↠ (FrN )∗ Q is a quotient line bundle of degree
N p deg Q. By Proposition B.2.33(1), (FrN )∗ E and (FrN )∗ F are nef. Taking N
such that N p deg Q < −d gives a quotient line bundle whose degree is less than −d,
contradicting the fact above.
Assume that char(k) = 0. Since X is of finite type over k, its defining equations
involve finitely many coefficients of k. Thus there exists a finitely generated Z-
subalgebra A ⊆ k, a scheme X e of finite type over A, and a cartesian diagram

X /X
e

□
 
Spec k / Spec A.

By Descent of Properties of Morphisms under Limits (B.3.7), we may further arrange

that Xe → Spec A is a family of smooth curves. Finally, by restricting along a map
Spec R → Spec A, we may assume that A is a DVR whose closed and generic points
have characteristic p and 0, respectively. We may therefore reduce to the positive
characteristic case by usingVthat Nefness for Bundles is Stable under Generization
k
(B.2.34). The nefness of E, Symk E, and Symk (E ∨ )∨ follow from the same
argument, and in fact shows more generally that for nef vector bundle E1 , . . . , Em
of ranks r1 , . . . , rm and for a representation ρ : GLr1 × · · · × GLrm → GLN which is
semipositive, i.e., extends to a map Matr1 ,r1 × · · · × Matrm ,rm → MatN,N , then the
induced vector bundle ρ(E1 , . . . , Em ) is also nef. See also [Kol90, Prop. 3.5], [Bar71,
Thm. 3.3], and [Har66a, Thm. 5.2].

Proposition B.2.34 (Nefness for Bundles is Stable under Generization). Let X

be a proper and flat scheme over a DVR R and E be a vector bundle on X. If the
restriction E0 of E to the central fiber X0 is nef, then so is the restriction Eη to the
generic fiber Xη .

Proof. This follows from Proposition B.2.17.

526
B.3 Limits of schemes
In moduli theory, we often need to deal with non-noetherian schemes for the simple
reason that moduli functors and stacks are defined over the category of all schemes.
Trying to work instead with the category of locally noetherian schemes has the
limitation that it is not closed under fiber products, while the category of schemes
finite type over a field or Z doesn’t contain local rings of schemes or their completions.
In any case, it is usually straightforward to reduce properties of schemes and their
morphisms to the noetherian case using the limit methods introduced in this section.

B.3.1 Existence of limits and noetherian approximation

The limit of an inverse system of schemes with affine transition maps exists.

Proposition B.3.1 (Existence of Limits). If (Sλ , fλµ )λ∈Λ is an inverse system

of schemes with affine transition maps, then the limit S = limλ∈Λ Sλ exists in the
category of schemes such that each morphism fλ : S → Sλ is affine.

Proof. If each Sλ = Spec Aλ is affine, one takes S = Spec(colimλ Aλ ). In general,

choose an element 0 ∈ Λ and set S = SpecS0 (colimλ≥0 fλ0,∗ OSλ ). Details can be
found in [EGA, IV.8.2] and [SP, Tag 01YX].

Every affine scheme Spec A is the limit of affine schemes Spec Aλ of finite type
over Z. This follows from the fact that the ring A is the union of its finitely generated
Z-subalgebras. More generally, we have:

Proposition B.3.2 (Relative Noetherian Approximation). Let X → S be a mor-

phism of schemes with X quasi-compact and quasi-separated and with S quasi-
separated. Then X = limλ∈Λ Xλ is a limit of an inverse system (Xλ , fλµ ) of schemes
of finite presentation over S with affine transition maps over S.

Proof. See [SP, Tag 09MV]. When S = Spec Z, this is often referred to as Absolute
Noetherian Approximation and was first established in [TT90, Thm. C.9].

Proposition B.3.3 (Descent of Morphisms under Limits). Let S = limλ∈Λ Sλ be a

limit of an inverse system of quasi-compact and quasi-separated schemes with affine
transition maps.
(1) For a finitely presented morphism X → S of schemes, there exists an index
0 ∈ Λ and a finitely presented morphism X0 → S0 of schemes such that
X ∼= X0 ×S0 S. Moreover, if we define Xλ := X0 ×S0 Sλ for λ > 0, then
X = limλ≥0 Xλ is the limit of the inverse system (Xλ , fλµ ) where the (affine)
transition map fλµ : Xλ → Xµ is the base change of Sλ → Sµ for λ ≥ µ.
(2) Let X0 and Y0 be finitely presented schemes over S0 for some index 0 ∈ Λ.
For λ > 0, set Xλ = X0 ×S0 Sλ and Yλ = Y0 ×S0 Sλ , and let X = limλ Xλ
and Y = limλ Yλ be the limits (Proposition B.3.1). Then the natural map

colimλ≥0 MorSλ (Xλ , Yλ ) → MorS (X, Y )

is bijective.

In other words, the category of schemes finitely presented over S is the colimit
of the categories of schemes finitely presented over Sλ .

Proof. See [EGA, IV.8.8] and [SP, Tag 01ZM].

527
 Quasi-coherent sheaves also descend under limits.

Proposition B.3.4 (Descent of Quasi-Coherent Sheaves under Limits). Let (Sλ , fλµ )
be an inverse system of quasi-compact and quasi-separated schemes with affine
transition maps and limit S = limλ∈Λ Sλ . Denote the projection maps by fλ : S → Sλ .
(1) If F is a quasi-coherent OS -module of finite presentation (resp., vector bundle,
line bundle), then there exists an index λ ∈ Λ and an OSλ module Fλ of finite
presentation (resp., vector bundle, line bundle) such that F ∼ = fλ∗ Fλ .
(2) For an index 0 ∈ Λ, let F0 and G0 be OS0 -modules of finite presentation, and
let F and G be the pullbacks to S via f0 and Fλ and Gλ be the pullbacks to Sλ
via f0λ . The natural map

colimλ≥0 HomOSλ (Fλ , Gλ ) → HomOS (F, G)

is bijective.
(3) For an index 0 ∈ Λ, let f0 : X0 → Y0 be a finitely presented morphism of
schemes over S0 and let F0 be a quasi-coherent sheaf on X0 of finite presenta-
tion. If the pullback of F0 under X0 ×S0 S → X0 is flat over Y0 ×S0 S, then
the pullback of F0 under X0 ×S0 Sλ → X0 is flat over Y0 ×S0 Sλ for λ ≫ 0.

In other words, the category of finitely presented modules over S is the colimit
of the categories of finitely presented modules over Sλ . Note that applying (2) with
F0 = OS0 implies Γ(S, F ) = colimλ≥0 Γ(Sλ , Fλ ).

Proof. See [EGA, IV.8.5.2] and [SP, Tags 01ZR, 0B8W, and 05LY].

B.3.2 Descent of properties under limits

Proposition B.3.5 (Descent of Properties of Schemes under Limits). Let S =
limλ Sλ be a limit of an inverse system of quasi-compact and quasi-separated schemes
with affine transition maps. If S is affine (resp., quasi-affine, separated), then so is
Sλ for λ ≫ 0.

Proof. See [SP, Tags 01Z6, 086Q, and 01Z5] and [TT90, Props C.6-7].

Definition B.3.6. We say that a property P of morphisms of schemes descends

under limits if for every limit S = limλ∈Λ Sλ of an inverse system of quasi-compact
and quasi-separated schemes with affine transition maps, the following holds: for
every index 0 ∈ Λ, and for every morphism g0 : X0 → Y0 of quasi-compact and
quasi-separated schemes with base changes gλ : Xλ → Yλ over Sλ and g : X → Y
over S, if g has P, then gλ has P for λ ≫ 0.

Proposition B.3.7 (Descent of Properties of Morphisms under Limits). The

following properties of morphisms of schemes descend under limits: isomorphism,
closed immersion, open immersion, affine, quasi-affine, finite, quasi-finite, proper,
projective, quasi-projective, separated, monomorphism, surjective, flat, locally of
finite presentation, unramified, étale, smooth, syntomic, and the property that every
fiber is connected and has pure dimension d for a fixed integer d.

Proof. See [EGA, IV 8.10.5] and [SP, Tags 081C and 05M5].

528
B.3.3 Spreading out and other applications
Limits methods of schemes allows us to ‘spread out’ objects defined over colimits
of rings (e.g., a local ring Rp ) to an object defined over a finite type ring (e.g., a
localization Rf ).

Proposition B.3.8 (Spreading out). Let R be a ring and p ⊆ R be a prime ideal.

If X → Spec Rp is a finitely presented morphism, there exists an element f ∈
/ p and
a finitely presented morphism X ′ → Spec Rf such that X ∼ = X ′ ×Rf Rp .
Proof. This is a direct consequence of Descent of Morphisms under Limits (B.3.3)
applied to Rp = colimf ∈p
/ Rf .

For example, if Y is an integral scheme with function field K(Y ), every finite
presented scheme defined over K(Y ) can be extended to a scheme defined over
an nonempty open subscheme. Similarly, a finitely presented scheme over the
henselization Rph (resp., strict henselization Rpsh ) as defined in §B.5.3 can be spread
out to a finitely presented scheme X ′ over an étale neighborhood (resp., residually
trivial étale neighborhood) Spec R′ → Spec R of p.
For another typical application of noetherian approximation, we illustrate how
properties of an arbitrary family of curves can be reduced to a family over a
noetherian base.

Proposition B.3.9. Let S be a quasi-compact and quasi-separated scheme (e.g., an

affine scheme), and let C → S be a proper, flat, and finitely presented morphism of
schemes such that every geometric fiber has dimension at most 1. Then there exists
a cartesian diagram
C / C′

□
 
S / S′

where S ′ is a scheme of finite type over Z and C ′ → S ′ is a proper flat morphism

of schemes such that every geometric fiber has dimension at most 1. Moreover, if
C → S is smooth, then C ′ → S ′ can also be arranged to be smooth.

Proof. Write S = limλ∈Λ Sλ as a limit of an inverse system of schemes of finite

type over Z (Proposition B.3.1). Since C → S is finitely presented, there exists an
index 0 ∈ Λ and a finitely presented morphism C0 → S0 such that C ∼ = C0 ×S0 S
(Proposition B.3.3). For each λ > 0, we can define Cλ = C0 ×S0 Sλ and we have a
cartesian diagram
C / Cλ / C0

□ □
  
S / Sλ / S0 .

Since C → S is flat and proper with fiber of dimension at most 1 (resp., smooth),
then there exists λ0 ∈ Λ such that the same is true for Cλ → Sλ for all λ ≥ λ0
(Proposition B.3.7). We now take S ′ = Sλ and C ′ = Cλ for some λ ≥ λ0 .

The upshot is that if we can establish properties of the morphism C ′ → S ′ of

noetherian schemes and the properties are stable under base change, then they hold
for C → S. In Lemma 5.2.24, we show the same property for nodal families of
curves.

529
B.4 Pushouts of schemes
Pushouts are the dual notion of fiber products. Unlike fiber products, pushouts may
not always exist. However, Ferrand identified a general situation where they do exist:
one map is a closed immersion and the other is affine. Ferrand pushouts of Artin
rings are especially important in deformation theory such as in the homogeneity
conditions appearing in Rim–Schlessinger’s Criteria (C.4.6) and Artin’s Axioms
for Algebraicity (C.7.4). We also use pushouts to construct the Gluing Morphisms
(5.6.16) between moduli spaces of stable curves.

B.4.1 Existence of pushouts

Theorem B.4.1 (Ferrand Pushout). Consider a diagram

X0  /X
i

f0 f (B.4.2)
  
Y0  /Y
j

of schemes where i : X0 ,→ X is a closed immersion and f0 : X0 → Y0 is affine. If

(⋆) for every point y0 ∈ Y0 , there exists an affine open subscheme U ⊆ X with
f0−1 (y) ⊆ U ∩ X0 ,
then there exists a scheme Y , a closed immersion j : Y0 ,→ Y , and an affine morphism
f : X → Y of schemes such that (B.4.2) is a pushout. Moreover,
(a) the square (B.4.2) is cartesian, X → Y restricts to an
` isomorphism X ∖ X0 →
Y ∖ Y0 , and |Y | is identified with the pushout |X| |X0 | |Y0 | as a topological
space,
(b) the induced map
OY → j∗ OY0 ×(j◦f0 )∗ OX0 f∗ OX
is an isomorphism of sheaves, and
(c) if f0 is finite, then so is f . In this case, if X0 , X, and Y0 are locally of finite
type over a noetherian scheme, then so is Y .
Proof. See [Fer03, Thm. 5.4 and 7.1], [Art70, Thm. 6.1], and [SP, Tag 0ECH].
We call Y = X X0 Y0 the Ferrand pushout. Note that if there is a cartesian
`
diagram of schemes as in (B.4.2) with f : X → Y affine, then condition ((⋆)) is
satisfied.
Remark B.4.3 (Existence of pushouts in general). Condition (⋆) does not always
hold: for example, consider f0 : X0 → Spec k and i : X0 ,→ X, where X is a smooth
proper (but not projective) 3-fold X over an algebraically closed field k such that
there is a set X0 two k-points not contained in an affine. However, the pushout
always exists as an algebraic space and is a pushout in the category of algebraic
spaces.
Example B.4.4 (Affine case). In the affine case where X = Spec A, X0 = Spec A0 ,
Y0 = Spec B0 , then Spec(A ×A0 B0 ) is the pushout X ⨿X0 Y0 .
Example B.4.5 (Gluing and pinching). If X0 ,→ X and X0 ,→ Y0 are closed
immersions, the pushout X ⨿X0 Y0 can be viewed as the gluing of X and Y0 along
X0 . For example, the nodal curve Spec k[x, y]/xy is the union of A1 and A1 along

530
their origins. If X0 = Z ⨿ Z is the union of two isomorphic disjoint subschemes of
X and X0 → Z is the projection, then the pushout X ⨿Z⨿Z Z can be viewed as the
pinching of the two copies of Z in X. For example, the nodal cubic curve is the
pinching of 0 and ∞ in P1 .
Exercise B.4.6. If X is the scheme-theoretic union of two closed subschemes Z1
and Z2 , show that X = Z1 ⨿Z1 ∩Z2 Z2 .
Example B.4.7 (Non-noetherianness). When f0 : X0 → Y0 is affine but not finite,
the pushout X ⨿X0 Y0 is often not noetherian. For example, if X0 = V (x) ⊆ X = A2k
and f0 : X0 → Spec k, the pushout is the non-noetherian affine scheme defined by

k[x, y] ×k[x] k = k[x, xy, xy 2 , xy 3 , . . .] ⊆ k[x, y].

On the other hand, we shouldn’t expect a finite type pushout: the y-axis in A2k
cannot be contracted.

B.4.2 Properties of pushouts

Proposition B.4.8 (Properties of Pushouts). Let X0 ,→ X be a closed immersion
and X0 → Y0 be an affine morphism of schemes.
(1) If Y ∼
`
=X Y0 is a Ferrand pushout of schemes, then the natural functor
X0

QCoh(Y ) → QCoh(Y0 ) ×QCoh(X0 ) QCoh(X),

restricts to an equivalence on the full subcategories of flat O-modules (resp.,

finite type and flat O-modules, finitely presented and flat O-modules).
Consider a commutative cube of schemes

X0′  / X′
x
X0  /Xx
  
Y0′  / Y′
 x 
Y0  /Y x

of schemes where X0 ,→ X is a closed immersion and X0 → Y0 is affine.

′
` Y → Y is flat
(2) Assume that such`that X0′ , Y0′ , and X ′ are the base changes.
If Y ∼= X Y
X0 0 , then Y ′ ∼
= X ′ X0′ Y0′ . If Y ′ → Y is fppf, the converse is
true.
(3) If the top and left faces are cartesian, and the front and back faces are Ferrand
pushouts, then all faces are cartesian. Moreover, if Y0′ → Y0 and X ′ → X are
étale (resp., smooth), so is Y ′ → Y .
(4) Suppose that Y is defined over a scheme S. Let S ′ → S be morphisms of
′ ′ ′
X ′ be the base changes. If X0 → S is flat
` let Y , X0 , Y0′ ,∼and′ `
schemes, and
∼
and Y = X X0 Y0 , then Y = X X ′ Y0′ .
0

Proof. Parts (1)–(3) follow from [Fer03, Thm. 2.2]; see also [SP, Tag 0D2K] and
[AHHLR24, §4]. Part (4) is elementary: one reduces to the affine case Y =
Spec B0 ×A0 A and S = Spec R, and since A0 is flat over R, the exact sequence
0 → B → A × B0 → A0 → 0 of R-modules remains exact after tensoring with an
R-algebra.

531
B.5 Completions, henselizations, and Artin Approx-
imation
After reviewing properties of complete, henselian, and strictly henselian local rings,
we discuss Artin Approximation (Theorem B.5.18) which can vaguely formulated
as: every algebraic object defined over the completion O bS,s of the local ring of a
finite type scheme S at a point s can be approximated to an object defined over a
residually-trivially étale neighborhood (S ′ , s′ ) → (S, s). While difficult to prove (and
we do not prove it here!), the statement of Artin Approximation is at least very easy
to digest and teaches us how to think about étale maps. It is a fundamental tool in
local-to-global arguments in moduli theory, but its use can usually be avoided (by
more direct but less conceptual methods).

B.5.1 Complete rings

Definition B.5.1. A ring R is complete with respect to an ideal I if the natural
map
R → lim R/I n (B.5.2)
←−
n

is an isomorphism. The completion with respect to an ideal I of R is defined as

b = lim R/I n .
R
← −
n

More generally, we say that an R-module M is complete if M → limn M/I n M is an

←−
isomorphism, and we define the completion of M as Mc = lim M/I n M .
←−n
The most important case is when I = m ⊆ R is a maximal ideal.
Caution B.5.3. When R is non-noetherian, the completion of a local ring may not
even be complete; see [SP, Tag 05JC]. In the literature, ‘complete’ sometimes refers
to
T only the surjectivity of (B.5.2), while ‘separated’ refers to the injectivity, i.e.,
n
n I = 0.
The Artin–Rees Lemma plays an important role in establishing basic properties
of complete noetherian local rings.
Lemma B.5.4 (Artin–Rees Lemma). Let R be a noetherian ring, I ⊆ R be an
ideal, M be a finitely generated A-module, and M = M0 ⊇ M1 ⊇ M2 ⊇ · · · be a
stable I-filtration (i.e., IMn = Mn+1 for n ≫ 0). If M ′ ⊆ M is a submodule, then
M ′ = M ′ ∩ M0 ⊇ M ′ ∩ M1 ⊇ · · · is a stable I-filtration. In particular, there exists
an integer k such that

M ′ ∩ (I n M ) = I n−k (M ′ ∩ (I k M ))

for all n ≥ k.
Proof. See [AM69, Prop. 10.9 and Cor. 10.10] and [Eis95, Lem. 5.1].
Proposition B.5.5 (Properties of Noetherian Complete Local Rings). Let (R, m)
be a noetherian local ring.
(1) R b = m ⊗R R;
b = mR
b is a complete noetherian local ring with maximal ideal m b
b n and mn /mn+1 = m
cn = m
(2) m b n /m
b n+1 ;
(3) R → Rb is faithfully flat; and

532
 c = M ⊗R R.
(4) For a finitely generated R-module M , M b

Proof. See [AM69, Prop. 10.13–16].

The following provides variants of Nakayama’s lemma that hold for complete
local rings without finite generated hypotheses.
Lemma B.5.6 (Complete Nakayma’s Lemma).
(1) If (A, m) is a complete noetherianT local rings and M is a (possibly not finitely
generated) A-module such that k mk M = 0 and m1 , . . . , mn ∈ M generate
M/mM , then m1 , . . . , mn also generate M .
(2) If (A, mA ) is a local ring and M → N is a homomorphism of A-modules such
that M/mA M → N/mA N is surjective, then M c→N b is surjective.
(3) Let (A, mA ) → (B, mB ) be a local homomorphism of complete noetherian local
rings such that A/mA ∼ = B/mB . If mA /m2A → mB /m2B is surjective, so is
A → B. If in addition A = B, then A → B is an isomorphism.
Proof. For (1) and (2), see [Eis95, Exc. 7.2] and [SP, Tag 0315]. To see (3), (2)
implies that the inclusion mA B ,→ mA is surjective. Thus, mA B = mA and B is
complete as an A-module with respect to mA . As A/mA → B/mB is surjective,
applying (2) again shows that A → B is surjective. The final statement follows from
the fact that a surjective endomorphism of a noetherian ring is an isomorphism.
Theorem B.5.7 (Cohen Structure Theorem). If (R, m) is a complete noetherian
local ring containing a field, then R ∼ = (R/m)[[y1 , . . . , yr ]]/J. If in addition R is
regular, then R ∼
= (R/m)[[y1 , . . . , yr ]].
Proof. See [Eis95, Thm. 7.7] and [SP, Tags 032A and 0C0S].

B.5.2 Henselian and strictly henselian local rings

Let (R, m) be a local ring with residue field κ. We will denote the image of a ∈ R
(resp., f ∈ R[x]) as a ∈ κ (resp., f ∈ κ[x]). If f ∈ R[t], we denote its derivative by
′
f ′ ∈ R[t]. Note that f ′ = f .
Definition B.5.8. Let (R, m) be a local ring with residue field κ.
(1) We say that R is henselian if for every monic polynomial f ∈ R[t], every root
α0 ∈ κ of f with f ′ (α0 ) ̸= 0 lifts to a root α ∈ R of f .
(2) We say that R is strictly henselian if R is henselian and κ is separably closed.
Hensel’s lemma states that every complete DVR R, e.g., Zp , is henselian.
Proposition B.5.9 (Henselian Equivalences). The following are equivalent for a
local ring (R, m) with residue field κ:
(1) R is henselian;
(2) for every polynomial f ∈ R[t], every factorization f = g0 h0 with gcd(g0 , h0 ) = 1
lifts to a factorization f = gh with g = g0 and h = h0 ;
(3) every finite R-algebra is a finite product of local rings finite over R;
(4) every quasi-finite R-algebra A is isomorphic to a product A ∼ = B × C where B
is a finite over R and C ⊗R κ = 0;
(5) every étale ring homomorphism ϕ : R → A and a prime p ⊆ A with ϕ−1 (p) = m
and κ = κ(p) has a unique section s : A → R with s−1 (m) = p.

533
Moreover, R is strictly henselian if and only if for every étale ring homomorphism
ϕ : R → A and prime p ⊆ A with ϕ−1 (p) = m, there is a unique section s : A → R
with s−1 (m) = p.

Proof. See [EGA, IV.18.5.11], [Mil80, Thm. I.4.2], and [SP, Tag 04GG].

Proposition B.5.10. Let (R, m) be a henselian (resp., strictly henselian) local ring
with residue field κ.
(1) Every finite R-algebra is a product of finite henselian local (resp., strictly
henselian) R-algebras.
(2) Every complete local ring is henselian.
(3) The functor A 7→ A ⊗R κ gives an equivalence of categories between finite étale
A-algebras and finite étale κ-algebras.

Proof. See [EGA, IV.18.5.10-15], [Mil80, 4.3-4.5], and [SP, Tag 04GE].

Remark B.5.11. Although it is not used in this text, there is a general notion of
henselian pairs that is sometimes useful. A pair (X, X0 ) consisting of a scheme X and
a closed subscheme X0 ⊆ X is henselian if every finite morphism f : U → X induces
a bijection ClOpen(U ) → ClOpen(f −1 (X0 )) between open and closed subschemes of
U and those of f −1 (X0 ). If (R, m) is a henselian local ring, then (Spec R, Spec(R/m))
is a henselian pair by Proposition B.5.9(3). See [EGA, IV.18.5.5] or [SP, Tag 09XD]
for a further discussion and equivalences.

B.5.3 Henselizations and strict henselizations

Definition B.5.12. Let (R, m) be a local ring with residue field κ. The henselization
of R is a local homomorphism R → Rh into a henselian local ring Rh such that
every other local homomorphism R → A into a henselian local ring factors uniquely
through R → Rh .
Given a separable closure κ → κs , the strict henselization of R with respect to κ →
κ is a local homomorphism R → Rsh into a strictly henselian local ring (Rsh , msh )
s

inducing κ → κs on residue fields such that every other local homomorphism

R → A into a strictly henselian local ring (A, mA ) factors through R → Rsh and the
factorization is uniquely determined by the inclusion Rsh /msh → A/mA of residue
fields.

Proposition B.5.13. Let (R, mR ) be a local ring with residue field. The henseliza-
tion R → Rh (resp., strict henselization R → Rsh ) exist and can be constructed as
colim A, where the colimit is taken over all étale local R-algebras A with R/mR ∼=
A/mA (resp., over diagrams R → A → (R/mR )s where (R/mR )s is a fixed separable
closure of R/mR and A is an étale local R-algebra). Moreover,
(1) the residue fields of Rh and Rsh are R/mR and (R/mR )s , respectively,
(2) the maps R → Rh and R → Rsh are faithfully flat local ring homomorphisms,
and
(3) if R is noetherian, then so is Rh and Rsh .

Proof. See [EGA, IV.18.5-8], [Mil80, I.4], and [SP, Tags 0BSK and 07QL].

For a scheme X and a point x ∈ X with a choice of separable closure κ(x) → κs ,

the henselization OX,x
h
and strict henselization OX,x
sh
are the colimits of Γ(U, OU )

534
taken over diagrams

Spec κ(x) /U Spec κ(x)s /U

x
ét and ét
$   
X Spec κ(x)
x / X,

where U → X is étale. We can view OX,x

h
and OX,x
sh
as local rings in the étale
topology.

B.5.4 Néron–Popescu Desingularization

Artin Approximation (B.5.18) is closely related to Néron–Popescu Desingularization
(B.5.15), another equally deep and powerful theorem. We do not attempt to
prove Néron–Popescu Desingularization, but we do show how it implies Artin
Approximation.
Definition B.5.14. A ring homomorphism A → B of noetherian rings is called
geometrically regular if A → B is flat and for every prime ideal p ⊆ A and every
finite field extension κ(p) → κ′ (where κ(p) = Ap /p), the fiber B ⊗A κ′ is regular.
If A → B is of finite type, then A → B is geometrically regular if and only
if Spec B → Spec A is smooth (Theorem A.3.1). A k-algebra B is geometrically
regular if and only if A ⊗k k′ is regular for every field extension (equivalently, for
every finite purely inseparable extension) k → k′ ; see [EGA, IV2 .6.7.8], [Mat89,
Thm. 23.7], or [SP, Tag 0381]. Note that a field extension is separable if and only if
it is geometrically regular.
Theorem B.5.15 (Néron–Popescu Desingularization). A homomorphism A → B
of noetherian rings is geometrically regular if and only if there is a directed system
Bλ of smooth A-algebras over a directed set Λ such that B = colim Bλ∈Λ .
Proof. This result was proved by Néron in [Nér64] in the case of DVRs and in
general by Popescu in [Pop85], [Pop86], and [Pop90]. We recommend [Swa98] and
[SP, Tag 07GC] for expositions.
Definition B.5.16. A noetherian local ring A is a G-ring if the homomorphism
A→A b is geometrically regular.

One of the defining properties of an excellent scheme is that the local rings are
G-rings. Fortunately, most local rings that we care about in algebraic geometry are
G-rings.
Theorem B.5.17. The localization of a finitely generated algebra over a field or Z
is a G-ring.
Proof. While substantially easier than Néron–Popescu Desingularization, this result
also requires some effort. See [EGA, IV.7.4.4] or [SP, Tag 07PX].

B.5.5 Artin Approximation

Recall from Definition A.1.4 that a contravariant functor F : Sch/S → Sets is limit
∼
preserving (or locally of finite presentation) if colim F (Aλ ) → F (colim Aλ ) for all
inverse systems {Spec Aλ } over S. When F is representable by X, this is equivalent
to X → S being of locally of finite presentation.

535
Theorem B.5.18 (Artin Approximation). Let S be a scheme and s ∈ S be a point
such that OS,s is a G-ring (Definition B.5.16), e.g., a scheme of finite type over a
field or Z. Let
F : Sch/S → Sets

be a limit preserving contravariant functor and ξb ∈ F (Spec O bS,s ). For every integer
N ≥ 0, there exists an étale morphism (S , s ) → (S, s) with κ(s) = κ(s′ ) and an
′ ′

object ξ ′ ∈ F (S ′ ) such that the restrictions of ξb and ξ ′ to Spec(OS,s /mN s

+1
) are
equal.

The restriction ξ ′ to Spec(OS,s /mN s

+1
) is well-defined because of the identification
N +1 ∼
OS,s /ms = OS ′ ,s′ /ms′ . It is not possible in general to find ξ ′ ∈ F (S ′ ) that
N +1

precisely restricts to ξ,
b or even such that the restrictions of ξ ′ and ξb to Spec OS,s /mn+1
s
agree for all n ≥ 0. For instance, consider F = Mor(−, A1 ) and a non-algebraic
power series ξb ∈ O bA1 ,0 .

Proof. The theorem was originally proven in [Art69a, Cor. 2.2] in the case that S is of
finite type over a field or an excellent dedekind domain. We also recommend [BLR90,
§3.6] for an accessible account of the case of excellent and henselian DVRs. We
prove only how Artin Approximation follows from Néron–Popescu Desingularization
(B.5.15). By Néron–Popescu, we may write O bS,s = colimλ∈Λ Bλ as a directed
colimit of smooth OS,s -algebras. Since F is limit preserving, there exists λ ∈ Λ, a
factorization OS,s → Bλ → O bS,s , and an element ξλ ∈ F (Spec Bλ ) whose restriction
to F (Spec ObS,s ) is ξ.
b Letting B = Bλ and ξ = ξλ , we have a commutative diagram

ξb

Spec O
bS,s g
/ Spec B ξ
/' F

& 
Spec OS,s

where Spec B → Spec OS,s is smooth. We claim that we can find a commutative
diagram
 / Spec B
S′
(B.5.19)
$ 
Spec OS,s

where S ′ ,→ Spec B is a closed immersion, (S ′ , s′ ) → (Spec OS,s , s) is étale with

κ(s) = κ(s′ ) such that Spec OS,s /mN s
+1
→ S ′ → Spec B agrees with the restric-
tion of g : Spec OS,s → Spec B. To see this, since ΩB/OS,s is a locally free B-
b 5

module, after replacing Spec B with an affine open neighborhood of g(s), we may
assume that ΩB/OS,s is free with basis db1 , . . . , dbr . This induces a homomorphism
OS,s [x1 , . . . , xr ] → B defined by xi 7→ bi and provides a factorization

Spec B → ArOS,s → Spec OS,s

5 This is where the approximation occurs. It is not possible to find a an étale map S ′ →
bS,s → S ′ → Spec B is equal to g.
Spec B → Spec OS,s such that Spec O

536
such that Spec B → ArOS,s is étale. Choosing a lift of the composition

3 OS,s


OS,s [x1 , . . . , xr ] /B /O
bS,s / OS,s /mNs
+1

defines a section s : Spec OS,s → ArOS,s and we define S ′ as the fibered product

S ′ _ / Spec OS,s
_
□ s
 
Spec B / Ar .
OS,s

ξ
This gives the desired diagram (B.5.19), and the composition ξ ′ : S ′ → Spec B − →F
is an element which agrees with ξb up to order N . Finally, we use limit methods
to ‘spread out’ the étale map (S ′ , s′ ) → (Spec OS,s , s) and element ξ ′ ∈ F (S ′ ) to
an étale map (S ′′ , s′′ ) → (S, s) and an element ξ ′′ ∈ F (S ′′ ). Assuming S = Spec A
is affine and writing OS,s = colimg∈m / s Ag , we may use Propositions B.3.3, B.3.5
and B.3.7 (or a direct argument) to construct g ∈ / ms and an étale affine morphism
S ′′ → Spec Ag such that S ′′ ×Ag Ams ∼ = S ′ . As F is limit preserving and Γ(S ′ , OS ′ ) =
′′ ), after replacing g with hg for some h ∈ / ms , we can find an
′′
colimh∈m/ s Γ(Sh , OSh
element ξ ∈ F (S ) restricting to ξ and, in particular, agreeing with ξb up to order
′′ ′′ ′

N.
Exercise B.5.20 (Alternative formulations). Let (A, m) be a henselian local G-ring.
(1) Let f1 , . . . , fm ∈ A[x1 , . . . , xn ] and b
a = (b b⊕n be a solution. Show
an ) ∈ A
a1 , . . . , b
that for every N ≥ 0, there is a solution a = (a1 , . . . , an ) ∈ A⊕n such that
a∼=ba mod mN +1 .
Hint: Apply Artin Approximation to the functor representing Spec A[x1 , . . . , xn ]/(f1 , . . . , fm ).
(2) Show that (1) implies Artin Approximation.
Hint: Use that F is limit preserving to find a finitely generated A-subalgebra
B⊆O bS,s and an element ξ ∈ F (B) restricting to ξ.b

B.5.6 A first application of Artin Approximation

Two pointed schemes with isomorphic completions have isomorphic étale neighbor-
hoods.
Corollary B.5.21. Let X and Y be schemes of finite type over a scheme S and let
s ∈ S be a point such that OS,s is a G-ring. If x ∈ X and y ∈ Y are points over
s such that O
bX,x and ObY,y are isomorphic as OS -algebras, then there exists étale
morphisms

(U, u)

{ #
(X, x) (Y, y)

with κ(x) = κ(u) = κ(y).

537
Proof. The functor

F : Sch/X → Sets, (T → X) 7→ MorS (T, Y )

is limit preserving as it can be identified with the representable functor MorX (−, Y ×S
X). The isomorphism O bX,x ∼
=O bY,y gives an element of F (Spec O bX,x ). Applying
Artin Approximation with N = 1 yields an étale map (U, u) → (X, x) with κ(x) =
κ(u) and a map (U, u) → (Y, y) such that OY,y /m2y → OU,u /m2u is an isomorphism.
Since O bU,u is abstractly isomorphic to O bY,y , Complete Nakayama’s Lemma (B.5.6(3))
implies that OY,y → OU,u is an isomorphism. This further implies that U → Y is
b b
étale at u and the statement follows after replacing U with an open neighborhood
of u. See also [SP, Tag 0CAV].
∼ b
If ϕ : O
bY,y → OX,x is a specified isomorphism, it is not always possible to find
étale neighborhoods (U, u) → (X, x) and (U, u) → (Y, y) such that the composition
ObY,y ∼
=O bU,u ∼
=ObX,x agrees with ϕ.

538
Appendix C

Deformation theory

Deformation theory is the study of the local geometry of a moduli space M near an
object E0 ∈ M(k). We focus primarily on the following three deformation problems:
(A) Embedded deformations of a closed subscheme Z0 in a fixed projective scheme
X over k. Here the moduli problem is the Hilbert functor HilbP (X) and the
object is E0 = [Z0 ⊆ X] ∈ HilbP (X)(k).
(B) Deformations of a scheme E0 over k. The motivating example for us is
when E0 is a smooth curve, in which case the moduli problem is Mg and
[E0 ] ∈ Mg (k), or more generally when E0 is a proper curve, in which case the
moduli problem is the stack Mallg of all curves.
(C) Deformations of a coherent sheaf E0 on a fixed projective scheme X over k.
The moduli problem is Coh(X) and [E0 ] ∈ Coh(X)(k).
Deformation theory provides a local-to-global perspective of moduli. By zooming
in around E0 ∈ M(k), we studying successively first-order neighborhoods of M at
E0 , higher-order deformations of E0 , formal neighborhoods of E0 , and finally étale
or smooth neighborhoods of E0 . Before getting started, we give a quick overview of
the seven sections of this appendix.
(1) A first-order deformation of E0 is an object E ∈ M(k[ϵ]) over the dual numbers
k[ϵ] := k[ϵ]/(ϵ2 ) together with an isomorphism α : E0 → E|Spec k , or in other
words a commutative diagram

Spec k  / Spec k[ϵ]

[E]
[E0 ]
% 
M,
allowing us to view E as a tangent vector of M at E0 . We classify first-order
deformations of Problems (A)–(C) in §C.1.
(2) Given a surjection A′ ↠ A of artinian local k-algebras with residue field k and
an object E ∈ M(A) with an isomorphism E0 → E|Spec k , a deformation of
E over A′ is an object E ′ ∈ M(A′ ) with an isomorphism α : E → E ′ |Spec A .
Pictorially, this corresponds to a commutative diagram
 
Spec k  / Spec A  / Spec A′
[E]
[E ′ ]
[E0 ]
%* 
M.

539
 In §C.2, for Problems (A)–(C), we determine the obstruction for the existence
of a deformation E ′ of E over A′ , and we classify all such deformations in the
case that there is no obstruction.
These first two sections are essential for our study of Mg and Bun(C), e.g., for
establishing smoothness and computing their dimensions; see Theorem 5.4.14. Since
the study of deformation theory is inextricably connected to stacks and moduli, we
have included §C.3 - C.7 for completeness.
(3) In §C.3, we offer a survey of the cotangent complex, and in particular how
it governs infinitesimal deformation theory, recovering some of the results of
§C.1-C.2.
(4) Given a complete noetherian local k-algebra (R, m), a formal deformation of
E0 over R is a compatible collection of deformations En ∈ M(R/mn+1 ) of E0 ,
and a formal deformation {En } is versal if every deformation over a thickening
of artin rings factors through one of the En (see Definition C.4.2 for a precise
definition). Rim–Schlessinger’s Criteria (Theorem C.4.6) provides criteria for
the existence of a versal deformation {En } of E0 , and in §C.4 we verify the
criteria for the Problems (A)–(C).
(5) A formal deformation {En } over (R, m) is effective if there exists an object
b ∈ M(R) extending the {En }, or in other words there exists a commutative
E
diagram
 / Spec R/m2   / Spec R/m3   / · · ·  / Spec R
Spec R/m
[E2 ]
[E1 ] [E]
b

[E0 ]
.) M.

In §C.5, we prove Grothendieck’s Existence Theorem (C.5.3) and show how it

implies that formal deformations are effective for Problems (A)–(C).
(6) Given an effective versal formal deformation E b over R, Artin Algebraization
(C.6.8) ensures the existence of a finite type k-scheme U with a point u ∈
U (k) and an object E ∈ M(U ) such that R ∼ = O b Spec R/mn+1 ∼
bU,u and E| =
E|Spec R/mn+1 for all n.
(7) Artin’s Axioms for Algebraicity (C.7.1 and C.7.4) provide criteria to verify the
algebraicity of a moduli problem M. Namely, it provides conditions to ensure
that the morphism [E] : U → M constructed above is a smooth morphism in
an open neighborhood of E0 .
For additional algebraic treatments of deformation theory, we recommend [Art76],
[Kol96, §I.2], [Vis97], [FGI+ 05, §6], [Ser06], [Nit09], [Har10], and [SP, Tag 0ELW].
We also recommend [Kod86] for deformations of manifolds, and [Ill71, Ill72] for an
exhaustive treatment of the cotangent complex.

C.1 First order deformations

For a field k, denote the dual numbers by k[ϵ] := k[ϵ]/(ϵ2 ).

C.1.1 First order embedded deformations

Definition C.1.1. Let X be a projective scheme over a k and Z0 ⊆ X be a
closed subscheme. A first-order deformation of Z0 ⊆ X is a closed subscheme

540
Z ⊆ Xk[ϵ] := X ×k k[ϵ] flat over k[ϵ] such that Z0 = Z ×k[ϵ] k. Pictorially, a
first-order deformation is a filling of the diagram
 /X
:X 9 k[ϵ]
cl cl
,  +
Z0 /Z
flat
$   % 
Spec k  / Spec k[ϵ].

with a scheme Z flat over k[ϵ] and dotted arrows making the diagram cartesian.
Note that since both Z0 and the central fiber Z ×k[ϵ] k of Z are embedded in X,
it makes sense in the definition to require that they are equal. We say that Z ⊆ Xk[ϵ]
is trivial if Z = Z0 ×k k[ϵ].
Remark C.1.2. The closed subscheme Z0 ⊆ X defines a k-point [Z0 ⊂ X] ∈ HilbP (X)
of the Hilbert scheme, where P is the Hilbert polynomial of Z0 with respect to a fixed
ample line bundle on X. A first-order deformation corresponds to a commutative
diagram
[Z0 ⊆X]
Spec _ k / HilbP (X)
8

 [Z⊆Xk[ϵ] ]

Spec k[ϵ],
or in other words a tangent vector [Z ⊆ Xk[ϵ] ] ∈ THilbP (X),[Z0 ⊆X] .
Proposition C.1.3. Let X be a scheme over a k and Z0 ⊆ X be a closed subscheme
defined by a sheaf of ideals I0 ⊆ OX . There is a bijection

{first-order deformations Z ⊆ Xk[ϵ] } ∼

= H0 (Z0 , NZ0 /X )

where NZ0 /X = H omOZ0 (I0 /I02 , OZ0 ) is the normal sheaf. Under this correspon-
dence, the trivial deformation corresponds to 0 ∈ H0 (Z0 , NZ0 /X ).
Remark C.1.4. In light of Remark C.1.2, this proposition gives a bijection THilbP (X),[Z0 ⊆X] ∼
=
H0 (Z0 , NZ0 /X ).
Proof. We first handle the case when X = Spec B and Z0 = Spec B/I0 , and show
that the set of first-order deformations is bijective to

H0 (Z0 , NZ0 /X ) ∼
= HomB/I0 (I0 /I02 , B/I0 ) ∼
= HomB (I0 , B/I0 ).

Given a first-order deformation Z = Spec B[ϵ]/I, the flatness of Z over k[ϵ] implies
that tensoring the exact sequence 0 → I → B[ϵ] → B[ϵ]/I → 0 of k[ϵ]-modules
with k = k[ϵ]/(ϵ) yields an exact sequence 0 → I0 → B → B/I0 → 0. We define
α : I0 → B/I0 as follows: for x0 ∈ I0 , choose a preimage x = a + bϵ ∈ I and set
α(x0 ) := b ∈ B/I0 . Conversely, given a B-module homomorphism α : I0 → B/I0 ,
we define

I = {a + bϵ | a ∈ I0 , b ∈ B such that b = α(a) ∈ B/I0 } ⊆ B[ϵ].

Then (B[ϵ]/I) ⊗k[ϵ] k = B/I0 . To see that B[ϵ]/I is flat over k[ϵ], it suffices by the
flatness criterion for the dual numbers (see Remark A.2.7) to check that the map

541
 ϵ
B/I0 →− B[ϵ]/I is injective: given b ∈ B with bϵ ∈ I, then b ∈ I0 by the definition
of I. Thus Z = Spec B[ϵ]/I defines a first-order deformation of Z0 . For a general
k-scheme X, after choosing an affine cover {Ui }, one checks that the bijections
between deformations of Ui and H0 (Ui ∩ Z0 , NUi ∩Z0 /Ui ) glue to the desired bijection.
See also [Art76, Thm. 6.1], [Kol96, Thm. 2.8], [Ser06, Prop. 3.2.1], [Har10, Prop. 2.3]
for details.

C.1.2 First-order deformations of schemes

Definition C.1.5. Let X0 be a scheme over a field k. A first-order deformation of
X0 is a scheme X flat over k[ϵ] together with an isomorphism α : X0 → X ×k[ϵ] k,
or in other words a cartesian diagram
 /X
X0
□ flat (C.1.6)
  
Spec k  / Spec k[ϵ].

A morphism of first-order deformations (X, α) and (X ′ , α′ ) is a morphism

β : X → X ′ of schemes over k[ϵ] such that (β ×k[ϵ] k) ◦ α = α′ , or in other words
considering X and X ′ in cartesian diagrams (C.1.6), we require the restriction of β
to central fiber X0 to be the identity.
We say that X is trivial if X is isomorphic to X0 ×k S
k[ϵ] as first-order deformations
and locally trivial if there exists a Zariski-cover X = i Ui such that Ui is a trivial
first-order deformation of Ui ×k[ϵ] k ⊆ X0 .
Every morphism of deformations is necessarily an isomorphism. This is a
consequence of the following algebraic fact.
Lemma C.1.7. Let A be a ring, m ⊆ A be a nilpotent ideal (e.g., (A, m) is an
artinian local ring), and M → N be a homomorphism of A-modules. Assume that
N is flat over A. If M/mM → N/mN is an isomorphism, so is M → N .
Proof. If C := coker(M → N ), then C/mC = coker(M/mM → N/mN ) = 0. As
mn = 0 for some n, we obtain that C = mC = m2 C = · · · = mn C = 0. If
K := ker(M → N ), then the flatness of N implies that K/mK = ker(M/mM →
N/mN ) = 0. Thus K = mK = · · · = mn K = 0, and we see that M → N is an
isomorphism.
Proposition C.1.8. Every first-order deformation of a smooth affine scheme X0
over k is trivial. In other words, X0 is rigid.
Proof. Let X be a first-order deformation of X0 . Since X0 → Spec k is smooth, we
may apply the Infinitesimal Lifting Criterion for Smoothness (A.3.1) to construct a
lift X → X0 making the diagram

X0 _
id / X0
;
smooth
 
X / Spec k

commute. This induces a morphism X → X0 ×k k[ϵ] over k[ϵ] which restricts to

the identity on X0 , and thus is an isomorphism by Lemma C.1.7. See also [Har77,
Exc. II.8.7].

542
 If X0 is not smooth or affine, then first-order deformations are not necessarily
trivial. For example, if X0 = Spec k[x, y]/(xy) is the nodal affine plane curve, then
X = Spec k[x, y, ϵ]/(xy − ϵ) → Spec k[ϵ] is a non-trivial first-order deformation. On
the other hand, considering an elliptic curve E0 = V (y 2 z − x(x − z)(x − λz)) ⊆ P2
for λ ̸= 0, 1,
Eα = V (y 2 z − x(x − z)(x − (λ + αϵ)z)) ⊆ P2k[ϵ] (C.1.9)
defines a first-order deformation of E0 for every α ∈ k, and the assignment α 7→ Eα
defines a bijection between k and the set of isomorphism classes of first-order defor-
mations; this follows from Proposition C.1.11 since H1 (E0 , TE0 ) = H1 (E0 , OE0 ) = k.
In fact, the same formula (C.1.9) defines a global deformation of E0 over A1 with
two singular fibers.
To get a handle on locally trivial deformations, we need to understand automor-
phisms of the trivial deformation.

Lemma C.1.10. Let X0 = Spec A be an affine scheme over k and X = Spec A[ϵ]
be the trivial first-order deformation. There are identifications

{automorphisms X → X of first-order defs} ∼

= Derk (A, A) ∼
= HomA (ΩA/k , A).

Proof. The second equivalence is given by the universal property of the module
of differentials. An automorphism of X (as a first-order deformation) corresponds
to a k[ϵ]-algebra isomorphism ϕ : A ⊕ Aϵ → A ⊕ Aϵ which is the identity modulo
ϵ. Therefore, ϕ is determined by the images ϕ(a) = a + d(a)ϵ of elements a ∈ A
where d : A → A is k-linear map. The map ϕ is a ring homomorphism if and only if
aa′ +d(aa′ )ϵ = (a+d(a)ϵ)(a′ +d(a′ )ϵ) = aa′ +(ad(a′ )+a′ d(a))ϵ for elements a, a′ ∈ A,
which this translates into the condition that d : A → A is a k-derivation.

For a scheme X0 over k, let Def(X0 ) and Def lt (X0 ) denote the sets of isomorphism
classes of first-order and locally trivial first-order deformations.

Proposition C.1.11. For a scheme X0 of finite type over k with affine diagonal,
there is a bijection
Def lt (X0 ) ∼
= H1 (X0 , TX0 ),
where TX0 = H omOX0 (ΩX0 /k , OX0 ). The trivial deformation corresponds to 0 ∈
H1 (X0 , TX0 ). If in addition X0 is smooth over k, there is a bijection

Def(X0 ) ∼
= H1 (X0 , TX0 ).

Proof. Let X → Spec k[ϵ] be a locally trivial first-order deformation of X0 . Choose

∼
an affine cover {Ui } of X0 and isomorphisms ϕi : Ui ×k k[ϵ] → X ∩ Ui . Letting Uij =
−1
Ui ∩ Uj , we have automorphisms ϕj |Uij ×k k[ϵ] ◦ ϕi |Uij ×k k[ϵ] of the trivial deformation
Uij ×k k[ϵ], which corresponds by Lemma C.1.10 to elements ϕij ∈ TX0 (Uij ). Since
ϕij ◦ ϕjk = ϕik on Uijk := Ui ∩ Uj ∩ Uk , we have that ϕij + ϕjk = ϕik ∈ TX0 (Uijk ).
Recall that H1 (X0 , TX0 ) can be computed using the Čech complex

0 / L TX0 (Ui ) d0
/L TX0 (Uij )
d1
/L TX0 (Uijk )
i i,j i,j,k

 / (sij |U
(sij ) ijk
− sik |Uijk + sjk |Uijk )ijk

As {ϕij } ∈ i,j TX0 (Uij ) is in the kernel of d1 , it defines an element of H1 (X0 , TX0 ) =
L

ker(d1 )/ im(d0 ). Conversely, given an element of H1 (X0 , TX0 ) and a choice of

543
representative {ϕij } ∈ ker(d1 ), then viewing each ϕij as an automorphism of the
trivial deformation of Uij , we may glue together the trivial deformations Ui ×k k[ϵ]
along Uij ×k k[ϵ] via ϕij to construct a global first-order deformation X of X0 . The
final statement follows as all first-order deformations are locally trivial when X0 is
smooth (Proposition C.1.8). See also [Har77, Exc. III.4.10 and Ex. III.9.13.2].
Example C.1.12. If C is a smooth projective curve of genus g ≥ 2, then
SD
TMg ,[C] = H1 (C, TC ) = H0 (C, Ω⊗2
C/k ),

which by Riemann–Roch is a 3g − 3 dimensional vector space.

Exercise C.1.13 (easy). Use the Euler exact sequence to show that H1 (Pn , TPn ) = 0
and conclude that every first-order deformation of Pn is trivial, i.e., Pn is rigid.
Exercise C.1.14 (moderate). Let C0 be a smooth and proper curve over k with
marked points p1 , . . . , pn ∈ C0 (k). Let Def(C0 , pi ) denote the set of first-order
deformations of (C0 , pi ), i.e., flat morphisms C → Spec k[ϵ] with n sections σi
extending (C0 , pi ). Show that

Def(C0 , pi ) ∼
= H1 C0 , TC0 (− i pi ) .
P

Remark C.1.15. More generally, if X0 is generically smooth and a local complete

intersection over k, then

Def(X0 ) = Ext1OX0 (ΩX0 , OX0 ).

Observe that when X0 is smooth, this recovers Proposition C.1.11. This also holds
if X0 = Spec R is the spectrum a complete noetherian local ring with an isolated
singularity such that R is a complete intersection ring. It is easy to see how a
first-order deformation X of X0 induces an extension class: the closed immersion
X0 ,→ X is defined by an ideal J ⊂ OX with J 2 = 0 such that J ∼ = OX0 as an
OX0 -module, and the usual right exact sequence

OX0 → ΩX |X0 → ΩX0 → 0

is also left exact because the kernel of the left map is on one hand torsion free (as a
subsheaf of OX0 ) and on the other hand torsion (as generic smoothness implies that
the map is generically injective).
Exercise C.1.16 (good practice).
(1) Show that there is a bijection Def(k[x, y]/(xy)) ∼
= k, where an element t ∈ k
corresponds to the first-order deformation Spec k[x, y, ϵ]/(xy − tϵ).
(2) Classify first-order deformations of the Ak -singularity k[x, y]/(y 2 − xk+1 ).

C.1.3 First order deformations of vector bundles and coherent

sheaves
Definition C.1.17. Let X be a scheme over k and E0 be a coherent sheaf. A
first-order deformation of E0 is a pair (E, α) where E is a coherent sheaf on X ×k k[ϵ]
∼
flat over k[ϵ] and α : E0 → E|X is an isomorphism. Pictorially, we have

E0 E
flat/k[ϵ]

X / Xk[ϵ] .

544
 A morphism (E, α) → (E ′ , α′ ) of first-order deformations is a morphism β : E →
E (equivalently an isomorphism by Lemma C.1.7) of coherent sheaves on X ′ such
′

that α′ = β|X ◦ α. We say that (E, α) is trivial if it isomorphic as first-order

deformations to (p∗ E0 , id) where p : Xk[ϵ] → X.
Proposition C.1.18. Let X be a scheme over k and E0 be a coherent sheaf. There
is a bijection

{first-order deformations (E, α) of E0 }/∼ ∼

= Ext1OX (E0 , E0 ).

Under this correspondence, the trivial deformation corresponds to 0 ∈ Ext1OX (E0 , E0 ).

If in addition E0 is a vector bundle (resp., line bundle), then each first-order
deformation is a vector bundle (resp., line bundle), and the set of isomorphism
classes of first-order deformations of E0 is bijective to H1 (X, E ndOX (E0 )) (resp.,
H1 (X, OX )).
Proof. If (E, α) is a first-order deformation, then since E is flat over k[ϵ], we may
ϵ
tensor the exact sequence 0 → k → − k[ϵ] → k → 0 of k[ϵ]-modules with E to obtain
ϵ
an exact sequence 0 → E0 → − E → E0 → 0 (after identifying E ⊗k[ϵ] k with E0 via
α). Since ExtOX (E0 , E0 ) parameterizes isomorphism classes of extensions [Har77,
1

Exc. III.6.1], we have constructed an element of Ext1OX (E0 , E0 ). Conversely, given

α
an exact sequence 0 → E0 − → E → E0 → 0, then E is a coherent sheaf on Xk[ϵ] and
is flat over k[ϵ] by the flatness criterion over the dual numbers (see Remark A.2.7).
The restriction E|X is isomorphic to E0 via α. See also [Har10, Thm. 2.7].
Remark C.1.19. The classifications of Proposition C.1.3, C.1.11, and C.1.18 give a vec-
tor space structure to the set of isomorphism classes of first-order deformations. This
vector space structure can also be witnessed as a consequence of Rim–Schlessinger’s
homogeneity condition; see Lemma C.4.9. For each case, the operations of scalar
multiplication and addition afford geometric descriptions.

C.2 Higher-order deformations and obstructions

We need some new obstructions!
Bjorn Poonen [Poo17]

Let M be a moduli problem, E ∈ M(A) be an object defined over a ring A, and

let A′ ↠ A be a surjection of rings with square-zero kernel. This section addresses
the following two questions:
(1) Does E deform to an object E ′ ∈ M(A′ )?
(2) If so, can we classify all such deformations?
Pictorially, we have:
E E′

Spec A  / Spec A′ .

Note that since J = ker(A′ → A) is square-zero, J = J/J 2 is naturally a module

over A = A′ /J. Question (1) asks whether there is an ‘obstruction’ to the existence
of a deformation E ′ while (2) seeks to classify all higher-order deformations given
that there is no obstruction.

545
 An interesting case is when A and A′ are local artinian algebras with residue
field k and the kernel J = ker(A′ → A) satisfies mA′ J = 0 (which implies that
J 2 = 0). In this case, J = J/mA′ J is naturally a vector space over k = A′ /mA′ .
Setting E0 := E|k ∈ M(k), we can view E as a deformation of E0 over A, and we
are attempting to classify the higher-order deformations over A′ . If there are no
obstructions to deforming, then the Infinitesimal Lifting Criterion for Smoothness
(3.7.1) implies that M is smooth at [E0 ].
The previous section studied the specific case when A = k and A′ = k[ϵ] in
which case deformations of an object E0 ∈ M(k) over A′ correspond to first-order
deformations. In this case, the obstruction vanishes as there is always the trivial
deformation (i.e., the pullback of E0 along Spec k[ϵ] → Spec k). Other examples of
A′ → A to keep in mind are k[x]/xn+1 ↠ k[x]/xn and Z/pn+1 ↠ Z/pn where we
inductively attempt to deform E0 over the nilpotent thickenings Spec k[x]/xn+1 ,→ A1
and Spec Z/pn+1 ,→ Spec Z.

C.2.1 Higher-order embedded deformations

Definition C.2.1. Let A′ ↠ A be a surjection of rings. Let X ′ be a scheme over
A′ and set X := X ′ ×A′ A. Let Z ⊆ X be a closed subscheme flat over A. A
deformation of Z ⊆ X over A′ is a closed subscheme Z ′ ⊆ X ′ flat over A′ such that
Z ′ ×A′ A = Z as closed subschemes of X. Pictorially, a deformation is a filling of
the cartesian diagram
 / ′
9X 9X
cl cl
,  +
Z /Z ′

flat flat
$   % 
Spec A  / Spec A′ .

The formulation of the next proposition uses the following notion: a torsor of an
abstract group G is set with a free and transitive of G.

Proposition C.2.2. Let X be a scheme over k with affine diagonal (e.g., separated).
Let A′ ↠ A be a surjection of artinian k-algebras with residue field k such that
mA′ J = 0 where J = ker(A′ → A). Let Z ⊆ XA be a closed subscheme flat over A,
and let Z0 = Z ×A k. Then
(1) If there exists a deformation Z ′ ⊆ XA′ of Z ⊆ XA over A′ , then the set of
such deformations is a torsor under Ext0OX (IZ0 , OZ0 ⊗k J).
(2) There exists an element obZ ∈ Ext1OX (IZ0 , OZ0 ⊗k J) (depending on Z and
A′ ↠ A) such that there exists a deformation of Z ⊆ X over A′ if and only if
obZ = 0.
More generally, if A′ ↠ A is a surjection of k-algebras with square-zero kernel J,
then deformations and obstructions of Z ⊂ XA are classified by ExtiOX (IZ , OZ ⊗A J)
A
for i = 0, 1.

Note that if Z0 ⊆ X is a local complete intersection, then IZ0 /IZ0 is a vector

bundle and ExtiOX (IZ0 , OZ0 ⊗k J) = Hi (Z0 , NZ0 /X ⊗k J).

Proof. We first handle the affine case. Write X = Spec B0 , X = Spec B, X ′ =

Spec B ′ , Z = Spec B/I, and Z0 = Spec B0 /I0 . If there exists a deformation Z ′ =

546
Spec B ′ /I ′ , then there is an exact diagram

0 0 0

  
0 / I ⊗A J / I′ /I /0

  
0 / B ⊗A J / B′ /B /0

  
0 / (B/I) ⊗A J / B ′ /I ′ / B/I /0

  
0 0 0

The exactness of the bottom row (resp., middle row) is equivalent to the flatness
of B ′ /I ′ (resp., B ′ ) over A′ by the Local Criterion of Flatness (A.2.6), while the
exactness of the left column follows from the flatness of B/I over A. Conversely, an
exact diagram defines a deformation Z ′ = Spec B ′ /I ′ .
We will define an action HomB (I, (B/I) ⊗A J) on the set of deformations. Given
ϕ ∈ HomB (I, (B/I) ⊗A J) and a deformation Z ′ = Spec B ′ /I ′ , define I ′′ ⊆ B ′ as
the set of elements x′′ ∈ B ′ with the following property: the image x′′ ∈ B of
x′′ lies in I and if x′ ∈ I ′ is a lifting of x′′ ∈ I, the image of x′′ − x′ ∈ B ⊗A J
in (B/I) ⊗A J is equal to ϕ(x′′ ) (noting that this condition is independent of
the choice of lifting x′ ). One checks that Spec B ′ /I ′′ is another deformation. On
the other hand, given two deformations defined by ideals I ′ and I ′′ , we define
ϕ : I → (B/I) ⊗A J by ϕ(x) = x′′ − x′ , where x′ ∈ I ′ and x′′ ∈ I ′′ are lifts of x
(which forces x′′ − x′ ∈ B ⊗A J). One checks that this is a B-module homomorphism
providing an inverse to the above construction. Since J is a k-vector space, there is an
identification HomB (I, (B/I) ⊗A J) = HomB0 (I0 , B0 /I0 ⊗k J), and this construction
globalizes to X to establish (1).
We will prove (2) under the hypothesis that there is an open cover {Ui } of X
and deformations Zi′ ⊆ XA′ ∩ Ui of Z ∩ Ui ,→ XA ∩ Ui over A′ . This is satisfies if
Z0 ,→ X is a local complete intersection; see [Kol96, Lem. 2.12]. The restrictions
Zi′ ∩ Uij and Zj′ ∩ Uij are related by an element ϕij ∈ H0 (Uij , NZ0 /X ⊗A J) which
in turn defines a Čech 1-cocycle (ϕij ) ∈ H1 (X, NZ0 /X ⊗A J). We leave the reader
to check that the vanishing of (ϕij ) characterizes whether there is a deformation
of Z ⊆ XA over A′ . See also [FGAIII, §5], [Art69b, Lem. 6.7], [Kol96, Prop. 2.5],
[Vis97, Thm. 2.5], and [Har10, Thm. 6.2].

C.2.2 Higher-order deformations of schemes

Definition C.2.3. Let A′ ↠ A be a surjection of rings and X → Spec A be a flat
morphism of schemes. A deformation of X → Spec A over A′ is a flat morphism
∼
X ′ → Spec A′ together with an isomorphism α : X → X ′ ×A′ A over A, or in other
words a cartesian diagram

X / X′

flat □ flat
  
Spec A  / Spec A′ .

547
 A morphism of deformations over A′ is a morphism of schemes over A′ restricting
to the identity on X. By Lemma C.1.7, every morphism of deformations is an
isomorphism.

Proposition C.2.4 (Higher-order Deformations of Complete Intersections). Let

X0 be a scheme of finite type over a field k such that X0 is generically smooth and
a local complete intersection. Let A′ ↠ A be a surjection of artinian local rings with
residue field k such that mA′ J = 0 where J := ker(A′ → A). If X → Spec A is a
deformation of X0 , then:
(1) The group of automorphisms of a deformation X → Spec A over A′ is bijective
to Ext0OX0 (ΩX0 , J).
(2) If there exists a deformation of X → Spec A over A′ , then the set of isomor-
phism classes of all such deformations is a torsor under Ext1OX0 (ΩX0 , J).
(3) There is an element obX ∈ Ext2OX0 (ΩX0 , J) with the property that there exists
a deformation of X → Spec A over A′ if and only if obX = 0.

In particular, if X0 is smooth, then automorphisms, deformations and obstruc-

tions are classified by Hi (X0 , TX0 ⊗k J) for i = 0, 1, 2.

Proof. We will prove the smooth case. For the general case, an explicit argument
is given in [Vis97, Thm. 4.4]; alternatively, since X0 is generically smooth and a
local complete intersection, the cotangent complex X0 is quasi-isomorphic to ΩX0
(Theorem C.3.1(3)) and thus the result follows from the fact that the cotangent
complex controls automorphisms, deformations, and obstructions (Theorem C.3.6).
When X0 = Spec B0 is an affine scheme, the same argument of Lemma C.1.10
shows that group of automorphisms of X ′ is identified with HomB (ΩB/k , J). Since
U 7→ HomOU (ΩX0 /k |U , J) and U 7→ Aut(X ′ ∩ U/X ∩ U ) are sheaves, Part (1) follows.
Part (2) follows from a similar argument to Proposition C.1.11. Indeed, fix a de-
formation X ′ → Spec A′ . If {Ui } is an affine cover of X0 , by the Infinitesimal Lifting
∼
Criterion for Smoothness (A.3.1), there are trivializations ϕi : Ui ×k A′ → X ′ ∩ Ui .
Then ϕij = ϕ−1j ◦ ϕi defines an automorphism of the trivial deformation correspond-
ing by (1) to an element ϕij ∈ H0 (Uij , TX0 ⊗ J). An element of H1 (X0 , TX0 ⊗k J)
defines a Čech 1-cocycle (ψij ) with respect to the covering {Ui }, and ϕij + Ψij
defines isomorphisms of the restrictions of the trivial deformations over Ui , which
glue to a global deformation X ′′ → Spec A′ . Conversely, if X ′′ → Spec A′ is an-
other deformation, there are isomorphisms ϕi : X ′ ∩ Ui → X ′′ ∩ Ui for each i,
and ϕij = ϕ−1j |X ′ |Uij ◦ ϕi |X ′ |Uij ∈ H (Uij , TX0 ⊗ J) defines a Čech 1-cycle (ϕij ) in
0

H1 (X0 , TX0 ⊗ J).

For (3), we again let {Ui } be an affine cover. Each deformation X ∩ Ui of
∼
X0 ∩ Ui is trivial, and induces automorphisms ϕij : Uij ×k A → Uij ×k A. By the
Infinitesimal Lifting Criterion for Smoothness (A.3.1), we may choose extensions
∼
ϕ′ij : Uij ×k A′ → Uij ×k A′ of ϕij . This defines gluing data for a deformation X ′
of X if the cocycle condition ϕ′jk ◦ ϕ′ij = ϕ′ik on the triple intersections Uijk ×k A′
is satisfied. The automorphism Ψ′ijk = (ϕ′ik )−1 ◦ ϕ′jk ◦ ϕ′ij restricts to the identity
on Uijk ×k A and thus defines an element of H0 (Uijk , TX0 ⊗ J). Consider the Čech

548
complex for F = TX0 ⊗ J with respect to {Ui }:
d1
/ d2
/
L L L
i,j F (Uij ) i,j,k F (Uijk ) i,j,k,l F (Uijkl )

(sij )  / (sij − sik + sjk )ijk

 / (sijk − sijl + sikl − sjkl )ijkl .
(sijk )

One checks that d2 (Ψ′ijk ) = 0 and thus (Ψ′ijk ) is a Čech 2-cocycle defining an element
of H2 (X, TX/A ⊗ J). If this element is zero, i.e., there exists (sij ) mapping to (Ψ′ijk ),
then modifying the automorphisms ϕ′ij by sij defines isomorphisms ϕ′′ij satisfying
the cocycle condition. See also [Ser06, Prop. 1.2.12] and [Har10, Cor. 10.3].
Exercise C.2.5 (Interpretation of deformations and obstruction using gerbes). Let
X → Spec A be a smooth morphism of schemes. Consider the category G over
Sch /X whose objects over S → X are cartesian diagrams
 / S′
S
□
  
Spec A  / Spec A′

where S → Spec A is the composition S → X → Spec A. A morphism (S →

X, S ,→ S ′ → Spec A′ ) → (T → X, T ,→ T ′ → Spec A′ ) is the data of a morphism
ϕ : S ′ → T ′ over A′ such that ϕ restricts to a morphism S → T over X.
(a) Show that G is a gerbe banded by the sheaf of groups TX/A ⊗A J on X. (Hint:
See Definition 6.4.10 for the definition of a banded gerbe.)
(b) Give an alternate proof of Proposition C.2.4. (Hint: For part C.2.4(3), use
Exercise 6.4.31.)
Exercise C.2.6 (Deformations of principal G-bundles). Let G be a smooth affine
algebraic group over a field k with Lie algebra g. Let X ,→ X ′ be a closed immersion
of finite type k-schemes defined by a square-zero sheaf of ideals J and assume that
X has affine diagonal. Show that
(1) The group of automorphisms of a deformation P ′ → X ′ of P → X is bijective
to H0 (X, g ⊗k J).
(2) If there exists a deformation over X ′ , then the set of isomorphism classes of
all such deformations is a torsor under H1 (X, g ⊗k J).
(3) There is an element obX ∈ H2 (X, g ⊗k J) with the property that there exists
a deformation over X ′ if and only if obX = 0.
Example C.2.7 (Abelian varieties). If X0 is an abelian variety over C of dimension
n, then it turns out that deforming X0 as an abstract scheme is equivalent to
deforming it as an abelian variety, and thus obstructions to deforming X0 as an
abelian variety also live in H2 (X0 , TX0 ). Using that ΩX0 = OX
n
0
is trivial and the
Hodge symmetries,
2
^
H2 (X0 , TX0 ) = H2 (X0 , OX0 )⊕n = H0 (X0 , n ⊕n
OX0
)

is nonzero. Nevertheless, Grothendieck and Mumford showed that the obstruction

obX ∈ H2 (X, TX/A ⊗A J) vanishes for every deformation problem! This shows that
abelian varieties are unobstructed, and their moduli space is smooth. See [Oor71].

549
C.2.3 Higher-order deformations of morphisms
For the deformation theory of pointed stable curves, we will need the following
enhancement of Higher-order Deformations of Complete Intersections (C.2.4) to the
following deformation problem where we are assuming that a deformation Z ′ over
A′ already exists:
 / Z ′
Z _ _
□
  
X / X′

□
  
Spec A  / Spec A′ .

Proposition C.2.8 (Higher-order Deformations of Morphisms with Fixed Source).

Let X0 be a scheme of finite type over a field k such that X0 is generically smooth
and a local complete intersection, and let Z0 ⊂ X0 be a closed subscheme with ideal
sheaf IZ0 . Let A′ ↠ A be a surjection of artinian local rings with residue field k
such that mA′ J = 0 where J := ker(A′ → A). If X → Spec A is a deformation of
X0 and Z ′ → Spec A′ is a deformation of Z0 , then:
(1) The group of automorphisms of a deformation Z ′ ⊆ X ′ of Z ⊆ X over A′ is
bijective to Ext0OX0 (ΩX0 , IZ0 ⊗k J).
(2) If there exists a deformation Z ′ ⊆ X ′ over A′ , then the set of isomorphism
classes of all such deformations is a torsor under Ext1OX0 (ΩX0 , IZ0 ⊗k J).
(3) There is an element obZ/X ∈ Ext2OX0 (ΩX0 , IZ0 ⊗k J) with the property that
there exists a deformation Z ′ ⊆ X ′ over A′ if and only if obX = 0.
Proof. See [Vis97, Thm. 5.4].
Exercise C.2.9 (Higher-order deformations of morphisms). Assume that X and
Y are proper A, and that f∗ OX = OY and R1 f∗ OX = 0. Show that the functor
taking a deformation f ′ : X ′ → Y ′ of f : X → Y over A′ to the deformation X ′ over
X over A′ induces an isomorphism of categories.
Hint: Given a deformation X ′ over X, define Y ′ as the ringed space (Y, f∗ OX ′ ).
Use the conditions of f and the flatness of X ′ over A′ to show that Y ′ is a scheme
flat over A′ . See also [Ran89, Thm. 3.3], [Vak06, §5.3], and [SP, Tag 0E3X].
For further background on deformations of morphisms, see [Ser06, §3.4].

C.2.4 Higher-order deformations of vector bundles

Definition C.2.10. Let A′ ↠ A be a surjection of rings. Let X ′ be a scheme over
A′ and set X := X ′ ×A′ A. Given a coherent sheaf E on X flat over A, a deformation
of E over A′ → A is a pair (E ′ , α) where E ′ is a coherent sheaf on X ′ flat over A′
and α : E → E ′ |X is an isomorphism. Pictorially, we have

E E′
flat/A′

flat/A
X / X ′.

A morphism (E, α) → (E ′ , α′ ) of deformations is a morphism β : E → E ′ of

coherent sheaves on XA′ such that α′ = β|X ◦ α. By Lemma C.1.7, every morphism
of deformations is an isomorphism.

550
Proposition C.2.11. Let X be a scheme over a field k. Let A′ ↠ A be a surjection
of artinian local rings with residue field k such that mA′ J = 0 where J := ker(A′ →
A). Let E be a coherent sheaf on XA and set E0 = E|X .
(1) The group of automorphisms of a deformation E ′ of E over A′ is bijective to
Ext0OX (E0 , E0 ⊗k J).
(2) If there exists a deformation of E over A′ , then the set of isomorphism classes
of all such deformations is a torsor under Ext1OX (E0 , E0 ⊗k J)..
(3) There is an element obE ∈ Ext2OX (E0 , E0 ⊗k J) with the property that there
exists a deformation of E over A′ if and only if obE = 0.
More generally, if A′ ↠ A is a surjection of k-algebras with square-zero kernel J,
automorphisms, deformations, and obstructions are classified by ExtiOX (E, E ⊗k J)
A
for i = 0, 1, 2.

Remark C.2.12. If E is a vector bundle (resp., line bundle), automorphisms, defor-

mations, and obstructions are classified by Hi (X, E ndOX (E0 ) ⊗k J) (resp., Hi (X, J))
for i = 0, 1, 2. In the case of line bundles, the obstruction can be realized by the
exact sequence Pic(X ′ ) → Pic(X) → H2 (X, J) induced from taking cohomology of
the short exact sequence 0 → J → Gm,X ′ → Gm,X → 1.

Proof. For (1), since E ′ is flat over A′ , tensoring the exact sequence 0 → OX ′ ⊗k J →
OX ′ → OX → 0 with E yields a short exact sequence

0 → E ⊗k J → E ′ → E → 0.

∼
If α : E ′ → E ′ is an automorphism with α|E = id, then α − id defines a map
E ′ → E ⊗k J, which factors to give a map ϕ : E0 → E0 ⊗k J. Conversely, if ϕ is a
ϕ
homomorphism, then the sum of the identity and E → E0 − → E0 ⊗k J defines an
automorphism.
For the rest of the proof, we will assume that E is a vector bundle. For (2), let E ′
∼
be a deformation. Let {Ui } be an affine covering of X with trivializations ϕi : E ′ |Ui →
⊕ri
OUi ×k A′ . Then ϕij = ϕ−1
j ◦ϕi is an automorphism of E |Uij , which by (1) corresponds
′

to an element of H (Uij , E ndOX (E)). A element of H (X, E ndOX (E)) defines a Čech
0 1

1-cocycle (ψij ) with ψij ∈ H0 (Uij , E ndOX (E)), and ϕij + ψij defines isomorphisms
of the trivial vector bundle over Uij ×k A′ which glue to a deformation E ′ over XA′ .
Conversely, if E ′ and E ′′ are two deformations, there exists a covering {Ui } and
isomorphisms αi : E ′ |Ui → E ′′ |Ui . The automorphisms ϕij = ϕ−1
j ◦ϕi defines elements
in H (Uij , E ndOX (E)), and (ϕij ) defines a Čech 1-cycle in H (X, E ndOX (E)).
0 1

⊕ri ∼
For (3), let {Ui } be an affine covering with trivializations ϕi : OUi ×Spec A
→ E|Ui
−1
yielding automorphisms ϕij = ϕj ◦ ϕi of the trivial vector bundles on Uij ×k A.
Choose automorphisms ϕ′ij of the trivial vector bundles on Uij ×k A extending
ϕij . The automorphisms Ψ′ijk = (ϕ′ik )−1 ◦ ϕ′jk ◦ ϕ′ij correspond to elements of
H0 (Uijk , TX0 ⊗ J) and define a Čech 2-cocycle obE := (Ψ′ijk ) ∈ H2 (X, E ndOX (E)).
See also [Har10, Thm. 7.1], [HL10, §2.A.6], and [SP, Tag 08VW].

Exercise C.2.13. Give an alternative proof of Proposition C.2.11 using the tech-
nique outlined in Exercise C.2.5. bundle case

551
C.3 Cotangent complex

Computers are useless. They can only give you answers.

Pablo Picasso

In this chapter, we summarize properties of the cotangent complex of a mor-

phism of schemes as introduced in [Ill71], globalizing work of André [And67] and
Quillen [Qui68, Qui70] on the cotangent complex of a ring homomorphism. A major
advantage of the cotangent complex is that it allows us to describe the deformations
and obstruction of singular schemes (Theorem C.3.6).

C.3.1 Properties of the cotangent complex

Theorem C.3.1. For every morphism f : X → Y of schemes (resp., finite type
morphism of noetherian schemes), there exists a complex

LX/Y : · · · → L−1 0
X/Y → LX/Y → 0

of flat OX -modules with quasi-coherent (resp., coherent) cohomology, whose image

− −
in DQCoh (OX ) (resp., DCoh (OX )) is also denoted by LX/Y . It satisfies the following
properties.
(1) H0 (X, LX/Y ) ∼
= ΩX/Y .
(2) The morphism f is smooth if and only if f is locally of finite presentation
and LX/Y is a perfect complex supported in degree 0. In this case, there is a
∼
quasi-isomorphism LX/Y → ΩX/Y with ΩX/Y in degree 0.
(3) If f is flat and finitely presented, then f is syntomic (Definition A.3.17) if and
only if LX/Y is a perfect complex supported in degrees [−1, 0]. Explicitly, if f
factors as a complete intersection X ,→ X e defined by a sheaf of ideals I and
a smooth morphism X e → Y , then LX/Y is quasi-isomorphic to 0 → I/I 2 →
ΩX/Y
e |X → 0 with ΩX/Y in degree 0. If in addition f is generically smooth,
then LX/Y ≃ ΩX/Y .
(4) If
g′
X′ /X

□ f
 
Y′
g
/Y

is a cartesian diagram with either f or g flat (or more generally f and g are
tor-independent), then there is a quasi-isomorphism g ′∗ LX/Y → LX ′ /Y ′ . (Note
that without any flatness condition g ′∗ ΩX/Y ≃ ΩX ′ /Y ′ .)
f
(5) If X −→ Y → Z is a composition of morphisms of schemes, then there is an
−
exact triangle in DQCoh (OX )

f ∗ LY /Z → LX/Z → LX/Y → f ∗ LY /Z [1].

552
 This induces a long exact sequence on cohomology

··· H−2 (LX/Y )

H−1 (f ∗ LX/Z ) H−1 (LX/Z ) H−1 (LX/Y )

f ∗ ΩY /Z ΩX/Z ΩX/Y 0,

extending the usual right exact sequence on differentials [Har77, II.8.12]. (Note
that if f is smooth, then H −1 (LX/Y ) = 0 and f ∗ ΩY /Z → ΩX/Z is injective.)

Proof. See [Ill71, II.1.2.3] and [SP, Tag 08T2] for the definition and construction of
the cotangent complex of a morphism of schemes (and more generally for morphisms
of ringed topoi). For (1)–(5), see [Ill71, II.[Link], II.3.1.2, II.3.2.6, II.2.2.3 and
II.2.1.2] and [SP, Tags 08UV, 0D0N, 0FK3, 08QQ, and 08T4] (noting that [SP,
Tag 08RB] relates the naive cotangent complex N LX/Y to LX/Y ). For the final
statement of (3), we observe that the right exact sequence
d
I/I 2 −
→ ΩX/Y
e |X → ΩX/Y → 0, (C.3.2)

is also left exact: as X ,→ X e is a complete intersection, I/I 2 is a vector bundle

and thus ker(d) is torsion free, but, on the other hand, since X → Y is generically
smooth, d is generically injective so that ker(d) is torsion. It follows ker(d) = 0 and
d
that LX/Y = [I/I 2 −
→ ΩX/Y
e |X ] is quasi-isomorphic to ΩX/Y .

C.3.2 Truncations of the cotangent complex

The definition of the cotangent complex relies on simplicial techniques and we will
not attempt an exposition here. We will however give an explicit description of its
truncation, which often suffices for applications. First, if X → Y factors as a closed
immersion X ,→ P defined by a sheaf of ideals I and a smooth morphism P → Y ,
then the truncation τ≥−1 (LX/Y ) of LX/Y in degrees [−1, 0] is quasi-isomorphic
to 0 → I/I 2 → ΩX/Y → 0 (with ΩX/Y in degree 0). If X → Y is syntomic
(e.g., smooth), then X ,→ Ye is a regular immersion, I/I 2 is a vector bundle, and
LX/Y ∼ = τ≥−1 (LX/Y ) (Theorem C.3.1(3)).
For a morphism X = Spec A → Spec B = Y of affine schemes, Lichtenbaum–
Schlessinger [LS67] offer an explicit description of the truncation τ≥−2 (LA/B ) of
LX/Y = LB/A . Choose a polynomial ring P = B[xi ] (with possibly infinitely many
generators) and a surjection P ↠ A as B-algebras with kernel I. Choose a free
P -module F and a surjection p : F ↠ I of P -modules with kernel K = ker(p).
Let K ′ ⊆ K be the submodule generated by p(x)y − p(y)x for x, y ∈ F . Then
τ≥−2 (LB/A ) is quasi-isomorphic to the complex of A-modules

K/K ′ → F ⊗P A → ΩP/B ⊗P A, (C.3.3)

with the last term in degree 0. See also [SP, Tag 09CG].
For i = 0, 1, 2, one defines the T i functors on the category of A-modules by

T i (A/B, −) := Hi (HomA (LA/B , −)) = Hi (HomA (τ≥2 LA/B , −)),

which describe deformations of schemes. See also [LS67, §2.3] and [Har10, §1.3].

553
C.3.3 Extensions of algebras and schemes
Definition C.3.4. An extension of a morphism X → S of schemes by a quasi-
coherent OX -module J is a short exact sequence
0 → J → OX ′ → OX → 0
where X ,→ X ′ is a closed immersion of schemes defined by the sheaf of ideals
J ⊆ OX ′ with J 2 = 0. (Note that the condition J 2 = 0 implies that the J ⊂ OX ′ is
naturally a OX -module.) The trivial extension is X[J] := (X, OX ⊕ J) where the
ring structure is defined by J 2 = 0.
A morphism of extensions is a morphism of short exact sequences which is
the identity on J and OX . We let ExalS (X, J) be the category of extensions of
X → S by J, and ExalS (X, J) be the set of isomorphism classes. If X = Spec A
and S = Spec R is affine, we write ExalR (A, J).
Geometrically, an extension is a commutative diagram of schemes

X / X′

 {
Spec R

such that J ∼ = ker(OX ′ → OX ) and J 2 = 0. The set of extensions ExalS (X, J)

is functorial with respect to OX -module maps J → J ′ and morphisms X ′ → X
of S-schemes, and inherits the structure of a module over Γ(X, OX ). In fact, the
groupoid ExalS (X, J) is a Picard category, and the prestack over Sch/S whose fiber
category over f : T → S is ExalT (XT , f ∗ J) is a Picard stack ; see [Ill71, III.1.1.5] and
[SGA4, XVIII.1.4]. Given an exact sequence 0 → J ′ → J → J ′′ → 0 of OX -modules,
there is an exact sequence

0 DerOS (OX , J ′ ) DerOS (OX , J) DerOS (OX , J ′′ )

ExalS (X, J) ExalS (X, J) ExalS (X, J ′′ ).

Given a morphism f : X → Y of S-schemes, there is an exact sequence

0 DerOY (OX , J) DerOS (OX , J) DerOS (OY , f∗ J)

ExalY (X, J) ExalS (X, J) ExalS (Y, f∗ J).

See [EGA, [Link]] and [Ill71, III.[Link], III.[Link]]. The top row of the second
diagram above is realized by applying HomOX (−, J) to the right exact sequence
f ∗ ΩY /S → ΩX/S → ΩX/Y → 0 and using the identifications HomOX (ΩX/Y , J) =
DerOY (OX , J), HomOX (ΩX/S , J) = DerOS (OX , J), and HomOX (f ∗ ΩY /S , J) =
HomOY (ΩY /S , f∗ J) = DerOS (OY , f∗ J).

C.3.4 The cotangent complex and deformation theory

Theorem C.3.5. If X → Y is a morphism of schemes and J is a quasi-coherent
OY -module, there is a natural isomorphism
ExalY (X, J) ∼
= Ext1OX (LX/Y , J).

554
Proof. See [Ill71, III.1.2.3].
By applying HomOX (LX/Y , −) to the exact sequence 0 → J ′ → J → J ′′ and
HomOX (−, J) to the exact triangle f ∗ LY /Z → LX/Z → LX/Y allows us to extend
the two above six-term exact sequences to long exact sequences. When X =
Spec A → Spec B = Y is a morphism of affine schemes, using the T i functors of
§C.3.2, the above equivalence translates to ExalB (A, J) = T 1 (A/B, J), which can
also be explicitly using the truncated cotangent complex (C.3.3); see [LS67, 4.2.2]
and [Har10, Thm. 5.1]. See [LS67, 2.3.5-6] and [Har10, Thms. 3.4-5] for how the T i
functors extend the above six-term sequences nine-term sequences.
Theorem C.3.6. Consider the following deformation problem
 / X′
X
f f′
  
Y / Y′
i

where f : X → Y is a morphism of schemes and i : Y ,→ Y ′ is a closed immersion

of schemes defined by an ideal sheaf J ⊆ OY ′ with J 2 = 0. A deformation is a
morphism f ′ : X ′ → Y ′ making the above diagram cartesian, and a morphism of
deformations is a morphism over Y ′ restricting to the identity on X.
(1) The group of automorphisms of a deformation f ′ : X ′ → Y ′ is isomorphic to
Ext0OX (LX/Y , f ∗ J).
(2) If there exists a deformation, then the set of deformations is a torsor under
Ext1OX (LX/Y , f ∗ J).
(3) There exists an element obX ∈ Ext2OX (LX/Y , f ∗ J) with the property that there
exists a deformation if and only if obX = 0.
Proof. See [Ill71, III.2.1.7] and [SP, Tag 08UZ]. See also [LS67, 4.2.5] and [Har10,
Thm. 10.1] for descriptions in the affine case using the truncated cotangent complex.

As a reality check, if f : X → Spec A is syntomic (e.g., smooth) and A′ ↠ A is a

surjection of rings with square-zero kernel J, then the identification

ExtiOX (LX/A , f ∗ J) = ExtiOX (ΩX/A , J)

recovers Proposition C.2.4.

Remark C.3.7. There are analogous results for other deformation problems. For
instance, for the deformation problem
 / X′
X
f

"  #
Y / Y′

    
Z / Z ′,

where the horizontal morphisms are closed immersions defined by square-zero ideal
sheaves JX , JY and JZ , then automorphisms, deformations, and obstructions are
classified by ExtiOX (Lf ∗ LY /Z , JX ) for i = −1, 0, 1 [Ill71, III.2.2.4].

555
C.4 Versal formal deformations and Rim–Schlessinger’s
Criteria
C.4.1 Functors and prestacks over artin rings
We work over a fixed field k for simplicity, but the definitions below and Rim–
Schlessinger’s Criteria can be formulated more generally (see Remark C.4.5). Let
Artk denote the category of artinian local k-algebras with residue field k.

Definition C.4.1 (Prorepresentability). We say that a covariant functor F : Artk →

Sets is prorepresentable if there exists a complete noetherian local k-algebra R and
∼
an isomorphism F → hR where hR := Homk−alg (R, −).

If F : Sch/k → Sets is a contravariant functor and x0 ∈ F (k), then we can

consider the induced functor of artin rings

Fx0 : Artk → Sets, A 7→ {x ∈ F (A) | x|k = x0 ∈ F (k)}

where x|k denotes the image of x under F (A) → F (A/mA ). If F is representable by a

scheme X and x ∈ X is the k-point corresponding to x0 , then Fx0 is prorepresentable
by ObX,x . It is possible that Fx be prorepresentable, but F not be representable.
0

Many functors of artin rings are not prorepresentable. For example, if C0 is a

smooth, connected, and projective curve with a non-trivial automorphism group,
then the covariant functor FC0 : Artk → Sets where FC0 (A) consists of isomorphism
classes of smooth proper families of curves C → Spec A such that C ×A A/mA is
isomorphic to C0 , is not prorepresentable. Nevertheless, many moduli functors admit
versal deformations. As it is important to keep track of automorphisms, we will
formulate the definition for prestacks over Artop
k .

Definition C.4.2 (Versality). Let X be a prestack over Artop

k such that the groupoid
X (k) is equivalent to a set {x0 }.
(1) A formal deformation (R, {xn }) of x0 is the data of a complete noetherian
local k-algebra (R, mR ) together with objects xn ∈ X (R/mn+1
R ) and morphisms
xn−1 → xn over Spec R/mnR → Spec R/mn+1 R .
(2) A formal deformation (R, {xn }) is versal if for every surjection A ↠ A0
in Artk with mn+1
A = 0, object η ∈ X (A), and k-algebra homomorphism
∼
ϕ0 : R/mn+1
R → A 0 with an isomorphism α0 : xn |A0 → η|A0 in X (A0 ), there
exists a k-algebra homomorphism ϕ : R/mR → A extending ϕ0 and an
n+1
∼
isomorphism α : xn |A → η in X (A) extending α0 .
(3) A versal formal deformation (R, {xn }) is miniversal (or a prorepresentable hull )
if the induced map hR (k[ϵ]) → X (k[ϵ])/∼ on isomorphism classes is bijective.

In other words, a formal deformation is an element of lim X (R/mn ). When X = F

←−
is a covariant functor Artk → Sets, a formal deformation is a compatible sequence
of elements xn ∈ F (R/mR ). If F is prorepresentable by R and xn ∈ F (R/mn+1
n+1
R )
is the corresponding element, then {xn } is a miniversal formal deformation, in
which case there is a unique lift in (C.4.3). Just as a deformation xn ∈ X (R/mn+1R )
can be viewed via Yoneda’s 2-Lemma as a morphism Spec R/mn+1 R → X , a formal
deformation can be viewed as a morphism {xn } : hR → X of prestacks. From this

556
 perspective, {xn } is versal if there exists a lift for every commutative diagram

Spec A / hR
_ 0 ;
{xn } (C.4.3)
 
Spec A
η
/ X,

where A ↠ A0 is a surjection in Artk . This should be compared with the Infinitesimal

Lifting Criterion for Smoothness (A.3.1 and 3.7.1). Note that since every surjection
A → A0 factors as a composition of surjections with one dimensional kernels,
versality can be checked on surjections with ker(A → A0 ) ∼
= k. A versal deformation
is miniversal if it induces an isomorphism on tangent spaces hR (k[ϵ]) → X (k[ϵ])/∼.
Remark C.4.4 (Global prestacks to local deformation prestacks). If X is a prestack
over Sch/k and x0 ∈ X (k), we define the local deformation prestack Xx0 at x0 as
the prestack, where an object is a morphism α : x0 → x in X to an object x over
a ring A ∈ Artopk ; in other words, an object is a pair (x, α) where x ∈ X (A) and
α : x0 → x|k is an isomorphism. A morphism (α : x0 → x) → (α′ : x0 → x′ ) is a
morphism β : x → x′ such that α′ = α ◦ β. Note that the fiber category Xx0 (k) is
equivalent to the set {id : x0 → x0 }.
If X is an algebraic stack with a smooth presentation U → X from a k-scheme
U and u ∈ U (k) is a lift of x0 , setting xn ∈ X (OU,u /mn+1u ) to be the composi-
tion Spec OU,u /mn+1
u ,→ U → X defines a versal formal deformation {xn }. Rim–
Schlessinger’s Criteria (Theorem C.4.6) provides criteria for the existence of a
versal formal deformation and Artin’s Axioms for Algebraicity (Theorem C.7.1)
provides further criteria for a versal formal deformation to be induced by a smooth
presentation U → X as above.
Remark C.4.5. To work over a more general base, consider a complete noetherian
local ring Λ with residual field k, and define ArtΛ to be the category of artinian
∼
local Λ-algebras (A, m) with an identification k → A/m. In practice, Λ is often
taken to be a ring of Witt vectors, e.g., Λ = Zp . This generality is important for
many applications, e.g., for lifting objects from characteristic p to characteristic 0;
see §C.5.3. Even more generally, one can consider the setup where A → k is a finite
(but not necessarily surjective); see [SP, Tag 06GB].

C.4.2 Rim–Schlessinger’s Criteria

Rim–Schlessinger’s Criteria provides necessary and sufficient conditions for a prestack
X over Artop
k (or a covariant functor F : Artk → Sets as in Schlessinger’s original
formulation) to admit a versal formal deformation.

Theorem C.4.6 (Rim–Schlessinger’s Criteria). Let X be a prestack over Artopk such

that the groupoid X (k) is equivalent to the set {x0 }. For morphisms B0 → A0 and
A → A0 in Artk , consider the natural functor

X (B0 ×A0 A) → X (B0 ) ×X (A0 ) X (A) (C.4.7)

Then X admits a miniversal formal deformation if and only if

(RS1 ) the functor (C.4.7) is essentially surjective whenever A ↠ A0 is surjection
with kernel k;

557
(RS2 ) the map (C.4.7) is essentially surjective when A0 = k and A = k[ϵ], and given
two commutative diagrams

x0 / y0  / Spec B0
Spec k
′
α
α
 over
 γ
/ y β
/ y′  

Spec k[ϵ] 
x ; / Spec(k[ϵ] ×k B0 ),
′
γ

there exists an isomorphism β : y → y ′ in X (k[ϵ] ×k B0 ) such that α′ = β ◦ α

(but we do not require that γ ′ = γ ◦ β).
(RS3 ) dimk TX < ∞ where TX := X (k[ϵ])/∼ (which is a vector space by Lemma C.4.9).

Moreover, X is prorepresentable if and only if X is equivalent to a functor and

(RS4 ) the map (C.4.7) is an equivalence whenever A ↠ A0 is a surjection with kernel
k.

Conditions (RS2 )–(RS3 ) (sometimes referred to as semi-homogeneity) may be

difficult to parse, but in practice, it is in fact often just as easy to verify the stronger
condition (RS4 ) (called homogeneity), and in fact the even stronger condition (RS∗4 )
(called strong homogeneity); see §C.4.3. When A → A0 is surjective, Spec B0 ×A0 A is
the pushout of Spec A0 ,→ Spec A and Spec B0 → Spec B in the category of schemes
(see §B.4). Therefore, homogeneity conditions translate into gluing conditions of
objects over the pushout.
Remark C.4.8 (Schlessinger’s Criteria). When X is a covariant functor
F : Artk → Sets with F (k) = {x0 }, then (RS1 )–(RS4 ) translate into Schlessinger’s
conditions as introduced in [Sch68]:
(H1 ) the map (C.4.7) is surjective whenever A ↠ A0 is a surjection with kernel k;
(H2 ) the map (C.4.7) is bijective when A0 = k and A = k[ϵ];
(H3 ) dimk F (k[ϵ]) < ∞; and
(H4 ) the map (C.4.7) is bijective whenever A ↠ A0 is a surjection with kernel k.
The functor F admits a miniversal formal deformation if (H1 )–(H3 ) hold and is
prorepresentable if (H3 )–(H4 ) hold.
If a prestack X over Artk satisfies (RS1 )–(RS3 ), then the functor FX : Artk →
Sets parameterizing isomorphism classes of objects satisfies (H1 )–(H3 ) but the con-
verse does not always hold. Moreover, the essential surjectivity of X (B0 ×A0 A) →
X (B0 ) ×X (A0 ) X (A) implies the surjectivity of FX (B0 ×A0 A) → FX (B0 ) ×FX (A0 )
FX (A) and the fully faithfulness for X implies the injectivity of FX as long as
AutX (B0 ) (y0 ) → AutX (A0 ) (y0 |A0 ) is surjective for an object y0 ∈ X (B0 ). This latter
condition holds in the case when FX (A0 ) is a set, e.g., when A0 = k. If X is the local
deformation prestack arising from an object x0 ∈ Xe(k) of an algebraic stack Xe over
Sch/k as in Remark C.4.4, then the surjectivity condition on automorphisms trans-
lates by the Infinitesimal Lifting Criterion for Smoothness (3.7.1) to the smoothness
of the inertia stack IX → X at e(x0 ), where e : X → IX is the identity section.
While the existence of a miniversal formal deformation of FX suffices for many
applications, for moduli problems with automorphisms it is more natural to ask for
the existence of a miniversal formal deformation of X and this generality is needed
for some applications, e.g., Artin’s Algebraization (Theorem C.6.8) and Artin’s
Axioms for Algebraicity (Theorem C.7.4).

558
 Before proceeding to the proof, we first indicate some properties of the conditions
(RS1 )–(RS4 ).

Lemma C.4.9. Let X be a prestack over Artop k such that the groupoid X (k) is
equivalent to the set {x0 }, and let FX : Artk → Sets be the covariant functor assigning
A ∈ Artk to the set of isomorphism classes X (A)/∼. Assume that Condition (RS2 )
holds for X .
(1) The tangent space TX = FX (k[ϵ]) has a natural structure of a k-vector space.
More generally, for every finite dimensional k-vector space V , denoting k[V ] as
the k-algebra k ⊕ V defined by V 2 = 0, the set FX (k[V ]) has a natural structure
of a k-vector space and there is a functorial bijection FX (k[V ]) = TX ⊗k V .
(2) Consider a surjection A ↠ A0 in Artk with square-zero kernel I and an
element x0 ∈ X (A0 ), and let Liftx (A) be the set of morphisms α : x0 → y
α′
over Spec A0 → Spec A where (α : x0 → x) ∼ (α′ : x0 −→ x′ ) if there is an
isomorphism β : x → x′ such that α′ = β ◦ α. There is an action of TX ⊗ I
on Liftx0 (A) which is functorial in X . Assume that Liftx0 (A) is non-empty.
Then the action is transitive if Condition (RS1 ) holds for X , and is free and
transitive (i.e., Liftx0 (A) is a torsor under TX ⊗ I) if Condition (RS4 ) holds
for X .

Proof. We first note if V is a finite dimensional vector space, then k[V ] = k[ϵ] ×k
· · · ×k k[ϵ] and by applying (RS2 ) inductively, we see that the conclusion of (RS2 )
also holds for A0 = k and A = k[V ]. For B0 ∈ Artk , the first part of (RS2 )
∼
implies that FX (B0 ×k k[V ]) → FX (B0 ) × FX (k[V ]) is a bijection. In particular,
∼
FX (k[V ] ×k k[W ]) → FX (k[V ]) × FX (k[W ]) is bijective for every pair of finite
dimensional vector spaces, or in other words the functor V 7→ FX (k[V ]) commutes
with finite products. The vector space structure of TX = FX (k[ϵ]) follows from the
bijectivity of
∼
FX (k[ϵ] ×k k[ϵ′ ]) → FX (k[ϵ]) × FX (k[ϵ′ ]). (C.4.10)
Indeed, if τ1 , τ2 ∈ FX (k[ϵ]), then we may use (C.4.10) to view (τ1 , τ2 ) ∈ FX (k[ϵ] ×k
k[ϵ′ ]), and we define τ1 + τ2 as the image of (τ1 , τ2 ) under F (k[ϵ] ×k k[ϵ′ ]) → F (k[ϵ])
induced by the ring map k[ϵ] ×k k[ϵ′ ] → k[ϵ] taking (ϵ, 0) and (0, ϵ′ ) to ϵ. Scalar
multiplication of c ∈ k on τ ∈ FX (k[ϵ]) is defined by taking the image of τ under
FX (k[ϵ]) → FX (k[ϵ]) induced by the map k[ϵ] → k[ϵ] taking ϵ to cϵ.
The same argument shows that V 7→ FX (k[V ]) defines a k-linear functor Vectfd k →
Vectk . The natural map

FX (k[ϵ]) × Homk (k[ϵ], k[V ]) → FX (k[V ]), (τ, ϕ) 7→ ϕ∗ τ := FX (ϕ)(τ )

For (2), observe that the natural map

A ×A0 A → k[I] ×k A, (a1 , a2 ) 7→ (a1 + a2 − a1 , a1 )

is an isomorphism. We therefore have a diagram

∗
p1
X (k[I]) × X (A) ↞ X (k[I] ×k A) ∼
= X (A ×A0 A) −→ X (A)

where the left functor is essentially surjective by the first part of (RS2 ). To define
the action, let τ ∈ TX ⊗ I = FX (k[I]) and (α : x0 → x) ∈ Liftx0 (A). Choose a
representative τe ∈ X (k[I]) of τ . We define τ · (α : x0 → x) ∈ Liftx0 (A) as p∗1 y, where

559
y ∈ X (k[I] ×k A) is a preimage of (e τ , x). To see that this is well-defined, suppose
that y ′ ∈ X (k[I] ×k A) is another preimage. This yields a diagram

x0 /x Spec k  / Spec A

over
   
τ / y β
7/ y
′
Spec k[I]
 / Spec(k[I] ×k A).

By the second part of (RS2 ), there exists β : y → y ′ filling in the diagram, and
thus p∗1 y and p∗1 y ′ in X (A) define the same element in Liftx0 (A). Finally, if (RS1 )
holds (resp., (RS4 ) holds), then X (A ×A0 A) → X (A) ×X (A0 ) X (A) is essentially
surjective (resp., an equivalence), and we see that the action is transitive (resp., free
and transitive).

Proof Theorem C.4.6. We establish the sufficiency of the criteria, leaving the neces-
sity to the reader. The tangent space TX := X (k[ϵ])/∼ has the structure of a vector
space by Lemma C.4.9(1) and is finite dimensional by (RS2 ). Let N = dimk TX and
x1 , . . . , xN ∈ TX be a basis. Define S = k[[x1 , . . . , xN ]] with mS = (x1 , . . . , xn ). We
will construct inductively a decreasing sequence of ideals J0 ⊇ J1 ⊇ · · · and objects
ηn ∈ X (S/Jn ) together with maps ηn → ηn+1 over Spec S/Jn ,→ Spec S/Jn+1 . We
set J0 = mS and η0 = x0 ∈ X (k). We also set J1 = m2S so that S/J1 ∼ = k[T X ]. Using
the bijection FX (k[TX ]) ∼ = TX ⊗k TX of Lemma C.4.9(1), the element i xi ⊗ xi
P
defines an isomorphism class of an object η1 ∈ X (S/J1 ) such that the induced map
Spec S/J1 → X induces a bijection on tangent spaces. By construction,there is a
map η0 → η1 over Spec k ,→ Spec S/J1 .
Suppose we have constructed Jn and ηn−1 → ηn . We claim that the set of ideals

mS Jn ⊆ J ⊆ Jn and there exists ηn → η

 
Σ= J ⊆S (C.4.11)
over Spec S/Jn ,→ Spec S/J

has a minimal element. Indeed, it is non-empty since Jn ∈ Σ so it suffices to check

that J ∩ K ∈ Σ for J, K ∈ Σ. To achieve this, note that Jn /mS Jn is a k-vector
space with subspaces J/mS Jn and K/mS Jn . We may therefore choose an ideal
J ⊆ J ′ ⊆ Jn with J ∩ K = J ′ ∩ K and J ′ + K = Jn . We have a diagram

ηJ∩K o ηJ |S/J ′ S/(J ∩ K) / S/J ′

O O
over □
 
ηK o ηn S/K / S/Jn ,

where ηJ ∈ X (S/J) and ηK ∈ X (S/K) are the objects corresponding to J and

K. Condition (RS1 ) implies that the diagram can be filled in, which shows that
J ∩ K ∈ Σ.
Define J = n Jn , R = S/J, and In = Jn /J. We claim that for every n, there
T
exists Nn with JNn ⊆ mn+1S + J, or in other words the topology on R defined by (In )
is the mR -adic topology on R. For each n, since S/mnS is artinian, there exists Nn
with JNn +mnS = JNn+1 +mnS = · · · . We set E = lim(JNn +mnS )/mnS ⊆ lim S/mnS = S.
←− ←−
We claim that E ⊆ J: for f ∈ E and any m, f ∈ Jm + mnS for n ≫ 0, and the claim
follows from Krull’s intersection theorem. Since E surjects onto (JNn + mn+1
S )/mn+1
S ,
the natural map J → (JNn + mS )/mS is also surjective, and the claim follows.
n+1 n+1

Since INn ⊆ mn+1

R , we can define ξn := ηNn |R/mn+1 . R

560
 It remains to show that the formal deformation ξ := {ξn } over R is versal.
Suppose B ↠ A is a surjection in Artk with kernel k and that we have a diagram

x /ξ g
/
@ Spec _ A ; hR
over
  g
e
y Spec B.
We must construct a morphism y → ξ extending x → ξ. We claim that it suffices to
construct a morphism ge : Spec B → hR (i.e., a ring map R → B) extending g. Since
hR (k[ϵ]) → TX is bijective, Lemma C.4.9(2) implies that there are actions of TX on
the sets Liftx (B) and Liftg (B) of isomorphism classes of lifts of x and g to objects
in X (B) and hR (B). Moreover, since (RS1 ) holds for X , the action on Liftx (B) is
transitive. Thus, we can find τ ∈ TX such that y = τ · (e g ∗ ξ) = (τ · ge)∗ ξ. This gives
an arrow y → ξ over τ · ge : Spec B → hR .
To construct ge, choose n such that R → A factors as R → R/In = S/Jn → A. It
suffices to show that Spec A → Spec S/Jn extends to a map Spec B → Spec S/Jn+1
and for this, it suffices to show the existence of a dotted arrow making the diagram
Spec _ A / Spec S/Jn
_

  (
Spec B / Spec B ×A (S/Jn ) / Spec S/Jn+1

commutative. Note that since ker(B → A) = k, ker(B ×A (S/Jn ) → S/Jn ) = k. As

S is a power series ring, we may choose an extension S → B of S → S/Jn → A. This
induces a map S → B ×A (S/Jn ). If this map is not surjective, then every element
of Jn must map to 0 in B ×A (S/Jn ), which implies that B ×A (S/Jn ) → S/Jn has a
section giving the desired lift. Otherwise, B ×A (S/Jn ) = S/K where K = ker(S →
B ×A (S/Jn )). The ideal K lies in the set of ideals defined in (C.4.11): the inclusion
K ⊂ Jn is clear, the inclusion mS Jn ⊆ K is implied by the equality ker(B → A) = k,
and the existence of ηn → η over Spec S/Jn ,→ Spec S/K follows from applying
(RS1 ) to the above square. By minimality of Jn+1 , we have a containment Jn+1 ⊆ K
and thus a ring map S/Jn+1 → S/K = B ×A (S/Jn ) inducing the desired dotted
arrow.
Finally, suppose that X is equivalent to a functor F and (RS4 ) holds. Given a
surjection B → A with kernel k and x ∈ F (A), we need to show the existence of a
unique lift in every diagram

Spec _ A
g
/ hR
;

 
Spec B / F.

By Lemma C.4.9(2), the map Liftg (B) → Liftx (B) is bijective as both are torsors
under TX . This implies the existence of a unique lift. See also [Sch68, Thm. 2.11],
[SGA7-I, Thm. VI.1.11], [Har10, Thm. 16.2], [Ser06, Thm. 2.3.2], and [SP, Tag 06IX].

C.4.3 Verifying Rim–Schlessinger’s Criteria

We apply Rim–Schlessinger’s Criteria (C.4.6) to construct miniversal formal defor-
mations for our three main moduli problems by verifying (RS1 )-(RS3 ). In fact, we

561
 will verify the following strong homogeneity condition:
(RS∗4 ) X (B0 ×A0 A) → X (B0 ) ×X (A0 ) X (A) is an equivalence for every map B0 → A0
and surjection A ↠ A0 of (not necessarily artinian) rings with square-zero
kernel.
It is often just as easier to verify (RS∗4 ) as the weaker conditions (RS1 )-(RS2 ). Strong
homogeneity will also appear as one of the axioms in our second version of Artin’s
Axioms for Algebraicity (C.7.4), as it will be useful to verify openness of versality.
It turns out that every algebraic stack satisfies (RS∗4 ) (see [SP, Tag 07WN]), or, in
other words, the Ferrand pushout Spec(B0 ×A0 A) is a pushout in the category of
algebraic stacks.
Verifying (RS∗4 ) relies on properties of modules over fiber products of rings.
Lemma C.4.12. Let A → A0 be a surjection of rings with square-zero kernel,
and B0 → A0 be a maps of rings. Let M, M0 , N0 be flat modules over A, A0 , B0 ,
M → M0 be an A-module map, and N0 → M0 be a B0 -module map. Assume that
M ⊗A A0 → M0 and N0 ⊗B0 A0 → M0 are isomorphisms. Set B := B0 ×A0 A and
N = N0 ×M0 M . Then
(1) the maps N ⊗B B0 → N0 and N ⊗B A → M are isomorphisms,
(2) N is flat over B, and
(3) The modules N0 and M are finitely presented if and only if N is.
(4) Let B0′ be a B0 -algebra, A′ be an A-algebra, and B0′ ⊗B0 A0 → A′ ⊗A A0 be
an isomorphism. Set B ′ = B0′ ×A′0 A. Then B → B ′ is flat and of finite
presentation if and only if B0 → B0′ and A → A′ are.
Proof. We verify (2), leaving the remaining claims to the reader. Let J = ker(A →
A0 ). Since M is flat over A, the Local Criterion of Flatness (A.2.6) implies that
0 → J ⊗A0 M0 → M → M0 → 0 is exact, and thus
0 → J ⊗A0 M0 → N → N0 → 0 (C.4.13)
is also exact. To show the flatness of N , by the Local Criterion of Flatness (A.2.6), it
suffices to verify that (a) N ⊗B B0 is flat over B0 and (b) J ⊗B0 N0 → N is injective.
The module in (a) is identified with N0 by (2), which is flat over B0 by hypothesis.
Since J ⊗A0 M0 = J ⊗A0 (N0 ⊗B0 A0 ) = J ⊗B0 N0 , (b) follows from (C.4.13). See
[Sch68, Lem. 3.4], [Har10, Prop. 16.4], and [SP, Tags 08KG, 0D2G, and 0D2K].
We say that a prestack X over Sch/k admits miniversal formal deformations
(resp., is locally prorepresentable, satisfies (RS3 )) if for every x0 ∈ X (k), the local
deformation prestack Xx0 (as defined in Remark C.4.4) admits a miniversal formal
deformation (resp., is prorepresentable, satisfies (RS3 )).
Proposition C.4.14. Let X be a proper scheme over a field k.
(1) The Hilbert functor Hilb(X) : Sch/k → Sets, whose objects over S are closed
subschemes Z ⊆ XS flat and finitely presented over S, satisfies (RS3 ) and
(RS∗4 ), and is therefore locally prorepresentable.
(2) The stack Fam over Sch/k, whose objects over S are proper, flat, and finitely
presented morphisms Y → S of algebraic spaces,1 admits miniversal formal
deformations. In particular, the stack Mall
g of all curves and the stack Mg of
smooth curves admit miniversal formal deformations.
1We need to allow Y to be an algebraic space if we want F am to be a stack; see Example 2.5.12

and Exercise 4.5.15. On the other hand, Rim–Schlessinger’s Criteria (C.4) equally applies to the
prestack parameterizing proper, flat, and finitely presented morphisms Y → S of schemes.

562
 (3) The stack Coh(X) over Sch/k, whose objects over S are finitely presented
quasi-coherent OXS -modules flat over S, satisfies (RS3 ) and (RS∗4 ), and there-
fore admits miniversal formal deformations. In particular, if C is a smooth,
connected, and projective curve, Bun(C) admits miniversal formal deforma-
tions.

Proof. For (1), Proposition C.1.3 identifies the tangent space of Hilb(X) at Z0 ⊆ X
with H1 (Z, NZ0 /X ). Since X is proper, H1 (Z, NZ0 /X ) is finite-dimensional and (RS3 )
holds. To check (RS∗4 ), A → A0 be a surjection of rings with square-zero kernel,
and B0 → A0 be a maps of rings, and suppose that W0 ⊆ XB0 and Z ⊆ XA are
closed subschemes flat over the base such that Z0 := W0 ×B0 A0 = Z ×A A0 ⊆ XA0 .
Then OW0 ×OZ0 OZ is a quotient of OXB , and defines a closed subscheme W ⊆ XB .
For each affine open Spec B ′ ⊆ XB with restrictions Spec B0′ ⊆ XB0 , Spec A′ ⊆ XA ,
and Spec A′0 ⊆ XA0 , we have that B ′ = B0′ ×A′0 A′ . By Lemma C.4.12(4), B → B ′
is flat and finitely presented, and thus W is flat and finitely presented over B.
For (2), the tangent space of Fam at Y0 is identified with Ext1OY0 (Y0 , OY0 ) by
Theorem C.3.6, and is therefore finite dimensional. (When Y0 is smooth, the tangent
space is H1 (Y0 , TY0 ).) If [Z0 → Spec B0 ] ∈ Fam(B0 ) and [Y → Spec A] ∈ Fam(A)
∼
with an isomorphism (Z0 )A0 → YA0 , then the Ferrand pushout Z := Z0 ⨿YA0 Y
exists by Theorem B.4.1. Applying Lemma C.4.12(4) to an affine cover of Z shows
that Z → Spec B is flat and finitely presented. Moreover, since Z is a pushout,
compatible isomorphisms of Z0 and Y extend uniquely to an isomoprhism of Z.
For (3), the tangent space of Coh(X) at a coherent sheaf is identified with
Ext1OX (E, E) by Proposition C.1.18, which is finite dimensional. The base change
XB is the pushout of XA0 → XB0 and XA0 ,→ XA , and for each affine open
Spec B ′ ⊆ XB with restrictions Spec B0′ ⊆ XB0 , Spec A′ ⊆ XA , and Spec A′0 ⊆
XA0 , B ′ = B0′ ×A′0 A′ . For G0 ∈ Coh(X)(B0 ) and F ∈ Coh(X)(A) restricting to
F0 ∈ Coh(X)(A), we define G := G0 ×F0 F on XB . This is finitely presented by
Lemma C.4.12(3) (applied to G|Spec B ′ over B ′ = B0′ ×A′0 A′ ) and flat over B by
Lemma C.4.12(2) (applied to G|Spec B ′ over B = B0 ×A0 A). This gives essential
surjectivity for (RS∗4 ), and the fully faithfullness is clear.

Exercise C.4.15. If X0 is a smooth proper scheme over k with no infinitesimal

automorphisms, i.e., H0 (X0 , TX0 ) = 0, show that the functor of deformations of X0
is prorepresentable.

Exercise C.4.16. Let X = Spec A be an affine scheme with isolated singularities

over a field k, and let FX : Artk → Sets be the functor, where FX (R) is the set of
isomorphism classes of pairs (B, β) where B is a flat R-algebra and β : A → B ⊗R
R/mR is an isomorphism. Show that FX admits a miniversal formal deformations.

C.5 Effective formal deformations and Grothendieck’s

Existence Theorem
Grothendieck’s Existence Theorem is a powerful technique for showing that formal
deformations are effective. It is sometimes referred to as Formal GAGA, as it is
analogous to Serre’s GAGA Theorem [Ser56] that for a proper scheme X over C with
analytification X an , the natural functor Coh(X) → Coh(X an ), taking a coherent
sheaf F to its analytification F an , is an equivalence. The proofs follow very similar
strategies.

563
C.5.1 Effective formal deformations
Definition C.5.1. Let X be a prestack (or functor) over Sch/k. Let x0 ∈ X (k)
and consider a formal deformation {xn } over a complete noetherian local k-algebra
with residue field k of x0 (or more precisely a formal deformation of the deformation
stack Xx0 at x0 as defined in Remark C.4.4). We say that {xn } is effective if there
∼
exists an object xb ∈ X (R) and compatible isomorphisms xn → x b|Spec R/mn+1 .
A formal deformation (R, {xn }) is effective if it is in the essential image of the
natural functor X (R) → lim X (R/mn ), or in other words if there exists a dotted
←−
arrow making the diagram
 / Spec R/m2   / Spec R/m3   / · · ·  / Spec R
Spec R/m
x2
x1 x
b
x0 
.) X

commutative.
Example C.5.2. If F : Sch/k → Sets is a contravariant functor representable by a
scheme X over k, then every formal deformation (R, {xn }) is effective. Indeed, xn
corresponds to a morphism Spec R/mn+1 → X with image x ∈ X(k) and thus to
a k-algebra homomorphism ϕn : O bX,x → R/mn+1 . By taking the inverse image of
ϕn , we have a local homomorphism O bX,x → R which in turn defines a morphism
b : Spec R → X extending {xn }.
x
More generally, if X is an algebraic stack over k, every formal deformation is
effective. Indeed, there exists a smooth presentation U → X and a lift u ∈ U (k) of
x0 ∈ X (k). By applying the Infinitesimal Lifting Criterion for Smoothness (3.7.1),
we may inductively construct lifts

Spec R/m n /9 U
_

 
Spec R/mn+1
xn
/ X.

Since U is a scheme, the maps Spec R/mn → U extend to a map Spec R → U , and
the composition Spec R → U → X effectivizes the formal deformation x b = {xn }.
Since the diagonal of X is representable, it follows that a compatible automorphisms
αn of xn extend to a unique automorphism of X, b i.e., the functor

X (R) → lim X (R/mn+1 )

←−
is an equivalence of categories.

C.5.2 Grothendieck’s Existence Theorem

The following is frequently applied when (R, m) is a complete local ring.
Theorem C.5.3 (Grothendieck’s Existence Theorem). Let X be a scheme proper
over a noetherian ring A which is complete with respect to an ideal m ⊆ A. Let
Xn := X ×R R/mn+1 . The functor
Coh(X) → lim Coh(Xn ), E 7→ {E/mn+1 E} (C.5.4)
←−
is an equivalence of categories.

564
 The essential surjectivity of (C.5.4) translates to an extension of the diagram

E0 E1 E2 E
 / X1   / X2   / · · · 
X0  /X

       
Spec R/m  / Spec R/m2  / Spec R/m3  / · · · / Spec R.

Using the language of formal schemes and setting X

b = X ×Spec R Spf R to be the
m-adic completion of X, then Grothendieck’s Existence Theorem asserts that the
functor Coh(X) → Coh(X),b defined by E 7→ E,
b is an equivalence.
Remark C.5.5 (Limits of categories). An object of lim Coh(Xn ) is a sequence {En }
←− ∼
of coherent OXn -modules En together with isomorphisms En+1 /mn+1 En+1 → En .
A morphism ϕ = {ϕn } : {En } → {Fn } is the data of compatible OXn -module
homomorphisms ϕn : En → Fn . If X ∼ = Spec R, then lim Coh(Xn ) is equivalent to
←−
the category of finite A-modules.
The category lim Coh(Xn ) is abelian and (C.5.4) is an exact functor. While the
←−
cokernel of a map {ϕn } : {En } → {Fn } in lim Coh(Xn ) is simply {coker ϕn }, kernels
←−
are more subtle. The Artin–Rees Lemma (B.5.4) implies that for each m, the image
of ker(αl ) → Em stabilizes for l ≫ m to a subsheaf Em ′
⊆ Em , and that for each
n, the quotients Em /m
′
Em stabilizes for m ≫ n to a coherent sheaf Kn on Xn .
n+1 ′

The kernel of {ϕn } is {Kn } (which is usually different from {ker ϕn }). See [SP,
Tag 0EHN].
∼
Proof. Fully faithfulness of (C.5.4) translates into the bijection Ext0OX (E, F ) →
limn Ext0OXn (E|Xn , F |Xn ), which is a consequence of Formal Functions (see Exer-
←−
cise A.5.5). It remains to show essential surjectivity.
Projective case: The key claim is that for every {En } ∈ Coh(Xn ), there exists
integers m and r together with compatible surjections OXn (−m)⊕r ↠ En . Consider
the finite type OX0 -algebra A := i≥0 m OX0 /mi+1 OX0 and the finitely generated
i
L

OA -module G := i≥0 mi Ei (noting that when {En } = {E/mn+1 } for a coherent

L

sheaf E on X, then G is the associated graded i i+1

E). Viewing G
L
i≥0 m E/m
as a coherent sheaf on SpecX0 A and applying Serre’s Vanishing Theorem [Har77,
Thm. II.5.2] to the projective morphism SpecX0 A → Spec i≥0 mi /mi+1 gives
L

an integer m0 such that for all m ≥ m0 H1 (X0 , G(m)) = 0. This yields that
H1 (X0 , mn+1 En (m)) = 0 for all n, which in turn implies that

H0 (X0 , En+1 (m)) ↠ H0 (X0 , En (m)) (C.5.6)

is surjective. After possibly increasing m0 , we can assure that E0 (m) is globally

generated by sections s0,1 , . . . , s0,r . By the surjectivity of (C.5.6), we can find
compatible lifts sn,i ∈ H0 (X0 , En (m)) of the sections s0,i . This gives compatible
⊕r
maps OX n
→ En (m), each which is surjective by Nakayama’s Lemma.
With the claim established, let {Kn } ∈ lim Coh(Xn ) be the kernel {ϕn : OXn (−m)⊕r ↠
←−
En } as in Remark C.5.5. Applying the claim again to {Kn } induces compatible
right exact sequences
′ α
OXn (−m′ )⊕r −−→
n
OXn (−m)⊕r → En → 0.

565
 ′
By fully faithfulness, the morphism {αn } is induced by a morphism αe : OX (−m′ )⊕r →
OX (−m) , and it follows that E := coker(e
⊕r
α) is a coherent sheaf on X such that
{E/mn+1 E} ∼ = {En } in lim Coh(Xn ).
←−
Proper case: By Chow’s Lemma [Har77, Exc. II.4.10], there exists a projective
morphism g : X ′ → X which is an isomorphism over a dense open subset U ⊆ X
such that X ′ is projective over the Spec R. Given {En } ∈ Coh(Xn ), consider the
pullback {g ∗ En } ∈ Coh(Xn′ ), where Xn′ := X ′ ×R R/mn+1 . Since X ′ is projective
over R, there exists a coherent sheaf E ′ on X ′ an an isomorphism
∼
{βn } : {g ∗ En } → {E ′ /mn+1 E ′ }

in lim Coh(Xn′ ). By Finiteness of Cohomology (A.5.3), g∗ E ′ is coherent. Conceptu-

←−
ally, the argument is now very straightforward: adjunction En → g∗ g ∗ En induces
an exact sequence
{αn }  {qn }
0 → {Kn } → {En } −−−→ (g∗ E ′ )/mn+1 (g∗ E ′ ) −−−→ {Qn } → 0 (C.5.7)

in lim Coh(Xn ) such that the ker(α) = {Kn } and coker(α) = {Qn } are supported
←−
on X ∖ U , i.e., each Kn and Qn are supported on X ∖ U . By noetherian induction,
we can assume that there are coherent sheaves K and Q on X and isomorphisms
{Kn } ∼ = {K/mn+1 K} and {Qn } ∼ = {Q/mn+1 Q}. By full faithfulness and exactness
of (C.5.4), the map {qn } is induced by a surjection g∗ E ′ → Q, and we define
∼
F := ker(g∗ E ′ → Q). Since ExtiOX (K, F ) → limn ExtiOXn (Kn , F/mn+1 F ), there is
←−
an extension 0 → K → E → F → 0 giving a coherent sheaf E on X such that
{En } ∼ = (E/mn+1 E).
The existence of {αn } in (C.5.7), however, takes some work. The formalism of
formal schemes can be useful here as one can consider the map of formal schemes
gb : Xb′ → X b over Spf R, the coherent sheaf E b = lim En on X,
b and the adjunc-
←−
tion morphism E b → gb∗ gb∗ Eb in Coh(X),b and apply a version of formal functions
[EGA, III1 .4.1.5]—sometimes called the ‘comparison theorem’—giving identifications
b ∼
g ∗ E)
Ri gb∗ (b = lim Rn g∗ (g ∗ En ) ∼
= (Rn g∗ E ′ )b.
←−
We argue more directly. We first show that there are unique maps αn filling in
the diagram
En
cn
/ g∗ g ∗ En
∼

αn g∗ βn
 
(g∗ E ′ )/mn+1 (g∗ E ′ )
dn
/ g∗ (E ′ /mn+1 E ′ ),

where cn and dn are the natural maps. By the uniqueness, the existence of {αn } is
local on X, so we may assume that X = Spec B. Since B → B b := lim B/mn+1 B is
←−
flat and all the coherent sheaves in the diagram are annihilated by a power of m, Flat
Base Change (A.2.12) further reduces us to the case that B is complete with respect
to mB. In this case E := lim En corresponds to a coherent sheaf on X mapping to
←−
{En } in lim Coh(Xn ). Applying Formal Functions (A.5.4) to X ′ → Spec B and E ′
←−
yields that g∗ E ′ ∼
= lim g (E ′ /mn+1 E ′ ), which shows that lim dn is an isomorphism.
←− ∗ ←−
Therefore, the composition α := (lim dn )−1 ◦ lim(g∗ βn ◦ cn ) defines a map E → g∗ E ′ ,
←− ←−
which induces the desired maps αn . We now show that the kernel and cokernel
of (C.5.7) are supported on X ∖ U . Since g ∗ E ∼ = lim g ∗ En , the isomorphism
←− ∼
∗
{βn } : (g En ) → (E /m′
E ) comes
n+1 ′

from an isomorphism∗ g ∼
∗
E → E ′ such that
{αn } : {En } → (g∗ E )/m′ n+1
(g∗ E ) comes from E → g∗ g E → g∗ E ′ . Since g is
′

566
an isomorphism over U , the adjunction map E → g∗ g ∗ E is an isomorphism over U ,
and it follows that ker{αn } and coker{αn } are supported on X ∖ U . See also [EGA,
III1 .5.1.4], [Ill05, Thm. 8.4.2], and [SP, Tag 088E].
Corollary C.5.8. Let (R, m) be a complete noetherian local ring and Xn →
Spec R/mn+1 be a sequence of proper morphisms such that Xn ×R/mn+1 R/mn ∼ =
Xn−1 . If Ln is a compatible sequence of line bundles on Xn such that L0 is ample,
then there exists a projective morphism X → Spec R and an ample line bundle L on
∼
X and compatible isomorphisms Xn ∼ = X ×R R/mn+1 and Ln → L|Xn .
In other words, there is an extension in the cartesian diagram
 / X1   / X2   / · · · 
X0  /X

    
Spec R/m  / Spec R/m2   / Spec R/m3   / · · ·  / Spec R

with X projective over R. We say that the formal deformation {Xn → Spec R/mn+1 }
of X0 is effective.2
Proof. Let k = R/m. Consider the finitely generated graded k-algebra B =
mn /mn+1 and the quasi-coherent graded OX0 -algebra A = B ⊗k OX0 . By apply-
L
ing Serre’s Vanishing Theorem to SpecX0 A and the ample line bundle L0 ⊗OX0 OX0′ ,
there exists d0 such that H1 (X0 , A ⊗ L⊗d
0 ) = 0 for d ≥ d0 . By possibly enlarging
d0 , we can assume that Ld00 is very ample. Let s0,0 , . . . , s0,N of H0 (X0 , L⊗d
0
0
) be
sections defining a closed immersion X0 ,→ P . There is an exact sequence
N

0 → mn OXn+1 /mn+1 OXn+1 → OXn+1 → OXn → 0.

Tensoring by L⊗d
n+1 yields a short exact sequence

0 → (mn OXn+1 /mn+1 OXn+1 ) ⊗ L⊗d ⊗d ⊗d

0 → Ln+1 → Ln → 0,

where we have used that (mn OXn+1 /mn+1 OXn+1 ) is supported on X0 along with
the identifications Ln+1 ⊗ OXm ∼ = Lm for m ≤ n. The vanishing of H1 (X0 , A ⊗ L⊗d 0 )
implies that we may lift the sections s0,0 , . . . , s0,N inductively to compatible sections
sn,0 , . . . , sn,N of H0 (Xn , L⊗d
n ). By Nakayama’s Lemma, the induced morphisms
Xn ,→ PN R/m n+1 are closed immersions giving a commutative diagram

N
 / Pn 2   / · · ·  / PN
;P R/m ; R
cl cl 9 R
cl
- +  -
X0  / X1  / · · · /X

$   &    # 
Spec k  / Spec R/m2  / · · · / Spec R

Grothendieck’s Existence Theorem (C.5.3) gives an equivalence Coh(PN

R ) → lim Coh(PN
R/mn+1 ).
←−
Essential surjectivity gives a coherent sheaf E on PR extending {OXn } and full
N

faithfulness gives a surjection OPNR

→ E extending OPN n+1 → OXn . We take
R/m

X ⊆ PN R to be the closed subscheme defined by ker(OPN

R
→ E). See also [EGA,
III.5.4.5], [Ill05, Thm. 8.4.10], and [SP, Tag 089A].
2 This is also sometimes referred to as algebraizable, but we reserve this term a deformation over

a finite type k-scheme, e.g., the output of Artin’s Algebraization (Theorem C.6.8).

567
Remark C.5.9. Assume in addition that each Xn is flat over R/mn+1 . If we are
only given an ample line bundle L0 on X0 (but not the line bundles Ln ), then the
obstruction to deforming Ln−1 to Ln is an element obLn−1 ∈ H2 (X, OX ⊗k mn )
(Proposition C.2.11). If these cohomology groups vanish (e.g., if X is of dimension
1), then there exists compatible extensions Ln , and thus the formal deformation
{Xn → Spec R/mn+1 } are effective.
Without the existence of deformations Ln of L0 , it is not necessarily true that
formal deformations are effective. For instance, there is a projective K3 surface
(X0 , L0 ) and a first-order deformation X1 → Spec k[ϵ] which is not projective (so L0
does not deform to X1 ), and a formal deformation which is not effective; see [Har10,
Ex. 21.2.1], [Sta06, Claim 3.5], and [SP, Tag 0D1Q]. Similarly, formal deformations
of abelian varieties may not be effective. Note that for the moduli of abstract K3
surfaces or abelian varieties, Rim–Schlessinger’s Criteria (C.4.6) applies to construct
versal formal deformations, but the lack of effectivity implies that the corresponding
stacks are not algebraic (Example C.5.2).
We apply Grothendieck’s Existence Theorem to the Hilbert functor Hilb(X), the
stack Mall
g of all curves, and the stack Coh(X) of coherent sheaves, each defined
over Sch/k as Proposition C.4.14.
Proposition C.5.10. Every formal deformation is effective for the Hilb(X), Mall g
and Coh(X). In particular, there exist effective miniversal formal deformations.
Proof. For Hilb(X), let {Zn ⊆ XR/mn+1 } be a formal deformation. Grothendieck’s
Existence Theorem (C.5.3) implies the existence of a coherent sheaf E on XR extend-
ing the structure sheaves {OZn }, and moreover that there is a surjection OXR → E
extending {OXn → OZn }. The subscheme Z ⊆ XR defined by ker(OXR → E)
effectivizes the formal deformation.
For Mallg , let {Cn → Spec R/m
n+1
} be a formal deformation. As C0 is a proper
curve over a field, it is projective. Let L0 be an ample line bundle on C0 . The
obstruction to deforming a line bundle Ln on Cn to a line bundle Ln+1 on Cn+1
is an element obLn−1 ∈ H2 (X, OX ⊗k mn ) (Proposition C.2.11). Since dim C = 1,
this cohomology group is zero and thus there are a compatible family of line bundle
{Ln }. We may therefore apply Corollary C.5.8.
For Coh(X), the effectivity of a formal deformation follows directly from Grothendieck’s
Existence Theorem (C.5.3). The last statement follows from the existence of miniver-
sal formal deformations (Proposition C.4.14).
Exercise C.5.11. Let X and Y be proper schemes over the spectrum S = Spec R
of a complete noetherian local ring R. Denote by Xn and Yn the restrictions of
X and Y to Sn = Spec R/mn+1R . Show that a compatible sequence of morphisms
fn : Xn → Yn over Sn extends to a unique morphism f : X → Y .

C.5.3 Lifting to characteristic 0

One striking application of deformation theory is to “lift” schemes X0 over a field
k of char(k) = p to characteristic 0. We say that X0 is liftable to characteristic 0
if there exists a complete noetherian local ring (R, m) of characteristic 0 such that
R/m = k and a smooth scheme X → Spec R such that X0 ∼ = X ×R k.3 One can
hope to then use characteristic 0 techniques (e.g., Hodge theory) on X and deduce
properties of X0 . The strategy to lift X0 is to inductively deform X0 to schemes
3 There are some variants to this definition, e.g., when R is already given as a complete DVR

with residue field k.

568
Xn over R/mn+1 and then apply Grothendieck’s Existence Theorem to effective the
formal deformation.
Smooth curves are liftable as obstructions to deforming both the curve and
the ample line bundle both vanish. Serre produced an example of a non-liftable
projective threefold (see [Har10, Thm. 22.4]), which Mumford extended to a non-
liftable projective surface (see [Ill05, Cor. 8.6.7]). On the other hand, Mumford
showed that principally polarized abelian varieties are liftable [Mum69] while Deligne
showed that K3 surfaces are liftable [Del81]. These examples are quite interesting as,
in both cases, formal deformations are not necessarily effective (see Remark C.5.9).

C.6 Artin Algebraization

Artin Algebraization states that every effective versal formal deformation “alge-
braizes”, i.e., extends to an an object over a finite type k-scheme. In this section,
we show how Artin Algebraization follows from Artin Approximation following the
ideas of Conrad and de Jong [CdJ02].

C.6.1 Limit preserving prestacks

Extending the definition of a limit preserving functor §B.5.5, we say that a prestack
X over Sch/k is limit preserving (or locally of finite presentation) if for every system
Bλ of k-algebras, the natural functor

colim X (Bλ ) → X (colim Bλ )

is an equivalence of categories. When X is an algebraic stack over k, then this

equivalent to the morphism X → Spec k being locally of finite presentation; see
Exercise 3.3.33.
Exercise C.6.1. Use the Limit Methods of §B.3 to show that Hilb(X), Mall
g and
Coh(X) are each limit preserving over Sch/k.

C.6.2 Conrad–de Jong Approximation

In Artin Approximation (B.5.18), the initial data is an object over a complete
noetherian local k-algebra which is assumed to be the completion of a finitely
generated k-algebra at a maximal ideal. We will now see that a similar approximation
result still holds if this latter hypothesis is dropped. The idea is to approximate both
the complete local ring and the object.
Recall also that if (A, m) is a local ring and MLis an A-module, then the associated
graded module of M is defined as Grm (M ) = n≥0 mn M/mn+1 M ; it is a graded
module over the graded ring Grm (A).
Theorem C.6.2 (Conrad–de Jong Approximation). Let X be a limit preserving
prestack over Sch/k, (R, mR ) be a complete noetherian local k-algebra with residue
field k, and ξ ∈ X (R). Then for every integer N ≥ 0, there exist
(1) an affine scheme Spec A of finite type over k and a k-point u ∈ Spec A;
(2) an object η ∈ X (A);
+1 ∼
(3) an isomorphism ϕN : R/mN R = A/mNu
+1
;
(4) an isomorphism of ξ|R/mN +1 and η|A/mNu
+1 via ϕN ; and
R

(5) an isomorphism Grm (R) = ∼ Grm (A) of graded k-algebras.

R u

569
 The proof of this theorem will proceed by simultaneously approximating equations
and relations defining R and the object ξ. The statements (1)–(4) will be easily
obtained as a consequence of Artin Approximation. It is a nice insight of Conrad
and de Jong that condition (5) can also be ensured by Artin Approximation, and
moreover that this condition suffices to imply the isomorphism of complete local
k-algebras in Artin Algebraization. Unsurprisingly, (5) takes the most work to
establish.
We will need some preparatory results controlling the constant appearing in the
Artin–Rees Lemma (B.5.4).
Definition C.6.3 (Artin–Rees Condition). Let (A, m) be a noetherian local ring,
φ : M → N be a morphism of finite A-modules, and c ≥ 0 be an integer. We say
that (AR)c holds for φ if

φ(M ) ∩ mn N ⊆ φ(mn−c M ), ∀n ≥ c.

The Artin–Rees Lemma (B.5.4) implies that (AR)c holds for φ if c ≫ 0.

Lemma C.6.4. Let (A, m) be a noetherian local ring. Let

α β α′ β′
L−
→M −
→N and L′ −→ M −→ N

be two complexes of finite A-modules. Let c be a positive integer. Assume that

(a) the first sequence is exact,
(b) the complexes are isomorphic modulo mc+1 , and
(c) (AR)c holds for α and β.
Then there exists an isomorphism Grm (coker β) → Grm (coker β ′ ) of graded Grm (A)-
modules.
Proof. The proof, while technical, is rather straightforward. First, by taking free
presentations of L and L′ , we can assume that L = L′ . One shows that (AR)c holds
for β ′ and that the second sequence is exact. Then one establishes the equality

mn+1 N + β(M ) ∩ mn N = mn+1 N + β ′ (M ) ∩ mn N

by using that (AR)c holds for β to show the containment “⊆”, and then using that
(AR)c holds for β ′ to get the other containment. The statement then follows from
the description Grm (coker β)n = mn N/(mn+1 N + β(M ) ∩ mn N ) and the similar
description of Grm (coker β ′ )n . For details, see [CdJ02, §3] and [SP, Tag 07VF].
Proof of Conrad–de Jong Approximation (Theorem C.6.2). Since X is limit preserv-
ing and R is the colimit of its finitely generated k-subalgebras, there is an affine
scheme V = Spec B of finite type over k and an object γ of X over V together with
a 2-commutative diagram

&
Spec R /V γ
/ X.

Let v ∈ V be the image of the maximal ideal m ⊆ R. After adding generators to

the ring B if necessary, we can assume that the composition ObV,v → R → R/m2
is surjective. This implies that O
bV,v → R is surjective by Complete Nakayama’s
Lemma (B.5.6(3)). The goal now is to simultaneously approximate over V the

570
equations and relations defining the closed immersion Spec R ,→ Spec O
bV,v and the
object ξ. To accomplish this goal, we choose a resolution

b⊕r − b b ⊕s β b
α
(C.6.5)
b
O V,v → OV,v −
→ OV,v → R → 0

as O
bV,v -modules and consider the functor

F : (Sch /V ) → Sets
α β
(T → V ) 7→ complexes OT⊕r −
→ OT⊕s −

→ OT .

It is not hard to check that this functor is limit preserving. The resolution in (C.6.5)
yields an element of F (ObV,v ). Applying Artin Approximation (B.5.18) gives an étale
morphism (V = Spec B ′ , v ′ ) → (V, v) and an element
′

α′ β′
(B ′⊕r −→ B ′⊕s −→ B ′ ) ∈ F (V ′ ) (C.6.6)

such that α′ , β ′ are equal to α

b, βb modulo mN +1 .
Let U = Spec A ,→ Spec B ′ = V ′ be the closed subscheme defined by im β ′ and
let u = v ′ ∈ U . Consider the composition
γ
η : U ,→ V ′ → V −
→X

As R = coker βb and A = coker β ′ , we have an isomorphism R/mN +1 ∼ = A/mN u

+1

together with an isomorphism of ξ|R/mN +1 and η|A/mN u

+1 . This gives statements

(1)–(4).
To establish (5), we need to show that there are isomorphisms mn /mn+1 ∼ =
N +1 ∼
n
mu /mn+1
u . For n ≤ N , this is guaranteed by the isomorphism R/m = A/m N +1
u .
On the other hand, for n ≫ 0, this can be seen to be a consequence of the Artin–Rees
Lemma (B.5.4). To handle the middle range of n, we need to control the constant
appearing in the Artin–Rees Lemma. First, note that before we applied Artin
Approximation, we could have increased N to ensure that (AR)N holds for α b and
β. We are thus free to assume this. Now statement (5) follows directly if we apply
b
b⊕r − b b ⊕s β b
α
Lemma C.6.4 to the exact complex O → OV,v of (C.6.5) and the complex
b
V,v → OV,v −
′ b′
b⊕r −
O
α b⊕s β→ ObV,v obtained by restricting (C.6.6) to F (O
bV,v ). See also [CdJ02]
V,v → OV,v −
b

and [SP, Tag 07XB].

Exercise C.6.7. Show that Conrad–de Jong Approximation implies Artin Approxi-
mation.

C.6.3 Artin Algebraization

Artin Algebraization has a stronger conclusion than Artin Approximation or Conrad–
de Jong Approximation in that no approximation is necessary. It guarantees the
existence of an object η over a pointed affine scheme (Spec A, u) of finite type over k,
which agrees with the given effective formal deformation ξ to all orders. To ensure
this, we need to impose that ξ is versal at u, i.e., that the restrictions ξn = ξ|A/mn+1
u
define a versal formal deformation {ξn } over A (Definition C.4.2).
Theorem C.6.8 (Artin Algebraization). Let X be a limit preserving prestack over
Sch/k. Let (R, m) be a complete noetherian local k-algebra and ξ ∈ X (R) be an
effective versal formal deformation. There exist

571
 (1) an affine scheme Spec A of finite type over k and a k-point u ∈ Spec A;
(2) an object η ∈ X (A);
∼ b
(3) an isomorphism α : R → A mu of k-algebras; and
(4) a compatible family of isomorphisms ξ|R/mn+1 = ∼ η| (under the identifi-
A/mn+1
u
n+1 ∼ n+1
cation R/m = A/m u ) for n ≥ 0.
Note that we are not asserting that ξ = ∼ η| b , although this does hold in some
Amu
settings, e.g., if X is an algebraic stack. See Example C.7.17 for an example where
the effective versal formal deformation is not unique, in which case there are two
algebraizations which are not étale-locally isomorphic.
Remark C.6.9. If R is known to be the completion of a finitely generated k-algebra,
this theorem can be viewed as an easy consequence of Artin Approximation. Indeed,
one applies Artin Approximation with N = 1 and then uses versality to obtain com-
patible maps R → A/mn+1 u and therefore a map R → A bm which is an isomorphism
u

modulo m2 . As R and A bm are abstractly isomorphic, the homomorphism R → A

u
bm
u
is an isomorphism by Complete Nakayama’s Lemma (B.5.6(3)) and the statement
follows. The argument in the general case is analogous, except we use Conrad–de
Jong Approximation instead of Artin Approximation.
Proof. Applying Conrad–de Jong Approximation (C.6.2) with N = 1, we obtain an
affine scheme Spec A of finite type over k with a k-point u ∈ Spec A, an object η ∈
X (A), an isomorphism ϕ2 : Spec A/m2u → Spec R/m2 , an isomorphism α2 : ξ|R/m2 →
η|A/m2u , and an isomorphism Grm (R) ∼ = Grmu (A) of graded k-algebras. We claim
that ϕ2 and α2 can be extended inductively to a compatible family of morphisms
ϕn : Spec A/mn+1
u → Spec R and isomorphisms αn : ξ|A/mn+1u
→ η|A/mn+1
u
. Indeed,
given ϕn and αn , versality of ξ implies that there is a lift ϕn+1 filling in the
commutative diagram
Spec A/mnu
ϕn
/ Spec R
8
ξ
 ϕn+1

Spec A/mn+1
u
/ X.
η|A/mn+1
u

By taking the limit, we have a homomorphism ϕb : R → A bm which is surjective

u
by Complete Nakayama’s Lemma (B.5.6(3)). On the other hand, for each n the
k-vector spaces mN /mN +1 and mN u /mu
N +1
have the same dimension. This implies
that ϕ is an isomorphism. See also [Art69b, Thm. 1.6] and [CdJ02, §4].
b

C.7 Artin’s Axioms for Algebraicity

As a general fact, our knowledge of nonprojective existence theorems is
exceedingly poor, and I hope this will change eventually.

Grothendieck, letter to Murre, 1962 [Mum10, p. 663]

Artin’s Axioms for Algebraicity provide criteria, often verifiable in practice,

ensuring that a given stack is algebraic. This foundational result was proved by
Artin in the very same paper [Art74b] where he introduced algebraic stacks. We
provide two versions below: Theorems C.7.1 and C.7.4. The first version is a fairly
easy consequence of Artin Algebraization (C.6.8).

572
Theorem C.7.1. (Artin’s Axioms for Algebraicity—first version) Let X be a stack
over (Sch/k)ét . Then X is an algebraic stack locally of finite type over k if and only
if the following conditions hold:
(1) (Limit preserving) The stack X is limit preserving over Sch/k, i.e., for every
system Bλ of k-algebras, the functor
colim X (Bλ ) → X (colim Bλ )
is an equivalence of categories.
(2) (Representability of the diagonal) The diagonal X → X × X is representable.
(3) (Existence of versal formal deformations) Every x0 ∈ X (k) has a versal formal
deformation {xn } over a complete noetherian local k-algebra (R, m) with residue
field k.
(4) (Effectivity) For every complete noetherian local k-algebra (R, m) with residue
field k, the natural functor
X (Spec R) → lim X (Spec R/mn )
←−
is an equivalence of categories.
(5) (Openness of versality) For every morphism U → X from a finite type k-scheme
bU,u /mn+1 →
which is versal at u ∈ U (k) (i.e., the formal deformation {Spec O u
X } is versal), there exists an open neighborhood V of u such that U → X is
versal at every k-point of V .
Proof. We first note that for a representable and locally of finite type morphism
U → X from a finite type k-scheme U , the Infinitesimal Lifting Criterion for
Smoothness (3.7.1) implies that U → X is smooth if and only if it is versal at all
k-points u ∈ U . For (⇒), (1) holds by Exercise 3.3.33, (2) holds by Theorem 3.2.1,
and (4) holds by Example C.5.2. If U → X is a morphism from a finite type
k-scheme, then it is necessarily representable and locally of finite type. Part (3)
holds by choosing a smooth presentation U → X and a preimage u ∈ U (k) of x0 and
taking the formal deformation {Spec OU,u /mn+1 u → X }. Part (5) holds by openness
of smoothness.
For the converse, we first note that representability of the diagonal, i.e., condition
(2), implies that every morphism U → X from a scheme U is representable, and the
limit preserving property (1) implies that U → X is locally of finite type. For every
object x0 ∈ X (k), we will construct a smooth morphism U → X from a scheme
and a preimage u ∈ U (k) of x0 . Conditions (3)–(4) guarantee that there exists an
effective versal formal deformation xb : Spec R → X of x0 where (R, m) is a complete
noetherian local k-algebra with residue field k. By Artin Algebraization (C.6.8),
there exists a finite type k-scheme U , a point u ∈ U (k), a morphism p : U → X , an
∼
isomorphism R ∼ =ObU,u , and compatible isomorphisms p|R/mn+1 → xb|R/mn+1 . By
(5), we can replace U with an open neighborhood of u so that U → X is versal
(or in other words smooth) at every k-point of U . See also [Art74b, §5], [LMB00,
Cor. 10.11], and [SP, Tag 07Y4].
Remark C.7.2. In practice, condition (1)–(4) are often easy to verify directly with
(3) a consequence of Rim–Schlessinger’s Criteria (C.4.6) and (4) a consequence of
Grothendieck’s Existence Theorem (C.5.3). Also note that (2) can sometimes be
established by applying the theorem to the diagonal X → X × X , i.e., to the Isom
sheaves IsomT (x, y) of objects x, y ∈ X (T ) over a scheme T . In some specialized
cases, (5) can be checked directly, but it is frequently verified as a consequence of a
well-behaved deformation and obstruction theory, as we explain in the next section.

573
C.7.1 Artin’s Axioms via an obstruction theory
We state a refinement of Artin’s Axioms for Algebraicity that is often easier to verify
in practice. Namely, we show that openness of versality (C.7.1(5)) holds if the stack
has a well-behaved deformation theory and if there exists a well-behaved obstruction
theory.
To formulate the statements, we will need a bit of notation. Let ξ ∈ X (A) be an
object over a finitely generated k-algebra A. Let M be a finite A-module and denote
by A[M ] the ring A ⊕ M defined by M 2 = 0. Let Def ξ (M ) the set of isomorphism
classes of diagrams

_ A
Spec
ξ
/: X
η


Spec A[M ],

where an isomorphism of two extensions η, η ′ : Spec A[M ] → X is by definition an

∼
isomorphism η → η ′ in X (A[M ]) restricting to the identity on ξ. Let Autξ (M ) be the
group of automorphisms of the trivial deformation ξ ′ : Spec A[M ] → Spec A → X .
Note that when ξ ∈ X (k), then Def ξ (k) is precisely the tangent space of X at ξ,
while Autξ (k) is the group of infinitesimal automorphism of ξ, i.e. the kernel of
AutX (k[ϵ]) (ξ ′ ) → AutX (k) (ξ).

Lemma C.7.3. Suppose that X is a prestack over Sch/k satisfying the strong
homogeneity condition (RS∗4 ). Let ξ ∈ X (A) be an object over a finitely generated
k-algebra A.
(1) For every A-module M , Def ξ (M ) and Autξ (M ) are naturally A-modules, and
the functors
Autξ (−) : Mod(A) → Mod(A)
Def ξ (−) : Mod(A) → Mod(A)
are A-linear.
(2) Consider a surjection B ↠ A of k-algebras with square-zero kernel I, and
let Liftξ (B) be the set of morphisms ξ → η over Spec A → Spec B where
(α : ξ → η) ∼ (α′ : ξ → η ′ ) if there is an isomorphism β : η → η ′ such that
α′ = β ◦ α. There is an action of Def ξ (I) on Liftξ (B) which is functorial in
B and I. Assuming Liftξ (B) is non-empty, this action is free and transitive.

Proof. This can be established by arguing as in Lemma C.4.9. For instance, scalar
multiplication by x ∈ A is defined by pulling back along the morphism Spec A[M ] →
Spec A[M ] induced by the A-algebra homomorphism A[M ] → A[M ], a+m 7→ a+xm.
Condition (RS∗4 ) implies that the functor X (A[M ⊕ M ]) → X (A[M ]) ×X (A) X (A[M ])
is an equivalence. Addition M ⊕ M → M induces an A-algebra homomorphism
A[M ⊕ M ] → A[M ] and thus a functor

X (A[M ]) ×X (A) X (A[M ]) ∼

= X (A[M ⊕ M ]) → X (A[M ])

which defines addition on Def ξ (M ) and Autξ (M ).

Unlike the automorphism Def ξ (−) and deformation Def ξ (−) functors which are
intrinsic to the moduli problem, an obstruction functor is an additional piece of
data.

574
 Theorem C.7.4 (Artin’s Axioms for Algebraicity—second version). A stack X over
(Sch/k)ét is an algebraic stack locally of finite type over k if the following conditions
hold.
(AA1 ) (Limit preserving) The stack X is limit preserving.
(AA2 ) (Representability of the diagonal) The diagonal X → X × X is representable.
(AA3 ) (Finiteness of tangent spaces) For every object ξ : Spec k → X , Def ξ (k) is a
finite dimensional k-vector space.
(AA4 ) (Strong homogeneity) For every k-algebra homomorphism B0 → A0 and
surjection A ↠ A0 of k-algebras with square-zero kernel, the functor

X (B0 ×A0 A) → X (B0 ) ×X (A0 ) X (A)

is an equivalence, i.e., (RS∗4 ) holds.

(AA5 ) (Effectivity) For every complete noetherian local k-algebra (R, m) with residue
field k, the natural functor

X (Spec R) → lim X (Spec R/mn )

←−
is an equivalence of categories.
(AA6 ) (Coherent deformation theory) For every object ξ ∈ X (A) over a k-algebra A,
the functor Def ξ (−) commutes with products.
(AA7 ) (Existence of a coherent obstruction theory) For every object ξ ∈ X (A) over a
k-algebra A, there exists the following data
(a) there is an A-linear functor

Obξ (−) : Mod(A) → Mod(A),

and for every surjection B → A with square-zero kernel I, there is an

element obξ (B) ∈ Obξ (I) such that there is an extension

_ A
Spec
ξ
/X
<


Spec B

if and only if obξ (B) = 0, and

(b) for every composition B → B ′ → A of k-algebras such that B ↠ A and
B ′ ↠ A are surjective with square-zero kernels I and I ′ , the image of
obξ (B) under Obξ (I) → Obξ (I ′ ) is obξ (B ′ ).
(c) For every object ξ ∈ X (A) over a k-algebra A, the functor Obξ (−)
commutes with products.

Moreover, (AA2 ) can be replaced with

(AA2′ ) For every object ξ : Spec k → X , Autξ (k) is a finite dimensional k-vector
space, and for every object ξ ∈ X (A) over a k-algebra A, the functor Autξ (−)
commutes with products.
Proof. We verify the conditions of Theorem C.7.1. By (AA3 )–(AA4 ), we may
apply Rim–Schlessinger’s Criteria (C.4.6) to deduce the existence of versal formal
deformations, i.e., C.7.1(3) holds. It remains to check openness of versality, i.e.,

575
C.7.1(5). Let ξ0 : U0 → X be a morphism from an affine scheme U0 = Spec B0
of finite type over k such that ξ0 is versal at a point u0 ∈ U0 (k). By (AA1 )–
(AA2 ), the morphism ξ0 : U0 → X is representable and locally of finite type. Let
Σ = {u ∈ U0 (k) | ξ0 : U0 → X is not versal at u}. If openness of versality does not
hold, then u0 ∈ Σ and there exists a countably infinite subset Σ′ = {u1 , u2 , . . .} ⊆ Σ
of distinct points with u0 ∈ Σ′ .
Step 1. We claim that there exists a commutative diagram
  
U0  / U1  / U2  / ···
ξ1
ξ0
 ~v ξ2
X

where each closed immersion Un−1 ,→ Un is defined by a short exact sequence

0 → κ(un ) → OUn → OUn−1 → 0,

and there exists open neighborhoods Wn ⊆ Un ∖ {u0 , . . . , un−1 } of un such that each
restriction ξn |Wn is not the trivial deformation of ξ0 |Wn ∩U0 . By construction, each
Un = Spec Bn is an affine scheme and each closed immersion Un ,→ Um for m ≥ n is
defined by a square-zero ideal. Suppose that we’ve already constructed ξ0 , . . . , ξn−1 .
Since ξ0 : U0 → X and ξn−1 : Un−1 → X are isomorphic in an open neighborhood of
un , the morphism ξn−1 : Un−1 → X is also not versal at un . Therefore, there exists
a surjection A → A0 in Artk with ker(A → A0 ) = k and a commutative diagram

un

 (
Spec k / Spec A0 / Un−1
_ : (C.7.5)
̸∃
ξn−1
 
Spec A / X,

such that un is the image of Spec A0 → Un−1 , which does not admit a lift Spec A →
Un−1 . Using strong homogeneity (AA4 ), there exists an extension of the commutative
diagram
Spec A / Un−1 = Spec Bn−1
_ 0 _
ξn−1
 
Spec A / Un = Spec(A ×A0 Bn−1 )
ξn


/( X

yielding an object ξn over Un = Spec Bn with Bn := A ×A0 Bn−1 . If ξn were the

trivial deformation of ξ0 in an open neighborhood of un , then Spec A → X would be
the trivial deformation of Spec A0 contradicting the obstruction to a lift of (C.7.5).
Finally note that ker(Bn → Bn−1 ) = k since ker(A → A0 ) = k. This establishes the
claim.
Step 2. Letting Bb = lim Bn and U b we claim that there exists an object
b = Spec B,
←−
b ) extending each ξn ∈ X (Un ). Let Mn = ker(Bn → B0 ) (noting that
ξb ∈ X (U

576
M0 = 0). Since Mn2 = 0, we can view Mn as a B0 -module. The k-algebra
 
 Y 
B
e := (b0 , b1 , . . .) ∈ Bn the image of each bn under Bn → B0 is b0
 
n≥0

has the following properties.

– The surjective
Q k-algebra homomorphism B → B0 defined by (bi ) 7→ b0 has
e
kernel M := n≥0 Mn .

– The map B e → B0 [M ] defined by (b0 , b1 , b2 , . . .) 7→ (b0 , b1 −b0 , b2 −b1 , b3 −b2 , . . .)

is a surjective k-algebra homomorphism with square-zero kernel.

– The composition B e → B0 [M ] induces a short exact sequence

b→B

0 / ker(B
b → B0 ) / ker(B
e → B0 ) / ker(B0 [M ] → B0 ) /0

0 / lim M /Q Mn /Q Mn /0
←−n≥0 n n≥0 n≥0

(b0 , b1 , . . .)  / (b1 − b0 , b2 − b1 , . . .).

– There is an identification B e ×B [M ] B0 .
b=B
0

Since the lift ξn ∈ X (Bn ) of ξ0 exists for each n, obξ0 (Bn ) = 0 ∈ Obξ0 (Mn ). By
(AA7 )(b), the element obξ0 (B) e maps to obξ (Bn ) under Obξ (M ) → Obξ (Mn ). By
0 0 0

, the map Obξ0 (M ) ,→ n Obξ0 (Mn ) is injective4 and thus obξ0 (B)
Q e = 0 ∈ Obξ (M )
0

which shows that there exists a lift ξe ∈ X (B) e of ξ0 .

The restrictions ξ| e B are not necessarily isomorphic to ξn . However, we may
n
use the free and transitive action Def ξ0 (Mn ) = Liftξ0 (B0 [Mn ]) on the non-empty
set of liftings Liftξ0 (B
fn ) to find elements tn ∈ Def ξ (Mn ) such that ξn = tn · ξ|
0
eB
n
∼ Q
(Lemma C.7.3). Since Def ξ0 (M ) → n Def ξ0 (Mn ) by (AA6 ), there exists e t ∈
Def ξ0 (M ) mapping to (tn ). After replacing ξe with e e we can arrange that ξ|
t · ξ, eB
n
and ξn are isomorphic for each n.
We now show that each restriction ξ| e B [M ] ∈ Def ξ (Mn ) under the composition
0 n 0

Be → B0 [M ] → B0 [Mn ] is the trivial deformation. Indeed, the map M = ker(B e→

B0 ) → ker(B0 [Mn ] → B0 ) = Mn induces a map Def ξ0 (M ) → Def ξ0 (Mn ) on
∼ Q
deformation modules, which under the identification Def ξ0 (M ) → n Def ξ0 (Mn ) of
(AA6 ), sends an element (η0 , η1 , . . .) to (ηn+1 |Bn − ηn ). The ring map B e → B0 [Mn ]
also induces a map Liftξ0 (B) e → Liftξ (B0 [Mn ]) which is equivariant with respect
0

to Def ξ0 (M ) → Def ξ0 (Mn ). It follows that the image of ξe in Liftξ0 (B0 [Mn ]) =
Def ξ0 (Mn ) is ξn+1 |Bn − ξn = 0.
The existence of ξb ∈ X (B)b extending (ξn ) ∈ lim X (Bn ) now follows from applying
←−

4
Q The hypotheses of (AA7 )(c) can be weakened to only require the injectivity of Obξ0 (M ) ,→
n Obξ0 (Mn ), although in practice one usually verifies bijectivity.

577
the identity B e ×B [M ] B0 and strong homogeneity (AA4 ) to the diagram
b=B
0

Spec B 0 [M ] / Spec B0
_ _

  ξ0
Spec B
e / Spec B
b
ξb
ξe
# 
/ X.

Step 3. We now use the versality of ξ0 : U0 → X at u0 to arrive at a contradiction.

Since X is limit preserving (AA1 ), there exists a finitely generated k-subalgebra
∼ ′
b and an object ξ ′ ∈ X (B ′ ) together with an isomorphism ξb →
B′ ⊆ B ξ |Bb . After
possibly enlarging B , we may assume that the composition B ,→ B → B0 is
′ ′ b
surjective. Since ker(Bb → B0 ) is square-zero, so is ker(B ′ → B0 ). This defines
a closed immersion U0 ,→ U ′ := Spec B ′ , and we can consider the commutative
diagram

U0
i

$ %
U 0 ×X U ′ / U ′ = Spec B ′
id
ξ′
%  ξ0 
/ X.
U0
where the fiber product U0 ×X U ′ is an algebraic space locally of finite type over
k. Since ξ0 : U0 → X is versal at u0 , it follows from (the artinian version of)
the Infinitesimal Lifting Criterion for Smoothness (3.7.1) that U0 ×X U ′ → U ′
is smooth at i(u0 ). After replacing U0 and U ′ with affine open neighborhoods
and {u1 , u2 , . . .} with an infinite subsequence contained in these open subsets, we
can arrange that U0 ×X U ′ → U ′ is smooth. The (non-artinian version) of the
Infinitesimal Lifting Criterion for Smoothness (A.3.1) for schemes implies that the
section of U0 ×X U ′ → U ′ over U0 extends to a global section U ′ → U0 ×X U ′ . This
implies that ξ ′ is the trivial deformation of ξ0 , which in turn implies that each ξn is
a trivial deformation of ξ0 , a contradiction.
For the addendum, we show that (AA2′ ) implies (AA2 ), i.e., the representability
of the diagonal. By using the Limit Methods of §B.3, it suffices to consider maps
(a, b) : T → X ×k X from a finite type k-scheme. The deformation functor Def ξ (−)
for the base change IsomT (a, b) → T corresponds to the automorphism functor
Autξ (−) for X , and we take obstruction functor Obξ (−) for IsomT (a, b) to be the
deformation functor Def ξ (−) for X . Our exposition follows [SP, Tag 0CYF] and
[Hal17, Thm. A]. See also [Art69b, ], [Art74b, Thm. 5.3] and [HR19a, Main Thm.]
for alternative versions of Artin’s Criteria, and [Mur95, Thm. 1] and [Mur64, Thm. 1]
for criteria for functors to abelian groups to be representable.

Remark C.7.6. The converse of the theorem also holds. For the necessity of the
conditions, we only need to check (AA3 ), (AA4 ), (AA6 ), and (AA7 ). Condition
(AA3 ) (finiteness of the tangent spaces) holds as X is of finite type over k, and (AA4 )
(strong homogeneity) holds by [SP, Tag 07WN]. Condition (AA7 ) (existence of an
obstruction theory) follows from the existence of a cotangent complex LX /k for X

578
satisfying properties analogous to Theorem C.3.1; see [Ols06]. If ξ : Spec A → X is
a morphism from a finitely generated k-algebra A and I is an A-module, then we set
Obξ (I) := Ext1A (ξ ∗ LX /k , I). Finally, it follows from cohomology and base change
(see [Hal14a]) that Def ξ (−) and Obξ (−) commute with products.

C.7.2 Verifying Artin’s Axioms

Theorem C.7.7. Let X be a proper scheme over a field k.
(1) The Hilbert functor Hilb(X) : Sch/k → Sets, whose objects over S are closed
subschemes Z ⊆ XS flat and finitely presented over S, is an algebraic space
locally of finite type over k.
(2) The prestack Mall g over Sch/k, whose objects over S are proper, flat, and
finitely presented morphisms Y → S of algebraic spaces with one dimensional
fibers, is an algebraic stack locally of finite type over k.
(3) The prestack Coh(X) over Sch/k, whose objects over S are finitely presented
quasi-coherent OXS -modules flat over S, is an algebraic stack locally of finite
type over k.
When X is projective, (1) and (3) are established by more explicit methods in
Theorem 1.1.2 and Exercise 3.1.23, while (2) was established in Theorem 5.4.6.
Proof. Fpqc Descent for Quasi-Coherent Sheaves (2.1.4) implies that Hilb(X) is a
sheaf and that Coh(X) is a stack, and Exercise 4.5.15 implies that Mall g is a stack.
We check the conditions of Theorem C.7.4. Condition (AA1 ) (limit preserving)
was verified in Exercise C.6.1. Condition (AA3 ) and the first part of (AA2′ ), i.e.,
the finite dimensionality of Autξ (k) and Def ξ (k), follow from the identifications
with 0 and Ext0OX (IZ0 , OZ0 ) for ξ = [Z0 ⊆ X] ∈ Hilb(X)(k) (Proposition C.2.2),
with ExtiOC (C , OC ) for i = 0, 1 for ξ = [C] ∈ Mall
g (k) (Theorem C.3.6), and with
ExtiOX (E, E) for ξ = [E] ∈ Coh(X)(k) for i = 0, 1 (Proposition C.2.11). Condition
(AA4 ) (the strong homogeneity condition of (RS∗4 )) was verified in Proposition C.4.14.
Condition (AA5 ) (effectivity) was checked in Proposition C.5.10 as a consequence of
Grothendieck’s Existence Theorem. For (AA7 ), we define obstruction theories as
follows: for a k-algebra A and an A-module M , we set
– Obξ (M ) := Ext1OX (IZ , OZ ⊗A M ) for ξ = [Z ⊆ XA ] ∈ HilbP (X)(A),
A

– Obξ (M ) := Ext2OC (LC/A , M ) = 0 for ξ = [C → Spec A] ∈ Mall

g (A), and

– Obξ (M ) := Ext2OX (E, E ⊗A M ) for ξ = [E] ∈ Coh(X)(A).

A

Condition (AA7 )(a)–(b) follow from Proposition C.2.2, Theorem C.3.6, and Proposi-
tion C.2.11, which also provides cohomological identifications with Autξ (M ) and
Def ξ (M ). Condition (AA6 ), (AA7 )(c), and the second part of (AA2′ ) (Autξ (−),
Def ξ (−), and Obξ (−) commutes with products) follows from Lemma C.7.8. See also
[Art69b, Cor. 6.2], [Lie06, Thm. 2.11], and [SP, Tags 09TU, 0D5A, and 08KA].
Lemma C.7.8. Let X → Spec A be a flat proper morphism of schemes. Let E and
F be coherent sheaves on X with F flat over A. The functors
Hi (X, F ⊗A −) : Mod(A) → Mod(A)
ExtiOX (LX/A , −) : Mod(A) → Mod(A)
ExtiOX (E, F ⊗A −) : Mod(A) → Mod(A)
commute with products.

579
Proof. Since F is flat over A, there is a perfect complex K • of A-modules such that
Hi (X, F ⊗A −) ∼
= Hi (K • ⊗A −) (Theorem A.6.2). Write K d = A⊕rd . For every set
of A-modules {Mα } we have an identification of complexes

0 / Q Mα⊕r0 / Q Mα⊕r1 / ··· / Q Mα⊕rn /0

α α α

0 / (Q Mα )⊕r0 / (Q Mα )⊕r1 / ··· / (Q Mα )⊕rn / 0.

α α α

The top row isQthe product of the complexes K • ⊗A Mα and its cohomology is
identified with α Hi (X, F ⊗A MQ α ), while the bottom row is K ⊗A ( α Mα ) with
•
Q
cohomology groups Hi (X, F ⊗A ( α Mα )). For the remaining statements, one needs
to apply more sophisticated versions of cohomology and base change; see [EGA,
III.7.7.5], [SP, Tag 08JR], and [Hal14b, Thm. E].

Exercise C.7.9 (Hom stacks, hard). Let X and Y be proper Deligne–Mumford

stacks over a field k. Show that the stack Mor(X , Y) over (Sch/k)ét , whose fiber
category over a k-scheme S is Mor(XS , YS ), is an algebraic stack locally of finite
type.

Exercise C.7.10 (Weil restriction, hard). Let S ′ → S is be a proper flat morphism

of Deligne–Mumford stacks locally of finite type over k. Let X ′ → S ′ be a locally of
finite type morphism. Show that the stack ReS ′ /S (X ′ ) over (Sch/S)ét whose fiber
category over an S-scheme T is X ′ (T ×S S ′ ) is representable by an algebraic stack
locally of finite type over S. This is a generalization of Exercise 0.3.23.

C.7.3 Counterexamples
We prove examples of non-algebraic sheaves and stacks failing Artin’s Axioms.

Example C.7.11 (Automorphism of modules). For a module M over a ring A,

consider
AutA (M ) : Sch /A → Sets, S 7→ AutS (M ⊗A S).
If M is flat and finitely presented, then AutA (M ) is representable by a group scheme
(and in fact a finite type affine group scheme). This L fails if M is not flat or not
finitely presented. For example, if A = k and VS= n∈N k · ⟨ei ⟩, then Autk (V ) is
not limit preserving: letting B = k[x1 , x2 , . . .] = n Bn with Bn = k[x1 , . . . , xn ], the
B-automorphism of V ⊗k B, defined by ei 7→ ei + xi+1 ei+1 , is not induced by a Bn -
automorphism of V ⊗k Bn for any n, i.e., colimn AutBn (V ⊗k Bn ) → AutB (V ⊗k B)
is not surjective.
On the other hand, let A = k[x] and M = A/(x) = k. Suppose that G =
Autk[x] (k) is representable by an algebraic space. Sections of G → Spec k[x] corre-
spond to Autk[x] (k) = k× . Two distinct sections must restrict to distinct sections
over Spec k[x]x , but this contradicts that G restricts to the trivial group scheme over
Spec k[x]x . The sheaf G fails the strong homogeneity condition (RS∗4 ) of axiom (AA4 ).
In fact, it fails Rim–Schlessinger’s Condition (RS2 ) (or equivalently Schlessinger’s
Condition (H2 )): G(k[ϵ] ×k k[ϵ)) → G(k[ϵ]) ×G(k) G(k[ϵ]) is not bijective, where k[ϵ]
has the k[x]-algebra structure via x 7→ ϵ. Indeed, G(k[ϵ]) = Autk[ϵ] (k) = k× = G(k),

×
but G(k[ϵ] ×k k[ϵ]) = k[ϵ] ×k k[ϵ]/(ϵ, ϵ) = ∼ k[ϵ]× =
∼ k× × k.

Example C.7.12 (Stacks of quasi-coherent sheaves / non-flat coherent sheaves). If

X is a proper scheme over a field k, the prestack QCoh(X), whose objects over S are
quasi-coherent sheaves F flat over S, is not algebraic. The previous example shows

580
that QCoh(X) is not limit preserving (because the requisite functor is not even fully
faithful) and that the diagonal is not representable, i.e., both (AA1 ) and (AA2 ) fail.
Similarly, the stack of finitely presented (but not necessarily flat) quasi-coherent
sheaves is not algebraic nor limit preserving. By the previous example, (AA2 ) and
the fully faithfulness of (AA4 ) both fail.

Example C.7.13 (Automorphisms of schemes). If X is a scheme over k, consider

Aut(X) : Sch/k → Sets, S 7→ AutS (XS ).

If X is proper, this is representable by an algebraic space. Without properness, this

may fail. For example, Aut(A1 ) is not representable and fails (AA3 ): if ξ = {id} ∈
Aut(A1 )(k), then by Proposition C.2.4, there is an identification of the tangent space
Def ξ (k) with H0 (A1 , TA1 ) = k[x].

Example C.7.14 (Stack of all algebraic spaces). Let X be the prestack over Sch/k,
whose objects over a k-scheme S is a morphism X → S of algebraic spaces, and
where a morphism (X → S) → (X ′ → S ′ ) is a cartesian diagram. This is a stack
over (Sch/k)ét , but it is not limit preserving and the diagonal X → X × X is not
representable, i.e., both (AA1 ) and (AA2 ) fail. It also fails (AA2′ ) as Autξ (k) may
be infinite dimensional, as we saw in the previous example.

Example C.7.15 (Stack of K3 surfaces). The moduli stack K3 over (Sch/k)ét ,

whose objects over a k-scheme S are smooth and proper morphisms X → S of
algebraic spaces such that every fiber is a K3 surface, is not algebraic. It fails the
effectivity axiom (AA5 ); see Remark C.5.9.

We now describe Artin’s counterexamples from [Art69c]. In each case, the setup
is an inductive system of affine schemes

X1 → X2 → X3 → · · · with Xq = Spec Aq

and we consider the functor

colim Xq : AffSch/k → Sets, Spec R 7→ colim Mork (Spec R, Xi ),

where an element of F (R) is an equivalence class of a pair (q, ϕ : Aq → R) of an

positive integer and an k-algebra homomorphism, where (q, ϕ) ∼ (q ′ , ϕ′ ) if there
ϕ ϕ′
exists Q ≥ q, q ′ such that AQ → Aq −
→ R and AQ → Aq′ −→ R agree.

581
 (A) (B)

(C) (D)

Figure 3.7.16: Counterexamples to Artin’s Axioms

Example C.7.17 (Infinitely tangential curves). For Figure 3.7.16(A), let Xq =

Spec Aq , where Aq = k[x, y]/(y(y−xq )) and Aq+1 → Aq is defined by (x, y) 7→ (x, xy).
In F (k[t]/tn+1 ), consider the elements αn : Aq → k[t]/tn+1 (for any q), defined by
(x, y) 7→ (t, 0), and βn : Aq → k[t]/tn+1 (for any q), defined by (x, y) 7→ (t, tq ). For
each n, αn ∼ βn as we can take q ≥ n + 1. Then {αn } is a versal formal deformation
that effectivizes to two distinct elements α b, βb ∈ F (k[[t]]) defined on Aq (for any
q) by α b(x, y) 7→ (t, 0) and β(x,
b y) 7→ (t, t ). In this way, we see that the functor
q

F (k[[t]]) → lim F (k[t]/tq ) is not injective, so that the effectivity axiom (AA5 ) fails.
←−
In this
(A) case, C.7.1(5) (openness of versality) also fails. Likewise, the diagonal on F
(A)
∼
is not representable: the formal deformation of Isomk[[t]] (b b given by {αn →
α, β) βn }
is not effective. ('A)

Example C.7.18 (Infinitely many nodes). For Figure 3.7.16(B), let Xq be the
affine scheme over C with q nodes obtained from A1 by nodally-identifying q pairs of
points. Then A1 → colim Xq is formally versal at any point that doesn’t map to a
node, but this is not formally versal in any open neighborhood of A1 as one must
remove infinitely many points. Thus C.7.1(5) (openness of versality) fails.
Q1
Similarly, for Figure 3.7.16(C), if Xq = Spec k[x, y]/(y i=1 (x − q)), then the
inclusion A1 → colim Xq is formally versal at any non-nodal point, but is not formally
versal in an open neighborhood.
Qq
Example C.7.19 (Infinitely many lines). For Figure 3.7.16(D), let Xq = Spec k[x, y]/ i=1 (iy−
x) be the union of q lines in the plane, and Xq ,→ Xq+1 be the induced closed
immersions. Then colim Xq satisfyies Schlessinger’s Axioms (H1 )–(H4 ) (or equivalent
the Rim–Schlessinger Axioms (RS1 )–(RS4 )). Thus Schlessinger’s Theorem applies to
produce a formal versal deformation {xn }, where xn ∈ (colim Xq )(k[x, y]/(x, y)n+1 )
is defined by the closed immersion Spec k[x, y]/(x, y)n+1 ,→ Xn+1 . The effectivity
axiom (AA5 ) fails as there is no element of (colim Xq )(k[[x, y]]) extending {xn }.
Example C.7.20 (Hom stacks). The Hom stack Mor(X , Y) over (Sch/k)ét is
algebraic if X and Y are proper over k (Exercise C.7.9). This holds more generally
over an arbitrary base if X → S is proper and flat, and if Y → S is only assumed to
be locally of finite presentation, quasi-separated, and with affine stabilizer groups;

582
see [HR19b, Thm. 1.2] and [BHL17, Cor. 1.6]. In these settings, a version of Tannaka
∼
Duality (6.6.1) holds, i.e., Mor(X , Y) → Mor⊗ (Coh(Y), Coh(X )), and this reduces
the effectivity axiom (AA5 ) to a version of Grothendieck’s Existence Theorem.
It is essential however that Y have affine stabilizers. If Y is the classifying
stack of an abelian variety, then Tannaka Duality may not hold and the Hom stack
Mor(X , Y) may fail to be algebraic; see [SP, Tag 0AF8] and [HR19b, §10].
Exercise C.7.21 (Sheafification of the functor of smooth curves). Let F be the
sheafification in (Sch/k)ét of the functor assigning S to the set Mg (S)/∼ of isomor-
phism classes of families of smooth curves over S. Which of Artin Axioms fails for
F?

583
Bibliography

[AB60] Lars Ahlfors and Lipman Bers, Riemann’s mapping theorem for variable
metrics, Ann. of Math. (2) 72 (1960), 385–404.

[AB83] Michael Atiyah and Raoul Bott, The Yang-Mills equations over Riemann
surfaces, Philos. Trans. Roy. Soc. London Ser. A 308 (1983), no. 1505,
523–615.

[Abh56] Shreeram Abhyankar, Local uniformization on algebraic surfaces over

ground fields of characteristic p ̸= 0, Ann. of Math. (2) 63 (1956), 491–526.

[ACG11] Enrico Arbarello, Maurizio Cornalba, and Phillip A. Griffiths, Geometry

of algebraic curves. Volume II, Grundlehren der Mathematischen Wis-
senschaften [Fundamental Principles of Mathematical Sciences], vol. 268,
Springer, Heidelberg, 2011, With a contribution by Joseph Daniel Harris.

[ACGH85] Enrico Arbarello, Maurizio Cornalba, Phillip A. Griffiths, and Joseph

Harris, Geometry of algebraic curves. Vol. I, Grundlehren der mathematis-
chen Wissenschaften [Fundamental Principles of Mathematical Sciences],
vol. 267, Springer-Verlag, New York, 1985.

[ACV03] Dan Abramovich, Alessio Corti, and Angelo Vistoli, Twisted bundles
and admissible covers, vol. 31, 2003, Special issue in honor of Steven L.
Kleiman, pp. 3547–3618.

[AGV08] Dan Abramovich, Tom Graber, and Angelo Vistoli, Gromov-Witten theory
of Deligne-Mumford stacks, Amer. J. Math. 130 (2008), no. 5, 1337–1398.

[AHHL21] Jarod Alper, Jochen Heinloth, and Daniel Halpern-Leistner, Cartan-

Iwahori-Matsumoto decompositions for reductive groups, Pure Appl. Math.
Q. 17 (2021), no. 2, 593–604.

[AHHLR24] Jarod Alper, Jack Hall, Daniel Halpern-Leistner, and David Rydh,
Artin algebraization for pairs with applications to the local structure of
stacks and Ferrand pushouts, Forum Math. Sigma 12 (2024), Paper No.
e20, 25.

[Ahl53] Lars V. Ahlfors, Development of the theory of conformal mapping and

Riemann surfaces through a century, Contributions to the theory of
Riemann surfaces, Annals of Mathematics Studies, no. 30, Princeton
University Press, Princeton, N.J., 1953, pp. 3–13.

[Ahl61] , Some remarks on Teichmüller’s space of Riemann surfaces, Ann.

of Math. (2) 74 (1961), 171–191.

584
[AHLH23] Jarod Alper, Daniel Halpern-Leistner, and Jochen Heinloth, Existence
of moduli spaces for algebraic stacks, Invent. Math. 234 (2023), no. 3,
949–1038.

[AHR19] Jarod Alper, Jack Hall, and David Rydh, The étale local structure of
algebraic stacks, 2019.

[AHR20] , A Luna étale slice theorem for algebraic stacks, Ann. of Math.
(2) 191 (2020), no. 3, 675–738.

[AI97] Michael Atiyah and Daniel Iagolnitzer (eds.), Fields Medallists’ lectures,
World Scientific Series in 20th Century Mathematics, vol. 5, World Sci-
entific Publishing Co., Inc., River Edge, NJ; Singapore University Press,
Singapore, 1997.

[AJP16] Norbert A’Campo, Lizhen Ji, and Athanase Papadopoulos, On the early
history of moduli and teichmüller spaces, 2016, arXiv:1602.07208.

[AK80] Allen B. Altman and Steven L. Kleiman, Compactifying the Picard scheme,
Adv. in Math. 35 (1980), no. 1, 50–112.

[AK10] Valery Alexeev and Allen Knutson, Complete moduli spaces of branchva-
rieties, J. Reine Angew. Math. 639 (2010), 39–71.

[Ale86] Pappus of Alexandria, Book 7 of the collection, Sources in the History of

Mathematics and Physical Sciences, vol. 8, Springer-Verlag, New York,
1986, Part 1. Introduction, text, and translation, Part 2. Commentary,
index, and figures, Edited and with translation and commentary by
Alexander Jones.

[Alp13] Jarod Alper, Good moduli spaces for Artin stacks, Ann. Inst. Fourier
(Grenoble) 63 (2013), no. 6, 2349–2402.

[Alp14] , Adequate moduli spaces and geometrically reductive group

schemes, Algebr. Geom. 1 (2014), no. 4, 489–531.

[AM69] Michael F. Atiyah and Ian G. Macdonald, Introduction to commutative

algebra, Addison-Wesley Publishing Co., Reading, Mass.-London-Don
Mills, Ont., 1969.

[And67] Michel André, Méthode simpliciale en algèbre homologique et algèbre

commutative, Lecture Notes in Mathematics, Vol. 32, Springer-Verlag,
Berlin-New York, 1967.

[AOV08] Dan Abramovich, Martin Olsson, and Angelo Vistoli, Tame stacks in
positive characteristic, Ann. Inst. Fourier (Grenoble) 58 (2008), no. 4,
1057–1091.

[Ari15] Dima Arinkin, Existense of semi-stable vector bundles on

smooth curves in positive characteristic, MathOverflow ques-
tion at [Link]
existense-of-semi-stable-vector-bundles-on-smooth-curves-in-positive-characteris,
2015.

[Art62] Michael Artin, Grothendieck topolgoies, Notes from a seminar of M. Artin,

Harvard U., 1962.

585
[Art69a] , Algebraic approximation of structures over complete local rings,
Inst. Hautes Études Sci. Publ. Math. (1969), no. 36, 23–58.

[Art69b] , Algebraization of formal moduli. I, Global Analysis (Papers in

Honor of K. Kodaira), Univ. Tokyo Press, Tokyo, 1969, pp. 21–71.

[Art69c] , The implicit function theorem in algebraic geometry, Algebraic

Geometry (Internat. Colloq., Tata Inst. Fund. Res., Bombay, 1968), Tata
Inst. Fundam. Res. Stud. Math., vol. 4, Published for the Tata Institute
of Fundamental Research, Bombay by Oxford University Press, London,
1969, pp. 13–34.

[Art70] , Algebraization of formal moduli. II. Existence of modifications,

Ann. of Math. (2) 91 (1970), 88–135.

[Art74a] , Algebraic construction of Brieskorn’s resolutions, J. Algebra 29

(1974), 330–348.

[Art74b] , Versal deformations and algebraic stacks, Invent. Math. 27 (1974),

165–189.

[Art76] , Lectures on deformations of singularities, Lecture Notes, vol. 54,

Tata Institute of Fundamental Research, 1976.

[AS78] Enrico Arbarello and Edoardo Sernesi, Petri’s approach to the study of
the ideal associated to a special divisor, Invent. Math. 49 (1978), no. 2,
99–119.

[Ati56] Michael Atiyah, On the Krull-Schmidt theorem with application to sheaves,

Bull. Soc. Math. France 84 (1956), 307–317.

[Ati57] M. F. Atiyah, Vector bundles over an elliptic curve, Proc. London Math.
Soc. (3) 7 (1957), 414–452.

[Ati04] Michael Atiyah, Collected works. Vol. 6, Oxford Science Publications,

The Clarendon Press, Oxford University Press, New York, 2004.

[AW71] Michael Artin and Gayn Winters, Degenerate fibres and stable reduction
of curves, Topology 10 (1971), 373–383.

[Bab39] D. W. Babbage, A note on the quadrics through a canonical curve, J.

London Math. Soc. 14 (1939), 310–315.

[Bai58] Walter L. Baily, Jr., Satake’s compactification of Vn , Amer. J. Math. 80

(1958), 348–364.

[Bai60a] , On the moduli of Jacobian varieties, Ann. of Math. (2) 71 (1960),

303–314.

[Bai60b] , On the moduli of Jacobian varieties and curves, Contributions

to function theory (Internat. Colloq. Function Theory, Bombay, 1960),
Tata Institute of Fundamental Research, Bombay, 1960, pp. 51–62.

[Bal09] Vikraman Balaji, Lectures on principal bundles, Moduli spaces and vector
bundles, London Math. Soc. Lecture Note Ser., vol. 359, Cambridge Univ.
Press, Cambridge, 2009, pp. 2–28.

586
[Bar55a] Iacopo Barsotti, Structure theorems for group-varieties, Ann. Mat. Pura
Appl. (4) 38 (1955), 77–119.

[Bar55b] , Un teorema di struttura per le varietà gruppali, Atti Accad. Naz.

Lincei Rend. Cl. Sci. Fis. Mat. Nat. (8) 18 (1955), 43–50.

[Bar71] Charles M. Barton, Tensor products of ample vector bundles in character-

istic p, Amer. J. Math. 93 (1971), 429–438.

[Bay11] Arend Bayer, A tour to stability conditions on derived categories, 2011.

[BB63] Andrzej Białynicki-Birula, On homogeneous affine spaces of linear alge-

braic groups, Amer. J. Math. 85 (1963), 577–582.

[BB66] Walter L. Baily, Jr. and Armand Borel, Compactification of arithmetic

quotients of bounded symmetric domains, Ann. of Math. (2) 84 (1966),
442–528.

[BB73] Andrzej Białynicki-Birula, Some theorems on actions of algebraic groups,

Ann. of Math. (2) 98 (1973), 480–497.

[BB76] , Some properties of the decompositions of algebraic varieties de-

termined by actions of a torus, Bull. Acad. Polon. Sci. Sér. Sci. Math.
Astronom. Phys. 24 (1976), no. 9, 667–674.

[BBPGPR09] Steven B. Bradlow, Leticia Brambila-Paz, Oscar García-Prada, and

S. Ramanan (eds.), Moduli spaces and vector bundles, London Mathemat-
ical Society Lecture Note Series, vol. 359, Cambridge University Press,
Cambridge, 2009, Including papers from the Newstead Fest held in honor
of Peter E. Newstead’s 65th birthday in Guanajuato, December 2006.

[BCHM10] Caucher Birkar, Paolo Cascini, Christopher D. Hacon, and James McK-
ernan, Existence of minimal models for varieties of log general type, J.
Amer. Math. Soc. 23 (2010), no. 2, 405–468.

[Beh03] Kai Behrend, Derived l-adic categories for algebraic stacks, Mem. Amer.
Math. Soc. 163 (2003), no. 774, viii+93.

[Beh04] , Cohomology of stacks, Intersection theory and moduli, ICTP

Lect. Notes, XIX, Abdus Salam Int. Cent. Theoret. Phys., Trieste, 2004,
pp. 249–294.

[Beh14] , Introduction to algebraic stacks, Moduli spaces, London Math.

Soc. Lecture Note Ser., vol. 411, Cambridge Univ. Press, Cambridge, 2014,
pp. 1–131.

[Ber60] Lipman Bers, Spaces of Riemann surfaces, Proc. Internat. Congress Math.
1958, Cambridge Univ. Press, New York, 1960, pp. 349–361.

[Ber72] , Uniformization, moduli, and Kleinian groups, Bull. London Math.

Soc. 4 (1972), 257–300.

[Ber90] Vladimir G. Berkovich, Spectral theory and analytic geometry over non-
Archimedean fields, Mathematical Surveys and Monographs, vol. 33, Amer-
ican Mathematical Society, Providence, RI, 1990.

587
[BF97] Kai Behrend and Barbara Fantechi, The intrinsic normal cone, Invent.
Math. 128 (1997), no. 1, 45–88.

[BFMV14] Gilberto Bini, Fabio Felici, Margarida Melo, and Filippo Viviani, Geomet-
ric invariant theory for polarized curves, Lecture Notes in Mathematics,
vol. 2122, Springer, Cham, 2014.

[BGLM24] Lukas Braun, Daniel Greb, Kevin Langlois, and Joaquín Moraga, Reduc-
tive quotients of klt singularities, 2024, arXiv:2111.02812.

[BH93] Winfried Bruns and Jürgen Herzog, Cohen-Macaulay rings, Cambridge

Studies in Advanced Mathematics, vol. 39, Cambridge University Press,
Cambridge, 1993.

[BHL17] Bhargav Bhatt and Daniel Halpern-Leistner, Tannaka duality revisited,

Adv. Math. 316 (2017), 576–612.

[BHPV04] Wolf P. Barth, Klaus Hulek, Chris A. M. Peters, and Antonius Van de Ven,
Compact complex surfaces, second ed., Ergebnisse der Mathematik und
ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics
[Results in Mathematics and Related Areas. 3rd Series. A Series of Modern
Surveys in Mathematics], vol. 4, Springer-Verlag, Berlin, 2004.

[Bir09] George D. Birkhoff, Singular points of ordinary linear differential equa-

tions, Trans. Amer. Math. Soc. 10 (1909), no. 4, 436–470.

[BL24] Daniel Bragg and Max Lieblich, Murphy’s law for algebraic stacks, 2024,
arXiv:2402.00862.

[BLR90] Siegfried Bosch, Werner Lütkebohmert, and Michel Raynaud, Néron

models, Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results
in Mathematics and Related Areas (3)], vol. 21, Springer-Verlag, Berlin,
1990.

[BN06] Kai Behrend and Behrang Noohi, Uniformization of Deligne-Mumford

curves, J. Reine Angew. Math. 599 (2006), 111–153.

[Bol87] Oskar Bolza, On binary sextics with linear transformations into themselves,
Amer. J. Math. 10 (1887), no. 1, 47–70.

[Boo41] George Boole, Exposition of a general theory of linear transformations,

Cambridge Math J. 3 (1841), 106–119.

[Bor91] Armand Borel, Linear algebraic groups, second ed., Graduate Texts in
Mathematics, vol. 126, Springer-Verlag, New York, 1991.

[Bou87] Jean-François Boutot, Singularités rationnelles et quotients par les groupes

réductifs, Invent. Math. 88 (1987), no. 1, 65–68.

[Bou89] Nicolas Bourbaki, Commutative algebra. Chapters 1–7, Elements of Mathe-

matics (Berlin), Springer-Verlag, Berlin, 1989, Translated from the French,
Reprint of the 1972 edition.

[BOW24] Daniel Bragg, Martin Olsson, and Rachel Webb, Ample vector bundles
and moduli of tame stacks, 2024, arXiv:2407.01743.

588
[Bri97] Michel Brion, Equivariant Chow groups for torus actions, Transform.
Groups 2 (1997), no. 3, 225–267.

[Bri17] , Some structure theorems for algebraic groups, Algebraic groups:

structure and actions, Proc. Sympos. Pure Math., vol. 94, Amer. Math.
Soc., Providence, RI, 2017, pp. 53–126.

[Bro94] Kenneth S. Brown, Cohomology of groups, Graduate Texts in Mathematics,

vol. 87, Springer-Verlag, New York, 1994, Corrected reprint of the 1982
original.

[BS03] Holger Brenner and Stefan Schröer, Ample families, multihomogeneous

spectra, and algebraization of formal schemes, Pacific J. Math. 208 (2003),
no. 2, 209–230.

[BS08] Elizabeth Baldwin and David Swinarski, A geometric invariant theory

construction of moduli spaces of stable maps, Int. Math. Res. Pap. IMRP
(2008), no. 1, Art. ID rpn 004, 104.

[BT72] François Bruhat and Jacques Tits, Groupes réductifs sur un corps local,
Inst. Hautes Études Sci. Publ. Math. (1972), no. 41, 5–251.

[Bum13] Daniel Bump, Lie groups, second ed., Graduate Texts in Mathematics,
vol. 225, Springer, New York, 2013.

[Cad07] Charles Cadman, Using stacks to impose tangency conditions on curves,

Amer. J. Math. 129 (2007), no. 2, 405–427.

[Cap94] Lucia Caporaso, A compactification of the universal Picard variety over

the moduli space of stable curves, J. Amer. Math. Soc. 7 (1994), no. 3,
589–660.

[Car62] Pierre Cartier, Groupes algébriques et groupes formels, Colloq. Théorie des
Groupes Algébriques (Bruxelles, 1962), Librairie Universitaire, Louvain;
Gauthier-Villars, Paris, 1962, pp. 87–111.

[Cay45] Arthur Cayley, On the theory of linear transformations, Cambridge Math

J. 5 (1845), 193–209.

[Cay60] , A new analytic representation of curves in space, Cambridge

Math J. 3 (1860), 225–236.

[Cay62] , A new analytic representation of curves in space, Cambridge

Math J. 5 (1862), 81–86.

[CdJ02] Brian Conrad and A. Johan de Jong, Approximation of versal deforma-

tions, J. Algebra 255 (2002), no. 2, 489–515.

[CH88] Maurizio Cornalba and Joe Harris, Divisor classes associated to families
of stable varieties, with applications to the moduli space of curves, Ann.
Sci. École Norm. Sup. (4) 21 (1988), no. 3, 455–475.

[Che60] Claude Chevalley, Une démonstration d’un théorème sur les groupes
algébriques, J. Math. Pures Appl. (9) 39 (1960), 307–317.

589
[Cho57] Wei-Liang Chow, On the projective embedding of homogeneous varieties,
Algebraic geometry and topology. A symposium in honor of S. Lefschetz,
Princeton University Press, Princeton, N.J., 1957, pp. 122–128.

[Cle73] Alfred Clebsch, Zur Theorie der Riemann’schen Fläche, Math. Ann. 6
(1873), no. 2, 216–230.

[Cle70] , Zur Theorie der binären algebraischen Formen, Math. Ann. 3

(1870), no. 2, 265–267.

[CLM22] Raymond Cheng, Carl Lian, and Takumi Murayama, Projectivity of the
moduli of curves, Stacks Project Expository Collection (SPEC), London
Math. Soc. Lecture Note Ser., vol. 480, Cambridge Univ. Press, Cambridge,
2022, pp. 1–43.

[CLS11] David A. Cox, John B. Little, and Henry K. Schenck, Toric varieties,
Graduate Studies in Mathematics, vol. 124, American Mathematical
Society, Providence, RI, 2011.

[CLV23] Izzet Coskun, Eric Larson, and Isabel Vogt, The normal bundle of a general
canonical curve of genus at least 7 is semistable, 2023, arXiv:2203.13211.

[Con02] Brian Conrad, A modern proof of Chevalley’s theorem on algebraic groups,

J. Ramanujan Math. Soc. 17 (2002), no. 1, 1–18.

[Con05a] , Formal GAGA for Artin stacks, [Link]

~conrad/papers/[Link], 2005.

[Con05b] Brian Conrad, Keel-mori theorem via stacks, 2005.

[Con14] Brian Conrad, Reductive group schemes, Autour des schémas en groupes.
Vol. I, Panor. Synthèses, vol. 42/43, Soc. Math. France, Paris, 2014,
pp. 93–444.

[Cor93] M. D. T. Cornalba, On the projectivity of the moduli spaces of curves, J.

Reine Angew. Math. 443 (1993), 11–20.

[Cox95] David A. Cox, The homogeneous coordinate ring of a toric variety, J.

Algebraic Geom. 4 (1995), no. 1, 17–50.

[CP21] Giulio Codogni and Zsolt Patakfalvi, Positivity of the CM line bundle for
families of K-stable klt Fano varieties, Invent. Math. 223 (2021), no. 3,
811–894.

[CS01] Pierre Colmez and Jean-Pierre Serre (eds.), Correspondance Grothendieck-

Serre, Documents Mathématiques (Paris) [Mathematical Documents
(Paris)], vol. 2, Société Mathématique de France, Paris, 2001.

[CS79] James B. Carrell and Andrew John Sommese, C∗ -actions, Math. Scand.
43 (1978/79), no. 1, 49–59.

[CW37] Wei-Liang Chow and B. L. van der Waerden, Zur algebraischen Geometrie.
IX, Math. Ann. 113 (1937), no. 1, 692–704.

[DC71] Jean A. Dieudonné and James B. Carrell, Invariant theory, old and new,
Academic Press, New York-London, 1971.

590
[Del81] Pierre Deligne, Relèvement des surfaces K3 en caractéristique nulle,
Algebraic surfaces (Orsay, 1976–78), Lecture Notes in Math., vol. 868,
Springer, Berlin-New York, 1981, Prepared for publication by Luc Illusie,
pp. 58–79.

[Del85] , Le lemme de Gabber, Astérisque (1985), no. 127, 131–150, Seminar

on arithmetic bundles: the Mordell conjecture (Paris, 1983/84).

[DG70] Michel Demazure and Pierre Gabriel, Groupes algébriques. Tome I:

Géométrie algébrique, généralités, groupes commutatifs, Masson & Cie,
Éditeurs, Paris; North-Holland Publishing Co., Amsterdam, 1970, Avec
un appendice Corps de classes local par Michiel Hazewinkel.

[DH98] Igor V. Dolgachev and Yi Hu, Variation of geometric invariant theory

quotients, Inst. Hautes Études Sci. Publ. Math. (1998), no. 87, 5–56, With
an appendix by Nicolas Ressayre.

[dJ03] A. Johan de Jong, A result of gabber, preprint, [Link]

[Link]/~dejong/papers/[Link], 2003.

[DK15] Harm Derksen and Gregor Kemper, Computational invariant theory, en-
larged ed., Encyclopaedia of Mathematical Sciences, vol. 130, Springer,
Heidelberg, 2015, With two appendices by Vladimir L. Popov, and an ad-
dendum by Norbert A’Campo and Popov, Invariant Theory and Algebraic
Transformation Groups, VIII.

[DM69] Pierre Deligne and David Mumford, The irreducibility of the space of
curves of given genus, Inst. Hautes Études Sci. Publ. Math. (1969), no. 36,
75–109.

[DMOS82] Pierre Deligne, James S. Milne, Arthur Ogus, and Kuang-yen Shih, Hodge
cycles, motives, and Shimura varieties, Lecture Notes in Mathematics,
vol. 900, Springer-Verlag, Berlin-New York, 1982.

[Dol03] Igor Dolgachev, Lectures on invariant theory, London Mathematical Soci-

ety Lecture Note Series, vol. 296, Cambridge University Press, Cambridge,
2003.

[Dri13] Vladimir Drinfeld, On algebraic spaces with an action of Gm , 2013,

arXiv:1308.2604.

[EG84] David Eisenbud and Shiro Goto, Linear free resolutions and minimal
multiplicity, J. Algebra 88 (1984), no. 1, 89–133.

[EG98] Dan Edidin and William Graham, Equivariant intersection theory, Invent.
Math. 131 (1998), no. 3, 595–634.

[EGA] Alexander Grothendieck, Éléments de géométrie algébrique, I.H.E.S. Publ.

Math. 4, 8, 11, 17, 20, 24, 28, 32 (1960, 1961, 1961, 1963, 1964, 1965,
1966, 1967).

[EH87] David Eisenbud and Joe Harris, The Kodaira dimension of the moduli
space of curves of genus ≥ 23, Invent. Math. 90 (1987), no. 2, 359–387.

[EH00] , The geometry of schemes, Graduate Texts in Mathematics, vol.

197, Springer-Verlag, New York, 2000.

591
[EHKV01] Dan Edidin, Brendan Hassett, Andrew Kresch, and Angelo Vistoli,
Brauer groups and quotient stacks, Amer. J. Math. 123 (2001), no. 4,
761–777.

[EHM92] Philippe Ellia, André Hirschowitz, and Emilia Mezzetti, On the number
of irreducible components of the Hilbert scheme of smooth space curves,
Internat. J. Math. 3 (1992), no. 6, 799–807.

[Eis95] David Eisenbud, Commutative algebra, Graduate Texts in Mathematics,

vol. 150, Springer-Verlag, New York, 1995, With a view toward algebraic
geometry.

[Eke88] Torsten Ekedahl, Canonical models of surfaces of general type in positive

characteristic, Inst. Hautes Études Sci. Publ. Math. (1988), no. 67, 97–144.

[Enr19] Federigo Enriques, Sulle curve canoniche di genere p dello spazio a p − 1

dimensioni, Rend. Accad. Sci. Ist. Bologna 23 (1919), 80–82.

[ER21] Dan Edidin and David Rydh, Canonical reduction of stabilizers for Artin
stacks with good moduli spaces, Duke Math. J. 170 (2021), no. 5, 827–880.

[ES87] Geir Ellingsrud and Stein Arild Strømme, On the homology of the Hilbert
scheme of points in the plane, Invent. Math. 87 (1987), no. 2, 343–352.

[ES14] Torsten Ekedahl and Roy Skjelnes, Recovering the good component of the
Hilbert scheme, Ann. of Math. (2) 179 (2014), no. 3, 805–841.

[Fal93] Gerd Faltings, Stable G-bundles and projective connections, J. Algebraic

Geom. 2 (1993), no. 3, 507–568.

[Fal03] , Finiteness of coherent cohomology for proper fppf stacks, J. Alge-

braic Geom. 12 (2003), no. 2, 357–366.

[FC90] Gerd Faltings and Ching-Li Chai, Degeneration of abelian varieties, Ergeb-
nisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics
and Related Areas (3)], vol. 22, Springer-Verlag, Berlin, 1990, With an
appendix by David Mumford.

[Fer03] Daniel Ferrand, Conducteur, descente et pincement, Bull. Soc. Math.

France 131 (2003), no. 4, 553–585.

[FGAI] Alexander Grothendieck, Technique de descente et théorèmes d’existence

en géométrie algébrique. I. Généralités. Descente par morphismes fidèle-
ment plats, Séminaire Bourbaki, Vol. 5, Soc. Math. France, Paris, 1959–60,
pp. Exp. No. 190, 299–327.

[FGAIII] , Techniques de construction et théorèmes d’existence en géométrie

algébrique. III. Préschemas quotients, Séminaire Bourbaki, Vol. 6, Soc.
Math. France, Paris, 1960–61, pp. Exp. No. 212, 99–118.

[FGAIV] , Techniques de construction et théorèmes d’existence en géométrie

algébrique. IV. Les schémas de Hilbert, Séminaire Bourbaki, Vol. 6, Soc.
Math. France, Paris, 1960–61, pp. Exp. No. 221, 249–276.

592
[FGAV] , Technique de descente et théorèmes d’existence en géométrie
algébrique. V. Les schémas de Picard: théorèmes d’existence, Séminaire
Bourbaki, Vol. 7, Soc. Math. France, Paris, 1961–62, pp. Exp. No. 232,
143–161.

[FGAVI] , Techniques de construction et théorèmes d’existence en géométrie

algébrique. VI. Les schémas de Picard: propriétés générales, Séminaire
Bourbaki, Vol. 7, Soc. Math. France, Paris, 1961–62, pp. Exp. No. 236,
221–243.

[FGI+ 05] Barbara Fantechi, Lothar Göttsche, Luc Illusie, Steven L. Kleiman, Nitin
Nitsure, and Angelo Vistoli, Fundamental algebraic geometry, Mathemati-
cal Surveys and Monographs, vol. 123, American Mathematical Society,
Providence, RI, 2005, Grothendieck’s FGA explained.

[FK12] Robert Fricke and Felix Klein, Vorlesungen über die Theorie der auto-
morphen Funktionen. Band II: Die funktionentheoretischen Ausführungen
und die Andwendungen, B. G. Teubner, Leipzig, 1912.

[FK92] , Vorlesungen über die Theorie der automorphen Funktionen. Band

1: Die gruppentheoretischen Grundlagen., B. G. Teubner, Leipzig, 1892.

[FM12] Benson Farb and Dan Margalit, A primer on mapping class groups, Prince-
ton Mathematical Series, vol. 49, Princeton University Press, Princeton,
NJ, 2012.

[FM13] Gavril Farkas and Ian Morrison (eds.), Handbook of moduli. Vol. I,II,III,
Advanced Lectures in Mathematics (ALM), vol. 24, 25, 26, International
Press, Somerville, MA; Higher Education Press, Beijing, 2013.

[FMN10] Barbara Fantechi, Etienne Mann, and Fabio Nironi, Smooth toric Deligne-
Mumford stacks, J. Reine Angew. Math. 648 (2010), 201–244.

[FO10] William Fulton and Martin Olsson, The Picard group of M1,1 , Algebra
Number Theory 4 (2010), no. 1, 87–104.

[Fog68] John Fogarty, Algebraic families on an algebraic surface, Amer. J. Math.

90 (1968), 511–521.

[FS82] Walter Ricardo Ferrer Santos, A note on reductive groups, An. Acad.
Brasil. Ciênc. 54 (1982), no. 3, 469–471.

[Fuc66] L. Fuchs, Zur Theorie der linearen Differentialgleichungen mit veränder-

lichen Coefficienten, J. Reine Angew. Math. 66 (1866), 121–160.

[Ful69] William Fulton, Hurwitz schemes and irreducibility of moduli of algebraic

curves, Ann. of Math. (2) 90 (1969), 542–575.

[Ful93] , Introduction to toric varieties, Annals of Mathematics Studies,

vol. 131, Princeton University Press, Princeton, NJ, 1993, The William
H. Roever Lectures in Geometry.

[Ful98] , Intersection theory, second ed., Ergebnisse der Mathematik und

ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics
[Results in Mathematics and Related Areas. 3rd Series. A Series of Modern
Surveys in Mathematics], vol. 2, Springer-Verlag, Berlin, 1998.

593
[Ful10] Damiano Fulghesu, The stack of rational curves, Comm. Algebra 38
(2010), no. 7, 2405–2417.
[Gab62] Pierre Gabriel, Des catégories abéliennes, Bull. Soc. Math. France 90
(1962), 323–448.
[GH78] Phillip Griffiths and Joseph Harris, Principles of algebraic geometry, Wiley-
Interscience [John Wiley & Sons], New York, 1978, Pure and Applied
Mathematics.
[Gie82] David Gieseker, Lectures on moduli of curves, Tata Institute of Fundamen-
tal Research Lectures on Mathematics and Physics, vol. 69, Published for
the Tata Institute of Fundamental Research, Bombay by Springer-Verlag,
Berlin-New York, 1982.
[Gir64] Jean Giraud, Méthode de la descente, Bull. Soc. Math. France Mém. 2
(1964), viii+150.
[Gir71] , Cohomologie non abélienne, Springer-Verlag, Berlin-New York,
1971, Die Grundlehren der mathematischen Wissenschaften, Band 179.
[GIT] David Mumford, Geometric invariant theory, Ergebnisse der Mathematik
und ihrer Grenzgebiete, Neue Folge, Band 34, Springer-Verlag, Berlin-New
York, 1965.
[GLP83] Laurent Gruson, Robert Lazarsfeld, and Christian Peskine, On a theorem
of Castelnuovo, and the equations defining space curves, Invent. Math. 72
(1983), no. 3, 491–506.
[God58] Roger Godement, Topologie algébrique et théorie des faisceaux, Actualités
Scientifiques et Industrielles [Current Scientific and Industrial Topics], No.
1252, Hermann, Paris, 1958, Publ. Math. Univ. Strasbourg. No. 13.
[Gor68] Paul Gordan, Beweis, dass jede Covariante und Invariante einer binären
Form eine ganze Function mit numerischen Coefficienten einer endlichen
Anzahl solcher Formen ist, J. Reine Angew. Math. 69 (1868), 323–354.
[Got78] Gerd Gotzmann, Eine Bedingung für die Flachheit und das Hilbertpolynom
eines graduierten Ringes, Math. Z. 158 (1978), no. 1, 61–70.
[GR71] Laurent Gruson and Michel Raynaud, Critères de platitude et de projec-
tivité, Invent. Math. 13 (1971), 1–89.
[Gre82] Mark L. Green, The canonical ring of a variety of general type, Duke
Math. J. 49 (1982), no. 4, 1087–1113.
[Gre84] , Koszul cohomology and the geometry of projective varieties, J.
Differential Geom. 19 (1984), no. 1, 125–171.
[Gre89] , Restrictions of linear series to hyperplanes, and some results
of Macaulay and Gotzmann, Algebraic curves and projective geometry
(Trento, 1988), Lecture Notes in Math., vol. 1389, Springer, Berlin, 1989,
pp. 76–86.
[Gre98] , Generic initial ideals, Six lectures on commutative algebra (Bel-
laterra, 1996), Progr. Math., vol. 166, Birkhäuser, Basel, 1998, pp. 119–
186.

594
[Gro57a] Alexander Grothendieck, Sur la classification des fibrés holomorphes sur
la sphère de Riemann, Amer. J. Math. 79 (1957), 121–138.

[Gro57b] , Sur quelques points d’algèbre homologique, Tohoku Math. J. (2)

9 (1957), 119–221.

[Gro61] , Techniques de construction en géométrie analytique. I-X, Sémi-

naire Henri Cartan (1960-1961).

[Gro68] , Le groupe de Brauer I, II, III, Dix exposés sur la cohomologie

des schémas, Adv. Stud. Pure Math., vol. 3, North-Holland, Amsterdam,
1968, pp. 46–188.

[Gro86] Alexandre Grothendieck, Récoltes et semailles: Réflexions et témoignages

sur un passé de mathématicien, unpublished, 1986.

[Gro17] Philipp Gross, Tensor generators on schemes and stacks, Algebr. Geom.
4 (2017), no. 4, 501–522.

[GS15] Anton Geraschenko and Matthew Satriano, Toric stacks I: The theory of
stacky fans, Trans. Amer. Math. Soc. 367 (2015), no. 2, 1033–1071.

[GS17] , A “bottom up” characterization of smooth Deligne-Mumford

stacks, Int. Math. Res. Not. IMRN (2017), no. 21, 6469–6483.

[Göt90] Lothar Göttsche, The Betti numbers of the Hilbert scheme of points on a
smooth projective surface, Math. Ann. 286 (1990), no. 1-3, 193–207.

[Göt94] , Hilbert schemes of zero-dimensional subschemes of smooth vari-

eties, Lecture Notes in Mathematics, vol. 1572, Springer-Verlag, Berlin,
1994.

[Hab75] William J. Haboush, Reductive groups are geometrically reductive, Ann.

of Math. (2) 102 (1975), no. 1, 67–83.

[Hai98] Mark Haiman, t, q-Catalan numbers and the Hilbert scheme, vol. 193, 1998,
Selected papers in honor of Adriano Garsia (Taormina, 1994), pp. 201–224.

[Hal13] Jack Hall, Moduli of singular curves, appendix to Towards a classification

of modular compactifications of Mg,n by D. Smyth, Invent. Math. 192
(2013), no. 2, 459–503.

[Hal14a] , Cohomology and base change for algebraic stacks, Math. Z. 278
(2014), no. 1-2, 401–429.

[Hal14b] , Cohomology and base change for algebraic stacks, Math. Z. 278
(2014), no. 1-2, 401–429.

[Hal17] , Openness of versality via coherent functors, J. Reine Angew.

Math. 722 (2017), 137–182.

[Har66a] Robin Hartshorne, Ample vector bundles, Inst. Hautes Études Sci. Publ.
Math. (1966), no. 29, 63–94.

[Har66b] , Connectedness of the Hilbert scheme, Inst. Hautes Études Sci.

Publ. Math. (1966), no. 29, 5–48.

595
[Har66c] , Residues and duality, Lecture notes of a seminar on the work of A.
Grothendieck, given at Harvard 1963/64. With an appendix by P. Deligne.
Lecture Notes in Mathematics, No. 20, Springer-Verlag, Berlin-New York,
1966.

[Har70] , Ample subvarieties of algebraic varieties, Lecture Notes in Math-

ematics, vol. Vol. 156, Springer-Verlag, Berlin-New York, 1970, Notes
written in collaboration with C. Musili.

[Har77] , Algebraic geometry, Springer-Verlag, New York-Heidelberg, 1977,

Graduate Texts in Mathematics, No. 52.

[Har10] , Deformation theory, Graduate Texts in Mathematics, vol. 257,

Springer, New York, 2010.

[Has03] Brendan Hassett, Moduli spaces of weighted pointed stable curves, Adv.
Math. 173 (2003), no. 2, 316–352.

[Hei17] Jochen Heinloth, Hilbert-Mumford stability on algebraic stacks and appli-

cations to G-bundles on curves, Épijournal Géom. Algébrique 1 (2017),
Art. 11, 37.

[Hes79] Wim H. Hesselink, Desingularizations of varieties of nullforms, Invent.

Math. 55 (1979), no. 2, 141–163.

[Hes81] , Concentration under actions of algebraic groups, Paul Dubreil

and Marie-Paule Malliavin Algebra Seminar, 33rd Year (Paris, 1980),
Lecture Notes in Math., vol. 867, Springer, Berlin, 1981, pp. 55–89.

[HH09] Brendan Hassett and Donghoon Hyeon, Log canonical models for the
moduli space of curves: the first divisorial contraction, Trans. Amer.
Math. Soc. 361 (2009), no. 8, 4471–4489.

[HH13] , Log minimal model program for the moduli space of stable curves:
the first flip, Ann. of Math. (2) 177 (2013), no. 3, 911–968.

[Hil93] David Hilbert, Theory of algebraic invariants, Cambridge University Press,

Cambridge, 1993, Translated from the German and with a preface by
Reinhard C. Laubenbacher, Edited and with an introduction by Bernd
Sturmfels.

[Hil90] , Ueber die Theorie der algebraischen Formen, Math. Ann. 36

(1890), no. 4, 473–534.

[Hil93] , Ueber die vollen Invariantensysteme, Math. Ann. 42 (1893),

no. 3, 313–373.

[HK00] Yi Hu and Sean Keel, Mori dream spaces and GIT, vol. 48, 2000, Dedicated
to William Fulton on the occasion of his 60th birthday, pp. 331–348.

[HL10] Daniel Huybrechts and Manfred Lehn, The geometry of moduli spaces
of sheaves, second ed., Cambridge Mathematical Library, Cambridge
University Press, Cambridge, 2010.

[HL14] Daniel Halpern-Leistner, On the structure of instability in moduli theory,

2014.

596
[HM82] Joe Harris and David Mumford, On the Kodaira dimension of the moduli
space of curves, Invent. Math. 67 (1982), no. 1, 23–88, With an appendix
by William Fulton.

[HM98] Joe Harris and Ian Morrison, Moduli of curves, Graduate Texts in Mathe-
matics, vol. 187, Springer-Verlag, New York, 1998.

[HN75] Günter Harder and Mudumbai S. Narasimhan, On the cohomology groups

of moduli spaces of vector bundles on curves, Math. Ann. 212 (1974/75),
215–248.

[Hoc65] Gerhard Hochschild, The structure of Lie groups, Holden-Day, Inc., San
Francisco-London-Amsterdam, 1965.

[HP47] William V. D. Hodge and Daniel Pedoe, Methods of Algebraic Geometry.

Vol. I-III, Cambridge, at the University Press, 1947.

[HR74] Melvin Hochster and Joel L. Roberts, Rings of invariants of reductive

groups acting on regular rings are Cohen-Macaulay, Advances in Math.
13 (1974), 115–175.

[HR15] Jack Hall and David Rydh, Algebraic groups and compact generation of
their derived categories of representations, Indiana Univ. Math. J. 64
(2015), no. 6, 1903–1923.

[HR19a] , Artin’s criteria for algebraicity revisited, Algebra Number Theory

13 (2019), no. 4, 749–796.

[HR19b] , Coherent Tannaka duality and algebraicity of Hom-stacks, Alge-

bra Number Theory 13 (2019), no. 7, 1633–1675.

[Hub06] John Hamal Hubbard, Teichmüller theory and applications to geometry,

topology, and dynamics. Vol. 1, Matrix Editions, Ithaca, NY, 2006, Teich-
müller theory, With contributions by Adrien Douady, William Dunbar,
Roland Roeder, Sylvain Bonnot, David Brown, Allen Hatcher, Chris
Hruska and Sudeb Mitra, With forewords by William Thurston and
Clifford Earle.

[Hum75] James E. Humphreys, Linear algebraic groups, Graduate Texts in Mathe-

matics, No. 21, Springer-Verlag, New York-Heidelberg, 1975.

[Hum78] , Introduction to Lie algebras and representation theory, Graduate

Texts in Mathematics, vol. 9, Springer-Verlag, New York-Berlin, 1978,
Second printing, revised.

[Hur92] Adolf Hurwitz, Ueber algebraische Gebilde mit eindeutigen Transforma-

tionen in sich, Math. Ann. 41 (1892), no. 3, 403–442.

[Hur91] , Ueber Riemann’sche Flächen mit gegebenen Verzweigungspunkten,

Math. Ann. 39 (1891), no. 1, 1–60.

[Igu55] Jun-ichi Igusa, On some problems in abstract algebraic geometry, Proc.

Nat. Acad. Sci. U.S.A. 41 (1955), 964–967.

[Igu60] , Arithmetic variety of moduli for genus two, Ann. of Math. (2)
72 (1960), 612–649.

597
[Ill71] Luc Illusie, Complexe cotangent et déformations. I, Lecture Notes in
Mathematics, Vol. 239, Springer-Verlag, Berlin, 1971.

[Ill72] , Complexe cotangent et déformations. II, Lecture Notes in Math-

ematics, vol. Vol. 283, Springer-Verlag, Berlin-New York, 1972.

[Ill05] , Grothendieck’s existence theorem in formal geometry, Fundamen-

tal algebraic geometry, Math. Surveys Monogr., vol. 123, Amer. Math.
Soc., Providence, RI, 2005, With a letter (in French) of Jean-Pierre Serre,
pp. 179–233.

[IM65] Nagayoshi Iwahori and Hideya Matsumoto, On some Bruhat decomposition

and the structure of the Hecke rings of p-adic Chevalley groups, Inst.
Hautes Études Sci. Publ. Math. (1965), no. 25, 5–48.

[Ive72] Birger Iversen, A fixed point formula for action of tori on algebraic
varieties, Invent. Math. 16 (1972), 229–236.

[JHS11] A. Johan de Jong, Xuhua He, and Jason Michael Starr, Families of ratio-
nally simply connected varieties over surfaces and torsors for semisimple
groups, Publ. Math. Inst. Hautes Études Sci. (2011), no. 114, 1–85.

[JP13] Lizhen Ji and Athanase Papadopoulos, Historical development of Teich-

müller theory, Arch. Hist. Exact Sci. 67 (2013), no. 2, 119–147.

[JS21] Joachim Jelisiejew and Łukasz Sienkiewicz, Białynicki-Birula decomposi-

tion for reductive groups in positive characteristic, J. Math. Pures Appl.
(9) 152 (2021), 189–210.

[Kee99] Seán Keel, Basepoint freeness for nef and big line bundles in positive
characteristic, Ann. of Math. (2) 149 (1999), no. 1, 253–286.

[Kem78] George R. Kempf, Instability in invariant theory, Ann. of Math. (2) 108
(1978), no. 2, 299–316.

[Kin94] Alastair D. King, Moduli of representations of finite-dimensional algebras,

Quart. J. Math. Oxford Ser. (2) 45 (1994), no. 180, 515–530.

[Kir84] Frances Clare Kirwan, Cohomology of quotients in symplectic and alge-

braic geometry, Mathematical Notes, vol. 31, Princeton University Press,
Princeton, NJ, 1984.

[Kir85] , Partial desingularisations of quotients of nonsingular varieties

and their Betti numbers, Ann. of Math. (2) 122 (1985), no. 1, 41–85.

[KKLV89] Friedrich Knop, Hanspeter Kraft, Domingo Luna, and Thierry Vust,
Local properties of algebraic group actions, Algebraische Transformations-
gruppen und Invariantentheorie, DMV Sem., vol. 13, Birkhäuser, Basel,
1989, pp. 63–75.

[KKMS73] George Kempf, Finn Faye Knudsen, David Mumford, and B. Saint-Donat,
Toroidal embeddings. I., Springer-Verlag, Berlin-New York„ , 1973.

[KKV89] Friedrich Knop, Hanspeter Kraft, and Thierry Vust, The Picard group of a
G-variety, Algebraische Transformationsgruppen und Invariantentheorie,
DMV Sem., vol. 13, Birkhäuser, Basel, 1989, pp. 77–87.

598
[Kle82] Felix Klein, Über Riemann’s theorie der algebraischen functionen, Leipzig:
Teubner, 1882.

[Kle66] Steven L. Kleiman, Toward a numerical theory of ampleness, Ann. of

Math. (2) 84 (1966), 293–344.

[Kle05] , The Picard scheme, Fundamental algebraic geometry, Math.

Surveys Monogr., vol. 123, Amer. Math. Soc., Providence, RI, 2005,
pp. 235–321.

[KM76] Finn Faye Knudsen and David Mumford, The projectivity of the moduli
space of stable curves. I. Preliminaries on “det” and “Div”, Math. Scand.
39 (1976), no. 1, 19–55.

[KM97] Seán Keel and Shigefumi Mori, Quotients by groupoids, Ann. of Math. (2)
145 (1997), no. 1, 193–213.

[KM98] János Kollár and Shigefumi Mori, Birational geometry of algebraic vari-
eties, Cambridge Tracts in Mathematics, vol. 134, Cambridge University
Press, Cambridge, 1998, With the collaboration of C. H. Clemens and A.
Corti, Translated from the 1998 Japanese original.

[Knu71] Donald Knutson, Algebraic spaces, Lecture Notes in Mathematics, Vol.

203, Springer-Verlag, Berlin-New York, 1971.

[Knu83a] Finn F. Knudsen, The projectivity of the moduli space of stable curves. II.
The stacks Mg,n , Math. Scand. 52 (1983), no. 2, 161–199.

[Knu83b] , The projectivity of the moduli space of stable curves. III. The line
bundles on Mg,n , and a proof of the projectivity of M g,n in characteristic
0, Math. Scand. 52 (1983), no. 2, 200–212.

[Kod86] Kunihiko Kodaira, Complex manifolds and deformation of complex struc-

tures, Grundlehren der mathematischen Wissenschaften [Fundamental
Principles of Mathematical Sciences], vol. 283, Springer-Verlag, New York,
1986, Translated from the Japanese by Kazuo Akao, With an appendix
by Daisuke Fujiwara.

[Kol90] János Kollár, Projectivity of complete moduli, J. Differential Geom. 32

(1990), no. 1, 235–268.

[Kol92] , Cone theorems and bug-eyed covers, J. Algebraic Geom. 1 (1992),

no. 2, 293–323.

[Kol96] , Rational curves on algebraic varieties, Ergebnisse der Mathe-

matik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in
Mathematics [Results in Mathematics and Related Areas. 3rd Series. A
Series of Modern Surveys in Mathematics], vol. 32, Springer-Verlag, Berlin,
1996.

[Kol97] , Quotient spaces modulo algebraic groups, Ann. of Math. (2) 145
(1997), no. 1, 33–79.

[Kol06] , Review of ‘Selected papers on the classification of varieties and

moduli spaces’, by David Mumford, Bull. Amer. Math. Soc. (N.S.) 43
(2006), no. 1, 111–114.

599
[Kol07] , Lectures on resolution of singularities, Annals of Mathematics
Studies, vol. 166, Princeton University Press, Princeton, NJ, 2007.

[Kol12] , Quotients by finite equivalence relations, Current developments

in algebraic geometry, Math. Sci. Res. Inst. Publ., vol. 59, Cambridge
Univ. Press, Cambridge, 2012, With an appendix by Claudiu Raicu,
pp. 227–256.

[Kol13] , Singularities of the minimal model program, Cambridge Tracts

in Mathematics, vol. 200, Cambridge University Press, Cambridge, 2013,
With a collaboration of Sándor Kovács.

[Kol21] , Mumford’s influence on the moduli theory of algebraic varieties,

Pure Appl. Math. Q. 17 (2021), no. 2, 619–647.

[Kon82] Jerzy Konarski, A pathological example of an action of k ∗ , Group actions

and vector fields (Vancouver, B.C., 1981), Lecture Notes in Math., vol.
956, Springer, Berlin, 1982, pp. 72–78.

[KP17] Sándor J. Kovács and Zsolt Patakfalvi, Projectivity of the moduli space of
stable log-varieties and subadditivity of log-Kodaira dimension, J. Amer.
Math. Soc. 30 (2017), no. 4, 959–1021.

[Kra84] Hanspeter Kraft, Geometrische Methoden in der Invariantentheorie, As-

pects of Mathematics, D1, Friedr. Vieweg & Sohn, Braunschweig, 1984.

[Kre99] Andrew Kresch, Cycle groups for Artin stacks, Invent. Math. 138 (1999),
no. 3, 495–536.

[Kre09] , On the geometry of Deligne-Mumford stacks, Algebraic geometry—

Seattle 2005. Part 1, Proc. Sympos. Pure Math., vol. 80, Part 1, Amer.
Math. Soc., Providence, RI, 2009, pp. 259–271.

[Kre13] , Flattening stratification and the stack of partial stabilizations of

prestable curves, Bull. Lond. Math. Soc. 45 (2013), no. 1, 93–102.

[Lam99] Tsit Y. Lam, Lectures on modules and rings, Graduate Texts in Mathe-
matics, vol. 189, Springer-Verlag, New York, 1999.

[Lan75] Stacy G. Langton, Valuative criteria for families of vector bundles on

algebraic varieties, Ann. of Math. (2) 101 (1975), 88–110.

[Lan83] Herbert Lange, Universal families of extensions, J. Algebra 83 (1983),

no. 1, 101–112.

[Lan13] Adrian Langer, On positivity and semistability of vector bundles in finite

and mixed characteristics, J. Ramanujan Math. Soc. 28A (2013), 287–309.

[Laz69] Daniel Lazard, Autour de la platitude, Bull. Soc. Math. France 97 (1969),
81–128.

[Laz04a] Robert Lazarsfeld, Positivity in algebraic geometry. I, Ergebnisse der

Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys
in Mathematics [Results in Mathematics and Related Areas. 3rd Series.
A Series of Modern Surveys in Mathematics], vol. 48, Springer-Verlag,
Berlin, 2004, Classical setting: line bundles and linear series.

600
[Laz04b] , Positivity in algebraic geometry. II, Ergebnisse der Mathematik
und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathe-
matics [Results in Mathematics and Related Areas. 3rd Series. A Series of
Modern Surveys in Mathematics], vol. 49, Springer-Verlag, Berlin, 2004,
Positivity for vector bundles, and multiplier ideals.

[Les14] John Lesieutre, The diminished base locus is not always closed, Compos.
Math. 150 (2014), no. 10, 1729–1741.

[Lie06] Max Lieblich, Remarks on the stack of coherent algebras, Int. Math. Res.
Not. (2006).

[Lie18] , Moduli of sheaves: a modern primer, Algebraic geometry: Salt

Lake City 2015, Proc. Sympos. Pure Math., vol. 97.2, Amer. Math. Soc.,
Providence, RI, 2018, pp. 361–388.

[Lip78] Joseph Lipman, Desingularization of two-dimensional schemes, Ann. of

Math. (2) 107 (1978), no. 1, 151–207.

[Liu02] Qing Liu, Algebraic geometry and arithmetic curves, Oxford Graduate
Texts in Mathematics, vol. 6, Oxford University Press, Oxford, 2002,
Translated from the French by Reinie Erné, Oxford Science Publications.

[LMB00] Gérard Laumon and Laurent Moret-Bailly, Champs algébriques, Ergeb-

nisse der Mathematik und ihrer Grenzgebiete. 3. Folge., vol. 39, Springer-
Verlag, Berlin, 2000.

[LP97] Joseph Le Potier, Lectures on vector bundles, Cambridge Studies in

Advanced Mathematics, vol. 54, Cambridge University Press, Cambridge,
1997, Translated by A. Maciocia.

[LS67] Stephen Lichtenbaum and Michael Schlessinger, The cotangent complex

of a morphism, Trans. Amer. Math. Soc. 128 (1967), 41–70.

[Lun73] Domingo Luna, Slices étales, Sur les groupes algébriques, Mém. Soc. Math.
France 33, 1973, pp. 81–105.

[Lur04] Jacob Lurie, Coherent tannaka duality for geometric stacks, 2004.

[LW15] Jun Li and Xiaowei Wang, Hilbert-Mumford criterion for nodal curves,
Compos. Math. 151 (2015), no. 11, 2076–2130.

[Lür71] Jacob Lüroth, Note über Verzweigungsschnitte und Querschnitte in einer

Riemann’schen Fläche, Math. Ann. 4 (1871), no. 2, 181–184.

[Mac71] Saunders MacLane, Categories for the working mathematician, Graduate

Texts in Mathematics, Vol. 5, Springer-Verlag, New York-Berlin, 1971.

[Mac07] Diane Maclagan, Notes on hilbert schemes, 2007.

[Mar78] Masaki Maruyama, Moduli of stable sheaves. II, J. Math. Kyoto Univ. 18
(1978), no. 3, 557–614.

[Mat60] Yozô Matsushima, Espaces homogènes de Stein des groupes de Lie com-
plexes, Nagoya Math. J. 16 (1960), 205–218.

601
[Mat64] Teruhisa Matsusaka, Theory of Q-varieties, Mathematical Society of
Japan, Tokyo, 1964.

[Mat89] Hideyuki Matsumura, Commutative ring theory, second ed., Cambridge

Studies in Advanced Mathematics, vol. 8, Cambridge University Press,
Cambridge, 1989, Translated from the Japanese by M. Reid.

[May69] Alan Mayer, Compactification of the variety of moduli of curves, Semi-

nar on degeneration of algebraic varieties (conducted by P.A. Griffiths),
Institute for Advanced Study, Princeton, NJ, 1969.

[Maz75] Joseph Mazur, Conditions for the existence of contractions in the category
of algebraic spaces, Trans. Amer. Math. Soc. 209 (1975), 259–265.

[Maz08] Barry Mazur, When is one thing equal to some other thing?, Proof and
other dilemmas, MAA Spectrum, Math. Assoc. America, Washington,
DC, 2008, pp. 221–241.

[Mes00] Bachuki Mesablishvili, Pure morphisms of commutative rings are effective

descent morphisms for modules—a new proof, Theory Appl. Categ. 7
(2000), No. 3, 38–42.

[MF82] David Mumford and John Fogarty, Geometric invariant theory, second
ed., Ergebnisse der Mathematik und ihrer Grenzgebiete [Results in Math-
ematics and Related Areas], vol. 34, Springer-Verlag, Berlin, 1982.

[MFK94] David Mumford, John Fogarty, and Frances Clare Kirwan, Geometric
invariant theory, third ed., Ergebnisse der Mathematik und ihrer Gren-
zgebiete (2) [Results in Mathematics and Related Areas (2)], vol. 34,
Springer-Verlag, Berlin, 1994.

[Mil80] James S. Milne, Étale cohomology, Princeton Mathematical Series, vol. 33,
Princeton University Press, Princeton, N.J., 1980.

[Mil17] , Algebraic groups, Cambridge Studies in Advanced Mathematics,

vol. 170, Cambridge University Press, Cambridge, 2017, The theory of
group schemes of finite type over a field.

[Mir95] Rick Miranda, Algebraic curves and Riemann surfaces, Graduate Studies
in Mathematics, vol. 5, American Mathematical Society, Providence, RI,
1995.

[MM64a] Alan Mayer and David Mumford, The boundary of moduli schemes, Lec-
ture notes prepared in connection with the Summer Institute on Algebraic
Geometry held at Woods Hole, MA, American Mathematical Society,
1964 (unpublished).

[MM64b] , Further comments on boundary points, Lecture notes prepared

in connection with the Summer Institute on Algebraic Geometry held at
Woods Hole, MA, American Mathematical Society, 1964 (unpublished).

[MO10] Brian Conrad ([Link] fpqc cov-

ers of stacks, MathOverflow, arXiv:[Link]
URL:[Link] (version: 2010-02-14).

602
[Mor80] Ian Morrison, Projective stability of ruled surfaces, Invent. Math. 56
(1980), no. 3, 269–304.
[MS72] David Mumford and Kalevi Suominen, Introduction to the theory of moduli,
Algebraic geometry, Oslo 1970 (Proc. Fifth Nordic Summer School in
Math.), Wolters-Noordhoff Publishing, Groningen, 1972, pp. 171–222.
[Muk03] Shigeru Mukai, An introduction to invariants and moduli, Cambridge
Studies in Advanced Mathematics, vol. 81, Cambridge University Press,
Cambridge, 2003, Translated from the 1998 and 2000 Japanese editions
by W. M. Oxbury.
[Mum61] David Mumford, An elementary theorem in geometric invariant theory,
Bull. Amer. Math. Soc. 67 (1961), 483–487.
[Mum62] , Further pathologies in algebraic geometry, Amer. J. Math. 84
(1962), 642–648.
[Mum63] , Projective invariants of projective structures and applications,
Proc. Internat. Congr. Mathematicians (Stockholm, 1962), Inst. Mittag-
Leffler, Djursholm, 1963, pp. 526–530.
[Mum65] , Picard groups of moduli problems, Arithmetical Algebraic Geom-
etry (Proc. Conf. Purdue Univ., 1963), Harper & Row, New York, 1965,
pp. 33–81.
[Mum66a] , Lectures on curves on an algebraic surface, Annals of Mathe-
matics Studies, No. 59, Princeton University Press, Princeton, N.J., 1966,
With a section by G. M. Bergman.
[Mum66b] , On the equations defining abelian varieties. I, Invent. Math. 1
(1966), 287–354.
[Mum67] , Pathologies. III, Amer. J. Math. 89 (1967), 94–104.
[Mum69] , Bi-extensions of formal groups, Algebraic Geometry (Internat.
Colloq., Tata Inst. Fund. Res., Bombay, 1968), Oxford Univ. Press,
London, 1969, pp. 307–322.
[Mum70a] , Abelian varieties, Tata Institute of Fundamental Research Stud-
ies in Mathematics, No. 5, Published for the Tata Institute of Fundamental
Research, Bombay; Oxford University Press, London, 1970.
[Mum70b] , Varieties defined by quadratic equations, Questions on Algebraic
Varieties (C.I.M.E., III Ciclo, Varenna, 1969), Edizioni Cremonese, Rome,
1970, pp. 29–100.
[Mum75a] , Curves and their Jacobians, University of Michigan Press, Ann
Arbor, Mich., 1975.
[Mum75b] , Pathologies IV, Amer. J. Math. 97 (1975), no. 3, 847–849.
[Mum76] , Hilbert’s fourteenth problem–the finite generation of subrings such
as rings of invariants, Mathematical developments arising from Hilbert
problems (Proc. Sympos. Pure Math., Northern Illinois Univ., De Kalb,
Ill., 1974), Proc. Sympos. Pure Math., vol. Vol. XXVIII, Amer. Math.
Soc., Providence, RI, 1976, pp. 431–444.

603
[Mum77] , Stability of projective varieties, Enseign. Math. (2) 23 (1977),
no. 1-2, 39–110.

[Mum83] , Towards an enumerative geometry of the moduli space of curves,

Arithmetic and geometry, Vol. II, Progr. Math., vol. 36, Birkhäuser
Boston, Boston, MA, 1983, pp. 271–328.

[Mum04] , Selected papers on the classification of varieties and moduli spaces,

Springer-Verlag, New York, 2004, With commentaries by David Gieseker,
George Kempf, Herbert Lange and Eckart Viehweg.

[Mum10] , Selected papers, Volume II, Springer, New York, 2010, On alge-
braic geometry, including correspondence with Grothendieck, Edited by
Ching-Li Chai, Amnon Neeman and Takahiro Shiota.

[Mur64] Jaap P. Murre, On contravariant functors from the category of pre-schemes

over a field into the category of abelian groups (with an application to
the Picard functor), Inst. Hautes Études Sci. Publ. Math. (1964), no. 23,
5–43.

[Mur95] , Representation of unramified functors. Applications (according

to unpublished results of A. Grothendieck), Séminaire Bourbaki, Vol. 9,
Soc. Math. France, Paris, 1995, pp. Exp. No. 294, 243–261.

[Nag59] Masayoshi Nagata, On the purity of branch loci in regular local rings,
Illinois J. Math. 3 (1959), 328–333.

[Nag62] , Local rings, Interscience Tracts in Pure and Applied Mathematics,

No. 13, Interscience Publishers a division of John Wiley & Sons New
York-London, 1962.

[Nag62] , Complete reducibility of rational representations of a matrix

group, J. Math. Kyoto Univ. 1 (1961/62), 87–99.

[Nag64] , Invariants of a group in an affine ring, J. Math. Kyoto Univ. 3

(1963/64), 369–377.

[Nak97] Hiraku Nakajima, Heisenberg algebra and Hilbert schemes of points on

projective surfaces, Ann. of Math. (2) 145 (1997), no. 2, 379–388.

[Nak99a] , Lectures on Hilbert schemes of points on surfaces, University

Lecture Series, vol. 18, American Mathematical Society, Providence, RI,
1999.

[Nak99b] , Lectures on Hilbert schemes of points on surfaces, University

Lecture Series, vol. 18, American Mathematical Society, Providence, RI,
1999.

[Nér64] André Néron, Modèles minimaux des variétés abéliennes sur les corps
locaux et globaux, Inst. Hautes Études Sci. [Link]. No. 21 (1964), 128.

[Nes84] Linda Ness, A stratification of the null cone via the moment map, Amer.
J. Math. 106 (1984), no. 6, 1281–1329, With an appendix by David
Mumford.

604
[New78] Peter E. Newstead, Introduction to moduli problems and orbit spaces,
Tata Institute of Fundamental Research Lectures on Mathematics and
Physics, vol. 51, Tata Institute of Fundamental Research, Bombay; Narosa
Publishing House, New Delhi, 1978.

[Nit05] Nitin Nitsure, Construction of Hilbert and Quot schemes, Fundamental

algebraic geometry, Math. Surveys Monogr., vol. 123, Amer. Math. Soc.,
Providence, RI, 2005, pp. 105–137.

[Nit09] , Deformation theory for vector bundles, Moduli spaces and vector
bundles, London Math. Soc. Lecture Note Ser., vol. 359, Cambridge Univ.
Press, Cambridge, 2009, pp. 128–164.

[Nit11] , Schematic Harder-Narasimhan stratification, Internat. J. Math.

22 (2011), no. 10, 1365–1373.

[NM64] Masayoshi Nagata and Takehiko Miyata, Note on semi-reductive groups,

J. Math. Kyoto Univ. 3 (1963/64), 379–382.

[Noe80] Max Noether, Ueber die invariante Darstellung algebraischer Functionen,

Math. Ann. 17 (1880), no. 2, 263–284.

[NR69] M. S. Narasimhan and S. Ramanan, Moduli of vector bundles on a compact

Riemann surface, Ann. of Math. (2) 89 (1969), 14–51.

[NS65] M. S. Narasimhan and C. S. Seshadri, Stable and unitary vector bundles

on a compact Riemann surface, Ann. of Math. (2) 82 (1965), 540–567.

[Ols03] Martin Olsson, Logarithmic geometry and algebraic stacks, Ann. Sci. École
Norm. Sup. (4) 36 (2003), no. 5, 747–791.

[Ols05] Martin C. Olsson, On proper coverings of Artin stacks, Adv. Math. 198
(2005), no. 1, 93–106.

[Ols06] Martin Olsson, Deformation theory of representable morphisms of alge-

braic stacks, Math. Z. 253 (2006), no. 1, 25–62.

[Ols07] , Sheaves on Artin stacks, J. Reine Angew. Math. 603 (2007),

55–112.

[Ols16] , Algebraic spaces and stacks, American Mathematical Society

Colloquium Publications, vol. 62, American Mathematical Society, Provi-
dence, RI, 2016.

[Oor66] Frans Oort, Algebraic group schemes in characteristic zero are reduced,
Invent. Math. 2 (1966), 79–80.

[Oor71] , Finite group schemes, local moduli for abelian varieties, and
lifting problems, Compositio Math. 23 (1971), 265–296.

[Oor81] , Coarse and fine moduli spaces of algebraic curves and polarized
abelian varieties, Symposia Mathematica, Vol. XXIV (Sympos., INDAM,
Rome, 1979), Academic Press, London-New York, 1981, pp. 293–313.

[Pet23] K. Petri, Über die invariante Darstellung algebraischer Funktionen einer

Veränderlichen, Math. Ann. 88 (1923), no. 3-4, 242–289.

605
[Poo17] Bjorn Poonen, Rational points on varieties, Graduate Studies in Mathe-
matics, vol. 186, American Mathematical Society, Providence, RI, 2017.

[Pop69] Herbert Popp, The singularities of the moduli schemes of curves, J.

Number Theory 1 (1969), 90–107.

[Pop74] , On moduli of algebraic varieties. II, Compositio Math. 28 (1974),

51–81.

[Pop85] Dorin Popescu, General Néron desingularization, Nagoya Math. J. 100

(1985), 97–126.

[Pop86] , General Néron desingularization and approximation, Nagoya

Math. J. 104 (1986), 85–115.

[Pop90] , Letter to the editor: “General Néron desingularization and ap-

proximation”, Nagoya Math. J. 118 (1990), 45–53.

[PS85] Ragni Piene and Michael Schlessinger, On the Hilbert scheme compactifi-
cation of the space of twisted cubics, Amer. J. Math. 107 (1985), no. 4,
761–774.

[Qui68] Daniel Quillen, Homology of commutative rings, Unpublished, 1968, pp. 1–

81.

[Qui70] , On the (co-) homology of commutative rings, Applications of

Categorical Algebra (Proc. Sympos. Pure Math., Vol. XVII, New York,
1968), Amer. Math. Soc., Providence, R.I., 1970, pp. 65–87.

[Ram74] C. P. Ramanujam, On a geometric interpretation of multiplicity, Invent.

Math. 22 (1973/74), 63–67.

[Ran89] Ziv Ran, Deformations of maps, Algebraic curves and projective geometry
(Trento, 1988), Lecture Notes in Math., vol. 1389, Springer, Berlin, 1989,
pp. 246–253.

[Rau62] H. Ernest Rauch, The singularities of the modulus space, Bull. Amer.
Math. Soc. 68 (1962), 390–394.

[Ray70] Michel Raynaud, Faisceaux amples sur les schémas en groupes et les
espaces homogènes, Lecture Notes in Mathematics, Vol. 119, Springer-
Verlag, Berlin-New York, 1970.

[Rei89] Zinovy Reichstein, Stability and equivariant maps, Invent. Math. 96

(1989), no. 2, 349–383.

[Ric77] Roger W. Richardson, Affine coset spaces of reductive algebraic groups,

Bull. London Math. Soc. 9 (1977), no. 1, 38–41.

[Rie57] Bernhard Riemann, Theorie der Abel’schen Functionen, J. Reine Angew.

Math. 54 (1857), 115–155.

[Rom05] Matthieu Romagny, Group actions on stacks and applications, Michigan

Math. J. 53 (2005), no. 1, 209–236.

[Ros56] Maxwell Rosenlicht, Some basic theorems on algebraic groups, Amer. J.

Math. 78 (1956), 401–443.

606
[Ryd11] David Rydh, Étale dévissage, descent and pushouts of stacks, J. Algebra
331 (2011), 194–223.

[Ryd13] , Existence and properties of geometric quotients, J. Algebraic

Geom. 22 (2013), no. 4, 629–669.

[Ryd15] , Noetherian approximation of algebraic spaces and stacks, J. Al-

gebra 422 (2015), 105–147.

[Ryd20] , A generalization of luna’s fundamental lemma for stacks with

good moduli spaces, 2020.

[Sai87] Takeshi Saito, Vanishing cycles and geometry of curves over a discrete
valuation ring, Amer. J. Math. 109 (1987), no. 6, 1043–1085.

[Sam55] Pierre Samuel, Méthodes d’algèbre abstraite en géométrie algébrique, Ergeb-

nisse der Mathematik und ihrer Grenzgebiete, (N.F.), vol. Heft 4, Springer-
Verlag, Berlin-Göttingen-Heidelberg, 1955.

[Sat56] Ichiro Satake, On the compactification of the Siegel space, J. Indian Math.
Soc. (N.S.) 20 (1956), 259–281.

[Sch68] Michael Schlessinger, Functors of Artin rings, Trans. Amer. Math. Soc.
130 (1968), 208–222.

[SD73] Bernard Saint-Donat, On Petri’s analysis of the linear system of quadrics

through a canonical curve, Math. Ann. 206 (1973), 157–175.

[Ser88] Jean-Pierre Serre, Algebraic groups and class fields, Graduate Texts in
Mathematics, vol. 117, Springer-Verlag, New York, 1988, Translated from
the French.

[Ser06] Edoardo Sernesi, Deformations of algebraic schemes, Grundlehren der

Mathematischen Wissenschaften [Fundamental Principles of Mathematical
Sciences], vol. 334, Springer-Verlag, Berlin, 2006.

[Ser56] Jean-Pierre Serre, Géométrie algébrique et géométrie analytique, Ann.

Inst. Fourier (Grenoble) 6 (1955/56), 1–42.

[Ses67] C. S. Seshadri, Space of unitary vector bundles on a compact Riemann

surface, Ann. of Math. (2) 85 (1967), 303–336.

[Ses72] , Quotient spaces modulo reductive algebraic groups, Ann. of Math.

(2) 95 (1972), 511–556; errata, ibid. (2) 96 (1972), 599.

[Ses77] , Geometric reductivity over arbitrary base, Advances in Math. 26

(1977), no. 3, 225–274.

[Sev15] Francesco Severi, Sul la classificazione del le curve algebriche e sul teorema
d’esistenza di riemann i–ii, Rend. della R. Acc dei Lincei 24 (1915), 877–
888,1011–1020.

[Sev21] , Vorlesungen über algebraische geometrie: Geometrie auf einer

kurve, riemannsche flächen, abelsche integrale, 1921, Berechtigte Deutsche
Übersetzung von Eugen Löffler. Mit einem Einführungswort von A. Brill.

607
[SGA1] Alexander Grothendieck and Michel Raynaud, Revêtements étales et
groupe fondamental, Séminaire de Géométrie Algébrique, I.H.E.S., 1963.

[SGA3I ] Schémas en groupes. I: Propriétés générales des schémas en groupes,

Séminaire de Géométrie Algébrique du Bois Marie 1962/64 (SGA 3). Dirigé
par M. Demazure et A. Grothendieck. Lecture Notes in Mathematics, Vol.
151, Springer-Verlag, Berlin, 1970.

[SGA3II ] Schémas en groupes. II: Groupes de type multiplicatif, et structure des

schémas en groupes généraux, Séminaire de Géométrie Algébrique du Bois
Marie 1962/64 (SGA 3). Dirigé par M. Demazure et A. Grothendieck.
Lecture Notes in Mathematics, Vol. 152, Springer-Verlag, Berlin, 1970.

[SGA3III ] Schémas en groupes. III: Structure des schémas en groupes réductifs,

Séminaire de Géométrie Algébrique du Bois Marie 1962/64 (SGA 3). Dirigé
par M. Demazure et A. Grothendieck. Lecture Notes in Mathematics, Vol.
153, Springer-Verlag, Berlin, 1970.

[SGA4] Théorie des topos et cohomologie étale des schémas., Lecture Notes in
Mathematics, Vol. 269, 270, 305, Springer-Verlag, Berlin, 1972, Séminaire
de Géométrie Algébrique du Bois-Marie 1963–1964 (SGA 4), Dirigé par
M. Artin, A. Grothendieck, et J. L. Verdier. Avec la collaboration de N.
Bourbaki, P. Deligne et B. Saint-Donat.

[SGA4 12 ] Pierre Deligne, Cohomologie étale, Lecture Notes in Mathematics, Vol.

569, Springer-Verlag, Berlin, 1977.

[SGA6] Théorie des intersections et théorème de Riemann-Roch, Lecture Notes in

Mathematics, Vol. 225, Springer-Verlag, Berlin-New York, 1971, Séminaire
de Géométrie Algébrique du Bois-Marie 1966–1967 (SGA 6), Dirigé par
P. Berthelot, A. Grothendieck et L. Illusie. Avec la collaboration de D.
Ferrand, J. P. Jouanolou, O. Jussila, S. Kleiman, M. Raynaud et J. P.
Serre.

[SGA7-I] Alexander Grothendieck, Michel Raynaud, and Dock Sang Rim, Groupes
de monodromie en géométrie algébrique. I, Lecture Notes in Mathematics,
Vol. 288, Springer-Verlag, 1972, Séminaire de Géométrie Algébrique du
Bois-Marie 1967–1969 (SGA 7 I).

[SGA7-II] Pierre Deligne and Nicholas Katz, Groupes de monodromie en géométrie

algébrique. II, Lecture Notes in Mathematics, Vol 340, Springer-Verlag,
1973, Séminaire de Géométrie Algébrique du Bois-Marie 1967–1969 (SGA
7, II).

[Sha77] Stephen S. Shatz, The decomposition and specialization of algebraic fami-

lies of vector bundles, Compositio Math. 35 (1977), no. 2, 163–187.

[Shi72] Goro Shimura, On the field of rationality for an abelian variety, Nagoya
Math. J. 45 (1972), 167–178.

[Sie35] Carl Ludwig Siegel, Über die analytische Theorie der quadratischen For-
men, Ann. of Math. (2) 36 (1935), no. 3, 527–606.

[Sil09] Joseph H. Silverman, The arithmetic of elliptic curves, second ed., Grad-
uate Texts in Mathematics, vol. 106, Springer, Dordrecht, 2009.

608
[Som82] Andrew J. Sommese, Some examples of C∗ actions, Group actions and
vector fields (Vancouver, B.C., 1981), Lecture Notes in Math., vol. 956,
Springer, Berlin, 1982, pp. 118–124.

[SP] The Stacks Project Authors, Stacks Project, [Link]

[Link], 2024.

[Spr77] Tonny A. Springer, Invariant theory, Lecture Notes in Mathematics, Vol.

585, Springer-Verlag, Berlin-New York, 1977.

[Spr98] , Linear algebraic groups, second ed., Progress in Mathematics,

vol. 9, Birkhäuser Boston, Inc., Boston, MA, 1998.

[SR72] Neantro Saavedra Rivano, Catégories tannakiennes, Bull. Soc. Math.

France 100 (1972), 417–430.

[SS11] Pramathanath Sastry and C. S. Seshadri, Geometric reductivity—a quo-

tient space approach, J. Ramanujan Math. Soc. 26 (2011), no. 4, 415–477.

[SS20] Roy Skjelnes and Gregory G. Smith, Smooth hilbert schemes: their classi-
fication and geometry, 2020, arXiv:2008.08938.

[Sta06] Jason Michael Starr, Artin’s axioms, composition and moduli spaces, 2006,
arXiv:math/0602646.

[Sto95] Michael Stoll, Vector bundles on curves, Lectures by Gerd Faltings, Bonn,
unpublished, 1995.

[Stu08] Bernd Sturmfels, Algorithms in invariant theory, second ed., Texts and
Monographs in Symbolic Computation, SpringerWienNewYork, Vienna,
2008.

[Sum74] Hideyasu Sumihiro, Equivariant completion, J. Math. Kyoto Univ. 14

(1974), 1–28.

[Sum75] , Equivariant completion. II, J. Math. Kyoto Univ. 15 (1975),

no. 3, 573–605.

[Swa98] Richard G. Swan, Néron-Popescu desingularization, Algebra and geometry

(Taipei, 1995), Lect. Algebra Geom., vol. 2, Int. Press, Cambridge, MA,
1998, pp. 135–192.

[Tei40] Oswald Teichmüller, Extremale quasikonforme Abbildungen und quadratis-

che Differentiale, Abh. Preuss. Akad. Wiss. Math.-Nat. Kl. 1939 (1940),
no. 22, 197.

[Tei44] , Veränderliche Riemannsche Flächen, Deutsche Math. 7 (1944),

344–359.

[Tha96] Michael Thaddeus, Geometric invariant theory and flips, J. Amer. Math.
Soc. 9 (1996), no. 3, 691–723.

[Tho87] Robert W. Thomason, Equivariant resolution, linearization, and Hilbert’s

fourteenth problem over arbitrary base schemes, Adv. in Math. 65 (1987),
no. 1, 16–34.

609
[Tor13] Ruggiero Torelli, Sulle varietà di jacobi, Rend. R. Acc. Lincei (V) 22-2
(1913), 98–103,437–441.

[Tot99] Burt Totaro, The Chow ring of a classifying space, Algebraic K-theory
(Seattle, WA, 1997), Proc. Sympos. Pure Math., vol. 67, Amer. Math.
Soc., Providence, RI, 1999, pp. 249–281.

[TT90] Robert W. Thomason and Thomas Trobaugh, Higher algebraic K-theory

of schemes and of derived categories, The Grothendieck Festschrift, Vol.
III, Progr. Math., vol. 88, Birkhäuser Boston, Boston, MA, 1990, pp. 247–
435.

[Tu93] Loring W. Tu, Semistable bundles over an elliptic curve, Adv. Math. 98
(1993), no. 1, 1–26.

[Vak06] Ravi Vakil, Murphy’s law in algebraic geometry: badly-behaved deforma-

tion spaces, Invent. Math. 164 (2006), no. 3, 569–590.

[Vak17] , The rising sea: Foundations of algebraic geometry,

[Link], 2017.

[Vie95] Eckart Viehweg, Quasi-projective moduli for polarized manifolds, Ergeb-

nisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics
and Related Areas (3)], vol. 30, Springer-Verlag, Berlin, 1995.

[Vis89] Angelo Vistoli, Intersection theory on algebraic stacks and on their moduli
spaces, Invent. Math. 97 (1989), no. 3, 613–670.

[Vis91] , The Hilbert stack and the theory of moduli of families, Geometry
Seminars, 1988–1991 (Italian) (Bologna, 1988–1991), Univ. Stud. Bologna,
Bologna, 1991, pp. 175–181.

[Vis97] , The deformation theory of local complete intersections, 1997.

[Vis05] , Grothendieck topologies, fibered categories and descent theory,

Fundamental algebraic geometry, Math. Surveys Monogr., vol. 123, Amer.
Math. Soc., Providence, RI, 2005, pp. 1–104.

[Wat75] William C. Waterhouse, Basically bounded functors and flat sheaves,

Pacific J. Math. 57 (1975), no. 2, 597–610.

[Wat79] , Introduction to affine group schemes, Graduate Texts in Mathe-

matics, vol. 66, Springer-Verlag, New York-Berlin, 1979.

[Wei48] André Weil, Variétés abéliennes et courbes algébriques, Publ. Inst. Math.
Univ. Strasbourg, vol. 8, Hermann & Cie, Paris, 1948.

[Wei56] , The field of definition of a variety, Amer. J. Math. 78 (1956),

509–524.

[Wei57] , Sur le théorème de Torelli, Séminaire Bourbaki, Vol. 4, Soc.

Math. France, Paris, 1957, pp. Exp. No. 151, 207–211.

[Wei58] , On the moduli of riemann surfaces, To Emil Artin on his sixtieth

birthday. Unpublished manuscript; Collected Papers, Vol. II (1958), 381–
389.

610
[Wei62] , Foundations of algebraic geometry, American Mathematical Soci-
ety, Providence, R.I., 1962.
[Wei94] Charles A. Weibel, An introduction to homological algebra, Cambridge
Studies in Advanced Mathematics, vol. 38, Cambridge University Press,
Cambridge, 1994.
[Wey25] Hermann Weyl, Theorie der Darstellung kontinuierlicher halb-einfacher
Gruppen durch lineare Transformationen. I, Math. Z. 23 (1925), no. 1,
271–309.
[Wey26] , Theorie der Darstellung kontinuierlicher halb-einfacher Gruppen
durch lineare Transformationen. II, Math. Z. 24 (1926), no. 1, 328–376.
[XZ20] Chenyang Xu and Ziquan Zhuang, On positivity of the CM line bundle
on K-moduli spaces, Ann. of Math. (2) 192 (2020), no. 3, 1005–1068.
[Zar39] Oscar Zariski, The reduction of the singularities of an algebraic surface,
Ann. of Math. (2) 40 (1939), 639–689.
[Zar58] , On the purity of the branch locus of algebraic functions, Proc.
Nat. Acad. Sci. U.S.A. 44 (1958), 791–796.

611