HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS

SAM RASKIN

Contents
Preface 1
1. Introduction 2
2. Monoidal structures 24
3. Modules and comodules 27
4. Renormalization 30
5. Weak actions of group schemes 36
6. Ind-coherent sheaves on some infinite dimensional spaces 48
7. Weak actions of group indschemes 69
8. Strong actions 80
9. Semi-infinite cohomology 96
10. Harish-Chandra data 108
11. Application to the critical level 128
References 136

Preface
In different forms and in disparate settings, loop spaces have been central objects of study in the
mathematics of the last century.
In algebraic topology, their study is ubiquitous: the connected components of a based loop space
are the fundamental group; modern methods emphasize the algebraic structure of loop spaces
themselves, providing more information than was available by classical means. In geometry, the
Morse theory of free loop spaces has played a distinguished role since the work [Flo] of Floer thirty
years ago. Witten [Wit1] studied loop spaces via geometric analysis, previewing later developments
in elliptic cohomology. Representations of of loop groups and affine Lie algebras have driven large
parts of the representation theory of the last four decades. In mathematical physics, loop spaces
are manifestly tied to string theory, and also arise when compactifying quantum field theories. A
great deal of exciting mathematics has arisen at this interfaces between these different subjects.
Sometimes, one imagines that a loop space is like a manifold, only infinite dimensional. However,
there are some peculiar phenomena characteristic of this setting and that do not appear in finite
dimensions. We use the term semi-infinite from the title of this work to refer to these characteristic
features. For our purposes, its meaning is somewhat vague, but it is meant to evoke splitting an
infinite dimensional space into two pieces, where the difference of the two infinite pieces is finite
dimensional.
For instance, in algebra, Laurent series split in such a way, as the sum of Taylor series and
polynomials in 𝑡´1 . In algebraic topology, Atiyah’s [Ati] proof of Bott periodicity using Fredholm
operators uses such a splitting. Floer’s Morse complex has this flavor, where indices of critical
1
2 SAM RASKIN

points are infinite but their differences are finite. And loop spaces of manifolds have polarizations
of this flavor, (see [Seg] S4, for example).In representation theory, semi-infinite cohomology of affine
Lie algebras has this flavor. In number theory and arithmetic geometry, the tame symbol arises by
such a construction [BBE].
This manuscript develops some foundational aspects of semi-infinite algebraic geometry. For in-
stance, we develop a theory of coherent sheaves on suitable semi-infinite spaces. Much attention is
given to symmetries of semi-infinite spaces and their categories of sheaves: a substantial portion
of our study relates to group actions in this setting. In addition, we relate our theory to topo-
logical algebras in Beilinson’s sense [Bei], which allows us to circumvent higher categorical issues
when applying our theory to treat concrete problems. Finally, we apply our theory to settle a foun-
dational issue of interest in the geometric Langlands program involving critical level Kac-Moody
representations.

1. Introduction
1.1. What is this work about? Briefly, this manuscript develops foundational aspects of the
algebraic geometry of loop spaces and the (higher) representation theory of loop groups.
At this point, the reader may naturally ask a number of other questions: what aspects? How do
the present foundations compare to existing works? What are the main constructions and results
of this work? Why are they useful, and what are they useful for? Why was this text written?
We ask the reader’s patience as we defer these questions. The bulk of this introduction is intended
to address them. However, given the technical nature of this work, we begin this introduction with
a more informal discussion of loop spaces and their place in algebraic geometry.

1.1.1. Loop spaces in topology. By way of introduction, suppose 𝑀 is a manifold. In this case, there
are two possible interpretations of the loop space of 𝑀 .
Define L𝑀 as the space of smooth maps t𝛾 : 𝑆 1 Ñ 𝑀 u equipped with its standard topology.
One can consider L𝑀 as a sort of infinite dimensional manifold: at a point 𝛾, its tangent space
should be Γp𝑆 1 , 𝛾 ˚ p𝑇𝑀 qq. We can therefore expect to find interesting differential geometry associ-
ated with this space.
There are many instances of this idea in the literature. To name a few:
‚ Witten [Wit1] proposed to study elliptic operators on L𝑀 to obtain geometric invariants in
the spirit of Atiyah-Singer. This idea fostered the development of elliptic cohomology (see
[Lur1] or [Seg] for example).
‚ If 𝑀 “ 𝐺 is a compact Lie group, then there is a rich theory of (projective) Hilbert space
representations of L𝑀 , cf. [PS]. This theory mimics the representation theory of compact
Lie groups in many respects.
‚ Bott [Bot1], [Bot2] used Morse theory on (based) loop spaces to study homology of Ω𝐺 “
L𝐺 ˆ𝐺 ˚ and to prove his celebrated periodicity theorem.
‚ Floer [Flo] used Morse theory on loop spaces to prove a special case of Arnold’s conjecture.
In what follows, the specifics of the above examples are not so important. But we highlight
a few key points. First, each of the constructions above are geometric, not homotopy theoretic.
Second, there are evident functional analytic questions at every stage. Finally, there are significant
difficulties (not all surmounted yet) in importing ideas from finite dimensional geometry into this
infinite dimensional setting.

1.1.2. Loop spaces in algebraic geometry. Broad features of differential geometry often have coun-
terparts in algebraic geometry. Loop spaces provide such an example, as we discuss below.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 3

In what follows, we work over a field 𝑘 fixed once and for all. We assume once and for all that 𝑘
has characteristic 0, though this is not literally needed for every point of our discussion. All schemes
are assumed to be 𝑘-schemes. We speak in absolute terms about relative properties of schemes to
imply reference to Specp𝑘q; e.g., a smooth scheme is a 𝑘-scheme that is smooth over Specp𝑘q.
˝
We let 𝐾 :“ 𝑘pp𝑡qq and 𝑂 :“ 𝑘rr𝑡ss. We let D :“ Specp𝐾q and D :“ Specp𝑂q denote the formal
punctured disc and the formal disc respectively.
˝
We may heuristically think of D as an algebro-geometric version of the circle 𝑆 1 . Here we un-
derstand the geometric circle, not merely its homotopy type.
1.1.3. For 𝑌 an affine scheme of finite type (over 𝑘), there is an indscheme (resp. scheme) 𝑌 p𝐾q
˝
(resp. 𝑌 p𝑂q), the loop (resp. arc) space of 𝑌 , that parametrizes maps D Ñ 𝑌 (resp. D Ñ 𝑌 ).
Below, we first give explicit constructions in the case 𝑌 “ A1 . We then give the definitions in
general.
1.1.4. First, suppose 𝑌 “ A1 .
Then A1 p𝑂q is meant to parametrize maps D Ñ A1 , i.e., Taylor series 𝑖ě0 𝑎𝑖 𝑡𝑖 . We take:
ř

A1 p𝑂q “ Specp𝑘r𝑎0 , 𝑎1 , . . .sq.

That is, of a polynomial algebra with generators labelled by Zě0 . At the risk of redundancy: the
function 𝑎𝑖 : A1 p𝑂q Ñ A1 takes the 𝑖th Taylor coefficient.
1.1.5. Similarly, A1 p𝐾q should parametrize Laurent series 𝑘pp𝑡qq “ colim𝑛 𝑡´𝑛 𝑘rr𝑡ss.
We take:

A1 p𝐾q “ colim Specp𝑘r𝑎´𝑛 , 𝑎´𝑛`1 , . . .sq.

𝑛
Here the structural morphisms:

Specp𝑘r𝑎´𝑛 , 𝑎´𝑛`1 , . . .sq Ñ Specp𝑘r𝑎´𝑛´1 , 𝑎´𝑛`1 , . . .sq

correspond to the evident ring maps:

𝑘r𝑎´𝑛´1 , 𝑎´𝑛 , . . .sq  𝑘r𝑎´𝑛 , 𝑎´𝑛`1 , . . .s

𝑎´𝑛´1 ÞÑ 0
𝑎𝑖 ÞÑ 𝑎𝑖 , 𝑖 ě ´𝑛.
But how should this colimit be understood? We do not mean it in the category of schemes, affine
or otherwise. Rather, it should be understood in a formal categorical sense, as an indscheme. More
precisely, we may understand this colimit in the category of prestacks.
We refer to [GR3] for a introduction to indschemes (in the general setting of derived algebraic
geometry).
1.1.6.
ś𝑛 An obvious variantś of the above discussions hold for 𝑌 “ A𝑛 in place of A1 : take 𝑌 p𝑂q “
1 𝑛 1
𝑖“1 A p𝑂q, and 𝑌 p𝐾q “ 𝑖“1 A p𝐾q.
More generally, if 𝑌 is an affine scheme of finite type, we may embed 𝑌 into A𝑛 for some 𝑛 and
thereby embed 𝑌 p𝑂q (resp. 𝑌 p𝐾q) into A𝑛 p𝑂q (resp. A𝑛 p𝐾q) to deduce that the result is an affine
scheme (resp. indscheme).
Remark 1.1.1. If 𝑌 is not affine, it is well-known that 𝑌 p𝑂q still behaves well, but 𝑌 p𝐾q does not.
See [KV] for further discussion.
4 SAM RASKIN

1.1.7. As the case 𝑌 “ A1 already makes clear, 𝑌 p𝑂q is (almost always) non-Noetherian, i.e., it
is of infinite type, and 𝑌 p𝐾q is ind-infinite type.
Therefore, 𝑌 p𝐾q is infinite dimensional in two regards. It is an indscheme, which is infinite
dimensionality in the ind-direction. Moreover, it is a union of subschemes like 𝑌 p𝑂q, which reflects
infinite dimensionality in the pro-direction. This parallels Laurent series 𝑘pp𝑡qq, which are similarly
infinite dimensional in two ways and in two directions.
In one interpretation, the word semi-infinite from the title of this work refers to this flavor of
geometry. We expand the usual landscape of algebraic geometry in two respects: we wish to consider
infinite type schemes (alias: pro-finite dimensional) and indschemes of ind-finite type (alias: ind-
finite dimensional), and need a class that contains both.
We remark that the intersection of these two classes: schemes (possibly of infinite type) that
are ind-finite type are exactly schemes of finite type. This parallels the linear algebra fact that
topological vector spaces that are discrete and pro-finite dimensional are finite dimensional.
1.1.8. In this manuscript, we develop some foundational aspects of algebraic geometry for such
semi-infinite spaces.
We are particularly interested in studying such spaces in the context of geometric representation
theory, and much of our emphasis reflects this. For instance, if 𝐺 acts on 𝑌 , then 𝐺p𝐾q acts
on 𝑌 p𝐾q, and we might consider 𝐺p𝐾q-equivariant sheaves on 𝑌 p𝐾q, or other implications of
these symmetries for sheaves on 𝑌 p𝐾q. Or we might replace 𝐺p𝐾q by its Lie algebra and consider
infinitesimal versions.
More broadly, we emphasize non-commutative geometry, derived categories of sheaves, group
symmetries, and Lie algebra symmetries. Of course, the overall goal is to recover as closely as
possible classical finite dimensional constructions from these theories in semi-infinite settings.
1.1.9. Below, we begin discussing the contents of this manuscript in more detail, making reference
to loop spaces as motivating examples.
At this point, we might have instead surveyed appearances of algebro-geometric loop spaces in
the literature, in parallel with S1.1.1. We prefer to incorporate connections with recent research
below in our discussion of the present work.
1.2. Brief remarks on categorical conventions. Before delving into the contents of this work,
we comment on some of our conventions.
First, as remarked above, we always work over a field 𝑘 of characteristic 0.
As the title of this work suggests, we use a great deal of homological algebra here. Our preferred
foundations is the 8-categorical approach to DG categories; we refer to [GR4] SI.1 for a detailed
introduction to this perspective.
DG categories are a more robust substitute for triangulated categories. Informally, DG categories
are categories enriched over chain complexes of 𝑘-vector spaces. The derived categories one typically
runs into in algebraic geometry and representation theory all naturally come from DG categories,
and we consider them as such.
DG categories are more readily manipulated than triangulated categories. For instance, if one
wishes to form limits of derived categories, i.e., categories of compatible systems of complexes up to
quasi- isomorphism, the homotopy limit of the corresponding DG categories provides an answer with
suitable properties, while there is no answer using triangulated categories alone. This construction
is quite useful for the purposes of this text: we generally define categories of sheaves on infinite
dimensional spaces as compatible systems on finite dimensional ones.
As indicated above, DG categories are objects of homotopical nature. Therefore, we consider
them as 8-categories in the sense of [Lur2] with extra structure, again, following [GR4]. For us,
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 5

8-categories provide a convenient, easily manipulated, and unified foundation for homotopical
mathematics.
We often drop extra decorations in the terminology, and so we simply refer to categories and
co/limits for 8-categories and homotopy co/limits within them. We speak of 1-categories when we
wish to emphasize categories enriched over sets rather than 8-groupoids.
Remark 1.2.1. To some extent, the reader can probably ignore our choice of foundations during this
introduction and prefer their own models. But to read the body of the work itself, it is necessary
to first be acquainted with [GR4] SI.1. Therefore, during this introduction, we choose not to make
every effort to avoid the terminology of DG categories and 8-categories.
1.3. Coherent sheaves. In S6 of this work, we develop a theory of (ind-)coherent sheaves on
semi-infinite spaces.
Below, we give motivation for this theory and describe some aspects of it.
1.3.1. Finite dimensional recollections. The role of coherent sheaves in conventional, finite dimen-
sional algebraic geometry is well-known.
Many classical invariants and constructions from the Italian1 et al.) found their home in Serre’s
theory [Ser1] of coherent sheaf cohomology and Serre’s duality theorem; see [Die] SVIII for a dis-
cussion.
Work of Auslander-Buchsbaum [AB] and Serre [Ser2] highlighted the interplay between geometry
and derived categories of sheaves, as we discuss further in S1.3.3.
In recent years, the above constructions have been abstracted, most notably in [GR4], via the
functoriality of ind-coherent sheaves on Noetherian schemes. We provide a brief introduction to
this circle of ideas below, and refer to loc. cit. for more context.
1.3.2. We now provide some more technical detail and notation to flesh out the discussion above.
This discussion may be skipped at a first pass.
Suppose 𝑆 is a (classical) scheme of finite type (over our characteristic 0 field 𝑘).
There is a traditional abelian (1-)category QCohp𝑆q♡ of quasi-coherent sheaves on 𝑆. Let Cohp𝑆q♡ Ď
QCohp𝑆q♡ denote the subcategory of coherent sheaves.
We recall that every object of QCohp𝑆q♡ can be realized as a colimit of coherent sheaves. This
can be strengthened with a categorical assertion: the natural functor IndpCohp𝑆q♡ q Ñ QCohp𝑆q♡
is an equivalence. Here for a category C, IndpCq denotes its ind-category; in the higher categorical
context, we refer to [Lur2] S5.3 for an introduction. In particular, there is a canonical categorical
procedure that recovers QCohp𝑆q♡ from Cohp𝑆q♡ .
1.3.3. We now let QCohp𝑆q denote the derived3 category of QCohp𝑆q♡ , which we consider here as
a DG category following our conventions.
Unlike the abelian categorical situation, there are two choices of “small” subcategory in QCohp𝑆q.
First, let Cohp𝑆q Ď QCohp𝑆q denote the subcategory of cohomologically bounded objects with
cohomologies lying in Cohp𝑆q♡ .
Next, let Perfp𝑆q Ď QCohp𝑆q denote the objects that locally on 𝑆 can be represented by bounded
complexes of vector bundles.
Clearly Perfp𝑆q Ď Cohp𝑆q. We have the following standard result in commutative algebra, refer-
enced above.
1Castelnuovo, Cremona, Enriques, Segre, Severi, Zariski2
3Generally speaking, it is better to use the definition of [GR4] SI.3 rather than thinking in terms of derived
categories.
6 SAM RASKIN

Theorem 1.3.1 (Auslander-Buchsbaum, Serre). The inclusion Perfp𝑆q Ď Cohp𝑆q is an equivalence

if and only if 𝑆 is smooth.
Next, we recall the following result of Thomason-Trobaugh.
Theorem 1.3.2. The natural functor IndpPerfp𝑆qq Ñ QCohp𝑆q is an equivalence.
Remark 1.3.3. For the sake of providing references for these statements in our homotopical setting,
we refer to [Gai5] Proposition 1.6.4 and [Lur4] Proposition [Link].
1.3.4. Ind-coherent sheaves. Above, we can also form IndpCohp𝑆qq, which we denote instead as
IndCohp𝑆q.
By the universal property of ind-categories, there is a canonical functor IndCohp𝑆q Ñ QCohp𝑆q,
often denoted as Ψ “ Ψ𝑆 . If 𝑆 is singular, this functor is not an equivalence. In general, if one
has a (small) DG category C closed under direct summands (i.e., idempotent complete), C is the
subcategory of IndpCq consisting of compact objects. Therefore, Ψ being an equivalence would
contradict Theorems 1.3.1 and 1.3.2.
The distinction between IndCohp𝑆q and QCohp𝑆q is not visible classically, i.e., working only with
bounded derived categories. More precisely, IndCohp𝑆q has a canonical 𝑡-structure; we recall that
this means we have subcategories4 IndCohp𝑆qě𝑛 , IndCohp𝑆qď𝑛 of objects in cohomological degrees
ě 𝑛 and ď 𝑛, and the intersection IndCohp𝑆q♡ :“ IndCohp𝑆qě0 XIndCohp𝑆qď0 is an abelian category.
»
Then Ψ is 𝑡-exact, and induces an equivalence IndCohp𝑆qě0 Ý Ñ QCohp𝑆qě0 , hence on bounded below
»
objects more generally. In particular, IndCohp𝑆q♡ Ý Ñ QCohp𝑆q♡ . We remark the close parallel to
»
the more classical equivalence IndpCohp𝑆q♡ q Ý
Ñ QCohp𝑆q♡ .
Therefore, the difference between IndCoh and QCoh is only relevant for unbounded derived cat-
egories.
Remark 1.3.4. Let 𝑆 “ Specp𝑘r𝜀s{𝜀2 q. The complex:
𝜀¨ 𝜀¨ 𝜀¨
... Ý Ñ 𝑘r𝜀s{𝜀2 Ý
Ñ 𝑘r𝜀s{𝜀2 Ý Ñ . . . P 𝑘r𝜀s{𝜀2 –mod “ QCohp𝑆q
is obviously acyclic. But the formal colimit of its stupid truncations:
𝜀¨ 𝜀¨
colim 𝑘r𝑛s “ colim . . . 0 Ñ 0 Ñ 𝑘r𝜀s{𝜀2 Ý
Ñ 𝑘r𝜀s{𝜀2 Ý
` ˘
Ñ . . . P IndCohp𝑆q
𝑛 𝑛 degree ´𝑛
is non-zero, e.g. as may be seen by computing Hom out of the augmentation module:

𝑘 P 𝑘r𝜀s{𝜀2 –mod♡ ♡
𝑓.𝑔. “ Cohp𝑆q Ď Cohp𝑆q.
1.3.5. To summarize, IndCoh exists due to a somewhat natural construction, but the distinction
with QCoh is somewhat subtle. What is its role in algebraic geometry?
We present several answers below.
‚ IndCoh is the natural setting to develop Grothendieck’s functorial approach to Serre duality
and upper-! functors. For instance, it is necessary to work in this setting for the upper-! func-
tor to commute with direct sums. We refer to [Gai5] and [GR4] for a detailed development
of this theory.
‚ IndCoh appears in Koszul duality problems, cf. [BGS], [Pos2], [Lur4] S13-14, [GR4].
‚ IndCoh appears in some problems in geometric representation theory. See e.g. [AG1], [Bez],
[BF1], [BZN]. See also the discussion of S1.3.11 below.
In short, for questions for which QCoh is close-but-wrong, IndCoh often provides the answer.
4We use cohomological gradings and indexing conventions throughout this work.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 7

1.3.6. There is another setting in which IndCoh behaves better than QCoh: when we consider
indschemes rather than schemes.
For abelian categories, this is implicit already in [BD1] 7.11.4. For derived categories, this is
developed in detail in [GR3] S7 and [GR4]. In short, for an indscheme 𝑆 locally of finite type,
IndCohp𝑆q has a nice 𝑡-structure with a nice corresponding abelian category, where there is not
generally one on QCohp𝑆q. And QCohp𝑆q is not generally compactly generated, while IndCohp𝑆q
clearly is. So IndCohp𝑆q can be studied using classical, more finitary methods, while QCohp𝑆q
generally is more pathological.
Remark 1.3.5. In the countable ind-affine case 𝑆 “ colim𝑖ě0 Specp𝐴𝑖 q, we can think of 𝑆 via
the topological algebra 𝐴 “ lim𝑖 𝐴𝑖 (with the pro-topology). In this case, the abelian category
IndCohp𝑆q♡ is the category of discrete 𝐴-modules, i.e., those 𝐴-modules 𝑀 for which every 𝑣 P 𝑀
has open annihilator in 𝐴.
1.3.7. Non-Noetherian settings. In the above situation, we have assumed finite type hypotheses.
For instance, our indschemes above were assumed to be ind-finite type. However, for semi-infinite
mathematics, this is too restrictive.
In S6, we introduce a class of DG indschemes 𝑆 (without finiteness hypotheses) that we call
reasonable indschemes. For instance, this class includes any quasi-compact quasi-separated DG
scheme that is eventually coconnective,5 or any indscheme of ind-finite type. For any reasonable
indscheme 𝑆, we associate a corresponding DG category IndCoh˚ p𝑆q with many similar properties
to the finite type situation.
Remark 1.3.6. The class of reasonable indschemes, which is defined by analogy with a similar notion
in [BD1], may be considered as an answer to the (implicit) call of S1.1.7, to provide for a general
category of “semi-infinite spaces.” Indeed, it is a class containing infinite type schemes (formally:
that are eventually coconnective and quasi-compact quasi-separated), and indschemes of ind-finite
type. Moreover, loop spaces into smooth affine targets are reasonable, cf. Example 6.8.4.
1.3.8. The construction IndCoh˚ is covariantly functorial: for a map 𝑓 : 𝑆 Ñ 𝑇 , we have an induced
functor IndCoh˚ p𝑆q Ñ IndCoh˚ p𝑇 q. Following [Gai5], we denote this functor 𝑓˚IndCoh .
Remark 1.3.7. Unlike in finite type, there is not generally a pullback functor for such a map.
This is the reason for the notation: it is the version of IndCoh with ˚-pushforwards. As in [Gai4],
there is a formally dual DG category IndCoh! p𝑆q :“ IndCoh˚ p𝑆q_ (in the notation of loc. cit.)
with (contravariant) upper-! functors instead. In (ind-)finite type, there is a canonical equivalence
IndCoh! p𝑆q » IndCoh˚ p𝑆q given by Serre duality, cf. [GR4] SII.2.
We remark that our notation here is directly parallel to that of [Ras3].
Remark 1.3.8. The above discussion reflects a general principle in semi-infinite algebraic geometry:
Duality for DG categories is a convenient organizational tool in finite dimensional situations, but
is largely inessential. That is, it provides an interpretation of many standard constructions, and it
sometimes provides helpful structure to arguments.
But in semi-infinite situations, working with DG category duality becomes more essential. More-
over, many of the subtle aspects of the subject have to do with non-trivial duality statements.
For example, as discussed below, we interpret semi-infinite cohomology for Lie algebras as a du-
ality statement. Similarly, one can interpret CDOs for 𝑌 as coming from suitable equivalences
IndCoh˚ p𝑌 p𝐾qq » IndCoh! p𝑌 p𝐾qq, i.e., self-duality for IndCoh˚ .
5This is a hypothesis particular to derived algebraic geometry: it means that the structure sheaf is bounded. Note
that this condition is satisfied for any classical scheme.
8 SAM RASKIN

1.3.9. We also introduce some equivariant versions of IndCoh˚ , i.e., we IndCoh˚ on suitable stacks.
The theory is somewhat more subtle in this regime, and we refer to S6 and S7 for further discussion.
Ignoring some technical points, our theory in particular covers “most” quotients of reasonable
indschemes by groups such as 𝐺p𝐾q for 𝐺 reductive, or by 𝐺p𝑂q for 𝐺 arbitrary.
1.3.10. Applications. Below, we discuss some (anticipated) applications of this theory.
1.3.11. 3𝑑 mirror symmetry. In the last five years, there have been significant advances in the
mathematical understanding of 3𝑑 mirror symmetry conjectures.
We refer to [BF2] S7 and the introduction to [HR] for detailed discussion of this area, and
defer attributions to Remark 1.3.10. But briefly, and somewhat heuristically, certain fundamental
conjectures in this area take the form:
˝
IndCoh˚ pM𝑎𝑝𝑠pD𝑑𝑅 , 𝑌1 qq » 𝐷˚ p𝑌2 p𝐾qq. (1.3.1)
Some remarks on the notation are in order.
‚ Here 𝑌1 and 𝑌2 are certain algebraic stacks of finite type; typically, they are quotients of
smooth affine varieties by the action of a reductive group.
‚ The relationship between 𝑌1 and 𝑌2 is not arbitrary; they should be 3d mirror dual pairs.
We refer to [BF2] S4 for some examples.
˝ ˝
‚ In (1.3.1), M𝑎𝑝𝑠pD𝑑𝑅 , 𝑌1 q is the moduli stack of flat maps from D to 𝑌1 . For instance, if
𝑌1 “ B𝐺, then this is the space of de Rham 𝐺-local systems on the punctured disc. This
space is an alternative to the derived loop space of a stack, and it has similar properties
(and the two coincide if 𝑌1 is an affine scheme).
‚ In (1.3.1), 𝐷˚ indicates a suitable category of 𝐷-modules, as defined in infinite type in
[Ras3].
‚ For physics purposes, the right (resp. left) hand side of (1.3.1) is the category of line
operators in the 𝐴-twist (resp. 𝐵-twist) of the 3𝑑 N “ 4 quantum field theory defined by
𝑌𝑖 (namely, the sigma model of maps into its cotangent stack). Physics predicts that the
𝐴-twist of the theory defined by 𝑌2 is equivalent (as a QFT) to the 𝐵-twist of the theory
defined by 𝑌1 for a mirror dual pair p𝑌1 , 𝑌2 q.
The left hand side of (1.3.1) is not a priori defined, so this conjecture is not precisely formulated
above (as acknowledged in [BF2]).
Example 1.3.9. In [HR], which is joint with Justin Hilburn, we consider the case 𝑌1 “ A1 {G𝑚 and
˝
𝑌2 “ A1 . We show that M𝑎𝑝𝑠pD𝑑𝑅 , A1 {G𝑚 q is the quotient of a reasonable indscheme by an action
of G𝑚 p𝐾q, so the present text makes sense of IndCoh˚ on this mapping space. We then prove the
equivalence (1.3.1) in this case, using the definitions provided in the present work for the left hand
side.
We expect the theory of IndCoh˚ in S6 leads more generally to accurate and precise conjectures.
Remark 1.3.10. We now address some of the lineage of 3d mirror symmetry. In physics, the general
idea that certain supersymmetric 3𝑑 theories might be non-trivially equivalent first appeared in [IS],
and was further developed by [HW]. In unpublished work, Hilburn-Yoo gave the algebro-geometric
description of the categories of line operators in 𝐴 and 𝐵-twists of 3𝑑 N “ 4 sigma models of the
type considered above, leading to conjectures of the above types. In addition, Costello, Dimofte,
and Gaitto (at least) played important roles in these developments. Connections between 3𝑑 mirror
symmetry with geometric Langlands began in physics with work [GW] of Gaitto-Witten, and was
further developed by Hilburn-Yoo, Braverman-Finkelberg-Nakajima [BFN], Braverman-Finkelberg
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 9

[BF2], and Ben-Zvi-Sakellaridis-Venkatesh. See also [BDGH] and [DGGH] for recent discussion in
the mathematical physics literature.
1.3.12. Factorizable Satake. In [CR], joint with Justin Campbell, we prove a factorizable (cf. [Ras5])
version of the derived geometric Satake theorem of [BF1].
This result was anticipated over a decade ago by Gaitsgory-Lurie, and is discussed in [Gai6] S4.7.
As in loc. cit., this result plays a key role in Gaitsgory’s approach to global geometric Langlands
conjectures.
One reason such a result was not proved earlier is that, unlike the non-factorizable version,
the theorem involves IndCoh on stacks of infinite type, so a definition of one side was not readily
available.
In [CR], we again see that the theory of IndCoh˚ provided here yields the “right” answer for
(derived, factorizable) geometric Satake.

1.3.13. Cautis-Williams. In [CW], Cautis and Harold propose a definition for the category of
half-BPS line operators in 4𝑑 𝒩 “ 2 gauge theories via coherent sheaves on spaces R𝑉,𝐺 :“
𝑉 p𝑂q{𝐺p𝑂q ˆ𝑉 p𝐾q{𝐺p𝐾q 𝑉 p𝑂q{𝐺p𝑂q considered in [BFN]. Here 𝐺 is a reductive group and 𝑉 is
a finite-dimensional 𝐺-representation.
A BFN space R𝑉,𝐺 is a quotient of a reasonable DG indscheme by an action of 𝐺p𝑂q, although
they are of infinite type and highly DG (i.e., non-classical). As such, Cautis-Williams use our theory
of IndCoh˚ to study coherent sheaves on these spaces.

1.3.14. Weak loop group actions. The application of IndCoh˚ within the present work is to develop
a theory of weak loop group actions on categories and to provide a categorical framework for
semi-infinite cohomology. We discuss these applications at length in S1.4 below.

1.4. Loop group actions on categories. In S7, we develop a theory of weak loop group actions
on DG categories, which constitutes a major part of the present work.
Below, we recall the finite dimensional theory (due to Gaitsgory), motivate and describe our semi-
infinite theory, connect to more classical ideas in infinite dimensional algebra, and give applications.
In brief, group actions on categories provide a unifying framework for many constructions in geo-
metric representation theory, and the theory for loop groups plays a foundational role in geometric
Langlands.

1.4.1. Preliminary remarks. We begin with some attributions, references, and historical comments.
(Some of the discussion may only make sense after reading subsequent of S1.4.)
The idea began, apparently, with [BD1] S7. In loc. cit., Beilinson and Drinfeld developed some
aspects of the theory of weak group actions on categories. As we discuss in S1.4.25, they used their
constructions in the case of loop groups to construct Hecke eigensheaves via localization. That
is to say, the initial development of the theory were in the setting of loop groups and geometric
Langlands.
The ideas of Beilinson-Drinfeld were developed by Gaitsgory in a series of works, sometimes
with co-authors, and sometimes only in informally distributed works. In [Gai2], he introduced
abelian categories over stacks; specializing to B𝐺, one obtains weak group actions on categories.
The Beilinson-Drinfeld ideas on Hecke patters were generalized in the appendices to [FG1], which
developed a theory of algebraic groups acting on abelian categories, and some parts of the theory
of loop groups acting on abelian categories. In the latter setting, it is inadequate to use bounded
derived categories, leading to some deficiencies in the generality of the results in loc. cit. Finally,
in finite dimensions, a robust derived theory was developed in [Gai8].
10 SAM RASKIN

The theory of strong loop group actions on DG categories was developed by Beraldo in [Ber].
This work also develops some consequences of the results in [Gai8], and therefore is a convenient
reference for the finite dimensional theory.
However, a well-developed theory of weak actions has not appeared before the present work. Even
if one is only interested in strong actions, the weak theory is needed to connect with Kac-Moody
representations; this is crucial for applications of Beilinson-Drinfeld type.
Finally, we refer to [ABC` ] for further discussion.
1.4.2. Algebraic group actions. Let 𝐺 be an affine algebraic group (in particular, finite type).
There are two flavors of 𝐺-actions on categories: weak and strong. We discuss each below.
1.4.3. Let DGCat𝑐𝑜𝑛𝑡 denote the symmetric monoidal (8-)category of cocomplete DG categories;
see S1.8 and S2.2 for more details.
Let QCohp𝐺q P DGCat𝑐𝑜𝑛𝑡 denote the DG category of quasi-coherent sheaves on 𝐺. The group
structures on 𝐺 induces a convolution monoidal structure on QCohp𝐺q.
Definition 1.4.1. A weak 𝐺-action on C P DGCat𝑐𝑜𝑛𝑡 is a QCohp𝐺q-module structure. We let
𝐺–mod𝑤𝑒𝑎𝑘 denote the category of (DG) categories with strong 𝐺-actions, i.e., QCohp𝐺q–modpDGCat𝑐𝑜𝑛𝑡 q.
Remark 1.4.2. For a 𝑘-point 𝑔 : Specp𝑘q Ñ 𝐺, we can form a skyscraper sheaf 𝑔˚ p𝑘q P QCohp𝐺q♡ .
For C P 𝐺–mod𝑤𝑒𝑎𝑘 , acting by 𝑔˚ p𝑘q defines an automorphism 𝑔 ¨ ´ : C Ñ C. That is, we get a map
𝐺p𝑘q :“ HompSpecp𝑘q, 𝐺q Ñ AutpCq. The above definition of a weak 𝐺-action can heuristically be
understood as refining such a map 𝐺p𝑘q Ñ AutpCq to allow 𝑘-points to vary “continuously” in the
natural sense of algebraic geometry.
For C P 𝐺–mod𝑤𝑒𝑎𝑘 , we define the corresponding weak invariants and weak coinvariants cate-
gories as:

C𝐺,𝑤 :“ Hom𝐺–mod𝑤𝑒𝑎𝑘 pVect, Cq C𝐺,𝑤 :“ Vect b C.

QCohp𝐺q
As recalled in Theorem 5.10.1, a theorem of Gaitsgory constructs functorial equivalences C𝐺,𝑤 »
C𝐺,𝑤 .
Remark 1.4.3. Above, Vect is considered as a categorification of the trivial 1-dimensional repre-
sentation of a group. The identity “invariants = coinvariants” is a categorification of Maschke’s
theorem, considering 𝐺 as analogous to a finite group.
1.4.4. Let 𝐷p𝐺q P DGCat𝑐𝑜𝑛𝑡 denote the category of 𝐷-modules on 𝐺, which again carries a
convolution monoidal structure.
Definition 1.4.4. A strong 𝐺-action on C P DGCat𝑐𝑜𝑛𝑡 is a 𝐷p𝐺q-module structure. We let 𝐺–mod
denote the category of (DG) categories with strong 𝐺-actions, i.e., 𝐷p𝐺q–modpDGCat𝑐𝑜𝑛𝑡 q.
As in the notation 𝐺–mod above, we often omit the term strong, and refer simply to actions of
𝐺 on DG categories.
We again have invariants and coinvariants categories C𝐺 and C𝐺 defined as in the weak setting.
As in [Ber], there is again a canonical equivalence C𝐺 » C𝐺 .
1.4.5. We now discuss examples.
If 𝐺 acts on some scheme 𝑋, then 𝐺 (canonically) acts weakly on QCohp𝑋q. If 𝑋 is locally of
finite type, then 𝐺 acts weakly on IndCohp𝑋q and strongly on 𝐷p𝑋q (the DG category of 𝐷-modules
on 𝑋).
The weak invariant categories:
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 11

QCohp𝑋q𝐺,𝑤 , IndCohp𝑋q𝐺,𝑤
are tautologically DG categories of equivariant sheaves and coincide with suitable sheaves on the
stack 𝑋{𝐺. The category:

𝐷p𝑋q𝐺
of (strong) invariants is the category of 𝐷-modules on the stack 𝑋{𝐺, and coincides with the
classical equivariant derived category.
1.4.6. We can also mix the two settings.
First, the functor of (right) 𝐷-module induction ind : QCohp𝐺q Ñ 𝐷p𝐺q is monoidal. This
provides a forgetful functor:

Oblv “ Oblv𝑠𝑡𝑟Ñ𝑤 : 𝐺–mod Ñ 𝐺–mod𝑤𝑒𝑎𝑘 . (1.4.1)

In other words, categories with strong 𝐺-actions have underlying weak 𝐺-actions.
1.4.7. Now consider 𝐷p𝐺q with its natural strong 𝐺 ˆ 𝐺-action. Taking weak invariants on the
right, we obtain a strong 𝐺-action on:

𝐷p𝐺q𝐺,𝑤 » g–mod.
Clearly g–mod𝐺 “ Vect𝐺,𝑤 “ Repp𝐺q. More generally, given 𝐻 Ď 𝐺, g–mod𝐻 is the (DG)
category of Harish-Chandra modules for the pair pg, 𝐻q; at the abelian categorical level, these are
representations of g and a lift6 of the action of h Ď g to an action of 𝐻.
1.4.8. It is often convenient to work with Hecke actions.
Given a weak (resp. strong) action of 𝐺 on C, there is an induced action of the monoidal category
QCohp𝐻z𝐺{𝐻q on C𝐻,𝑤 (resp. 𝐷p𝐻z𝐺{𝐻q on C𝐻 ). Indeed, C𝐻,𝑤 “ Hom𝐺–mod𝑤𝑒𝑎𝑘 pQCohp𝐺{𝐻q, Cq
and QCohp𝐻z𝐺{𝐻q “ End𝐺–mod𝑤𝑒𝑎𝑘 pQCohp𝐺{𝐻qq.
Example 1.4.5. Let 𝐺 be semisimple and let 𝐵 Ď 𝐺 be a Borel subgroup. We obtain an action
of 𝐷p𝐵z𝐺{𝐵q on g–mod𝐵 . Let 𝑗𝑤 : 𝐵z𝐵𝑤𝐵{𝐵 ãÑ 𝐵z𝐺{𝐵 denote the locally closed embedding of
a Bruhat cell. The actions of 𝑗𝑤,˚,𝑑𝑅 pICq, 𝑗𝑤,! pICq P 𝐷p𝐵z𝐺{𝐵q on g–mod𝐵 are Arkhipov twisting
functors from [Ark2] (essentially by definition). Because these objects are well-known to be invertible
»
in the monoidal category 𝐷p𝐵z𝐺{𝐵q, they define auto-equivalences 𝑗𝑤,˚,𝑑𝑅 ‹´, 𝑗𝑤,! ‹´ : g–mod𝐵 Ý Ñ
g–mod𝐵 .
These automorphisms play a fundamental role in some approaches to studying the BGG category
O, cf. [Hum]. In other words, the Hecke action on O :“ g–mod𝐵 (or a variant with generalized central
character) is a crucial structure that is non-obvious from classical perspectives and transparent from
the perspective of group actions on categories.
1.4.9. The setting of loop groups. Now suppose we replace 𝐺 by 𝐺p𝐾q.
In this case, a monoidal category7 𝐷˚ p𝐺p𝐾qq of 𝐷-modules on 𝐺p𝐾q was defined in [Ber] (see
also [Ras3]).
Therefore, we may define 𝐺p𝐾q–mod as 𝐷˚ p𝐺p𝐾qq–modpDGCat𝑐𝑜𝑛𝑡 q, just as in the finite dimen-
sional setting.
6If 𝐻 is connected, such a lift is unique if it exists.
7Here the notation 𝐷 ˚ appears instead of 𝐷 because of the semi-infinite nature of 𝐺p𝐾q. The notation is exactly
in parallel with that of S1.3.4.
12 SAM RASKIN

Remark 1.4.6. The definition of invariants and coinvariants categories make sense in this setting.
However, for group indschemes such as 𝐺p𝐾q, there is not a natural equivalence between invariants
and coinvariants, although there is for 𝐺p𝑂q. Following the heuristic of Remark 1.4.3, this is because
𝐺 behaves like a finite group, 𝐺p𝑂q behaves like a profinite group, and 𝐺p𝐾q behaves like a non-
compact topological group.
However, in the special case when 𝐺 is reductive, one may construct a (somewhat non-canonical)
equivalence C𝐺p𝐾q » C𝐺p𝐾q , bootstrapping from 𝐺p𝑂q using the ind-properness of the affine Grass-
mannian Gr𝐺 “ 𝐺p𝐾q{𝐺p𝑂q.
For a special (Whittaker ) setting in which equivalences of this form hold for unipotent 𝐺, see (in
increasing orders of generality) [Gai3], [Ber], and [Ras6].
1.4.10. The setting of weak actions of loop groups on DG categories has proved more elusive, and
is developed here in S7. We provide a brief overview of the theory here. For simplicity, we assume8
𝐺 is reductive in the discussion that follows.

1.4.11. We define a category 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 in S7.

As in finite dimensions, there are fundamental functors:

Oblv, p´q𝐺p𝐾q,𝑤 , p´q𝐺p𝐾q,𝑤 : 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 Ñ DGCat𝑐𝑜𝑛𝑡

that are the forgetful functor, (weak) coinvariants, and (weak) invariants functors.
But a major technical subtlety in this setting is that the functor:

Oblv : 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 Ñ DGCat𝑐𝑜𝑛𝑡

is not conservative. This means that there are morphisms C1 Ñ C2 in 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 that are not
equivalences, but are at the level of their underlying DG categories. (In this regard, 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘
behaves somewhat like IndCoh on a singular affine scheme, but with one additional categorical level
of complexity.)
Therefore, we cannot simply define 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 as IndCoh˚ p𝐺p𝐾qq–modpDGCat𝑐𝑜𝑛𝑡 q or some-
thing along these lines. This presents technical challenges in the development of the theory and its
functoriality, and even in the construction of natural objects here.
Remark 1.4.7. In S7, we refer to objects of 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 as categories with genuine (weak) actions.
Relatedly, we actually denote the above functor Oblv by Oblv𝑔𝑒𝑛 . The terminology is borrowed
from stable homotopy theory; in some cosmetic regards, the non-conservativeness of Oblv𝑔𝑒𝑛 makes
𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 appear somewhat analogous to category of genuine 𝐺-spectra for a finite group 𝐺.
1.4.12. With the above in mind, for the purposes of the introduction, we largely ignore the non-
conservativeness of the forgetful functor, and essentially assume the constructions one naively ex-
pects to have (e.g., a symmetric monoidal structure on 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 with unit the “trivial”
representation Vect P 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 ) exist.
Of course, we refer to S7 for a careful development.

8More generally, our theory works for any Tate group indscheme, cf. S7.2. The loop group 𝐺p𝐾q satisfies this
hypothesis only when 𝐺 is reductive.
But for more general affine algebraic 𝐺, one can also allow the formal completion of 𝐺p𝐾q along 𝐺p𝑂q, or any
other congruence subgroup. This allows one to apply our methods to recover semi-infinite cohomology for Tate Lie
algebras in full generality, cf. S1.4.19.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 13

1.4.13. What becomes of invariants and coinvariants for weak 𝐺p𝐾q-actions?

Here one discovers the peculiar phenomenon characteristic of infinite dimensions: what we refer
to as the modular character 𝜒𝑇 𝑎𝑡𝑒 of 𝐺p𝐾q.
By definition, this is a certain canonical object 𝜒𝑇 𝑎𝑡𝑒 P 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 . Non-canonically, Oblvp𝜒𝑇 𝑎𝑡𝑒 q P
DGCat𝑐𝑜𝑛𝑡 is isomorphic to Vect, so we can think of 𝜒𝑇 𝑎𝑡𝑒 as a categorification of an action of 𝐺p𝐾q
on a line. (As the name suggests, we regard this object as a categorification of the modular character
of a Lie group, defined as usual via its Haar measure.)
The key property of 𝜒𝑇 𝑎𝑡𝑒 is the existence of functorial equivalences:

pC b 𝜒𝑇 𝑎𝑡𝑒 q𝐺p𝐾q,𝑤 » C𝐺p𝐾q,𝑤 (1.4.2)

for C P 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 . In words: (weak) invariants and coinvariants for 𝐺p𝐾q coincide up to a
twist by the modular character.
Remark 1.4.8. If we replace 𝐺p𝐾q by an affine algebraic group 𝐺 (or an affine group scheme such
as 𝐺p𝑂q), the modular character is canonically trivial: i.e., one has a canonical equivalence with
Vect equipped with its trivial action.
In S1.4.19, we discuss how this modular character recovers classical constructions in semi-infinite
algebra.

1.4.14. Next, let us discuss the relationship with strong 𝐺p𝐾q-actions and affine Lie algebras.
In S8, we construct a forgetful functor:

𝐺p𝐾q–mod Ñ 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘
refining the natural forgetful functor 𝐺p𝐾q–mod Ñ DGCat𝑐𝑜𝑛𝑡 . (This is only non-trivial because of
the subtleties in the definition of 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 .)

1.4.15. By definition, there is a canonical object 𝐷˚ p𝐺p𝐾qq P 𝐺p𝐾q–mod. Let 𝐷! p𝐺p𝐾qq be the
dual DG category, cf. [Ras3] or Remark 1.3.7. This object is canonically a 𝐷˚ p𝐺p𝐾qq-module, so
an object of 𝐺p𝐾q–mod as well.
Let g denote the Lie algebra of 𝐺 and gpp𝑡qq :“ g b 𝑘pp𝑡qq.
In Lemma 9.13.1, we construct a canonical equivalence:

gpp𝑡qq–mod » 𝐷! p𝐺p𝐾qq𝐺p𝐾q,𝑤 . (1.4.3)

Here the definition of the (unbounded) DG category gpp𝑡q–mod of (smooth) gpp𝑡qq-modules, due
to Frenkel-Gaitsgory [FG2], is somewhat subtle due to semi-infinite issues; we review it in S4.
That aside, in the above equivalence, the trivial representation 𝑘 P gpp𝑡qq–mod corresponds to the
!-pullback of 𝑘 along the structure map 𝐺p𝐾q Ñ Specp𝑘q.

1.4.16. Applications. We now relate the theory outlined above to other literature and provide some
applications.

1.4.17. Semi-infinite cohomology revisited. First, we indicate how the theory of weak 𝐺p𝐾q-actions
provides a new perspective on classical semi-infinite cohomology, as introduced in [Fei]. This ma-
terial is the subject of S9.
Remark 1.4.9. There have been various previous attempts to provide conceptual constructions of
semi-infinite cohomology: see [Vor], [Ark1], and [Pos1] for example. Our perspective emphasizes the
connection to the higher categorical representation theory of the loop group.
14 SAM RASKIN

1.4.18. First, we construct an equivalence:

g´𝑇 𝑎𝑡𝑒 –mod » 𝐷˚ p𝐺p𝐾qq𝐺p𝐾q,𝑤

p (1.4.4)
similar to (1.4.3), where the notation deserves more explanation. Here p
g´𝑇 𝑎𝑡𝑒 indicates the canonical
Kac-Moody central extension of gpp𝑡qq (or its opposite); see e.g. [BD2] S2.7, S3.8, and [BD1] S7.13
for convenient reviews (including the relation to semi-infinite cohomology, cf. S1.4.19). We remind
that this Lie algebra is the Kac-Moody extension defined by minus the Killing form for g. By
g´𝑇 𝑎𝑡𝑒 –mod, we mean the Frenkel-Gaitsgory DG category of smooth modules on which the central
p
element acts by the identity.
The above equivalence is actually non-canonical: it depends on a choice of a compact open 9
subgroup of 𝐺p𝐾q; in the discussion that follows, we suppose we have chosen 𝐺p𝑂q for this purpose.
Here is the idea of the construction. First, the choice of compact open subgroup induces an
equivalence 𝐷! p𝐺p𝐾qq » 𝐷˚ p𝐺p𝐾qq P 𝐺p𝐾q–mod. We then have:

(1.4.2)
𝐷˚ p𝐺p𝐾qq𝐺p𝐾q,𝑤 » 𝐷! p𝐺p𝐾qq𝐺p𝐾q,𝑤 » p𝐷! p𝐺p𝐾qq b 𝜒b´1
𝑇 𝑎𝑡𝑒 q
𝐺p𝐾q,𝑤
.
This is similar to (1.4.3) except for the twist by 𝜒b´1
𝑇 𝑎𝑡𝑒 . We explain in S9 that this twist amounts to
considering modules over the canonical central extension of gpp𝑡qq.

1.4.19. Next, as 𝐷˚ p𝐺p𝐾qq is dual to 𝐷! p𝐺p𝐾qq and invariants are dual to coinvariants, it follows
formally from (1.4.3) and (1.4.4) that p g´𝑇 𝑎𝑡𝑒 –mod is dual as a DG category to gpp𝑡qq–mod.
In particular, gpp𝑡qq–mod identifies with the category of (continuous DG) functors pg´𝑇 𝑎𝑡𝑒 –mod Ñ
Vect. Therefore, the trivial representation 𝑘 P gpp𝑡qq–mod defines a functor:

g´𝑇 𝑎𝑡𝑒 –mod Ñ Vect.

p (1.4.5)
There is a well-known such functor that plays a key role in the study of affine Lie algebras:
semi-infinite cohomology. We recall this is a functor:
8
𝐶 2 pgpp𝑡qq, grr𝑡ss; ´q : p
g´𝑇 𝑎𝑡𝑒 –mod Ñ Vect
where, as the notation indicates, there is also (mild) dependence on the choice of compact open
subgroup grr𝑡ss.
We show that under our duality, the functor (1.4.5) coincides with classical semi-infinite coho-
mology. This is done in S9. It follows more generally that the pairing:

g´𝑇 𝑎𝑡𝑒 –mod Ñ Vect

gpp𝑡qq–mod b p
underlying the duality gpp𝑡qq–mod “ ppg´𝑇 𝑎𝑡𝑒 –modq_ is calculated by tensoring two modules and
forming semi-infinite cohomology of the resulting p
g´𝑇 𝑎𝑡𝑒 -module.
Remark 1.4.10. The existence of a duality with this property was anticipated in [AG3] S2.2. Our
treatment differs from loc. cit.: we construct a duality above on abstract grounds and then show
it recovers classical semi-infinite cohomology, while Arkhipov-Gaitsgory use the latter to establish
duality.
In addition, some details are missing in arkhipov-gaitsgory-localization. Moreover, our approach
makes it essentially tautological that the duality is strongly 𝐺p𝐾q-equivariant, which is of funda-
mental importance (see below) and difficult to see from the perspective of [AG3].
9See S7.2 for the definition.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 15

Remark 1.4.11. There is a generalization of the above for general affine Kac-Moody algebras:

g𝜅 –modq_ » p
pp g´𝜅´𝑇 𝑎𝑡𝑒 –mod.
The ideas are the same, so we have not emphasized it in our present discussion. See S11 for more
details.
1.4.20. Construction of a 𝐺p𝐾q-action on gpp𝑡qq–mod. The equivalence (1.4.3) induces a strong ac-
tion of 𝐺p𝐾q on gpp𝑡qq–mod. (There is a similar generalization allowing Kac-Moody twists, cf. S11.)
As we will discuss, this construction plays a key role in some studies of Kac-Moody representations.
Note that this assertion do not mention weak 𝐺p𝐾q-actions: gpp𝑡qq–mod can be defined as a
DG category directly, and we are discussing strong 𝐺p𝐾q-actions. And indeed, the existence of a
𝐺p𝐾q-action on gpp𝑡qq–mod was previously outlined in [Gai7].
However, as we discuss below, our perspective here is more robust and fills a number of gaps in
the literature.
1.4.21. The action defined by (1.4.3) implies a universal property for gpp𝑡qq–mod as a strong
𝐺p𝐾q-category: given C P 𝐺p𝐾q–mod, a 𝐺p𝐾q-equivariant functor C Ñ gpp𝑡qq–mod is equivalent to
a weakly 𝐺p𝐾q-equivariant functor C Ñ Vect, i.e., a functor C𝐺p𝐾q,𝑤 Ñ Vect.
In other words, although we are discussing a strong action, when thinking about gpp𝑡qq–mod as
a 𝐺p𝐾q-module, it is helpful to know about 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 .
1.4.22. In addition, our perspective on the 𝐺p𝐾q-action here make various functoriality results
that have tacitly been assumed in the literature.
For instance, [AG3] Theorem-Construction 4.2.2 is unproved in loc. cit. It asserts the existence
of a (strongly) 𝐺p𝐾q-equivariant (coherent) global sections functor:

ΓpBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 , ´q : 𝐷˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 p𝑋qq Ñ gpp𝑡qq–mod (1.4.6)
where 𝑋 is a smooth projective curve, 𝑥 P 𝑋p𝑘q is a 𝑘-point with formal completion identified10
with Spfp𝑘rr𝑡ssq, and Bun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 is the (infinite type) scheme parametrizing 𝐺-bundle on 𝑋 with a
trivialization on the formal neighborhood of 𝑥.
Remark 1.4.12. Although the above functor is constructed in [AG3], the construction of its 𝐺p𝐾q-
equivariance is not shown, although it plays a key role in loc. cit. And it appears difficult to establish
using the techniques of loc. cit.
From our perspective, the 𝐺p𝐾q-equivariance is readily established. Because this plays a key role
in other literature, we briefly outline the construction below.
Remark 1.4.13. As we discuss in S1.4.25, the above functor is closely related to the Beilinson-
Drinfeld localization functor, which plays a key role in the de Rham geometric Langlands program
and for which there are also gaps in the literature.
1.4.23. As just stated, we digress to outline the construction of a 𝐺p𝐾q-equivariant functor (1.4.6).
We find this example illustrates some important ideas in this text, but this material may safely be
skipped by the reader. In particular, it relies on some basic familiarity with 𝐺-bundles on curves
that is not assumed elsewhere. Moreover, it assumes some working knowledge of 𝐷-modules and
IndCoh in infinite type, as developed in [Ras3] and S6 of this text.
Let Bun𝐺 denote the finite type Artin stack of 𝐺-bundles on 𝑋. Recall that Bun𝐺 is not quasi-
compact: rather, we can write it as a union Bun𝐺 “ colim𝑖 𝑈𝑖 for 𝑈𝑖 Ď Bun𝐺 open quasi-compact
10To say it better: our ambient formal disc is assumed to be the one based around 𝑥.
16 SAM RASKIN

substacks. Recall that Bun𝑙𝑒𝑣𝑒𝑙,𝑥

𝐺 Ñ Bun𝐺 is a 𝐺p𝑂q-torsor. We let 𝑈 r𝑖 Ď Bun𝑙𝑒𝑣𝑒𝑙,𝑥 denote the
𝐺
inverse image of 𝑈𝑖 , which is an open quasi-compact subscheme of Bun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 .
Because each 𝑈 r𝑖 q » 𝐷! p𝑈
r𝑖 is a 𝐺p𝑂q-torsor over a smooth stack 𝑈𝑖 , it follows that 𝐷˚ p𝑈 r𝑖 q and
˚ r ! r
IndCoh p𝑈𝑖 q » IndCoh p𝑈𝑖 q canonically; indeed, for the former equivalence, see [Ras3] Proposition
4.8.1, and for the latter, one notes that the natural functor Ψ : IndCoh˚ p𝑈r𝑖 q Ñ QCohp𝑈r𝑖 q (cf. S6) is
˚ r
an equivalence for pro-smooth schemes, so IndCoh p𝑈𝑖 q is canonically self-dual as a DG category, as
this is always true for QCoh on a quasi-compact scheme. For convenience, we denote these categories
simply by 𝐷p𝑈 r𝑖 q and IndCohp𝑈r𝑖 q in what follows.
We then have:

𝐷˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q “ colim 𝐷p𝑈
r𝑖 q
𝑖
(1.4.7)
IndCoh ˚
pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q “ colim 𝐷p𝑈
r𝑖 q
𝑖

where each of these colimits is formed in DGCat𝑐𝑜𝑛𝑡 and the structure maps are pushforwards.
Therefore, there is a canonical11 forgetful functor:

Oblv : 𝐷˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q Ñ IndCoh˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q.

Because 𝐺p𝐾q acts on Bun𝑙𝑒𝑣𝑒𝑙,𝑥

𝐺 , it acts strongly on 𝐷˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q and weakly on IndCoh˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q.
12
One can readily verify that there is a canonical weak 𝐺p𝐾q-equivariant structure on Oblv.

11This forgetful functor is a little funny: naturally, one has a forgetful functor Oblv : 𝐷 ! p𝑆q Ñ IndCoh! p𝑆q by
dualizing the discussion in S6.20. This gives forgetful functors 𝐷p𝑈 r𝑖 q “ 𝐷! p𝑈
r𝑖 q Ñ IndCoh! p𝑈r𝑖 q “ IndCohp𝑈 r𝑖 q. These
functors are naturally compatible under the structure functors in the above colimit, so give a forgetful functor of the
desired type.
In other words, we use the identifications 𝐷! p𝑈 r𝑖 q “ 𝐷˚ p𝑈
r𝑖 q (and similarly for IndCoh) to construct the forgetful
functor above. (This may be compared to [BD1] S7.14.3.)
12Let us outline the argument, since it is a little technical.
First, one more naturally shows that the forgetful functor Oblv : 𝐷! pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q Ñ IndCoh! pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q is canonically
weakly 𝐺p𝐾q-equivariant; this follows from general functoriality considerations. The difference here is that, as in the
previous discussion, this forgetful functor exists more generally and is more natural, so its functoriality is more easily
established.
Then we claim that there is a unique naive (in the sense of S7) weak 𝐺p𝐾q-equivariant structure on Oblv :
𝐷˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q Ñ IndCoh˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q fitting into a commutative diagram:

𝐷˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q / IndCoh˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q

 
𝐷! pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q / IndCoh! pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q
where the vertical maps are the canonical arrows resulting from (1.4.7).
For this, one uses Step (2) from the proof of Lemma 8.16.1 (taking A𝑟𝑒𝑛 “ A “ IndCoh˚ p𝐺p𝐾qq in loc. cit). This
construction produces a certain monoidal DG category B with a quotient functor B Ñ IndCoh˚ p𝐺p𝐾qq. It is easy to
establish B-linearity of the above functors: one reduces to the (non-cocomplete) monoidal subcategory B𝑐 , and then
because compact objects in IndCoh˚ p𝐺p𝐾qq have finite support, the claim is essentially formal from the definitions.
Because B Ñ IndCoh˚ p𝐺p𝐾qq is a quotient functor, this suffices.
Finally, using (a simple version of) the constructions from S8.14, one sees that this structure canonically upgrades
to a genuine weak 𝐺p𝐾q-equivariant structure.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 17

1.4.24. Duality for Kac-Moody algebras. In the discussion above, we have emphasized an aesthetic
feature of our theory of weak 𝐺p𝐾q-actions: it reinterprets semi-infinite cohomology as a duality be-
tween DG categories. However, this result also has practical significance, especially in its (strongly)
𝐺p𝐾q-equivariant form.
‚ In [Dhi], Dhillon uses Kac-Moody duality to construct semi-infinite cohomology for W-
algebras using group actions on categories and [Ras6].
‚ In [AG3], Arkhipov-Gaitsgory use this duality to show Kashiwara-Tanisaki localization for
thin flags (at negative levels) implies their localization theorem for thick flags (at positive
levels).
‚ In [Ras7] Appendix A, we use duality to provide a conceptual interpretation of the main
construction of [AG2].
‚ Kac-Moody duality appears in studies of the positive level cases of Gaitsgory’s FLE; see
[ABC` ], [CDR], and [Liu].
‚ In [Gai9], Gaitsgory uses Kac-Moody duality throughout his study of the Kac-Moody side
of Kazhdan-Lusztig style equivalences. (Note that in loc. cit. S1.3, introducing this duality,
Gaitsgory writes “The material in this subsection does not admit adequate references in
the published literature.”)
To varying degrees, the above references make implicit and explicit use of 𝐺p𝐾q-equivariance
property of Kac-Moody duality, which was not known before the present work.

1.4.25. Beilinson-Drinfeld localization revisited. There is a 𝐺p𝐾q-equivariant localization functor:

g´𝑇 𝑎𝑡𝑒 Ñ 𝐷! pBun𝑙𝑒𝑣𝑒𝑙,𝑥

Loc : p 𝐺 q
obtained by duality from (1.4.6).
More generally, one can incorporate a level13 𝜅 as in S11 to obtain dual level 𝜅 strongly 𝐺p𝐾q-
equivariant functors:

ΓpBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 , ´q : 𝐷𝜅˚ pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 g𝜅 –mod
p𝑋qq Ñ p
𝑙𝑒𝑣𝑒𝑙,𝑥
Loc : p !
g𝜅 Ñ 𝐷𝜅`𝑇 𝑎𝑡𝑒 pBun𝐺 q » 𝐷𝜅! pBun𝑙𝑒𝑣𝑒𝑙,𝑥
𝐺 q
where the last isomorphism follows from integrality of the Tate level; we refer to S11 for the relevant
notions.
In particular, passing to 𝐺p𝑂q-invariants, one obtains a localization functor:

Loc𝐺p𝑂q : p
g𝜅 –mod𝐺p𝑂q Ñ 𝐷𝜅! pBun𝐺 q
that is a morphism of module categories for the spherical Hecke category of bi-𝐺p𝑂q-equivariant
𝜅-twisted 𝐷-modules on 𝐺p𝐾q.
A weaker version of this latter functor was considered in [BD1] S7.14. More precisely, it was
constructed at the triangulated level; higher coherence data regarding the Hecke action was not
considered in loc. cit.
Remark 1.4.14. In Beilinson-Drinfeld’s construction, semi-infinite linear algebra plays a major role:
cf. [BD1] 7.14.2-5. Our construction provides a conceptual explanation for this: the more natural
object is the global sections functor, whose categorical dual is the localization functor.
13At some points in this introduction, we assume the reader has some familiarity with standard notions from
Kac-Moody representations. Briefly, we remind that a level is a 𝐺-invariant symmetric bilinear form on g, and defines
a central extension of gpp𝑡qq in a standard way.
18 SAM RASKIN

Remark 1.4.15. To indicate the importance of the Hecke equivariance of Loc, recall its role in [BD1].
ˇ
For a 𝐺-oper ˇ
𝜒 on the curve 𝑋, which we recall is a 𝐺-local system with extra structure, Beilinson-
Drinfeld form a corresponding quotient V𝑐𝑟𝑖𝑡,𝜒 P p g𝑐𝑟𝑖𝑡 –mod𝐺p𝑂q of the vacuum representation via
the Feigin-Frenkel isomorphism (here the level is critical, cf. S11). Moreover, they show that this
quotient satisfies the local Hecke property (see loc. cit. Theorem 5.4.11, or [Ras1]). By applying
the localization functor, they deduce a global version of this result and thereby construct a Hecke
eigensheaf.
Remark 1.4.16. For another perspective on the Beilinson-Drinfeld localization functor (without
emphasis on Hecke symmetry), see [Roz].
1.4.26. Localization at critical level. In [FG1], Frenkel-Gaitsgory initiated an ambitious program to
study critical level Kac-Moody representations using ideas from local geometric Langlands. Many
of their conjectures resemble Beilinson-Bernstein-style localization theorems.
In [Ras7], we recently proved one of their outstanding conjectures for 𝐺𝐿2 using the (critical
level) 𝐺p𝐾q-action on p g𝑐𝑟𝑖𝑡 –mod constructed here. We refer to loc. cit. for more details on this
application.
Remark 1.4.17. This application was the genesis of this text: in writing [Ras7], we found that there
were a number of gaps in the literature that needed to be addressed.
1.4.27. Local geometric Langlands. Finally, we highlight that weak loop group actions play a dis-
tinguished role in the conjectural local geometric Langlands program.
Fix a level 𝜅 as above. Let 𝐺p𝐾q–mod𝜅 denote the category of DG categories with a strong 𝐺p𝐾q-
action with level 𝜅, cf. S11. For instance, for 𝜅 “ 0, this is the cateogry we previously denoted as
𝐺p𝐾q–mod.
Quantum local geometric Langlands predicts that there is something like an equivalence:

ˇ
𝐺p𝐾q–mod𝜅 « 𝐺p𝐾q–mod ˇ.
𝜅
For instance, if 𝜅 is generic, there is expected to be an honest equivalence of this type. We ignore
the difference between « and » below.
The basic feature of this (almost) equivalence is that the diagram:

𝐺p𝐾q–mod𝜅 ˇ
𝐺p𝐾q–mod 𝜅
´ˇ
p´q𝐺p𝐾q,𝑤 Whit
' w
DGCat𝑐𝑜𝑛𝑡
should (canonically) commute. Here 𝐺 ˇ is the Langlands dual group to 𝐺 and 𝜅 ˇ is a (suitably
normalized) dual level.
In the diagram above, the functor on the right is the Whittaker functor, well studied in local
and global geometric Langlands; see in [Ras6] in the local context.
More saliently, the functor on the left is defined as the composition:

p´q𝐺p𝐾q,𝑤
𝐺p𝐾q–mod𝜅 Ñ 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 ÝÝÝÝÝÝÑ DGCat𝑐𝑜𝑛𝑡 .
As the various terms here were not previously14 well-studied, this functor was previously not well-
understood, even at a formal level.
14For instance, the construction of the forgetful functor 𝐺p𝐾q–mod Ñ 𝐺p𝐾q–mod
𝑤𝑒𝑎𝑘 is the subject of S8, and is
somewhat involved.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 19

In this way, weak loop group actions play a defining role in local geometric Langlands.

Remark 1.4.18. For an introduction to the above ideas, we refer to [ABC` ].

1.5. Topological algebras and Harish-Chandra data. In S1.4, we introduced the theory of
loop group actions on categories and argued that it provides useful perspectives on Kac-Moody
representations.
However, the theory is a priori quite abstract. For instance, in concrete circumstances, it can be
quite difficult to construct a (strong, say) 𝐺p𝐾q-action on a category C: this was implicit in S1.4.20.

Remark 1.5.1. If C is a category of modules for a vertex algebra with Kac-Moody symmetry, we
should expect C to have a strong 𝐺p𝐾q-action (with suitable level).

Remark 1.5.2. We have seen in S1.4.27 that the (level 𝜅) strong 𝐺p𝐾q-action on p g𝜅 –mod plays an
important role in local geometric Langlands.
However, the construction of this 𝐺p𝐾q-action was quite abstract. What if we wish to prove
concrete results, say of the sort predicted by local geometric Langlands, about this 𝐺p𝐾q-action?
Certainly one needs to use classical, non-derived structures to be able to study the 𝐺p𝐾q-action
on p
g𝜅 –mod.

Our theories of topological DG algebras and Harish-Chandra data address these issues. We
describe salient parts of this theory below.

1.5.1. Classical Harish-Chandra data. What finite-dimensional theory are we trying to imitate for
loop groups?
Classically, suppose 𝐺 is an affine algebraic group and 𝐴 is an associative algebra. We suppose
𝐴 is classical, i.e., not DG.
If 𝐺 acts on 𝐴, then 𝐺 acts weakly on the category 𝐴–mod.
To upgrade this weak action to a strong action, it is equivalent to specify a Harish-Chandra
datum, i.e., a map:

𝑖:gÑ𝐴
that is a 𝐺-equivariant morphism of Lie algebras, and such that the induced adjoint action of g on
𝐴 is the infinitesimal action defined by the 𝐺-action on 𝐴.
For instance, if 𝐴 “ 𝑈 pgq with the adjoint 𝐺-action, we can take 𝑖 as the structural map
g Ñ 𝑈 pgq to obtain the strong 𝐺-action on g–mod. Or, if 𝑋 is a smooth affine variety with a 𝐺-
action, we can take 𝑖 as the composition g Ñ Γp𝑋, 𝑇𝑋 q Ñ Γp𝑋, 𝐷𝑋 q to obtain the strong 𝐺-action
on Γp𝑋, 𝐷𝑋 q–mod “ 𝐷p𝑋q.
We refer to [Neg] and [FG1] S20.4 for discussion in the classical (intrinsically 1-categorical)
context. I am not aware of a suitable reference in the DG setting, but it is not difficult to directly
deduce the above assertion from standard theory.

Remark 1.5.3. In summary: in the above setting, certain15 strong 𝐺-actions categories C “ 𝐴–mod
can be completely encoded in terms of classical abstract algebra.

15More precisely, those actions for which the forgetful functor 𝐴–mod Ñ Vect is given a weakly 𝐺-equivariant
structure; this is equivalent to specifying our initial 𝐺-action on the algebra 𝐴 itself.
20 SAM RASKIN

1.5.2. We desire a parallel theory suitable for loop groups.

There is an immediate difference: for loop groups, we need to work with topological algebras.
Indeed, a fundamental example should be the category gpp𝑡qq–mod, where 𝐴 “ 𝑈 pgpp𝑡qqq is the
completed enveloping algebra of gpp𝑡qq (cf. [Bei]).
Note that at the abelian categorical level, gpp𝑡qq–mod♡ consists only of discrete modules over
𝑈 pgpp𝑡qqq, i.e., the definition of the category of modules is sensitive to the topology on gpp𝑡qq.
Remark 1.5.4. The relevant class of topological algebras was introduced in [BD2] and [Bei]. In
the latter, they were given the name topological chiral algebras. In this text, we prefer the name
Ñ
b-algebras.
1.5.3. The theory of loop group actions on DG categories inherently involves derived categories.
Ñ
However, to the author’s knowledge, previous literature has avoided relating b-algebras and DG
categories.
Ñ
Instead, other works have only considered abelian categories of modules over b-algebras, and
then have passed to the corresponding derived categories of modules.
Unfortunately, this approach is inadequate for setting up a theory of Harish-Chandra data for
Ñ
loop groups: one needs a more direct connection between b-algebras and their corresponding derived
categories of modules.
Ñ
To overcome this issue, we develop an inherently derived theory of b-algebras and their modules
in S3, S2, and S4. (As in standard, we consider pro-complexes instead of topological vector spaces.)
Remark 1.5.5. Clausen-Scholze [Sch] have recently put forward a different theory of topological
˚
algebras in homotopical settings. Their theory in effect gives a derived version of the b-algebras
considered in [Bei].
˚
We do not consider b-algebras in this text, but one could formally define such a tensor structure
on ProVect (or pro-abelian groups, or pro-spectra) by taking [Bei] Corollary 1.1 as a definition
(using the constructions from our S3). It would be interesting to contrast the resulting theory with
the Clausen-Scholze approach, which uses direct topological methods rather than pro-objects (and
certainly produces an inequivalent theory).
1.5.4. Our theory is largely parallel to that of [Bei], but there are some discrepancies.
The major one relates to the finer points of the Frenkel-Gaitsgory definition of gpp𝑡qq–mod refer-
enced in S1.4.15. The issue is that the forgetful functor gpp𝑡qq–mod Ñ Vect is not conservative: it
sends some non-zero objects to zero. (It is conservative on the bounded below derived category.)
Therefore, it is not literally possible to think of general objects of gpp𝑡qq–mod as vector spaces
with extra structure (like an gpp𝑡qq-action).
Following [FG2], we refer to these sorts of phenomena as renormalization problems. The finer
points of the theory all relate to renormalization difficulties.

1.5.5. Finally, in S10, we introduce a suitable theory of Harish-Chandra data for loop groups.
Due to the complications above, the theory is quite technical, and there are additional technical
hypotheses that do not appear in the finite-dimensional setting.
Ñ
Still, at the end of the day, for a classical (i.e., non-DG) b-algebra 𝐴 (which we also assume
comes from a topological vector space), and equipped with a suitable renormalization datum (as
above), we can characterize suitable actions of 𝐺p𝐾q on the renormalized category 𝐴–mod𝑟𝑒𝑛 of
𝐴-modules in terms of morphisms 𝑖 : gpp𝑡qq Ñ 𝐴 satisfying some identities.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 21

In other words, we characterize the action of 𝐺p𝐾q on 𝐴–mod𝑟𝑒𝑛 using classical linear algebraic
data: an action of 𝐺p𝐾q on 𝐴 and a map 𝑖 : gpp𝑡qq Ñ 𝐴 satisfying some properties. Therefore,
the theory of Harish-Chandra data in S10 provides a way to study certain categories with 𝐺p𝐾q-
actions using purely 1-categorical methods. In this way, we can relate classical infinite-dimensional
representation theory and 𝐺p𝐾q-actions on categories. (See Remark 1.5.7 for one explicit situation
in which this principle applies.)
Remark 1.5.6. Our theory covers the following examples:
‚ The action of 𝐺p𝐾q of gpp𝑡qq–mod.
‚ The action of 𝐺p𝐾q on 𝐷! p𝑌 p𝐾qq for 𝑌 a smooth affine variety equipped with a 𝐺p𝐾q-
equivariant CDO (cf. [BD2] S3.9).
‚ Suitable extensions of the above including a level 𝜅.
More heuristically, we expect that given a “nice” vertex algebra V with Kac-Moody symmetry,
Ñ
our theory applies to the 𝐺p𝐾q-action on the b-algebra associated with V.
1.5.6. Applications. We now discuss applications of this material to other sources.

1.5.7. Critical level. Suppose 𝜅 “ 𝑐𝑟𝑖𝑡 is the so-called critical level, which we remind is ´ 12 times the
Killing form. In this case, Feigin-Frenkel showed that there is a large center Z of the corresponding
(twisted, completed) enveloping algebra.
In S11, we apply our theory of Harish-Chandra data to give a categorical realization of these
symmetries of the Feigin-Frenkel center.
The main result is Theorem 11.18.1. Roughly, this result says that p g𝑐𝑟𝑖𝑡 –mod has an action of
IndCoh˚ pSpf Zq, and that this action naturally commutes with the canonical (critical level) 𝐺p𝐾q-
action on pg𝑐𝑟𝑖𝑡 –mod.
Remark 1.5.7. Similar constructions are easy to perform in the finite dimensional setting. However,
the author is unaware of a simpler approach than the one presented here in the affine setting.
The basic issue is that we need to relate the center Z, which is defined using topological algebras
and studied using representation theory, to derived category constructions (as in the definition of
Ñ
𝐺p𝐾q-actions). This is what our theories of b-algebras and Harish-Chandra data were designed to
do.
Remark 1.5.8. A weaker version of the above construction was given in [FG1] S23. The construction
is loc. cit. was given before there was a good theory of 𝐺p𝐾q-actions on DG categories, so is
inherently weaker than the construction we give here. For instance, our construction manifestly
accounts for all higher homotopy coherence data, so is compatible with constructions on 𝐺p𝐾q-
categories.
Remark 1.5.9. This material is used in [Ras7], as mentioned in S1.4.26. As in Remark 1.4.17,
much of the present text arose from trying to fill in gaps in the literature regarding categorical
Ñ
symmetries for Kac-Moody algebras. In particular, this is true for our theories of b-algebras and
Harish-Chandra data; we needed the commuting actions of 𝐺p𝐾q and the critical center for the
methods of loc. cit.
1.5.8. BRST-style constructions. One sometimes finds the following situation in the physics liter-
ature.
One is given a vertex algebra V with a 𝐺p𝑂q-action and a morphism Vg,𝜅 Ñ V defining Kac-
Moody symmetry on V for some integral level 𝜅. One wishes to form the BRST reduction BRSTpVq
22 SAM RASKIN

of V with respect to this Kac-Moody action. However, this is only possible if 𝜅 “ ´𝑇 𝑎𝑡𝑒 “ 2 ¨ 𝑐𝑟𝑖𝑡.
What should be done at other levels? Ñ
In this case, there is a natural map pg𝜅 Ñ 𝐴pVq, where 𝐴pVq is the b-algebra controlling vertex
modules for V. This map satisfies the identities to be a Harish-Chandra datum. One should expect
the technical conditions of S10 to be satisfied, so 𝐺p𝐾q acts strongly on the DG category V–mod
(which is assumed suitably renormalized).
As the level is integral, we can ignore it. Therefore, V–mod𝐺p𝐾q is defined. Heuristically, it would
be modules over BRSTpVq if that reduction were defined. In this sense, the derived category of
modules over BRSTpVq is defined, even if there is an anomaly; it is the forgetful functor to Vect
that is missing. (Moreover, one should be able to extend this to work with factorization categories
in the sense of [Ras2], providing a substitute for the VOA structure on BRSTpVq.)

Remark 1.5.10. For one example in which the above principle BRSTpVq–mod « V–mod𝐺p𝐾q can
be made quite precise, see [Ras6]. Note that in loc. cit., every object of the relevant equivariant
category lives in cohomological degree ´8; i.e., it is essential to confront the finer points of the
homological algebra.

Remark 1.5.11. The issue described here occurs many places in the literature: see [CG], [GR1], and
[BLL` ] for some recent examples of different flavors.

Remark 1.5.12. We do not claim that this construction is suited for all purposes. For instance,
important invariants such as conformal blocks are not a priori defined for categories such as
V–mod𝐺p𝐾q , only for the VOA BRST (if it exists).

1.5.9. Weil representations and Coulomb branches. In [Ras8], we use the theory of Harish-Chandra
data to construct an analogue of the Weil representation for symplectic loop groups. We apply this
to construct Coulomb branches for symplectic representations of reductive groups,16 generalizing
the BFN construction [BFN] to the case when there is no Lagrangian subspace.

1.5.10. Harish-Chandra bimodules. Finally, we briefly want to draw the interested reader’s atten-
tion to the fact that S10 implicitly provides tools to study the category of affine Harish-Chandra
bimodules.
For a level 𝜅 of g, define HCaff
𝐺,𝜅 as End𝐺p𝐾q–mod𝜅 pp
g𝜅 –modq, i.e., as the monoidal DG category
of endomorphisms of p g𝜅 –mod considered as a category with a level 𝜅 𝐺p𝐾q-action (this notion is
defined in S11). By the results of S8, this category may also be calculated as p g𝜅 –mod𝐺p𝐾q,𝑤 , the
category of weak 𝐺p𝐾q-coinvariants.
The category HCaff𝐺,𝜅 plays a central role in quantum local geometric Langlands, but is difficult
to study explicitly. Theorem 11.18.1 amounts to a construction of a monoidal functor:

IndCoh! pOp𝐺ˇ q Ñ HCaff

𝐺,𝑐𝑟𝑖𝑡

so the proof must provide some basic study of the right hand side.
The main technical work in this study is implicit in S10. The careful reader will find that the
mosttechnical results in S10 are about categories p g𝜅 –mod𝐾,𝑤 for 𝐾 Ď 𝐺p𝐾q compact open, and
g𝜅 –mod𝐺p𝐾q,𝑤 “ HCaff
understanding these categories is an essential prerequisite to understanding p 𝐺,𝜅 .

16In the absence of a certain anomaly involving 𝜋 p𝐺q; cf. [Wit2] for 𝐺 “ 𝑆𝐿 .
4 2
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 23

1.6. Relation to older approaches. It is roughly fair to say that this text is an update of the
appendices to [FG1], incorporating modern homotopical techniques and working with unbounded
derived categories.
We remark that the extension to unbounded derived categories is essential in applications (see
already [FG2]), and the reader will observe that most of the difficulties that come up in our setting
exactly have to do with the difference between bounded below derived categories and unbounded
ones.
Many of our constructions are also close in spirit to [Pos1], although our perspective and emphasis
are somewhat different.

1.7. Leitfaden. Here are the (essential) logical dependencies.

S2 S5 S6

  ~
S3 S7

 
S4 S8

  
S9 S10


S11
Ñ
Briefly, S2-4 develops the theory of b-algebras and renormalization data. The theory of weak
group actions on categories is developed in S5 and S7, and is related to strong group actions in
S8. We apply these ideas to semi-infinite cohomology in S9. The material we need on ind-coherent
sheaves is developed in S6. Finally, S10 introduces Harish-Chandra data for group indschemes acting
Ñ
on b-algebras, and S11 gives an application at the critical level.

1.8. Conventions. We always work over the base field 𝑘 of characteristic zero.
We use higher categorical language without mention: by category, we mean 8-category in the
sense of [Lur2], and similarly for monoidal category and so on. We let Cat denote the category of
p8´q-categories, and Gpd denote the category of p8´qgroupoids. We also refer to S2.2 for some
essential notation used throughout the paper.
Similarly, by scheme we mean derived scheme over 𝑘 in the sense of [GR4], or spectral scheme
over 𝑘 in the sense of [Lur4]. Similarly, by indscheme, we mean what [GR3] calls DG indscheme.
Algebras of all flavors are assumed to be derived unless otherwise stated. We emphasize that when
we speak of DG objects or chain complexes of vector spaces or the like, we really understand
objects of suitable 8-categories, not explicit cochain models for them; we refer to [GR4] S1 for an
introduction to this way of thinking.
24 SAM RASKIN

For C a DG category, we let HomC pF, Gq P Vect denote the Hom-complex between objects, and
we let HomC pF, Gq P Gpd denote groupoid of maps in C regarded as an abstract category, i.e.,
forgetting the DG structure. We remind that Ω8 HomC pF, Gq “ HomC pF, Gq, where on the left hand
side we are regarding Vect as the 8-category of 𝑘-module spectra.
For a DG category C with a 𝑡-structure, we let 𝜏 ě𝑛 and 𝜏 ď𝑛 denote the truncation functors; we
use cohomological gradings throughout (as indicated by the use of superscripts).

1.9. Acknowledgements. We thank Sasha Beilinson, Dylan Butson, Justin Campbell, Gurbir
Dhillon, Dennis Gaitsgory, Nick Rozenblyum, and Harold Williams for a number of essential dis-
cussions related to this text.

2. Monoidal structures
! Ñ
2.1. In this section, we define monoidal structures b and b on ProVect. This material follows [Bei],
the appendices to [FG1], [GK], and [Pos1] Appendix D. The main difference with those sources is
that we work in the derived setting, which requires somewhat restructuring the usual definitions.

2.2. Notation. Let ProVect denote the pro-category of (the DG category) Vect; we refer to [Lur2]
S7.1.6 and [Lur4] SA.8.1 for details on pro-categories.
Remark 2.2.1. There are cardinality issues to keep in mind when working with a category such as
ProVect. Let us remind some relevant ideas from [Lur2] S5. A category is accessible if it satisfies
a certain hypothesis involving cardinalities (and is idempotent complete); for instance, compactly
generated categories are accessible, where the cardinality condition is the hypothesis that the sub-
category of compact objects is essentially small. We also remind that presentable accessible and
cocomplete (i.e., admitting colimits). This hypothesis is designed so the “naive” proof of the ad-
joint functor theorem (involving potentially large limits) goes through; in particular, presentable
categories admit limits.
Now for an accessible category C admitting finite limits, (e.g., C is presentable), PropCq𝑜𝑝 is
defined to be the category of accessible functors C Ñ Gpd preserving finite limits. Note that PropCq
is not presentable, so the adjoint functor theorem and its relatives do not apply.
By a DG category, we mean a stable (8-)category with a Vect𝑐 -module category structure where
the action is exact in each variable separately. Here Vect𝑐 Ď Vect is the subcategory of compact
objects (i.e., perfect objects, i.e., bounded complexes with finite dimensional cohomologies). By a
DG functor, we mean an exact functor compatible between the Vect𝑐 -module structures. We let
DGCat𝑏𝑖𝑔 denote the 2-category of such.
We let DGCat Ď DGCat𝑏𝑖𝑔 denote the 2-category of accessible DG categories under accessible DG
functors. Recall that DGCat𝑐𝑜𝑛𝑡 denotes the 2-category of cocomplete, presentable DG categories
under continuous functors.
For the set up of topological algebras, it would be more natural to work with spectra and stable
categories, but given our convention that we work over 𝑘, we stick to the language of DG categories.
We remind (c.f. [Lur3] S4.8.1) that DGCat𝑏𝑖𝑔 has a canonical symmetric monoidal structure with
unit Vect𝑐 . We denote this monoidal structure by b. If we worked in the spectral setting, functors
CbD Ñ E would be the same as functors C ˆ D Ñ E exact in each variable separately; in the DG
setting, they should be called bi-DG functors.
Similarly, DGCat𝑐𝑜𝑛𝑡 has a symmetric monoidal structure b such that functors C b D Ñ E are
the same as bi-DG functors that commute with colimits in each variable separately.
For F P C and G P D, we let F b G denote the induced object of CbD or C b D as appropriate.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 25

For C a compactly generated DG category, we let C𝑐 denote its subcategory of compact objects,
as in the case of Vect above.
2.3. Review of topological tensor products. Following the above references, we seek two tensor
! Ñ
products b and b on ProVect. Ignoring homotopy coherences issues for the moment, we recall the
basic formulae characterizing these two tensor products concretely.
Roughly, if 𝑉 “ lim𝑖 𝑉𝑖 , 𝑊 “ lim𝑗 𝑊𝑗 P ProVect are filtered limits with 𝑉𝑖 , 𝑊𝑗 P Vect, then:
!
𝑉 b 𝑊 “ lim 𝑉𝑖 b 𝑊𝑗 .
𝑖,𝑗
Ñ
The (non-symmetric) monoidal product b is characterized by the fact that it is a bi-DG functor,
and the functor:
Ñ
𝑉 b ´ : ProVect Ñ ProVect
commutes with limits, while the functor:
Ñ
𝑉 b ´ : Vect Ñ ProVect
commutes with colimits.
Explicitly, if 𝑊𝑗 “ colim𝑘 𝑊𝑗,𝑘 with 𝑊𝑗,𝑘 P Vect𝑐 , we have:
Ñ
𝑉 b 𝑊 “ lim colim 𝑉 b 𝑊𝑗,𝑘 .
𝑗 𝑘
(Clearly we should allow the indexing set for the terms “𝑘” to depend on 𝑗.)
These two tensor products are connected as follows. For 𝑉1 , 𝑉2 , 𝑊1 , 𝑊2 , there is a natural map:
! Ñ ! Ñ ! Ñ
p𝑉1 b 𝑉2 q b p𝑊1 b 𝑊2 q Ñ p𝑉1 b 𝑊1 q b p𝑉2 b 𝑊2 q. (2.3.1)
In particular, there is a natural map:
Ñ !
𝑉 b 𝑊 Ñ 𝑉 b 𝑊.
2.4. Topological tensor products in the derived setting. We now formally define the above
structures and characterize their categorical properties.
!
The tensor product b on ProVect is easy: as Ind of a monoidal category has a canonical tensor
product, so does Pro.
Ñ
2.5. To construct b, first note that PropVectq𝑜𝑝 is by definition the category HompVect, Gpdq of
accessible functors Vect Ñ Gpd. Any such functor factors canonically as:
𝐹 Oblv Ω8
Vect Ý
Ñ Vect ÝÝÝÑ Spectra ÝÝÑ Gpd
with 𝐹 a DG functor, i.e., ProVect “ HomDGCat pVect, Vectq𝑜𝑝 .
Notation 2.5.1. For 𝑉 P ProVect, we let 𝐹𝑉 denote the induced functor Vect Ñ Vect. Clearly
Ñ
𝐹𝑉 “ HomProVect p𝑉, ´q. Define 𝑉 b ´ : ProVect Ñ ProVect as the “partially-defined left adjoint”
to 𝐹𝑉 , i.e., for 𝑊 P Vect and 𝑈 P ProVect, we have functorial isomorphisms:
Ñ
HomProVect p𝑉 b 𝑊, 𝑈 q » HomVect p𝑊, 𝐹𝑉 p𝑈 qq. (2.5.1)
We extend this construction to general 𝑊 P ProVect by right Kan extension.
26 SAM RASKIN

We extend this construction to a monoidal structure by reinterpreting it as composition of

functors in HomDGCat pVect, Vectq. That is, we observe:

𝐹𝑉 ˝ 𝐹𝑊 » 𝐹 Ñ .
𝑉 b𝑊
The left hand side extends to the evident monoidal structure on HomDGCat pVect, Vectq𝑜𝑝 .
Ñ !
2.6. Comparison of tensor products. We now wish to give compatibilities between b and b.
Roughly, we claim that these form a “lax E2 ” structure.17
Let AlgpCatq denote the category of monoidal categories and lax monoidal functors, which we
Ñ
consider as a symmetric monoidal category under products. We claim that pProVect, bq is a com-
!
mutative algebra in this category with operation b. Note that this structure encodes the natural
transformations (2.3.1). (In S3.3, we give some simple consequences, and the reader may wish to
skip ahead.)
!
To construct this compatibility, note that if we write ProVect as EndDGCat pVectq𝑜𝑝 , then b cor-
responds to Day convolution. Then this follows from formal facts about Day convolution.
With respect to the symmetric monoidal structure b on DGCat, we have the internal Hom
functor:

HomDGCat pC, Dq P DGCat𝑏𝑖𝑔

which is the usual (DG) category of DG functors. If C is a monoidal DG category and D P
AlgpDGCat𝑐𝑜𝑛𝑡 q, then recall that HomDGCat pC, Dq has the usual Day convolution monoidal structure.
It is characterized by the fact that:

HomAlg𝑙𝑎𝑥 pDGCatq pE, HomDGCat pC, Dqq “ HomDGCat pCbE, Dq

where by Alg𝑙𝑎𝑥 pDGCatq we mean monoidal DG categories under lax monoidal functors.
It is straightforward to see that Day convolution has the property that the composition functor:

HomDGCat pC, DqbHomDGCat pD, Eq Ñ HomDGCat pC, Eq

is lax monoidal (assuming D, E P AlgpDGCat𝑐𝑜𝑛𝑡 , bq). This immediately implies our claim about
the monoidal structures on ProVect.
2.7. First, suppose that C, D P DGCat𝑐𝑜𝑛𝑡 with D compactly generated. Then observe that there
is a canonical bi-DG functor:
Ñ
´ b´
PropCq ˆ PropDq ÝÝÝÑ PropC b Dq
computed as follows.
Ñ
´ bG
If G P D𝑐 , then the induced functor ÝÝÝÑ: PropCq Ñ PropC b Dq is the right Kan extension of the
Ñ
functor ´ b G : C Ñ C b D Ď PropC b Dq. In general, for F P PropCq, the functor b´ : PropDq Ñ
PropC b Dq is computed by first left Kan extending the above functor from D𝑐 to D, and then right
Kan extending to PropDq.
This operation is functorial in the sense that for 𝐹 : C1 Ñ C2 P DGCat𝑐𝑜𝑛𝑡 , the diagram:
17For what follows, it is important to think of the E operad as Eb2 and not as the little discs operad: the laxness
2 1
evidently breaks the 𝑆𝑂p2q-symmetry.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 27

PropC1 q ˆ PropDq / PropC2 q ˆ PropDq

Ñ Ñ
´ b´ ´ b´
 
PropC1 b Dq / PropC2 b Dq

canonically commutes. Indeed, this follows immediately from:

Lemma 2.7.1. For 𝐹 : C Ñ D P DGCat𝑐𝑜𝑛𝑡 , Prop𝐹 q : PropCq Ñ PropDq commutes with limits and
colimits.
Proof. Prop𝐹 q tautologically commutes with limits. For the commutation with colimits, note that
𝐹 admits an (accessible) right adjoint 𝐺 by the adjoint functor theorem, so Prop𝐹 q admits the right
adjoint Prop𝐺q.


3. Modules and comodules

Ñ
3.1. We let Alg b denote the category of (associative, unital) algebras in ProVect with respect to
Ñ Ñ Ñ
b. We refer to objects of Alg b as b-algebras. We remark that (lower categorical analogues of) such
objects have been variously referred to as topological algebras or topological chiral algebras in the
literature. Ñ
In this section, we give basic definitions about modules over b-algebras. Note that we are exclu-
sively interested in discrete modules, i.e., modules in Vect, not in ProVect, and our notation will
always take this for granted.
Terminology 3.1.1. We generally use the term discrete to refer to objects of Vect Ď ProVect. For
Ñ
example, we say a b-algebra is discrete if its underlying object lies in Vect (in which case this
structure is equivalent to a usual associative DG algebra structure).
This should not be confused with the usage of this phrase in homotopy theory, where it is often
used for an object in the heart of a 𝑡-structure. In that setting, we prefer the term classical, so e.g.,
Ñ
a classical b-algebra is one whose underlying object lies in ProVect♡ .
3.2. Comparison with comonads. First, note that by construction, we have:
Ñ
Alg b » taccessible DG comonads on Vect.u𝑜𝑝
Here DG indicates that we have compatible comonad and DG functor structures; in the stable
setting, this would simply mean the underlying functor of our comonad is exact.
Ñ
Let 𝐴 be a b-algebra. Define 𝐴–mod𝑡𝑜𝑝 as 𝐴–modpProVectq. We define 𝐴–mod𝑛𝑎𝑖𝑣𝑒 to be the
“naive” category of discrete 𝐴-modules:

𝐴–mod𝑡𝑜𝑝 ˆ Vect.
ProVect
Ñ
That is, an object of 𝐴–mod𝑛𝑎𝑖𝑣𝑒 has an underlying vector space 𝑀 P Vect, an action map 𝐴𝑀 Ñ
𝑀 P ProVect, and the usual (higher) associativity data. (We use the notation “naive” by comparison
with the renormalization setting introduced below.)
By (2.5.1), if 𝑆 :“ 𝐹𝐴 is the comonad corresponding to 𝐴, we have a canonical equivalence:

𝐴–mod𝑛𝑎𝑖𝑣𝑒 » 𝑆–comod
28 SAM RASKIN

compatible with forgetful functors to Vect.

As a consequence, 𝐴–mod𝑛𝑎𝑖𝑣𝑒 is presentable and the forgetful functor Oblv : 𝐴–mod𝑛𝑎𝑖𝑣𝑒 Ñ Vect
is continuous, conservative, and, of course, comonadic.
Ñ
Remark 3.2.1. To conclude: the language of b-algebras is equivalent to the language of (DG)
comonads on Vect. Therefore, the wisdom in using this language may be reasonably questioned by
the reader; we use it here to connect to older work, and because pro-vector spaces are typically
nicer to describe than their corresponding comonads.
Ñ
3.3. Tensor products. We now spell out what the material of S2.6 means for b-algebras and
their modules. (We remind that S2.6 is a souped up version of (2.3.1), which may be more helpful
to refer to.)
Ñ
By S2.6, Alg b is symmetric monoidal with tensor product:

!
𝐴, 𝐵 ÞÑ 𝐴 b 𝐵.
Similarly, we have the bi-DG functor:

!
𝐴–mod𝑡𝑜𝑝 ˆ 𝐵–mod𝑡𝑜𝑝 Ñ 𝐴 b 𝐵–mod𝑡𝑜𝑝
!
p𝑀, 𝑁 q ÞÑ 𝑀 b 𝑁.
Clearly this induces a bi-DG functor:

!
𝐴–mod𝑛𝑎𝑖𝑣𝑒 ˆ 𝐵–mod𝑛𝑎𝑖𝑣𝑒 Ñ 𝐴 b 𝐵–mod𝑛𝑎𝑖𝑣𝑒
p𝑀, 𝑁 q ÞÑ 𝑀 b 𝑁.
This functor commutes with colimits in each variable separately, so induces:

!
𝐴–mod𝑛𝑎𝑖𝑣𝑒 b 𝐵–mod𝑛𝑎𝑖𝑣𝑒 Ñ p𝐴 b 𝐵q–mod𝑛𝑎𝑖𝑣𝑒 . (3.3.1)
To properly encode all higher categorical data, note that we have upgraded 𝐴 ÞÑ 𝐴–mod𝑛𝑎𝑖𝑣𝑒 to
Ñ
a contravariant lax symmetric monoidal functor from b-algebras to DGCat𝑐𝑜𝑛𝑡 .

3.4. Forgetful functors. Like every functor in DGCat𝑐𝑜𝑛𝑡 , 𝐹 is pro-representable, i.e., there is a
filtered projective system 𝑖 ÞÑ F𝑖 P 𝐴–mod𝑛𝑎𝑖𝑣𝑒 such that:

colim𝑖 Hom𝐴–mod𝑛𝑎𝑖𝑣𝑒 pF𝑖 , ´q “ Oblv .

In fact, we claim that lim𝑖 OblvpF𝑖 q P ProVect is the pro-vector space underlying 𝐴.
Indeed, let Φ : Vect Ñ 𝐴–mod𝑛𝑎𝑖𝑣𝑒 be the functor right adjoint to the forgetful functor. Note
that Oblv Φ “ 𝐹𝐴 (:“ HomProVect p𝐴, ´q), as is clear in the comonadic picture.
Then for any 𝑉 P Vect, we obtain:

HomProVect plim OblvpF𝑖 q, 𝑉 q “ colim HomVect pOblvpF𝑖 q, 𝑉 q “

𝑖 𝑖
colim HomVect pF𝑖 , Φp𝑉 qq “ Oblv Φp𝑉 q “ HomProVect p𝐴, ´q
𝑖
as desired.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 29

3.5. 𝑡-structures. Recall that ProVect has a natural 𝑡-structure with pProVectqď0 “ PropVectď0 q
and pProVectqě0 “ PropVectě0 q; we omit the parentheses in the sequel as there can be no confusion.
Ñ
In the remainder of the section, we will be interested in connective b-algebras, i.e., such algebras
𝐴 in ProVectď0 . Clearly this hypothesis is equivalent to the comonad 𝐹𝐴 being left 𝑡-exact.
From this latter description, we see that 𝐴–mod𝑛𝑎𝑖𝑣𝑒 carries a canonical 𝑡-structure such that
Oblv : 𝐴–mod𝑛𝑎𝑖𝑣𝑒 Ñ Vect is 𝑡-exact. Because Oblv commutes with colimits, this 𝑡-structure is
necessarily right complete.
3.6. Convergence. In order to formulate Proposition 3.7.1 below, we introduce the following
terminology.
For 𝑉 P ProVect, the convergent completion of 𝑉 is:

lim𝑛 𝜏 ě´𝑛 p𝑉 q P ProVect.

We say that 𝑉 is convergent if the natural map 𝑉 Ñ 𝑉p is an isomorphism. Note that 𝑉 is convergent
if and only if it lies in PropVect` q Ď PropVectq (or equivalently: 𝐹𝑉 is left Kan extended from Vect` ).
In particular, we obtain that connective convergent pro-vector spaces are (contravariantly) equiv-
Ñ
alent to left 𝑡-exact functors Vect` Ñ Vect` P DGCat. Under this dictionary, connective b-algebras
are the same as left 𝑡-exact (accessible) DG comonads on Vect` .
Ñ
Remark 3.6.1. If 𝐴 is a connective b-algebra, then its convergent completion 𝐴
p is as well, and
` » `
𝐴–mod𝑛𝑎𝑖𝑣𝑒 Ý
p Ñ 𝐴–mod𝑛𝑎𝑖𝑣𝑒 .
3.7. Comparison with categorical data. We have the following psychologically important re-
sult.
Proposition 3.7.1. The functor:
Ñ
tconvergent, connective b-algebrasu Ñ DGCat{Vect`
` ˘
𝐴 ÞÑ Oblv : 𝐴–mod` 𝑛𝑎𝑖𝑣𝑒 Ñ Vect
`

is fully-faithful. A DG category C with structural functor 𝐹 : C Ñ Vect` lies in the essential image
of this map if and only if:
‚ 𝐹 is conservative.
‚ C admits a (necessarily unique) 𝑡-structure for which 𝐹 is 𝑡-exact.
‚ Cě0 admits arbitrary colimits, and the functor 𝐹 : Cě0 Ñ Vectě0 preserves such colimits.
Under this equivalence, C is the bounded below derived category of its heart C♡ with 𝐹 the derived
Ñ
functor of its restriction C♡ Ñ Vect♡ if and only if the corresponding b-algebra 𝐴 is classical (i.e.,
lies in ProVect♡ ).
We first recall the following standard result about simplicial objects, see e.g. [Lur3] Remark
[Link].
Lemma 3.7.2. For a cosimplicial object F‚ in a stable (e.g., DG) category C, let Totď𝑛 F‚ be the
limit over the subcategory Δď𝑛 Ď Δ of totally ordered sets of cardinality ď 𝑛 ` 1.
Then for 𝑛 ą 0:

KerpTotď𝑛 F‚ Ñ Totď𝑛´1 F‚ q
is isomorphic to a direct summand of F𝑛 r´𝑛s.
30 SAM RASKIN

Proof of Proposition 3.7.1. First it is straightforward to see that 𝐴–mod`

𝑛𝑎𝑖𝑣𝑒 actually satisfies the
above conditions.
Suppose 𝐹 : C Ñ Vect` with the above properties is given. Clearly the 𝑡-structure on C is
bounded from below (i.e., C “ C` ), compatible with filtered colimits, and right complete. Clearly
the equivalence follows if we can show such 𝐹 is comonadic; the argument is well-known, but we
reproduce it here for convenience.
First, we claim 𝐹 |Cě0 : Cě0 Ñ Vectě0 commutes with arbitrary totalizations. By Lemma 3.7.2,
if F‚ is a cosimplicial diagram in C with F𝑖 P Cě0 for all 𝑖, then the totalization exists and is
calculated by:

𝜏 ď𝑛 Tot F‚ “ 𝜏 ď𝑛 Totď𝑛`1 F‚ .
Since Totď𝑛`1 is a finite limit, 𝑡-exactness of 𝐹 implies the claim.
Now observe that 𝐹 admits a left 𝑡-exact (possibly non-continuous) right adjoint 𝐺, as 𝐹 |Cě0
admits a left exact right adjoint. Then for any F P C, we have F P Cě´𝑁 , for 𝑁 large enough, so
p𝐺𝐹 q𝑛 pFq P Cě´𝑁 for any 𝑛, so the totalization Totpp𝐺𝐹 q‚`1 pFqq exists and is preserved by the
conservative functor 𝐹 , implying comonadicity.
It remains to show the compatibility with abelian categories. Suppose A is a 𝑘-linear abelian
category with a 𝑘-linear functor 𝐹 ♡ : A Ñ Vect♡ that is exact, continuous, conservative, and
accessible. Then there is a pro-object lim F𝑖 P PropAq (F𝑖 P A) corepresenting 𝐹 ♡ . It immediately
follows that this pro-object also corepresents the derived functor 𝐹 p:“ 𝑅𝐹 ♡ q : 𝐷` pAq Ñ Vect`
(because the functor this pro-system defines maps injectives in A♡ into Vect♡ ). By S3.4, this implies
Ñ
that the corresponding b- algebra has underlying object lim 𝐹 pF𝑖 q P ProVect. Because 𝐹 ♡ is exact,
𝐹 is 𝑡-exact, so 𝐹 pF𝑖 q P Vect♡ , implying lim 𝐹 pF𝑖 q P ProVect♡ .
Conversely, suppose 𝐴 is classical. Let Φ : Vect Ñ 𝐴–mod𝑛𝑎𝑖𝑣𝑒 denote the (possibly discontinuous)
right adjoint to the forgetful functor. For 𝑉 P Vect♡ ,18 Oblv Φp𝑉 q “ 𝐹𝐴 p𝑉 q “ HomProVect p𝐴, 𝑉 q P
Vect♡ , so Φp𝑉 q P 𝐴–mod♡ 𝑛𝑎𝑖𝑣𝑒 . Moreover, Φp𝑉 q is obviously injective in 𝐴–mod𝑛𝑎𝑖𝑣𝑒 in the sense that
for any F P 𝐴–mod𝑛𝑎𝑖𝑣𝑒 , HompF, Φp𝑉 qq “ HomVect pOblvpFq, 𝑉 q P Vectď0 . For F P 𝐴–mod♡
ě0
𝑛𝑎𝑖𝑣𝑒 , the
♡
map F Ñ Φ OblvpFq is a monomorphism in 𝐴–mod𝑛𝑎𝑖𝑣𝑒 (as it splits after applying Oblv), so such
there are “enough” injective objects, implying 𝐴–mod` 𝑛𝑎𝑖𝑣𝑒 is the bounded below derived category
of its heart. Moreover, this reasoning immediately shows that the forgetful functor is the derived
functor of its restriction to the hearts.


4. Renormalization
4.1. In our applications, the naive category 𝐴–mod𝑛𝑎𝑖𝑣𝑒 is typically not the one we want. For
example, the forgetful functor p g𝜅 –mod Ñ Vect is not conservative, so the above construction does
not recover the correct category p g𝜅 –mod, i.e., 𝑈 pp g𝜅 –mod.
g𝜅 q–mod𝑛𝑎𝑖𝑣𝑒 ‰ p
Following [FG2], a key role is played by renormalization of derived categories. We refer to loc. cit.,
[Gai7], and [Ras6] for introductions to this notion in the setting of Kac-Moody algebras. The basics
of the theory of ind-coherent sheaves also play an instructional role: see [Gai5] for an introduction.
In this section, we give an introduction to this formalism.

4.2. Renormalization data.

18If we worked with a general commutative ring 𝑘 P Ab♡ , 𝑉 should be an injective 𝑘-module.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 31

Ñ
Definition 4.2.1. A renormalization datum for a connective b-algebra 𝐴 is a DG category 𝐴–mod𝑟𝑒𝑛 P
»
DGCat𝑐𝑜𝑛𝑡 , equipped with a 𝑡-structure and an equivalence 𝜌 : 𝐴–mod` 𝑛𝑎𝑖𝑣𝑒 ÝÑ 𝐴–mod` 𝑟𝑒𝑛 P DGCat,
such that:
‚ 𝜌 is 𝑡-exact.
‚ 𝐴–mod𝑟𝑒𝑛 is compactly generated with compact generators lying in 𝐴–mod` 𝑟𝑒𝑛 .
‚ The 𝑡-structure on 𝐴–mod𝑟𝑒𝑛 is compactly generated: i.e., G P 𝐴–modě0 𝑟𝑒𝑛 if and only
Hom𝐴–mod𝑟𝑒𝑛 pF, Gq “ 0 for every compact F P 𝐴–modă0 𝑟𝑒𝑛 .
Ñ Ñ
We will also say 𝐴 is a renormalized b-algebra to mean 𝐴 is a connective b-algebra equipped
with a renormalization datum.
Remark 4.2.2. Once and for all, we emphasize: if 𝐴 is renormalized, it is in particular connective.
Remark 4.2.3. The subcategory 𝐴–mod𝑐𝑟𝑒𝑛 of compact objects in 𝐴–mod𝑟𝑒𝑛 embeds canonically into
𝐴–mod𝑛𝑎𝑖𝑣𝑒 as 𝐴–mod𝑐𝑟𝑒𝑛 Ď 𝐴–mod` `
𝑟𝑒𝑛 » 𝐴–mod𝑛𝑎𝑖𝑣𝑒 . It is immediate to see that a renormalization
datum is equivalent to a choice of such a subcategory satisfying some conditions.
Ñ Ñ
b
Remark 4.2.4. By Proposition 3.7.1, the category Alg𝑐𝑜𝑛𝑣,𝑟𝑒𝑛 of convergent, renormalized b-algebras
are equivalent to some categorical data: C P DGCat𝑐𝑜𝑛𝑡 , a continuous functor 𝐹 : C Ñ Vect, and
a 𝑡-structure on C such that 𝐹 is 𝑡-exact and conservative on C` , and the 𝑡-structure on C is
generated by eventually coconnective compact objects. (We remark that 𝐹 completely determines
the 𝑡-structure in this case.) As we will show in Theorem 4.6.1, this equivalence canonically upgrades
to a symmetric monoidal one.
Remark 4.2.5. Suppose C is a compactly generated DG category with a continuous functor 𝐹 : C Ñ
Vect. Then 𝐹 may be pro-represented by a pro-compact object. Comparing with S3.4, we see that
Ñ
this puts significant restrictions on which b-algebras 𝐴 admit renormalization data. (For example,
up to convergent completion, 𝐴 P ProVect must be expressible as a filtered limit of some discrete
𝐴-modules that are almost compact, i.e., whose truncations are compact in 𝐴–modě´𝑛 𝑛𝑎𝑖𝑣𝑒 for all 𝑛.)

4.3. Examples. We now give some examples of renormalization data.

We begin with examples when 𝐴 is discrete, i.e., 𝐴 P Vectď0 Ď ProVectď0 .
Example 4.3.1 (Ind-coherent sheaves). Let 𝐴 be a commutative, connective 𝑘-algebra (almost)
of finite type and let 𝑆 “ Specp𝐴q. Recall that IndCohp𝑆q (:“ IndpCohp𝑆qq) equipped with the
tautological embedding Cohp𝑆q ãÑ QCohp𝑆q` “ 𝐴–mod` defines a renormalization datum for 𝐴.
Example 4.3.2. More generally, if 𝐴 is a left coherent19 𝑘-algebra, then define 𝐴–mod𝑐𝑜ℎ Ď 𝐴–mod
as the subcategory of bounded complexes with finitely presented cohomologies. Then 𝐴–mod𝑟𝑒𝑛 :“
Indp𝐴–mod𝑐𝑜ℎ q defines a renormalization datum.
It is straightforward to show that this renormalization datum is initial among all renormalization
data for 𝐴.
Example 4.3.3 (Quasi-coherent sheaves). If 𝐴 is a connective associative 𝑘-algebra, then 𝐴–mod
itself underlies a renormalization datum if and only if 𝐴 is eventually coconnective, i.e., 𝐴 is also
bounded below as a complex of vector spaces. Indeed, recall that for renormalization data, there is
an assumption that the category be compactly generated by eventually coconnective objects, and
𝐴–mod is compactly generated by perfect ones.
19Recall that an algebra is left coherent if it is connective; the category 𝐴–mod
𝑐𝑜ℎ defined below is actually a DG
category, i.e., it is closed under cones; and 𝜏 ě´𝑛 𝐴 P 𝐴–mod𝑐𝑜ℎ for all 𝑛. For example, this is the case if 𝐴 is left
Noetherian.
32 SAM RASKIN

We now give some examples involving honestly topological algebras.

Example 4.3.4. Suppose 𝑆 “ colim𝑖 𝑆𝑖 is an ind-affine indscheme of ind-finite type. Then IndCohp𝑆q
is naturally a renormalization for the pro-algebra of functions on 𝑆. Indeed, this is a special case
of Example 4.3.5.
Example 4.3.5 (Pro-algebras). Suppose 𝑖 ÞÑ 𝐴𝑖 P AlgpVectď0 q is a projective system of algebras.
! Ñ
Let 𝐴 “ lim𝑖 𝐴𝑖 P PropVectq. Then 𝐴 is a b-algebra, and a posteriori a b-algebra.
Suppose that:
‚ 𝐴𝑖 is left coherent.
‚ Each structural map 𝜙𝑖𝑗 : 𝐴𝑖 Ñ 𝐴𝑗 is surjective on 𝐻 0 with finitely generated kernel.
Let 𝐴𝑖 –mod𝑟𝑒𝑛 be as in Example 4.3.2. Note that our assumptions imply that restriction along
𝜙𝑖𝑗 maps 𝐴𝑗 –mod𝑐𝑜ℎ to 𝐴𝑖 –mod𝑐𝑜ℎ . By ind-extension, we obtain 𝑡-exact functors 𝐴𝑗 –mod𝑟𝑒𝑛 Ñ
𝐴𝑖 –mod𝑟𝑒𝑛 .
Then define:

𝐴–mod𝑟𝑒𝑛 :“ colim 𝐴𝑖 –mod𝑟𝑒𝑛 P DGCat𝑐𝑜𝑛𝑡 .

𝑖
Here the structural functors are the above functors. We claim that 𝐴–mod𝑟𝑒𝑛 naturally defines a
renormalization datum.
As noted above, these functors preserve compact objects, so 𝐴–mod𝑟𝑒𝑛 is compactly generated.
Moreover, by [Ras6] Lemma 5.4.3 (1), there is a canonical 𝑡-structure on 𝐴–mod𝑟𝑒𝑛 such that
each functor res𝑖 : 𝐴𝑖 –mod𝑟𝑒𝑛 Ñ 𝐴–mod𝑟𝑒𝑛 is 𝑡-exact. This 𝑡-structure is tautologically compactly
generated and right complete.
Moreover, there is a canonical functor Oblv : 𝐴–mod𝑟𝑒𝑛 Ñ Vect P DGCat𝑐𝑜𝑛𝑡 pro-represented by
the object:

lim res𝑖 𝜏 ě´𝑛 𝐴𝑖 P Prop𝐴–mod𝑟𝑒𝑛 q.

𝑖,𝑛
It is straightforward to show that each composition Oblv ˝ res𝑖 : 𝐴𝑖 –mod𝑟𝑒𝑛 Ñ Vect is the canonical
forgetful functor on 𝐴𝑖 –mod𝑟𝑒𝑛 . It immediately follows that Oblv is 𝑡-exact.
Moreover, we claim that Oblv is conservative on bounded below objects. Indeed, in the above
pro-system, all objects are connective and all structural maps are surjective on 𝐻 0 . As the objects
res𝑖 𝐻 0 p𝐴𝑖 q tautologically generate 𝐴–mod♡ 𝑟𝑒𝑛 under colimits, this implies the claim.
Finally, it suffices to note that at the level of bounded below derived categories, this functor Oblv
defines 𝐴 under the dictionary of Proposition 3.7.1: indeed, Oblv of this pro-generator is manifestly
the Postnikov completion of 𝐴 in ProVect.
Remark 4.3.6. In Example 4.3.5, compact objects in 𝐴–mod𝑟𝑒𝑛 are closed under truncations.
Example 4.3.7. This example appears somewhat in the wrong place: it uses some terminology from
S4.4, and is really motivated by Example 4.3.8.
Ñ
Suppose 𝐴 is a connective b-algebra and that we are given a morphism 𝜙 : 𝐴0 Ñ 𝐴 of connective
Ñ
20
b-algebras such that the forgetful functor 𝐴–mod` `
𝑛𝑎𝑖𝑣𝑒 Ñ 𝐴0 –mod𝑛𝑎𝑖𝑣𝑒 is monadic.
Denote this monad by 𝑇 . Now suppose moreover that the composition:
𝑇 𝜌
𝐴0 –mod` Ñ 𝐴0 –mod`
𝑛𝑎𝑖𝑣𝑒 Ý 𝑛𝑎𝑖𝑣𝑒 ãÑ 𝐴0 –mod𝑟𝑒𝑛
20In particular, this functor admits a left adjoint. So if 𝐴 is discrete and 𝐴 “ 𝑘, this forces 𝐴 to be eventually
0
coconnective.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 33

renormalizes in the sense of S4.4. (E.g., this is automatic if 𝐴0 –mod𝑟𝑒𝑛 is given by Example 4.3.5.)
Then 𝑇 clearly induces a monad on 𝐴0 –mod𝑟𝑒𝑛 , and 𝐴–mod𝑟𝑒𝑛 :“ 𝑇 –modp𝐴0 –mod𝑟𝑒𝑛 q obviously
defines a renormalization datum for 𝐴.
Example 4.3.8 (Tate Lie algebras). Suppose h P ProVect♡ is a Tate Lie algebra. By this, we mean
that the dual Tate vector space h_ P ProVect is given a coLie algebra structure with respect to the
!
b symmetric monoidal structure. Recall that in this case, h necessarily admits an open profinite
dimensional subalgebra h0 Ď h, where these hypotheses force h0 “ lim𝑖 h𝑖 for h𝑖 ranging over the
finite dimensional Lie algebra quotients of h0 .
(For example, we might have h “ gpp𝑡qq for finite dimensional g; then h0 may be taken as grr𝑡ss
and h𝑖 “ grr𝑡ss{𝑡𝑖 grr𝑡ss.)
Then 𝐴0 “ 𝑈 ph0 q :“ lim𝑖 𝑈 ph𝑖 q satisfies the hypotheses of Example 4.3.5 (c.f. Example 4.4.4
regarding renormalization of the monad). Note that each 𝑈 ph𝑖 q–mod𝑟𝑒𝑛 “ 𝑈 ph𝑖 q–mod here, so
objects restrictions of modules 𝑈 ph𝑖 q give compact generators of 𝑈 ph0 q–mod𝑟𝑒𝑛 .
Moreover, 𝐴 “ 𝑈 phq the completed enveloping algebra of h, 𝐴0 Ñ 𝐴 satisfies the hypotheses of
Example 4.3.7 (say, by the PBW theorem). In particular, we obtain 𝑈 phq–mod𝑟𝑒𝑛 .
Following Gaitsgory, we denote these DG categories by h0 –mod and h–mod, leaving renormaliza-
tion out of the notation.
Note that the construction of h–mod recovers the format of [FG2] S23. Indeed, unwinding the
constructions, we find that compact generators are given by inducing trivial modules from k𝑖 to h
for k𝑖 :“ Kerph0 Ñ h𝑖 q.
Example 4.3.9. Renormalization data is given for the affine W-algebra in [Ras6]: the compact
generators are denoted W𝑛𝜅 in loc. cit. Outside of the Virasoro case, this example does not fit into
any of the above patterns. (This is closely related to the fact that the W-algebra chiral algebras
are generally neither commutative nor chiral envelopes.)
Ñ
4.4. Construction of functors. Let 𝐴 be a renormalized b-algebra.
Suppose C P DGCat𝑐𝑜𝑛𝑡 and that we are given a DG functor 𝐹 : 𝐴–mod`
𝑟𝑒𝑛 » 𝐴–mod𝑛𝑎𝑖𝑣𝑒 Ñ C.
`

Definition 4.4.1. 𝐹 renormalizes if it is left Kan extended from 𝐴–mod𝑐𝑟𝑒𝑛 .

For 𝐹 as above (not necessarily assumed to renormalize), we define 𝐹𝑟𝑒𝑛 : 𝐴–mod𝑟𝑒𝑛 Ñ C as
the ind-extension of 𝐹 |𝐴–mod𝑐𝑟𝑒𝑛 . Note that 𝐹𝑟𝑒𝑛 |𝐴–mod`
𝑟𝑒𝑛
is the left Kan extension of 𝐹 |𝐴–mod𝑐𝑟𝑒𝑛 ;
therefore, 𝐹 renormalizes if and only if the natural map 𝐹𝑟𝑒𝑛 |𝐴–mod` 𝑟𝑒𝑛
Ñ 𝐹 is an isomorphism.
Suppose now that C admits a 𝑡-structure compatible with filtered colimits, that 𝐹 is 𝑡-exact, and
that 𝐹 |𝐴–modě0
𝑟𝑒𝑛
commutes with filtered colimits.
Warning 4.4.2. It is not true in this generality that 𝐹 necessarily renormalizes: 𝐹𝑟𝑒𝑛 may fail to be
(left) 𝑡-exact. (See Counterexample 4.5.4.)
However, we claim:

𝜏 ě0 𝐹𝑟𝑒𝑛 “ 𝜏 ě0 𝐹 (4.4.1)
when restricted to 𝐴–mod`
𝑟𝑒𝑛 .
Indeed, for F P 𝐴–mod`
𝑟𝑒𝑛 , write F “ colim𝑖 F𝑖 with F𝑖 compact. Then:

𝜏 ě0 𝐹𝑟𝑒𝑛 pFq “ 𝜏 ě0 colim 𝐹 pF𝑖 q “ colim 𝐹 p𝜏 ě0 F𝑖 q “ 𝐹 p𝜏 ě0 colim F𝑖 q “ 𝐹 p𝜏 ě0 Fq.

𝑖 𝑖 𝑖
There are two general settings in which 𝐹 does renormalize.
34 SAM RASKIN

Example 4.4.3. If the 𝑡-structure on C is left separated, then (4.4.1) clearly implies that 𝐹 renor-
malizes.
Example 4.4.4. Suppose merely that 𝐹 is left 𝑡-exact (or left 𝑡-exact up to shift) and that compact
objects of 𝐴–mod𝑟𝑒𝑛 are closed under truncations. Then we claim that 𝐹 renormalizes. Indeed,
then every F P 𝐴–modě0 𝑟𝑒𝑛 can be written as a filtered colimit F “ colim𝑖 F𝑖 with F𝑖 compact and in
𝐴–modě0𝑟𝑒𝑛 : write F as a filtered colimit of arbitrary compacts and then apply 𝜏 ě0 . Then we obtain:
»
𝐹𝑟𝑒𝑛 pFq “ colim 𝐹 pF𝑖 q Ý
Ñ 𝐹 pFq
𝑖
by assumption that 𝐹 commutes with filtered colimits in 𝐴–modě0
𝑟𝑒𝑛 .
Example 4.4.5 (Forgetful functors). The forgetful functor Oblv : 𝐴–mod`𝑛𝑎𝑖𝑣𝑒 Ñ Vect renormalizes
to give a functor Oblv𝑟𝑒𝑛 : 𝐴–mod𝑟𝑒𝑛 Ñ Vect by Example 4.4.3. In what follows, we typically
abbreviate the notation Oblv𝑟𝑒𝑛 to simply Oblv. (Although we call this functor forgetful, it is not
generally conservative.)
Example 4.4.6 (Identity functor). The embedding 𝐴–mod`𝑛𝑎𝑖𝑣𝑒 ãÑ 𝐴–mod𝑛𝑎𝑖𝑣𝑒 renormalizes to give
a continuous functor id𝑟𝑒𝑛 : 𝐴–mod𝑟𝑒𝑛 Ñ 𝐴–mod𝑛𝑎𝑖𝑣𝑒 P DGCat𝑐𝑜𝑛𝑡 , again by Example 4.4.3.
4.5. Morphisms. We have the following notion of compatibility between algebra morphisms and
renormalization data.
Ñ Ñ
Definition 4.5.1. A morphism of renormalized b-algebras is a map 𝑓 : 𝐴 Ñ 𝐵 of b such that
the (𝑡-exact) functor Oblv : 𝐵–mod` `
𝑛𝑎𝑖𝑣𝑒 Ñ 𝐴–mod𝑛𝑎𝑖𝑣𝑒 Ď 𝐴–mod𝑟𝑒𝑛 renormalizes to a functor
Oblv “ Oblv𝑟𝑒𝑛 : 𝐵–mod𝑟𝑒𝑛 Ñ 𝐴–mod𝑟𝑒𝑛 .
Ñ
b
We let Alg𝑟𝑒𝑛 denote the category of renormalized algebras and such morphisms.
Remark 4.5.2. We emphasize that this is a property, not a structure, for the underlying map of
Ñ
b-algebras.
Ñ
Example 4.5.3. Example 4.4.5 says that the unit map 𝑘 Ñ 𝐴 is a morphism of renormalized b-
algebras. More generally, this is true for any map from an eventually coconnective algebra with the
“trivial” renormalization from Example 4.3.3.
Counterexample 4.5.4. Let 𝐴 be a (discrete) almost finite type, eventually coconnective commuta-
tive 𝑘-algebra with 𝑆 “ Specp𝐴q singular.
Take 𝐴–mod𝑟𝑒𝑛1 “ IndCohp𝑆q and 𝐴–mod𝑟𝑒𝑛2 “ QCohp𝑆q, and let us pedantically write 𝐴𝑟𝑒𝑛1 ,
𝐴𝑟𝑒𝑛2 for the corresponding renormalized algebras. Then the identity map for 𝐴 defines a morphism
𝐴𝑟𝑒𝑛1 Ñ 𝐴𝑟𝑒𝑛2 of renormalized algebras, but not a morphism 𝐴𝑟𝑒𝑛2 Ñ 𝐴𝑟𝑒𝑛1 .
4.6. Tensor products. We now revisit the material of S3.3 in the presence of renormalizations.
Ñ
So suppose 𝐴 and 𝐵 are renormalized b-algebras.
!
Then we claim that 𝐴–mod𝑟𝑒𝑛 b 𝐵–mod𝑟𝑒𝑛 defines a renormalization datum for 𝐴 b 𝐵.
More precisely, define:
! !
𝐴 b 𝐵–mod𝑐𝑟𝑒𝑛 Ď 𝐴 b 𝐵–mod𝑛𝑎𝑖𝑣𝑒
as the DG subcategory Karoubi generated by the essential image of the composition:

(3.3.1) !
𝐴–mod𝑐𝑟𝑒𝑛 ˆ 𝐵–mod𝑐𝑟𝑒𝑛 Ñ 𝐴–mod𝑛𝑎𝑖𝑣𝑒 b 𝐵–mod𝑛𝑎𝑖𝑣𝑒 ÝÝÝÝÑ 𝐴 b 𝐵–mod𝑛𝑎𝑖𝑣𝑒 .
! !
Now define 𝐴 b 𝐵–mod𝑟𝑒𝑛 as Indp𝐴 b 𝐵–mod𝑐𝑟𝑒𝑛 q.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 35

! !
Theorem 4.6.1. (1) 𝐴 b 𝐵–mod𝑟𝑒𝑛 is a renormalization datum for 𝐴 b 𝐵.21
(2) The natural functor:
!
𝐴–mod𝑟𝑒𝑛 b 𝐵–mod𝑟𝑒𝑛 Ñ 𝐴 b 𝐵–mod𝑟𝑒𝑛
is an equivalence.
Lemma 4.6.2. Suppose C, D1 , D2 P DGCat𝑐𝑜𝑛𝑡 have 𝑡-structures compatible with filtered colimits
and 𝐹 : D1 Ñ D2 P DGCat𝑐𝑜𝑛𝑡 is a functor.
Recall that C b D𝑖 admits a canonical 𝑡-structure with pC b D𝑖 qď0 generated under colimits by
objects F b G for F P Cď0 and G P Dď0 𝑖 .
(1) If 𝐹 is right 𝑡-exact, then so is idC b𝐹 : C b D1 Ñ C b D2 .
(2) If the 𝑡-structure on C is compactly generated and 𝐹 is left 𝑡-exact, then idC b𝐹 is left
𝑡-exact.
(3) Under the assumptions of (2), if the 𝑡-structure on C is right complete and 𝐹 |Dě0 is con-
1
servative, then idC b𝐹 |pCbD1 qě0 is conservative.
Proof. (1) is immediate. (2) is shown e.g. in [Ras6] Lemma B.6.2, but we recall the argument as it
is used also for (3).
Let F P Cď0 be compact. Then F defines a continuous functor DF :“ HomC pF, ´q : C Ñ Vect.
We can tensor to obtain:

DF b idD𝑖 : C b D𝑖 Ñ D𝑖 .
As in the proof of [Ras6] Lemma B.6.2, if F P Cď0 , then this functor is left 𝑡-exact, and conversely,
G P C b D𝑖 lies in cohomological degrees ě 0 if and only if DF b idD𝑖 pGq P Dě0 𝑖 for each such F.
These facts immediately imply (2).
Now for (3), suppose G P pC b D1 qě0 with pidC b𝐹 qpGq “ 0. Then for any F as above, we claim:

pDF b idD1 qpGq “ 0 P D1 .

Indeed, this object lies in degrees ě 0, so it suffices to show that 𝐹 applied to it is zero. Then:

𝐹 pDF b idD1 qpGq “ pDF b 𝐹 qpGq “ DFpidC b𝐹 qpGq “ 0.

Now right completeness of the (compactly generated) 𝑡-structure on C is equivalent to C being
compactly generated by objects F of the above type (and their shifts), so this implies that G “ 0
as desired.

Proof of Theorem 4.6.1. Note that 𝐴–mod𝑟𝑒𝑛 b 𝐵–mod𝑟𝑒𝑛 admits a canonical 𝑡-structure, as in the
statement of Lemma 4.6.2. This 𝑡-structure is obviously compactly generated by objects bounded
from below, since this is true for each of the tensor factors.
By Lemma 4.6.2, the functor:

Oblv “ Oblv𝐴 b Oblv𝐵 : 𝐴–mod𝑟𝑒𝑛 b 𝐵–mod𝑟𝑒𝑛 Ñ Vect

is 𝑡-exact and conservative on p𝐴–mod𝑟𝑒𝑛 b 𝐵–mod𝑟𝑒𝑛 qě0 .22
! !
21C.f. Remark 4.2.3: because 𝐴 b 𝐵–mod𝑐𝑟𝑒𝑛 is tautologically embedded into 𝐴 b 𝐵–mod`
𝑛𝑎𝑖𝑣𝑒 , this is a property,
not a structure.
22Note that Lemma 4.6.2 as formulated should be applied to the functor:
36 SAM RASKIN

Ñ
By Remark 4.2.4, there is some convergent, connective b-algebra 𝐶 such that 𝐴–mod𝑟𝑒𝑛 b
𝐵–mod𝑟𝑒𝑛 with its forgetful functor defines a renormalization datum for 𝐶.
!
Since the forgetful functor 𝐴–mod𝑟𝑒𝑛 b 𝐵–mod𝑟𝑒𝑛 Ñ Vect lifts to 𝐴 b 𝐵–mod𝑛𝑎𝑖𝑣𝑒 (by Example
! Ñ
4.4.6 and S3.3), we have a canonical map 𝐴 b 𝐵 Ñ 𝐶 of b-algebras. To prove the theorem, it
!
suffices to show that this map realizes 𝐶 as the convergent completion of 𝐴 b 𝐵.
For this, suppose 𝑖 ÞÑ F𝑖 P 𝐴–mod𝑐𝑟𝑒𝑛 and 𝑗 ÞÑ G𝑗 P 𝐵–mod𝑐𝑟𝑒𝑛 pro-represent the forgetful
functors. Clearly lim𝑖,𝑗 F𝑖 b G𝑗 P Prop𝐴–mod𝑟𝑒𝑛 b 𝐵–mod𝑟𝑒𝑛 q pro-represents the forgetful functor to
vector spaces. As in S3.4, the object:

lim OblvpF𝑖 b G𝑗 q
𝑖,𝑗

is canonically isomorphic to 𝐶 P ProVect. But we can calculate this object as:23

lim OblvpF𝑖 b G𝑗 q “ lim colim Hom𝐴–mod𝑟𝑒𝑛 b𝐵–mod𝑟𝑒𝑛 pF𝑘 b Gℓ , F𝑖 b G𝑗 q “

𝑖,𝑗 𝑖,𝑗 𝑘,ℓ
´ ¯
lim colim Hom𝐴–mod𝑟𝑒𝑛 pF𝑘 , F𝑖 q b Hom𝐵–mod𝑟𝑒𝑛 pGℓ , G𝑗 q “
𝑖,𝑗 𝑘,ℓ
!
lim OblvpF𝑖 q b lim OblvpG𝑗 q.
𝑖 𝑗
!
This last term is the b-tensor product of the convergent completions of 𝐴 and 𝐵 respectively,
giving the claim.

Ñ
b
This construction obviously equips Alg𝑟𝑒𝑛 with a unique symmetric monoidal structure such that
Ñ
the forgetful functor to Alg b is symmetric monoidal. For this symmetric monoidal structure, the
functor:
Ñ
b
Alg𝑟𝑒𝑛 Ñ DGCat𝑐𝑜𝑛𝑡
𝐴 ÞÑ 𝐴–mod𝑟𝑒𝑛
is symmetric monoidal by construction.

5. Weak actions of group schemes

Ñ
5.1. In this section, we begin a study of action of (suitable) group indschemes 𝐻 on b-algebras
and on categories.
We will explain, following Gaitsgory, that (under suitable hypotheses) there is a naive notion of
weak 𝐻-action on a DG category, and a less naive notion, which we call genuine weak actions.
The bulk of this section is devoted to developing the notion of genuine actions when 𝐻 is a
classical affine group scheme. With that said, this section begins with a general discussion of naive
Ñ
actions on categories and b-algebras in the case of general ind-affine group indschemes.

id b Oblv : 𝐴–mod𝑟𝑒𝑛 b 𝐵–mod𝑟𝑒𝑛 Ñ 𝐴–mod𝑟𝑒𝑛 .

23The second equality is a general fact about maps out of external products of two compact objects.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 37

Ñ
5.2. Topological bialgebras. A b-bialgebra 𝐵 is a coalgebra 𝐵 in the symmetric monoidal cat-
Ñ ! Ñ
egory pAlg b , bq. In particular, such a 𝐵 is equipped with an b-algebra structure and is equipped
! Ñ
with a coproduct ∆ : 𝐵 Ñ 𝐵 b 𝐵 that is a morphism of b-algebras.
Ñ
There is a natural notion of coaction of such a 𝐵 on a b-algebra 𝐴. Here we have a coaction
! Ñ
map coact : 𝐴 Ñ 𝐵 b 𝐴, which is a map of b-algebras (and satisfies higher compatibilities with ∆
and so on).
! !
Variant 5.2.1. A b-bialgebra is a bialgebra in the symmetric monoidal category pProVect, bq. Any
Ñ
such object has an underlying b-bialgebra structure.
! !
Note that in the b-setting, commutative and cocommutative b-bialgebra structures have evident
Ñ Ñ
meaning, while in the b-setting, only cocommutative b-bialgebra structures make sense.
5.3. In the above setting, note that 𝐵–mod𝑛𝑎𝑖𝑣𝑒 P DGCat𝑐𝑜𝑛𝑡 inherits a canonical monoidal DG
structure. For example, the monoidal operation is given by:

S3.3 ! Δ˚
𝐵–mod𝑛𝑎𝑖𝑣𝑒 b 𝐵–mod𝑛𝑎𝑖𝑣𝑒 ÝÝÑ 𝐵 b 𝐵–mod𝑛𝑎𝑖𝑣𝑒 ÝÝÑ 𝐵–mod𝑛𝑎𝑖𝑣𝑒
where ∆˚ is restriction of module structures along the map ∆.
Similarly, if 𝐵 coacts on 𝐴, then 𝐵–mod𝑛𝑎𝑖𝑣𝑒 acts on 𝐴–mod𝑛𝑎𝑖𝑣𝑒 .
Ñ
b
5.4. Now suppose that 𝐵 is given a renormalization datum. Recall from S4.6 that Alg𝑟𝑒𝑛 is a
symmetric monoidal category.
Therefore, it makes sense to say that a bialgebra structure on 𝐵 is compatible with the renor-
malization datum on 𝐵: this means that the counit and comultiplication maps are morphisms of
Ñ Ñ
b
renormalized b-algebras. Similarly, for 𝐴 P Alg𝑟𝑒𝑛 , we may speak of a coaction of 𝐵 on 𝐴 being
compatible with the given renormalization data: this means the coaction data makes 𝐴 a comodule
Ñ
b
for 𝐵 in the symmetric monoidal category Alg𝑟𝑒𝑛 .
In such cases, 𝐵–mod𝑟𝑒𝑛 inherits a canonical monoidal structure and 𝐵–mod𝑟𝑒𝑛 acts on 𝐴–mod𝑟𝑒𝑛 .
5.5. Group setting. Now suppose that 𝐻 is an ind-affine group indscheme. We suppose 𝐻 is
reasonable in the sense of [BD1] (or S6.8 below): that is, 𝐻 “ colim 𝐻𝑖 for 𝐻𝑖 Ď 𝐻 eventually
coconnective quasi-compact quasi-separated subschemes24 with all maps 𝐻𝑖 Ñ 𝐻𝑗 almost finitely
presented.
!
Then 𝐵 “ Funp𝐻q :“ lim𝑖 Γp𝐻𝑖 , O𝐻𝑖 q P ProVect is a commutative b-bialgebra, and in particular
Ñ
inherits a b-bialgebra structure.
Ñ Ñ
We say that 𝐻 naively acts on 𝐴 P Alg b if 𝐵 coacts on 𝐴. We let Alg b,𝐻ñ denote the category
Ñ Ñ
of b-algebras with naive 𝐻-actions (i.e., the category of 𝐵-comodules in Alg b ).
5.6. Naive group actions on categories. Assume in the above notation that each of the (com-
mutative) algebras Γp𝐻𝑖 , O𝐻𝑖 q are coherent, as in Example 4.3.2. Then 𝐵 admits a canonical renor-
malization as in loc. cit. We define IndCoh˚ p𝐻q :“ 𝐵–mod𝑟𝑒𝑛 .
Remark 5.6.1. In S6, we will define IndCoh˚ in much greater generality. However, this elementary
definition coincides in the present setting.
24For emphasis: the 𝐻 may not necessarily be group subschemes.
𝑖
38 SAM RASKIN

Remark 5.6.2. The notation is taken from [Ras3] (see also [Gai7]), to which we refer for an expla-
nation. The main purpose of this notation is to remind us that to avoid the pitfalls inherent in
working with IndCoh in the infinite type setting.
Example 5.6.3. Suppose that 𝐻 is the loop group 𝐺p𝐾q for 𝐺 an affine algebraic group.25 By
[GR3], there is a canonical equivalence QCohp𝐺p𝐾qq » IndCoh˚ p𝐺p𝐾qq defined as such. But we
note that this equivalence uses the compact open subgroup 𝐺p𝑂q in an essential way: the functor
Oblv : IndCoh˚ p𝐺p𝐾qq Ñ Vect, which tautologically exists in the above definition, corresponds to
the composition:

𝜋˚ ´b𝜔Gr𝐺 ΓIndCoh pGr𝐺 ,´q

QCohp𝐺p𝐾qq ÝÑ QCohpGr𝐺 q » IndCohpGr𝐺 q ÝÝÝÝÝÝÝÝÝÑ Vect.
Here 𝜋 : 𝐺p𝐾q Ñ Gr𝐺 “ 𝐺p𝐾q{𝐺p𝑂q is the projection, and IndCoh is defined in the standard sense
on Gr𝐺 because it is of ind-finite type; the rest of the notation is standard in the subject, and the
functor of tensoring with the dualizing sheaf is an equivalence by a theorem of [GR3] (and formal
smoothness of Gr𝐺 ).
Definition 5.6.4. A naive weak action of 𝐻 on C P DGCat𝑐𝑜𝑛𝑡 is an IndCoh˚ p𝐻q-module26 structure
for C.
Remark 5.6.5. The antipode for 𝐻 induces a canonical equivalence between left and right modules
for IndCoh˚ p𝐻q, so we often ignore the distinction going forward.
Remark 5.6.6. We sometimes omit “weak”: the distinction between naive and genuine actions in
this section only occurs for weak group actions, not for strong group actions.
Example 5.6.7. 𝐻 has a canonical naive action on IndCoh˚ p𝐻q.
Example 5.6.8. 𝐻 has a canonical naive action on Vect. Indeed, 𝐻 naively acts on 𝑘 with coaction
given by the unit map, and this is compatible with renormalization.
Example 5.6.9. For any indscheme 𝑋 of ind-finite type with an 𝐻 action, 𝐻 naively acts on
IndCohp𝑋q.
Example 5.6.10. If 𝐻 acts on a Tate Lie algebra k, then 𝐻 naively weakly acts on k–mod, (defined
as in Example 4.3.8). In particular, 𝐻 weakly acts on h–mod.
Ñ Ñ
b,𝐻ñ
5.7. We let Alg𝑟𝑒𝑛 denote the category of renormalized b-algebras with naive 𝐻-actions com-
Ñ
patible with the renormalization, i.e., the category of 𝐵-comodules in Alg b for 𝐵 “ Γp𝐻, O𝐻 q
equipped with the renormalization datum IndCoh˚ p𝐻q.
Ñ
b,𝐻ñ
Note that for 𝐴 P Alg𝑟𝑒𝑛 , 𝐻 acts naively on 𝐴–mod𝑟𝑒𝑛 (c.f. S5.4).

5.8. We let 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 denote the 2-category of categories with a naive weak action of 𝐻,
i.e., IndCoh˚ p𝐻q–modpDGCat𝑐𝑜𝑛𝑡 q.
For C P IndCoh˚ p𝐻q–mod, we define the naive weak invariants and coinvariants as:

C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 :“ Hom𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 pVect, Cq, C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 :“ Vect b C.

IndCoh˚ p𝐻q

25In particular, 𝐺 is classical and finite type.

26Here we are considering IndCoh˚ p𝐻q as an algebra object in DGCat
so e.g. the action functor is IndCoh˚ p𝐻qb
𝑐𝑜𝑛𝑡 ,
C Ñ C. In particular, the induced action IndCoh p𝐻q ˆ C Ñ C commutes with colimits in each variable separately.
˚
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 39

5.9. Genuine actions. In the remainder of this section and in S7, we study a more robust variant
of the above notion, under somewhat more restrictive hypotheses. In this section, we focus on the
case where 𝐻 is profinite dimensional, which contains the main phenomena.
5.10. Finite dimensional reminder. We first remind the reader of the following foundational
result, which will play a key role.
Let 𝐻 be an affine algebraic group. In this case, we remove the label “naive” from the notation,
e.g., 𝐻–mod𝑤𝑒𝑎𝑘 “ 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 — the naiveté is only in the infinite type setting.
Theorem 5.10.1 (Gaitsgory, [Gai8]). For 𝐻 an affine algebraic group, the functor:

𝐻–mod𝑤𝑒𝑎𝑘 Ñ Repp𝐻q–mod “ Repp𝐻q–modpDGCat𝑐𝑜𝑛𝑡 q

C ÞÑ C𝐻,𝑤
is an equivalence. (Here the functor exists because Repp𝐻q :“ QCohpB𝐻q is tautologically isomorphic
to Hom𝐻–mod𝑤𝑒𝑎𝑘 pVect, Vectq as a monoidal category.)
5.11. Profinite dimensional setting. In the remainder of this section, we suppose 𝐻 is a classical
affine group scheme.
5.12. Let 𝐵 “ Funp𝐻q P Vect♡ as before. Because 𝐻 can be written as a limit of smooth schemes
under smooth morphisms, the tautological functor 𝐵–mod𝑟𝑒𝑛 “ IndCoh˚ p𝐻q Ñ QCohp𝐻q “
𝐵–mod𝑛𝑎𝑖𝑣𝑒 is an equivalence.
5.13. We begin with a remark in the naive setting.
Note that the Beck-Chevalley conditions apply for the cosimplicial diagram defining C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 .
Therefore, Oblv : C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ C admits a continuous right adjoint Av𝑤,𝑛𝑎𝑖𝑣𝑒
˚ , and Oblv is comonadic.
The comonad on C is given by convolution with the coalgebra O𝐻 in the monoidal category
QCohp𝐻q.
In particular, for C “ Vect, we obtain that Repp𝐻q𝑛𝑎𝑖𝑣𝑒 :“ Vect𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 p“ QCohpB𝐻qq is canon-
ically equivalent to the category of 𝐵-comodules (with 𝐵 as above).
5.14. We now define Repp𝐻q “ Repp𝐻q𝑟𝑒𝑛 as IndpRepp𝐻q𝑐 qq for Repp𝐻q𝑐 Ď Repp𝐻q𝑛𝑎𝑖𝑣𝑒 the
full subcategory generated by finite dimensional representations, i.e., objects whose image in Vect
is compact.27 Since Repp𝐻q𝑐 is closed under tensor products in Repp𝐻q𝑛𝑎𝑖𝑣𝑒 , Repp𝐻q is a rigid
symmetric monoidal DG category.
Note that Repp𝐻q carries a canonical 𝑡-structure for which the forgetful functor to Vect is 𝑡-exact.
We have Repp𝐻q` » Repp𝐻q` 𝑛𝑎𝑖𝑣𝑒 .

5.15. The following definition plays a key role.

Definition 5.15.1. The category 𝐻–mod𝑤𝑒𝑎𝑘 of categories with a genuine 28 weak 𝐻-action is Repp𝐻q–mod “
Repp𝐻q–modpDGCat𝑐𝑜𝑛𝑡 q.
27Note that this example fits into the formalism of S4. Indeed, 𝐵 is the union of its finite dimensional sub-coalgebras,
Ñ
so 𝐵 _ P ProVect is a profinite dimensional algebra, in particular, an b-algebra. Its modules are tautologically the
same as 𝐵-comodules. This definition of Repp𝐻q is then obtained by applying Example 4.3.5.
28The terminology is borrowed from equivariant homotopy theory. In that context, for finite 𝐻, one extends
the naive notion of 𝐻-action on a spectrum in such a way that the trivial representation (and more generally,
any permutation representation) becomes compact. This is somewhat analogous to the present context, where we
renormalize 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 so that the trivial representation Vect becomes (completely) compact.
Although the subtleties in our context only occur for group schemes (which are analogous to profinite groups) and
group indschemes (which are analogous to locally compact totally disconnected groups), we still find this analogy to
be somewhat evocative.
40 SAM RASKIN

Construction 5.15.2. Note that Repp𝐻q𝑛𝑎𝑖𝑣𝑒 “ Hom𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 pVect, Vectq. In particular, Vect
admits commuting actions of Repp𝐻q𝑛𝑎𝑖𝑣𝑒 and QCohp𝐻q. In particular, since Repp𝐻q𝑟𝑒𝑛 Ñ Repp𝐻q𝑛𝑎𝑖𝑣𝑒
is symmetric monoidal, Vect is a bimodule for Repp𝐻q𝑟𝑒𝑛 and QCohp𝐻q, and therefore tensoring
defines a functor:

𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 .
Notation 5.15.3. Following the case of finite dimensional 𝐻, we think of the underlying object of
DGCat𝑐𝑜𝑛𝑡 as the weak 𝐻-invariants of a DG category acted on by 𝐻.
To accommodate this, suppose we are given an object of 𝐻–mod𝑤𝑒𝑎𝑘 . By definition, this means
that we are given an object D P Repp𝐻q–mod. We use the notation C𝐻,𝑤 in place of D, where we
let C denote the underlying object of 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 . We then abusively write C P 𝐻–mod𝑤𝑒𝑎𝑘 to
summarize the situation.
Roughly, the reader should think C P 𝐻–mod𝑤𝑒𝑎𝑘 means that C P 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 , and we are
given a “correction” C𝐻,𝑤 to C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 .
We emphasize that this “forgetful functor” 𝐻–mod𝑤𝑒𝑎𝑘 Ñ DGCat𝑐𝑜𝑛𝑡 (factoring through 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 )
is not conservative.
Remark 5.15.4. Because Repp𝐻q is rigid monoidal, Vect is dualizable over Repp𝐻q. Therefore, the
functor 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 admits left and right adjoints. It is immediate to see
that they are computed as strong and weak invariants respectively, with Repp𝐻q acting through
Repp𝐻q𝑛𝑎𝑖𝑣𝑒 .
In particular, for C P 𝐻–mod𝑤𝑒𝑎𝑘 , there is a canonical functor:

C𝐻,𝑤 Ñ C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 .
It is not difficult to see that each of these functors 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 Ñ 𝐻–mod𝑤𝑒𝑎𝑘 are fully-
faithful. Indeed, it is well-known that it suffices to verify this for either functor, and for the right
adjoint this is the content of [Ras4] Proposition 3.5.1 (which is proved by a standard Beck-Chevalley
argument).
Example 5.15.5. We have a canonical object Vect P 𝐻–mod𝑤𝑒𝑎𝑘 with Vect𝐻,𝑤 “ Repp𝐻q. Clearly
Hom𝐻–mod𝑤𝑒𝑎𝑘 pVect, Cq “ C𝐻,𝑤 .
Example 5.15.6. By Theorem 5.10.1, genuine and naive ś8actions coincide in the finite dimen-
sional case. It is straightforward to show that if 𝐻 “ 𝑖“1 G𝑎 , then Repp𝐻q is not equivalent
to Repp𝐻q𝑛𝑎𝑖𝑣𝑒 , so the two notions do not coincide in this case.
Remark 5.15.7. The relationship between 𝐻–mod𝑤𝑒𝑎𝑘 and 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 is somewhat analogous
to the relationship between IndCoh and QCoh, though it occurs a categorical level higher. Namely,
there is a non-conservative functor 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 analogous to the functor Ψ :
IndCohp𝑆q Ñ QCohp𝑆q for an eventually coconnective Noetherian scheme 𝑆, and in both cases,
there are fully-faithful left and right adjoints.
5.16. The key advantage of genuine 𝐻-actions is that the theory completely reduces to the finite
dimensional setting, as we now discuss.
Indeed, recall that 𝐻 is a limit lim𝑖 𝐻𝑖 of affine algebraic groups under smooth29 surjective maps.
Let 𝐾𝑖 Ď 𝐻 denote the kernel of the map 𝐻 Ñ 𝐻𝑖 . Note that there is a canonical functor
𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻𝑖 –mod𝑤𝑒𝑎𝑘 , sending C to:
29Of course we are using characteristic zero in an essential way.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 41

C𝐾𝑖 ,𝑤 :“ C𝐻,𝑤 b Vect.

Repp𝐻𝑖 q

That is, we apply the restriction along the tensor functor Repp𝐻𝑖 q Ñ Repp𝐻q and the inverse to
Theorem 5.10.1.

Proposition 5.16.1. The induced functor:

𝐻–mod𝑤𝑒𝑎𝑘 Ñ lim 𝐻𝑖 –mod𝑤𝑒𝑎𝑘

𝑖

is an equivalence.

This follows immediately from the next lemma.

Lemma 5.16.2. The morphism:

colim Repp𝐻𝑖 q Ñ Repp𝐻q P ComAlgpDGCat𝑐𝑜𝑛𝑡 q

𝑖

is an equivalence.

Proof. This is a special case of Example 4.3.5.



Corollary 5.16.3. For any C P 𝐻–mod𝑤𝑒𝑎𝑘 , the functor:

colim C𝐾𝑖 ,𝑤 Ñ C P DGCat𝑐𝑜𝑛𝑡

𝑖

is an equivalence. Moreover, each of the structural functors in this colimit admits a continuous right
adjoint.

Corollary 5.16.4. The functor Oblv : C𝐻,𝑤 Ñ C admits a continuous right adjoint Av𝑤
˚.

5.17. Functoriality. Suppose 𝑓 : 𝐻1 Ñ 𝐻2 is a morphism of classical affine group schemes.

We claim that there are induced adjoint functors:

ind𝑤 : 𝐻1 –mod𝑤𝑒𝑎𝑘 Õ 𝐻2 –mod𝑤𝑒𝑎𝑘 : Res

with the weak induction functor ind𝑤 also canonically isomorphic to the right adjoint to Res.
Indeed, unwinding the definitions, 𝐻𝑖 –mod𝑤𝑒𝑎𝑘 » Repp𝐻𝑖 q–mod, and we take ind𝑤 to correspond
to restriction of module categories along the symmetric monoidal functor Repp𝐻2 q Ñ Repp𝐻1 q.
This functor obviously admits a left adjoint Repp𝐻1 q bRepp𝐻2 q ´, which is defined to be Res. Then
Res is both left and right adjoint because Repp𝐻1 q is self-dual as a Repp𝐻2 q-module category by
general properties of rigid monoidal DG categories, c.f. [Gai4].
In particular, taking 𝐻 Ñ Specp𝑘q, we see Res : DGCat𝑐𝑜𝑛𝑡 Ñ 𝐻–mod𝑤𝑒𝑎𝑘 sends Vect to itself
with the trivial 𝐻-action. The equality of left and right adjoints here should be interpreted as an
“invariants = coinvariants” statement for genuine 𝐻-actions. We remark that the corresponding
statement is false in the setting of naive weak actions.
42 SAM RASKIN

5.18. Genuine actions via canonical renormalization. The following is a typical construction
of genuine weak 𝐻-actions.
Suppose 𝐻 acts naively on C. Suppose moreover that C is equipped with a 𝑡-structure such
that Oblv Av𝑤,𝑛𝑎𝑖𝑣𝑒
˚ : C Ñ C is 𝑡-exact. Then C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 has a (unique) 𝑡-structure such that
Oblv : C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ C is 𝑡-exact (c.f. the proof of Proposition 5.18.3 (1) below). Note that the
functor Av𝑤,𝑛𝑎𝑖𝑣𝑒
˚ is also 𝑡-exact in this case.
In what follows, we let C𝐻,𝑤,𝑐 Ď C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 denote the (non-cocomplete) DG subcategory of
objects F P C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 with OblvpFq compact in C.
Definition 5.18.1. In the above setting, we say the naive action of 𝐻 on C canonically renormalizes
(compatibly with the 𝑡-structure) if:
(1) C and its 𝑡-structure are compactly generated.
(2) Compact objects in C are bounded (i.e., eventually connective and coconnective).
(3) The essential image of the functor Oblv : C𝐻,𝑤,𝑐 X C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,ď0 Ñ C𝑐 X Cď0 Karoubi gener-
ates. (Here C𝑐 Ď C is the subcategory of compact objects.)
Remark 5.18.2. Note that under the above hypotheses, the functor C𝐻,𝑤,𝑐 Ñ C𝑐 Karoubi generates.
The following result summarizes the main features of this setting.
Proposition 5.18.3. Suppose 𝐻 acts naively on C, C is equipped with a 𝑡-structure compatible
with the 𝐻-action, and suppose the 𝐻-action canonically renormalizes.
Define C𝐻,𝑤 as IndpC𝐻,𝑤,𝑐 q; as C𝐻,𝑤,𝑐 Ď C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 is a Repp𝐻q𝑐 -submodule category, C𝐻,𝑤 has a
canonical Repp𝐻q-module structure.
Let 𝜓 denote the canonical functor:

𝜓 : C𝐻,𝑤 Ñ C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 P DGCat𝑐𝑜𝑛𝑡

ind-extending the embedding C𝐻,𝑤,𝑐 ãÑ C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 . Note that 𝜓 is a morphism of Repp𝐻q-module
categories.
𝜓
We use a standard abuse of notation in letting Oblv : C𝐻,𝑤 Ñ C denote the composition C𝐻,𝑤 Ý Ñ
Oblv
C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 ÝÝÝÑ C.
(1) C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 admits a unique 𝑡-structure such that such that the (conservative) forgetful functor
to C is 𝑡-exact.
(2) C𝐻,𝑤 has a unique compactly generated 𝑡-structure such that the forgetful functor to C is
𝑡-exact and conservative on eventually coconnective subcategories.
(3) For 𝑉 P Repp𝐻q♡ , the action functors 𝑉 ‹ ´ : C𝐻,𝑤 Ñ C𝐻,𝑤 and 𝑉 ‹ ´ : C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ
C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 are 𝑡-exact.
(4) The functor 𝜓 is 𝑡-exact and an equivalence on eventually coconnective subcategories:
»
𝜓 : C𝐻,𝑤,` Ý
Ñ C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,` .
(5) The composition:30

C𝐻,𝑤 b Vect Ñ C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 b Vect Ñ C

Repp𝐻q Repp𝐻q

is an equivalence. In particular, C𝐻,𝑤 P Repp𝐻q–mod induces a canonical genuine 𝐻-action

on C.
30In fact, each functor here is an equivalence.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 43

˚ : C Ñ C
(6) Let Av𝑤 𝐻,𝑤 denote the right adjoint to Oblv. Then the induced natural transfor-

mation:

𝜓 ˝ Av𝑤
˚ Ñ Av˚
𝑤,𝑛𝑎𝑖𝑣𝑒

(of functors C Ñ C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 ) is an isomorphism.

Proof. (1) follows by noting that C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 is the totalization Tot C b QCohp𝐻qb‚ and all of the
` ˘

structural maps in the underlying semi-cosimplicial diagram are 𝑡-exact.31

In (2), the uniqueness is clear: the 𝑡- structure must have C𝐻,𝑤,ď0 generated under colimits by
C 𝐻,𝑤,𝑐 X C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,ď0 . As is standard, this does define a 𝑡-structure, and the forgetful functor to C
is clearly right 𝑡-exact. We will complete the proof of (2) later in the argument; but now, we will
verify that (3) holds for this 𝑡-structure (without relying on any as yet unproved parts of (2)).
For 𝑉 P Repp𝐻q♡ , we first show that the functor 𝑉 ‹ ´ : C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 is 𝑡-exact. Here
it suffices to verify this after applying Oblv. But Oblv ˝p𝑉 ‹ ´q “ 𝑉 b Oblvp´q, which is clearly
𝑡-exact.
To see 𝑉 ‹ ´ : C𝐻,𝑤 Ñ C𝐻,𝑤 is 𝑡-exact, note that we can assume 𝑉 is finite dimensional (because
the 𝑡-structure on C𝐻,𝑤 , being compactly generated, is compatible with filtered colimits). By the
naive case, this functor preserves C𝐻,𝑤,𝑐 XC𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,ď0 , so we obtain right 𝑡-exactness in the genuine
setting. Now 𝑉 ‹ ´ is right adjoint to the left 𝑡-exact functor 𝑉 _ ‹ ´, giving the left 𝑡-exactness.
We now make an auxiliary observation. Suppose G P C𝐻,𝑤,𝑐 Ď C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 . By assumption, G lies
in cohomological degrees ě ´𝑁 for 𝑁 " 0. We claim that G is actually compact as an object of
C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,ě´𝑁 . Indeed, because G is eventually connective, the functor:

HomC𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 pG, ´q : C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,ě´𝑁 Ñ Gpd

factors through the subcategory of 𝑀 -truncated groupoids for some 𝑀 " 0 (depending on 𝑁 and
G). Moreover, we have:
»
Ñ Tot HomCbQCohp𝐻qb‚ pOblvpGq b Ob‚
HomC𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 pG, ´q Ý 𝐻 , ´q
»
where each of these functors factors through 𝑀 -truncated groupoids. Therefore, we have Tot Ý Ñ
Totď𝑀 `1 here, so commuting finite limits with filtered colimits in Gpd gives the claim about G.
Using this observation, we will now show (6). Note that the canonical natural transformation:

Funp𝐻q ‹ ´ Ñ Av𝑤,𝑛𝑎𝑖𝑣𝑒
˚ Oblv P HompC𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 , C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 q (5.18.1)
is an isomorphism, where here ‹ denotes the action of Repp𝐻q on C 𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 and Funp𝐻q is the
regular representation in Repp𝐻q♡ Ď Repp𝐻q. Indeed, the identification32 C “ C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 bRepp𝐻q
Vect and the Beck-Chevalley formalism imply this.
We now similarly claim that there is a canonical isomorphism:
»
Ñ Av𝑤 ˚ Oblv : C Ñ C𝐻,𝑤 .
𝐻,𝑤
Funp𝐻q ‹ ´ Ý
Both functors commute with colimits, so it suffices to verify that for every F P C𝐻,𝑤,𝑐 , the natural
map:

Funp𝐻q ‹ F Ñ Av𝑤
˚ OblvpFq

31Note that this argument is general for naive 𝐻-actions and compatible 𝑡-structures. I.e., it is not specific to
canonical renormalization.
32This follows from identifying both sides with Funp𝐻q–modpCq using Barr-Beck and the Beck-Chevalley formalism.
44 SAM RASKIN

is a natural isomorphism. Let G P C𝐻,𝑤,𝑐 be given. Write Funp𝐻q as a filtered colimit colim𝑖 𝑉𝑖
where 𝑉𝑖 P Repp𝐻q♡ are finite dimensional. We claim that:

»
colim HomC𝐻,𝑤,𝑐 pG, 𝑉𝑖 ‹ Fq Ý
Ñ HomC𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 pG, colim 𝑉𝑖 ‹ Fq.
𝑖 𝑖

Indeed, because F and G are eventually coconnective, by (3) (in the naive case), we have 𝑉𝑖 ‹F, F, G P
C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,ě´𝑁 for 𝑁 " 0 (and for all 𝑖). Then the fact that G is compact in C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,ě´𝑁 ´𝑟 for
all 𝑟 ě 0 gives the claim.
Therefore, we have:

HomC𝐻,𝑤 pG, Funp𝐻q ‹ Fq “ HomC𝐻,𝑤 pG, colim 𝑉𝑖 ‹ Fq “ colim HomC𝐻,𝑤 pG, 𝑉𝑖 ‹ Fq “

𝑖 𝑖
colim HomC𝐻,𝑤,𝑐 pG, 𝑉𝑖 ‹ Fq “ HomC𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 pG, colim 𝑉𝑖 ‹ Fq “ HomC𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 pG, Av𝑤,𝑛𝑎𝑖𝑣𝑒
˚ pFqq “
𝑖 𝑖
HomC pOblvpGq, OblvpFqq.

To complete the argument, note that both functors Av𝑤,𝑛𝑎𝑖𝑣𝑒

˚ and 𝜓 Av𝑤 ˚ commute with colimits,
so it suffices to show that our natural transformation is an equivalence when evaluated on compact
objects in C. Moreover, because the naive 𝐻-action on C canonically renormalizes, it suffices to
check this on compact objects of the form OblvpFq for F P C𝐻,𝑤,𝑐 . But then Av𝑤,𝑛𝑎𝑖𝑣𝑒˚ Oblv and
𝜓 Av𝑤 ˚ Oblv are each canonically given by the action of Funp𝐻q, and 𝜓 is Repp𝐻q-linear, giving the
claim.
Returning to (2), we claiom that for F P C𝐻,𝑤,ě0 we have OblvpFq P Cě0 . Note that Av𝑤 ˚ OblvpFq “
Funp𝐻q ‹ F as before, and since Funp𝐻q is in degree 0, Funp𝐻q ‹ F is also in degrees ě 0. Therefore,
for G P C𝐻,𝑤,𝑐 X C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,ă0 , we have:

HomC pOblv G, Oblv Fq “ HomC𝐻,𝑤 pG, Av𝑤

˚ Oblv Fq “ 0 P Gpd

By hypothesis, Că0 is generated under colimits by such objects Oblv G, giving the claim.
To conclude (2), we need to show that Oblv is conservative on eventually coconnective objects.
Suppose F P C𝐻,𝑤,ě0 with OblvpFq “ 0. Then we have:

1ÞÑ1
Fr1s “ CokerpF Ñ Av𝑤 ÝÝÑ Funp𝐻qq ‹ F.
˚ OblvpFqq “ Cokerp𝑘 Ý

By (3), the right hand side is in C𝐻,𝑤,ě0 , so Fr1s P C𝐻,𝑤,ě0 , so F P C𝐻,𝑤,ě1 . Iterating this, we obtain
F “ 0 as desired.
Then (4) follows from (2) and (6) by observing that these results imply that the forgetful functor
C𝐻,𝑤,` Ñ C` is comonadic with comonad given by the action of Funp𝐻q P QCohp𝐻q.
FÞÑFbRepp𝐻q 𝑘
It remains to show (5). By the Beck-Chevalley formalism, C𝐻,𝑤 ÝÝÝÝÝÝÝÝÝÑ C𝐻,𝑤 bRepp𝐻q Vect
admits a conservative right adjoint, and the corresponding monad on C𝐻,𝑤 is the action of Funp𝐻q P
Repp𝐻q. Using our canonical identification of that convolution with Av𝑤 ˚ Oblv, we obtain that the
functor from (5) is fully-faithful. By Remark 5.18.2, it is also essentially surjective.


5.19. Canonical renormalization for IndCoh. We also have the following variant.
Lemma 5.19.1. Suppose that 𝑋 is an indscheme locally almost of finite type acted on by 𝐻. Then
the naive 𝐻-action on IndCohp𝑋q canonically renormalizes (relative to its canonical 𝑡-structure).
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 45

Proof. We use the following construction. Suppose F P IndCohp𝑋q𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,♡ and G Ď OblvpFq is

a subobject. Define a subobject G r Ď F P IndCohp𝑋q𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,♡ as the fiber product (in the abelian
category IndCohp𝑋q 𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒,♡ ):

G
r“F ˆ Av𝑤 /F
˚ pGq
Av𝑤
˚ OblvpFq

coact
 
Av𝑤 / Av𝑤 OblvpFq.
˚ pGq ˚

Observe that OblvpGq

r Ď G Ď OblvpFq (using the counit of the adjunction).33
Now if G is coherent, then OblvpGq
r is as well (because we are in a Noetherian setup). Moreover,
GÞÑG
r
the map SubObjpOblvpFqq ÝÝÝÑ SubObjpFq commutes with filtered colimits (because we are taking
fiber products in a Grothendieck abelian category). Therefore, we have:

F“ colim G.
r
GĎOblvpFq coherent

Applying Oblv, we see that OblvpFq is a filtered colimit of objects coming from IndCohp𝑋q𝐻,𝑤,𝑐 .
Since such objects OblvpFq generate IndCohp𝑋qď0 under colimits (since Av𝑤˚ is 𝑡-exact and conser-
vative), this gives the claim.


5.20. Varying the group. We record the following result for later use.
Lemma 5.20.1. Suppose 𝑓 : 𝐻1 Ñ 𝐻2 is a morphism of classical affine group schemes. Suppose
C P DGCat𝑐𝑜𝑛𝑡 is equipped with a 𝑡-structure and a compatible naive action of 𝐻2 that renormalizes.
Then:
(1) The induced naive 𝐻1 -action also renormalizes.
(2) The category:

Repp𝐻1 q b C𝐻2 ,𝑤
Repp𝐻2 q
is compactly generated, and the natural functor:

Repp𝐻1 q b C𝐻2 ,𝑤 Ñ C𝐻1 ,𝑤,𝑛𝑎𝑖𝑣𝑒 (5.20.1)

Repp𝐻2 q

maps compact objects to objects in C𝐻1 ,𝑤,𝑐 (with this subcategory defined as in Proposition
5.18.3).
(3) The functor:
˘𝑐
C𝐻2 ,𝑤 Ñ C𝐻1 ,𝑤,𝑐
`
Repp𝐻1 q b
Repp𝐻2 q

induced by (5.20.1) is fully-faithful, so induces a fully-faithful functor:

Repp𝐻1 q b C𝐻2 ,𝑤 Ñ C𝐻1 ,𝑤 . (5.20.2)

Repp𝐻2 q

33In fact, G
r is maximal among subobjects of F with this property.
46 SAM RASKIN

(4) If C “ IndCohp𝑋q for 𝑋 locally almost of finite type and acted on by 𝐻2 (as in Lemma
5.19.1), then (5.20.2) is an equivalence.
(5) If 𝐻1 and 𝐻2 are algebraic groups, then (5.20.2) is an equivalence.
Proof. (1) is immediate from the definitions.
In (2), note that all the categories appearing in the construction computing the tensor prod-
uct Repp𝐻1 q bRepp𝐻2 q C𝐻2 ,𝑤 are compactly generated, and each of the functors in the underlying
semisimplicial diagram preserve compact objects. Therefore, this tensor product is compactly gen-
erated by objects of the form 𝑉 bRepp𝐻2 q F for 𝑉 P Repp𝐻1 q𝑐 and F P C𝐻2 ,𝑤,𝑐 . Moreover, the functor
(5.20.1) sends this object to 𝑉 ‹ OblvpFq, where Oblv denotes the functor C𝐻2 ,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ C𝐻1 ,𝑤,𝑛𝑎𝑖𝑣𝑒
and 𝑉 denotes the action of Repp𝐻1 q on C𝐻1 ,𝑤,𝑛𝑎𝑖𝑣𝑒 . Clearly this object lies in C𝐻1 ,𝑤,𝑐 , giving the
claim.
For (3), let Ind𝐻 2
𝐻1 : Repp𝐻1 q Ñ Repp𝐻2 q denote the (continuous) right adjoint to the restric-
tion functor. Because Ind𝐻 2
𝐻1 is a morphism of Repp𝐻2 q-module categories (by rigid monoidality of
Repp𝐻2 q), we see that Ind𝐻 2
𝐻1 b idC𝐻2 ,𝑤 is right adjoint to the functor pF P C𝐻2 ,𝑤 q ÞÑ 𝑘bRepp𝐻2 q F
Repp𝐻2 q
(for 𝑘 the trivial representation).
Now let F, G P C𝐻2 ,𝑤 and let 𝑉, 𝑊 P Repp𝐻1 q be given with 𝑊 P Repp𝐻1 q𝑐 . By the above, we
have:

HomRepp𝐻1 q b C𝐻2 ,𝑤 p𝑊 b G, 𝑉 b Fq “
Repp𝐻2 q Repp𝐻2 q Repp𝐻2 q

HomRepp𝐻1 q b C𝐻2 ,𝑤 p𝑘 b G, p𝑊 _ b 𝑉 q b Fq “
Repp𝐻2 q Repp𝐻2 q Repp𝐻2 q

HomC𝐻2 ,𝑤 pG, Ind𝐻

𝐻1 p𝑊 b 𝑉 q ‹ Fq.
2 _

Now if we assume F, G P C𝐻2 ,𝑤,` and 𝑉 P Repp𝐻1 q` , then by Proposition 5.18.3 the above Hom
maps isomorphically (via the functor 𝜓 from loc. cit.) onto:

HomC𝐻2 ,𝑤,𝑛𝑎𝑖𝑣𝑒 pG, Ind𝐻

𝐻1 p𝑊 b 𝑉 q ‹ Fq.
2 _
(5.20.3)
Before calculating this term further, let Oblv𝐻2 Ñ𝐻1 : C𝐻2 ,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ C𝐻1 ,𝑤,𝑛𝑎𝑖𝑣𝑒 be the restric-
tion functor and let Av𝑤,𝑛𝑎𝑖𝑣𝑒,𝐻
˚
1 Ñ𝐻2
: C𝐻1 ,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ C𝐻2 ,𝑤,𝑛𝑎𝑖𝑣𝑒 denote its right adjoint. Note
that although this latter functor may not commute with colimits,34 its restriction to C𝐻2 ,𝑤,𝑛𝑎𝑖𝑣𝑒,ě0
commutes with filtered colimits.35 We claim that for any 𝑈 P Repp𝐻1 q` , the natural map:

Ind𝐻
𝐻1 p𝑈 q ‹ F Ñ Av˚
𝑤,𝑛𝑎𝑖𝑣𝑒,𝐻1 Ñ𝐻2
𝑈 ‹ Oblv𝐻2 Ñ𝐻1 pFq
2
` ˘

is an isomorphism. First, if 𝑈 is the regular representation, this follows from the identification
Av𝑤,𝑛𝑎𝑖𝑣𝑒
˚ Oblv “ Funp𝐻q‹´ from (5.18.1). If 𝑈 “ Funp𝐻qb𝑄 for 𝑄 P Vect` , then the claim follows
from commutation the fact that Av𝑤,𝑛𝑎𝑖𝑣𝑒,𝐻
˚
1 Ñ𝐻2
commutes with colimits bounded uniformly from
`
below. Finally, general 𝑈 P Repp𝐻1 q follows using the cobar resolution for 𝑈 , using the 𝑡-structure
to justify commuting the totalization with various functors.36
Applying this to 𝑈 “ 𝑊 _ b 𝑉 from above, we calculate (5.20.3) as:
34For example, if 𝐻 is a point and 𝐻 “ ś8 G .
2 1 𝑖“1 𝑎
35Indeed, compact generation of C𝐻2 ,𝑤 implies C𝐻2 ,𝑤,ě0 Ý
»
Ñ C𝐻2 ,𝑤,𝑛𝑎𝑖𝑣𝑒,ě0 is compactly generated by 𝜏 ě0 pC𝐻2 ,𝑤,𝑐 q;
these clearly map into 𝜏 ě0 pC𝐻1 ,𝑤,𝑐 q under Oblv. So Oblv : C𝐻2 ,𝑤,ě0 Ñ C𝐻1 ,𝑤,ě0 preserves compacts, so its right
adjoint preserves filtered colimits.
36More precisely, any truncation 𝜏 ď𝑁 applied to this totalization coincides with a suitable finite totalization.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 47

HomC𝐻2 ,𝑤,𝑛𝑎𝑖𝑣𝑒 pG, Ind𝐻

𝐻1 p𝑊 b 𝑉 q ‹ Fq “
2 _

HomC𝐻2 ,𝑤,𝑛𝑎𝑖𝑣𝑒 pG, Av𝑤,𝑛𝑎𝑖𝑣𝑒,𝐻 𝑊 b 𝑉 ‹ Oblv𝐻2 Ñ𝐻1 pFqq “

` _
1 Ñ𝐻2
˚
HomC𝐻1 ,𝑤,𝑛𝑎𝑖𝑣𝑒 pOblv𝐻2 Ñ𝐻1 pGq, p𝑊 _ b 𝑉 q ‹ Oblv𝐻2 Ñ𝐻1 pFqq “
HomC𝐻2 ,𝑤,𝑛𝑎𝑖𝑣𝑒 p𝑊 ‹ Oblv𝐻2 Ñ𝐻1 pGq, 𝑉 ‹ Oblv𝐻2 Ñ𝐻1 pFqq.
As the left hand side of (5.20.1) is compactly generated by objects of the form 𝑉 bRepp𝐻2 q F for
F P C𝐻2 ,𝑤,𝑐 and 𝑉 P Repp𝐻2 q𝑐 , this gives fully-faithfulness of (5.20.1) when restricted to compact
objects.
Next, (5) follows from Lemma 5.20.2 (applied to 𝐻 “ 𝐻1 and D the essential image of (5.20.2)).
Finally, we show (4). It suffices to show that any object F P IndCohp𝑋q𝐻1 ,𝑤,𝑐 lies in the essential
image of (5.20.2). Moreover, we can assume F lies in the heart of the 𝑡-structure.
In the course of reductions, we use the following (simple) observation repeatedly: if 𝑓 : 𝑌 Ñ 𝑋
is an equivariant map of locally almost of finite type indschemes acted on by 𝐻2 , and F is of the
form 𝑓˚IndCoh pGq for some G P IndCohp𝑌 q𝐻2 ,𝑤 such that G lies in the essential image of:

Repp𝐻1 q b IndCohp𝑌 q𝐻2 ,𝑤 Ñ IndCohp𝑌 q𝐻1 ,𝑤

Repp𝐻2 q

then F lies in the essential image of the functor:

Repp𝐻1 q b IndCohp𝑋q𝐻2 ,𝑤 Ñ IndCohp𝑋q𝐻1 ,𝑤 .

Repp𝐻2 q

Because IndCohp𝑋q♡ “ IndCohp𝑋 𝑐𝑙 q♡ and similarly for equivariant categories, we may assume
(by the above) that 𝑋 is classical. Then 𝑋 is a colimit 𝑋 “ colim 𝑋𝑖 under closed embeddings of
finite type classical schemes acted on by 𝐻2 . Therefore, applying the above reduction technique
again, we may assume 𝑋 is a classical scheme of finite type.
Now observe that there exists a map 𝐻2 Ñ 𝐻21 of group schemes with 𝐻21 an affine algebraic
group and such that 𝐻2 acts on 𝑋 through 𝐻21 . We claim that there is a commutative diagram:

𝐻1 / 𝐻1
1

 
𝐻2 / 𝐻1
2

with 𝐻11 again an affine algebraic group and such that the 𝐻1 -equivariant structure on F comes
from an 𝐻11 -equivariant structure. Indeed, we can write 𝐻1 “ lim𝑖 𝐻1,𝑖 where each 𝐻1,𝑖 an affine
algebraic group over 𝐻21 (so in particular, it acts on 𝑋). Then the 𝐻1 -equivariant structure on F is
encoded by the coaction map:

OblvpFq Ñ Funp𝐻1 q ‹ OblvpFq P IndCohp𝑋q♡ .

The right hand side is colim𝑖 Funp𝐻1,𝑖 q ‹ OblvpFq, and because F is coherent, the above map factors
through Funp𝐻1,𝑖 q ‹ OblvpFq for some 𝑖. Since we are in a 1-categorical context here, it suffices to
take 𝐻11 “ 𝐻1,𝑖 for such 𝑖.
We then have a commutative diagram:
48 SAM RASKIN

Repp𝐻11 q b
1
IndCohp𝑋q𝐻2 ,𝑤 / IndCohp𝑋q𝐻11 ,𝑤
Repp𝐻21 q

 
Repp𝐻1 q b IndCohp𝑋q𝐻2 ,𝑤 / IndCohp𝑋q𝐻1 ,𝑤 .
Repp𝐻2 q

We have lifted F along the right vertical arrow. By (5), the top arrow is an equivalence. Therefore,
F lies in the essential image of the bottom functor, giving the result.

In the course of the above, we appealed to the following result.
Lemma 5.20.2. Let 𝐻 be an affine algebraic group acting weakly37 on C P DGCat𝑐𝑜𝑛𝑡 . Suppose
D Ď C𝐻,𝑤 be a DG subcategory such that:
‚ D is closed under colimits and the Repp𝐻q-action.
‚ D is compactly generated and the inclusion D ãÑ C𝐻,𝑤 preserves compact objects.
Oblv
‚ The composite D ãÑ C𝐻,𝑤 ÝÝÝÑ C generates C under colimits.
Then D “ C𝐻,𝑤 .
Proof. We will repeatedly use the fact shown in the course of the proof of Proposition 5.18.3 that
Av𝑤 ˚ Oblv “ Funp𝐻q ‹ ´ (as endofunctors on C
𝐻,𝑤 ).

First, note that for G P C, Av𝑤 ˚ pGq P D. Indeed, since the functor Oblv |D : D Ñ C generates
˚ :C
under colimits, it suffices to see that Oblv Av𝑤 𝐻,𝑤 Ñ C𝐻,𝑤 maps D into itself. But this functor

is given by the action of the regular representation in Repp𝐻q, so by assumption preserves D.

Now let 𝑘 P Repp𝐻q denote the trivial representation. By Lemma 3.7.2, we have:
»
𝑘Ý Ñ 𝜏 ď𝑛 Totď𝑛`1 pAv𝑤˚ Oblvq
‚`1
p𝑘q
for all 𝑛 ě 0. Because 𝐻 is finite type and we are in characteristic 0, Repp𝐻q has finite cohomological
dimension. Therefore, for 𝑛 " 0, the boundary map:

𝜏 ą𝑛 Totď𝑛`1 pAv𝑤
˚ Oblvq
‚`1
p𝑘q Ñ 𝑘r1s P Repp𝐻q
is nullhomotopic. Therefore, 𝑘 is a direct summand of Totď𝑛`1 pAv𝑤 ˚ Oblvq
‚`1 p𝑘q for 𝑛 " 0.

Now for F P C𝐻,𝑤 , we have F “ 𝑘 ‹ F (for ‹ denoting the action of Repp𝐻q), which implies that
F is a direct summand of:

Totď𝑛`1 pAv𝑤˚ Oblvq

‚`1
pFq.
By the above, this object lies in D, so F does as well.


6. Ind-coherent sheaves on some infinite dimensional spaces

6.1. This section, wedged as it is between S5 and S7, is an extended digression.
To orient the reader, we provide a somewhat extended introduction.
37As 𝐻 is finite type, this simply means C is a QCohp𝐻q-module category.
Similarly, when we refer to weak invariants in this lemma, this is what we would call naive weak invariants in an
infinite type setting. I.e., we are forming these weak invariants without any canonical renormalization or any such.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 49

6.2. First, let us explain the role this material plays in S7.
In loc. cit., a certain monoidal category denoted IndCoh˚𝑟𝑒𝑛 p𝐾z𝐻{𝐾q plays a key role, where 𝐻 a
Tate group indscheme and 𝐾 Ď 𝐻 is a compact open subgroup (see loc. cit. for the terminology).
The definition of this category is not hard: IndCoh˚𝑟𝑒𝑛 p𝐾z𝐻{𝐾q is compactly generated, and
compact objects are objects of IndCohp𝐻{𝐾q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 that map to Cohp𝐻{𝐾q Ď IndCohp𝐾z𝐻{𝐾q.
However, this description breaks symmetry, so the monoidal structure is not so evident. This prob-
lem becomes compounded when we try to compare these categories for different compact open
subgroups 𝐾, or different groups 𝐻, and so on.
Therefore, the ultimate goal of this section is to introduce a class of prestacks we call renor-
malizable, which includes prestacks of the form 𝐾z𝐻{𝐾, and for which there is a robust theory of
ind-coherent sheaves (denoted IndCoh˚𝑟𝑒𝑛 ). As the above description indicates, the most important
application of this material in S7 is to resolve some homotopy coherence issues.

6.3. In the above example, we could avoid breaking the symmetry by regarding IndCoh˚𝑟𝑒𝑛 p𝐾z𝐻{𝐾q
as 𝐾 ˆ 𝐾-equivariant ind-coherent sheaves on 𝐻. However, 𝐻 is of ind-infinite type, so is outside
the usual framework of [Gai5] and [GR4]. Therefore, we first develop IndCoh˚ on schemes (possibly
of infinite type, but qcqs and eventually coconnective) and reasonable indschemes (see S6.8 for the
definition).

6.4. Definition for schemes. Let ą´8Sch𝑞𝑐𝑞𝑠 denote the category of quasi-compact quasi-separated
eventually coconnective schemes.
For 𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 , we define Cohp𝑆q Ď QCohp𝑆q as the full subcategory of objects F P QCohp𝑆q`
such that F P QCohp𝑆qě´𝑁 implies F is compact in QCohp𝑆qě´𝑁 . We define IndCoh˚ p𝑆q as
IndpCohp𝑆qq, and we define a 𝑡-structure on IndCoh˚ p𝑆q by taking connective objects to be gen-
erated under colimits by Cohp𝑆q X QCohp𝑆qď0 . Note that there is a canonical continuous functor
Ψ “ Ψ𝑆 : IndCoh˚ p𝑆q Ñ QCohp𝑆q ind-extending the embedding Cohp𝑆q ãÑ QCohp𝑆q.
Lemma 6.4.1. Under the above hypotheses, the functor Ψ : IndCoh˚ p𝑆q Ñ QCohp𝑆q is 𝑡-exact and
an equivalence on eventually coconnective subcategories.
Proof. Clearly Ψ is right 𝑡-exact.
Let G P Perfp𝑆q Ď Cohp𝑆q be given. Then the functors:

HomQCohp𝑆q pG, Ψp´qq, HomIndCoh˚ p𝑆q pG, ´q : IndCoh˚ p𝑆q Ñ Vect

are canonically isomorphism as both commute with colimits and the restriction of the two to Cohp𝑆q
are clearly equal.
Therefore, for F P IndCoh˚ p𝑆qą0 and G P Perfp𝑆q X QCohp𝑆qď0 , we have:

HomQCohp𝑆q pG, ΨpFqq “ HomIndCoh˚ p𝑆q pG, Fq “ 0.

As QCohp𝑆qď0 is generated under colimits by such G, this implies ΨpFq P IndCoh˚ p𝑆qą0 , giving the
𝑡-exactness.
Now take F P Cohp𝑆q. We claim that the natural transformation:

HomIndCoh˚ p𝑆q pF, 𝜏 ě0 p´qq Ñ HomQCohp𝑆q pF, Ψp𝜏 ě0 p´qq

of functors IndCoh˚ p𝑆q Ñ Gpd is an isomorphism. Indeed, recall that there exists F1 P Perfp𝑆q and
a map F1 Ñ F inducing an isomorphism on 𝜏 ě0 ; therefore, we may assume F P Perfp𝑆q, and the
result follows from the (evident) identity:
50 SAM RASKIN

»
HomQCohp𝑆q pF, Ψp´qq Ý
Ñ HomIndCoh˚ p𝑆q pF, ´q
for such perfect F.
We immediately obtain that for any F P IndCoh˚ p𝑆q, the natural transformation:

HomIndCoh˚ p𝑆q pF, 𝜏 ě0 p´qq Ñ HomQCohp𝑆q pF, Ψp𝜏 ě0 p´qq

is an isomorphism. Clearly this is equivalent to fully-faithfulness of Ψ|IndCoh˚ p𝑆q` .
As QCohp𝑆qě0 is generated under colimits by 𝜏 ě0 pPerfp𝑆qq and this category is in the essential
image of Ψ by 𝑡-exactness, we obtain that Ψ induces an equivalence on coconnective objects as
desired.

Remark 6.4.2. The notation IndCoh˚ is parallel to similar notation from [Ras3] and is used to
emphasize the differences between the Noetherian and non-Noetherian situations. Note that one
can dualize to obtain a theory IndCoh! parallel to 𝐷! from loc. cit. Because IndCoh is canonically
self-dual on indschemes locally almost of finite type, we do not include superscripts when working
with such objects (since our theory manifestly recovers that of [GR4] in this case).

6.5. For 𝑓 : 𝑆 Ñ 𝑇 in ą´8Sch𝑞𝑐𝑞𝑠 , define 𝑓˚IndCoh : IndCoh˚ p𝑆q Ñ IndCoh˚ p𝑇 q P DGCat𝑐𝑜𝑛𝑡 to be

the unique left 𝑡-exact (continuous DG) fitting into a commutative diagram:

𝑓˚IndCoh
IndCoh˚ p𝑆q / IndCoh˚ p𝑇 q

Ψ Ψ
 𝑓˚ 
QCohp𝑆q / QCohp𝑇 q.

As in [Gai5] Proposition 3.2.4, this construction canonically upgrades to a functor ą´8Sch𝑞𝑐𝑞𝑠 Ñ

DGCat𝑐𝑜𝑛𝑡 .
Remark 6.5.1. For 𝑓 affine, 𝑓˚IndCoh is 𝑡-exact.
Notation 6.5.2. For 𝑓 the projection map 𝑆 Ñ Specp𝑘q, we use the notation ΓIndCoh p𝑆, ´q :
IndCoh˚ p𝑆q Ñ Vect for the corresponding pushforward functor (and similarly for the more gen-
eral 𝑆 considered later in this section).
6.6. Next, we discuss behavior with respect to flat morphisms.
Lemma 6.6.1. Suppose 𝑓 : 𝑆 Ñ 𝑇 P ą´8Sch𝑞𝑐𝑞𝑠 is flat. Then 𝑓˚IndCoh admits a left adjoint.
Proof. In this case, the adjoint functors 𝑓 ˚ : QCohp𝑇 q Õ QCohp𝑆q : 𝑓˚ preserve the subcategories
QCohp´qě´𝑛 for all integers 𝑛, and in particular induce an adjunction. It immediately follows that
𝑓 ˚ maps Cohp𝑇 q Ď QCohp𝑇 q to Cohp𝑆q, and that the ind-extension of this functor is the sought-after
left adjoint.


For such flat 𝑓 , we denote this left adjoint by 𝑓 ˚,IndCoh .

Lemma 6.6.2. For a Cartesian diagram:
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 51

𝜙
𝑆1 / 𝑇1

𝜓 𝑔
 𝑓 
𝑆 /𝑇
with 𝑔 flat and 𝑆, 𝑇, 𝑇 1 in ą´8Sch
𝑞𝑐𝑞𝑠 (so 𝑆 1 P ą´8Sch𝑞𝑐𝑞𝑠 as well), the natural transformation:

𝑔 ˚,IndCoh 𝑓˚IndCoh Ñ 𝜙IndCoh

˚ 𝜓 ˚,IndCoh
is an isomorphism.
Proof. Clear from the corresponding statement for QCoh.

Now by [GR4] Theorem V.[Link], the functor IndCoh˚ extends canonically to a functor:

IndCoh˚ : Corrpą´8Sch𝑞𝑐𝑞𝑠 q𝑎𝑙𝑙;𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡 .

Here we are using the notation from loc. cit. We remind that the source category is the correspon-
dence category for ą´8Sch𝑞𝑐𝑞𝑠 in which morphisms from one eventually coconnective scheme 𝑆 to
another 𝑇 are diagrams:

𝐻
𝛼 𝛽


𝑆 𝑇
with 𝛼 flat; the functor IndCoh attaches to such a correspondence the functor 𝛽˚IndCoh 𝛼˚,IndCoh . (We
˚

have omitted the “admissible” morphism data from loc. cit.; one may take only isomorphisms for
our purposes, i.e., only work with a 1-category of correspondences.)
Remark 6.6.3. The above material extends if we replace flatness by finite Tor- dimension. However,
we do not need this extension and therefore do not emphasize it.
6.7. We have the following basic result.
Lemma 6.7.1. IndCoh˚ satisfies Zariski descent on ą´8Sch
𝑞𝑐𝑞𝑠 .

Proof. The argument from [Gai5] Proposition 4.2.1 applies in this setting.

More generally, we have:
Proposition 6.7.2. IndCoh˚ satisfies flat descent on ą´8Sch
𝑞𝑐𝑞𝑠 (for upper-˚ functors).
Proof. Let 𝑓 : 𝑇 Ñ 𝑆 be a faithfully flat map in ą´8Sch
𝑞𝑐𝑞𝑠 . By definition, we need to show that:

IndCoh˚ p𝑆q Ñ Tot𝑠𝑒𝑚𝑖 pIndCoh˚ p𝑇 ˆ𝑆 ‚`1 qq

is an isomorphism, where Tot𝑠𝑒𝑚𝑖 indicates the limit over the semisimplicial category Δ𝑖𝑛𝑗 (which
we use because only the semisimplicial part of the Cech nerve has flat structural maps).
Next, observe that by construction, IndCoh˚ p𝑆q is naturally a QCohp𝑆q-module category (in
DGCat𝑐𝑜𝑛𝑡 ), and similarly for 𝑇 . Moreover, 𝑓 ˚,IndCoh : IndCoh˚ p𝑆q Ñ IndCoh˚ p𝑇 q is QCohp𝑆q-linear,
where IndCoh˚ p𝑇 q is a QCohp𝑆q-module category via 𝑓 ˚ : QCohp𝑆q Ñ QCohp𝑇 q.
Therefore, we obtain a functor:
52 SAM RASKIN

𝑓 ˚,IndCoh,𝑒𝑛ℎ : IndCoh˚ p𝑆q b QCohp𝑇 q Ñ IndCoh˚ p𝑇 q. (6.7.1)

QCohp𝑆q
The same argument as in [Gai5] Proposition 4.4.2 shows that this functor is fully-faithful. Note that
the essential image of this functor is the subcategory generated under colimits and the QCohp𝑇 q-
action by the essential image of 𝑓 ˚,IndCoh .
Next, observe that the above constructions are suitably functorial and therefore induce a fully-
faithful functor:

Tot𝑠𝑒𝑚𝑖 pIndCoh˚ p𝑆q b QCohp𝑇 ˆ𝑆 ‚`1 qq Ñ Tot𝑠𝑒𝑚𝑖 pIndCoh˚ p𝑇 ˆ𝑆 ‚`1 qq. (6.7.2)
QCohp𝑆q
Below, we will show that this functor is actually an equivalence.
Assuming this, let us deduce the descent claim. As IndCoh˚ p𝑆q is compactly generated, hence
dualizable, and QCohp𝑆q is rigid monoidal, we obtain that IndCoh˚ p𝑆q is dualizable as a QCohp𝑆q-
module category. Therefore:

»
IndCoh˚ p𝑆q b Tot𝑠𝑒𝑚𝑖 pQCohp𝑇 ˆ𝑆 ‚`1 qq Ý
Ñ Tot𝑠𝑒𝑚𝑖 pIndCoh˚ p𝑆q b QCohp𝑇 ˆ𝑆 ‚`1 qq.
QCohp𝑆q QCohp𝑆q

The left hand side is then IndCoh˚ p𝑆q bQCohp𝑆q QCohp𝑆q by flat descent for QCoh (see [Lur4] Corol-
lary D.6.3.3).
We now show that (6.7.2) is an equivalence. Suppose we are given an object of the right hand side.
»
In particular, we are given F P IndCoh˚ p𝑇 q with an isomorphism 𝛼 : 𝑝˚,IndCoh
1 pFq ÝÑ 𝑝˚,IndCoh
2 pFq for
𝑝𝑖 : 𝑇 ˆ𝑆 𝑇 Ñ 𝑇 the projections.
We will show that the map F Ñ 𝑝IndCoh
1,˚ 𝑝˚,IndCoh
2 pFq adjoint to 𝛼 realizes F as a summand. Assum-
ing this claim, we obtain that F is a direct summand of 𝑓 ˚,IndCoh 𝑓˚IndCoh pFq “ 𝑝IndCoh
1,˚ 𝑝˚,IndCoh
2 pFq, in
particular, a summand of an object lying in the essential image of (6.7.1). This implies that F is in
the essential image of (6.7.1), our original object lies in the essential image of (6.7.2), completing
the argument.
Let ∆ : 𝑇 Ñ 𝑇 ˆ𝑆 𝑇 denote the diagonal map and let ∆˚,IndCoh : IndCoh˚ p𝑇 ˆ𝑆 𝑇 q Ñ
PropIndCoh˚ p𝑇 qq denote the “partially-defined” left adjoint to ∆IndCoh˚ , noting that ∆˚,IndCoh is
defined on 𝑝˚,IndCoh
𝑖 pFq. Then observe that ∆˚,IndCoh p𝛼q “ idF . Indeed, the standard argument in
a simplicial setting applies in our setting: applying the partially-defined ˚-restriction along the
diagonal 𝑇 Ñ 𝑇 ˆ𝑆 𝑇 ˆ𝑆 𝑇 to the cocycle relation here gives the claim.
Therefore, the diagram:
𝛼
𝑝˚,IndCoh pFq
» / 𝑝˚,IndCoh pFq
1 2

' w
∆IndCoh
˚ pFq
commutes, where the diagonal arrows are the obvious ones induced by adjunction and the obser-
vation 𝑝𝑖 ∆ “ id. By adjunction, this means that the composition map:

F Ñ 𝑝IndCoh
1,˚ pFq𝑝˚,IndCoh
2 pFq Ñ 𝑝IndCoh
1,˚ ∆IndCoh
˚ pFq “ F
is the identity for F. But this composition is clearly the map under consideration.

 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 53

6.8. Indschemes. We now extend the above to the setting of indschemes.

Let PreStk𝑐𝑜𝑛𝑣 denote the category of convergent prestacks. Recall that these are by definition
accessible functors ą´8AffSch𝑜𝑝 Ñ Gpd; the natural functor PreStk Ñ PreStk𝑐𝑜𝑛𝑣 admits fully-
faithful left and right adjoints38 given by Kan extensions. We recall that the composition Sch ãÑ
PreStk Ñ PreStk𝑐𝑜𝑛𝑣 is still fully-faithful; we regard Sch as a subcategory of convergent prestacks
via this functor.

Definition 6.8.1. A reasonable indscheme is an object 𝑆 P PreStk𝑐𝑜𝑛𝑣 that can be written as a

filtered colimit colim𝑖 𝑆𝑖 in PreStk𝑐𝑜𝑛𝑣 of quasi-compact quasi-separated eventually coconnective
schemes 𝑆𝑖 under almost finitely presented closed embeddings.

Let IndSch𝑟𝑒𝑎𝑠 Ď PreStk𝑐𝑜𝑛𝑣 denote the subcategory of reasonable indschemes.

Remark 6.8.2. By [Lur4] Corollary [Link], a closed embedding 𝑇1 ãÑ 𝑇2 P Sch𝑞𝑐𝑞𝑠 is almost finitely
presented if and only if for every 𝑛, 𝜏 ě´𝑛 𝑖˚ pO𝑇1 q P QCohp𝑇2 qě´𝑛 is compact (in this category).
In particular, if 𝑇1 is eventually coconnective, this is equivalent to 𝑖˚ pO𝑇1 q lying in Cohp𝑇2 q Ď
QCohp𝑇2 q.

Remark 6.8.3. Our terminology here is borrowed from [BD1] S7.11, where a similar notion was
introduced for classical indschemes. However, our derived version here is much more restrictive,
because almost finite presentation is much more restrictive than classical finite presentation.

Example 6.8.4. For 𝑌 a smooth affine scheme, its algebraic loop space 𝑌 p𝐾q is reasonable by [Dri]
Theorem 6.3. Indeed, loc. cit. shows that 𝑌 p𝐾q is a filtered colimit of classical39 affine schemes that
are spectra of coherent commutative rings. Moreover, the structure maps are classically of finite
presentation. For such rings, classical finite presentation is equivalent to almost finite presentation,
giving the claim.

6.9. We let Sch𝑟𝑒𝑎𝑠 denote Sch𝑞𝑐𝑞𝑠 X IndSch𝑟𝑒𝑎𝑠 Ď PreStk𝑐𝑜𝑛𝑣 . We refer to objects of this category
as reasonable schemes. Note that any any of the following classes of quasi-compact quasi-separated
schemes is reasonable:
‚ Eventually coconnective.
‚ Locally coherent.
‚ Locally eventually coherent 40 in the sense of [Gai5] S2.

38We remind that PreStk

𝑐𝑜𝑛𝑣 is typically (e.g., in [GR4]) regarded as a full subcategory of PreStk via this right
adjoint. This is because under this embedding, PreStk𝑐𝑜𝑛𝑣 then contains many subcategories of PreStk of interest,
e.g., Sch and IndSch.
We mostly ignore this embedding in what follows and only consider the projection PreStk Ñ PreStk𝑐𝑜𝑛𝑣 , but it
may help guide the reader to keep this in mind.
39There is a subtle point here: that 𝑌 p𝐾q is classical for smooth affine 𝑌 is stated as Conjecture 9.2.10 in [GR3],
and shown when 𝑌 is a group. So we are implicitly assuming that conjecture, or are by fiat interpreting 𝑌 p𝐾q as the
classical indscheme underlying the corresponding DG indscheme.
With that said, the conjecture from [GR3] may be proved as follows. Let 𝑆 be a classical ℵ0 -indscheme that is
Tate-smooth in the sense of [Dri] S6.3.7. By [Dri] Theorem 6.3, and as in [GR3] S9.1, it suffices to show that sucgh 𝑆 is
formally smooth when considered as a DG indscheme. This may be shown using an appropriate variant of Proposition
9.5.2 from [GR3].
As we do not use this result here, we do not supply further details.
40This condition for 𝑆 P Sch ě´𝑛´1
𝑆 Ñ 𝜏 ě´𝑛 𝑆 are almost finitely presented
𝑞𝑐𝑞𝑠 means that the Postnikov maps 𝜏
for 𝑛 " 0.
54 SAM RASKIN

6.10. We define:

IndCoh˚ : IndSch𝑟𝑒𝑎𝑠 Ñ DGCat𝑐𝑜𝑛𝑡

𝑆 ÞÑ IndCoh˚ p𝑆q
p𝑓 : 𝑆 Ñ 𝑇 q ÞÑ p𝑓˚IndCoh : IndCoh˚ p𝑆q Ñ IndCoh˚ p𝑇 qq.
by left Kan extension.
Remark 6.10.1. This definition of IndCoh˚ evidently extends to all convergent prestacks (as the
relevant left Kan extension). However, this definition does not recover the category we are after for
the class of (weakly) renormalizable prestacks introduced below. Therefore, we do not consider this
total left Kan extension here.
6.11. Explicitly, for 𝑆 “ colim𝑖 𝑆𝑖 with 𝑆𝑖 eventually coconnective and quasi-compact quasi-
separated schemes and structural maps being almost finitely presented closed embeddings, we have
IndCoh˚ p𝑆q “ colim𝑖 IndCoh˚ p𝑆𝑖 q P DGCat𝑐𝑜𝑛𝑡 .
As each of the structure functors in this colimit is 𝑡-exact (since pushforward for affine morphisms
is), IndCoh˚ p𝑆q inherits a canonical 𝑡-structure (see e.g. [Ras6] Lemma 5.4.3 (1)). This 𝑡-structure
is characterized by the fact that each pushforward functor IndCoh˚ p𝑆𝑖 q Ñ IndCoh˚ p𝑆q is 𝑡-exact.
In addition, by Remark 6.8.2, each of these functors preserves compact objects. In particular,
IndCoh˚ p𝑆q is compactly generated, and so is IndCoh˚ p𝑆qď0 .
Definition 6.11.1. Cohp𝑆q Ď IndCoh˚ p𝑆q is the subcategory of compact objects. We refer to such
objects as coherent.
We record the following characterization of coherent sheaves for future use.
Lemma 6.11.2. For 𝑆 a reasonable indscheme, F P IndCoh˚ p𝑆q is coherent if and only if F P
IndCoh˚ p𝑆q` and for all 𝑁 " 0 with F P IndCoh˚ p𝑆qě´𝑁 , F is compact in the category IndCoh˚ p𝑆qě´𝑁 .
Proof.
Step 1. First, we remark that this result is immediate from the definitions and Lemma 6.4.1 when
𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 .
Step 2. For convenience, we introduce the following terminology. Suppose C P DGCat𝑐𝑜𝑛𝑡 is equipped
with a right separated 𝑡-structure compatible with filtered colimits. We say F P C is almost compact
if F P C` and for all 𝑁 " 0 with F P Cě´𝑁 , F is compact in Cě´𝑁 .
In this terminology, our goal is to show that compactness is equivalent to almost compactness in
IndCoh˚ p𝑆q. One direction is evident: compactness implies almost compactness as compact objects
in IndCoh˚ p𝑆q are eventually connective.
Step 3. Suppose C, D P DGCat𝑐𝑜𝑛𝑡 are equipped with right separated 𝑡-structures compatible with
filtered colimits. Let 𝐹 : C Ñ D be a 𝑡-exact functor admitting a continuous right adjoint 𝐺 and
such that 𝐹 |C` conservative. We claim that for F P C` with 𝐹 pFq almost compact, F is itself almost
compact.
First, note that the 𝑡-structures are automatically right complete. Therefore, almost compactness
of 𝐹 pFq implies that it is eventually connective. As 𝐹 |C` is conservative and 𝑡-exact, this implies
F is also eventually connective.
By Lemma 3.7.2, 𝐹 |C` is comonadic (c.f. the proof of Proposition 3.7.1). Therefore, for an integer
𝑁 and G P Cě´𝑁 , we have:

HomC pF, Gq “ Tot HomD p𝐹 pFq, 𝐹 p𝐺𝐹 q‚ pGqq P Gpd. (6.11.1)

 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 55

There exists an integer 𝑀 (depending on 𝑁 and F) such that each term in this totalization is an 𝑀 -
truncated groupoid (as 𝐹 pFq is eventually connective). Therefore, the above totalization commutes
with a finite totalization. As HomD p𝐹 pFq, ´q and 𝐹 and 𝐺 all commute with filtered colimits in
Dě´𝑁 , and as filtered colimits commute with finite limits in Gpd, this implies that the left hand
side of (6.11.1) commutes with filtered colimits in the variable GpP Cě´𝑁 q as desired.
Step 4. Let 𝑆 “ colim𝑖 𝑆𝑖 as in the definition of reasonable indscheme. Let 𝛼𝑖 : 𝑆𝑖 Ñ 𝑆 denote the
structural morphisms.
By definition, we have IndCoh˚ p𝑆q “ colim𝑖 IndCoh˚ p𝑆𝑖 q (under pushforwards), with the colimit
being taken in DGCat𝑐𝑜𝑛𝑡 .
We will also need the following variant. Let Cat𝑝𝑟𝑒𝑠 denote the category of presentable categories
and functors commuting with colimits. For any integer 𝑛, we claim:

IndCoh˚ p𝑆qě´𝑛 “ colim IndCoh˚ p𝑆𝑖 qě´𝑛 P Cat𝑝𝑟𝑒𝑠 (6.11.2)

𝑖
with the colimit being taken in Cat𝑝𝑟𝑒𝑠 . Indeed, we have IndCoh˚ p𝑆q “ lim𝑖 IndCoh˚ p𝑆𝑖 q under right
adjoints, where this limit may be formed in any of DGCat𝑐𝑜𝑛𝑡 , Cat𝑝𝑟𝑒𝑠 , and Cat. As these right
adjoints are left 𝑡-exact, we find that IndCoh˚ p𝑆qě´𝑛 “ lim𝑖 IndCoh˚ p𝑆𝑖 qě´𝑛 . The functors in this
limit also admit left adjoints, so the limit coincides with the colimit in Cat𝑝𝑟𝑒𝑠 .
Step 5. We now conclude the argument. Suppose F P IndCoh˚ p𝑆q is almost compact. By assumption,
F P IndCoh˚ p𝑆qě´𝑛 for some 𝑛 P Z.
Write F “ colim𝑗 F𝑗 for F𝑗 P Cohp𝑆q. We obtain F “ colim𝑗 𝜏 ě´𝑛 pF𝑗 q. By almost compactness
of F, there exists an index 𝑗 such that F is a summand of 𝜏 ě´𝑛 pF𝑗 q.
As Cohp𝑆q “ colim𝑖 Cohp𝑆𝑖 q P Cat (c.f. [Lur3] Lemmas [Link]-13), there exists an index 𝑖 and
some Fr 𝑗 P Cohp𝑆𝑖 q such that 𝛼IndCoh pF
𝑖,˚
r 𝑗 q “ F𝑗 . Moreover, by (6.11.2) and [Lur3] Lemma [Link],
after possibly increasing the index 𝑖 there exists F r P IndCoh˚ p𝑆𝑖 qě´𝑛 (a summand of 𝜏 ě´𝑛 pF𝑗 q)
with:

pFq “ F
IndCoh r
𝛼𝑖,˚
(as summands of F𝑗 ).
By Step 3, F
r is almost compact in IndCoh˚ p𝑆𝑖 q. As in Step 1, this means F
r P Cohp𝑆𝑖 q. As 𝛼IndCoh
𝑖,˚
admits a continuous right adjoint, we obtain the result.

6.12. The following technical result is convenient for comparing different possible presentations of
a reasonable indscheme.
Proposition 6.12.1. Let 𝑆 “ colim𝑖PI 𝑆𝑖1 “ colim𝑗PJ 𝑆𝑗2 be two expressions of 𝑆 as a reasonable
indscheme, i.e., these colimits are filtered, 𝑆𝑖1 , 𝑆𝑗2 P ą´8Sch𝑞𝑐𝑞𝑠 and the structure maps in each of
these colimits are almost finitely presented. Let 𝛼𝑖 : 𝑆𝑖1 Ñ 𝑆 and 𝛽𝑗 : 𝑆𝑗2 Ñ 𝑆 denote the structure
maps.
Then for any choice of indices 𝑖 P I and 𝑗 P J such that the map 𝛼𝑖1 : 𝑆𝑖1 Ñ 𝑆 factors as
𝜄 𝛽𝑗
𝑆𝑖1 Ñ
Ý 𝑆𝑗2 ÝÑ 𝑆, the map 𝜄 is almost of finite presentation.
Proof. By Remark 6.8.2, it suffices to show that 𝜄˚ pO𝑆 1 q P Cohp𝑆𝑗2 q. Clearly this object lies in
𝑖
IndCoh˚ p𝑆𝑗2 q` . By Lemma 6.11.2 and Step 3 from its proof, it therefore suffices to show that
IndCoh 𝜄 pO q P Cohp𝑆q, but this is clear as 𝛽 IndCoh 𝜄 “ 𝛼IndCoh .
𝛽𝑗,˚ ˚ 𝑆𝑖1 𝑗,˚ ˚ 𝑖,˚

56 SAM RASKIN

6.13. Reasonable schemes. Observe that we have a functor:

QCohp´q : Sch𝑞𝑐𝑞𝑠 Ñ DGCat𝑐𝑜𝑛𝑡

encoding pushforward of quasi-coherent sheaves. By left Kan extension, there is a canonical natural
transformation:

Ψ : IndCoh˚ p´q|Sch𝑟𝑒𝑎𝑠 Ñ QCohp´q|Sch𝑟𝑒𝑎𝑠

of functors Sch𝑟𝑒𝑎𝑠 Ñ DGCat𝑐𝑜𝑛𝑡 .
»
Lemma 6.13.1. For every 𝑆 P Sch𝑟𝑒𝑎𝑠 , Ψ induces an equivalence IndCoh˚ p𝑆q` Ý Ñ QCohp𝑆q` .
˚
Moreover, Ψ identifies Cohp𝑆q Ď IndCoh p𝑆q` (as defined in Definition 6.11.1) with the subcategory
of cohomologically bounded objects in QCohp𝑆q that are compact in QCohp𝑆qě´𝑁 for all 𝑁 " 0.
Proof. In what follows, for C a DG category with a 𝑡-structure and 𝑛 ě 0, we let Cr´𝑛,0s denote the
subcategory C of objects in cohomological degrees r´𝑛, 0s.
Because 𝑆 is reasonable, we have 𝑆 “ colim 𝑆𝑖 a filtered colimit under almost finitely presented
closed embeddings. As in Step 4 from the proof of Lemma 6.11.2, we have IndCoh˚ p𝑆qr´𝑛,0s “
colim𝑖 IndCoh˚ p𝑆𝑖 qr´𝑛,0s with the colimit being taken in Cat𝑝𝑟𝑒𝑠 (the category of presentable cate-
gories and functors commuting with colimits).
Recall that e.g. QCohp𝑆qr´𝑛,0s “ QCohp𝜏 ě´𝑛 𝑆qr´𝑛,0s . Therefore, we obtain:
Ψ
»
Ñ colim QCoh˚ p𝑆𝑖 qr´𝑛,0s “
IndCoh˚ p𝑆qr´𝑛,0s “ colim IndCoh˚ p𝑆𝑖 qr´𝑛,0s Ý
𝑖 𝑖
˚ ě´𝑛 r´𝑛,0s ˚ ě´𝑛 r´𝑛,0s
colim QCoh p𝜏 𝑆𝑖 q “ colim IndCoh p𝜏 𝑆𝑖 q “ IndCoh˚ p𝜏 ě´𝑛 𝑆qr´𝑛,0s
𝑖 𝑖
with all colimits being taken in Cat𝑝𝑟𝑒𝑠 , and where we have used that 𝜏 ě´𝑛 𝑆 “ colim𝑖 𝜏 ě´𝑛 𝑆𝑖 .
By right completeness of the 𝑡-structures on IndCoh˚ p𝑆q and QCohp𝑆q, we obtain the claims.

6.14. Proper morphisms. We now discuss proper morphisms. We refer to [Lur4] Part II for an
extensive discussion of such morphisms in derived algebraic geometry. However, we take a more
restrictive definition than loc. cit. (to simplify terminology): we say 𝑓 : 𝑆 Ñ 𝑇 P Sch𝑞𝑐𝑞𝑠 is proper
if it is proper in the sense of [Lur4] and almost of finite presentation (only finite type is required
in loc. cit.).
Lemma 6.14.1. Suppose 𝑓 : 𝑆 Ñ 𝑇 P Sch𝑟𝑒𝑎𝑠 is proper. Then 𝑓˚IndCoh admits a continuous right
adjoint 𝑓 ! .
Proof. This is immediate from Lemma 6.13.1 and [Lur4] Theorem [Link].

Lemma 6.14.2. Suppose we are given a Cartesian diagram of schemes:
𝜙
𝑆1 / 𝑇1

𝜓 𝑔
 𝑓 
𝑆 /𝑇
with all terms lying in Sch𝑟𝑒𝑎𝑠 41 and 𝑓 proper.
41This is not automatic for 𝑆 1 “ 𝑆 ˆ 𝑇 1 even if 𝑆, 𝑇 , and 𝑇 1 lie in Sch
𝑇 𝑟𝑒𝑎𝑠 .
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 57

(1) The natural map:

𝜓˚IndCoh 𝜙! Ñ 𝑓 ! 𝑔˚IndCoh
is an isomorphism.
(2) Suppose that 𝑔 is flat. Then the natural42 map:

𝜓 ˚,IndCoh 𝑓 ! Ñ 𝜙! 𝑔 ˚,IndCoh
is an isomorphism.
Proof. The same argument from [Gai5] Proposition 3.4.2 applies for (1). Similarly, again using
Lemma 6.13.1, the same argument as in [Gai5] Proposition 7.1.6 applies43 for (2).

6.15. Flatness. We say a morphism 𝑓 : 𝑇1 Ñ 𝑇2 P PreStk𝑐𝑜𝑛𝑣 is flat if for any eventually cocon-
nective affine 𝑆 P ą´8AffSch and any map 𝑆 Ñ 𝑇2 , the fiber product 𝑇1 ˆ𝑇2 𝑆 lies in ą´8Sch𝑞𝑐𝑞𝑠
and its structure map to 𝑆 is flat. Similarly, we say 𝑓 is a flat cover if it is flat and 𝑇1 ˆ𝑇2 𝑆 Ñ 𝑆
is faithfully flat.
Remark 6.15.1. As our definition requires flat morphisms to be schematic (in the relevant sense
for convergent prestacks) and to be quasi-compact quasi-separated, it is much more stringent than
usual notions of flatness. We hope the reader will forgive this abuse, which we find unburdens the
terminology and notation to some degree.
Clearly flat morphisms (resp. covers) are closed under compositions and base-change.
Remark 6.15.2. If 𝑓 : 𝑆 Ñ 𝑇 is flat and 𝑇 is a reasonable indscheme, then 𝑆 is a reasonable
indscheme as well.
6.16. We will now study a variety of base-change results for flat morphisms.
Lemma 6.16.1. Let 𝑓 : 𝑆 Ñ 𝑇 P IndSch𝑟𝑒𝑎𝑠 be a flat morphism.
(1) 𝑓˚IndCoh admits a left adjoint 𝑓 ˚,IndCoh .
(2) Suppose 𝑇 “ colim𝑖 𝑇𝑖 as in the definition of reasonable indscheme. For any index 𝑖, let
𝑆𝑖 :“ 𝑆 ˆ𝑇 𝑇𝑖 and denote the relevant structural maps as:

𝑓𝑖
𝑆𝑖 / 𝑇𝑖

𝛽𝑖 𝛼𝑖
 𝑓 
𝑆 / 𝑇.
Let 𝛼𝑖! and 𝛽𝑖! denote the (continuous) right adjoints to 𝛼𝑖,˚
IndCoh and 𝛽 IndCoh .
𝑖,˚

42We remark that the existence of this map depends on (1).

43There is one small modification to make. The argument in [Gai5] uses Proposition 3.6.11 from loc. cit., which
in turn uses Lemma 3.6.13 from loc. cit. The argument from loc. cit. for this lemma does not literally work: [Gai5]
reduces to the classical case by means which are not available to us here. The difference in the argument is not
significant, but we provide the details below.
The lemma in question (in our setup) asserts that for flat 𝑓 : 𝑆 P Sch𝑟𝑒𝑎𝑠 and F P QCohp𝑆q flat, the functor
F b ´IndCoh˚ p𝑆q Ñ IndCoh˚ p𝑆q is 𝑡-exact. (Here we are using the natural action of QCohp𝑆q on IndCoh˚ p𝑆q obtained
by ind-extension from the action of Perfp𝑆q on Cohp𝑆q.)
In our setup, Zariski descent (Lemma 6.7.1 below, which is independent of Lemma 6.14.2), reduces to considering
the case where 𝑆 is affine. Then the result is immediate from Lazard’s theorem in the derived setup: see [Lur3]
Theorem [Link].
58 SAM RASKIN

Then the natural map:

IndCoh !
𝑓𝑖,˚ 𝛽𝑖 Ñ 𝛼𝑖! 𝑓˚IndCoh
is an isomorphism.
(3) The natural map:

𝑓𝑖˚,IndCoh 𝛼𝑖! Ñ 𝛽𝑖! 𝑓 ˚,IndCoh

is an isomorphism.
(4) Given a Cartesian diagram in IndSch𝑟𝑒𝑎𝑠 :
𝜙
𝑆1 / 𝑇1

𝜓 𝑔
 𝑓 
𝑆 /𝑇
(with 𝑓 flat), the natural morphism:

𝑓 ˚,IndCoh 𝑔˚IndCoh Ñ 𝜓˚IndCoh 𝜙˚,IndCoh (6.16.1)

is an isomorphism.
Proof. By Lemma 6.14.2 (1)-(2), the adjoint functors:

𝑓𝑖˚,IndCoh : IndCoh˚ p𝑇𝑖 q Õ IndCoh˚ p𝑆𝑖 q : 𝑓𝑖,˚

IndCoh
(6.16.2)
are (canonically) compatible with the structure functors in the limits IndCoh p𝑆q “ lim𝑖 IndCoh˚ p𝑆𝑖 q
˚

and IndCoh˚ p𝑇 q “ lim𝑖 IndCoh˚ p𝑇𝑖 q (the limits being under upper-! functors), and therefore induce
an adjunction p𝑓 ˚,IndCoh , 𝑓˚IndCoh q satisfying (2) and (3).
For (4), we are immediately reduced to the case where 𝑇 1 P ą´8Sch𝑞𝑐𝑞𝑠 (as IndCoh˚ p𝑇 1 q is neces-
sarily generated under colimits by objects pushed forward from eventually coconnective schemes).
𝑔 𝛼
𝑖
Then 𝑔 factors as 𝑇 1 ÑÝ 𝑇𝑖 ÝÑ 𝑇 . By Lemma 6.6.2, we are reduced to the case where 𝑇 1 “ 𝑇𝑖 and
𝑔 “ 𝛼𝑖 .
In this case, the claim follows from the fact that the adjoint functors (6.16.2) are (canoni-
cally) compatible with the structural functors in the colimits IndCoh˚ p𝑆q “ colim𝑖 IndCoh˚ p𝑆𝑖 q and
IndCoh˚ p𝑇 q “ colim𝑖 IndCoh˚ p𝑇𝑖 q (in DGCat𝑐𝑜𝑛𝑡 , under lower-* functors) by Lemma 6.6.2.

Corollary 6.16.2. Let 𝑓 : 𝑆 Ñ 𝑇 P IndSch𝑟𝑒𝑎𝑠 be flat. Then 𝑓 ˚,IndCoh : IndCoh˚ p𝑇 q Ñ IndCoh˚ p𝑆q
is 𝑡-exact.
Proof. We use the notation from Lemma 6.16.1 (2). Then IndCoh˚ p𝑇 qď0 is generated under colimits
IndCoh pFq for F P IndCoh˚ p𝑇 qď0 . We then have:
by objects of the form 𝛼𝑖,˚ 𝑖

IndCoh IndCoh ˚,IndCoh

𝑓 ˚,IndCoh p𝛼𝑖,˚ pFqq “ 𝛽𝑖,˚ 𝑓𝑖 pFq
by Lemma 6.16.1 (4) and this is in IndCoh˚ p𝑆qď0 because 𝑓𝑖 is a flat map of schemes. Therefore,
𝑓 ˚,IndCoh is right 𝑡-exact.
For left 𝑡-exactness, suppose F P IndCoh˚ p𝑇 qě0 . To see 𝑓 ˚,IndCoh pFq P IndCoh˚ p𝑆qě0 , it is equiva-
lent to show that 𝛽𝑖! 𝑓 ˚,IndCoh pFq P IndCoh˚ p𝑆𝑖 qě0 for all 𝑖. But by Lemma 6.14.2 (2), we have:

𝛽𝑖! 𝑓 ˚,IndCoh pFq “ 𝑓𝑖˚,IndCoh 𝛼𝑖! pFq.

 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 59

Then 𝛼𝑖! pFq P IndCoh˚ p𝑇𝑖 qě0 by assumption on F, so the same is true after applying 𝑓𝑖˚,IndCoh to it.


6.17. We say that a morphism 𝑓 : 𝑆 Ñ 𝑇 of reasonable indschemes is ind-proper if for some

(equivalently,44 any) presentations 𝑆 “ colim𝑖PI 𝑆𝑖 and 𝑇 “ colim𝑗PJ 𝑇𝑗 as in the definitions of
reasonable indschemes, and any index 𝑖 P I there exists 𝑗 P J such that 𝑆𝑖 Ñ 𝑆 Ñ 𝑇 factors
through a proper morphism 𝑆𝑖 Ñ 𝑇𝑗 .

Lemma 6.17.1. Let 𝑓 : 𝑆 Ñ 𝑇 be an ind-proper morphism of reasonable indschemes. Then 𝑓˚IndCoh

admits a continuous right adjoint 𝑓 ! .

Proof. As IndCoh˚ “ IndpCohq here, the continuity of the right adjoint 𝑓 ! is equivalent to 𝑓˚IndCoh
preserving compacts. Then we are immediately reduced to the case where 𝑆, 𝑇 P ą´8Sch𝑞𝑐𝑞𝑠 , which
is covered by Lemma 6.14.1.


We have the following (somewhat partial) generalization of Lemma 6.14.2.

Lemma 6.17.2. Suppose we are given a Cartesian diagram of reasonable indschemes:

𝜙
𝑆1 / 𝑇1

𝜓 𝑔
 𝑓 
𝑆 /𝑇

with 𝑓 ind-proper and 𝑔 flat.

(1) The natural map:

𝜓˚IndCoh 𝜙! Ñ 𝑓 ! 𝑔˚IndCoh
is an isomorphism.
(2) The natural map:

𝜓 ˚,IndCoh 𝑓 ! Ñ 𝜙! 𝑔 ˚,IndCoh
is an isomorphism.

Proof. Writing 𝑆 “ colim𝑗 𝑆𝑗 as in the definition of reasonable indscheme, both statements imme-
diately reduce to the case where 𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 .
Now take 𝑇 “ colim𝑖 𝑇𝑖 as in the definition of reasonable indscheme. By assumption on 𝑓 , the
𝑓 𝛼𝑗
map 𝑓 factors as 𝑆 ÝÑ 𝑇𝑗 ÝÑ 𝑇 for 𝛼𝑗 : 𝑇𝑗 Ñ 𝑇 the structure map and 𝑓 proper. By Lemma 6.14.2
(1) and (2), both statements reduce to the case where 𝑆 “ 𝑇𝑗 and 𝑓 “ 𝛼𝑗 (Here the results follow
from Lemma 6.16.1 (2)-(3).


44C.f. Proposition 6.12.1.

60 SAM RASKIN

6.18. We record the following basic result for later use.

Lemma 6.18.1. For a flat cover 𝑓 : 𝑆 Ñ 𝑇 P IndSch𝑟𝑒𝑎𝑠 , F P IndCoh˚ p𝑇 q lies in Cohp𝑇 q if and
only 𝑓 ˚,IndCoh pFq lies in Cohp𝑆q.
Proof. First, note that 𝑓 ˚,IndCoh is conservative. Indeed, if 𝑇 P ą´8Sch𝑞𝑐𝑞𝑠 , this follows from Propo-
sition 6.7.2, and the general case results from this using Lemma 6.14.2 (2). (See also Theorem
6.25.1).
Therefore, if 𝑓 ˚,IndCoh pFq P Cohp𝑆q, we find in particular that F P IndCoh˚ p𝑇 q` . Now we obtain
the result from Lemma 6.11.2 and Step 3 from the proof of loc. cit.

6.19. We remark that because we can ˚-pullback along flat maps, the functor IndCoh˚ : IndSch𝑟𝑒𝑎𝑠 Ñ
DGCat𝑐𝑜𝑛𝑡 extends (again by [GR4] Theorem V.[Link]) to a functor:

CorrpIndSch𝑟𝑒𝑎𝑠 q𝑎𝑙𝑙;𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡 .

6.20. Relationship to 𝐷-modules. Recall the functor 𝐷˚ : IndSch𝑟𝑒𝑎𝑠 Ñ DGCat𝑐𝑜𝑛𝑡 constructed
in [Ras3]. We will construct a canonical natural transformation:

IndCoh˚ Ñ 𝐷˚
that can be thought of as inducing an ind-coherent sheaf to a 𝐷-module.
Each of these functors is by definition left Kan extended from ą´8Sch𝑞𝑐𝑞𝑠 , so it suffices to
define the natural transformation for the restrictions of these functors here. Moreover, each of
these functors is a Zariski sheaf, so it suffices to define the natural transformation on ą´8AffSch.
This in turn is equivalent to specifying a compatible sequence of natural transformations on each
ě´𝑛AffSch “ Propě´𝑛AffSch q.
𝑓𝑡
By definition, 𝐷˚ |ě´𝑛AffSch is right Kan extended from ě´𝑛AffSch𝑓 𝑡 . Therefore, we need to spec-
ify the natural transformation on ě´𝑛AffSch𝑓 𝑡 compatibly over all 𝑛. Here we define our natural
transformation as the (“right45”) 𝐷-module induction functor ind : IndCoh Ñ 𝐷 constructed in
[GR4].
Remark 6.20.1. By construction, this natural transformation upgrades to a natural transformation
of lax symmetric monoidal functors; see S6.30 below.
6.21. Weakly renormalizable prestacks. We now introduce a convenient class of prestacks.46
Definition 6.21.1. A convergent prestack 𝑆 P PreStk𝑐𝑜𝑛𝑣 is weakly renormalizable if there exists a
flat covering map 𝑇 Ñ 𝑆 with 𝑇 P IndSch𝑟𝑒𝑎𝑠 .
We let PreStk𝑤.𝑟𝑒𝑛 Ď PreStk𝑐𝑜𝑛𝑣 denote the subcategory of weakly renormalizable prestacks.
6.22. Morphisms. We now introduce some classes of morphisms between weakly renormalizable
prestacks.
Definition 6.22.1. (1) A morphism 𝑓 : 𝑆1 Ñ 𝑆2 P PreStk𝑤.𝑟𝑒𝑛 is reasonable indschematic if for
any flat morphism 𝑇 Ñ 𝑆2 with 𝑇 P IndSch𝑟𝑒𝑎𝑠 , the fiber product 𝑇1 ˆ𝑇2 𝑆 P PreStk𝑐𝑜𝑛𝑣 is
a reasonable indscheme.
(2) A morphism 𝑆1 Ñ 𝑆2 P PreStk𝑤.𝑟𝑒𝑛 is locally flat if for any 𝑇 P IndSch𝑟𝑒𝑎𝑠 and any flat
morphism 𝑇 Ñ 𝑆1 , the composition 𝑇 Ñ 𝑆1 Ñ 𝑆2 is flat.
45As opposed to left.
46It is formally convenient to not have to sheafify in forming quotients such as 𝑋{𝐻, so we prefer to work with
prestacks.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 61

Example 6.22.2. Any morphism from a reasonable indschemes to a weakly renormalizable prestack is
reasonable indschematic. In particular, any morphism between reasonable indschemes is reasonable
indschematic.
Remark 6.22.3. It is straightforward to show that to check 𝑆1 Ñ 𝑆2 is locally flat, it suffices to
check the condition from the definition for some flat cover 𝑇 of 𝑆1 . In particular, a morphism of
reasonable indschemes is locally flat if and only if it is flat.
Example 6.22.4. If 𝐻 is a classical affine group scheme, B𝐻 Ñ Specp𝑘q is locally flat.
The following is immediate:
Lemma 6.22.5. Reasonable indschematic and locally flat morphisms are closed under compo-
sitions. Any base-change of a reasonable indschematic morphism by a locally flat morphism is
again reasonable indschematic, and any base-change of a locally flat morphism by a reasonable
indschematic morphism is again locally flat.
6.23. Set-theoretic remarks. In what follows, we will not explicitly address certain set-theoretic
issues. More precisely, we will want to form limits e.g. over all reasonable indschemes flat over a
given weakly renormalizable prestack. This indexing category is not essentially small, so there are
set-theoretic issues.
To address these, fix a regular cardinal 𝜅 and replace “flat” everywhere by “flat and locally
𝜅-presented.” One should understand weakly renormalizable prestacks in this sense (i.e., these are
prestacks admitting a locally 𝜅-presented flat cover by a reasonable indscheme), and so on.
As will follow from Theorem 6.25.1, all of our constructions are invariant under extension of 𝜅,
i.e., if 𝑆 is a weakly renormalizable prestack relative to 𝜅 and 𝜅1 ě 𝜅 is another regular cardinal,
then the categories IndCoh˚ defined using 𝜅 and 𝜅1 coincide.
Again, since the cutoff 𝜅 plays such a minor role, in order to simplify the exposition we do not
mention it again.
6.24. IndCoh˚ on weakly renormalizable prestacks. Define PreStk𝑤.𝑟𝑒𝑛,𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 as the 1-full sub-
category of PreStk𝑤.𝑟𝑒𝑛 where we only allow locally flat morphisms, and define IndSch𝑟𝑒𝑎𝑠,𝑓 𝑙𝑎𝑡 simi-
larly.
Definition 6.24.1. IndCoh˚ : PreStk𝑜𝑝𝑤.𝑟𝑒𝑛,𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡 is the right Kan extension of the
𝑜𝑝
functor IndSch𝑟𝑒𝑎𝑠,𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡 (which sends 𝑆 to IndCoh˚ p𝑆q and sends flat 𝑓 : 𝑇1 Ñ 𝑇2 to
𝑓 ˚,IndCoh ).
By [GR4] Theorem V.[Link],47 the above construction upgrades canonically to a functor:

IndCoh˚ : CorrpPreStk𝑤.𝑟𝑒𝑛 q𝑟𝑒𝑎𝑠.𝑖𝑛𝑑𝑠𝑐ℎ;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡 .

Here “𝑟𝑒𝑎𝑠.𝑖𝑛𝑑𝑠𝑐ℎ” is shorthand for reasonable indschematic and “𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡” is shorthand for locally
flat. Therefore, the notation indicates that for a reasonable indschematic morphism 𝑓 : 𝑆 Ñ 𝑇
between weakly renormalizable prestacks, we have a pushforward functor 𝑓˚IndCoh : IndCoh˚ p𝑆q Ñ
IndCoh˚ p𝑇 q; for 𝑓 locally flat, we have a functor 𝑓 ˚,IndCoh ; and the two satisfy base-change.
Remark 6.24.2. Using the 2-category of correspondences as in [GR4], one can further encode that
𝑓 ˚,IndCoh is left adjoint to 𝑓˚IndCoh for flat 𝑓 .
Remark 6.24.3. By [Gai5] Proposition 11.4.3, for 𝑆 weakly renormalizable and locally almost of
finite type, IndCoh˚ p𝑆q is canonically isomorphic to the usual category IndCohpp𝑆q.
47We remark that the hypotheses from loc. cit. are trivially verified in this setting, c.f. Lemma 6.22.5.
62 SAM RASKIN

6.25. The following result justifies the definition of IndCoh˚ for weakly renormalizable prestacks.
Theorem 6.25.1. IndCoh˚ satisfies flat descent on PreStk𝑤.𝑟𝑒𝑛 .
Proof. Let 𝑆 P PreStk𝑤.𝑟𝑒𝑛 be given and let 𝑓 : 𝑇 Ñ 𝑆 be a flat cover. We need to show that:

IndCoh˚ p𝑆q Ñ Tot𝑠𝑒𝑚𝑖 pIndCoh˚ p𝑇 ˆ𝑆 ‚`1 qq

is an isomorphism.
We proceed in increasing generality.
Step 1. First, suppose 𝑆 P IndSch𝑟𝑒𝑎𝑠 .
Let 𝑆 “ colim𝑖 𝑆𝑖 as in the definition of reasonable indscheme. We then have:

𝑃 𝑟𝑜𝑝.6.7.2
IndCoh˚ p𝑆q “ lim IndCoh˚ p𝑆𝑖 q “ lim Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑆𝑖 ˆ𝑆 𝑇 ˆ𝑆 ‚`1 q “
𝑖,upper-! 𝑖,upper-! upper-˚

Tot𝑠𝑒𝑚𝑖 lim IndCoh˚ p𝑆𝑖 ˆ𝑆 𝑇 ˆ𝑆 ‚`1 q “ Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑇 ˆ𝑆 ‚`1 q

upper-˚ 𝑖,upper-! upper-˚

as desired, where we have used Lemma 6.14.2 (2) to commute the limits.
Step 2. Next, suppose 𝑆 is a general weakly renormalizable prestack and 𝑇 is a reasonable ind-
scheme.
We denote the functor under consideration by:

𝐹 : IndCoh˚ p𝑆q :“ lim IndCoh˚ p𝑈 q Ñ Tot𝑠𝑒𝑚𝑖 pIndCoh˚ p𝑇 ˆ𝑆 ‚`1 qq

𝑈 PIndSch𝑟𝑒𝑎𝑠
𝑈 Ñ𝑆 flat
We will show 𝐹 is an equivalence by explicitly constructing an inverse functor 𝐺.
Namely, we have a functor (induced by ˚-pullback):

Tot𝑠𝑒𝑚𝑖 pIndCoh˚ p𝑇 ˆ𝑆 ‚`1 qq Ñ Tot𝑠𝑒𝑚𝑖 lim IndCoh˚ p𝑈 ˆ𝑆 𝑇 ˆ𝑆 ‚`1 q

𝑈 PIndSch𝑟𝑒𝑎𝑠
𝑈 Ñ𝑆 flat

Exchanging the order of limits on the right hand side and noting that 𝑈 ˆ𝑆 𝑇 ˆ𝑆 ‚`1 is the Cech
nerve of the flat cover 𝑈 ˆ𝑆 𝑇 Ñ 𝑈 P IndSch𝑟𝑒𝑎𝑠 , the previous step implies that the right hand side
is canonically isomorphic to:

lim IndCoh˚ p𝑈 q “: IndCoh˚ p𝑆q.

𝑈 PIndSch𝑟𝑒𝑎𝑠
𝑈 Ñ𝑆 flat

Therefore, we obtain our functor 𝐺 : Tot𝑠𝑒𝑚𝑖 pIndCoh˚ p𝑇 ˆ𝑆 ‚`1 qq Ñ IndCoh˚ p𝑆q.

To verify that 𝐺 and 𝐹 are inverses, it suffices to show 𝐺𝐹 » id and 𝐹 𝐺 » id. We construct
such isomorphisms by straightforward means below.
First, note that for 𝑈 P IndSch𝑟𝑒𝑎𝑠 equipped with a flat map to 𝑆, we have a projection morphism
of augmented simplicial prestacks:

𝑈 ˆ 𝑇 ˆ𝑆 ‚`1 Ñ 𝑇 ˆ𝑆 ‚`1 .
𝑆
This is functorial in 𝑈 , so passing to the limits and using the augmentation to obtain the horizontal
arrows, we get the commutative diagram:
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 63

IndCoh˚ p𝑆q
𝐹 / Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑇 ˆ𝑆 ‚`1 q


lim IndCoh˚ p𝑈 q
» / Tot𝑠𝑒𝑚𝑖 lim IndCoh˚ p𝑈 ˆ𝑆 𝑇 ˆ𝑆 ‚`1 q
𝑈 PIndSch𝑟𝑒𝑎𝑠 𝑈 PIndSch𝑟𝑒𝑎𝑠
𝑈 Ñ𝑆 flat 𝑈 Ñ𝑆 flat

By definition, 𝐺 is the composition of the right horizontal arrow and the inverse to the bottom
arrow. The commutativity of this diagram therefore gives 𝐺𝐹 » id.
To construct an isomorphism 𝐹 𝐺 » id, it suffices to do so after further composition with the
functor:
»
𝜂 : Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑇 ˆ𝑆 ‚`1 q Ý
Ñ Tot𝑠𝑒𝑚𝑖 Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑇 ˆ𝑆 ‚`‚`2 q.
Note that 𝜂 is an isomorphism because 𝑇 is a reasonable indscheme.
The target of 𝜂 is the double totalization of the bi-semi-cosimplicial object obtained from
IndCoh˚ p𝑇 ˆ𝑆 ‚`1 q by restricting along the join (alias: concatenation) map join : Δ𝑖𝑛𝑗 ˆΔ𝑖𝑛𝑗 Ñ Δ𝑖𝑛𝑗 .
Moreover, by construction, the functor 𝜂𝐹 𝐺 is the natural map in such a situation (from the limit
of a functor to the limit of its restriction to another category).
Let 𝑝1 : Δ𝑖𝑛𝑗 ˆ Δ𝑖𝑛𝑗 Ñ Δ𝑖𝑛𝑗 be the first projection. There is an evident natural transformation
𝑝1 Ñ join inducing a commutative diagram:

Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑇 ˆ𝑆 ‚`1 q

»
 ,
Tot𝑠𝑒𝑚𝑖 Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑇 ˆ𝑆 ‚`1 q / Tot𝑠𝑒𝑚𝑖 Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑇 ˆ𝑆 ‚`‚`2 q.

The diagonal arrow is 𝜂𝐹 𝐺 by the above discussion, while the left and bottom arrows compose to
give 𝜂. This gives the claim.
Step 3. Finally, we treat the general case in which 𝑆 and 𝑇 are both weakly renormalizable prestacks.
By assumption on 𝑆, there exists 𝑆 1 P IndSch𝑟𝑒𝑎𝑠 and 𝑆 1 Ñ 𝑆 a flat cover. We then obtain a
commutative diagram:

IndCoh˚ p𝑆q / Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑇 ˆ𝑆 ‚`1 q

 
Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑆 1,ˆ𝑆 ‚`1 q / Tot𝑠𝑒𝑚𝑖 Tot𝑠𝑒𝑚𝑖 IndCoh˚ p𝑆 1,ˆ𝑆 ‚`1 ˆ𝑆 𝑇 ˆ𝑆 ‚`1 q

The left, bottom and right arrows are isomorphisms by the previous step, so the top arrow is as
well.

Corollary 6.25.2. Let 𝑆 “ 𝑇 {𝐻 for 𝐻 a classical affine group scheme acting on 𝑇 P IndSch𝑟𝑒𝑎𝑠 .
Then the functor:

IndCoh˚ p𝑆q Ñ IndCoh˚ p𝑇 q𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒

is an equivalence.
64 SAM RASKIN

Proof. Clear from the Theorem 6.25.1 (e.g., using Proposition 6.36.4 and Proposition 6.35.2 (2) to
convert IndCoh˚ on the relevant products to tensor products).

6.26. 𝑡-structures. Let 𝑆 P PreStk𝑤.𝑟𝑒𝑛 be given. Then IndCoh˚ p𝑆q has a unique 𝑡-structure such
that for every 𝑈 P IndSch𝑟𝑒𝑎𝑠 and flat 𝑈 Ñ 𝑆, the pullback functor IndCoh˚ p𝑆q Ñ IndCoh˚ p𝑈 q is
𝑡-exact.
Indeed, by definition, we have:

IndCoh˚ p𝑆q “ lim IndCoh˚ p𝑈 q

𝑈 PIndSch𝑟𝑒𝑎𝑠
𝑈 Ñ𝑆 flat
and all of the structural functors are 𝑡-exact by Corollary 6.16.2.
6.27. Coherence. Next, for 𝑆 P PreStk𝑤.𝑟𝑒𝑛 , we define Cohp𝑆q Ď IndCoh˚ p𝑆q as:

Cohp𝑆q “ lim Cohp𝑈 q.

𝑈 PIndSch𝑟𝑒𝑎𝑠
𝑈 Ñ𝑆 flat
Clearly ˚-pullback along locally flat maps preserve Coh.
6.28. Renormalizable prestacks.
Definition 6.28.1. 𝑆 P PreStk𝑐𝑜𝑛𝑣 is renormalizable if there exists a flat cover 𝑓 : 𝑇 Ñ 𝑆 with
𝑇 P IndSch𝑙𝑎𝑓 𝑡 an indscheme locally almost of finite type.
For 𝑆 renormalizable, we let IndCoh˚𝑟𝑒𝑛 p𝑆q denote IndpCohp𝑆qq.
Remark 6.28.2. One might prefer a definition in greater generality (e.g., without finiteness hy-
potheses on 𝑇 ). However, this definition suffices for our applications, and this finiteness hypothesis
simplifies the theory (essentially by Lemma 6.28.4 below).
Example 6.28.3. By Lemma 6.18.1 and Theorem 6.25.1, F P IndCoh˚ p𝑆q is coherent if and only if
its ˚-pullback to some flat cover is so. In particular, 𝑆 “ 𝑇 {𝐻 for 𝑇 P IndSch𝑙𝑎𝑓 𝑡 and 𝐻 a classical
affine group scheme acting on 𝑇 , then 𝑆 is renormalizable with IndCoh˚𝑟𝑒𝑛 p𝑆q “ IndCoh˚ p𝑇 q𝐻,𝑤 .
Lemma 6.28.4. For 𝑆 a renormalizable prestack, coherent objects in IndCoh˚ p𝑆q are closed under
truncations.
Proof. By Theorem 6.25.1 and the definition, this reduces to the case of indschemes locally almost
of finite type where it is clear.

Proposition 6.28.5. Let 𝑆 be a renormalizable prestack. Define a 𝑡-structure on IndCoh˚𝑟𝑒𝑛 p𝑆q by
taking connective objects to be generated under colimits by Cohp𝑆q X IndCoh˚ p𝑆qď0 .
Then then canonical functor IndCoh˚𝑟𝑒𝑛 p𝑆q Ñ IndCoh˚ p𝑆q is 𝑡-exact and induces an equivalence
on eventually coconnective subcatgories.
Proof. Lemma 6.28.4 implies that Cohp𝑆q ãÑ IndCoh˚𝑟𝑒𝑛 p𝑆q is closed under truncations for this
𝑡-structure. This clearly implies that IndCoh˚𝑟𝑒𝑛 p𝑆q Ñ IndCoh˚ p𝑆q is 𝑡-exact.
Next, observe that if F P Cohp𝑆q, then F is compact in IndCoh˚ p𝑆qě´𝑁 for all 𝑁 " 0. Indeed, this
follows from Step 3 from the proof of Lemma 6.11.2. Combined with the fact that compact objects
in IndCoh˚𝑟𝑒𝑛 p𝑆q are closed under truncations, this implies that IndCoh˚𝑟𝑒𝑛 p𝑆q` Ñ IndCoh˚ p𝑆q` is
fully-faithful.
Finally, an argument as in Lemma 5.19.1 shows that the functor is essentially surjective.

 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 65

6.29. Let 𝑓 : 𝑆1 Ñ 𝑆2 P PreStk𝑟𝑒𝑛 be a reasonable indschematic morphism. Then 𝑓˚IndCoh :

IndCoh˚ p𝑆1 q Ñ IndCoh˚ p𝑆2 q is left 𝑡-exact. Therefore, by Lemma 6.28.4 and Proposition 6.28.5,
there exists a unique left 𝑡-exact functor 𝑓˚IndCoh𝑟𝑒𝑛 : IndCoh˚𝑟𝑒𝑛 p𝑆1 q Ñ IndCoh˚𝑟𝑒𝑛 p𝑆2 q fitting into a
commutative diagram:

𝑓˚IndCoh𝑟𝑒𝑛
IndCoh˚𝑟𝑒𝑛 p𝑆1 q / IndCoh˚ p𝑆2 q
𝑟𝑒𝑛

 𝑓˚IndCoh 
IndCoh˚ p𝑆1 q / IndCoh˚ p𝑆2 q

(with vertical arrows the canonical functors).

Similarly, if 𝑓 is locally flat, then 𝑓 ˚,IndCoh is 𝑡-exact, so there is a unique functor 𝑓 ˚,IndCoh𝑟𝑒𝑛 :
IndCoh˚𝑟𝑒𝑛 p𝑆2 q Ñ IndCoh˚𝑟𝑒𝑛 p𝑆1 q fitting into an analogous diagram to the above.
Observe that renoralizable prestacks are closed under fiber squares with one leg locally flat and
the other leg indschematic locally almost of finite type. As in [Gai5] Proposition 3.2.4, there is a
unique functor:

IndCoh˚𝑟𝑒𝑛 : CorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡

equipped with a natural transformation to the functor IndCoh˚ : CorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 Ñ
DGCat𝑐𝑜𝑛𝑡 (obtained by restriction from the functor in S6.24) that on every 𝑆 P PreStk𝑟𝑒𝑛 evaluates
to the canonical functor IndCoh˚𝑟𝑒𝑛 p𝑆q Ñ IndCoh˚ p𝑆q. Here [Link] is shorthand for indschematic
locally almost of finite presentation.48
6.30. Symmetric monoidal structures. Let C Ď PreStk𝑐𝑜𝑛𝑣 denote any one of the subcategories:

ą´8
Sch𝑞𝑐𝑞𝑠 Ď Sch𝑟𝑒𝑎𝑠 Ď IndSch𝑟𝑒𝑎𝑠 Ď PreStk𝑤.𝑟𝑒𝑛
or take C “ PreStk𝑟𝑒𝑛 .
It is direct from the various definitions that C is closed under pairwise Cartesian products in
PreStk𝑐𝑜𝑛𝑣 . In particular, the various correspondence categories we have considered admit symmetric
monoidal structures with monoidal product given objectwise by Cartesian product: see [GR4] SV.3.
We will upgrade the various versions of IndCoh˚ considered so far to have lax symmetric monoidal
structures. This construction is straightforward following [GR4]; we indicate the logic below.
6.31. First, observe that the functor:

pΨ : IndCoh˚ Ñ QCohq : ą´8Sch𝑞𝑐𝑞𝑠 Ñ Homp∆1 , DGCat𝑐𝑜𝑛𝑡 q

admits a unique lax symmetric monoidal structure upgrading the standard symmetric monoidal
structure on QCoh : ą´8Sch𝑞𝑐𝑞𝑠 Ñ DGCat𝑐𝑜𝑛𝑡 . Indeed, this follows by the same method as in [Gai5]
Proposition 3.2.4.
In particular, for 𝑆, 𝑇 P ą´8Sch𝑞𝑐𝑞𝑠 , there is an exterior product functor:

´ b ´ : IndCoh˚ p𝑆q b IndCoh˚ p𝑇 q Ñ IndCoh˚ p𝑆 ˆ 𝑇 q.

This functor is uniquely characterized by the fact that it maps Cohp𝑆q ˆ Cohp𝑇 q Ñ Cohp𝑆 ˆ 𝑇 q
and is compatible with Ψ and exterior product of coherent sheaves.
Remark 6.31.1. We will discuss when the above functor is an equivalence in S6.35.
48We remark that such a morphism is in particular reasonable indschematic.
66 SAM RASKIN

6.32. The above lax symmetric monoidal structure canonically extends to one on the functor
IndCoh˚ : Corrpą´8Sch𝑞𝑐𝑞𝑠 q𝑎𝑙𝑙;𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡 using [GR4] Theorem V.[Link] (and the definition
of the symmetric monoidal structure on correspondences from [GR4] SV.3.2.1).

6.33. Similar logic applies for reasonable indschemes: by Kan extension, IndCoh˚ : IndSch𝑟𝑒𝑎𝑠 Ñ
DGCat𝑐𝑜𝑛𝑡 has a canonical lax symmetric monoidal structure, and this extends canonically to a lax
symmetric monoidal structure on the functor CorrpIndSch𝑟𝑒𝑎𝑠 q𝑎𝑙𝑙;𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡 .

6.34. Next, we apply [GR4] Proposition V.[Link] to obtain a lax monoidal structure on IndCoh˚ :
CorrpPreStk𝑤.𝑟𝑒𝑛 q𝑟𝑒𝑎𝑠.𝑖𝑛𝑑𝑠𝑐ℎ;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡 .
Finally, by a similar argument as for eventually coconnective schemes (i.e., using 𝑡-structures),
the functor IndCoh˚𝑟𝑒𝑛 : CorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡 admits a canonical lax sym-
metric monoidal structure (characterized by compatibility with the one on IndCoh˚ and the natural
transformation IndCoh˚𝑟𝑒𝑛 Ñ IndCoh˚ ).

6.35. Strictness. We now study when our lax symmetric monoidal functors behave as honest
symmetric monoidal functors.
Definition 6.35.1. A weakly renormalizable prestack 𝑆 is strict if for every 𝑇 P ą´8Sch
𝑞𝑐𝑞𝑠 , the
functor:

´ b´ : IndCoh˚ p𝑆q b IndCoh˚ p𝑇 q Ñ IndCoh˚ p𝑆 ˆ 𝑇 q (6.35.1)

is an equivalence.
Before giving examples, we record some basic properties about this notion.
Proposition 6.35.2. (1) Suppose 𝑆 P PreStk𝑤.𝑟𝑒𝑛 is strict. Then for every 𝑇 P IndSch𝑟𝑒𝑎𝑠 , the
natural functor:

´ b ´ : IndCoh˚ p𝑆q b IndCoh˚ p𝑇 q Ñ IndCoh˚ p𝑆 ˆ 𝑇 q

is an equivalence.
(2) Suppose 𝑆 P IndSch𝑟𝑒𝑎𝑠 is strict. Then for every 𝑇 P PreStk𝑤.𝑟𝑒𝑛 , the natural functor:

´ b ´ : IndCoh˚ p𝑆q b IndCoh˚ p𝑇 q Ñ IndCoh˚ p𝑆 ˆ 𝑇 q

is an equivalence.
(3) Suppose 𝑆 P IndSch𝑟𝑒𝑎𝑠 is a filtered colimit under almost finitely presented closed embeddings
𝑆 “ colim𝑖 𝑆𝑖 with 𝑆𝑖 P ą´8Sch𝑞𝑐𝑞𝑠 strict. Then 𝑆 is strict.
(4) Suppose 𝑆 P PreStk𝑤.𝑟𝑒𝑛 admits a flat cover by a strict reasonable indscheme. Then 𝑆 is
strict.
Proof. (1) (resp. (3)) is immediate from the presentation of IndCoh˚ p𝑇 q (resp. IndCoh˚ p𝑆q) as a
colimit (using that 𝑇 , resp. 𝑆, is assumed to be in IndSch𝑟𝑒𝑎𝑠 ). Then (2) is similarly formal, noting
that we can commute the relevant tensor product and limit because IndCoh˚ p𝑆q is compactly
generated (hence dualizable) for 𝑆 P IndSch𝑟𝑒𝑎𝑠 . The same applies in (4): we have to commute a
limit with a tensor product against IndCoh˚ p𝑇 q, which we can do because the relevant 𝑇 here is
assumed to be in ą´8Sch𝑞𝑐𝑞𝑠 so IndCoh˚ p𝑇 q is compactly generated.

 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 67

Warning 6.35.3. In some contexts, people speak use a more general notion of indscheme in which
transition maps are not required to be closed embeddings, and use the term strict indscheme to refer
to the (more standard) notion of indscheme we have used. This terminology has no relationship to
the above notion of strict indscheme(/prestack); we hope our use of this terminology in the present
context does not create confusion.
6.36. We now give some examples of strictness. Note that Proposition 6.35.2 (3) and (4) reduce
us to constructing examples of strict 𝑆 with 𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 .
Lemma 6.36.1. For every 𝑆, 𝑇 P ą´8Sch𝑞𝑐𝑞𝑠 , the functor (6.35.1) is fully-faithful.
Proof. The same argument as in [Gai5] Proposition 4.6.2 applies.

Proposition 6.36.2. If 𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 is almost finite type (over Specp𝑘q49), then 𝑆 is strict.
Proof. Let 𝑇 P ą´8Sch𝑞𝑐𝑞𝑠 be given. By Lemma 6.36.1, we need to show (6.35.1) is essentially
surjective. By Zariski descent, we reduce to the case where 𝑆 is separated.
We have a standard convolution functor:

´ ‹ ´ : IndCohp𝑆 ˆ 𝑆q b IndCoh˚ p𝑆 ˆ 𝑇 q Ñ IndCoh˚ p𝑆 ˆ 𝑇 q

(6.36.1)
K P IndCohp𝑆 ˆ 𝑆q, F P IndCoh˚ p𝑆 ˆ 𝑇 q ÞÑ 𝑝IndCoh
23,˚ pid𝑆 ˆ∆𝑆 ˆ id𝑇 q pK b Fq
!
` ˘

(where e.g., ∆𝑆 : 𝑆 Ñ 𝑆 ˆ 𝑆 is the diagonal and 𝑝23 : 𝑆 ˆ 𝑆 ˆ 𝑇 Ñ 𝑆 ˆ 𝑇 is the projection onto

the last two coordinates).
By Lemma 6.14.2 (1), we have:

∆IndCoh
𝑆,˚ p𝜔𝑆 q ‹ F “ F
for any F P IndCoh˚ p𝑆 ˆ 𝑇 q. In particular, (6.36.1) is essentially surjective.
Now note the composition:

IndCohp𝑆q b IndCohp𝑆q b IndCoh˚ p𝑆 ˆ 𝑇 q Ñ

(6.36.1) (6.36.2)
IndCohp𝑆 ˆ 𝑆q b IndCoh˚ p𝑆 ˆ 𝑇 q ÝÝÝÝÑ IndCoh˚ p𝑆 ˆ 𝑇 q
factors through the subcategory IndCohp𝑆q b IndCoh˚ p𝑇 q. Now the first functor in (6.36.2) is an
equivalence by [Gai5] Proposition 4.6.2, so we are done.

Before proceeding, we need the following auxiliary result.
Lemma 6.36.3. Suppose 𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 is written as a filtered inverse limit 𝑆 “ lim𝑖 𝑆𝑖 under
flat affine structure maps with 𝑆𝑖 P ą´8Sch𝑞𝑐𝑞𝑠 . Then ˚-pullback induces an equivalence:
»
colim IndCoh˚ p𝑆𝑖 q Ý
Ñ IndCoh˚ p𝑆q P DGCat𝑐𝑜𝑛𝑡 .
𝑖,upper-˚

Proof. First, note that the functor:

colim QCohp𝑆𝑖 q Ñ QCohp𝑆q P DGCat𝑐𝑜𝑛𝑡 (6.36.3)

𝑖

49As in [Gai5], it is important here that 𝑘 have characteristic 0 or be a perfect field of characteristic 𝑝.
68 SAM RASKIN

is an equivalence. Indeed, both sides are monadic over QCohp𝑆𝑖0 q for any fixed index 𝑖0 , and the
induced map on monads is an isomorphism (here we are not using the flatness assumption).
Therefore, for any index 𝑖0 and any F P Perfp𝑆𝑖0 q, G P QCohp𝑆𝑖0 q, the natural map:

colim HomQCohp𝑆𝑖 q p𝛼𝑖˚0 ,𝑖 pFq, 𝛼𝑖˚0 ,𝑖 pGqq Ñ HomQCohp𝑆q p𝛼𝑖˚0 pFq, 𝛼𝑖˚0 pGqq P Gpd (6.36.4)
𝑖Ñ𝑖0

is an isomorphism; here 𝛼𝑖0 ,𝑖 : 𝑆𝑖 Ñ 𝑆𝑖0 and 𝛼𝑖0 : 𝑆 Ñ 𝑆𝑖0 denote the structural maps.
We now claim that (6.36.4) is an isomorphism G P QCohp𝑆𝑖0 q` and F P Cohp𝑆𝑖0 q. Indeed, if
»
F, G P QCohp𝑆𝑖0 qě´𝑁 , we choose F1 P Perfp𝑆𝑖0 q equipped with an isomorphism 𝜏 ě´𝑁 F1 ÝÑ F. By
flatness, we have:

𝛼𝑖˚0 ,𝑖 pGq P QCohp𝑆𝑖 qě´𝑁

»
𝜏 ě´𝑁 𝛼𝑖˚0 ,𝑖 pF1 q Ý
Ñ 𝛼𝑖˚0 ,𝑖 pFq
and similarly for 𝛼𝑖0 . Therefore, the sides of (6.36.4) are unchanged under replacing F by F1 , so we
are reduced to that case.
In particular, (6.36.4) is an isomorphism when F, G P Cohp𝑆𝑖0 q. Unwinding the above logic, it
follows that the natural functor:

»
colim IndCoh˚ p𝑆𝑖 q Ý
Ñ IndCoh˚ p𝑆q P DGCat𝑐𝑜𝑛𝑡 (6.36.5)
𝑖

is fully-faithful. To show that this functor is an equivalence, it suffices to show any F P Cohp𝑆q lies
in the essential image.
»
Suppose F lies in cohomological degrees ě ´𝑁 , and choose F1 P Perfp𝑆q with 𝜏 ě´𝑁 pF1 q Ý Ñ F.
Observe that we also have:

»
𝜏 ě´𝑁 F1 Ý
Ñ F P IndCoh˚ p𝑆q
where the truncation is for the 𝑡-structure on IndCoh˚ (and we are using the embeddings Perfp𝑆q Ď
Cohp𝑆q Ď IndCoh˚ p𝑆q). Indeed, both sides are bounded from below, so it suffices to check this after
applying the 𝑡-exact functor Ψ; then the corresponding isomorphism was a defining property of F1 .
By (6.36.3), there exists an index 𝑖0 and some F𝑖10 P Perfp𝑆𝑖0 q equipped with an isomorphism
»
Ñ F1 . We then obtain:
𝛼𝑖˚0 pF𝑖10 q Ý

F “ 𝜏 ě´𝑁 pF1 q “ 𝜏 ě´𝑁 𝛼𝑖˚,IndCoh

0
pF𝑖10 q “ 𝛼𝑖˚,IndCoh
0
𝜏 ě´𝑁 pF𝑖10 q P IndCoh˚ p𝑆q
(where all truncations are for 𝑡-structures on IndCoh˚ ). This shows F is in the essential image of
(6.36.5) as desired.


Next, we show:
Proposition 6.36.4. Suppose 𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 can be written50 as a filtered limit 𝑆 “ lim𝑖 𝑆𝑖 under
flat affine structure maps with each 𝑆𝑖 locally almost of finite type. Then 𝑆 is strict.
Proof. Let 𝑇 P ą´8Sch𝑞𝑐𝑞𝑠 be given. We have a commutative diagram in DGCat𝑐𝑜𝑛𝑡 :

50This is an analogue (better suited for IndCoh) of the notion of placid scheme introduced in [Ras3].
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 69

IndCoh˚ p𝑆q b IndCoh˚ p𝑇 q / IndCoh˚ p𝑆 ˆ 𝑇 q

O O

colim IndCoh˚ p𝑆𝑖 q b IndCoh˚ p𝑇 q / colim IndCoh˚ p𝑆𝑖 ˆ 𝑇 q.

𝑖 𝑖

By Lemma 6.36.3 (applied to 𝑆 and 𝑆 ˆ 𝑇 ), the vertical arrows are isomorphisms. By Proposition
6.36.2, the bottom arrow is an isomorphism. Therefore, the top arrow is an isomorphism, as desired.


7. Weak actions of group indschemes

7.1. In this section, we study weak group actions on categories for Tate group indschemes. This
is a convenient class of group indschemes containing loop groups and their relatives.
For a compact open subgroup 𝐾 Ď 𝐻, we define a weak Hecke category H𝐻,𝐾 𝑤 , which is a certain

monoidal DG category. We then define 𝐻–mod𝑤𝑒𝑎𝑘 in such a way that we have an equivalence:
»
Ñ H𝐻,𝐾
p´q𝐾,𝑤 : 𝐻–mod𝑤𝑒𝑎𝑘 Ý 𝑤
–mod. (7.1.1)
The main homotopical difficulty is to give a definition that is manifestly independent of 𝐾, which
we do by a sort of brute force argument.
The main new feature in the setting of (polarizable) group indschemes, as opposed to group
schemes, is the presence of the modular character ; see S7.17. We especially draw the reader’s
attention to Proposition 7.18.2. The reader who is not worried about homotopical issues may
essentially skip to S7.17, taking (7.1.1) as something like a definition.

7.2. Tate group indschemes. Let 𝐻 be a group indscheme.

Definition 7.2.1. A compact open subgroup of 𝐻 is a classical affine group scheme 𝐾 with a closed
embedding 𝐾 ãÑ 𝐻 that is a homomorphism, and such that 𝐻{𝐾 is an indscheme locally almost
of finite type (equivalently, presentation).
A Tate group indscheme is a reasonable group indscheme admitting a compact open subgroup.
Example 7.2.2. For 𝐺 an affine algebraic group, one can take 𝐻 “ 𝐺p𝐾q and 𝐻0 “ 𝐺p𝑂q. More
generally, a compact open subgroup is a closed group subscheme containing Kerp𝐺p𝑂q Ñ 𝐺p𝑂{𝑡𝑁 qq
for 𝑁 " 0.
Example 7.2.3. For 𝐺 an affine algebraic group and 𝐻0 Ď 𝐺p𝐾q a compact open subgroup, one can
take 𝐻 “ 𝐺p𝐾q^ 𝐻0 to be the formal completion of 𝐺p𝐾q along 𝐻0 .

Remark 7.2.4. If 𝐻 is an ind-affine Tate group indscheme, then 𝐻 satisfies the hypotheses of S5.6,
and the definition of IndCoh˚ p𝐻q from loc. cit. clearly coincides with the construction from S6 in
this case. We remark that the notion of naive action of 𝐻 from loc. cit. extends to the possibly
non-ind-affine setting considered here: this means an IndCoh˚ p𝐻q-module category.
Throughout this section, 𝐻 denotes a Tate group indscheme. Our main objective in this section
is to develop a theory of genuine 𝐻-actions on categories.
Remark 7.2.5. We remark that 𝐻 being a Tate group indscheme implies in particular that 𝐻 is
reasonable (as an indscheme). Moreover, by Proposition 6.36.4 and Proposition 6.35.2 (3), 𝐻 is
strict (in the sense of S6.35).
70 SAM RASKIN

7.3. Topological conventions. The reader may safely ignore this discussion at first pass.
Throughout this section, we impose a convention (implicit above) that all quotients are under-
stood as prestack quotients (i.e., geometric realizations of the bar construction) sheafified for the
Zariski topology. In particular, Bp´q indicates the Zariski sheafified classifying space.
Therefore, in the definition of compact open subgroup above, the condition that 𝐻{𝐾 be an
indscheme is a funny condition which is satisfied if e.g. the étale sheafification of this quotient is
an indscheme and the projection from 𝐻 to this quotient is a 𝐾-torsor locally trivial in the Zariski
topology. This is the case when 𝐾 is prounipotent or when 𝐻 “ 𝐺p𝐾q and 𝐾 “ 𝐺p𝑂q (as is
well-known, see e.g. [BD1] Theorem 4.5.1).
This convention can be relaxed to the étale topology at the cost of replacing the fundamental
role of quasi-compact quasi-separated schemes in S6 by quasi-compact quasi-separated algebraic
spaces. That be done without serious modification (using [Lur4] Proposition [Link] as a staring
point), but we content ourselves with the above restrictions since they suffice for our applications.
(We remark that changing Zariski to étale would mean that for any 𝐾 Ď 𝐻 with a prounipotent
tail, whenever 𝐾 “should” be regarded as a compact open subgroup, it is.)

7.4. The following terminology will be convenient in what follows.

Definition 7.4.1. A DG 2-category is a category enriched over DGCat𝑐𝑜𝑛𝑡 . We let 2–DGCat denote
the category of DG 2-categories (for our purposes, it is sufficient to view 2–DGCat as a 1-category).
7.5. Hecke categories. Let 𝐻 be a group indscheme and let 𝐾 be a group subscheme.
First, recall51 that the (Zariski sheafified) quotient 𝐾z𝐻{𝐾 “ B𝐾 ˆB𝐻 B𝐾 is an algebra in
CorrpPreStk𝑐𝑜𝑛𝑣 q𝑎𝑙𝑙;𝑎𝑙𝑙 with unit and multiplication maps defined by the correspondences:

𝐾
B𝐾 𝐾z𝐻 ˆ 𝐻{𝐾

{ # v &
Specp𝑘q 𝐾z𝐻{𝐾 𝐾z𝐻{𝐾 ˆ 𝐾z𝐻{𝐾 𝐾z𝐻{𝐾.
Now observe that by our assumptions on 𝐻 and 𝐾, each of the left arrows in the above diagrams
are locally flat while each of the right arrows are locally almost of finite presentation. Therefore,
𝐾z𝐻{𝐾 is canonically an algebra in CorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 .
Definition 7.5.1. The (genuine) weak Hecke category H𝐻,𝐾
𝑤 P AlgpDGCat𝑐𝑜𝑛𝑡 q is IndCohp𝐻{𝐾q𝐾,𝑤 “
IndCoh˚𝑟𝑒𝑛 p𝐾z𝐻{𝐾q, with monoidal structure induced from the above algebra (under correspon-
dences) structure on 𝐾z𝐻{𝐾 and by applying the symmetric monoidal functor:

IndCoh˚𝑟𝑒𝑛 : CorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 Ñ DGCat𝑐𝑜𝑛𝑡

constructed in S6.
Remark 7.5.2. We do not need the whole theory of S6 to construct the above monoidal structure
on H𝐻,𝐾
𝑤 . Indeed, this follows from Example 8.16.3 below.

When we introduce 𝐻–mod𝑤𝑒𝑎𝑘 below, we will have 𝐻–mod𝑤𝑒𝑎𝑘 » H𝐻,𝐾 𝑤 –mod, so this elementary

construction gives a quick construction of 𝐻–mod𝑤𝑒𝑎𝑘 .

51If 𝐻 is classical, which is our main case of interest (see Remark 7.6.2), what follows is complete. If 𝐻 is derived,
[GR4] SV.3.4 covers the homotopy coherence issues.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 71

However, it is not clearly independent of the choice of compact open subgroup 𝐾, and this leads
to difficulties with the functoriality in 𝐻. For our purposes, the most important use of the theory
of S6 is to deal with these functoriality problems.
7.6. Genuine actions. Let TateGp denote the category of Tate group indschemes.
Our first goal will be to show the following result. In what follows, we regard TateGp as a
symmetric monoidal category via products, and similarly for 2–DGCat.
Proposition-Construction 7.6.1. There is a canonical lax symmetric monoidal functor:

TateGp𝑜𝑝 Ñ 2–DGCat
𝐻 ÞÑ 𝐻–mod𝑤𝑒𝑎𝑘
with the following properties.
(1) For any 𝐻 P TateGp𝑜𝑝 and any 𝐾 Ď 𝐻 compact open, there is a canonical equivalence:

p´q𝐾,𝑤
𝐻–mod𝑤𝑒𝑎𝑘 ÝÝÝÝÑ H𝐻,𝐾
𝑤
–mod :“ H𝐻,𝐾
𝑤
–modpDGCat𝑐𝑜𝑛𝑡 q.
(2) For any morphism 𝑓 : 𝐻1 Ñ 𝐻2 P TateGp and any 𝐾𝑖 Ď 𝐻𝑖 compact open subgroups with
𝑓 p𝐾1 q Ď 𝐾2 , the functor:

𝐻2 –mod𝑤𝑒𝑎𝑘 Ñ 𝐻1 –mod𝑤𝑒𝑎𝑘
canonically fits into a commutative diagram:

𝐻2 –mod𝑤𝑒𝑎𝑘 / 𝐻1 –mod𝑤𝑒𝑎𝑘

» »
 
H𝐻
𝑤 –mod / H𝑤
2 ,𝐾2 𝐻1 ,𝐾1 –mod

where the bottom arrow is constructed using the Hecke-bimodule IndCoh˚𝑟𝑒𝑛 p𝐾1 z𝐻2 {𝐾2 q.
(3) For 𝐼 a finite set and t𝐻𝑖 P TateGpu𝑖P𝐼 equipped with compact open subgroups 𝐾𝑖 Ď 𝐻𝑖 , the
functor:
ź ź
p𝐻𝑖 –mod𝑤𝑒𝑎𝑘 q Ñ p 𝐻𝑖 –mod𝑤𝑒𝑎𝑘 q
𝑖P𝐼 𝑖P𝐼
coming from the lax symmetric monoidal structure corresponds under the equivalences from
(1) to the functor:
ź
pH𝐻𝑤
𝑖 ,𝐾𝑖
–modq Ñ Hś
𝑤
𝐻𝑖 ,
ś
𝐾𝑖 –mod » pb H𝐻𝑖 ,𝐾𝑖 q–mod
𝑤
𝑖P𝐼 𝑖P𝐼 𝑖
𝑖P𝐼
tC𝑖 P H𝐻
𝑤
𝑖 ,𝐾𝑖
–modu𝑖P𝐼 ÞÑ b C𝑖 .
𝑖

Remark 7.6.2. Let TateGp𝑐𝑙 Ď TateGp denote the subcategory of Tate group indschemes that are
classical as prestacks.52 Note that TateGp𝑐𝑙 is actually a p1, 1q-category.
To simplify the exposition, we actually only give the restriction of the functor from Proposition-
Construction 7.6.1 to TateGp𝑐𝑙 ; this suffices for our applications. The requisite homotopy coherence
needed to provide all of Proposition-Construction 7.6.1 can be given using [GR4] SV.3.4. But the
52As in [GR3] Theorem 9.3.4, any formally smooth Tate group indscheme that is weakly ℵ in the sense of loc.
0
cit. is automatically classical. In particular, this applies for a loop group, or for its formal completion at any compact
open subgroup scheme.
72 SAM RASKIN

argument (at least in the form the author has in mind) is more tedious than merits inclusion
(without real applications) here.
We will give the construction after some preliminary remarks.
7.7. Morita theory. We will need a review of Morita categories in a higher categorical context.
7.8. First, let C be a symmetric monoidal category with colimits and whose monoidal product
commutes with colimits in each variable.
Let C–mod denote the 2-category of C-module categories M with colimits and for which for every
F P C and G P M, the action functors F ‹ ´ : M Ñ M and ´ ‹ G : C Ñ M admit right adjoints53 (so
in particular, the action functor commutes with colimits in each variable).
In this case, let MoritapCq be the 2-category defined as the full subcategory of C–mod consisting of
objects of the form 𝐴–mod for 𝐴 P AlgpCq. As in [Lur3] Remark [Link], this recovers the standard
Morita 2-category if C is a p1, 1q-category.
In particular, we remind from loc. cit. that if we have M𝑖 P MoritapCq for 𝑖 “ 1, 2 and we choose
𝐴𝑖 P AlgpCq such that M𝑖 “ 𝐴𝑖 –mod, then the category of morphisms 𝐹 : M1 Ñ M2 is canonically
equivalent to the data of an p𝐴2 , 𝐴1 q-bimodule.
7.9. In what follows, we will use the (fully-faithful) Segal functor Seq‚ : 2–Cat Ñ HompΔ𝑜𝑝 , Catq;
we refer to [GR4] Appendix A.1 for details.
7.10. For C a symmetric monoidal category that may not have colimits, there is no hope for defining
its Morita category.54 However, we can still define the associated simplicial category Seq‚ pMoritapCqq
as follows.
First, if C is essentially small, embed C into its Yoneda category YopCq :“ HompC𝑜𝑝 , Gpdq. Note
that YopCq is equipped with a symmetric monoidal structure commuting with colimits in each
variable. In particular, MoritapYopCqq and Seq‚ pMoritapCqq are defined.
Now for r𝑛s P Δ𝑜𝑝 , define Seq𝑛 pCq to be the full subcategory of Seq𝑛 pYopCqq whose objects are
sequences M0 Ñ . . . Ñ M𝑛 where each M𝑖 is of the form 𝐴𝑖 –mod for 𝐴𝑖 P AlgpCq Ď AlgpYopCqq and
each morphism M𝑖 Ñ M𝑖`1 admits a right adjoint in the 2-category YopCq–mod.55 It is straightfor-
ward to see that this latter condition is equivalent to M𝑖 Ñ M𝑖`1 corresponding to an p𝐴𝑖`1 , 𝐴𝑖 q-
bimodule that lies in C Ď YopCq.
7.11. In general, we define Seq‚ pMoritapCqq : Δ𝑜𝑝 Ñ Cat e.g. by extending the universe so that C
is essentially small. It is straightforward to see that we do actually obtain a simplicial category56
in this way.
Moreover, if C does admit colimits and its monoidal product preserves such colimits, then the
simplicial category just defined canonically coincides with the same denoted simplicial category
defined by applying the Segal construction to the category MoritapCq from S7.8. Therefore, we are
justified in not distinguishing the two notationally.
Notation 7.11.1. In what follows, for C as above and 𝐴 P AlgpCq, we let r𝐴s denote the induced
object in Seq0 pMoritapCqq.
53In the language of [Lur3] S4.2.1, M is cotensored and enriched over C.
54Indeed, composition in a Morita category involves tensor products of bimodules, and this means certain geometric
realizations need to exist.
55By the definition in S7.8, this means that the underlying functor admits a right adjoint that commutes with all
colimits and is YopCq-linear.
56By definition, a category has essentially small Homs. In principle, universe extension risks breaking this property,
and the content of the claim is that this does not happen here.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 73

7.12. Finally, we conclude by remarking that Moritap´q is by construction functorial for (left) lax
symmetric monoidal functors.
7.13. Construction of the functor. We now return to the setting of S7.6.
Proof of Proposition-Construction 7.6.1. To simplify the exposition, we ignore the symmetric monoidal
structures until the last step.
Step 1. First, define the category57 TateGp𝑓 𝑟 Ď Homp∆1 , TateGpq as the subcategory consisting of
maps 𝐾 Ñ 𝐻 that are the embedding of a compact open subgroup in a Tate group indscheme.
p𝐾Ď𝐻qÞÑ𝐻
Note that the forgetful functor Oblv𝑓 𝑟 : TateGp𝑓 𝑟 ÝÝÝÝÝÝÝÑ TateGp is 1-fully-faithful (i.e., the
induced maps on Homs are fully-faithful morphisms of groupoids).
We claim that Oblv𝑓 𝑟 is a Verdier localization functor. By definition, this means that for every
C P Cat, the functor:

HompTateGp, Cq Ñ HompTateGp𝑓 𝑟 , Cq
of restriction along Oblv𝑓 𝑟 is fully-faithful functor with essential image consisting of functors sending
Oblv𝑓 𝑟 -local58 morphisms to isomorphisms.
Indeed, note that the left adjoint to Oblv𝑓 𝑟 is pro-representable: it maps 𝐻 P TateGp to the pro-
object “ lim ”𝐾Ď𝐻compact open p𝐾 Ñ 𝐻q,59 where 𝐾 Ñ 𝐻 is regarded as an object in TateGp𝑓 𝑟 and
the quotation marks emphasize that this (filtered) limit takes place in the relevant pro-category.
This pro-valued left adjoint is clearly fully-faithful, and this is well-known to imply the Verdier
localization property.
Step 2. Next, we construct a (right) lax morphism:

Seq‚ pTateGp𝑜𝑝
𝑓 𝑟 q Ñ Seq‚ pMoritapCorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 qq
of simplicial categories sending 𝐾 Ď 𝐻 P TateGp𝑓 𝑟 (regarded as a 0-simplex of the left hand side)
to r𝐾z𝐻{𝐾s (regarded as a 0-simplex of the right hand side as in Notation 7.11.1). Here by lax, we
mean that our morphism is merely a lax natural transformation, i.e., a functor over Δ𝑜𝑝 between
the corresponding coCartesian Grothendieck fibrations.
This step is where we use Remark 7.6.2 to ignore homotopy coherence issues. That is, we treat
TateGp as a p1, 1q-category (e.g,. by actually restricting to TateGp𝑐𝑙 ).
Our lax functor assigns to every r𝑛s P Δ and r𝑛s-shaped diagram p𝐾0 Ď 𝐻0 q Ñ . . . Ñ p𝐾𝑛 Ď 𝐻𝑛 q
in TateGp𝑓 𝑟 the r𝑛s-simplex of Seq‚ pMoritapCorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 qq defined by:

𝐾0 z𝐻0 {𝐾0 ñ 𝐾0 z𝐻1 {𝐾1 ð 𝐾1 z𝐻1 {𝐾1 ñ . . . 𝐾𝑛´1 z𝐻𝑛 {𝐾𝑛 ð 𝐾𝑛 z𝐻𝑛 {𝐾𝑛 .
The notation indicates that we consider 𝐾𝑖 z𝐻𝑖 {𝐾𝑖 as an algebra in CorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡
and 𝐾𝑖 z𝐻𝑖`1 {𝐾𝑖 as a bimodule in this same correspondence category for 𝐾𝑖 z𝐻𝑖 {𝐾𝑖 and 𝐾𝑖`1 z𝐻𝑖`1 {𝐾𝑖`1 .
We now need to specify where morphisms (in the relevant Grothendieck construction) are sent.
To simplify the notation, we spell out the construction only for morphisms lying over the active
0ÞÑ0,1ÞÑ2
morphism 𝛼 : r1s ÝÝÝÝÝÝÑ r2s P Δ. (The reader will readily see that this simplification really is
only cosmetic.)
57The subscript is an abbreviation of framed.
58This phrase refers to morphisms in TateGp that map to isomorphisms under Oblv𝑓 𝑟 .
𝑓𝑟
59The key point in verifying this formula is that for any 𝑓 : 𝐻 Ñ 𝐻 P TateGp and 𝐾 Ď 𝐻 a compact open
1 2 2 2
subgroup, the category of compact open subgroups 𝐾1 Ď 𝐻1 mapping into 𝐾2 is non-empty and filtered.
74 SAM RASKIN

Then a 2-simplex of the left hand side above corresponds to a datum p𝐾0 Ď 𝐻0 q Ñ p𝐾1 Ď
𝐻1 q Ñ p𝐾2 Ď 𝐻2 q P TateGp, and a map to a 1-simplex p𝐾 r0 Ď 𝐻r 0 q is equivalent to maps and
𝑓𝑖 : 𝐻𝛼p𝑖q Ñ 𝐻
r 𝑖 (𝑖 “ 0, 1) sending 𝐾𝛼p𝑖q to 𝐾
r𝑖.
The relevant map in the right hand side is induced by the augmented simplicial diagram:

𝐾0 z𝐻1 {𝐾1 ˆ 𝐾1 z𝐻1 {𝐾1 ˆ 𝐾1 z𝐻1 {𝐾2 Ñ 𝐾0 z𝐻1 {𝐾1 ˆ 𝐾1 z𝐻2 {𝐾2 Ñ 𝐾
r 0 z𝐻
r1𝐾
r1
P CorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 .
Here all arrows are morphisms in the correspondence category (i.e., they represent correspondences),
the underlying simplicial diagram is the bar construction for the relative tensor product of these
left and right 𝐾1 z𝐻1 {𝐾1 -modules, and the augmentation is given by the correspondence:

𝐾1
𝐾0 z𝐻1 ˆ 𝐻2 {𝐾2

u '
𝐾0 z𝐻1 {𝐾1 ˆ 𝐾1 z𝐻2 {𝐾2 𝐾
r 0 z𝐻
r 1 {𝐾
r1.
Here the left arrow is obvious, and the right arrow is the composition:
1 𝐾 2 𝐾
mult. 𝑓1 r
𝐾0 z𝐻1 ˆ 𝐻2 {𝐾2 Ñ 𝐾0 z𝐻2 ˆ 𝐻2 {𝐾2 ÝÝÝÑ 𝐾0 z𝐻2 {𝐾2 ÝÑ 𝐾 0 z𝐻1 {𝐾1 .
r r
Unwinding the constructions, this was exactly the sort of datum we needed to specify.
Step 3. Applying the lax symmetric monoidal functor IndCoh˚𝑟𝑒𝑛 : CorrpPreStk𝑟𝑒𝑛 q𝑖𝑛𝑑𝑠𝑐ℎ.𝑙𝑎𝑓 𝑝;𝑙𝑜𝑐.𝑓 𝑙𝑎𝑡 Ñ
DGCat𝑐𝑜𝑛𝑡 and the functoriality from S7.12, we obtain a lax functor:

Seq‚ pTateGp𝑜𝑝
𝑓 𝑟 q Ñ Seq‚ pMoritapDGCat𝑐𝑜𝑛𝑡 qq.
As these are each Segal categories for actual 2-categories, this is the same60 as a lax functor
TateGp𝑜𝑝
𝑓 𝑟 Ñ MoritapDGCat𝑐𝑜𝑛𝑡 q. The latter 2-category is by construction contained in DGCat𝑐𝑜𝑛𝑡 –mod,
which is itself contained in 2–DGCat.
Therefore, we obtain a lax functor of 2-categories:

TateGp𝑜𝑝𝑓 𝑟 Ñ 2–DGCat.
We claim that this lax functor is an actual functor.
Suppose we are given p𝐾0 Ď 𝐻0 q Ñ p𝐾1 Ď 𝐻1 q Ñ p𝐾2 Ď 𝐻2 q P TateGp. We obtain a diagram
that commutes up to a natural transformation:

H𝐻
𝑤 –mod KS / H𝑤 –mod
0 ,𝐾0
7 𝐻2 ,𝐾2

'
H𝐻
𝑤
1 ,𝐾1
–mod
corresponding to the map of Hecke bimodules:

IndCoh˚𝑟𝑒𝑛 p𝐾0 z𝐻1 {𝐾1 q b IndCoh˚𝑟𝑒𝑛 p𝐾1 z𝐻2 {𝐾2 q Ñ IndCoh˚𝑟𝑒𝑛 p𝐾0 z𝐻2 {𝐾2 q.
H𝐻
𝑤
1 ,𝐾1
that we need to show is an isomorphism.
60By definition of lax functor; c.f. [GR4] Appendix A.1.3.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 75

To verify this, note that we have a canonical monoidal functor Repp𝐾1 q Ñ H𝐻

𝑤
1 ,𝐾1
. By Lemma
5.20.1 (4), we have an equivalence:
»
Repp𝐾0 q b H𝐻
𝑤
1 ,𝐾1
Ñ IndCoh˚𝑟𝑒𝑛 p𝐾0 z𝐻1 {𝐾1 q
Ý
Repp𝐾1 q
of H𝐻
𝑤
1 ,𝐾1
-module categories.
Therefore, we can calculate:

IndCoh˚𝑟𝑒𝑛 p𝐾0 z𝐻1 {𝐾1 q b IndCoh˚𝑟𝑒𝑛 p𝐾1 z𝐻2 {𝐾2 q “

H𝐻
𝑤
1 ,𝐾1

Repp𝐾0 q b H𝐻
𝑤
1 ,𝐾1
b IndCoh˚𝑟𝑒𝑛 p𝐾1 z𝐻2 {𝐾2 q “
Repp𝐾1 q H𝐻
𝑤
1 ,𝐾1
»
Repp𝐾0 q b IndCoh˚𝑟𝑒𝑛 p𝐾1 z𝐻2 {𝐾2 q Ý
Ñ IndCoh˚𝑟𝑒𝑛 p𝐾0 z𝐻2 {𝐾2 q
Repp𝐾1 q
as desired.
Step 4. Next, suppose that we are given an Oblv𝑓 𝑟 -local morphism, or equivalently, 𝐻 P TateGp
with an embedding of compact open subgroups 𝐾1 Ď 𝐾2 Ď 𝐻.
Then the functor:

TateGp𝑜𝑝
𝑓 𝑟 Ñ 2–DGCat
sends this datum to the functor:

H𝐻,𝐾
𝑤
2
–mod Ñ H𝐻,𝐾
𝑤
1
–mod
˚
defined by the bimodule IndCoh𝑟𝑒𝑛 p𝐾1 z𝐻{𝐾2 q. By Step 1, to obtain the functor from Proposition-
Construction 7.6.1, it suffices to show the above is an equivalence. As 𝐾2 has a cofinal sequence of
normal compact open subgroups, it suffices to treat the case where 𝐾1 is normal in 𝐾2 .
Note that the composition:

H𝐻,𝐾
𝑤
2
–mod Ñ H𝐻,𝐾
𝑤
1
–mod Ñ DGCat𝑐𝑜𝑛𝑡 (7.13.1)
sends D P H𝐻,𝐾
𝑤
2
–mod to:

Repp𝐾1 q b D
Repp𝐾2 q
by the isomorphism:
»
H𝐻,𝐾
𝑤
2
b Ñ IndCoh˚𝑟𝑒𝑛 p𝐾2 z𝐻{𝐾1 q
Repp𝐾1 q Ý
Repp𝐾2 q
of H𝐻,𝐾
𝑤
2
-module categories (obtained as in the previous step from Lemma 5.20.1 (4)). By normality
of 𝐾1 in 𝐾2 , we may further identify:

Repp𝐾1 q b D “ Vect b D.
Repp𝐾2 q Repp𝐾2 {𝐾1 q
By Theorem 5.10.1, this implies that (7.13.1) is conservative and commutes with limits, so is
monadic.
Therefore, it suffices to show that the functor:

H𝐻,𝐾1 Ñ EndH𝐻,𝐾
𝑤
˚
–mod pIndCoh𝑟𝑒𝑛 p𝐾2 z𝐻{𝐾1 qq (7.13.2)
2
76 SAM RASKIN

is an equivalence (as the right hand side is the monad defined by (7.13.1)). This follows by similar
logic — the right hand side is:
𝑤
HomH𝐻,𝐾
𝑤 –mod pH𝐻,𝐾2 b Repp𝐾1 q, IndCoh˚𝑟𝑒𝑛 p𝐾2 z𝐻{𝐾1 qq “
2 Repp𝐾2 q
HomRepp𝐾2 q–mod pRepp𝐾1 q, IndCoh˚𝑟𝑒𝑛 p𝐾2 z𝐻{𝐾1 qq “
HomRepp𝐾2 {𝐾1 q–mod pVect, IndCoh˚𝑟𝑒𝑛 p𝐾2 z𝐻{𝐾1 qq “
𝑇 ℎ𝑚.5.10.1
HomRepp𝐾2 {𝐾1 q–mod pVect, IndCoh˚𝑟𝑒𝑛 p𝐾1 z𝐻{𝐾1 q𝐾2 {𝐾1 ,𝑤 q “
HomQCohp𝐾2 {𝐾1 q–mod pQCohp𝐾2 {𝐾1 q, IndCoh˚𝑟𝑒𝑛 p𝐾1 z𝐻{𝐾1 qq “ H𝐻,𝐾
𝑤
1

as desired (it is immediate to check that this identification is compatible with the functor (7.13.2)).
Step 5. Finally, we briefly remark that all of the above immediately upgrades to the (lax) symmetric
monoidal setting.
In detail, note that TateGp𝑓 𝑟 is symmetric monoidal under products. and the functor TateGp𝑓 𝑟 Ñ
TateGp is a symmetric monoidal Verdier localization.
The functor from Step 2 upgrades to a functor of simplicial symmetric monoidal categories, noting
that by the construction of S7.7-7.12, MoritapCq is canonically a simplicial symmetric monoidal
category.
This implies that the functor TateGp𝑜𝑝 𝑓 𝑟 Ñ DGCat𝑐𝑜𝑛𝑡 –mod is naturally a (left) lax symmetric
˚ 61
monoidal functor, since IndCoh is. Finally, the forgetful functor DGCat𝑐𝑜𝑛𝑡 –mod Ñ 2–DGCat is
by construction lax symmetric monoidal, giving the result.

7.14. Forgetful functors. As in S5.15, for 𝐻 a Tate group indscheme, there is a canonical non-
conservative functor:

Oblv𝑔𝑒𝑛 : 𝐻–mod𝑤𝑒𝑎𝑘 Ñ DGCat𝑐𝑜𝑛𝑡

C ÞÑ colim C𝐾,𝑤 .
𝐾Ď𝐻 compact open
We denote the colimit appearing on the right hand side also by C in a similar abuse of notation as
in the profinite dimensional setting. We remark that this forgetful functor Oblv𝑔𝑒𝑛 upgrades to a
functor to 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 ,which we also denote by Oblv𝑔𝑒𝑛 .
As in the profinite dimensional setting, where there is not confusion we often omit Oblv𝑔𝑒𝑛 from
the notation, i.e., we often speak of genuine 𝐻-actions on C P DGCat𝑐𝑜𝑛𝑡 by which we mean that
we are given an object of 𝐻–mod𝑤𝑒𝑎𝑘 that maps to C under Oblv𝑔𝑒𝑛 .
Lemma 7.14.1. The above forgetful functor commutes with limits and colimits.
CÞÑC𝐾,𝑤
Proof. First note that for every 𝐾 Ď 𝐻 compact open, the functor 𝐻–mod𝑤𝑒𝑎𝑘 ÝÝÝÝÝÑ DGCat𝑐𝑜𝑛𝑡
commutes with limits and colimits. Indeed, it may be calculated as the composition:
» Oblv
Ñ H𝐻,𝐾
𝐻–mod𝑤𝑒𝑎𝑘 Ý 𝑤
ÝÝÝÑ DGCat𝑐𝑜𝑛𝑡
and the latter functor commutes with limits and colimits (as this is always the case for a category
of modules over an algebra).
61To make the implicit explicit: we are using the tensor product on DGCat
𝑐𝑜𝑛𝑡 –mod obtained by viewing DGCat𝑐𝑜𝑛𝑡
as a commutative algebra in the symmetric monoidal category denoted Cat8 pKqb in [Lur3] Proposition [Link] (so
DGCat𝑐𝑜𝑛𝑡 –mod are the modules in this category).
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 77

It immediately follows that our forgetful functor commutes with colimits. Commutation with
limits follows by noting that each structural functor in the colimit admits a continuous right adjoint
given by ˚-averaging, so we may also calculate it as the functor:

C ÞÑ lim C𝐾,𝑤
𝐾Ď𝐻 compact open
(where the structural functors in the limit are these right adjoints).

7.15. Invariants and coinvariants. Let 𝐻 be a Tate group indscheme.
Define triv : DGCat𝑐𝑜𝑛𝑡 Ñ 𝐻–mod𝑤𝑒𝑎𝑘 as the restriction functor along the homomorphism 𝐻 Ñ
Specp𝑘q (regarding the target as the trivial group).
Remark 7.15.1. By Proposition-Construction 7.6.1 (2), for 𝐾 Ď 𝐻 compact open we have:

trivpVectq𝐾,𝑤 “ Repp𝐾q.
We define the functor of (genuine, weak) invariants:

𝐻–mod𝑤𝑒𝑎𝑘 Ñ DGCat𝑐𝑜𝑛𝑡
C ÞÑ C𝐻,𝑤
to be the right adjoint to triv, and we define (genuine, weak) the coinvariants functor C ÞÑ C𝐻,𝑤 to
be the left adjoint. These may be computed explicitly after a choice of compact open subgroup 𝐾
as:

C𝐻,𝑤 » HomH𝐻,𝐾
𝑤 –mod pRepp𝐾q, C
𝐾,𝑤
q
(7.15.1)
C𝐻,𝑤 » Repp𝐾q b C𝐾,𝑤 .
H𝐻,𝐾
𝑤

Remark 7.15.2. The comparison between invariants and coinvariants is more subtle in the group
indscheme setting. than in the group scheme setting.62
7.16. Rigid monoidal categories. Before proceeding, we review some constructions with rigid
monoidal DG categories, following [Gai4] S6. We refer to loc. cit. for the relevant notion of rigid
monoidal DG category; we remind that this is a property for some A P AlgpDGCat𝑐𝑜𝑛𝑡 q to satisfy.63
We will construct a canonical morphism 𝜙A : A Ñ A of monoidal categories that plays a key
role.
Let A_ be the dual of A as an object of DGCat𝑐𝑜𝑛𝑡 . Note that A_ is canonically A-bimodule in
DGCat𝑐𝑜𝑛𝑡 (as it is the dual of the A-bimodule A). Therefore, we obtain a monoidal functor:

A Ñ Endmod–A pA_ q P AlgpDGCat𝑐𝑜𝑛𝑡 q. (7.16.1)

(The notation indicates endomorphisms as a right A-module in DGCat𝑐𝑜𝑛𝑡 .) On the other hand, by
definition of rigidity, the functor:

A Ñ A_
F ÞÑ pG ÞÑ HomA p1A , F ‹ Gq
62And these functors behave less well. For example, they may fail to be conservative (c.f. [Gai8] Theorem 2.5.4).
63If A is compactly generated and rigid, then the subcategory A𝑐 Ď A is closed under the monoidal operation and
rigid according to the more standard notion of rigid monoidal (in terms of existence of duals).
78 SAM RASKIN

is an equivalence of right A-module categories (here 1A is the unit object). Therefore, the right
hand side of (7.16.1) identifies canonically with A, and we obtain the desired morphism 𝜙A .
Remark 7.16.1. By construction, there are natural isomorphisms:

»
HomA p1A , F ‹ Gq Ý
Ñ HomA p1A , G ‹ 𝜙A pFqq
for F, G P A.
For M P A–mod :“ A–modpDGCat𝑐𝑜𝑛𝑡 q, we let 𝜙A,˚ pMq P A–mod denote the restriction of M
along 𝜙A , and we let 𝜙˚A denote the inverse to the equivalence 𝜙A,˚ .
By [Gai4] Corollary 6.3.3, for M, N P A–mod with M dualizable in DGCat𝑐𝑜𝑛𝑡 , there is a canonical
equivalence:

»
HomA–mod pM, Nq Ý
Ñ M_ b 𝜙A,˚ pNq (7.16.2)
A
functorial in M and N.

7.17. Polarizations and the modular character. We now introduce the following terminology.
Definition 7.17.1. A polarization of 𝐻 P TateGp is a compact open subgroup 𝐾 Ď 𝐻 such that
𝐻{𝐾 is ind-proper. If a polarization exists, we say that 𝐻 is polarizable.
Example 7.17.2. The loop group of a reductive group is polarizable.
Example 7.17.3. If 𝐻 is formal in ind-directions, i.e., its reduced locus 𝐻 𝑟𝑒𝑑 Ď 𝐻 is a compact open
subgroup, then 𝐻 is polarizable (and equipped with the canonical polarization 𝐻 𝑟𝑒𝑑 ). In this case,
we say 𝐻 is a group indscheme 𝐻 of Harish-Chandra type.
We have the following result, which is evident from the definitions (and preservation of coherent
objects under flat pullbacks and proper pushforwards):
Lemma 7.17.4. For 𝐾 a polarization of 𝐻, the genuine Hecke category H𝐻,𝐾
𝑤 is rigid monoidal
(in the sense of [Gai4] S6).
Corollary 7.17.5. If 𝐻 is polarizable, the coinvariants functor 𝐻–mod𝑤𝑒𝑎𝑘 Ñ DGCat𝑐𝑜𝑛𝑡 is corep-
resentable.
Proof. Let 𝐾 be a polarization of 𝐻. Then we have the equivalence 𝐻–mod𝑤𝑒𝑎𝑘 » H𝐻,𝐾 𝑤 –mod, and

the right hand side is rigid monoidal. Using the notation of S7.16, we obtain functorial identifica-
tions:

(7.15.1) (7.16.2)
C𝐻,𝑤 “ Repp𝐾q b C𝐾,𝑤 “ HomH𝐻,𝐾
𝑤
˚
–mod pRepp𝐾q, 𝜙H𝑤 pC
𝐾,𝑤
qq “
H𝐻,𝐾
𝑤 𝐻,𝐾

,˚ pRepp𝐾qq, C
𝐾,𝑤
HomH𝐻,𝐾
𝑤 –mod p𝜙H𝐻,𝐾
𝑤 q
where we have repeatedly used that 𝜙H𝐻,𝐾 ˚
𝑤 ,˚ and 𝜙H𝑤 are mutually inverse equivalences. Applying
𝐻,𝐾
the equivalence H𝐻,𝐾 –mod » 𝐻–mod𝑤𝑒𝑎𝑘 now gives the result.
𝑤


Definition 7.17.6. For 𝐻 a polarizable Tate group indscheme, the modular character 𝜒𝑇 𝑎𝑡𝑒,𝐻 “
𝜒𝑇 𝑎𝑡𝑒 P 𝐻–mod𝑤𝑒𝑎𝑘 is the object corepresenting the functor of coinvariants.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 79

7.18. Next, observe that because the functor of Proposition-Construction 7.6.1 is lax symmetric
monoidal, 𝐻–mod𝑤𝑒𝑎𝑘 is naturally symmetric monoidal with unit trivpVectq. We denote the tensor
product by b; explicitly, for C1 , C2 P 𝐻–mod𝑤𝑒𝑎𝑘 , we have an object C1 b C2 P p𝐻 ˆ 𝐻q–mod𝑤𝑒𝑎𝑘
from the lax symmetric monoidal functoriality, and then we restrict along the diagonal map.
Lemma 7.18.1. If 𝐻 is a polarizable Tate group indscheme, the functor:

𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻–mod𝑤𝑒𝑎𝑘
given by tensoring with the modular character 𝜒𝑇 𝑎𝑡𝑒 is an equivalence.
Proof. Let 𝐾 Ď 𝐻 be a polarization. By the proof of Corollary 7.17.5 (and Proposition-Construction
7.6.1 (3)), we obtain a commutative diagram:

´b𝜒𝑇 𝑎𝑡𝑒
𝐻–mod𝑤𝑒𝑎𝑘 / 𝐻–mod𝑤𝑒𝑎𝑘

p´q𝐾,𝑤 p´q𝐾,𝑤 (7.18.1)

 𝜙H𝐻,𝐾 ,˚ 
H𝐻,𝐾
𝑤 –mod / H𝐻,𝐾 –mod.

Each of the vertical arrows and the bottom arrow are equivalences, so the top arrow is as well.


Let 𝐻 be a polarizable Tate group indscheme. By the lemma, there is a canonical object 𝜒´𝑇 𝑎𝑡𝑒 P
𝐻–mod𝑤𝑒𝑎𝑘 inverse to 𝜒𝑇 𝑎𝑡𝑒 under tensor product, i.e., we have:

𝜒´𝑇 𝑎𝑡𝑒 b 𝜒𝑇 𝑎𝑡𝑒 “ trivpVectq P 𝐻–mod𝑤𝑒𝑎𝑘 .

Proposition 7.18.2. For 𝐻 as above and C P 𝐻–mod𝑤𝑒𝑎𝑘 , there is a canonical isomorphism:

C𝐻,𝑤 » pC b 𝜒´𝑇 𝑎𝑡𝑒 q𝐻,𝑤

functorial in C.
Proof. We have:

pC b 𝜒´𝑇 𝑎𝑡𝑒 q𝐻,𝑤 “ Hom𝐻–mod𝑤𝑒𝑎𝑘 pVect, C b 𝜒´𝑇 𝑎𝑡𝑒 q “ Hom𝐻–mod𝑤𝑒𝑎𝑘 p𝜒𝑇 𝑎𝑡𝑒 , Cq “ C𝐻,𝑤
where the last equality was the definition of 𝜒𝑇 𝑎𝑡𝑒 .


7.19. The following result gives a somewhat non-canonical description of 𝜒𝑇 𝑎𝑡𝑒 .

Proposition 7.19.1. Let 𝐻 be a polarizable Tate group indscheme. Then for any compact open
subgroup 𝐾 Ď 𝐻, there exists a canonical isomorphism:

Oblv𝐻
𝐾 p𝜒𝑇 𝑎𝑡𝑒,𝐻 q » Vect P 𝐾–mod𝑤𝑒𝑎𝑘

for Oblv𝐻𝐾 : 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐾–mod𝑤𝑒𝑎𝑘 the restriction functor and for Vect P 𝐾–mod𝑤𝑒𝑎𝑘 regarded
with the trivial action.
80 SAM RASKIN

𝐻,𝑤
Proof. For, note that Oblv𝐻𝐾 admits a left adjoint ind𝐾 which also calculates its right adjoint.
Indeed, under the equivalences:

p´q𝐾,𝑤
𝐻–mod𝑤𝑒𝑎𝑘 » H𝐻,𝐾
𝑤
–mod
p´q𝐾,𝑤
𝐾–mod𝑤𝑒𝑎𝑘 » Repp𝐾q–mod
Oblv𝐻𝐾 corresponds to restriction along the monoidal functor Repp𝐾q Ñ H𝐻,𝐾 . This immediately
𝑤

gives the existence of the left adjoint, and the fact that it also calculates the right adjoint follows
from the fact that H𝐻,𝐾
𝑤 is canonically self-dual as a Repp𝐾q-module category (which is the case
because H𝐻,𝐾 is canonically self-dual as a DG category via Serre duality, and Repp𝐾q is rigid
symmetric monoidal).
Now for C P 𝐾–mod𝑤𝑒𝑎𝑘 , we obtain:
𝐻,𝑤
𝐾 p𝜒𝑇 𝑎𝑡𝑒,𝐻 q, Cq “ Hom𝐻–mod𝑤𝑒𝑎𝑘 p𝜒𝑇 𝑎𝑡𝑒,𝐻 , ind𝐾 pCqq “
Hom𝐾–mod𝑤𝑒𝑎𝑘 pOblv𝐻
ind𝐻,𝑤
𝐾 pCq𝐻,𝑤 “ C𝐾,𝑤 “ C
𝐾,𝑤

functorially in C, giving the claim.


Warning 7.19.2. Suppose 𝐾1 Ď 𝐾2 Ď 𝐻. Then we obtain isomorphisms:

𝛼𝑖 : Oblv𝐻
𝐾𝑖 p𝜒𝑇 𝑎𝑡𝑒,𝐻 q » Vect P 𝐾𝑖 –mod𝑤𝑒𝑎𝑘 , 𝑖 “ 1, 2
However, Oblv𝐾 2
𝐾1 p𝛼2 q ‰ 𝛼1 . Rather, one can check that the two isomorphisms differ by tensoring
with detpk2 {k1 qrdimpk2 {k1 qs P Repp𝐾1 q “ End𝐾1 –mod𝑤𝑒𝑎𝑘 pVectq.64

8. Strong actions
8.1. In this section, we relate weak actions for a Tate group indschemes 𝐻 to strong actions of 𝐻,
as defined in [Ber].
8.2. Let us spell out our goals more precisely. Let 𝐷˚ p𝐻q P AlgpDGCat𝑐𝑜𝑛𝑡 q be the monoidal DG
category defined (with the same notation) in [Ras3]. Let 𝐻–mod :“ 𝐷˚ p𝐻q–mod be the (2-)category
of categories with a strong 𝐻-action.
In this section, we will construct a restriction functor:

Oblv “ Oblv𝑠𝑡𝑟Ñ𝑤 : 𝐻–mod Ñ 𝐻–mod𝑤𝑒𝑎𝑘

compatible with forgetful functors to DGCat𝑐𝑜𝑛𝑡 (where for 𝐻–mod𝑤𝑒𝑎𝑘 , we are considering the
forgetful functor Oblv𝑔𝑒𝑛 of S7.14).
8.3. Moreover, we will show that Oblv : 𝐻–mod Ñ 𝐻–mod𝑤𝑒𝑎𝑘 admits a left and right adjoints
that are morphisms of DGCat𝑐𝑜𝑛𝑡 -module categories, and with the following property.
For C P 𝐻–mod𝑤𝑒𝑎𝑘 , define:
^
Cexpphq,𝑤 :“ colim C𝐻𝐾 ,𝑤 P DGCat𝑐𝑜𝑛𝑡
𝐾Ď𝐻 compact open
^ is the formal completion of 𝐻 along 𝐾, which is
under the obvious structural functors. Here 𝐻𝐾
necessarily a Tate group indscheme. That is, we consider the restriction of C along the forgetful
64This factor arises because the proof of Proposition 7.19.1 (necessarily) used Serre duality on IndCohp𝐻{𝐾q to
obtain the canonical self-duality for H𝐻,𝐾
𝑤
.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 81

functor 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻𝐾^ , apply the invariants construction of S7.15 for 𝐻 ^ , and pass to the
𝐾
colimit.
Similarly, define:

Cexpphq,𝑤 :“ lim C𝐻 𝐾
^ ,𝑤 .
𝐾Ď𝐻 compact open
As we will see, each of the structural functors in the limit (resp. colimit) defining Cexpphq,𝑤 (resp.
Cexpphq,𝑤 ) admits a left adjoint (resp. continuous right adjoint), so these two expressions can be
expressed as limits or colimits in DGCat𝑐𝑜𝑛𝑡 .
Then we will see that the composition of our right (resp. left) adjoint 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻–mod
with the forgetful functor 𝐻–mod Ñ DGCat𝑐𝑜𝑛𝑡 sends C P 𝐻–mod𝑤𝑒𝑎𝑘 to Cexpphq,𝑤 (resp. Cexpphq,𝑤 ).
In other words, we will show that 𝐻 acts strongly on Cexpphq,𝑤 and Cexpphq,𝑤 , and that these
categories satisfy the evident universal properties with respect to these actions and Oblv𝑠𝑡𝑟Ñ𝑤 .
Remark 8.3.1. If 𝐻 is polarizable, then it is straightforward to deduce from Proposition 7.18.2 that
the above functors p´qexpphq,𝑤 , p´qexpphq,𝑤 : 𝐻–mod𝑤𝑒𝑎𝑘 Ñ DGCat𝑐𝑜𝑛𝑡 are equivalent up to certain
twists. We will formulate this statement precisely in Proposition 8.21.1, where we will also show
that this isomorphism is strongly 𝐻-equivariant in a canonical way.
8.4. Strategy. To orient the reader in what follows, we give a brief overview of the approach.
To give a functor 𝐻–mod Ñ 𝐻–mod𝑤𝑒𝑎𝑘 the commutes with colimits and is a morphism of
DGCat𝑐𝑜𝑛𝑡 -module categories is equivalent to specifying an object of 𝐻–mod𝑤𝑒𝑎𝑘 with a right 𝐷˚ p𝐻q-
module structure.
In a suitable sense, this object is 𝐷˚ p𝐻q considered as weakly acted on via the left action of 𝐻,
and with the evident commuting strong action of 𝐻 on the right.
Implementing this strategy turns out the be somewhat involved. It is not so difficult to define
𝐷˚ p𝐻q as an object of 𝐻–mod𝑤𝑒𝑎𝑘 : this is done is S8.9. However, the commuting 𝐷˚ p𝐻q-action
takes some work, and will be given in S8.20.
8.5. Warmup. First, we discuss the case where 𝐻 is a classical affine group scheme. While do not
rely on this special case in the general construction, it is illustrative of the main ideas.
Let 𝐻 “ lim𝑖 𝐻𝑖 be a cofiltered limit of affine algebraic groups under smooth surjective homo-
morphisms. As above, to construct our functor:

𝐻–mod :“ 𝐷˚ p𝐻q–mod Ñ 𝐻–mod𝑤𝑒𝑎𝑘 “ Repp𝐻q–mod

is suffices to construct a pRepp𝐻q, 𝐷˚ p𝐻qq-bimodule in DGCat𝑐𝑜𝑛𝑡 .
This bimodule is h–mod P DGCat𝑐𝑜𝑛𝑡 (c.f. Example 4.3.8). We have:

h–mod “ lim h𝑖 –mod

𝑖
where each structural functor h𝑖 –mod Ñ h𝑗 –mod takes the Lie algebra invariants with respect to
Kerph𝑖  h𝑗 q. As is standard, 𝐻𝑖 acts strongly on h𝑖 –mod, and the above structural functors are
equivariant in the suitable homotopy coherent sense for the 𝐻𝑖 -action on h𝑗 –mod induced by the
homomorphism 𝐻𝑖 Ñ 𝐻𝑗 . Therefore, we obtain an action:

𝐷˚ p𝐻q :“ lim 𝐷p𝐻𝑖 q ñ lim h𝑖 –mod “ h–mod.

𝑖 𝑖
Now for any map of indices 𝑖 Ñ 𝑗, there is an action of Repp𝐻𝑗 q on h𝑖 –mod commuting with the
strong 𝐻𝑗 -action: it is given by restricting an 𝐻𝑗 -representation to 𝐻𝑖 and then tensoring with the
Lie algebra representation. Again, this is suitably homotopy coherent, so we obtain an action:
82 SAM RASKIN

𝐷˚ p𝐻q b Repp𝐻𝑗 q ñ h–mod.

Finally, these actions are suitably compatible with varying 𝑗, so we obtain:

𝐷˚ p𝐻q b Repp𝐻q “ 𝐷˚ p𝐻q b colim Repp𝐻𝑗 q ñ h–mod

𝑗
as desired.

8.6. A remark on naive coinvariants. Before proceeding, it is convenient to record the following
technical result. The reader may safely skip this material and refer back to it as needed.
Let 𝑆 be a reasonable indscheme and let P𝐾 Ñ 𝑆 be a 𝐾- torsor for 𝐾 a classical affine group
scheme. By Corollary 6.25.2, naive weak 𝐾-invariants in IndCoh˚ pP𝐾 q are given by IndCoh˚ p𝑆q.
Moreover, the naive 𝐾-action on IndCoh˚ pP𝐾 q clearly canonically renormalizes, and the corre-
sponding category of genuine 𝐾-invariants is IndCoh˚ p𝑆q by Lemma 6.18.1. This leaves the case of
naive coinvariants.
Lemma 8.6.1. In the above setting, the IndCoh-pushforward functor IndCoh˚ pP𝐾 q Ñ IndCoh˚ p𝑆q
induces an equivalence:

IndCoh˚ pP𝐾 q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ IndCoh˚ p𝑆q. (8.6.1)

Proof.
Step 1. First, note that we are reduced to the case where 𝑆 is a quasi-compact quasi-separated
eventually coconnective scheme. Indeed, if 𝑆 “ colim𝑖 𝑆𝑖 with 𝑆𝑖 P ą´8Sch𝑞𝑐𝑞𝑠 and structural maps
almost finitely presented, then:

IndCoh˚ p𝑆q “ colim IndCoh˚ p𝑆𝑖 q P DGCat𝑐𝑜𝑛𝑡

𝑖
IndCoh˚ pP𝐾 q “ colim IndCoh˚ pP𝐾 ˆ 𝑆𝑖 q P DGCat𝑐𝑜𝑛𝑡 .
𝑖 𝑆
This clearly gives the reduction. In the remainder of the argument, we therefore assume 𝑆 P
ą´8Sch
𝑞𝑐𝑞𝑠 .
»
Step 2. Next, suppose P𝐾 Ñ 𝑆 is trivial, i.e., P𝐾 Ý
Ñ 𝐾 ˆ 𝑆 𝐾-equivariantly. By Lemma 6.36.3
(applied to 𝐾), we have:
»
IndCoh˚ p𝐾q b IndCoh˚ p𝑆q Ý
Ñ IndCoh˚ pP𝐾 q.
»
Observe that Perfp𝐾q Ý Ñ Cohp𝐾q because 𝐾 is a limit of smooth schemes under flat affine mor-
phisms. Therefore, the above coincides with QCohp𝐾q b IndCoh˚ p𝑆q. This clearly gives the result
in this case.
Step 3. Next, we show the result when P𝐾 is Zariski-locally trivial. For this, we first establish some
general facts about IndCoh˚ .
Suppose 𝑗 : 𝑈 ãÑ 𝑆 is a quasi-compact open subscheme. Then the natural functor:

IndCoh˚ p𝑆q b QCohp𝑈 q Ñ IndCoh˚ p𝑈 q

QCohp𝑆q

is an equivalence. Indeed, IndCoh˚ p𝑈 q is the essential image of the functor 𝑗˚IndCoh 𝑗 ˚,IndCoh : IndCoh˚ p𝑆q Ñ
IndCoh˚ p𝑆q, while IndCoh˚ p𝑆q bQCohp𝑆q QCohp𝑈 q is the essential image of:
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 83

idIndCoh˚ p𝑆q b 𝑗˚ 𝑗 ˚ : IndCoh˚ p𝑆q “ IndCoh˚ p𝑆q b QCohp𝑆q Ñ

QCohp𝑆q QCohp𝑆q
IndCoh˚ p𝑆q b QCohp𝑆q “ IndCoh p𝑆q.˚
QCohp𝑆q

These endofunctors of IndCoh˚ p𝑆q coincide, giving the result.

As a consequence, suppose 𝑈1 , 𝑈2 Ď 𝑆 are quasi-compact opens covering 𝑆; then we claim that
the map:
ž
IndCoh˚ p𝑈1 q IndCoh˚ p𝑈2 q Ñ IndCoh˚ p𝑆q P DGCat𝑐𝑜𝑛𝑡
IndCoh˚ p𝑈1 X𝑈2 q

is an equivalence (this pushout being formed in DGCat𝑐𝑜𝑛𝑡 ). Indeed, it is well-known65 that we have:
ž
QCohp𝑈1 q QCohp𝑈2 q Ñ QCohp𝑆q P DGCat𝑐𝑜𝑛𝑡
QCohp𝑈1 X𝑈2 q

and therefore in QCohp𝑆q–mod. Tensoring IndCoh˚ p𝑆q over QCohp𝑆q preserves this colimit, giving
the claim from the above.
Now for any 𝑈1 , 𝑈2 Ď 𝑆 as above, we obtain:

ž »
IndCoh˚ pP𝐾 ˆ 𝑈1 q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 IndCoh˚ pP𝐾 ˆ 𝑈2 q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 Ý
Ñ IndCoh˚ pP𝐾 q𝐾𝑤
𝑆 𝑆
IndCoh˚ pP𝐾 ˆ𝑈1 X𝑈2 q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒
𝑆

by applying the above to the base-changed Zariski cover of P𝐾 and by commuting geometric
realizations with pushouts. We now clearly obtain the claim by induction on the number of opens
required to trivialize P𝐾 .
Step 4. Next, we show the result for 𝐾 prounipotent.
By the previous step, it suffices to note that any 𝐾-torsor on an affine scheme 𝑇 is trivial. This
is standard: prounipotent 𝐾 has a lower central series 𝐾 “ 𝐾 1 İ 𝐾 2 İ . . . where all subquotients
are (possibly infinite) products of copies of G𝑎 . For such products, the claim follows from vanishing
of higher (flat) cohomology of 𝑇 with coefficients in its structure sheaf. By induction, any 𝐾{𝐾 𝑛 -
torsor on an affine scheme is trivial, and then we deduce the same for 𝐾 using countability of this
filtration and surjectivity of 𝜋0 pHomp𝑇, 𝐾{𝐾 𝑛`1 qq Ñ 𝜋0 pHomp𝑇, 𝐾{𝐾 𝑛 qq.
Step 5. Finally, we show the result in general.
Let 𝐾 Ñ 𝐾 𝑟𝑒𝑑 be the proreductive66 quotient of 𝐾 𝑟𝑒𝑑 , and let 𝐾 𝑢 be the kernel of this homo-
morphism, i.e., the prounipotent radical of 𝐾.
Because representations of 𝐾 𝑟𝑒𝑑 are semisimple, for any C with a naive weak 𝐾 𝑟𝑒𝑑 action, the
𝑟𝑒𝑑
functor C𝐾 𝑟𝑒𝑑 ,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ C𝐾 ,𝑤,𝑛𝑎𝑖𝑣𝑒 is an equivalence. Indeed, the argument from [Gai8] S7.2 applies
just as well in the proreductive case as in the reductive one.
We then obtain:

» 𝑟𝑒𝑑 ,𝑤,𝑛𝑎𝑖𝑣𝑒
Ñ IndCoh˚ pP𝐾 𝑟𝑒𝑑 q𝐾
IndCoh˚ pP𝐾 q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 “ pIndCoh˚ pP𝐾 q𝐾 𝑢 ,𝑤,𝑛𝑎𝑖𝑣𝑒 q𝐾 𝑟𝑒𝑑 ,𝑤,𝑛𝑎𝑖𝑣𝑒 Ý

65This identity is implicit in the proof of [Gai1] Proposition 2.3.6. One can find this statement explicitly in [Gai8]
by combining Theorem 2.1.1 and Proposition 6.2.7 from loc. cit.
66Here we use proreductive as shorthand for pro-(algebraic group with reductive connected components).
84 SAM RASKIN

for P𝐾 𝑟𝑒𝑑 Ñ 𝑆 the induced 𝐾 𝑟𝑒𝑑 -torsor (appealing to the previous step here). We now obtain the
result by Corollary 6.25.2.


8.7. Induction. We begin with the following general lemma.

Lemma 8.7.1. Let 𝑓 : 𝐻1 Ñ 𝐻2 be a morphism in TateGp.

(1) The forgetful functor:

𝐻2 –mod𝑤𝑒𝑎𝑘 Ñ 𝐻1 –mod𝑤𝑒𝑎𝑘
admits a left adjoint ind𝑤 “ ind𝐻
𝐻1
2 ,𝑤
: 𝐻1 –mod𝑤𝑒𝑎𝑘 Ñ 𝐻2 –mod𝑤𝑒𝑎𝑘 .
(2) Suppose there exists 𝐾 Ď 𝐻1 compact open such that 𝑓 realizes 𝐾 as a compact open
subgroup of 𝐻2 as well.67
Then the diagram:

Oblv𝑔𝑒𝑛
𝐻1 –mod𝑤𝑒𝑎𝑘 / 𝐻1 –mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒
𝐻 ,𝑤 𝐻 ,𝑤,𝑛𝑎𝑖𝑣𝑒
ind𝐻2 ind𝐻2
1 1
 Oblv𝑔𝑒𝑛 
𝐻2 –mod𝑤𝑒𝑎𝑘 / 𝐻2 –mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒

commutes (where a priori it only commutes up to a natural transformation). Here the func-
tor on the right is tensoring over IndCoh˚ p𝐻1 q with IndCoh˚ p𝐻2 q.

Proof. As 𝐻𝑖 –mod𝑤𝑒𝑎𝑘 Ñ 𝐾𝑖 –mod𝑤𝑒𝑎𝑘 are (by construction) monadic functors, the first claim easily
reduces to the setting of S5.17.
Similarly, such considerations formally reduce the second claim to the case where 𝐻1 “ 𝐾. We
denote 𝐻2 simply by 𝐻 in this case. So we wish to show the diagram:

Oblv𝑔𝑒𝑛
𝐾–mod𝑤𝑒𝑎𝑘 / 𝐾–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒
𝐻 ,𝑤,𝑛𝑎𝑖𝑣𝑒
ind𝐻,𝑤
𝐾
ind𝐻2
1
 Oblv𝑔𝑒𝑛 
𝐻–mod𝑤𝑒𝑎𝑘 / 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒

commutes. Each of the functors involved commutes with colimits and is DGCat𝑐𝑜𝑛𝑡 -linear, so it
suffices to check that the diagram commutes when evaluated on the trivial representation Vect P
𝐾–mod𝑤𝑒𝑎𝑘 (since this object generates by definition). In this case, the claim is that the natural
map:

»
IndCoh˚ p𝐻q b Vect Ý
Ñ IndCohp𝐻{𝐾q.
˚
IndCoh p𝐾q

This follows from Lemma 8.6.1.



67Using Lemma 8.6.1, one can show that the conclusion holds more generally if there exist 𝐾 Ď 𝐻 compact open
𝑖 𝑖
subgroups such that 𝑓 maps 𝐾1 into 𝐾2 via a closed embedding.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 85

8.8. Now for 𝐻 a Tate group indscheme and 𝐾 Ď 𝐻 compact open, let 𝐻𝐾 ^ denote the formal

completion of 𝐻 along 𝐾. We can form ind𝐻,𝑤

𝐻𝐾
^
^ pVectq P 𝐻–mod𝑤𝑒𝑎𝑘 (where Vect P 𝐻𝐾 –mod𝑤𝑒𝑎𝑘 is

our standard trivial object). By Lemma 8.7.1, we have:

˚
Oblv𝑔𝑒𝑛 pind𝐻𝐾
^ pVectqq “ IndCoh p𝐻q b Vect.
IndCoh˚ p𝐻𝐾
^q

This tensor product evidently maps to 𝐷p𝐻{𝐾q, and we claim that the induced functor is an
equivalence. Indeed, we have a commutative diagram:

IndCohp𝐻{𝐾q IndCoh˚ p𝐻q b Vect / IndCoh˚ p𝐻q b Vect

IndCoh˚ p𝐾q IndCoh˚ p𝐻𝐾
^q

) u
𝐷p𝐻{𝐾q.
The diagonal arrows admit continuous, monadic right adjoints and the induced natural transfor-
mation on monads is an isomorphism, giving the claim.
In the above setting, we use our standard abuse of notation in letting 𝐷p𝐻{𝐾q P 𝐻–mod𝑤𝑒𝑎𝑘
denote the object ind𝐻,𝑤
𝐻 ^ pVectq.
𝐾

8.9. Note that this object is manifestly covariant in 𝐾, so we can form:

𝐷˚ p𝐻q :“ lim 𝐷p𝐻{𝐾q P 𝐻–mod𝑤𝑒𝑎𝑘 .

𝐾
Note that under Oblv𝑔𝑒𝑛 and the equivalence of S8.8, these structural functors map to de Rham
pushforward functors.
By definition (and Lemma 7.14.1), this object maps under Oblv𝑔𝑒𝑛 to the category 𝐷˚ p𝐻q P
DGCat𝑐𝑜𝑛𝑡 defined in [Ras3], justifying the notation.
Remark 8.9.1. Each of the structural functors in the above diagram admits a left adjoint in the
2-category 𝐻–mod𝑤𝑒𝑎𝑘 : indeed, these functors are given by 𝐷-module ˚-pullback along the smooth
maps 𝐻{𝐾1 Ñ 𝐻{𝐾2 .68 Therefore, this limit is also a colimit (in 𝐻–mod𝑤𝑒𝑎𝑘 ) under those left
adjoints.
8.10. By Remark 8.9.1, 𝐷˚ p𝐻q P 𝐻–mod𝑤𝑒𝑎𝑘 corepresents the functor:
^ ^
C ÞÑ lim C𝐻𝐾 ,𝑤 “ colim C𝐻𝐾 ,𝑤 “ Cexpphq,𝑤
𝐾Ď𝐻 compact open 𝐾Ď𝐻 compact open
where the structural functors in the colimit are the evident forgetful functors, and the structural
functors in the limit are their right adjoints.

8.11. Below, we will construct an action of 𝐷˚ p𝐻q P AlgpDGCat𝑐𝑜𝑛𝑡 q on this object 𝐷˚ p𝐻q P
𝐻–mod𝑤𝑒𝑎𝑘 encoding the right action of 𝐻 on itself. As in S8.4, this would suffice to construct a
functor of the desired type. By the discussion of S8.10, the formula from S8.3 for the right adjoint
would be immediate, and the formula for the left adjoint would follow dually.
Therefore, we will give this construction below following a sequence of digressions.

68This discussion is a bit informal, since it really applies after applying Oblv
𝑔𝑒𝑛 . But e.g., it easily follows from
Lemma 8.12.2 below that the left adjoints exist in the genuine setting as well.
86 SAM RASKIN

8.12. Some generalities on 𝐷-modules. We make the above construction somewhat more ex-
plicit. The reader may safely skip this material and refer back to it as needed.
First, it is convenient to extend the generality of the above construction. Let 𝑋 be an indscheme
locally almost of finite type, and suppose 𝑋 is acted on by 𝐻. Then we there is a canonical object
𝐷p𝑋q P 𝐻–mod𝑤𝑒𝑎𝑘 attached to 𝑋 (and mapping to the category of 𝐷-modules on 𝑋 under
Oblv𝑔𝑒𝑛 ).
We sketch the construction. A variant of Proposition-Construction 7.6.1 attaches an object
IndCohp𝑋q P 𝐻–mod𝑤𝑒𝑎𝑘 to 𝑋 such that for any congruence subgroup 𝐾, IndCohp𝑋q𝐾,𝑤 “ IndCoh˚𝑟𝑒𝑛 p𝑋{𝐾q
as an H𝐻,𝐾
𝑤 -module.
𝑜𝑝
Now let 𝑋‚𝑖𝑛𝑓 P IndSchΔ 𝑙𝑎𝑓 𝑡 be the infinitesimal groupoid of 𝑋, i.e., the simplicial indscheme
locally almost of finite type obtained as the Cech nerve of 𝑋 Ñ 𝑋𝑑𝑅 . By functoriality, this di-
agram is a simplicial diagram of indschemes (locally almost of finite) acted on by 𝐻. Therefore,
by the above construction, we obtain a simplicial diagram IndCohp𝑋‚𝑖𝑛𝑓 q P 𝐻–mod𝑤𝑒𝑎𝑘 . We define
𝐷p𝑋q P 𝐻–mod𝑤𝑒𝑎𝑘 as its colimit. By Lemma 7.14.1 and [GR4] Proposition III.[Link](b), this
object indeed maps to the usual category of 𝐷-modules 𝐷p𝑋q P DGCat𝑐𝑜𝑛𝑡 under the forgetful
functor 𝐻–mod𝑤𝑒𝑎𝑘 Ñ DGCat𝑐𝑜𝑛𝑡 .
Now that for any choice of compact open subgroup 𝐾 Ď 𝐻, we have 𝐷p𝑋q𝐾,𝑤 P H𝐻,𝐾 𝑤 –mod. By

construction 𝐷p𝑋q 𝐾,𝑤 is compactly generated with compact objects induced from IndCoh˚𝑟𝑒𝑛 p𝑋{𝐾q “
IndCohp𝑋q𝐾,𝑤 .
Lemma 8.12.1. For 𝐾 Ď 𝐻 compact open, the object 𝐷p𝐻{𝐾q defined in S8.8 coincides with the
object we have just constructed.
Proof. This is essentially a slight refinement of the argument from S8.8.
Let C1 P 𝐻–mod𝑤𝑒𝑎𝑘 denote the object from S8.8 and let C2 P 𝐻–mod𝑤𝑒𝑎𝑘 denote the object just
constructed (i.e., from the construction defined for any indscheme locally almost of finite type).
There is a natural map C1 “ ind𝐻,𝑤 𝐻𝐾^ pVectq Ñ C2 , and we claim it is an isomorphism. It suffices to

check this after applying weak 𝐾-invariants.

By construction, we have:

C𝐾,𝑤
1 “ Repp𝐾q b H𝐻,𝐾
𝑤
.
H𝐻
𝑤
^ ,𝐾
𝐾

This gives rise to a functor:

IndCoh˚𝑟𝑒𝑛 p𝐾z𝐻{𝐾q “ Repp𝐾q b H𝐻,𝐾

𝑤
Ñ Repp𝐾q b H𝐻,𝐾
𝑤
“ C𝐾,𝑤
1 .
Repp𝐾q H𝐻
𝑤
^ ,𝐾
𝐾

This functor admits a right adjoint that is continuous and conservative, so monadic. The further
composition with C𝐾,𝑤
1 Ñ C𝐾,𝑤
2 behaves similarly (by construction), and the induced maps on
monads is an isomorphism, giving the claim.


We now show the following result for 𝑋 any indscheme locally almost of finite type.
Lemma 8.12.2. 𝐷p𝑋q𝐾,𝑤 admits a (unique) compactly generated 𝑡-structure for which the for-
»
getful functor 𝐷p𝑋q𝐾,𝑤 Ñ 𝐷p𝑋q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 is 𝑡-exact and induces an equivalence 𝐷p𝑋q𝐾,𝑤,` Ý Ñ
𝐷p𝑋q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒,` . An object in 𝐷p𝑋q𝐾,𝑤 is compact if and only if it is eventually coconnective and
its image in 𝐷p𝑋q is compact.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 87

In other words, as a category with a genuine 𝐾-action, 𝐷p𝑋q is given by the canonical renor-
malization construction from S5.18.

Proof of Lemma 8.12.2. Define the 𝑡-structure on 𝐷p𝑋q𝐾,𝑤 by taking connective objects to be
generated by objects induced from IndCohp𝑋q𝐾,𝑤,ď0 . To see 𝐷p𝑋q𝐾,𝑤 Ñ 𝐷p𝑋q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 is 𝑡-exact,
it is equivalent to see that the further forgetful functor:

Oblv : 𝐷p𝑋q𝐾,𝑤 Ñ 𝐷p𝑋q

is 𝑡-exact. Clearly this functor is right 𝑡-exact. Now for F P 𝐷p𝑋q𝐾,𝑤,ě0 , the underlying object of
IndCohp𝑋q𝐾,𝑤 is coconnective by design, so the same is true for the underlying object of IndCohp𝑋q.
Therefore, OblvpFq P 𝐷p𝑋q maps to a coconnective object of IndCohp𝑋q; this is equivalent to
OblvpFq being coconnective, as desired.
Because the functor 𝐷p𝑋q𝐾,𝑤,` Ñ 𝐷p𝑋q` is 𝑡-exact and these 𝑡-structures are right complete,
this functor is comonadic. The forgetful functor 𝐷p𝑋q𝐾,𝑤 Ñ 𝐷p𝑋q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 induces an equivalence
on the corresponding comonads on 𝐷p𝑋q, and the latter category maps comonadically to 𝐷p𝑋q.
»
This implies 𝐷p𝑋q𝐾,𝑤,` Ý Ñ 𝐷p𝑋q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒,` .
The last part is proved similarly to Lemma 5.20.1 (4). We need to show that if F P 𝐷p𝑋q𝐾,𝑤,` “
𝐷p𝑋q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒,` has OblvpFq P 𝐷p𝑋q compact, then F is compact in 𝐷p𝑋q𝐾,𝑤 . We are clearly
reduced to the case where 𝑋 is classical. In this case, 𝑋 is a colimit under closed embeddings of
finite type schemes acted on by 𝐾, so we are further reduced to the case where 𝑋 is a finite type
scheme. Moreover, we can assume F lies in the heart of the 𝑡-structure, since it is bounded and
each of its cohomology groups satisfy the same hypothesis.
Now there exists 𝐾 1 C 𝐾 compact open (i.e., 𝐾{𝐾 1 is an affine algebraic group) with the action
of 𝐾 on 𝑋 factoring through 𝐾{𝐾 1 . Further, as in the proof of Lemma 5.20.1 (4), the hypothesis
on F implies that there is a compact open subgroup 𝐾 2 Ď 𝐾 1 also normal in 𝐾 such that F lies in
the essential image of the functor:
2 ,𝑤,𝑛𝑎𝑖𝑣𝑒,`
𝐷p𝑋q𝐾{𝐾 Ñ 𝐷p𝑋q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒,` .
Now the result follows from Lemma 5.20.2 (applied to 𝐾{𝐾 2 ).


8.13. Naive Hecke actions. Suppose that 𝐻 is a Tate group indscheme and 𝐾 Ď 𝐻 is compact
open. Suppose 𝐻 acts naively on C, i.e., C is a module for IndCoh˚ p𝐻q. Then we claim there is an
induced action of the monoidal category:69

𝑤,𝑛𝑎𝑖𝑣𝑒
H𝐻,𝐾 :“ IndCoh˚ p𝐾z𝐻{𝐾q “ IndCoh˚ p𝐻{𝐾q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒
on C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 .
Indeed, by Lemma 8.6.1, the IndCohp𝐻{𝐾q P 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 corepresents the functor of naive
𝐾-invariants, so we obtain:

𝑤,𝑛𝑎𝑖𝑣𝑒
H𝐻,𝐾 “ End𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 pIndCohp𝐻{𝐾qq ñ Hom𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 pIndCohp𝐻{𝐾q, Cq.
The following result is a formal consequence of Remark 5.15.4.

69We emphasize that the middle term uses non-renormalized IndCoh˚ .

88 SAM RASKIN

Lemma 8.13.1. The functor:

𝑤,𝑛𝑎𝑖𝑣𝑒
𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 Ñ H𝐻,𝐾 –mod
constructed above is fully-faithful.
Proof. First, suppose C is equipped with a naive action of the compact open subgroup 𝐾. Then
the natural functor:

Vect b C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ C
Rep𝑛𝑎𝑖𝑣𝑒 p𝐾q
is an equivalence. Indeed, this is the content of Remark 5.15.4 (and is shown in [Ras4] Proposition
3.5.1).
In particular, we obtain:
𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒
Vect b H𝐻,𝐾 » IndCohp𝐻{𝐾q.
Rep𝑛𝑎𝑖𝑣𝑒 p𝐾q
𝑤𝑚,𝑛𝑎𝑖𝑣𝑒
Clearly this is an equivalence of H𝐻,𝐾 - module categories.
Now for any C with a naive weak action of 𝐻, we calculate:
𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒
IndCoh˚ p𝐻{𝐾q b C𝐾,𝑤 “ Vect b H𝐻,𝐾 b C𝐾,𝑤 “
𝑤,𝑛𝑎𝑖𝑣𝑒 𝑤,𝑛𝑎𝑖𝑣𝑒
H𝐻,𝐾 Rep𝑛𝑎𝑖𝑣𝑒 p𝐾q H𝐻,𝐾
»
Vect b Ñ C.
C𝐾,𝑤 Ý
Rep𝑛𝑎𝑖𝑣𝑒 p𝐾q
𝑤,𝑛𝑎𝑖𝑣𝑒
This map is obviously the counit for the evident adjunction 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 Ô H𝐻,𝐾 –mod, so
we obtain the claim.

8.14. Canonical renormalization. We now wish to give an analogue of the construction from
S5.18 in the Tate setting.
8.15. We begin with some general results about renormalizing monoidal structures and module
structures.
8.16. We will need the following general constructions in what follows. The reader may safely skip
this material and refer back to it as necessary.
Lemma 8.16.1. Suppose we are given:
‚ pA, ‹q P AlgpDGCat𝑐𝑜𝑛𝑡 q a monoidal DG category.
‚ A𝑟𝑒𝑛 P DGCat𝑐𝑜𝑛𝑡 a compactly generated DG category with A𝑐𝑟𝑒𝑛 its subcategory of compact
objects.
‚ 𝑡-structures on A and A𝑟𝑒𝑛 compatible with filtered colimits.
‚ A 𝑡-exact functor Ψ : A𝑟𝑒𝑛 Ñ A commuting with colimits and inducing an equivalence
»
A` Ñ A` on eventually coconnnective subcategories.
𝑟𝑒𝑛 Ý
Suppose in addition that the following properties are satisfied:
(1) The unit object 1 P A lies in A` .
(2) A𝑐𝑟𝑒𝑛 is contained in A`𝑟𝑒𝑛 .
(3) For every F P A𝑐𝑟𝑒𝑛 , the functors ΨpFq ‹ ´ : A Ñ A and ´ ‹ ΨpFq : A Ñ A are left 𝑡-exact
up to shift.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 89

(4) For every F P A𝑐𝑟𝑒𝑛 , the continuous functors A𝑟𝑒𝑛 Ñ A𝑟𝑒𝑛 defined by ind-extension of:
ΨpFq‹Ψp´q
A𝑐𝑟𝑒𝑛 ÝÝÝÝÝÝÝÑ A` » A`
𝑟𝑒𝑛 Ď A𝑟𝑒𝑛
Ψp´q‹ΨpFq
A𝑐𝑟𝑒𝑛 ÝÝÝÝÝÝÝÑ A` » A` 𝑟𝑒𝑛 Ď A𝑟𝑒𝑛
are left 𝑡-exact up to shift.
Then A𝑟𝑒𝑛 admits a unique monoidal structure such that:
‚ The functor Ψ admits a monoidal structure.
‚ For every F P A𝑐𝑟𝑒𝑛 , the functors F ‹ ´ : A𝑟𝑒𝑛 Ñ A𝑟𝑒𝑛 and ´ ‹ F : A𝑟𝑒𝑛 Ñ A𝑟𝑒𝑛 preserve
A`𝑟𝑒𝑛 .

Proof.
Step 1. We begin with a general construction.
DGCat𝑐𝑜𝑛𝑡 pA𝑟𝑒𝑛 q Ď EndDGCat𝑐𝑜𝑛𝑡 pA𝑟𝑒𝑛 q the subcategory of functors 𝐹 : A𝑟𝑒𝑛 Ñ
Let us denote by Endą´8
A𝑟𝑒𝑛 that are left 𝑡-exact up to shift. Then the restriction functor:

Endą´8 `
DGCat𝑐𝑜𝑛𝑡 pA𝑟𝑒𝑛 q Ñ EndDGCat pA𝑟𝑒𝑛 q (8.16.1)
is fully-faithful. For this, define End𝐿𝐾𝐸 ` `
DGCat pA𝑟𝑒𝑛 q Ď EndDGCat pA𝑟𝑒𝑛 q to be the subcategory of functors
left Kan extended from their restrictions to A𝑟𝑒𝑛 . Then (8.16.1) clearly maps through this subcat-
𝑐

egory. Now the restriction functor End𝐿𝐾𝐸 DGCat pA𝑟𝑒𝑛 q Ñ HomDGCat pA𝑟𝑒𝑛 , A𝑟𝑒𝑛 q is fully-faithful, and so
` 𝑐 `

is its composition with (8.16.1), so (8.16.1) is fully-faithful.

We remark that the essential image of (8.16.1) consists of those DG functors 𝐹 : A` 𝑟𝑒𝑛 Ñ A𝑟𝑒𝑛
`

that are left Kan extended from A𝑟𝑒𝑛 and such that the resulting ind-extended functor A𝑟𝑒𝑛 Ñ A𝑟𝑒𝑛
𝑐

is left 𝑡-exact up to shift.

Finally, we remark that (8.16.1) is manifestly a monoidal DG functor (between non-cocomplete
DG categories).
Step 2. Next, we define an auxiliary category.
Let B𝑐 Ď A be the full subcategory Karoubi generated by objects of the form ΨpF1 q ‹ . . . ‹ ΨpF𝑛 q
for F1 , . . . , F𝑛 P A𝑐𝑟𝑒𝑛 . (We allow 𝑛 “ 0, i.e., 1 is one of our generators of B.)
Clearly B𝑐 is an essentially small monoidal DG category; let B :“ IndpB𝑐 q P AlgpDGCat𝑐𝑜𝑛𝑡 q.
Note that B𝑐 Ď A` by assumption. Define a continuous DG functor 𝜁 : B Ñ A𝑟𝑒𝑛 by ind-
extension from:

B𝑐 Ď A ` » A ` 𝑟𝑒𝑛 Ď A𝑟𝑒𝑛 .
We remark that 𝜁 is a colocalization functor, i.e., it admits a fully-faithful left adjoint. Namely,
this left adjoint is the ind-extension of the fully-faithful functor Ψ : A𝑐𝑟𝑒𝑛 Ñ B𝑐 Ď A` .
Step 3. We now construct a B-bimodule structure on A𝑟𝑒𝑛 in DGCat𝑐𝑜𝑛𝑡 .
Let e.g. B𝑚𝑜𝑛–𝑜𝑝 denote B with its monoidal structure reversed. So we wish to construct a
continuous monoidal DG functor B b B𝑚𝑜𝑛–𝑜𝑝 Ñ EndDGCat𝑐𝑜𝑛𝑡 pA𝑟𝑒𝑛 q. This is equivalent to giving a
monoidal DG functor:

pB b B𝑚𝑜𝑛–𝑜𝑝 q𝑐 “ B𝑐 bB𝑐,𝑚𝑜𝑛–𝑜𝑝 Ñ EndDGCat𝑐𝑜𝑛𝑡 pA𝑟𝑒𝑛 q.

(We remind that b indicates the tensor product on the category of small DG categories.)
Note that A` » A` 𝑟𝑒𝑛 is an B -bimodule (in DGCat) by our assumption (3). Therefore, we obtain
𝑐

a monoidal functor:
90 SAM RASKIN

B𝑐 bB𝑐,𝑚𝑜𝑛–𝑜𝑝 Ñ EndDGCat pA`

𝑟𝑒𝑛 q.
Moreover, by assumption (4), this functor maps into the essential image of (8.16.1). Therefore, it
lifts canonically to a monoidal functor:

B𝑐 bB𝑐,𝑚𝑜𝑛–𝑜𝑝 Ñ Endą´8
DGCat𝑐𝑜𝑛𝑡 pA𝑟𝑒𝑛 q Ď EndDGCat𝑐𝑜𝑛𝑡 pA𝑟𝑒𝑛 q
as desired.
Step 4. Next, observe that our functor 𝜁 from above is a morphism of B-bimodule categories (in
DGCat𝑐𝑜𝑛𝑡 ) Indeed, this results from the fact that the embedding B𝑐 ãÑ A` is a morphism of
B𝑐 -bimodule categories (in DGCat).
In particular, Kerp𝜁q is a two-sided monoidal ideal in B. As 𝜁 was a colocalization DG functor,
this means that A𝑟𝑒𝑛 admits a unique monoidal structure such that 𝜁 is monoidal. This monoidal
structure clearly has the desired properties.

Example 8.16.2. Note that the assumption (4) is automatic given the other assumptions (notably,
(3)) if compact objects in A𝑐𝑟𝑒𝑛 are closed under truncations.
𝑤,𝑛𝑎𝑖𝑣𝑒
Example 8.16.3. By Example 8.16.2, Lemma 8.16.1 applies for A𝑟𝑒𝑛 “ H𝐻,𝐾 𝑤 Ñ H𝐻,𝐾 “ A.
In particular, it may be used to directly construct the monoidal structure on H𝐻,𝐾
𝑤 from that of
𝑤,𝑛𝑎𝑖𝑣𝑒
H𝐻,𝐾 .
We will also need a variant of the above construction for module categories.
Lemma 8.16.4. In the setting of Lemma 8.16.1, suppose we are additionally given:
‚ M P DGCat𝑐𝑜𝑛𝑡 a module category (in DGCat𝑐𝑜𝑛𝑡 ) for A.
‚ M𝑟𝑒𝑛 P DGCat𝑐𝑜𝑛𝑡 a compactly generated DG category.
‚ 𝑡-structures on M and M𝑟𝑒𝑛 compatible with filtered colimits and such that M𝑐𝑟𝑒𝑛 (the sub-
category of compact objects) is contained in M`𝑟𝑒𝑛 .
‚ A 𝑡-exact functor 𝜓 : M𝑟𝑒𝑛 Ñ M commuting with colimits and inducing an equivalence
»
M` Ñ M` .
𝑟𝑒𝑛 Ý
Suppose that:
(1) For every F P A𝑐𝑟𝑒𝑛 , the functor ΨpFq ‹ ´ : M Ñ M preserves M` .
(2) For every F P A𝑐𝑟𝑒𝑛 , the continuous functor M𝑟𝑒𝑛 Ñ M𝑟𝑒𝑛 defined by ind-extension from:
ΨpFq‹𝜓p´q
M𝑐𝑟𝑒𝑛 ÝÝÝÝÝÝÝÑ M` » M` 𝑟𝑒𝑛 Ď M𝑟𝑒𝑛
is left 𝑡-exact up to shift.
(3) For every G P M𝑐𝑟𝑒𝑛 , the functor ´ ‹ 𝜓pGq : A Ñ M maps A` to M` .
(4) For every G P M𝑐𝑟𝑒𝑛 , the continuous functor A𝑟𝑒𝑛 Ñ M𝑟𝑒𝑛 defined by ind-extension from:
Ψp´q‹𝜓pGq
A𝑐𝑟𝑒𝑛 ÝÝÝÝÝÝÝÑ M` » M` 𝑟𝑒𝑛 Ď M𝑟𝑒𝑛
is left 𝑡-exact up to shift.
Then there is a unique action of A𝑟𝑒𝑛 on M𝑟𝑒𝑛 such that:
‚ The functor 𝜓 : M𝑟𝑒𝑛 Ñ M is a morphism of A𝑟𝑒𝑛 -module categories, where A𝑟𝑒𝑛 acts on
M by restriction along Ψ : A𝑟𝑒𝑛 Ñ A.
‚ For every F P A𝑐𝑟𝑒𝑛 , the functor F ‹ ´ : M𝑟𝑒𝑛 Ñ M𝑟𝑒𝑛 preserves M`
𝑟𝑒𝑛 .
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 91

Proof. As in the proof of Lemma 8.16.1, the restriction functor:

Endą´8 `
DGCat𝑐𝑜𝑛𝑡 pM𝑟𝑒𝑛 q Ñ EndDGCat pM𝑟𝑒𝑛 q (8.16.2)
is fully-faithful, using similar notation as in that argument.
We use the notation from the proof of Lemma 8.16.1 freely below. By assumption, the (non-
cocomplete) monoidal DG category B𝑐 is equipped with a monoidal DG functor to the right hand
side of (8.16.2) and maps into the essential image of that functor by assumption, so we obtain an
induced action of B on M𝑟𝑒𝑛 . We again denote this action using the notation ‹.
As in the proof of Lemma 8.16.1, the monoidal functor 𝜁 admits a fully-faithful left adjoint
𝜉 : A𝑟𝑒𝑛 ãÑ B.
𝜉p´q‹G
We will use the following observation. Let G P M𝑐𝑟𝑒𝑛 . By construction, the functor A𝑟𝑒𝑛 ÝÝÝÝÑ
M𝑟𝑒𝑛 is ind-extended from the composition:

´‹𝜓pGq
A𝑐𝑟𝑒𝑛 Ñ A` ÝÝÝÝÑ M` » M`
𝑟𝑒𝑛 Ď M𝑟𝑒𝑛 .

Therefore, our assumptions imply that this functor is left 𝑡-exact up to shift.
Now observe that 𝜉 is automatically left lax monoidal. Therefore, it suffices to show:
(1) For G P M𝑟𝑒𝑛 , the natural map:

𝜉p1A q ‹ G Ñ 1B ‹ G “ G
is an isomorphism.
(2) For F1 , F2 P A𝑟𝑒𝑛 and G P M𝑟𝑒𝑛 , the natural map:

𝜉pF1 q ‹ 𝜉pF2 q ‹ G Ñ 𝜉pF1 ‹ F2 q ‹ G

is an isomorphism.
As 𝜉 is a left adjoint, each of the functors appearing above commutes with colimits in each
variable. Therefore, we may assume G P M𝑐𝑟𝑒𝑛 Ď M` 𝑟𝑒𝑛 in each of the above cases, and F1 , F2 P A𝑟𝑒𝑛
𝑐

in (2).
For (1), note that 𝜉p1A q ‹ G P M` 𝑟𝑒𝑛 by the observation above, so as the same is true for G, it
suffices to check that the map is an isomorphism after applying 𝜓; this is clear.
For (2), the functors 𝜉pF𝑖 q ‹ ´ : M𝑟𝑒𝑛 Ñ M𝑟𝑒𝑛 are preserve M` 𝑟𝑒𝑛 by construction, so again the
two terms we are comparing lie in M𝑟𝑒𝑛 so it suffices to (trivially) observe that the relevant map
`

becomes an isomorphism after applying 𝜓.


Example 8.16.5. As in Example 8.16.2, assumption (2) (resp. (4)) is automatic if compact objects
in M𝑟𝑒𝑛 (resp. A𝑟𝑒𝑛 ) are closed under truncations.
8.17. Suppose 𝐻 is a Tate group indscheme and suppose C P DGCat𝑐𝑜𝑛𝑡 is acted on naively by 𝐻.
Suppose in addition that C is equipped with a 𝑡-structure.
Definition 8.17.1. The naive action of 𝐻 on C canonically renormalizes (relative to the 𝑡-structure)
if:
‚ For every compact open subgroup 𝐾 Ď 𝐻, the induced naive action of 𝐾 on C canonically
renormalizes (in the sense of S5.18).
92 SAM RASKIN

‚ For every compact open subgroup 𝐾 Ď 𝐻, the data:

𝑤,𝑛𝑎𝑖𝑣𝑒
Ψ : A𝑟𝑒𝑛 “ H𝐻,𝐾
𝑤
Ñ A “ H𝐻,𝐾
𝜓 : M𝑟𝑒𝑛 “ C𝐾,𝑤 Ñ M “ C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒
satisfy the hypotheses of Lemma 8.16.4. (Here C𝐾,𝑤 is defined as in S5.18.)
‚ For every pair 𝐾1 Ď 𝐾2 Ď 𝐻 of embedded compact open subgroups of 𝐻, the morphism:

C𝐾2 ,𝑤 b Repp𝐾1 q Ñ C𝐾1 ,𝑤

Repp𝐾2 q
of Lemma 5.20.1 is an equivalence.
Remark 8.17.2. Technically there is some room for confusion: if 𝐻 is itself a classical affine group
scheme, then this condition is a bit more stringent than the one from S5.18. The author hopes that
this will not cause any confusion.
Proposition 8.17.3. Suppose C is equipped with a 𝑡-structure and a naive action of 𝐻 that canon-
ically renormalize.
Define the category GenpCq to consist of objects D P 𝐻–mod𝑤𝑒𝑎𝑘 equipped with an isomorphism
Oblv𝑔𝑒𝑛 pDq » C P 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 (c.f. S7.14) and with the property that for any compact open
subgroup 𝐾, D𝐾,𝑤 is compactly generated and the induced functor:

D𝐾,𝑤 Ñ D𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 “ C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒

»
is fully-faithful on compact objects and induces an isomorphism D𝐾,𝑤,𝑐 Ý Ñ C𝐾,𝑤,𝑐 (the right hand
side being defined by S5.18).
Then the category GenpCq is contractible, i.e., equivalent to ˚ P Gpd Ď Cat.
Proof. Fix a compact open subgroup 𝐾 Ď 𝐻. Define Gen𝐾 pCq to consist of D P 𝐻–mod𝑤𝑒𝑎𝑘
equipped with an isomorphism Oblv𝑔𝑒𝑛 pDq » C and satisfying the similar property as for GenpCq,
but only for 𝐾 (not for all compact open subgroups). Clearly GenpCq Ď Gen𝐾 pCq is a full subcate-
gory. Therefore, it suffices to show that Gen𝐾 pCq is contractible and that GenpCq is non-empty.
We need to check that GenpCq is non-empty. Let C𝐾,𝑤 be defined as by canonical renormaliza-
tion for 𝐾. By Lemma 8.16.4 (and by definition of canonical renormalization for 𝐻), there is a
canonical H𝐻,𝐾
𝑤 -module structure on C𝐾,𝑤 . Let 𝜄 pCq P 𝐻–mod
𝐾 𝑤𝑒𝑎𝑘 be the corresponding object
with 𝜄𝐾 pCq 𝐾,𝑤 “C 𝐾,𝑤 (as H𝐻,𝐾 -modules). There is an evident isomorphism Oblv𝑔𝑒𝑛 𝜄𝐾 pCq “ C P
𝑤

𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 . This construction clearly defines an object of Gen𝐾 pCq, and contractibility of the
category is immediate from Lemmas 8.16.4 and 8.13.1.
We claim moreover that the above construction defines an object of GenpCq.
For any 𝐾 1 Ď 𝐾, we have:
1 1
𝜄𝐾 pCq𝐾 ,𝑤 “ 𝜄𝐾 pCq𝐾,𝑤 b Repp𝐾 1 q “ C𝐾,𝑤 b Repp𝐾 1 q Ñ C𝐾 ,𝑤
Repp𝐾q Repp𝐾q
and this functor satisfies the conclusions of Lemma 8.16.4 with respect to the Hecke categories
relative to 𝐾 1 . This gives an isomorphism 𝜄𝐾 pCq » 𝜄𝐾 1 pCq P 𝐻–mod𝑤𝑒𝑎𝑘 .
As the intersection of compact open subgroups is again compact open, we see that for any (pos-
sibly not nested) 𝐾, 𝐾 1 Ď 𝐻 compact open subgroups, there exists an isomorphism 𝜄𝐾 pCq𝜄𝐾 pCq »
𝜄𝐾 1 pCq. This implies that:
č
𝜄𝐾 pCq P Gen𝐾 1 pCq “: GenpCq
𝐾1
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 93

as desired.

Notation 8.17.4. In the above setting, we follow our standard abuses of notation in letting C P
𝐻–mod𝑤𝑒𝑎𝑘 denote the canonical object constructed via Proposition 8.17.3 (namely, the object
defined by the forgetful functor ˚ “ GenpCq Ñ 𝐻–mod𝑤𝑒𝑎𝑘 ).
We also need the following variant.
Proposition 8.17.5. Suppose70 A P AlgpDGCat𝑐𝑜𝑛𝑡 q is equipped with a 𝑡-structure such that id :
A Ñ A satisfies the hypotheses for the functor “Ψ” from Lemma [Link]
Suppose C P DGCat𝑐𝑜𝑛𝑡 is equipped with a 𝑡-structure and an IndCoh˚ p𝐻q b A-module structure
such that:
‚ The underlying naive 𝐻-action canonically renormalizes.
‚ For any 𝐾 Ď 𝐻 compact open, the evident A-action on C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 satisfies the hypotheses
of Lemma 8.16.4 relative to 𝜓 : C𝐾,𝑤 Ñ C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 P DGCat𝑐𝑜𝑛𝑡 (and id : A Ñ A).
Then:
(1) For any compact open subgroup 𝐾 Ď 𝐻, the morphism:
𝑤,𝑛𝑎𝑖𝑣𝑒
H𝐻,𝐾
𝑤
b A Ñ H𝐻,𝐾 bA
satisfies the hypotheses of Lemma 8.16.1, where both sides are equipped with the natural
tensor product 𝑡-structures.
𝑤,𝑛𝑎𝑖𝑣𝑒
Moreover, the corresponding action of H𝐻,𝐾 b A on C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 satisfies the hypotheses
of Lemma 8.16.4 (relative to the above functor and C𝐾,𝑤 Ñ C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 ).
(2) Define the category GenA pCq to consist of objects72 D P A–modp𝐻–mod𝑤𝑒𝑎𝑘 q equipped with
an isomorphism Oblv𝑔𝑒𝑛 pDq » C P A–modp𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 q (c.f. S7.14) and with the
property that on forgetting the A-action, D defines an object of GenpCq (as in the notation
of Proposition 8.17.3).
Then the category GenA pCq is contractible, i.e., equivalent to ˚.
Proof. (1) is immediate from Lemma 4.6.2 (2). Then (2) follows by the exact same argument as in
Proposition 8.17.3.

8.18. Preliminary remarks about 𝐷-modules. Let 𝐻 be a Tate group indscheme and fix a
compact open subgroup 𝐾0 .
Then 𝐾0 induces a 𝑡-structure on 𝐷˚ p𝐻q. Indeed, we have:

𝐷˚ p𝐻q “ colim 𝐷˚ p𝐻{𝐾q

𝐾Ď𝐾0 Ď𝐻 compact open
under ˚-pullback functors (which are defined as each pullback here is smooth). As this colimit is
filtered and these functors are all 𝑡-exact up to shift (being smooth pullbacks), we obtain the claim.
Explicitly, this 𝑡-structure is normalized by the fact that for each projection 𝜋𝐾 : 𝐻 Ñ 𝐻{𝐾, the
˚,𝑑𝑅
functor 𝜋𝐾 r´ dimp𝐾0 {𝐾qs : 𝐷p𝐻{𝐾q Ñ 𝐷˚ p𝐻q is 𝑡-exact.
70The notation is potentially misleading: this category A behaves more like the category A
𝑟𝑒𝑛 from Lemma 8.16.1.
71This is just a convenient way to say A is compactly generated with the action of compact objects being given
by functors that are left 𝑡-exact up to shift and with unit being eventually coconnective.
72Here A-modules in 𝐻–mod
𝑤𝑒𝑎𝑘 are defined because 𝐻–mod𝑤𝑒𝑎𝑘 is tensored over DGCat𝑐𝑜𝑛𝑡 .
94 SAM RASKIN

Remark 8.18.1. Note that the 𝑡-structures attached to different compact open subgroups differ
by shifts by a locally constant function on 𝐻, namely, their relative dimension. For our present
purposes, such differences are irrelevant, so we do not emphasize the choice of 𝐾0 in what follows.
8.19. We will use the following basic observation.
Lemma 8.19.1. The 𝑡-structure just constructed on 𝐷˚ p𝐻q satisfies the following properties.
‚ The 𝑡-structure is right complete.
‚ Any compact F P 𝐷˚ p𝐻q is eventually coconnective.
‚ Compact objects are closed under truncations.
‚ For any F P 𝐷˚ p𝐻q compact, the monoidal operations F ‹ ´ : 𝐷˚ p𝐻q Ñ 𝐷˚ p𝐻q and
´ ‹ F : 𝐷˚ p𝐻q Ñ 𝐷˚ p𝐻q are left 𝑡-exact up to shift.
Proof. The first three claims are evident from the construction. For the last one, note that there is a
compact open subgroup 𝐾 Ď 𝐻 and a coherent 𝐷-module F0 P 𝐷p𝐻{𝐾q such that F “ 𝜋𝐾 ˚,𝑑𝑅
pF0 q.
Then for G P 𝐷 p𝐻q, we have:
˚

𝐾
F ‹ G “ F0 ‹ Av𝐾
˚ pGq
𝐾
where ‹ is convolution on 𝐷˚ p𝐻q, Av𝐾 ˚ indicates (strong) 𝐾-averaging on the left, and ´ ‹ ´ :
𝐷p𝐻{𝐾q b 𝐷p𝐾{𝐻q Ñ 𝐷˚ p𝐻q is the relative convolution. The functor Av𝐾 ˚ is left 𝑡-exact up to
shift: it is right adjoint to a functor that is 𝑡-exact up to shift. Then the claim is evident from the
fact that F0 has support some finite type scheme and from standard cohomological estimates.

Remark 8.19.2. The upshot is that 𝐷˚ p𝐻q almost satisfies the hypotheses of the monoidal DG
category A from Proposition 8.17.5: its unit object is not eventually coconnective, but 𝐷˚ p𝐻q
otherwise satisfies the evident non-unital analogue.
8.20. Main construction. We are now equipped to give the main construction.
To avoid confusion, we let 𝐷˚ p𝐻q𝑔𝑒𝑛 P 𝐻–mod𝑤𝑒𝑎𝑘 denote the object constructed in S8.9, and
we use 𝐷˚ p𝐻q to indicate the underlying DG category Oblv𝑔𝑒𝑛 p𝐷˚ p𝐻q𝑔𝑒𝑛 q.
First, note that there is a canonical monoidal functor IndCoh˚ p𝐻q Ñ 𝐷˚ p𝐻q; indeed, this functor
is constructed in S6.20 with the monoidal structure coming from Remark 6.20.1.
In particular, 𝐷˚ p𝐻q is canonically a pIndCoh˚ p𝐻q, 𝐷˚ p𝐻qq-bimodule. The left IndCoh˚ p𝐻q-
module (i.e., naive weak 𝐻-module) structure here is by construction to one arising from realizing
𝐷˚ p𝐻q as Oblv𝑔𝑒𝑛 p𝐷˚ p𝐻q𝑔𝑒𝑛 q.
Next, observe that the left action action of IndCoh˚ p𝐻q on 𝐷˚ p𝐻q canonically renormalizes in the
sense of S8.17, and the corresponding object (via Proposition 8.17.3) of 𝐻–mod𝑤𝑒𝑎𝑘 is 𝐷˚ p𝐻q𝑔𝑒𝑛 .
Indeed, this is a routine verification by Lemma 8.12.2 and Examples 8.16.2 and 8.16.5. The last
axiom for canonical renormalization (on varying the compact open subgroups) reduces to Lemma
5.20.1 (4).
Therefore, by Proposition 8.17.5 and Lemma 8.19.1 (c.f. Remark 8.19.2) we obtain an a priori
non-unital73 action of 𝐷˚ p𝐻q on 𝐷˚ p𝐻q𝑔𝑒𝑛 P 𝐻–mod𝑤𝑒𝑎𝑘 .
By [Lur3] Proposition [Link], it is a property (not a structure) for 𝐷˚ p𝐻q to act unitally on
𝐷˚ p𝐻q𝑔𝑒𝑛 . We verify this explicitly as follows.

73It is clear that our discussion goes through in a non-unital setting, but this also follows directly from the unital
case by freely adjoining a unit, i.e., applying Proposition 8.17.5 with A “ 𝐷˚ p𝐻q ˆ Vect.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 95

Let 𝐾0 Ď 𝐻 be a fixed compact open subgroup, which we also use to normalize the 𝑡-structure
on 𝐷˚ p𝐻q. As 𝐻–mod𝑤𝑒𝑎𝑘 » H𝐻,𝐾 𝑤
0
–mod, we need to verify that the induced (right) 𝐷˚ p𝐻q-action
on 𝐷˚ p𝐻q𝐾0 ,𝑤 is unital.
»
Note that this is tautologically the case for 𝐷˚ p𝐻q𝐾0 ,𝑤,𝑛𝑎𝑖𝑣𝑒 . As 𝐷˚ p𝐻q𝐾0 ,𝑤,` Ý
Ñ 𝐷˚ p𝐻q𝐾0 ,𝑤,𝑛𝑎𝑖𝑣𝑒,`
˚
by construction, it suffices to show that the unit object 𝛿1 P 𝐷 p𝐻q acts by a left 𝑡-exact functor
on 𝐷˚ p𝐻q𝐾0 ,𝑤 .
Recall that 𝛿1 “ colim𝐾 𝛿𝐾 where the colimit runs over compact open subgroups and the term 𝛿𝐾
indicates the 𝛿 𝐷-module on 𝐻 supported on 𝐾.74 Each 𝛿𝐾 is compact, so (by the construction of
Lemma 8.16.4) acts on 𝐷˚ p𝐻q𝐾0 ,𝑤 by a functor that is left 𝑡-exact up to shift. In fact, these functors
are left 𝑡-exact as is: the induced functor 𝐷˚ p𝐻q𝐾0 ,𝑤,𝑛𝑎𝑖𝑣𝑒 is Oblv Av𝐾 ˚ , which is left 𝑡-exact.
75

By the above description of 𝛿1 , it also acts by a left 𝑡-exact functor, so our earlier remarks we
are done.
This completes the construction of a (right) 𝐷˚ p𝐻q-action on 𝐷˚ p𝐻q𝑔𝑒𝑛 P 𝐻–mod𝑤𝑒𝑎𝑘 , and
therefore (as in S8.4), induces a functor:

𝐻–mod “ 𝐷˚ p𝐻q–mod Ñ 𝐻–mod𝑤𝑒𝑎𝑘 .

8.21. Invariants vs. coinvariants. We now complete the promise from Remark 8.3.1, comparing
weak invariants and coinvariants for expphq in the polarizable case.
Recall the category 𝐷! p𝐻q :“ 𝐷˚ p𝐻q_ “ HomDGCat𝑐𝑜𝑛𝑡 p𝐷˚ p𝐻q, Vectq from [Ras3]. Clearly 𝐷! p𝐻q
is canonically a p𝐷˚ p𝐻q, 𝐷˚ p𝐻qq-bimodule. As 𝐻 is placid (in the sense of loc. cit.), any left (resp.
»
right) 𝐻-invariant dimension theory on 𝐻 defines an equivalence 𝐷˚ p𝐻q Ý Ñ 𝐷! p𝐻q of left (resp.
right) 𝐷˚ p𝐻q-modules.
Note that invariant dimension theories do exist on 𝐻: any choice of congruence subgroup defines
one (see [Ras3] Construction 6.12.6). In particular, 𝐷! p𝐻q is invertible as a bimodule.
Proposition 8.21.1. Let 𝐻 be a polarizable Tate group indscheme. Then there is a canonical
isomorphism of functors:

𝐷! p𝐻q b p´qexpphq,𝑤 » p´ b 𝜒´𝑇 𝑎𝑡𝑒 qexpphq,𝑤 : 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻–mod.

𝐷˚ p𝐻q

Proof. In what follows, we use the symmetric monoidal structure ´ b ´ on 𝐻–mod𝑤𝑒𝑎𝑘 from S7.18.
Let C P 𝐻–mod𝑤𝑒𝑎𝑘 . We claim:
´ ˘¯
Cexpphq,𝑤 “ C b Oblv 𝑠𝑡𝑟Ñ𝑤
` ˚
𝐷 p𝐻q
𝐻,𝑤
´ ¯𝐻,𝑤
Cexpphq,𝑤 “ C b Oblv𝑠𝑡𝑟Ñ𝑤 𝐷! p𝐻q
` ˘

as objects of76 𝐻–mod functorially in C. Indeed, the first identity is immediate, and the second
identity follows similarly the fact that p´q𝐻,𝑤 is DGCat𝑐𝑜𝑛𝑡 -linear for polarizable 𝐻.
Then the claim is straightforward:

74Note that this object of 𝐷 ˚ p𝐻q is in cohomological degree ´ dimp𝐾 {𝐾q if 𝐾 Ď 𝐾 .

0 0
75Here, of course, the averaging is taken on the right, i.e., it does not interact with the weak 𝐾 -invariants.
0
76Here we are using the fact that Oblv𝑠𝑡𝑟Ñ𝑤 p𝐷 ! p𝐻qq, Oblv𝑠𝑡𝑟Ñ𝑤 p𝐷 ˚ p𝐻qq P 𝐷 ˚ p𝐻q–modp𝐻–mod
𝑤𝑒𝑎𝑘 q, where this
structure arises from the bimodule structures on 𝐷! p𝐻q and 𝐷˚ p𝐻q.
96 SAM RASKIN

´ ˘¯ 𝑃 𝑟𝑜𝑝.7.18.2
Cexpphq,𝑤 “ C b Oblv𝑠𝑡𝑟Ñ𝑤 𝐷˚ p𝐻q
`
“
𝐻,𝑤
´ ˘¯𝐻,𝑤
C b 𝜒´𝑇 𝑎𝑡𝑒 b Oblv𝑠𝑡𝑟Ñ𝑤 𝐷˚ p𝐻q
`
“
´ ˘¯𝐻,𝑤
𝐷! p𝐻qb´1 b C b 𝜒´𝑇 𝑎𝑡𝑒 b Oblv𝑠𝑡𝑟Ñ𝑤 𝐷! p𝐻q
`
“
𝐷˚ p𝐻q

𝐷! p𝐻qb´1 b pC b 𝜒´𝑇 𝑎𝑡𝑒 qexpphq,𝑤

𝐷˚ p𝐻q

for 𝐷! p𝐻qb´1 :“ Hom𝐷˚ p𝐻q–mod p𝐷! p𝐻q, 𝐷˚ p𝐻qq the 𝐷˚ p𝐻q-bimodule inverse to 𝐷! p𝐻q. This clearly
gives the identity.

Example 8.21.2. Suppose C “ IndCoh˚ p𝐻q P 𝐻–mod𝑤𝑒𝑎𝑘 (i.e., the evident object that corepresents
Oblv𝑔𝑒𝑛 ). Then Cexpphq,𝑤 » 𝐷˚ p𝐻q P 𝐻–mod. If one takes77 IndCoh! p𝐻q :“ IndCoh˚ p𝐻q b 𝜒´𝑇 𝑎𝑡𝑒 P
𝐻–mod𝑤𝑒𝑎𝑘 , then Proposition 8.21.1 says IndCoh! p𝐻qexpphq,𝑤 “ 𝐷! p𝐻q, as expected.

9. Semi-infinite cohomology
9.1. Construction of semi-infinite cohomology. Let 𝐻 be a Tate group indscheme.
Definition 9.1.1. The absolute semi-infinite cohomology functor:
8
𝐶 2 ph, ´q : Vectexpphq,𝑤 Ñ Vect P 𝐻–mod
is the counit map corresponding to the adjunction constructed in S8.
The goal for this section is to show that for 𝐻 formally smooth, this functor identifies (in a
suitable sense) with the classical functor of semi-infinite cohomology for Tate Lie algebras.
8
Remark 9.1.2. As in indicated in the notation above, 𝐶 2 ph, ´q is strongly 𝐻-equivariant. This is
a non-obvious (if widely anticipated) property from the traditional construction of semi-infinite
cohomology via Clifford algebras.
Remark 9.1.3. The above functor is defined (and is strongly 𝐻-equivariant) for any Tate group
indscheme 𝐻. However, for the purposes of relating this functor to classical constructions, we may
assume 𝐻 is polarizable; indeed, replacing 𝐻 by its formal completion along any compact open
8
subgroup manifestly does not change 𝐶 2 ph, ´q as a morphism in DGCat𝑐𝑜𝑛𝑡 . Therefore, in the
analysis of this section, 𝐻 is frequently taken to be polarizable.
Remark 9.1.4. The above construction (hence our comparison theorem) only applies for those Tate
Lie algebras h P ProVect♡ arising as the Lie algebra of some formally smooth Tate group indscheme.
Equivalently, there must exist k Ď h a compact open Lie subalgebra arising that arises as the Lie
algebra of some affine group scheme. Certainly this is the case whenever h has a pro-nilpotent
compact open subalgebra, which covers all examples of interest.
We anticipate (as indicated in the notation) that there is a theory of weak actions for the “formal
group” expphq for a general Tate Lie algebra h. The argument given below for the comparison
theorem should then apply as is in that setup. However, as the applications we have in mind do
not require such a theory, we do not develop one in this text.
77Note that IndCoh! p𝐻q is also the internal Hom in the symmetric monoidal category 𝐻–mod from IndCoh˚ p𝐻q
𝑤𝑒𝑎𝑘
to the trivial object Vect.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 97

Remark 9.1.5. We thank Gurbir Dhillon for insisting that we include this material in the present
text and for helpful discussions related to it.
9.2. Throughout this section, 𝐻 denotes a Tate group indscheme.
We maintain the conventions of S7.3: all quotients (including classifying stacks) are Zariski sheafi-
fied.

9.3. Central extensions. We begin by discussing some general constructions relating to central
extensions.

9.4. We begin with the following construction.

First, note that there is a canonical action of BG𝑚 on Vect, i.e., an action of QCohpBG𝑚 q equipped
with the convolution monoidal structure on Vect. Indeed, this monoidal category is canonically
equivalent to Z-graded vector spaces with degree-wise. Our monoidal functor QCohpBG𝑚 q Ñ Vect
takes the p´1q-st degree component.78
Extend this construction to an action of BG𝑚 ˆ Z by having the generator 1 P Z act on Vect as
»
p´qr1s : Vect ÝÑ Vect.
Now for any 𝐻 a Tate group indscheme equipped with a homomorphism:

p𝜀, 𝛿q : 𝐻 Ñ BG𝑚 ˆ Z
of groups, we obtain an action of 𝐻 on Vect by restriction along the monoidal pushforward functor:79

IndCoh˚ p𝐻q Ñ IndCoh˚ pBG𝑚 ˆ Zq “ QCohpBG𝑚 ˆ Zq.

Remark 9.4.1. Under the above construction, any ℎ P 𝐻p𝑘q defines a skyscraper sheaf in IndCoh˚ p𝐻q,
so by extension, an automorphism of Vect. By construction, this automorphism is 𝜀pℎq b ´r𝛿pℎqs,
where 𝜀pℎq is the 𝑘-line defined by ℎ and 𝜀.
Proposition 9.4.2. The above construction gives an equivalence of groupoids:

HomGp p𝐻, BG𝑚 ˆ Zq Ñ HomAlgpDGCat𝑐𝑜𝑛𝑡 q pIndCoh˚ p𝐻q, Vectq. (9.4.1)

Proof.
Step 1. First, it is convenient to dualize (in the sense of [Gai4]). For 𝑆 a reasonable indscheme,
we let IndCoh! p𝑆q P DGCat𝑐𝑜𝑛𝑡 denote the dual to IndCoh˚ p𝑆q (which exists because IndCoh˚ p𝑆q is
compactly generated). Note that this construction is covariant in 𝑆; we denote pullback along a
map 𝑓 : 𝑆 Ñ 𝑇 by 𝑓 ! : IndCoh! p𝑇 q Ñ IndCoh! p𝑆q.80 In particular, there is a canonical object 𝜔𝑆 P
IndCoh! p𝑆q, the !-pullback of 𝑘 P Vect “ IndCoh! pSpecp𝑘qq along the structure map 𝑆 Ñ Specp𝑘q.
There is a standard natural transformation Υ𝑆 : QCohp´q Ñ IndCoh! p´q such that for any
𝑆 P IndSch𝑟𝑒𝑎𝑠 , Υ𝑆 pO𝑆 q “ 𝜔𝑆 . Indeed, for 𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 , Υ𝑆 is by definition dual to Ψ𝑆 :
IndCoh˚ p𝑆q Ñ QCohp𝑆q (using the standard self-duality of QCohp´q on qcqs schemes). In general,
the construction is obtained by right Kan extension from this one.
78The sign here makes normalizations for later constructions more convenient: see Remark 9.4.1.
79This functor is defined by the formalism of S6 because 𝐻 and BG ˆ Z are weakly renormalizable prestacks and
𝑚
this morphism is reasonable indschematic.
80This construction should not be confused with the one studied in the IndCoh˚ -setting of S6 for a proper (or ind-
proper) morphism. Because we only use this construction in the proof of the present proposition, we abuse notation
by using the same notation to mean different things (in somewhat different contexts).
98 SAM RASKIN

Note that Υ𝑆 is fully-faithful for any 𝑆 P IndSch𝑟𝑒𝑎𝑠 . Indeed, by construction, this reduces to
𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 , and in that case it follows because the dual functor Ψ𝑆 admits a fully-faithful left
adjoint.
Step 2. Now suppose that 𝑆 is strict in the sense of S6.35. By functoriality, IndCoh! p𝑆q is symmetric
!
monoidal with tensor product ´ b ´ with unit object 𝜔𝑆 (c.f. [Gai5] S5.6).
!
We remark that the (symmetric) monoidal category pIndCoh! p𝑆q, bq acts canonically on IndCoh˚ p𝑆q
by duality.
!
In what follows, we say L P IndCoh! p𝑆q is invertible if L b ´ : IndCoh! p𝑆q Ñ IndCoh! p𝑆q is an
equivalence.
Step 3. Suppose that 𝑆 P ą´8Sch𝑞𝑐𝑞𝑠 . We suppose that 𝑆 is strict and L P IndCoh! p𝑆q is invertible.
We claim that:
(1) L lies in the essential image of Υ𝑆 .
(2) The object L P QCohp𝑆q with Υ𝑆 pLq “ L (which is well-defined by the above) is invertible
in QCohp𝑆q and therefore corresponds to a Z-graded81 line bundle on 𝑆 (the Z-grading
being locally constant).
First, note that L P IndCoh! p𝑆q is compact: this follows from invertibility and compactness of 𝜔𝑆
(we emphasize that 𝑆 is eventually coconnective). We let DL P Cohp𝑆q denote the corresponding
object under the equivalence:

pIndCoh! p𝑆q𝑐 q𝑜𝑝 » IndCoh˚ p𝑆q𝑐 “ Cohp𝑆q.

For this, let 𝑠 : Specp𝐾q Ñ 𝑆 be a map with 𝐾 a field. Clearly 𝑠! pLq is invertible in IndCoh! pSpecp𝐾qq “
𝐾–mod. In particular, it corresponds to a graded line. We then obtain:

𝑠! pLq “ x𝐾, 𝑠! pLqy “ x𝑠IndCoh

˚ p𝐾q, Ly “ HomIndCoh˚ p𝑆q pDL, 𝑠IndCoh
˚ p𝐾qq “ 𝑠˚ pDLq_
where the brackets indicate pairings between evident dual categories, the upper-* fiber is the usual
quasi-coherent fiber of DL P Cohp𝑆q Ď QCohp𝑆q, and the dual indicates the dual as a 𝐾-vector
space.
In particular, the ˚-fibers of DL at field-valued points are concentrated in some single cohomo-
logical degree and 1-dimensional there. Now it follows from Nakayama’s lemma that DL is perfect
and a graded line bundle.
Applying duality again, this translates to saying that L P Υ𝑆 pPerfp𝑆qq, and moreover, is Υ𝑆 of
a graded line bundle.
Step 4. Next, we observe that the above immediately generalizes to the case where 𝑆 P IndSch𝑟𝑒𝑎𝑠
can be expressed as a filtered colimit of strict schemes 𝑆𝑖 P ą´8Sch𝑞𝑐𝑞𝑠 under almost finitely
presented closed embeddings. Indeed, this case immediately reduces to the schematic case considered
above.
In particular, this applies for 𝑆 “ 𝐻 our Tate group indscheme, as observed in Remark 7.2.5.
Step 5. We now complete the argument.
By duality and strictness of 𝐻, a monoidal functor IndCoh˚ p𝐻q Ñ Vect is equivalent to a
»
comonoidal functor Vect Ñ IndCoh! p𝐻q, i.e., an object L P IndCoh! p𝐻q with isomorphisms 𝑚! pLq Ý
Ñ
L b L, 𝑒! pLq “ 𝑘 equipped with higher homotopical compatibilities (for 𝑚 : 𝐻 ˆ 𝐻 Ñ 𝐻 the mul-
tiplication and 𝑒 : Specp𝑘q Ñ 𝐻 the unit).
81We normalize this identification by having the suspension p´qr1s correspond to increasing the grading by 1.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 99

Observe that L is invertible in IndCoh! p𝐻q: its inverse is the pullback of L along the inversion
»
map 𝐻 ÝÑ 𝐻.
Therefore, L defines a Z-graded line bundle on 𝐻, or equivalently, a map 𝐻 Ñ BG𝑚 ˆ Z. The
comonoidal structure above is equivalent to making the map into a map of group prestacks. It is
immediate to verify that this equivalence is the inverse to the functor (9.4.1).

9.5. The Tate canonical extension. We now construct a canonical central extension:

1 Ñ G𝑚 Ñ 𝐻𝑇 𝑎𝑡𝑒 Ñ 𝐻 Ñ 1
of any polarizable Tate group indscheme 𝐻.
Take 𝜒𝑇 𝑎𝑡𝑒 P 𝐻–mod𝑤𝑒𝑎𝑘 . Recall that 𝐻–mod𝑤𝑒𝑎𝑘 is naturally symmetric monoidal and that
𝜒𝑇 𝑎𝑡𝑒 is invertible for this monoidal structure.
Recall from Proposition 7.19.1 that Oblv𝑔𝑒𝑛 p𝜒𝑇 𝑎𝑡𝑒 q P DGCat𝑐𝑜𝑛𝑡 is a trivial gerbe, i.e., this DG
category is non-canonically isomorphic to Vect (the identification depends on a choice of compact
open subgroup 𝐾 of 𝐻). In particular, we have a canonical isomorphism EndDGCat𝑐𝑜𝑛𝑡 pOblv𝑔𝑒𝑛 p𝜒𝑇 𝑎𝑡𝑒 qq “
Vect.
As 𝐻 acts naively on Oblv𝑔𝑒𝑛 p𝜒𝑇 𝑎𝑡𝑒 q, this defines a canonical homomorphism IndCoh˚ p𝐻q Ñ
Vect. By Proposition 9.4.2, we obtain a homomorphism map:

p𝜀𝑇 𝑎𝑡𝑒 , 𝛿𝑇 𝑎𝑡𝑒 q : 𝐻 Ñ BG𝑚 ˆ Z.

By definition, 𝐻𝑇 𝑎𝑡𝑒 is the central extension defined by the homomorphism 𝜀𝑇 𝑎𝑡𝑒 .
Remark 9.5.1. Define an object82 Vect𝜒𝑇 𝑎𝑡𝑒 P 𝐻–mod𝑤𝑒𝑎𝑘 as:

Vect𝜒𝑇 𝑎𝑡𝑒 :“ 𝜒𝑇 𝑎𝑡𝑒 b trivpOblv𝑔𝑒𝑛 p𝜒b´1

𝑇 𝑎𝑡𝑒 qq.
Note that Vect𝜒𝑇 𝑎𝑡𝑒 maps canonically under Oblv𝑔𝑒𝑛 to Vect, and the induced naive 𝐻-action is
the one constructed above. By Proposition 7.19.1, any choice of compact open subgroup 𝐾 Ď 𝐻
induces an isomorphism 𝜒𝑇 𝑎𝑡𝑒 » Vect𝜒𝑇 𝑎𝑡𝑒 P 𝐻–mod𝑤𝑒𝑎𝑘 .
9.6. We now discuss basic properties of 𝐻𝑇 𝑎𝑡𝑒 .
Proposition-Construction 9.6.1. For any compact open subgroup 𝐾 of 𝐻, there is a canonical
splitting of 𝐻𝑇 𝑎𝑡𝑒 over 𝐾.
Proof. Immediate from Proposition 7.19.1.

Corollary 9.6.2. 𝐻𝑇 𝑎𝑡𝑒 is a Tate group indscheme.
Warning 9.6.3. In general, for 𝐾1 Ď 𝐾2 Ď 𝐻 compact open subgroups, the canonical splitting
for 𝐾2 may not restrict to the canonical splitting for 𝐾1 (although this is automatic if 𝐾1 is
pro-unipotent).
Proposition-Construction 9.6.4. Suppose 𝐻 has the property that 𝛿𝑇 𝑎𝑡𝑒 is identically 0.
Let Res : 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻𝑇 𝑎𝑡𝑒 –mod𝑤𝑒𝑎𝑘 denote the functor of restriction along 𝐻𝑇 𝑎𝑡𝑒 Ñ 𝐻, and
let 𝜒𝑇 𝑎𝑡𝑒 P 𝐻–mod𝑤𝑒𝑎𝑘 denote the modular character for 𝐻.
Then Resp𝜒𝑇 𝑎𝑡𝑒 q is canonically trivialized, i.e., there is a canonical isomorphism:

Resp𝜒𝑇 𝑎𝑡𝑒 q » triv Oblv𝑔𝑒𝑛 pResp𝜒𝑇 𝑎𝑡𝑒 q P 𝐻𝑇 𝑎𝑡𝑒 –mod𝑤𝑒𝑎𝑘 .

82By analogy with usual representations, Vect
𝜒𝑇 𝑎𝑡𝑒 is the character defined by the “1-dimensional” representation
𝜒𝑇 𝑎𝑡𝑒 .
100 SAM RASKIN

Proof. In the notation of Remark 9.5.1, it suffices to construct an isomorphism RespVect𝜒𝑇 𝑎𝑡𝑒 q »
Vect P 𝐻𝑇 𝑎𝑡𝑒 –mod𝑤𝑒𝑎𝑘 .
Note that by standard cohomological estimates, the underlying naive action of 𝐻 on Oblv𝑔𝑒𝑛 pVect𝜒𝑇 𝑎𝑡𝑒 q
canonically renormalizes in the sense of Proposition 8.17.3, and that Vect𝜒𝑇 𝑎𝑡𝑒 is obtained by this
canonical renormalization procedure. The same applies for 𝐻𝑇 𝑎𝑡𝑒 in place of 𝐻. Therefore, it suffices
to give the construction on underlying naive categories.
But here the result follows from Proposition 9.4.2 and the evident trivialization of the composite
homomorphism:

𝐻𝑇 𝑎𝑡𝑒 Ñ 𝐻 Ñ BG𝑚 ˆ Z.

Remark 9.6.5. By Theorem 5.10.1 for of G𝑚 , it is easy to see that Resp𝜒𝑇 𝑎𝑡𝑒 q P 𝐻𝑇 𝑎𝑡𝑒 –mod𝑤𝑒𝑎𝑘 is the
modular character for 𝐻𝑇 𝑎𝑡𝑒 . Therefore, the modular character of 𝐻𝑇 𝑎𝑡𝑒 is canonically trivialized.
9.7. We now wish to formulate in a precise way the following idea: for C P 𝐻–mod𝑤𝑒𝑎𝑘 , pC b
Vect𝜒𝑇 𝑎𝑡𝑒 q𝐻,𝑤 is the subcategory of C𝐻𝑇 𝑎𝑡𝑒 ,𝑤 of objects on which G𝑚 acts by homotheties.
There is a somewhat more satisfying formulation in the naive setting than the genuine one, so
we separate the two cases.

9.8. Let 𝑘p1q P ReppG𝑚 q denote the standard representation, and for 𝑛 P Z, let 𝑘p𝑛q denote its
𝑘ÞÑ𝑘p𝑛q
𝑛th tensor power. We let Vectp𝑛q Ď ReppG𝑚 q denote the image of Vect ÝÝÝÝÝÑ ReppG𝑚 q, i.e., the
category of graded vector spaces of pure degree 𝑛.
Suppose C P 𝐻–mod𝑤𝑒𝑎𝑘 and restrict C to 𝐻𝑇 𝑎𝑡𝑒 –mod𝑤𝑒𝑎𝑘 ; we now omit Res from the notation.
As the central G𝑚 Ď 𝐻𝑇 𝑎𝑡𝑒 acts trivially on C, there is a forgetful functor:

C𝐻𝑇 𝑎𝑡𝑒 ,𝑤 Ñ CG𝑚 ,𝑤 “ ReppG𝑚 q b Oblv𝑔𝑒𝑛 pCq.

We remark that this functor factors through C𝐻𝑇 𝑎𝑡𝑒 ,𝑤,𝑛𝑎𝑖𝑣𝑒 , and that the corresponding functor
C𝐻𝑇 𝑎𝑡𝑒 ,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ ReppG𝑚 q b Oblv𝑔𝑒𝑛 pCq is conservative.
𝐻𝑇 𝑎𝑡𝑒 ,𝑤,𝑛𝑎𝑖𝑣𝑒
For 𝑛 P Z, define Cp𝑛q Ď C𝐻𝑇 𝑎𝑡𝑒 ,𝑤,𝑛𝑎𝑖𝑣𝑒 as the full subcategory:

C𝐻𝑇 𝑎𝑡𝑒 ,𝑤,𝑛𝑎𝑖𝑣𝑒 ˆ Vectp𝑛q b Oblv𝑔𝑒𝑛 pCq.

ReppG𝑚 qbOblv𝑔𝑒𝑛 pCq

In words: this is the full subcategory of 𝐻𝑇 𝑎𝑡𝑒 -equivariant objects where the central G𝑚 acts by
the 𝑛th power of its canonical character, this notion being defined because G𝑚 acts trivially on C.
Note that:

Ñ C𝐻 𝑇 𝑎𝑡𝑒 ,𝑤
»
C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 Ý p0q (9.8.1)
by semi-simplicity of ReppG𝑚 q.
Proposition 9.8.1. Let 𝐻 be a Tate group indscheme with 𝛿𝑇 𝑎𝑡𝑒 identically 0.
Then for any C P 𝐻–mod𝑤𝑒𝑎𝑘 , the canonical functor:

𝑃 𝑟𝑜𝑝.´𝐶𝑜𝑛𝑠𝑡.9.6.4
pC b Vect𝜒𝑇 𝑎𝑡𝑒 q𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ pC b Vect𝜒𝑇 𝑎𝑡𝑒 q𝐻𝑇 𝑎𝑡𝑒 ,𝑤,𝑛𝑎𝑖𝑣𝑒 » C𝐻𝑇 𝑎𝑡𝑒 ,𝑤,𝑛𝑎𝑖𝑣𝑒
is fully-faithful with essential image C𝐻 𝑇 𝑎𝑡𝑒 ,𝑤,𝑛𝑎𝑖𝑣𝑒
p1q .
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 101

»
Proof. Consider the isomorphism Vect Ý Ñ Vect𝜒𝑇 𝑎𝑡𝑒 P 𝐻𝑇 𝑎𝑡𝑒 –mod𝑤𝑒𝑎𝑘 from Proposition-Construction
9.6.4 restricted to G𝑚 . Note that Vect𝜒𝑇 𝑎𝑡𝑒 |G𝑚 is canonically isomorphic to Vect with its trivial ac-
tion (as it was obtained by restriction from 𝐻). Therefore, this isomorphism is an equivalence:
»
Ñ Vect𝜒𝑇 𝑎𝑡𝑒 |G𝑚 “ Vect P G𝑚 –mod𝑤𝑒𝑎𝑘 .
Vect Ý
Therefore, this isomorphism amounts to specifying an invertible object of EndG𝑚 –mod𝑤𝑒𝑎𝑘 pVectq “
ReppG𝑚 q. It follows from the construction that this object is 𝑘p1q.
We now obtain the result from (9.8.1).

9.9. We now explain how to adapt the above to the setting of genuine actions.
Note that by Theorem 5.10.1 (for G𝑚 ), Res : 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻𝑇 𝑎𝑡𝑒 –mod𝑤𝑒𝑎𝑘 admits a right (and
left) DGCat𝑐𝑜𝑛𝑡 -linear adjoint commuting with colimits. We abuse notation somewhat in denoting
this functor by p´qG𝑚 ,𝑤 .
For any 𝑛 P Z, there is an adjunction map:

G𝑚 ,𝑤 𝑃 𝑟𝑜𝑝.´𝐶𝑜𝑛𝑠𝑡.9.6.4
Vectb𝑛 b𝑛
𝜒𝑇 𝑎𝑡𝑒 Ñ pVect𝜒𝑇 𝑎𝑡𝑒 q » VectG𝑚 ,𝑤 P 𝐻–mod𝑤𝑒𝑎𝑘 .
The induced map:
G𝑚 ,𝑤
‘ Vectb𝑛
𝜒𝑇 𝑎𝑡𝑒 Vect P 𝐻–mod𝑤𝑒𝑎𝑘
𝑛PZ
is an isomorphism.83
On tensoring, for any C P 𝐻–mod𝑤𝑒𝑎𝑘 , we obtain an isomorphism:
»
‘ pC b Vectb𝑛 Ñ RespCqG𝑚 ,𝑤 P 𝐻–mod𝑤𝑒𝑎𝑘 .
𝜒𝑇 𝑎𝑡𝑒 q Ý
𝑛PZ
Passing to invariants, we obtain:
»
‘ pC b Vectb𝑛
𝜒𝑇 𝑎𝑡𝑒 q
𝐻,𝑤
Ñ C𝐻𝑇 𝑎𝑡𝑒 ,𝑤 .
Ý
𝑛PZ
We record these observations as the following analogue of Proposition 9.8.1.
Proposition 9.9.1. For C P 𝐻–mod𝑤𝑒𝑎𝑘 , C𝐻𝑇 𝑎𝑡𝑒 ,𝑤 is canonically Z-graded with pC b Vect𝜒𝑇 𝑎𝑡𝑒 q𝐻,𝑤
as its degree 1 component.
9.10. We remark briefly on another interpretation of the above results. We allow ourselves to be
slightly imprecise here in speaking about BG𝑚 -actions on categories on equal footing with genuine
actions of Tate group indschemes, although this is not formally allowed in the theory developed in
S7 (though one could suitably extend the theory without difficulty).
By fiat, genuine BG𝑚 -actions on C P DGCat𝑐𝑜𝑛𝑡 are the same as naive ones, i.e., IndCoh˚ pBG𝑚 q-
actions. As IndCoh˚ pBG𝑚 q “ QCohpZq with convolution on the left corresponding to tensor products
on the right, such a datum is equivalent to a Z-grading C “ ‘𝑛PZ Cp𝑛q . Here Cp0q “ CBG𝑚 ,𝑤 .
Note that the BG𝑚 on Vect constructed in S9.4 has Vect “ Vectp´1q , i.e., it is Vect graded in
pure degree ´1. For simplicity, we denote this object by Vectp´1q P BG𝑚 –mod𝑤𝑒𝑎𝑘 .
We have a fiber sequence of groups:
𝜀
𝑇 𝑎𝑡𝑒
𝐻𝑇 𝑎𝑡𝑒 Ñ 𝐻 ÝÝÝÑ BG𝑚
83The ‘ denotes the coproduct in 𝐻–mod
𝑤𝑒𝑎𝑘 . Note that (even infinite) coproducts in 𝐻–mod𝑤𝑒𝑎𝑘 coincide with
products as the same is true in DGCat𝑐𝑜𝑛𝑡 .
102 SAM RASKIN

as 𝛿𝑇 𝑎𝑡𝑒 is assumed to be 0. By design, the pullback of Vectp´1q along 𝜀𝑇 𝑎𝑡𝑒 is Vect𝜒𝑇 𝑎𝑡𝑒 .
Therefore, for C with a genuine (or naive) action of 𝐻, BG𝑚 acts on C𝐻𝑇 𝑎𝑡𝑒 ,𝑤 , i.e., we obtain a
𝐻𝑇 𝑎𝑡𝑒 ,𝑤
grading C𝐻𝑇 𝑎𝑡𝑒 ,𝑤 “ ‘𝑛PZ Cp𝑛q . We then have:
´` ¯BG𝑚 ,𝑤
pC b Vect𝜒𝑇 𝑎𝑡𝑒 q𝐻,𝑤 “ ‘𝑛PZ C𝐻 𝑇 𝑎𝑡𝑒 ,𝑤
“ C𝐻 𝑇 𝑎𝑡𝑒 ,𝑤
˘
p𝑛q b Vect p´1q p1q
as desired.
9.11. Representations of Tate group indschemes. The following result describes the major
structures of Repp𝐻q.
Proposition 9.11.1. Let 𝐻 be polarizable.
(1) Repp𝐻q is compactly generated.
(2) Suppose 𝐻 is a classical indscheme. There is a unique 𝑡-structure on Repp𝐻q such that for
any compact open subgroup 𝐾 Ď 𝐻, the (conservative) forgetful functor Repp𝐻q Ñ Repp𝐾q
is 𝑡-exact.
(3) Suppose that 𝐻 is of Harish-Chandra type (c.f. Example 7.17.3) and formally smooth. Then
Repp𝐻q` is the bounded below derived category of Repp𝐻q♡ .
Proof. Let 𝐾 Ď 𝐻 be a polarization.
For any C P 𝐻–mod𝑤𝑒𝑎𝑘 , the forgetful functor C𝐻,𝑤 Ñ C𝐾,𝑤 is conservative and admits a con-
! : C
tinuous left adjoint Av𝑤 𝐾,𝑤 Ñ C𝐻,𝑤 . Indeed, the former property is true for any compact

open subgroup while the latter is true by ind-properness of 𝐻{𝐾. Therefore, if C𝐾,𝑤 is compactly
generated, then C𝐻,𝑤 is compactly generated. Applying this for C “ Vect gives (1).
Next, in the setting of (2), observe that it suffices to show Oblv Av𝑤! : Repp𝐾q Ñ Repp𝐾q is
right 𝑡-exact. Indeed, we are reduced to showing this by the monadicity of Oblv shown above. (We
remark that 𝑡-exactness of the restriction functor to some compact open subgroup clearly implies
the same for any compact open subgroup.)
Because 𝐻 is classical, the same is true of 𝐻{𝐾, i.e., we can write 𝐻{𝐾 “ colim𝑖 𝑆𝑖 a filtered
colimit of classical proper 𝑘-schemes.
Suppose 𝑉 P Repp𝐾q♡ is finite dimensional. It suffices to show that for such 𝑉 , Oblv Av𝑤! p𝑉 q P
Repp𝐾qď0 , or equivalently, that the underlying vector space of this 𝐾-representation is in Vectď0 .
Let E𝑉 P QCohp𝐻{𝐾q be the corresponding (naively 𝐻-equivariant) vector bundle. Then:

Oblv Av𝑤
! p𝑉 q “ Γ
IndCoh
p𝐻{𝐾, E𝑉 b 𝜔𝐻{𝐾 q “ colim ΓIndCoh p𝑆𝑖 , E𝑉 |𝑆𝑖 b 𝜔𝑆𝑖 q “ Γp𝑆𝑖 , E_ _
𝑉 |𝑆𝑖 q .
𝑖

We have Γp𝑆𝑖 , E_ ě0
P Vect as 𝑆𝑖 is classical, so we obtain the claim by dualizing.
𝑉 |𝑆𝑖 q
Finally, in the setting of (3), we suppose 𝐾 is chosen so 𝐻 is formally complete along it; note
that the above argument shows that Oblv Av𝑤 ! is 𝑡-exact by formal smoothness of 𝐻{𝐾.
`
In the case 𝐻 “ 𝐾, the fact that Repp𝐾q is the bounded below derived category of its heart
♡
is standard. Any object Av𝑤 ˚ p𝑉 q for 𝑉 P Vect is injective in Repp𝐾q♡ . Moreover, any object
of Repp𝐾q admits an injective resolution by such objects. Finally, for 𝑊 P Repp𝐾q♡ and 𝑉 as
♡

above, HomRepp𝐾q p𝑊, Av𝑤 ˚ p𝑉 qq “ HomVect p𝑊, 𝑉 q is concentrated in cohomological degree 0. These
observations imply the claim.
In general, the argument follows by Lemma 9.11.2 below (or see a variant of this argument in
[Ras6] Lemma A.18.1).

We used the following result above.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 103

Lemma 9.11.2. Suppose C, D P DGCat are equipped with 𝑡-structures. compatible with filtered
colimits. Let 𝐺 : C Ñ D P DGCat𝑐𝑜𝑛𝑡 be a conservative, 𝑡-exact functor with a 𝑡-exact left adjoint
𝐹.
Suppose D` is the bounded below derived category of D♡ . Then C` is the bounded below derived
category of C♡ .

Proof. Let 𝐼 P C♡ be an injective object. We need to show that for any F P C♡ , HomC pF, 𝐼q P Vect♡ .
Clearly this complex is in degrees ě 0. We will show by induction on 𝑖 ą 0 that 𝐻 𝑖 HomC pF, 𝐼q “
Ext𝑖C pF, 𝐼q vanishes for all F. For 𝑖 “ 1, we have Ext1C pF, 𝐼q “ Ext1C♡ pF, 𝐼q “ 0, giving the base
case. Suppose the result is true for 𝑖 ě 1, and we will deduce it for 𝑖 ` 1.
First, note that the counit map 𝐹 𝐺pFq Ñ F P C♡ is an epimorphism. Indeed, we can check this
after applying the conservative, 𝑡-exact functor 𝐺, and then the map splits.
Let F0 be the kernel of this counit. We obtain an exact sequence:

Ext𝑖C pF0 , 𝐼q Ñ Ext𝑖`1 𝑖`1 𝑖`1

C pF, 𝐼q Ñ ExtC p𝐹 𝐺pFq, 𝐼q “ ExtD p𝐺pFq, 𝐺p𝐼qq.

The first term vanishes by induction. The last term vanishes because 𝐺 : C♡ Ñ D♡ admits a 𝑡-exact
left adjoint so preserves injectives, and by assumption on D. This gives the claim.


Combining Propositions 9.11.1, 9.8.1 and 9.9.1, we obtain:

Corollary 9.11.3. For polarizable 𝐻, the categories Rep˘𝑇 𝑎𝑡𝑒 p𝐻q are compactly generated. If 𝐻
is classical, there is a unique compactly generated 𝑡-structure on Rep˘𝑇 𝑎𝑡𝑒 p𝐻q for which the for-
getful functor to Repp𝐾q is 𝑡-exact for any compact open subgroup 𝐾 Ď 𝐻 (using Proposition
7.19.1). The category Rep𝑇 𝑎𝑡𝑒 p𝐻q` (resp. Rep´𝑇 𝑎𝑡𝑒 p𝐻q` ) maps isomorphically onto the subcate-
gory of Repp𝐻𝑇 𝑎𝑡𝑒 q` consisting of objects on which the central G𝑚 acts by (direct sums of shifts of )
its standard representation (resp. the inverse to the standard representation). If 𝐻 is additionally of
Harish-Chandra type, then Rep˘𝑇 𝑎𝑡𝑒 p𝐻q` is the bounded below derived category of Rep˘𝑇 𝑎𝑡𝑒 p𝐻q♡ .

9.12. Passage to Lie algebras. Let 𝐻 be a Tate group indscheme of Harish-Chandra type. We
assume 𝐻 is polarizable in what follows (although most of the discussion generalizes to the non-
polarizable case by replacing 𝐻 with its formal completion along some compact open subgroup).
Define h–mod :“ Vectexpphq,𝑤 . Similarly, define h𝑇 𝑎𝑡𝑒 –mod (resp. h´𝑇 𝑎𝑡𝑒 –mod) as pVect𝜒𝑇 𝑎𝑡𝑒 qexpphq,𝑤
(resp. pVect𝜒´𝑇 𝑎𝑡𝑒 qexpphq,𝑤 ).84

Definition 9.12.1. For 𝐾 Ď 𝐻 a fixed compact open subgroup, the relative semi-infinite cohomology
functor:

8
𝐶 2 ph, k; ´q : h´𝑇 𝑎𝑡𝑒 –mod Ñ Vect P 𝐻–mod
8
corresponds to 𝐶 2 ph, ´q : Vectexpphq,𝑤 Ñ Vect under the equivalence:

84The notation is potentially confusing. There is a central extension h

𝑇 𝑎𝑡𝑒 around (the Lie algebra of 𝐻𝑇 𝑎𝑡𝑒 ), and
we are in effect considering modules over it on which the central element 1 P 𝑘 Ď h𝑇 𝑎𝑡𝑒 acts by the identity (or minus
the identity), of course in a suitable derived categorical sense.
This is a somewhat standard abuse, and we hope that it does not cause confusion. To be clear: we will never
consider all modules over the Tate Lie algebra h𝑇 𝑎𝑡𝑒 .
104 SAM RASKIN

𝑃 𝑟𝑜𝑝.8.21.1
h´𝑇 𝑎𝑡𝑒 –mod :“ pVect𝜒´𝑇 𝑎𝑡𝑒 qexpphq,𝑤 »
!
𝐷 p𝐻q b pVect𝜒´𝑇 𝑎𝑡𝑒 b 𝜒𝑇 𝑎𝑡𝑒 qexpphq,𝑤 »
𝐷˚ p𝐻q
Vectexpphq,𝑤
using the choice of 𝐾 both to identify Vect𝜒𝑇 𝑎𝑡𝑒 with 𝜒𝑇 𝑎𝑡𝑒 via Proposition 7.19.1 and to identify
𝐷! p𝐻q with 𝐷˚ p𝐻q via [Ras3] Construction 6.12.6.
Note that 𝐻 acts strongly on h–mod by the construction of S8. For 𝐾 Ď 𝐻 compact open,
we have h–mod𝐾 “ Repp𝐻𝐾 ^ q by construction. In particular, we have h–mod “ colim h–mod𝐾 .
𝐾
Moreover, for 𝐻 formally smooth, h–mod𝐾 has a canonical 𝑡-structure by Proposition 9.11.1.

9.13. We now suppose that 𝐻 is formally smooth. In this case, its Lie algebra h is naturally a
Tate Lie algebra in the sense of Example 4.3.8, and we have two possibly conflicting definitions of
h–mod. However, we claim that they do not in fact conflict.
Below, we understand h–mod in the sense defined immediately above, i.e., as Vectexpphq,𝑤 .
Lemma 9.13.1. (1) For each pair 𝐾1 Ď 𝐾2 Ď 𝐻 of compact open subgroups, the (conservative)
restriction functor:

h–mod𝐾2 Ñ h–mod𝐾1
is 𝑡-exact. In particular, the colimit h–mod over all such compact open subgroups admits a
canonical 𝑡-structure.
(2) The forgetful functor h–mod :“ Vectexpphq,𝑤 Ñ Vect is 𝑡-exact and conservative on eventually
Ñ
coconnective subcategories. The corresponding b-algebra (as defined by Proposition 3.7.1) is
the completed universal enveloping algebra of the Tate Lie algebra h. Moreover, the compact
generators of h–mod correspond to the renormalization datum specified in Example 4.3.8.
Proof. The 𝑡-exactness of the various restriction functors is clear from Proposition 9.11.1.
Moreover, for 𝐾1 Ď 𝐾2 Ď 𝐻 compact open subgroups and for 𝑉 P h–mod𝐾1 ,ě0 , we claim that the
adjunction map Oblv Av𝐾 ˚
1 Ñ𝐾2
p𝑉 q Ñ 𝑉 induces a monomorphism in h–mod𝐾1 ,♡ upon applying
𝐻 . Indeed, we can test this after applying the (conservative, 𝑡-exact) forgetful functor to k2 –mod𝐾
0
1 ,
where it is evident.
It follows that for any 𝑉 P h–modě0 and 𝐾 a congruence subgroup, the adjunction map
Oblv Av𝐾 0 𝐾
˚ p𝑉 q Ñ 𝑉 gives a monomorphism on 𝐻 . As 𝑉 “ colim𝐾 Oblv Av˚ p𝑉 q, this implies
that Oblv : h–mod Ñ Vect is conservative on eventually coconnective subcategories.
Now define an object:

P :“ lim indhk p𝑘q P Proph–mod♡ q Ď Proph–mod` q

𝐾
where the notation is understood as follows. First, the limit is formed in the pro-category, and is
indexed by compact open subgroups 𝐾 Ď 𝐻. Then 𝑘 P k–mod denotes the trivial representation
and indhk : k–mod Ñ h–mod𝐾 is the left adjoint to the forgetful functor.
Then P pro-corepresents the forgetful functor h–mod` Ñ Vect` . Moreover, under the forgetful
functor, P maps an object of PropVect♡ q. By Proposition 3.7.1, h–mod` is the bounded below
derived category of its heart, and this heart is the category of discrete modules for OblvpPq with
Ñ
respect to its natural b-algebra structure.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 105

Ñ
One can identify OblvpPq with its b-algebra structure as follows. Let h𝑑𝑖𝑠𝑐 P LieAlgpVectq denote
the Lie algebra obtained by forgetting the topology on h.85 We have a canonical map h𝑑𝑖𝑠𝑐 Ñ h
of Tate Lie algebras, giving rise to a forgetful functor h–mod Ñ h𝑑𝑖𝑠𝑐 –mod. By [GR4], the (non-
topological) algebra attached to h𝑑𝑖𝑠𝑐 is the usual enveloping algebra 𝑈 ph𝑑𝑖𝑠𝑐 q. Moreover, the natural
map:
𝑑𝑖𝑠𝑐
indhk𝑑𝑖𝑠𝑐 p𝑘q Ñ indhk p𝑘q P Vect
is an isomorphism. This immediately implies the claim.
Finally, it is immediate from the constructions to identify the compact generators.

Corollary 9.13.2. Under the above hypotheses, h–mod` is the bounded below derived category of
h–mod♡ .
Proof. Immediate from Lemma 9.13.1 and Proposition 3.7.1.


9.14. Classical semi-infinite cohomology. Let 𝐻 be a formally smooth polarizable Tate group
indscheme.
We let h𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 denote the central extension 0 Ñ 𝑘 Ñ h𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 Ñ h Ñ 0 of Tate Lie algebras
constructed e.g. in [BD1] S7.13. We abuse notation in letting h𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod denote not the category
of representations as is, but the analogue where we impose the requirement that the central 1 P 𝑘
act by the identity. We remind that h𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 is canonically split over any Lie subalgebra k0 Ď h that
is a lattice (in the usual sense of Tate vector spaces).
By Lemma 19.8.1 from [FG1] and Corollary 9.13.2 above, for 𝐾 Ď 𝐻 a compact open subgroup,
we have DG a functor:
8
2
𝐶𝑠𝑡𝑑,0 ph, k; ´q : h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod` Ñ Vect
of standard semi-infinite cohomology (defined in terms of Clifford algebras and spin representations)
whose restriction to h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –modě´𝑛 commutes with filtered colimits for any 𝑛. We also let:
8
2
𝐶𝑠𝑡𝑑 ph, k; ´q : h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod Ñ Vect
8
2
denote the functor obtained by restricting 𝐶𝑠𝑡𝑑,0 ph, k; ´q to h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod𝑐 and then ind-extending.

9.15. We will now show that the canonical natural transformation:

8 8
2 2
𝜂 : 𝐶𝑠𝑡𝑑 ph, k; ´q|h´𝑇 𝑎𝑡𝑒 –mod` Ñ 𝐶𝑠𝑡𝑑,0 ph, k; ´q
𝑠𝑡𝑑

of functors h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod` Ñ Vect is an isomorphism. (Combined with Theorem 9.16.1 below, this
8
means that 𝐶 2 ph, k; ´q|h´𝑇 𝑎𝑡𝑒 –mod` may be calculated using the standard semi-infinite complex.)
First, if h “ k, this follows immediately from the fact that compact objects in k–mod are closed
under truncations (c.f. Example 4.4.4).
8
In general, recall from [BD1] S7.13.29 that for 𝑀 P h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod♡ , 𝐶𝑠𝑡𝑑,0
2
ph, k; 𝑀 q has a canonical
ě0
increasing filtration indexed by Z with associated graded terms:
85In other words, we pass to the inverse limit of the pro-vector space underlying h and then apply 𝐻 0 if for some
pathological reason there are higher cohomology groups.
106 SAM RASKIN

8
2
gr𝑖 𝐶𝑠𝑡𝑑,0 ph, k; 𝑀 q “ 𝐶 ‚ pk, Λ𝑖 ph{kq b 𝑀 qr𝑖s (9.15.1)
for 𝐶 ‚ pk, ´q denoting the cohomological Chevalley complex. This is functorial in 𝑀 , so the functor
8
𝐶𝑠𝑡𝑑,0 ph, k; ´q : h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod` Ñ Vect upgrades to a functor valued in Fil ě0 Vect the (DG) category
2

of Zě0 -filtered vector spaces.

8
2
By construction, the functor 𝐶𝑠𝑡𝑑 ph, k; ´q also upgrades to Fil ě0 Vect compatibly with the above
and the map 𝜂. To check 𝜂 is an isomorphism, it is enough to do so on the associated graded level. As
the associated graded functors above factor through the forgetful functor h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod Ñ k–mod`
(c.f. (9.15.1)), we conclude as above.
9.16. For the remainder of this section, 𝐻 is a formally smooth polarizable Tate group indscheme
with 𝛿𝑇 𝑎𝑡𝑒 “ 0.
We have the following comparison theorem.
Theorem 9.16.1. For 𝐻 and 𝐾 as above, there is a canonical identification of h𝑇 𝑎𝑡𝑒 with h𝑇 𝑎𝑡𝑒𝑠𝑡𝑑
as central extensions of h. Moreover, under this identification, there is a canonical isomorphism
8 8
2
𝐶 2 ph, k; ´q » 𝐶𝑠𝑡𝑑 ph, k; ´q of functors h´𝑇 𝑎𝑡𝑒 –mod » h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod Ñ Vect. This pair of identifica-
tions is uniquely characterized by compatibility with the given splittings of these central extensions
over k and with the isomorphisms Lemmas 9.17.1 and 9.18.1 as formulated and proved below.
We will prove this result in the remainder of this section.
9.17. First, we review an important property of standard semi-infinite infinite cohomology.
Lemma 9.17.1. The composition:
h´𝑇 𝑎𝑡𝑒 8
indk 𝑠𝑡𝑑 2 ph,k;´q
𝐶𝑠𝑡𝑑
k–mod ÝÝÝÝÝÝÝÝÑ h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod ÝÝÝÝÝÝÝÑ Vect
is isomorphic to the functor 𝐶 ‚ pk, ´q :“ Homk–mod p𝑘, ´q : k–mod Ñ Vect of Lie algebra cohomology.
Proof. By [BD1] Remark 7.13.30, there is a canonical isomorphism of between the composite func-
tor:
h´𝑇 𝑎𝑡𝑒 8
𝑠𝑡𝑑 2
𝐶𝑠𝑡𝑑,0 ph,k;´q
` indk `
k–mod ÝÝÝÝÝÝÝÝÑ h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod ÝÝÝÝÝÝÝÝÑÑ Vect.
8
with 𝐶 ‚ pk, ´q| k–mod` .
2
Now the result follows by construction of 𝐶𝑠𝑡𝑑 .

8
9.18. We now establish similar results for 𝐶 ph, k; ´q. 2

First, observe that we have a duality functor:

8
2
Dh,𝐾 : h–mod_ » h´𝑇 𝑎𝑡𝑒 –mod P DGCat𝑐𝑜𝑛𝑡
(depending on the choice of compact open subgroup 𝐾). Indeed, our choice of 𝐾 identifies h´𝑇 𝑎𝑡𝑒 –mod »
Vectexpphq,𝑤 (c.f. S9.12). Note that this category is in fact dualizable as it is compactly generated,
and its dual is:

HomDGCat𝑐𝑜𝑛𝑡 pVectexpphq,𝑤 , Vectq “ HomDGCat𝑐𝑜𝑛𝑡 pVect, Vectqexpphq,𝑤 “ h–mod.

8
Under this duality, the functor 𝐶 2 ph, k; ´q clearly corresponds to the trivial representation 𝑘 P
h–mod♡ .
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 107

We now have:
Lemma 9.18.1. The composition:
h´𝑇 𝑎𝑡𝑒 8
ind 𝐶 2 ph,k;´q
k–mod ÝÝÝkÝÝÝÑ h´𝑇 𝑎𝑡𝑒 –mod ÝÝÝÝÝÝÑ Vect
is canonically isomorphic to the functor 𝐶 ‚ pk, ´q :“ Homk–mod p𝑘, ´q : k–mod Ñ Vect of Lie algebra
cohomology.
8
h
Proof. The induction functor indk ´𝑇 𝑎𝑡𝑒 is dual to Oblv : h–mod Ñ k–mod by construction of Dh,𝐾
2
.
We now obtain the result by duality.


9.19. We now prove Theorem 9.16.1.

Let h´𝑇 𝑎𝑡𝑒`𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 denote the Baer sum central extension of h´𝑇 𝑎𝑡𝑒 and h𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 . We maintain
our abuse of notation regarding modules over central extensions: the category h´𝑇 𝑎𝑡𝑒`𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod
is set up so the element 1 P 𝑘 Ď h´𝑇 𝑎𝑡𝑒`𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 acts by the identity on any object of it.86 Note
that this central extension is canonically split over k; in particular, we have a forgetful functor
Oblv : h´𝑇 𝑎𝑡𝑒`𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod Ñ k–mod.
8 8
2 2
The functor 𝐶𝑠𝑡𝑑 ph, k; ´q : h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod Ñ Vect defines by the duality Dh,𝐾 an object 𝐾 P
87
h´𝑇 𝑎𝑡𝑒`𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod.
Combining Lemmas 9.17.1 and 9.18.1, we find that Oblvp𝐾q “ 𝑘 P k–mod, where 𝑘 P k–mod indi-
cates the trivial module. In particular, as Oblv is 𝑡-exact and conservative, 𝐾 lies in h´𝑇 𝑎𝑡𝑒`𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod♡
and corresponds to a 1-dimensional representation.
Now observe that giving a 1-dimensional representation of a central extension 0 Ñ 𝑘 Ñ h5 Ñ
h Ñ 0 (on which the central element acts by the identity) is equivalent to splitting the central
extension: the induced map h5 Ñ h ˆ 𝑘 is an isomorphism of central extensions of h. Under this
splitting, the given 1-dimensional representation of h5 maps to the trivial representation of h.
Therefore, we obtain a trivialization of the central extension h´𝑇 𝑎𝑡𝑒`𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 of h such that 𝐾 maps
to the trivial object 𝑘 P h–mod. This is equivalent to giving an isomorphism h´𝑇 𝑎𝑡𝑒 » h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 of
central extensions such that the functor
8
2 ph,k;´q
𝐶𝑠𝑡𝑑
h´𝑇 𝑎𝑡𝑒 –mod » h´𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 –mod ÝÝÝÝÝÝÝÑ Vect
8
2
matches under the duality Dh,𝐾 to the trivial module 𝑘 P h–mod. This gives the desired isomorphism
8 8
2
𝐶 2 ph, k; ´q » 𝐶𝑠𝑡𝑑 ph, k; ´q.
Uniqueness follows as the map:

Auth–mod p𝑘q Ñ Autk–mod p𝑘q

is an isomorphism (both sides are 𝑘 ˆ , considered as group objects in Set Ď Gpd).

86This does not of course characterize the category. One can work with group indschemes and central extensions
by G𝑚 as above to give one quick definition. Alternatively, one can note that the centrality means QCohpA1 q “ 𝑘–mod
(regarding 𝑘 as an abelian Lie algebra in this notation) acts canonically on the category of all h´𝑇 𝑎𝑡𝑒`𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 -modules,
and we are taking the fiber of that category at 1 P A1 p𝑘q.
87As always, this notation abusively indicates that the central element 1 P 𝑘 Ď h
´𝑇 𝑎𝑡𝑒`𝑇 𝑎𝑡𝑒𝑠𝑡𝑑 acts by the identity
on our modules, understood in the appropriately derived sense.
108 SAM RASKIN

10. Harish-Chandra data

10.1. Suppose 𝐻 is an algebraic group and 𝐴 P Alg is equipped with an action of 𝐻, giving a
weak action of 𝐻 on 𝐴–mod. It is not difficult to see in this setup that upgrading88 this action to
a strong one is equivalent to specifying a Harish-Chandra datum in a suitable derived sense.
We remind that this means we are given an 𝐻-equivariant map of Lie algebras 𝑖 : h Ñ 𝐴
satisfying a number of “compatibilities” (which are actually extra data in a derived setting), most
notably, that the corresponding adjoint action of h on 𝐴 coincides with the infinitesimal action of
𝐻 on 𝐴.
10.2. The goal for this section is to develop such ideas in the setting where 𝐻 is a Tate group
Ñ
indscheme and 𝐴 is an b-algebra.
There are a number of subtleties compared to the finite dimensional setting discussed above
related to the ideas developed so far in this text.
First, 𝐴 needs to be equipped with a renormalization datum compatible with the 𝐻-action in
the sense of S5.4, and with the Harish-Chandra data in a suitable sense.
Second, we need to upgrade the naive action of 𝐻 on 𝐴–mod𝑟𝑒𝑛 to a genuine one. We do this
using the theory of canonical renormalization from S8.17.
With that said, the theory we develop has no89 homotopical complexity for 𝐴 and 𝐻 classical.90
The main example to have in mind is 𝐴 “ 𝑈 phq, the completed enveloping algebra of h (i.e., the
Ñ
b-algebra assigned to the 𝑡-exact functor h–mod :“ Vectexpphq,𝑤 Ñ Vect via Proposition 3.7.1).
For the above to make sense, we need a key technical result, Theorem 10.8.1, that (in particular)
says that the genuine weak action of 𝐻 on h–mod comes from canonical renormalization. We need
to impose two hypotheses on the group indschemes 𝐻 to obtain this result: that 𝐻 is polarizable,
and that it has a prounipotent tail, i.e., there exists a prounipotent compact open subgroup in 𝐻.
Therefore, these hypotheses trail us throughout this section. We remark that they are satisfied in
the main example of interest: when 𝐻 is the loop group of a reductive group (or a central extension
of such).
10.3. As this section is lengthy, we begin with a brief guide to its structure.
Ñ
In S10.4-10.7, we introduce the notion of genuine 𝐻-action on an b-algebra 𝐴; roughly, this
means there is a genuine 𝐻-action on 𝐴–mod𝑟𝑒𝑛 defined by canonical renormalization.
In S10.8, we formulate Theorem 10.8.1, which was mentioned above. The proof occupies S10.8-
10.17.
In S10.20, we formulate our definition of Harish-Chandra data, which relies on Theorem 10.8.1.
Finally, in S10.22-10.23, we discuss Harish-Chandra data explicitly in the case where 𝐴 is classical.
Ñ
10.4. Genuine actions and b-algebras. In what follows, let 𝐻 be an ind-affine Tate group
indscheme. (These hypotheses will be strengthened in S10.18.)
Ñ Ñ
b,𝐻ñ
Recall the notation Alg𝑟𝑒𝑛 from S5.7: this is the category of renormalized b-algebras with
naive 𝐻-actions that are compatible with the renormalization.
88Although the forgetful functor 𝐴–mod Ñ Vect is weakly 𝐻-equivariant, the “upgrade” in question does not
interact with the forgetful functor. For example, 𝐻 acts strongly on h–mod, but the forgetful functor h–mod Ñ Vect
is only weakly equivariant.
89Although the data is 1-categorical in nature, checking that an apparent Harish-Chandra datum actually defines
one in our sense involves non-trivial homological algebra (as we will see).
90In fact, the theory essentially requires 𝐻 to be classical from the start. More precisely, we require 𝐻 to be
formally smooth, which forces 𝐻 to be classical under mild countability assumptions; see [GR3].
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 109

Definition 10.4.1. The 1-full subcategory:

Ñ Ñ
1 b,𝐻ñ b,𝐻ñ
Alg𝑔𝑒𝑛 Ď Alg𝑟𝑒𝑛
Ñ Ñ
b,𝐻ñ
of b-algebras with nearly genuine 𝐻-actions has objects 𝐴 P Alg𝑟𝑒𝑛 such that the naive action
of 𝐻 on 𝐴–mod𝑟𝑒𝑛 (as in S5.7) canonically renormalizes (in the sense of S8.17) with respect to the
given 𝑡-structure on 𝐴–mod𝑟𝑒𝑛 .
Ñ Ñ
b,𝐻ñ b,𝐻ñ
Morphisms 𝐴1 Ñ 𝐴2 in 1Alg𝑔𝑒𝑛 are morphisms in Alg𝑟𝑒𝑛 with the property that for every
𝐾 Ď 𝐻 a compact open subgroup, the functor:

𝐴2 –mod𝐾,𝑤 𝐾,𝑤
𝑟𝑒𝑛 Ñ 𝐴1 –mod𝑟𝑒𝑛
obtained by ind-extension from:

𝐴2 –mod𝐾,𝑤,𝑐
𝑟𝑒𝑛 Ñ 𝐴2 –mod𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒,`
𝑟𝑒𝑛 Ñ 𝐴1 –mod𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒,`
𝑟𝑒𝑛 » 𝐴1 –mod𝐾,𝑤,`
𝑟𝑒𝑛 Ď 𝐴1 –mod𝐾,𝑤
𝑟𝑒𝑛

is 𝑡-exact (equivalently, left 𝑡-exact).

Finally, we define:
Ñ Ñ
b,𝐻ñ
Alg𝑔𝑒𝑛 Ď 1Alg𝑔𝑒𝑛
b,𝐻ñ

Ñ Ñ
b,𝐻ñ
of b-algebras with genuine 𝐻-actions as the full subcategory consisting of objects 𝐴 P 1Alg𝑔𝑒𝑛 such
Ñ Ñ
b,𝐻ñ b,𝐻ñ
that the unit morphism 𝑘 Ñ 𝐴 P Alg𝑟𝑒𝑛 is a morphism in the 1-full subcategory 1Alg𝑟𝑒𝑛 . (In
𝐾,𝑤
other words, the forgetful functor 𝐴–mod𝑟𝑒𝑛 Ñ Vect induces a 𝑡-exact functor 𝐴–mod𝑟𝑒𝑛 Ñ Repp𝐾q
for any compact open subgroup 𝐾 Ď 𝐻.)
Remark 10.4.2. By construction, each of the restriction functors:
Ñ Ñ Ñ
b,𝐻ñ b,𝐻ñ
Alg𝑔𝑒𝑛 Ñ Alg𝑟𝑒𝑛 Ñ Alg b,𝐻ñ
is 1-fully-faithful; the former is in addition conservative. (This is an abstract way of saying that a
genuine action of 𝐻 on 𝐴 is equivalent to specifying a naive action and a renormalization datum for
𝐴 satisfying some properties, and that genuinely equivariant morphisms are naively 𝐻-equivariant
morphisms satisfying some properties.)
Ñ Ñ
b,𝐻ñ b,𝐻ñ
Definition 10.4.3. For 𝐴1 , 𝐴2 P Alg𝑔𝑒𝑛 , we say a morphism 𝑓 : 𝐴1 Ñ 𝐴2 in Alg𝑔𝑒𝑛 is a genuinely
Ñ
𝐻-equivariant morphism. We refer to a morphisms in Alg b,𝐻ñ as naively 𝐻-equivariant.
10.5. There is an evident functor:
Ñ
b,𝐻ñ 𝑜𝑝
pAlg𝑔𝑒𝑛 q Ñ 𝐻–mod𝑤𝑒𝑎𝑘
𝐴 ÞÑ 𝐴–mod𝑟𝑒𝑛
given by canonical renormalization. Following our standard abuses for genuine 𝐻-actions, we denote
Ñ
b,𝐻ñ
this functor 𝐴 ÞÑ 𝐴–mod𝑟𝑒𝑛 . Moreover, as 𝑘 P Alg𝑔𝑒𝑛 is an initial object (by fiat in our definition
of genuine 𝐻-action), this functor upgrades to a functor to the overcategory p𝐻–mod𝑤𝑒𝑎𝑘 q{Vect : for
Ñ
b,𝐻ñ
𝐴 P Alg𝑔𝑒𝑛 , the structural map 𝐴–mod𝑟𝑒𝑛 Ñ Vect is the forgetful functor.
110 SAM RASKIN

Ñ Ñ
b,𝐻ñ b,𝐻ñ
Let Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 Ď Alg𝑔𝑒𝑛 be the full subcategory consisting of those objects whose underlying
Ñ
b-algebra is convergent.
Theorem 10.5.1. For 𝐻 polarizable, the functor:
Ñ
b,𝐻ñ
pAlg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 q𝑜𝑝 Ñ p𝐻–mod𝑤𝑒𝑎𝑘 q{Vect
is 1-fully-faithful and conservative.
We defer the proof to S10.7.
Ñ
b,𝐻ñ
Remark 10.5.2. Although the definition of the category Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 is weighty, this result gives a way
Ñ
to convert algebraic data in Alg b that may be quite concrete to abstract categorical data involving
genuine 𝐻-actions.
10.6. To prove Theorem 10.5.1, we will need the following result.
Proposition 10.6.1. Let 𝐻 be a polarizable Tate group indscheme. Let C, D P 𝐻–mod𝑤𝑒𝑎𝑘 be
equipped with 𝑡-structures compatible with the weak 𝐻-actions. Suppose the genuine 𝐻-actions on
each of C and D are obtained by canonical renormalization using these 𝑡-structures and the under-
lying naive 𝐻-actions (c.f. S8.17).
Let 𝐹, 𝐺 : C Ñ D P 𝐻–mod𝑤𝑒𝑎𝑘 be two genuinely 𝐻-equivariant functors, and suppose that 𝐺 is
left 𝑡-exact (at the level of its underlying functor C Ñ D P DGCat𝑐𝑜𝑛𝑡 ).
Then the natural map:

HomHom𝐻–mod p𝐹, 𝐺q Ñ HomHom𝐻–mod p𝐹, 𝐺q P Vect

𝑤𝑒𝑎𝑘 𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒

(induced by Oblv𝑔𝑒𝑛 ) is an equivalence. In other words, giving a genuinely 𝐻-equivariant natural

transformation between 𝐹 and 𝐺 is equivalent to giving a naively 𝐻-equivariant natural transfor-
mation between them.
We will use the following lemma.
Lemma 10.6.2. Let 𝐻 be a Tate group indscheme and let 𝐾 Ď 𝐻 be a polarization of 𝐻. Then
the forgetful functors:

C𝐻,𝑤 Ñ C𝐾,𝑤
C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒
𝑤,𝑛𝑎𝑖𝑣𝑒
admit left adjoints, denoted Av𝑤
! and Av! respectively. Moreover, the diagram:

C𝐾,𝑤 / C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒

Av𝑤
! Av𝑤,𝑛𝑎𝑖𝑣𝑒
  !

C𝐻,𝑤 / C𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒
commutes (a priori, it commutes up to a natural transformation).
Proof. The existence of Av𝑤
! , as we have appealed to at various points earlier in this text, follows
from (7.15.1) and the Beck-Chevalley formalism.
Let Φ : 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 Ñ 𝐻–mod𝑤𝑒𝑎𝑘 denote the (non-continuous) right adjoint to Oblv𝑔𝑒𝑛 .
Clearly Φp´q𝐻,𝑤 “ p´q𝐻,𝑤,𝑛𝑎𝑖𝑣𝑒 . Moreover, passing to right adjoints in Lemma 8.7.1 (2) it follows
that Φp´q𝐾,𝑤 “ p´q𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 . Now the commutativity of the diagram follows by rewriting it as:
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 111

C𝐾,𝑤 / ΦpOblv𝑔𝑒𝑛 pCqq𝐾,𝑤

Av𝑤
! Av𝑤
!
 
C𝐻,𝑤 / ΦpOblv𝑔𝑒𝑛 pCqq𝐻,𝑤 .

(Alternatively, the base-change follows directly by applying the Beck-Chevalley formalism in the
naive setting and comparing with (7.15.1).)

Proof of Proposition 10.6.1. In what follows, let 𝐾 Ď 𝐻 be a polarization.
Step 1. Let HompC, Dq P 𝐻–mod𝑤𝑒𝑎𝑘 denote the inner Hom object between C and D in the sym-
metric monoidal category 𝐻–mod𝑤𝑒𝑎𝑘 .
Note that Oblv𝑔𝑒𝑛 HompC, Dq is the category HompC, Dq :“ HomDGCat𝑐𝑜𝑛𝑡 pC, Dq of continuous
DG functors between C and D. Indeed, this follows from the fact that Oblv𝑔𝑒𝑛 admits a DGCat𝑐𝑜𝑛𝑡 -
continuous left adjoint ´ b IndCoh˚ p𝐻q (for IndCoh˚ p𝐻q P 𝐻–mod𝑤𝑒𝑎𝑘 as in Example 8.21.2).
Similarly, formation of inner Homs is intertwined by the forgetful functor 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐾–mod𝑤𝑒𝑎𝑘 :
this follows from the existence of the left adjoint ind𝐻,𝑤
𝐾 from Lemma 8.7.1 and the (evident) version
of the projection formula for this left adjoint.
Clearly HompC, Dq𝐻,𝑤 “ Hom𝐻–mod𝑤𝑒𝑎𝑘 pC, Dq. By the above, we just as well have HompC, Dq𝐾,𝑤 “
Hom𝐾–mod𝑤𝑒𝑎𝑘 pC, Dq. Finally, because Oblv𝑔𝑒𝑛 HompC, Dq is the category of functors between C and
D, we have:

HompC, Dq𝐻,𝑤 “ Hom𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 pC, Dq

HompC, Dq𝐾,𝑤 “ Hom𝐾–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 pC, Dq.

Step 2. Consider 𝐺 P HompC, Dq𝐾,𝑤 as above (though it lifts to 𝐻-invariants).

Let Oblv : HompC, Dq𝐾,𝑤 Ñ HompC, Dq P DGCat𝑐𝑜𝑛𝑡 be the forgetful functor, and let Av𝐾,𝑤
˚ de-
note its right adjoint. We claim that the natural map 𝐺 Ñ TotppAv𝐾,𝑤
˚ Oblvq‚`1 𝐺q P HompC, Dq𝐾,𝑤

is an equivalence. Here we emphasize that the totalization is calculated in the DG category

HompC, Dq𝐾,𝑤 .
Indeed, we have:

HompC, Dq𝐾,𝑤 “ Hom𝐾–mod𝑤𝑒𝑎𝑘 pC, Dq “ HomRepp𝐾q–mod pC𝐾,𝑤 , D𝐾,𝑤 q “

(10.6.1)
HomRepp𝐾q𝑐 –modpDGCatq pC𝐾,𝑤,𝑐 , D𝐾,𝑤 q
for C𝐾,𝑤,𝑐 Ď C𝐾,𝑤 the subcategory of compact objects: we remind that as part of the definition of C
being obtained from canonical renormalization, C𝐾,𝑤 is compactly generated with compact objects
being eventually coconnective.
The advantage of the last expression in (10.6.1) is that limits are manifestly computed termwise.
So for F P C𝐾,𝑤,𝑐 , we have:

TotppAv𝐾,𝑤
˚ Oblvq‚`1 𝐺qpFq “ TotppAv𝐾,𝑤
˚ Oblvq‚`1 𝐺pFqq P D𝐾,𝑤 .
By Proposition 5.18.3 (4), the natural map from 𝐺pFq to this limit is an equivalence.
Step 3. As an immediate consequence of Step 2, note that for any 𝐹r : C Ñ D genuinely 𝐾-
equivariant, the map:
112 SAM RASKIN

HomHom𝐾–mod p𝐹r , 𝐺q Ñ HomHom𝐾–mod p𝐹r , 𝐺q

𝑤𝑒𝑎𝑘 𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒

is an isomorphism.
Step 4. We now complete the argument. We again view HompC, Dq P 𝐻–mod𝑤𝑒𝑎𝑘 and 𝐹, 𝐺 as
objects in HompC, Dq𝐻,𝑤 .
Note that the forgetful functor Oblv : HompC, Dq𝐻,𝑤 Ñ HompC, Dq𝐾,𝑤 is conservative (by
(7.15.1)) and admits a left adjoint Av𝑤! (c.f. Lemma 10.6.2). Therefore, this forgetful functor is
monadic. The same applies in the naively equivariant setting.
We obtain that 𝐹 is a geometric realization:
»
Ñ 𝐹 P HompC, Dq𝐻,𝑤 “ Hom𝐻–mod𝑤𝑒𝑎𝑘 pC, Dq.
|pAv! Oblvq‚`1 p𝐹 q| Ý
Therefore, we have:

» ‚ »
HomHom𝐻–mod pC,Dq p𝐹, 𝐺q Ñ Tot HomHom𝐾–mod
Ý pC,Dq pOblvpAv! Oblvq p𝐹 q, 𝐺q Ý
Ñ
𝑤𝑒𝑎𝑘 𝑤𝑒𝑎𝑘
‚
Tot HomHom𝐾–mod pC,Dq pOblvpAv! Oblvq p𝐹 q, 𝐺q.
𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒

Here we are using Step 3 in the second isomorphism, and we are implicitly using Lemma 10.6.2 to
intertwine Av! functors in the genuine and naive settings. By the same logic, the last term above
computes Hom𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 p𝐹, 𝐺q, giving the claim.

10.7. As promised, we now prove the above theorem.
Proof of Theorem 10.5.1. First, let us verify that the functor is conservative. The composition of
this functor with the forgetful functors:
Oblv𝑔𝑒𝑛
p𝐻–mod𝑤𝑒𝑎𝑘 q{Vect Ñ 𝐻–mod𝑤𝑒𝑎𝑘 ÝÝÝÝÝÑ DGCat𝑐𝑜𝑛𝑡
Ñ
b,𝐻ñ
send 𝐴 P Alg𝑔𝑒𝑛 to 𝐴–mod𝑟𝑒𝑛 . This functor is conservative by Remark 4.2.4, giving the claim.
Ñ
b,𝐻ñ
Now for 𝐴1 , 𝐴2 P Alg𝑔𝑒𝑛 , we have the following commutative diagram:

Hom Ñ p𝐴1 , 𝐴2 q / Hom1 p𝐴2 –mod𝑟𝑒𝑛 , 𝐴1 –mod𝑟𝑒𝑛 q

b ,𝐻ñ p𝐻–mod𝑤𝑒𝑎𝑘 q {Vect
Alg𝑔𝑒𝑛

(10.7.1)
 
Hom p𝐴1 , 𝐴2 q / Hom1
p𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 q{Vect p𝐴2 –mod𝑟𝑒𝑛 , 𝐴1 –mod𝑟𝑒𝑛 q.
Ñ
b ,𝐻ñ
Alg𝑟𝑒𝑛

Here the decoration 1 on the bottom left term indicates the subcategory of those functors that
are 𝑡-exact after applying p´q𝐾,𝑤 for any compact open subgroup 𝐾, while the similar notation
on the bottom right term indicates the subcategory of 𝑡-exact functors. We wish to show that the
top arrow in (10.7.1) is fully-faithful. We will do so by showing that the other three arrows are
fully-faithful.
The left arrow of (10.7.1) is fully-faithful by definition of genuine 𝐻-actions.
The right arrow of (10.7.1) is fully-faithful by Proposition 10.6.1.
Finally, the bottom arrow of (10.7.1) is an equivalence by Remark 4.2.4; indeed, by loc. cit. (or
more precisely, by Proposition 3.7.1 and Theorem 4.6.1), the functor:
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 113

Ñ
b
pAlg𝑐𝑜𝑛𝑣,𝑟𝑒𝑛 q𝑜𝑝 Ñ pDGCat𝑐𝑜𝑛𝑡 q{Vect
𝐴 ÞÑ 𝐴–mod𝑟𝑒𝑛
is symmetric monoidal and fully-faithful.91

10.8. A construction of genuine 𝐻-actions. We now formulate a key result that allows us to
Ñ
construct many genuine 𝐻-actions on b- algebras.
Ñ
b,𝐻ñ
Suppose 𝐴 P Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 be given. We have the corresponding object 𝐴–mod𝑟𝑒𝑛 P 𝐻–mod𝑤𝑒𝑎𝑘 .
We define:

𝐴#𝑈 phq–mod𝑟𝑒𝑛 :“ Oblv𝑠𝑡𝑟Ñ𝑤 p𝐴–mod𝑟𝑒𝑛 expphq,𝑤

q P 𝐻–mod𝑤𝑒𝑎𝑘 .
(The notation will be justified in what follows.) Note that this category has a canonical genuinely
𝐻-equivariant functor to Vect:

𝐴#𝑈 phq–mod𝑟𝑒𝑛 “ Oblv𝑠𝑡𝑟Ñ𝑤 p𝐴–mod𝑟𝑒𝑛 expphq,𝑤

q Ñ 𝐴–mod𝑟𝑒𝑛 Ñ Vect
where the first functor is the counit for the adjunction and the second arrow is the standard forgetful
functor for 𝐴–mod𝑟𝑒𝑛 .
Theorem 10.8.1. Suppose that 𝐻 is formally smooth Tate group indscheme that is polarizable,
and has a prounipotent tail. Then 𝐴#𝑈 phq–mod𝑟𝑒𝑛 P p𝐻–mod𝑤𝑒𝑎𝑘 q{Vect lies in the essential image
of the functor from Theorem 10.5.1.
Example 10.8.2. Taking 𝐴 “ 𝑘, this result is already quite non-trivial: it says that under the
above hypotheses, the naive 𝐻-action on h–mod canonically renormalizes (with respect to the
standard 𝑡-structure), and that h–mod :“ Oblv𝑠𝑡𝑟Ñ𝑤 pVectexpphq,𝑤 q P 𝐻–mod𝑤𝑒𝑎𝑘 is its canonical
renormalization.
The proof of Theorem 10.8.1 is involved, so is deferred to S10.17 so that we may first give some
preliminary results.
Remark 10.8.3. To orient the reader, let us briefly discuss the importance of Theorem 10.8.1 for
defining Harish-Chandra data. Assume the result for now. Of course, we should let 𝐴#𝑈 phq denote
Ñ
b,𝐻ñ
the corresponding object of Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 .
As the notation indicates, 𝐴#𝑈 phq should be understood as the usual smash product. Then a
morphism 𝐴#𝑈 phq Ñ 𝐴 restricting to the identity along the canonical embedding 𝐴 Ñ 𝐴#𝑈 phq is
the “main” part of a Harish-Chandra datum (i.e., the rest of the data is interpreted as homotopy
compatibilities). We refer to S10.22-10.23 for more explicit discussion.
Ñ
Remark 10.8.4. Before proving the theorem, we do not make explicit reference to the b-algebra
𝐴#𝑈 phq, only its category of modules. That is, we treat 𝐴#𝑈 phq–mod𝑟𝑒𝑛 as alternative notation
expphq,𝑤
to Oblv𝑠𝑡𝑟Ñ𝑤 p𝐴–mod𝑟𝑒𝑛 q. We let Oblv : 𝐴#𝑈 phq–mod𝑟𝑒𝑛 Ñ Vect denote the forgetful functor
constructed above.
91Explicitly, the essential image of this functor consists of those compactly generated DG categories C equipped
with 𝐹 : C Ñ Vect such that for the 𝑡-structure on C defined by having Cď0 generated under colimits by compact
objects F P C𝑐 such that 𝐹 pFq P Vectď0 , the functor 𝐹 is 𝑡-exact and conservative on C` , and such that compact
objects in C are eventually coconnective.
114 SAM RASKIN

10.9. 𝑡-structures. As Theorem 10.8.1 concerns canonical renormalization and therefore 𝑡-structures,
it is convenient to have some convenient language regarding 𝑡-structures in the presence of 𝐻-
actions.
Therefore, we begin with an extended digression on this subject. The reader may safely skip
ahead to S10.14 and refer back as needed.

10.10. Suppose 𝐻 is a Tate group indscheme and C P 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 .

Note that the action functor:

actC : IndCoh˚ p𝐻q b C Ñ C

lifts canonically as:

𝛼C
IndCoh˚ p𝐻q b C / IndCoh˚ p𝐻q b C
actC ΓIndCoh p𝐻,´qbid C

& x
C
with 𝛼C an equivalence. Indeed, viewing IndCoh˚ p𝐻q as a coalgebra in DGCat𝑐𝑜𝑛𝑡 (via pushforwards
along diagonal maps, as works for any strict indscheme), there is a unique map of IndCoh˚ p𝐻q-
comodules 𝛼C fitting into a diagram as above. That this functor is an equivalence follows from the
case C “ IndCoh˚ p𝐻q, where it is follows from strictness of 𝐻.
Suppose now that C is equipped with a 𝑡-structure. We say this 𝑡-structure is compatible with
the (naive, weak) action of 𝐻 on C if it is compatible with filtered colimits and 𝛼C is 𝑡-exact when
both sides are equipped with the tensor product 𝑡-structures (as in Lemma 4.6.2).
Ñ
b,𝐻ñ
Example 10.10.1. Suppose that 𝐴 P Alg𝑟𝑒𝑛 . Then the induced naive, weak 𝐻-action on 𝐴–mod𝑟𝑒𝑛
is compatible with the 𝑡-structure. Indeed, this is a restatement of the definition of compatibility
between an 𝐻-action and a renormalization datum.
10.11. We now move to discuss 𝑡-structures in the presence of strong 𝐻-actions.
We will need the following result.
Lemma 10.11.1. Suppose 𝐾 is a classical affine group scheme. Suppose C P 𝐾–mod is a DG
category equipped with a strong 𝐾-action.
Let 𝐾 “ lim𝑗 𝐾{𝐾𝑗 with 𝐾{𝐾𝑗 an algebraic group (so 𝐾𝑗 Ď 𝐾 is a normal compact open
subgroup).
Below, we also write C for the induced object of 𝐾–mod𝑤𝑒𝑎𝑘 under Oblv𝑠𝑡𝑟Ñ𝑤 .
(1) The natural functor colim𝑗 pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 Ñ C𝐾,𝑤 P DGCat𝑐𝑜𝑛𝑡 is an equivalence.
(2) Each of the structural functors in this colimit admits a continuous right adjoint.
Proof. By construction, we have:

k–mod “ colim pk{k𝑖 q–mod “ colim 𝐷p𝐾{𝐾𝑖 q𝐾{𝐾𝑖 ,𝑤 P DGCat𝑐𝑜𝑛𝑡

𝑖 𝑖
as p𝐷p𝐾q, Repp𝐾qq-bimodules. Moreover, each structural functor in this colimit clearly admits a
continuous right adjoint (calculated as Lie algebra cohomology for the appropriate finite dimensional
Lie algebra), and that right adjoint is a morphism of p𝐷p𝐾q, Repp𝐾qq-bimodules. This gives the
claim by construction of Oblv𝑠𝑡𝑟Ñ𝑤 .

 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 115

10.12. Now suppose 𝐻 is a Tate group indscheme with a prounipotent tail, and that C P 𝐻–mod
is acted on strongly by 𝐻 and is equipped with a 𝑡-structure.
We say that this 𝑡-structure is strongly compatible with the 𝐻-action if it compatible with the
underlying weak, naive action of 𝐻 and for any prounipotent compact open subgroup 𝐾 Ď 𝐻, the
subcategory C𝐾 Ď C is closed under truncations.
Remark 10.12.1. For 𝐻 an algebraic group, this condition is clearly equivalent to the 𝑡-structure
being compatible with the underlying weak action. As discussed in [Ras6] SB.4, this is equivalent
to any other notion of a 𝑡-structure being compatible with a strong action essentially because the
forgetful functor Oblv : 𝐷p𝐻q Ñ QCohp𝐻q is conservative and 𝑡-exact up to shift.
From here, one deduces that in general, if a 𝑡-structure is compatible with the weak naive action
of a Tate group indscheme 𝐻, it is strongly compatible if and only if the above condition holds for
some prounipotent subgroup.
Remark 10.12.2. We do not know of an example of C P 𝐻–mod as above and a 𝑡-structure that is
compatible with the weak, naive action of 𝐻 but not strongly compatible.
10.13. Suppose C P 𝐻–mod is equipped with a 𝑡-structure that is strongly compatible with the
𝐻-action.
Observe that for 𝐾 Ď 𝐻 any compact open, C𝐾 inherits a unique 𝑡-structure such that C𝐾 Ñ C
is 𝑡-exact. Indeed, if 𝐾 is prounipotent, this is true by design; in general, choose 𝐾 𝑢 Ď 𝐾 a normal,
prounipotent compact open subgroup, and then observe that the action of the algebraic group
𝑢
𝐾{𝐾 𝑢 on C𝐾 is compatible with the 𝑡-structure there in the sense of [Ras6] Appendix B. By loc.
cit., we obtain the claim.
More generally, in the above notation, each pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 inherits a canonical 𝑡-structure, and
each of the structural functors in the colimit from Lemma 10.11.1 is 𝑡-exact.
Therefore, C𝐾,𝑤 admits a unique 𝑡-structure such that each functor pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 Ñ C𝐾,𝑤 is
𝑡-exact: see [Ras6] Lemma 5.4.3.
Lemma 10.13.1. In the above setting, the forgetful functor C𝐾,𝑤,` Ñ C` is conservative.
Proof. For every 𝑗, let 𝛼𝑗 : C𝐾,𝑤 Ñ pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 be the right adjoint to the structural functor
𝛽𝑗 : pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 Ñ C𝐾,𝑤 .
By 𝑡-exactness, it suffices to show that this functor is conservative on the heart. For F P C𝐾,𝑤,♡ ,
we have F “ colim𝑗 𝛽𝑗 𝛼𝑗 pFq. Note that 𝛽𝑗 is 𝑡-exact, so 𝛼𝑗 is left 𝑡-exact. On 𝐻 0 , each structural
map in this colimit is a monomorphism: indeed, 𝐻 0 p𝛽𝑗 𝛼𝑗 pFqq is the maximal subobject of F ly-
ing in pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤,♡ (which is a full subcategory of C𝐾,𝑤,♡ because we are working with abelian
categories).
Now if F ‰ 0, then 𝐻 0 p𝛼𝑗 pFqq ‰ 0 for some 𝑗, and as the composition pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 Ñ C𝐾,𝑤 Ñ C
is clearly conservative, we obtain the claim.

Corollary 10.13.2. In the above setting, the natural functor C𝐾,𝑤,` Ñ C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒,` is an equiva-
lence.
Proof. The 𝑡-exact conservative forgetful functor C𝐾,𝑤,` Ñ C` is comonadic by Lemma 3.7.2. The
comonads on C defined by C𝐾,𝑤 and C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 always coincide, so we obtain the claim.

10.14. Some results on smash products. We now suppose we are in the setup of Theorem
10.8.1.
116 SAM RASKIN

Lemma 10.14.1. (1) 𝐴#𝑈 phq–mod𝑟𝑒𝑛 is compactly generated.

(2) There exists a unique compactly generated 𝑡-structure on 𝐴#𝑈 phq–mod𝑟𝑒𝑛 such that compact
objects are eventually coconnective and such that the forgetful functor 𝐴#𝑈 phq–mod𝑟𝑒𝑛 Ñ
Vect is 𝑡-exact and conservative on 𝐴#𝑈 phq–mod` 𝑟𝑒𝑛 .
(3) The above 𝑡-structure is strongly compatible with the 𝐻-action on 𝐴#𝑈 phq–mod𝑟𝑒𝑛 .
(4) For any compact open subgroup 𝐾 Ď 𝐻, 𝐴#𝑈 phq–mod𝐾,𝑤 𝑟𝑒𝑛 is compactly generated.
(5) By (3) and S10.13, 𝐴#𝑈 phq–mod𝐾,𝑤 𝑟𝑒𝑛 is equipped with a canonical 𝑡-structure. This 𝑡-
structure is compactly generated, and compact objects are bounded from below. Moreover,
the forgetful functor 𝐴#𝑈 phq–mod𝐾,𝑤
𝑟𝑒𝑛 Ñ Repp𝐾q is 𝑡-exact.
Ñ
(6) Suppose 𝑓 : 𝐴 Ñ 𝐵 is a morphism of b-algebras with genuine 𝐻-actions.
Then for any 𝐾 Ď 𝐻 compact open, the functor:

𝐵#𝑈 phq–mod𝐾,𝑤 𝐾,𝑤

𝑟𝑒𝑛 Ñ 𝐴#𝑈 phq–mod𝑟𝑒𝑛
is 𝑡-exact (for the 𝑡-structures from (5)).
Proof. We proceed in steps.
Step 1. Recall that by construction, the (genuine) weak 𝐻-action on 𝐴#𝑈 phq–mod𝑟𝑒𝑛 canonically
upgrades to a strong action. We use the same notation 𝐴#𝑈 phq–mod𝑟𝑒𝑛 for the corresponding
object of 𝐻–mod; in particular, for 𝐾 Ď 𝐻 compact open, we let 𝐴#𝑈 phq–mod𝐾 𝑟𝑒𝑛 denote the
strong 𝐾-invariants for the action.
We begin by proving that 𝐴#𝑈 phq–mod𝐾 𝑟𝑒𝑛 is compactly generated for any compact open sub-
group 𝐾 Ď 𝐻.
Note that by definition, we have:
𝐻 ^ ,𝑤
𝐴#𝑈 phq–mod𝐾 expphq,𝑤 𝐾
𝑟𝑒𝑛 “ p𝐴–mod𝑟𝑒𝑛
𝐾
q “ 𝐴–mod𝑟𝑒𝑛
^ is the formal completion of 𝐻 along 𝐾. The forgetful functor:
where 𝐻𝐾
𝐻 ^ ,𝑤
𝐴#𝑈 phq–mod𝐾 𝐾
𝑟𝑒𝑛 “ 𝐴–mod𝑟𝑒𝑛 Ñ 𝐴–mod𝐾,𝑤
𝑟𝑒𝑛
is conservative and admits a left adjoint; as this forgetful functor is manifestly continuous, it is
monadic.
Because the 𝐻-action on 𝐴–mod𝑟𝑒𝑛 arises by canonical renormalization (by assumption on 𝐴),
𝐴–mod𝐾,𝑤
𝑟𝑒𝑛 is compactly generated. By the above, we immediately deduce the same for 𝐴#𝑈 phq–mod𝑟𝑒𝑛 .
𝐾

Step 2. Note that the monad on 𝐴–mod𝐾,𝑤 𝑟𝑒𝑛 constructed in Step 1 is 𝑡-exact for the 𝑡-structure
on 𝐴–mod𝐾,𝑤 𝑟𝑒𝑛 (coming from Proposition 5.18.3). Indeed, as 𝐻 is formally smooth, 𝐻𝐾 ^ is too, so

𝜔𝐻𝐾 ^
^ {𝐾 P IndCohp𝐻 {𝐾q
𝐾,𝑤 lies in the heart of the 𝑡-structure. The monad in question is given
𝐾
by convolution with this object. Convolution by an object in Cohp𝐻𝐾 ^ q𝐾,𝑤,♡ defines a functor

𝐴–mod𝐾,𝑤 𝐾,𝑤
𝑟𝑒𝑛 Ñ 𝐴–mod𝑟𝑒𝑛 that is left 𝑡-exact up to shift by definition of canonical renormalization,
and it is 𝑡-exact by the compatibility of the naive action of 𝐻 on 𝐴–mod𝑟𝑒𝑛 with the 𝑡-structure;
therefore, the same applies to convolution by arbitrary objects of IndCohp𝐻𝐾 ^ q𝐾,𝑤,♡ , giving the

claim. ^ ,𝑤
𝐻𝐾
It follows that 𝐴–mod𝑟𝑒𝑛 “ 𝐴#𝑈 phq–mod𝐾 𝑟𝑒𝑛 admits a unique 𝑡-structure such that the (monadic,
with monad the one in question) forgetful functor to 𝐴–mod𝐾,𝑤 𝑟𝑒𝑛 is 𝑡-exact.

Step 3. We now deduce (1) and (2).

We have:
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 117

𝐴#𝑈 phq–mod𝑟𝑒𝑛 “ colim 𝐴#𝑈 phq–mod𝐾

𝑟𝑒𝑛 P DGCat𝑐𝑜𝑛𝑡 .
𝐾Ď𝐻 compact open
Each structural functor in this colimit admits a continuous right adjoint so preserves compact
objects. We obtain that the colimit is compactly generated since each term is by Step 1.
Moreover, we claim that each of the structural functors in this colimit is 𝑡-exact for the 𝑡-
structures from Step 2. For 𝐾 1 Ď 𝐾 Ď 𝐻 compact open subgroups, we have a commutative
diagram of forgetful functors:

𝐴#𝑈 phq–mod𝐾 / 𝐴#𝑈 phq–mod𝐾 1

𝑟𝑒𝑛 𝑟𝑒𝑛

 
𝐴–mod𝐾,𝑤 / 𝐴–mod𝐾 1 ,𝑤 .
𝑟𝑒𝑛 𝑟𝑒𝑛

The vertical functors are 𝑡-exact and conservative by construction, and the bottom functor is clearly
𝑡-exact, so the claim follows.
Therefore, our (filtered) colimit inherits a canonical 𝑡-structure such that each functor 𝐴#𝑈 phq–mod𝐾
𝑟𝑒𝑛 Ñ
𝐴#𝑈 phq–mod𝑟𝑒𝑛 is 𝑡-exact. Let us show that this 𝑡-structure has the desired properties from (2).
It is clear from the construction that this 𝑡-structure is compactly generated and that compact
objects are eventually coconnective.
For 𝐾 any compact open subgroup of 𝐻, the functor:

𝐴#𝑈 phq–mod𝐾 𝐾,𝑤

𝑟𝑒𝑛 Ñ 𝐴–mod𝑟𝑒𝑛 Ñ 𝐴–mod𝑟𝑒𝑛 Ñ Vect
calculates the forgetful functor. The first functor in this sequence is 𝑡-exact and conservative by
design, while the remaining functors are 𝑡-exact and conservative on eventually coconnective sub-
categories by assumption. It remains to show this conservativeness survives passage to the colimit
in 𝐾.
First, suppose 𝐾 is a fixed prounipotent compact open. Observe that 𝐴#𝑈 phq–mod𝐾,♡ 𝑟𝑒𝑛 admits an
explicit description: it is the abelian category of discrete 𝐻 0 p𝐴q-modules equipped with a suitably
compatible smooth h-action such that k “ Liep𝐾q acts locally nilpotently. Indeed, this follows from
the description of the category as modules over a certain monad on 𝐴–mod𝐾,𝑤,♡ 𝑟𝑒𝑛 .
♡
In particular, the abelian category 𝐴#𝑈 phq–mod𝐾,♡ 𝑟𝑒𝑛 Ď 𝐴#𝑈 phq–mod𝑟𝑒𝑛 is closed under taking
subobjects.
It follows that for F P 𝐴#𝑈 phq–mod♡ 𝑟𝑒𝑛 , the map F Ñ Oblv Av˚ pFq induces a monomorphism
on 𝐻 0 ; here e.g. Av˚ indicates the functor of strong 𝐾-averaging. For such F, note that:

F“ colim Oblv Av𝐾

˚ pFq “ colim 𝐻 0 Oblv Av𝐾
˚ pFq
𝐾Ď𝐻 compact open 𝐾Ď𝐻 compact open

with each structural map in the latter colimit a monomorphism in 𝐴#𝑈 phq–mod♡ 𝑟𝑒𝑛 . Therefore, if
F is non-zero, 𝐻 0 pAv𝐾˚ pFqq is non-zero for some 𝐾. Now the desired conservativeness follows from
𝐾
the similar result for 𝐴#𝑈 phq–mod𝑟𝑒𝑛 .
Step 4. We now mildly generalize our earlier constructions.
Let 𝐾 be as before, and let 𝐾0 Ď 𝐾 be a normal compact open subgroup. We will study the
category:

p𝐴#𝑈 phq–mod𝐾 0 𝐾{𝐾0 ,𝑤

𝑟𝑒𝑛 q .
118 SAM RASKIN

^ ,𝑤
𝐻𝐾
As before, this category identifies with p𝐴–mod𝑟𝑒𝑛0 q𝐾{𝐾0 ,𝑤 . There is a canonical forgetful func-
tor:
^ ,𝑤
𝐻𝐾
p𝐴#𝑈 phq–mod𝐾 0 𝐾{𝐾0 ,𝑤
𝑟𝑒𝑛 q “ p𝐴–mod𝑟𝑒𝑛0 q𝐾{𝐾0 ,𝑤 Ñ p𝐴–mod𝐾 0 ,𝑤 𝐾{𝐾0 ,𝑤
𝑟𝑒𝑛 q “ 𝐴–mod𝐾,𝑤
𝑟𝑒𝑛

which is again monadic. Clearly the corresponding monad on 𝐴–mod𝐾,𝑤 𝑟𝑒𝑛 is again 𝑡-exact.
𝐾0 𝐾{𝐾0 ,𝑤
Therefore, we once again find that p𝐴#𝑈 phq–mod𝑟𝑒𝑛 q is again compactly generated, and
that it admits a unique 𝑡-structure for which the forgetful functor to 𝐴–mod𝐾,𝑤𝑟𝑒𝑛 is 𝑡-exact.

Step 5. We now show (4) from the statement of the lemma.

By Lemma 10.11.1, and using the notation of loc. cit., we have:
𝐾
𝐴#𝑈 phq–mod𝐾,𝑤 𝑗 𝐾{𝐾𝑗 ,𝑤
𝑟𝑒𝑛 “ colim p𝐴#𝑈 phq–mod𝑟𝑒𝑛 q P DGCat𝑐𝑜𝑛𝑡 .
𝑗
Each of the structural functors in this colimit admits a continuous right adjoint. Therefore,
compact generation of the colimit follows from compact generation of each term, which we have
shown in Step 4.
Step 6. Let us be in the general setup of S10.10, with 𝐻 acting naively on C, which is equipped
with a 𝑡-structure. Suppose that the 𝑡-structure on C is compactly generated.
We claim that the 𝑡-structure on C is compatible with the naive 𝐻-action if and only if for every
F P Coh˚ p𝐻qď0 and G P Cď0 compact, 𝛼C pF b Gq is connective.
Indeed, this condition clearly implies 𝛼C is right 𝑡-exact. It suffices to show that p𝛼C q´1 is similarly
right 𝑡-exact. Note that 𝛼C´1 is obtained by conjugating 𝛼C by the automorphism invIndCoh ˚ b idC for
» IndCoh
inv : 𝐻 Ý Ñ 𝐻 the inversion map. As this automorphism is 𝑡-exact (since inv˚ clearly is), we
obtain the result.
Step 7. We will show that the 𝑡-structure on C “ 𝐴#𝑈 phq–mod𝑟𝑒𝑛 is compatible with the naive
𝐻-action using the criterion of Step 6. Let 𝛼 “ 𝛼C in what follows.
^ ,𝑤
𝐾Ñ𝐻𝐾
Let 𝐾 Ď 𝐻 be a fixed compact open subgroup and let Av! “ Av! : 𝐴–mod𝐾,𝑤𝑟𝑒𝑛 Ñ
𝐴#𝑈 phq–mod𝑟𝑒𝑛 denote the left adjoint to the forgetful functor. Let G0 P 𝐴–mod𝑟𝑒𝑛
𝐾 𝐾,𝑤,ď0
be compact
and let G :“ Av! pG0 q. Note that objects of this form compactly generate 𝐴#𝑈 phq–modď0 𝑟𝑒𝑛 (letting
𝐾 vary, of course).
Let F P Cohp𝐻q♡ . By Step 6 and the above remarks, it suffices to show that 𝛼pFbGq is connective.
We will show this in Step 9 after some preliminary constructions.
Let us just note one special case. Suppose F “ 𝑘ℎ is the skyscraper sheaf at a 𝑘-point ℎ of 𝐻.
Ad p𝐾qÑ𝐻𝐾 ^ ,𝑤
Ad p𝐾q,𝑤,ď0
Then the object in question is Av! ℎ p𝑘ℎ ‹ G0 q. Clearly 𝑘ℎ ‹ G0 P 𝐴–mod𝑟𝑒𝑛ℎ , so the
same is true after !-averaging.
The general argument is similar, but has additional complications due to working in families.
Step 8. We need some constructions involving standard Chevalley-style constructions.
Let 𝐾 be any classical affine group scheme and let 𝐾0 Ď 𝐾 be a compact open subgroup. Suppose
C is equipped with a genuine 𝐾-action.
𝐾 ^ ,𝑤
Let H P C 𝐾0 be given. We will construct a canonical Chevalley filtration on H, which is
^ ,𝑤
𝐾0 Ñ𝐾𝐾
an increasing filtration with gr𝑖 H “ Av! 0
pΛ𝑖 pk{k0 q ‹ OblvpHqqr𝑖s. For clarity: the notation
means we forget H down to C 𝐾 0 ,𝑤 𝑖
, act by Λ pk{k0 qr𝑖s P Repp𝐾0 q, then !-average back.
^
Indeed, there is an action of the symmetric monoidal category Repp𝐾𝐾 ^ q “ k–mod𝐾0 on C𝐾𝐾0 .
0
The trivial representation 𝑘 P k–mod𝐾0 has a standard filtration with gr𝑖 p𝑘q “ indkk0 pΛ𝑖 pk{k0 qqr𝑖s
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 119

(see [GR4] SIV.5.2 for a much more general context for such constructions). As 𝑘 is the unit for the
monoidal structure here, we obtain a filtration of the desired type by functoriality.
We will actually need a more general, parametrized version of this construction. We sketch the
ideas below.
Suppose 𝑆 is an affine scheme. Let K be a compact 92 group scheme over 𝑆, meaning an affine
group scheme over 𝑆 that can be realized as a projective limit under smooth surjective structure
maps of affine group schemes that are smooth over 𝑆.
Let K0 Ď K be an (𝑆-family of) compact open subgroups of K, meaning K0 is compact in the
above sense and we are given K0 Ñ K a homomorphism of group schemes over 𝑆 that is a closed
embedding such that K{K0 is smooth over 𝑆 (in particular, of finite presentation over 𝑆).
For such K, there is a symmetric monoidal category ReppKq, defined as the evident colimit as in
the case where 𝑆 is a point. For example, Repp𝐾 ˆ 𝑆q “ Repp𝐾q b QCohp𝑆q for 𝐾 a classical affine
group scheme. Note that ReppKq is a QCohp𝑆q-module category, is a symmetric monoidal category
over QCohp𝑆q.
Let C be a genuine 𝐾-category, meaning we are given C𝐾,𝑤 a Repp𝐾q-module category. We obtain
K
C 0 ,𝑤 :“ C𝐾,𝑤 bReppKq ReppK0 q.
The ideas of S7 translate into this setting in an evident way. In particular, we a parametrized
category of Harish-Chandra modules ReppK^ K0 q, which is equipped with a monadic forgetful functor
K^ ,𝑤
to ReppK0 q. This allows us to make sense of C K0 , again by tensoring.
In particular, the construction of Chevalley filtrations goes through in this setting.
Step 9. We now conclude the argument. We use the notation from Step 7.
Let 𝑆 Ď 𝐻 be a classical affine subscheme on which F is scheme-theoretically supported.
In what follows, we use a subscript 𝑆 to indicate a product with 𝑆. For example, 𝐻𝑆 “ 𝐻 ˆ 𝑆.
Let K be the group scheme 𝐾 ˆ 𝑆 over 𝑆; in what follows, we always regard K as a (family of)
compact open subgroup(s) of 𝐻𝑆 via the map:

p𝑘,ℎqÞÑpAdℎ p𝑘q,ℎq
K “ 𝐾 ˆ 𝑆 ÝÝÝÝÝÝÝÝÝÝÝÑ 𝐻 ˆ 𝑆 “ 𝐻𝑆 .
By a standard Noetherian descent argument, there exists a compact open subgroup 𝐾0 Ď 𝐻
such that 𝐾0,𝑆 Ď K Ď 𝐻𝑆 . Note that K{𝐾0,𝑆 is smooth over 𝑆 in this case.
Let 𝐴–mod𝑟𝑒𝑛,𝑆 :“ QCohp𝑆q b 𝐴–mod𝑟𝑒𝑛 . We similarly have 𝐴–mod𝐾 𝑆 ,𝑤 𝐾,𝑤
𝑟𝑒𝑛,𝑆 “ 𝐴–mod𝑟𝑒𝑛 b QCohp𝑆q
and 𝐴–modK,𝑤
𝑟𝑒𝑛 , with 𝛼𝐴–mod𝑟𝑒𝑛 inducing an isomorphism:

𝐴–mod𝐾 𝑆 ,𝑤
Ñ 𝐴–modK,𝑤
»
𝑟𝑒𝑛,𝑆 Ý 𝑟𝑒𝑛,𝑆 .

Regarding F as a coherent sheaf on 𝑆, F bG0 P 𝐴–mod𝐾,𝑤

𝑟𝑒𝑛,𝑆 by definition. Let H :“ 𝛼𝐴–mod𝑟𝑒𝑛 pF b
G0 q P 𝐴–modK,𝑤
𝑟𝑒𝑛,𝑆 .
We clearly have:
^ ,𝑤
𝐾Ñ𝐻𝐾 KÑ𝐻𝐾,𝑆
^ ,𝑤
𝛼pF b Av! pG0 qq “ Av! pHq
where the left hand side is what we wish to show is connective.
Applying Step 8, we obtain an increasing filtration on H with:
𝐾0,𝑆 ÑK^
𝐾 ,𝑤
gr𝑖 H “ Av! 0,𝑆
pΛ𝑖 pLiepKq{ Liep𝐾0,𝑆 qq ‹ OblvpHqqr𝑖s.

92The terminology is admittedly bad: it is meant to evoke compact open, nothing about properness.
120 SAM RASKIN

We remark that LiepKq{ Liep𝐾0,𝑆 q is a finite-rank vector bundle on 𝑆. As in the previous step,
OblvpHq indicates that we forget down to weak 𝐾0,𝑆 -invariants.
KÑ𝐻𝐾,𝑆
^ ,𝑤 ^ ,𝑤
𝐻𝐾
Therefore, Av! pHq inherits a filtration in IndCoh˚ p𝐻q b 𝐴–mod𝑟𝑒𝑛0 with 𝑖th associated
graded term:

^ ,𝑤
𝐾0 Ñ𝐻𝐾 ^ ,𝑤
𝐻𝐾
pid b Av! 0
qpΛ𝑖 pLiepKq{ Liep𝐾0,𝑆 qq ‹ OblvpHqqr𝑖s P QCohp𝑆q b 𝐴–mod𝑟𝑒𝑛0 . (10.14.1)
𝐾 ,𝑤
Now observe that OblvpHq P 𝐴–mod𝑟𝑒𝑛,𝑆 0,𝑆
is connective: it suffices93 to check this after applying
the forgetful functor to 𝐴–mod𝑟𝑒𝑛,𝑆 where it is clear.
This clearly that implies (10.14.1) is connective (for the tensor product 𝑡-structure), giving our
claim.
Step 10. To complete the proof of (3), it remains to show that the 𝑡-structure on 𝐴#𝑈 phq–mod𝑟𝑒𝑛
is strongly compatible with the 𝐻-action.
Fortunately, this is evident from our constructions: the 𝑡-structure on 𝐴#𝑈 phq–mod𝑟𝑒𝑛 was de-
fined so that:
𝐻 ^ ,𝑤
𝐴#𝑈 phq–mod𝐾 𝐾
𝑟𝑒𝑛 “ 𝐴–mod𝑟𝑒𝑛 Ñ 𝐴#𝑈 phq–mod𝑟𝑒𝑛
is 𝑡-exact for every compact open subgroup 𝐾.
Step 11. We now show (5).
The analysis of Step 4 clearly shows that for any 𝐾0 Ď 𝐾 a normal compact open subgroup, the
𝑡-structure on p𝐴#𝑈 phq–mod𝐾 0 𝐾{𝐾0 ,𝑤
𝑟𝑒𝑛 q is compactly generated and compact objects are bounded
from below. By construction, this implies the same for 𝐴#𝑈 phq–mod𝐾,𝑤 𝑟𝑒𝑛 .
To show that 𝐴#𝑈 phq–mod𝐾,𝑤 𝑟𝑒𝑛 Ñ Repp𝐾q is 𝑡-exact, we claim that it suffices to show this for its
𝐾0 𝐾{𝐾0 ,𝑤
restriction to p𝐴#𝑈 phq–mod𝑟𝑒𝑛 q . Indeed, right 𝑡-exactness of this functor is evident (as the
forgetful functor 𝐴#𝑈 phq–mod𝑟𝑒𝑛 Ñ Vect is 𝑡-exact and all our forgetful functors are conservative
on eventually coconnective subcategories).
For left 𝑡-exactness, note that if F P 𝐴#𝑈 phq–mod𝐾,𝑤 𝑟𝑒𝑛 , then:

F “ colim 𝐿𝑗 pFq
𝑗

where 𝐿𝑗 is the comonad on 𝐴#𝑈 phq–mod𝐾,𝑤

𝑟𝑒𝑛 defined by the adjunction:

𝐾
𝑗 𝐾{𝐾𝑗 ,𝑤
p𝐴#𝑈 phq–mod𝑟𝑒𝑛 q Õ 𝐴#𝑈 phq–mod𝐾,𝑤
𝑟𝑒𝑛
using the notation of Lemma 10.11.1. Because the left adjoint in this adjunction is 𝑡-exact, 𝐿𝑗 is
left 𝑡-exact. As this colimit is filtered, we obtain the claim.
Now observe that the composition:
^
𝐻𝐾
p𝐴#𝑈 phq–mod𝐾 0 𝐾{𝐾0 ,𝑤
𝑟𝑒𝑛 q “ p𝐴–mod𝑟𝑒𝑛0 q𝐾{𝐾0 ,𝑤 Ñ 𝐴–mod𝐾,𝑤
𝑟𝑒𝑛 Ñ Repp𝐾q
calculates the forgetful functor in question. The second arrow is 𝑡-exact by assumption on 𝐴. The
^
𝐻𝐾
same holds for the first arrow is 𝑡-exact because 𝐴–mod𝑟𝑒𝑛0 Ñ 𝐴–mod𝐾 0 ,𝑤
𝑟𝑒𝑛 is 𝑡-exact by construction,
and forgetting from 𝐾-invariants to 𝐾0 -invariants is 𝑡-exact and conservative (because 𝐾{𝐾0 is
finite type).
93Unlike coconnectivity.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 121

Step 12. Finally, it remains to show (6).

First, note that the functor:

𝐵#𝑈 phq–mod𝐾 𝐾
𝑟𝑒𝑛 Ñ 𝐴#𝑈 phq–mod𝑟𝑒𝑛
is 𝑡-exact. Indeed, we have a commutative diagram:

𝐵#𝑈 phq–mod𝐾 / 𝐴#𝑈 phq–mod𝐾

𝑟𝑒𝑛 𝑟𝑒𝑛

 
𝐵–mod𝐾,𝑤 / 𝐴–mod𝐾,𝑤
𝐻 ^ ,𝑤
where, as always, the vertical arrows are given by rewriting e.g. 𝐴#𝑈 phq–mod𝐾 𝐾
𝑟𝑒𝑛 as 𝐴–mod𝑟𝑒𝑛 .
These vertical arrows are conservative and 𝑡-exact by construction, and the bottom horizontal arrow
Ñ
b,𝐻ñ
is 𝑡-exact by definition of morphism in Alg𝑔𝑒𝑛 .
More generally, we find that for any 𝐾0 Ď 𝐾 a compact open normal subgroup, the functor:

p𝐵#𝑈 phq–mod𝐾 0 𝐾{𝐾0 ,𝑤

𝑟𝑒𝑛 q Ñ p𝐴#𝑈 phq–mod𝐾 0 𝐾{𝐾0 ,𝑤
𝑟𝑒𝑛 q
is 𝑡-exact. We now conclude the result by the same logic as in Step 11.

10.15. Some results on canonical renormalization. We now develop some general results on
canonical renormalization in the setting of Tate group indschemes. The ultimate result is Corollary
10.16.3, which gives a convenient way of checking the hypotheses for canonical renormalization.
Proposition 10.15.1. Suppose that 𝐻 is a Tate group indscheme with a prounipotent tail.
Suppose C P 𝐻–mod is acted on strongly by 𝐻 and equipped with a 𝑡-structure strongly compatible
with the 𝐻-action.
Suppose that for every compact open subgroup 𝐾 Ď 𝐻, C𝐾,𝑤 is compactly generated with compact
objects lying in C𝐾,𝑤,` (with respect to the 𝑡-structure of S10.13).
Then for every compact open subgroup 𝐾 Ď 𝐻, an object F P C𝐾,𝑤 is compact if and only if
F P C𝐾,𝑤,` and OblvpFq is compact in C.
Remark 10.15.2. We remark that the result is clearly about 𝐾, and that 𝐻 is a bit of a red herring.
Also, note that the conclusion of the lemma may be stated as C P 𝐾–mod𝑤𝑒𝑎𝑘 is obtained by
canonical renormalization (in the sense of S5.18) from the underlying naive weak 𝐾-action on C.
Proof of Proposition 10.15.1. Clearly compact objects satisfy this property, as Oblv admits the
continuous right adjoint Av𝑤 ˚ . Therefore, suppose F P C
𝐾,𝑤,` with OblvpFq compact; we wish to

show that F is compact.

Choose 𝑁 " 0 such that F P C𝐾,𝑤,ě´𝑁 . Note that F is compact in C𝐾,𝑤,ě´𝑁 : see Step 3 from
the proof of Lemma 6.11.2.
Write F as a filtered colimit colim𝑖 F𝑖 with F𝑖 P C𝐾,𝑤 compact. As the 𝑡-structure is compatible
with filtered colimits, we obtain F “ colim𝑖 𝜏 ě´𝑁 F𝑖 . Because F is compact in C𝐾,𝑤,ě´𝑁 , we see
that F is a summand of 𝜏 ě´𝑁 F𝑖 for some 𝑖.
By Lemma 10.11.1 (and in the notation of loc. cit.), the map:

colim pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤,𝑐 Ñ C𝐾,𝑤,𝑐 P DGCat

𝑗
is an equivalence, where 𝐾𝑗 runs over compact open prounipotent subgroups of 𝐻 that are normal
in 𝐾; as is usual, the superscript p´q𝑐 indicates the subcategory of compact objects. Therefore, F𝑖
122 SAM RASKIN

lifts to pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤,𝑐 for some index 𝑗. As the forgetful functor pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 Ñ C𝐾,𝑤 is 𝑡-exact, the
same is true of F itself. We abuse notation in letting F also denote a lift to pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 .
As the forgetful functor pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 Ñ C𝐾,𝑤 admits a continuous right adjoint, it suffices to
show that F is compact as an object of pC𝐾𝑗 q𝐾{𝐾𝑗 ,𝑤 . Moreover, by Lemma 10.15.3, it suffices to
show that F is compact after forgetting to C𝐾𝑗 . As 𝐾𝑗 is prounipotent by assumption, the forgetful
functor C𝐾𝑗 Ñ C is fully-faithful, giving the claim.

Above, we used the following result.
Lemma 10.15.3. Let 𝐻 be an affine algebraic group (in particular, of finite type) acting weakly
on C. Then F P C𝐻,𝑤 is compact if and only if OblvpFq P C is compact.
Proof. Clearly if F P C𝐻,𝑤 is compact, then OblvpFq P C is compact. Suppose the converse.
Recall from the proof of Lemma 5.20.2 that for G P C𝐻,𝑤 , G is a summand of Totď𝑛 pAv𝑤
˚ Oblvq
‚`1 pGq

for some 𝑛; moreover, this is functorial in G by the construction of loc. cit.

Therefore, the functor HomC𝐻,𝑤 pF, ´q : C𝐻,𝑤 Ñ Vect is a summand of:

Totď𝑛 HomC pOblvpFq, OblvpAv𝑤 ‚

˚ Oblvq p´q.
A summand of a finite limit of continuous functors is continuous, giving the claim.

10.16. Next, we show the following result.
Proposition 10.16.1. Suppose that 𝐻 is a polarizable Tate group indscheme with a prounipotent
tail. Suppose that C P 𝐻–mod is equipped with a compactly generated 𝑡-structure strongly compatible
with the action of 𝐻 on C.
𝑤,ě0
Then for every compact open subgroup 𝐾 Ď 𝐻 and every F P H𝐻,𝐾 , the functor F ‹ ´ : C𝐾,𝑤 Ñ
C𝐾,𝑤 is left 𝑡-exact.
Proof.
Step 1. First, we claim that the conclusion of the for a compact open subgroup 𝐾 is equivalent
to asking that F P H𝐻,𝐾𝑤 compact acts on C𝐾,𝑤 by a functor that is left 𝑡-exact up to shift. This
property is clearly weaker than that in the statement of the proposition, so suppose it is satisfied.
We remind that Cohp𝐾z𝐻{𝐾q Ď H𝐻,𝐾 𝑤 is the subcategory of compact objects and is closed under
truncations. Therefore, it suffices to show that for F P Cohp𝐾z𝐻{𝐾qě0 , the functor F ‹ ´ : C𝐾,𝑤 Ñ
𝑤,ě0
C𝐾,𝑤 is left 𝑡-exact. Indeed, any object of H𝐻,𝐾 is a filtered colimit of such coconnective compact
objects. Fix F P Cohp𝐾z𝐻{𝐾q in what follows.
ě0

Under our assumption, F ‹ ´ maps C𝐾,𝑤,ě0 into C𝐾,𝑤,` . Therefore, by Lemma 10.13.1, it suffices
to show that the composition:
F‹´ Oblv
C𝐾,𝑤 ÝÝÑ C𝐾,𝑤 ÝÝÝÑ C
is left 𝑡-exact. Moreover, we can clearly replace C𝐾,𝑤 with C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 here. Then the corresponding
functor may be calculated as the composition:
𝛼C
Fb´ »
C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 ÝÝÝÑ IndCohp𝐻{𝐾q b C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ pIndCohp𝐻q b Cq𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 ÝÑ
ΓIndCoh p𝐻{𝐾,´q
IndCohp𝐻{𝐾q b C ÝÝÝÝÝÝÝÝÝÑ C.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 123

There are some things to explain in the above manipulations: we are regarding F P IndCohp𝐻{𝐾q by
forgetting the left 𝐾-equivariance; in the third term, the 𝐾-equivariance is taken for the diagonal
𝐾-action mixing the given action on C and the right action of 𝐻 on C; and the implicit commuting
of weak 𝐾-equivariance with the tensor product by C in the fourth term is justified by the fact that
C is assumed compactly generated and therefore is dualizable (or one may use Lemma 8.6.1). The
first functor is left 𝑡-exact by assumption on F; the second is clearly 𝑡-exact; the third is 𝑡-exact
because the naive 𝐻-action is compatible with 𝑡-structures; and the fourth by Lemma 4.6.2 (2),
using that ΓIndCoh p𝐻{𝐾, ´q is left 𝑡-exact. This gives the claim.
Step 2. Next, we check the above hypothesis in the case where 𝐾 is a polarization of 𝐻. In fact,
a little more generally, we will show that if 𝐾 is a polarization and 𝐾0 Ď 𝐻 is any other compact
open subgroup, then for any F P Cohp𝐾0 z𝐻{𝐾q, the functor:

F ‹ ´ : C𝐾0 ,𝑤 Ñ C𝐾,𝑤
is left 𝑡-exact up to shift.
Let G P Cohp𝐾z𝐻{𝐾0 q be obtained by applying Serre duality on 𝐻{𝐾 to F (considered with
»
its natural 𝐾0 -equivariant structure) and then pulling back along the inversion map 𝐾z𝐻{𝐾0 Ý Ñ
𝐾0 z𝐻{𝐾.
As is standard, G ‹ ´ : C𝐾,𝑤 Ñ C𝐾0 ,𝑤 is left adjoint to F ‹ ´ (by ind-properness of 𝐻{𝐾).
Therefore, it suffices to show G ‹ ´ is right 𝑡-exact up to shift.
This is straightforward: it suffices to show the composite with C𝐾0 ,𝑤 Ñ C is right 𝑡-exact up to
shift by Lemma 10.13.1, and this follows by a similar (in fact, simpler) argument to Step 1, using
that G is supported on a finite type subscheme of 𝐻{𝐾.
Step 3. Next, suppose 𝐾 is a compact open subgroup of 𝐻 that admits an embedding 𝐾 Ď 𝐾𝑝𝑜𝑙 Ď 𝐻
with 𝐾𝑝𝑜𝑙 a polarization of 𝐻 and 𝐾 normal in 𝐾𝑝𝑜𝑙 (so 𝐾𝑝𝑜𝑙 {𝐾 is an affine algebraic group). We
will prove the result for 𝐾 in this case.
𝑤,ě0
Let F P H𝐻,𝐾 and G P C𝐾,𝑤,ě0 be given. We need to show that F ‹ G P C𝐾,𝑤,ě0 .
˚ : C
As the functor Av𝑤 𝐾,𝑤 Ñ C𝐾𝑝𝑜𝑙 ,𝑤 of averaging from 𝐾 to 𝐾
𝑝𝑜𝑙 is conservative and 𝑡-exact
(by the normality assumption), it suffices to show that Av𝑤 ˚ pF ‹ Gq P C𝐾𝑝𝑜𝑙 ,𝑤,ě0 .
This term may clearly be calculated by averaging F P H𝐻,𝐾𝑤 on the left to obtain Fr P IndCoh𝑟𝑒𝑛 p𝐾𝑝𝑜𝑙 z𝐻{𝐾qě0 ,
and then convolving with F. r By the previous step, that object is eventually coconnective, and by
Step 1 it is honestly coconnective.
Step 4. Finally, we prove the claim for 𝐾 a general compact open subgroup of 𝐻.
By the previous step, there exists a compact open subgroup 𝐾0 Ď 𝐾 for which the conclusion of
the proposition holds.
𝑤,ě0
We again let F P H𝐻,𝐾 and G P C𝐾,𝑤,ě0 denote given objects, and we aim to show that their
convolution is eventually coconnective.
By the proof of Lemma 5.20.2, G is a direct summand of Totď𝑛 pAv𝑤 ˚ Oblvq
‚`1 pGq for some 𝑛;

here our functors denote the adjoint pair Oblv : C𝐾,𝑤 Õ C𝐾0 ,𝑤 : Av𝑤 ˚ . Each term in this finite limit
lies in Av𝑤 pC 𝐾0 ,𝑤,ě0 q, so we may assume G “ Av𝑤 pG q for G P C𝐾0 ,𝑤,ě0 .
˚ ˚ 0 0
It suffices to check that OblvpF ‹ Gq “ OblvpF ‹ Av𝑤 ˚ pG0 qq P C
𝐾0 ,𝑤 is eventually coconnective,

since Oblv : C 𝐾,𝑤 Ñ C 0 is conservative.

𝐾 ,𝑤

But the above object may be calculated by mapping F along the functor H𝐻,𝐾 𝑤 Ñ H𝐻,𝐾𝑤
0
of
forgetting equivariance on both sides and then convolving with G0 ; by assumption on 𝐾0 , this
object is coconnective as desired.

124 SAM RASKIN

Remark 10.16.2. The careful reader will see that we never really used the hypothesis that the 𝐻-
action on C is strong. Here are the actually relevant hypotheses. First, we need a genuine 𝐻-action
on C and a 𝑡-structure on C naively compatible with the 𝐻-action. In addition, for every compact
open subgroup 𝐾 Ď 𝐻, we need a 𝑡-structure on C𝐾,𝑤 for which C𝐾,𝑤 Ñ C𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 is 𝑡-exact and an
equivalence on eventually coconnective subcategories. Finally, we need that for 𝐾1 Ď 𝐾2 compact
open subgroups, C𝐾2 ,𝑤 Ñ C𝐾1 ,𝑤 is 𝑡-exact.
Corollary 10.16.3. Suppose 𝐻 is polarizable with a prounipotent tail. Suppose C P 𝐻–mod is
equipped with a 𝑡-structure strongly compatible with the weak 𝐻-action. Suppose that for every
𝐾 Ď 𝐻 compact open, C𝐾,𝑤 is compactly generated by objects lying in C𝐾,𝑤,` X C𝐾,𝑤,ď0 .
Then the naive weak action of 𝐻 on C canonically renormalizes, and Oblv𝑠𝑡𝑟Ñ𝑤 pCq P 𝐻–mod𝑤𝑒𝑎𝑘
is its canonical renormalization.
Proof. Immediate from the definition of canonical renormalization and from Proposition 10.15.1
and Proposition 10.16.1 (and Example 8.16.5 as applied to A𝑟𝑒𝑛 “ H𝐻,𝐾
𝑤 ).


10.17. Conclusion. We now combine the various ingredients above to conclude the proof of The-
orem 10.8.1.
By Lemma 10.14.1 (1), 𝐴#𝑈 phq–mod𝑟𝑒𝑛 is compactly generated. Moreover, this category has a
canonical compactly generated 𝑡-structure by Lemma 10.14.1 (2). The forgetful functor 𝐴#𝑈 phq–mod𝑟𝑒𝑛 Ñ
Vect from S10.8 is 𝑡-exact and conservative on eventually coconnective objects by Lemma 10.14.1
Ñ Ñ
b
(3). Therefore, by Remark 4.2.4, there is a corresponding connective b-algebra 𝐴#𝑈 phq P Alg𝑟𝑒𝑛 .
Moreover, the naive action of 𝐻 on 𝐴#𝑈 phq–mod𝑟𝑒𝑛 , its naive compatibility with the 𝑡-structure
(Lemma 10.14.1 (3)), and the naive 𝐻-equivariance of the forgetful functor 𝐴#𝑈 phq–mod𝑟𝑒𝑛 Ñ Vect
define a naive 𝐻-action on 𝐴#𝑈 phq compatible with its renormalization datum. Indeed, this follows
from Remark 4.2.4 (c.f. the end of the proof of Theorem 10.5.1). This upgrades 𝐴#𝑈 phq to an object
Ñ
b,𝐻ñ
of Alg𝑟𝑒𝑛 . We claim that this action is genuine.
First, we need to show that the genuine 𝐻-action on 𝐴#𝑈 phq–mod𝑟𝑒𝑛 is obtained by canonical
renormalization. We do this by applying Corollary 10.16.3 to C “ 𝐴#𝑈 phq–mod𝑟𝑒𝑛 . We check that
the various conditions from that corollary are satisfied.
By construction, the naive weak 𝐻-action on 𝐴#𝑈 phq–mod𝑟𝑒𝑛 upgrades to a strong action.
By Lemma 10.14.1 (4) and (5), 𝐴#𝑈 phq–mod𝐾,𝑤 𝑟𝑒𝑛 is compactly generated, and its 𝑡-structure is
as well, and compact objects are eventually coconnective.
Therefore, the corollary applies, and we find that the 𝐻-action on 𝐴#𝑈 phq is nearly genuine (in
the sense of S10.4). It is genuine by Lemma 10.14.1 (5).
10.18. Harish-Chandra data. In the remainder of this section, we suppose that 𝐻 is a formally
smooth polarizable ind-affine Tate group indscheme with prounipotent tail. In particular, Theorem
10.8.1 applies.
10.19. The reader may prefer to skip this material and refer back to it as needed.
Let C be a category, and suppose 𝑇 : C Ñ C is a comonad. Let F P C be a fixed object. We claim
that 𝑇 canonically upgrades to a comonad on the overcategory C{F .
For G P C{F , we have the map 𝑇 pGq Ñ G Ñ F where the first map is the counit for 𝑇 and the
second map is the structure map for G; this makes 𝑇 pGq into an object of C{F . We denote this
functor by 𝑇{F : C{F Ñ C{F .
We claim 𝑇{F has a natural comonad structure. Consider 𝑇 –comod{F , the category of 𝑇 -comodules
G in C equipped with a map 𝛼 : G Ñ F P C (with no hypotheses on how 𝛼 interacts with the
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 125

comodule structure). The forgetful functor 𝑇 –comod{F Ñ C{F is obviously conservative; we claim
that it is actually comonadic. Indeed, C{F Ñ C clearly commutes with contractible limits,94, so the
claim follows from Barr-Beck. It is clear the underlying endofunctor of this comonad on C{F is given
by 𝑇{F .
Note that by construction, the data of a 𝑇{F -comodule structure on G P C{F is equivalent to a
𝑇 -comodule structure on the underlying object G P C.
10.20. We apply the above for the comonad Oblv𝑠𝑡𝑟Ñ𝑤 ˝p´qexpphq,𝑤 : 𝐻–mod𝑤𝑒𝑎𝑘 Ñ 𝐻–mod𝑤𝑒𝑎𝑘
and Vect P 𝐻–mod𝑤𝑒𝑎𝑘 . This defines a comonad on 𝐻–mod𝑤𝑒𝑎𝑘,{Vect .
By Theorem 10.8.1 and Lemma 10.14.1 (6), this comonad preserves the 1-full subcategory:
Ñ 𝑇 ℎ𝑚.10.5.1
b,𝐻ñ
pAlg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 q𝑜𝑝 Ď p𝐻–mod𝑤𝑒𝑎𝑘 q{Vect .
Ñ
b,𝐻ñ
This induces a monad on Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 , which we denote by 𝐴 ÞÑ 𝐴#𝑈 phq.
We can now make the following definition.
Ñ
b,𝐻ñ
Definition 10.20.1. A Harish-Chandra datum for 𝐴 P Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 is a structure of module for the
above monad.
Remark 10.20.2. By definition, a Harish-Chandra datum gives rise to an “action” map 𝐴#𝑈 phq Ñ
𝐴.
10.21. We now make the following observation.
Lemma 10.21.1. The functor Oblv𝑠𝑡𝑟Ñ𝑤 : 𝐻–mod Ñ 𝐻–mod𝑤𝑒𝑎𝑘 is comonadic.
Proof. This functor is conservative as the composition:
Oblv𝑔𝑒𝑛
𝐻–mod Ñ 𝐻–mod𝑤𝑒𝑎𝑘 ÝÝÝÝÝÑ DGCat𝑐𝑜𝑛𝑡
computes the forgetful functor for 𝐻–mod, which is conservative.
To conclude, we simply note that as 𝐻 is polarizable, Oblv𝑠𝑡𝑟Ñ𝑤 admits a left adjoint by Propo-
sition 8.21.1 and therefore commutes with all limits.

Ñ
b,𝐻ñ
Therefore, by the discussion of S10.19, a Harish-Chandra datum for 𝐴 P Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 is equivalent
to upgrading the genuine 𝐻-action on 𝐴–mod𝑟𝑒𝑛 to a strong 𝐻-action with the property95 that the
expphq,𝑤
coaction functor 𝐴–mod𝑟𝑒𝑛 Ñ Oblv𝑠𝑡𝑟Ñ𝑤 p𝐴–mod𝑟𝑒𝑛 q “ 𝐴#𝑈 phq–mod𝑟𝑒𝑛 come from a genuinely
𝐻-equivariant morphism 𝐴#𝑈 phq Ñ 𝐴.
Ñ Ñ
b,𝐻ñ b,𝐻ñ
More precisely, let Alg𝐻𝐶 denote the category of 𝐴 P Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 equipped with a Harish-
Chandra datum (i.e., the category of modules for the monad 𝐴 ÞÑ 𝐴#𝑈 phq). Then we have:
Lemma 10.21.2. The above functor:
Ñ
b,𝐻ñ 𝑜𝑝 𝐴ÞÑ𝐴–mod
pAlg𝐻𝐶 q ÝÝÝÝÝÝÝÝ𝑟𝑒𝑛
ÝÑ 𝐻–mod ˆ 𝐻–mod𝑤𝑒𝑎𝑘,{Vect
𝐻–mod𝑤𝑒𝑎𝑘
is 1-fully-faithful.
94See [Lur2] Proposition [Link] for a complete proof.
Ñ
95This encodes the fact that 𝐴–mod b,𝐻ñ 𝑜𝑝
𝑟𝑒𝑛 is a comodule in the 1-full subcategory pAlg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 q Ďp𝐻–mod𝑤𝑒𝑎𝑘 q{Vect ,
i.e., it has to do with the fact that this is a 1-full subcategory and not an actual subcategory.
126 SAM RASKIN

Ñ
10.22. The classical case. Let 𝐴 be a classical, renormalized b-algebra equipped with a genuine
𝐻-action. As in Theorem 10.8.1, we can form the smash product 𝐴#𝑈 phq. We wish to explicitly
♡
describe the category 𝐴#𝑈 phq–mod♡ 𝑟𝑒𝑛 p“ 𝐴#𝑈 phq–mod𝑛𝑎𝑖𝑣𝑒 q.
Suppose 𝑉 is an object of this abelian category. The canonical map 𝐴 Ñ 𝐴#𝑈 phq (coming as
the unit for the monad structure) makes 𝑉 into a (discrete) 𝐻 0 p𝐴q-module. Also, the fact that the
unit map 𝑘 Ñ 𝐴 is 𝐻-equivariant gives a map 𝑘#𝑈 phq “ 𝑈 phq Ñ 𝐴#𝑈 phq, so 𝑉 also acquires an
h-module structure.
To describe the compatibility between these two actions, we need the following digression. Any
𝜉 P h defines a derivation 𝛿𝜉 : 𝐴 Ñ 𝐴. In detail, 𝜉 defines a homomorphism Funp𝐻q Ñ 𝑘r𝜀s{𝜀2
extending the augmentation on Funp𝐻q, so we obtain a map:

! ! Ñ
coact
𝐴 ÝÝÝÑ 𝐴 b Funp𝐻q Ñ 𝐴 b 𝑘r𝜀s{𝜀2 P Alg b
giving id𝐴 mod 𝜀. If we write the underlying map of pro-vector spaces as id𝐴 ˆ𝛿𝜉 𝜀, the map 𝛿𝜉 :
𝐴 Ñ 𝐴 P ProVect♡ is by definition our derivation.
Then we claim that the difference between the two maps:

Ñ act p𝜉¨´q
𝐴 b 𝑉 ÝÝÑ 𝑉 ÝÝÝÑ 𝑉,
Ñ (10.22.1)
Ñ id bp𝜉¨´q Ñ act
𝐴 b 𝑉 ÝÝÝÝÝÝÑ 𝐴 b 𝑉 ÝÝÑ 𝑉
is:
Ñ
Ñ 𝛿𝜉 bid𝑉 Ñ act
𝐴 b 𝑉 ÝÝÝÝÑ 𝐴 b 𝑉 ÝÝÑ 𝑉. (10.22.2)
96
More symbolically:

𝜉 ¨ 𝑓 ¨ 𝑣 ´ 𝑓 ¨ 𝜉 ¨ 𝑣 “ 𝛿𝜉 p𝑓 q ¨ 𝑣, 𝜉 P h, 𝑓 P 𝐴, 𝑣 P 𝑉.
Indeed, for 𝐾 Ď 𝐻 compact open, we have the canonical equivalence:

𝐻 ^ ,𝑤,♡
𝐴#𝑈 phq–mod𝐾,♡ 𝐾
𝑟𝑒𝑛 “ 𝐴–mod𝑟𝑒𝑛 “ 𝐴–mod𝐻 𝐾 ,𝑤,𝑛𝑎𝑖𝑣𝑒,♡
𝑛𝑎𝑖𝑣𝑒 .
This verifies the above identity for 𝑉 strongly 𝐾-equivariant. Every object in 𝐴#𝑈 phq–mod𝐾,♡𝑟𝑒𝑛 is
a filtered colimit of objects strongly equivariant for some congruence subgroup, so we obtain the
claim in general.
In addition, this same logic implies the converse. That is, we have the following result.
Proposition 10.22.1. For 𝑉 P Vect♡ , lifting 𝑉 to an object of 𝐴#𝑈 phq–mod♡ 𝑟𝑒𝑛 is equivalent via
the above constructions to specifying an action of 𝐴 on 𝑉 and an action of h on 𝑉 such that the
difference between the two maps in (10.22.1) is (10.22.2) for any 𝜉 P h.
We then obtain:
96This latter formula is only sufficient when 𝐴 is a topological vector space. By definition, this means that there is
a (non-derived) topological vector space 𝐴„ with a complete, separated, linear topology such that 𝐴 is the associated
pro-vector space, i.e., 𝐴 “ lim 𝐴„ {𝑈 P ProVect where 𝑈 runs over open subspaces of 𝐴„ . In this case, it is typically
sufficient to work with elements of 𝐴.
Note that h is necessarily a topological vector space, justifying working directly with its elements in some of these
formulae.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 127

Ñ Ñ
Corollary 10.22.2. Suppose 𝐵 P Alg b is classical. Then specifying a map 𝐴#𝑈 phq Ñ 𝐵 P Alg b
Ñ
is equivalent to giving maps 𝜙 : 𝐴 Ñ 𝐵 P Alg b and 𝑖 : h Ñ 𝐵 compatible with brackets97 such for
any 𝜉 P h, the difference between the two maps:
𝜙 𝑖p𝜉q¨´
𝐴Ý
Ñ 𝐵 ÝÝÝÑ 𝐵
𝜙 ´¨𝑖p𝜉q
𝐴Ý
Ñ 𝐵 ÝÝÝÑ 𝐵
is the composition:
𝛿𝜉 𝜙
𝐴 ÝÑ 𝐴 Ý
Ñ 𝐵.
Proof. Immediate from Proposition 3.7.1 (and [Lur3] Theorem [Link]).

10.23. In the above setting, we now show:
Lemma 10.23.1. Suppose that for any compact open subgroup 𝐾 Ď 𝐻, 𝐾 is the spectrum of a
countably generated ring. Then 𝐴#𝑈 phq is classical.
Proof. By Proposition 3.7.1, it suffices to show 𝐴#𝑈 phq–mod` 𝑟𝑒𝑛 is the bounded derived category of
its underlying abelian category. By [Ras6] Lemma 5.4.3 and our countability assumption, it suffices
^ ,𝑤,`
𝐻𝐾
to show that for any compact open subgroup 𝐾 Ď 𝐻, 𝐴#𝑈 phq–mod𝐾,` 𝑟𝑒𝑛 “ 𝐴–mod𝑟𝑒𝑛 is the
bounded derived category of its underlying abelian category. This category admits a monadic, 𝑡-
exact restriction functor to 𝐴–mod𝐾,𝑤,`
𝑟𝑒𝑛 , and the corresponding monad is 𝑡-exact; so Lemma 9.11.2
𝐾,𝑤,`
reduces to showing that 𝐴–mod𝑟𝑒𝑛 is the bounded derived category of its heart. As this category
is comonadic over 𝐴–mod` 𝑟𝑒𝑛 with 𝑡-exact comonad, that claim follows from the similar one for
𝐴–mod` 𝑟𝑒𝑛 .

Remark 10.23.2. It seems likely that the above result is true without the countability hypothesis.
By this lemma (and Yoneda), Corollary 10.22.2 gives a complete description of 𝐴#𝑈 phq. In
particular, using the 1-categorical nature of our setup, we find the following result.
Corollary 10.23.3. Under the above assumptions, a Harish-Chandra datum for 𝐴 (in the sense
of S10.20) is equivalent to specifying an 𝐻-equivariant map 𝑖 : h Ñ 𝐴 compatible with brackets
such that for 𝜉 P h, r𝑖p𝜉q, ´s “ 𝛿𝜉 as maps 𝐴 Ñ 𝐴 P ProVect♡ , and such that the induced (naively
𝐻-equivariant) map:

𝐴#𝑈 phq Ñ 𝐴
is genuinely 𝐻-equivariant.
Remark 10.23.4. The last condition in the corollary can be difficult to check in practice. There is
one simple case though: if compact objects in 𝐴–mod𝑟𝑒𝑛 are closed under truncations, then for any
Ñ
b,𝐻ñ
𝐵 P Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 , any naively 𝐻-equivariant morphism 𝐵 Ñ 𝐴 is genuinely 𝐻-equivariant. Indeed,
this follows from the definition of canonical renormalization and from the discussion of S4.4.
We record the following consequence of the above discussion for later reference.
97If 𝐵 is a topological vector space, then in the language of [Bei], we would say 𝑖 is a map of topological Lie
algebras.
128 SAM RASKIN

Ñ
b,𝐻ñ Ñ
Corollary 10.23.5. Suppose 𝐴, 𝐵 P Alg𝐻𝐶 are classical b-algebras equipped with genuine 𝐻-
Ñ
b,𝐻ñ
actions. Then giving a morphism 𝑓 : 𝐴 Ñ 𝐵 P Alg𝐻𝐶 is equivalent to giving a genuinely 𝐻-
Ñ
b,𝐻ñ
equivariant morphism 𝑓 : 𝐴 Ñ 𝐵 P Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 such that the diagram:

𝑓 
𝐴 /𝐵

commutes in ProVect♡ , where the diagonal morphisms encode the Harish-Chandra data as above.

11. Application to the critical level

11.1. In this section, we prove Theorem 11.18.1, providing a large class of symmetries for Kac-
Moody representations at critical level. We also show a compatibility result, Theorem 11.19.1, with
our previous work [Ras6]. The arguments are straightforward applications of the methods developed
in S10.

11.2. Let us describe the contents of this section in more detail.

Let 𝐺 be a split reductive group over 𝑘. Let98 𝐾 :“ 𝑘pp𝑡qq.
Recall that for any Ad-invariant symmetric bilinear form 𝜅 on g, we have the corresponding
Kac-Moody central extension 0 Ñ 𝑘 Ñ p g𝜅 Ñ gpp𝑡qq. As we will discuss in S11.6, 𝜅 defines a twisted
notion of strong 𝐺p𝐾q-actions; we denote the corresponding category by 𝐺p𝐾q–mod𝜅 . The theory
is barely different from the untwisted one. A basic object is p g𝜅 –mod P 𝐺p𝐾q–mod𝜅 .
Let 𝑐𝑟𝑖𝑡 :“ ´ 12 𝜅g , for 𝜅g the Killing form on g.
Let Op𝐺ˇ denote the indscheme of 𝐺-opersˇ on the punctured disc; we take as the definition of
opers what are called marked opers in [BTCZ] SA. (which is a slight modification of the definition
in [BD1]).
Our goal for this section is to construct a coaction of IndCoh˚ pOp𝐺ˇ q on pg𝑐𝑟𝑖𝑡 –mod P 𝐺p𝐾q–mod𝑐𝑟𝑖𝑡
via the Feigin-Frenkel isomorphism. In other words, we wish to show that in a suitable sense,
g𝑐𝑟𝑖𝑡 –mod is tensored over its center compatibly with the (critical level) strong 𝐺p𝐾q-action on it.
p
This result appears as Theorem 11.18.1.

11.3. Central extensions. We begin by generalizing some material from S8 in the presence of
central extensions and twisted 𝐷-modules.
We outline the main ideas and leave the verification that certain constructions generalize to the
reader.

11.4. Fix 𝑐 P 𝑘 once and for all; we refer to 𝑐 as the twisting parameter. Let BG𝑚 denote the
Zariski sheafified version of the classifying space.
Observe that AffSch𝑐𝑙 {BG𝑚 of classical affine schemes equipped with a line bundle embeds as a
full subcategory of PropAffSch𝑓.𝑡.,{BG𝑚 q, the pro-category of (classically) finite type classical affine
schemes with a line bundle; this follows by standard Noetherian approximation (specifically, [Gro]
Theorem 8.5.2). Then the procedure from [Ras3] S2 produces functors:

98We abuse notation in letting 𝐾 denote both Laurent series and compact open subgroups of 𝐺p𝐾q. This abuse
should never cause confusion, and we prefer it to various alternatives.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 129

𝐷𝑐˚ : PreStk{BG𝑚 Ñ DGCat𝑐𝑜𝑛𝑡

𝐷𝑐! : PreStk𝑜𝑝
{BG𝑚 Ñ DGCat𝑐𝑜𝑛𝑡 .

These functors are given by suitable Kan extensions from the finite type setup (c.f. loc. cit.), and
for 𝑆 affine, finite type, and equipped with a line bundle L, they each assign to 𝑆 the category of
pL, 𝑐q-twisted 𝐷-modules on 𝑆 (as defined e.g. in [GR2] S5).
Remark 11.4.1. We generally omit the line bundle from the notation since it can usually be taken
for granted.
As in S6.20, there is a canonical natural transformation:

IndCoh˚ Ñ 𝐷𝑐˚ P HompIndSch𝑟𝑒𝑎𝑠,{BG𝑚 , DGCat𝑐𝑜𝑛𝑡 q

defined by a formal extension process from the finite type case.
11.5. Because BG𝑚 is a commutative group, there is a canonical symmetric monoidal structure
on PreStk{BG𝑚 for which the forgetful functor to PreStk is symmetric monoidal for the Cartesian
monoidal product.
Explicitly, for p𝑆, L𝑆 q and p𝑇, L𝑇 q in PreStk{BG𝑚 , 𝑆 ˆ𝑇 is equipped with the line bundle L𝑆 bL𝑇 .
Then each of the functors 𝐷𝑐˚ and 𝐷𝑐! are naturally lax symmetric monoidal for this symmet-
ric monoidal structure, meaning that we have external products in either setup. Indeed, this lax
symmetric monoidal structure arises from the finite type setup by Kan extension. As in Remark
6.20.1, the natural transformation IndCoh˚ Ñ 𝐷𝑐˚ canonically upgrades (via our same extension
procedure) to a natural transformation of lax symmetric monoidal functors.
11.6. Now let 𝐻 be a Tate group indscheme and let 𝜆 : 𝐻 Ñ BG𝑚 be a homomorphism; equiva-
lently, we have a central extension:

1 Ñ G𝑚 Ñ 𝐻 1 Ñ 𝐻 Ñ 1.
We assume that there exists a compact open subgroup 𝐾 Ď 𝐻 on which 𝜆 is trivial as a homomor-
phism. Then note that 𝐻 1 is also a Tate group indscheme. If 𝐾 can be taken to be a polarization
of 𝐻, then 𝐻 1 is polarizable.
We obtain the category 𝐷𝑐˚ p𝐻q of twisted 𝐷-modules on 𝐻 for the underlying line bundle defined
by 𝜆. Because p𝐻, 𝜆q is an algebra object in PreStk{BG𝑚 (for the symmetric monoidal structure
described above), 𝐷𝑐˚ p𝐻q is canonically a monoidal DG category, i.e., an algebra object in DGCat𝑐𝑜𝑛𝑡 .
We let 𝐻–mod𝑐 denote the category of modules for 𝐷𝑐˚ p𝐻q in DGCat𝑐𝑜𝑛𝑡 , and we refer to these
as DG categories equipped with (strong) 𝑐-twisted 𝐻-actions.
11.7. Before proceeding, we will need the following digression on twisted invariants and coinvari-
ants. 1 ,𝑤
Suppose C P 𝐻–mod𝑤𝑒𝑎𝑘 . As in S9.9, we have a certain full subcategory C𝐻 p1q Ď C
𝐻𝑇 𝑎𝑡𝑒 ,𝑤

𝜆
p exp
Ñ BG^
Now suppose that 𝜆 factors as 𝐻 Ý 𝑎 ÝÝÑ BG𝑚 . In this case, we can define a new homo-
morphism:

𝜆𝑐 :“ expp𝑐 ¨ 𝜆q
p : 𝐻 Ñ BG𝑚 .
130 SAM RASKIN

1
𝐻 ,𝑤
We let C𝐻,𝑤,𝜒𝑐 denote the corresponding category of twisted invariants, i.e., we take C𝐻,𝑤,𝜒𝑐 :“ Cp1q𝑐
for 𝐻𝑐1 the central extension defined by 𝜆𝑐 .
This construction can also be understood as follows. We obtain a homomorphism IndCoh˚ p𝐻q Ñ
IndCohpBG^ 1
𝑎 q » QCohpA q where the right hand side is equipped with its usual tensor product
structure (as opposed to convolution). Our twisting parameter 𝑐 P 𝑘 defines a homomorphism
QCohpA1 q Ñ Vect (taking the fiber at99 ´𝑐), so an object 𝜒𝑐 P 𝐻–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 . It is easy to
see that this naive weak 𝐻-action canonically renormalizes, defining 𝜒𝑐 P 𝐻–mod𝑤𝑒𝑎𝑘 . Then by
Proposition 9.9.1, C𝐻,𝑤,𝜒𝑐 “ pC b 𝜒𝑐 q𝐻,𝑤 .
Similarly, we have a twisted coinvariants functor C𝐻,𝑤,𝜒𝑐 , defined by tensoring with 𝜒𝑐 and taking
coinvariants.

11.8. Combining the material of S11.4-11.5 with the methods of S8, we obtain a forgetful functor:

Oblv𝑠𝑡𝑟Ñ𝑤 : 𝐻–mod𝑐 Ñ 𝐻–mod𝑤𝑒𝑎𝑘 .

This functor admits left and right adjoints, which we denote by p´qexpphq,𝑤,𝜒𝑐 and p´qexpphq,𝑤,𝜒𝑐 .
There are explicit formulae for these functors, similar to S8.3. Before giving them, suppose that
𝐾 Ď 𝐻 is a compact open subgroup on which 𝜆 is trivial. In this case, 𝜆|𝐻𝐾 ^ clearly factors through

log
»
BG^ Ñ BG^
𝑚 Ý 𝑎 , so the discussion of S11.7 applies.
Now we have:
^
Cexpphq,𝑤 :“ colim C𝐻𝐾 ,𝑤,𝜒𝑐 P DGCat𝑐𝑜𝑛𝑡
𝐾Ď𝐻 compact open
𝜆 trivial on K
Cexpphq,𝑤 :“ lim C𝐻 𝐾
^ ,𝑤,𝜒 P DGCat𝑐𝑜𝑛𝑡 .
𝑐
𝐾Ď𝐻 compact open
𝜆 trivial on K
under the obvious structural functors.
Proposition 8.21.1 has an immediate counterpart in this setting: simply change pexpphq, 𝑤q-
invariants and coinvariants in loc. cit. to the corresponding 𝑐-twisted versions.

11.9. Now suppose that 𝐻 is formally smooth. Let h1 denote the Lie algebra of 𝐻 1 , considered as
a central extension of the Tate Lie algebra h by 𝑘. Let h1𝑐 denote the central extension obtained by
Baer-scaling by our twisting parameter 𝑐 P 𝑘.
We let h1𝑐 –mod :“ Vectexpphq,𝑤,𝑐 .
Note that the notation is abusive: this category should be understood not as modules of the
abstract Tate Lie algebra h1𝑐 , but as modules over the central extension, i.e., modules on which
1 P 𝑘 Ď h1𝑐 acts by the identity (in a suitable derived sense).
An evident version of Proposition 9.13.1 applies; this shows that h1𝑐 –mod has a natural 𝑡-structure
with the expected heart, and is the “renormalized” DG category of representations considered in
[FG2] S23.
Remark 11.9.1. By Proposition 8.21.1 (or its appropriate version here), h1𝑐 –mod is dualizable with
dual h1´𝑐´𝑇 𝑎𝑡𝑒 –mod, where the notation indicates the Baer sum of the inverse central extension to
h1𝑐 and h´𝑇 𝑎𝑡𝑒 . By Theorem 9.16.1, the pairing:

h1𝑐 –mod b h1´𝑐´𝑇 𝑎𝑡𝑒 –mod Ñ Vect

99The minus sign here matches the sign conventions of S9.

 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 131

is given on eventually coconnective objects by tensoring and applying semi-infinite cohomology.

Note that the equivalence ph1´𝑐´𝑇 𝑎𝑡𝑒 –modq_ » h1𝑐 –mod is of categories acted on by 𝐻 strongly with
𝑐-twist.100
11.10. Critical level. We now apply the above to 𝐻 “ 𝐺p𝐾q the loop group. Let 𝐺p𝑂q Ď 𝐺p𝐾q
be the subgroup, which is a compact open subgroup.
Let 𝜆 : 𝐺p𝐾q Ñ BG𝑚 be the map defining the Tate central extension of 𝐺p𝐾q, as in S9. Let
𝑐 “ ´ 21 above. Apply Theorem 9.16.1 relative to the compact open subgroup 𝐾 “ 𝐺p𝑂q Ď 𝐺p𝐾q
and [BD2] S2.7.5, the corresponding central extension is the critical level Kac-Moody extension.
Note that this Tate central extension is canonically trivialized over 𝐺p𝑂q.
We let 𝐺p𝐾q–mod𝑐𝑟𝑖𝑡 denote 𝐻–mod𝑐 and p g𝑐𝑟𝑖𝑡 –mod denote h1𝑐 –mod in this setting. We let 𝑈 ppg𝑐𝑟𝑖𝑡 q
Ñ
`
denote the twisted enveloping algebra, i.e., the b-algebra defined by Oblv : p g𝑐𝑟𝑖𝑡 –mod Ñ Vect (in
1
[BD1], it is usually denoted 𝑈 pg b 𝐾q).
g𝑐𝑟𝑖𝑡 –mod♡ denote the vacuum representation indggrr𝑡ss
p𝑐𝑟𝑖𝑡
We let V𝑐𝑟𝑖𝑡 P p p𝑘q. More generally, for 𝑛 ě 0,
we let V𝑐𝑟𝑖𝑡,𝑛 “ indg𝑡𝑛𝑐𝑟𝑖𝑡 ♡
grr𝑡ss p𝑘q P g𝑐𝑟𝑖𝑡 –mod . By construction, these objects compactly generated
p
p
g𝑐𝑟𝑖𝑡 –mod.
p

Remark 11.10.1. The above clearly applies as is to define 𝐺p𝐾q–mod𝜅 , the category of categories
with level 𝜅 (strong) 𝐺p𝐾q-actions, 𝜅 any scalar multiple of the Killing form. A simple modification
allows for arbitrary levels 𝜅 (for possibly non-simple 𝐺): see [Ras6] S1.29-30.
!
11.11. The center. Define Z P ComAlgpProVect♡ , bq as the non-derived center of 𝑈 pp g𝑐𝑟𝑖𝑡 q.
More precisely, Z is the pro-vector space corresponding to the the topological vector space
𝐻 0 p𝑈 pp g𝑐𝑟𝑖𝑡 q;101 the topology is the subspace topol-
g𝑐𝑟𝑖𝑡 q𝐺p𝐾q q for the adjoint action of 𝐺p𝐾q on 𝑈 pp
ogy. It is straightforward to see that the multiplication on Z is commutative, and therefore its
Ñ !
evident b-algebra structure extends canonically to a commutative b-algebra structure (c.f. [Bei]
S1.5).
In what follows, we treat Z and 𝑈 pg𝑐𝑟𝑖𝑡 q interchangeably as pro-vector spaces and as topological
vector spaces, following our custom from S

11.12. We now recall the finer structure of Z.

Define 𝐼𝑛 Ď Z as the ideal Z X 𝑈 pg𝑐𝑟𝑖𝑡 q ¨ 𝑡𝑛 grr𝑡ss Ď 𝑈 pg𝑐𝑟𝑖𝑡 q. By construction of the topology of
Z, this ideal is open in Z, and the ideals 𝐼𝑛 as 𝑛 varies provide a neighborhood basis of 0. We let
z𝑛 “ Z{𝐼𝑛 .102
We let Z𝑛 “ Specpz𝑛 q and define Z “ colim𝑛 ś Z𝑛 P IndSch (so Z “ SpfpZq).
By [BD1] S3.7.9-10, Z is103 isomorphic to 𝑟𝑖“1 A1 p𝐾q where 𝑟 is the rank of g. Moreover,
loc.
ś𝑟 cit.´𝑑constructs such an isomorphism such that Z𝑛 Ď Z corresponds to the closed subscheme
𝑖 A1 p𝑂q Ď
ś 𝑟 1
𝑖“1 𝑡 𝑖“1 A p𝐾q for 𝑡 our coordinate on the formal disc and 𝑑1 , . . . , 𝑑𝑟 the degrees
of the invariant polynomials on g.

100A priori, ph1 _

´𝑐´𝑇 𝑎𝑡𝑒 –modq may look like it is acted on strongly with twist 𝑐 ` 𝑇 𝑎𝑡𝑒. But since the Tate extension
is by definition integral, we can canonically identify 𝐻–mod𝑐`𝑇 𝑎𝑡𝑒 with 𝐻–mod𝑐 (though not compatibly with the
forgetful functors to 𝐻–mod𝑤𝑒𝑎𝑘 !).
101See [BD1] Theorem 3.7.7 (ii). In fact, Z is the non-derived invariants of 𝑈 pp
g𝑐𝑟𝑖𝑡 q with respect to any subgroup
of 𝐺p𝐾q containing a compact open subgroup; see [BD1] 3.7.11. We will not need this fact here.
102The notation follows [BD1], which denotes our z by z.
0
103Mildly non-canonically: there are various choices needed to obtain these coordinates.
132 SAM RASKIN

In particular, Z is a reasonable indscheme. Therefore, Z–mod𝑟𝑒𝑛 :“ IndCoh˚ pZq defines a renor-

malization datum for Z; the forgetful functor to Vect is ΓIndCoh pZ, ´q. (See also Example 4.3.5.) In
what follows, we always consider Z as equipped with this renormalization datum.
Remark 11.12.1. Using the isomorphism above, one can construct an equivalence QCohpZq »
morphism of QCohpZq-module categories sending OZ to the ˚-pullback
IndCoh˚ pZq: this is the unique ś
of 𝜔 𝑖“1 A1 p𝐾q{A1 p𝑂q P IndCohp 𝑟𝑖“1 A1 p𝐾q{A1 p𝑂qq along the evident projection (using the additive
ś 𝑟

structure on A1 to form the quotient). However, this construction is highly non-canonical and does
not behave well with respect to the changes of coordinates (which are non-linear).
11.13. Being the center of 𝑈 pg𝑐𝑟𝑖𝑡 q, Z acts on 𝑈 pg𝑐𝑟𝑖𝑡 q.
More precisely, we can consider Z as a commutative algebra object in the symmetric monoidal
Ñ !
category pAlg b , bq; then 𝑈 pg𝑐𝑟𝑖𝑡 q is a module for it in the usual sense of monoidal categories. For
! Ñ
example, the action map act : Z b 𝑈 pg𝑐𝑟𝑖𝑡 q Ñ 𝑈 pg𝑐𝑟𝑖𝑡 q P Alg b sends 𝑧 b 𝜉 ÞÑ 𝑧 ¨ 𝜉.
Lemma 11.13.1. The morphism act is compatible with renormalization data, i.e., it upgrades
Ñ
b
(necessarily uniquely) to a morphism in Alg𝑟𝑒𝑛 .
Proof. Let:
!
g𝑐𝑟𝑖𝑡 –mod`
coact𝑛𝑎𝑖𝑣𝑒 : p `
𝑛𝑎𝑖𝑣𝑒 Ñ Z b 𝑈 pg𝑐𝑟𝑖𝑡 q–mod𝑛𝑎𝑖𝑣𝑒
denote the standard forgetful functor.104 Define:
!
coact𝐿 `
g𝑐𝑟𝑖𝑡 –mod`
𝑛𝑎𝑖𝑣𝑒 : Z b 𝑈 pg𝑐𝑟𝑖𝑡 q–mod𝑛𝑎𝑖𝑣𝑒 Ñ Propp 𝑛𝑎𝑖𝑣𝑒 q
as its pro-left adjoint.
Let 𝑛, 𝑚 ě 0. We claim that:

coact𝐿 g𝑐𝑟𝑖𝑡 –mod`

𝑛𝑎𝑖𝑣𝑒 pz𝑛 b V𝑐𝑟𝑖𝑡,𝑚 q “ lim z𝑛 b V𝑐𝑟𝑖𝑡,𝑚 P Propp 𝑛𝑎𝑖𝑣𝑒 q. (11.13.1)
𝑟ě𝑛,𝑚 z𝑟
Here the limit is formed in this pro-category, and the terms make sense because the action of Z on
V𝑐𝑟𝑖𝑡,𝑚 factors through z𝑚 by definition. We emphasize that this is a derived tensor product, i.e.,
we view z𝑛 b V𝑐𝑟𝑖𝑡,𝑚 as an z𝑟 -bimodule in g𝑐𝑟𝑖𝑡 –mod𝑛𝑎𝑖𝑣𝑒 and then form its Hochschild homology
in this category; because the kernel of z𝑟 Ñ z𝑛 is generated by a regular sequence (by S11.12), this
tensor product does honestly lie in p g𝑐𝑟𝑖𝑡 –mod` 𝑛𝑎𝑖𝑣𝑒 . We note that this reasoning also shows that this
𝑐
object actually lies in p
g𝑐𝑟𝑖𝑡 –mod Ď p g𝑐𝑟𝑖𝑡 –mod` .
Indeed, there is a canonical map from the left hand side of (11.13.1) to the right hand side.
It suffices to show this map is an isomorphism when evaluated on objects in the heart of the 𝑡-
structure. We can explicitly calculate both sides using Koszul and Chevalley complexes, giving the
claim.
!
Now because the objects z𝑛 b V𝑐𝑟𝑖𝑡,𝑚 compactly generate Z b 𝑈 pg𝑐𝑟𝑖𝑡 q–modď0 𝑟𝑒𝑛 by definition (see
S4.6), the fact that coact𝐿 pz
𝑛𝑎𝑖𝑣𝑒 𝑛 b V 𝑐𝑟𝑖𝑡,𝑚 q is pro-(compact and connective) immediately implies
that coact𝑛𝑎𝑖𝑣𝑒 renormalizes (i.e., is left 𝑡-exact), i.e., it gives the conclusion of the lemma.

Ñ Ñ
b b
Therefore, we find that 𝑈 pg𝑐𝑟𝑖𝑡 q P Alg𝑟𝑒𝑛 is canonically a module for Z P Alg𝑟𝑒𝑛 .
104The notation is motivated by Remark 11.13.2.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 133

Ñ
b,𝑜𝑝 𝐴ÞÑ𝐴–mod
Remark 11.13.2. Applying the symmetric monoidal functor Alg𝑟𝑒𝑛 ÝÝÝÝÝÝÝÝ𝑟𝑒𝑛
ÝÑ DGCat𝑐𝑜𝑛𝑡 , we
˚ 105
find that p
g𝑐𝑟𝑖𝑡 –mod is canonically a comodule for Z–mod𝑟𝑒𝑛 “ IndCoh pZq.
11.14. Next, we include 𝐺p𝐾q-actions.
Ñ
For 𝐻 an ind-affine group indscheme the category Alg b,𝐻ñ from S5.5. Note that this category is
Ñ !
canonically a module category for the (symmetric) monoidal category pAlg b , bq: this is a general
feature for co/module categories in monoidal categories. The same discussion applies verbatim in
Ñ
the setting of renormalized b-algebras.
Now take 𝐻 “ 𝐺p𝐾q. Recall that 𝐺p𝐾q has an adjoint action on 𝑈 pp g𝑐𝑟𝑖𝑡 q as a renormalized
Ñ
b-algebra encoding the naive, weak 𝐺p𝐾q-action on pg𝑐𝑟𝑖𝑡 –mod.
Ñ ! Ñ
The action of Z P ComAlgpAlg b , bq on 𝑈 pp
g𝑐𝑟𝑖𝑡 q P Alg b above clearly upgrades to an action in
Ñ
the symmetric monoidal category Alg b,𝐺p𝐾qñ . Moreover, tracing the definitions, Lemma 11.13.1
Ñ !
b
immediately implies that this action upgrades to an action of Z P ComAlgpAlg𝑟𝑒𝑛 , bq on 𝑈 pp
g𝑐𝑟𝑖𝑡 q P
Ñ
b
Alg𝑟𝑒𝑛 .
Remark 11.14.1. In parallel to Remark 11.13.2, the above discussion implies that IndCoh˚ pZq coacts
on pg–mod𝑐𝑟𝑖𝑡 P 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘,𝑛𝑎𝑖𝑣𝑒 , where the latter category is considered as a module category
for pDGCat𝑐𝑜𝑛𝑡 , bq.
11.15. We now extend the discussion to the setting of genuine actions.
Suppose now that 𝐻 is an ind-affine Tate group indscheme. Recall from S10.4 that we have the
Ñ Ñ
b,𝐻ñ b,𝐻ñ
1-full subcategory Alg𝑔𝑒𝑛 of Alg𝑟𝑒𝑛 . It is immediate from the definitions and Lemma 4.6.2 that
Ñ !
b
this 1-full subcategory is closed under the pAlg𝑟𝑒𝑛 , bq-action. Therefore, there is a unique action of
Ñ ! Ñ Ñ
b b,𝐻ñ b,𝐻ñ
pAlg𝑟𝑒𝑛 , bq on Alg𝑔𝑒𝑛 compatible with the embedding into Alg𝑟𝑒𝑛 .
Ñ Ñ !
b,𝐻ñ,𝑜𝑝 𝐴ÞÑ𝐴–mod b
We remark that the functor Alg𝑔𝑒𝑛 ÝÝÝÝÝÝÝÝ𝑟𝑒𝑛
ÝÑ 𝐻–mod𝑤𝑒𝑎𝑘 is pAlg𝑟𝑒𝑛 , bq-linear, where
Ñ !
b
pAlg𝑟𝑒𝑛 , bq acts on 𝐻–mod𝑤𝑒𝑎𝑘 through its canonical symmetric monoidal functor to DGCat𝑐𝑜𝑛𝑡 .
Ñ
b,𝐺p𝐾qñ
11.16. Recall from Theorem 10.8.1 that the 𝐺p𝐾q-action on 𝑈 pp
g𝑐𝑟𝑖𝑡 q P Alg𝑟𝑒𝑛 is genuine.
We now have the following upgraded version of Lemma 11.13.1.
! Ñ
b,𝐻ñ
Lemma 11.16.1. The morphism act : Z b 𝑈 pg𝑐𝑟𝑖𝑡 q Ñ 𝑈 pg𝑐𝑟𝑖𝑡 q P Alg𝑟𝑒𝑛 is genuinely 𝐻-
Ñ
b,𝐻ñ
equivariant, i.e., it is a morphism in Alg𝑔𝑒𝑛 .
g𝑐𝑟𝑖𝑡 –mod has commuting106 IndCoh˚ p𝐺p𝐾qq-module and Z–mod𝑟𝑒𝑛 -
Proof. As in Remark 11.14.1, p
comodule structures.
For 𝐾 Ď 𝐺p𝐾q a compact open subgroup, we need to show that the coaction functor:107

coact𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 : p
g𝑐𝑟𝑖𝑡 –mod𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 Ñ pIndCoh˚ pZqbp
g𝑐𝑟𝑖𝑡 –modq𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 “ IndCoh˚ pZqbp
g𝑐𝑟𝑖𝑡 –mod𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒
105In geometric terms, note that diagonal pushforward equips IndCoh˚ pZq with a canonical coalgebra structure in
DGCat𝑐𝑜𝑛𝑡 , using the fact that Z is a strict indscheme.
106In homotopically precise terms, we should say this category is an pIndCoh˚ p𝐺p𝐾qq, IndCoh! pZqq-bimodule, where
IndCoh! pZq :“ IndCoh˚ pZq_ as usual.
107We can commute the weak invariants with the tensor product as IndCoh˚ pZq is dualizable.
134 SAM RASKIN

renormalizes to a left 𝑡-exact functor:

coact𝐾,𝑤 : p
g𝑐𝑟𝑖𝑡 –mod𝐾,𝑤 Ñ IndCoh˚ pZq b p
g𝑐𝑟𝑖𝑡 –mod𝐾,𝑤
(where the right hand side is equipped with the usual tensor product 𝑡-structure).
As restriction from genuine weak 𝐾-invariants to invariants for a small compact open subgroup
above is 𝑡-exact and conservative, we can assume for simplicity that 𝐾 is prounipotent and contained
in 𝐺p𝑂q.
Now the argument is essentially the same as in Lemma 11.13.1. Consider z𝑛 bV𝑐𝑟𝑖𝑡,𝑚 P IndCoh˚ pZqb
˘ď0
g𝑐𝑟𝑖𝑡 –mod𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒,` . Note that these objects compactly generate IndCoh˚ pZq bpg𝑐𝑟𝑖𝑡 –mod𝐾,𝑤
`
p as
𝐾 is prounipotent. Moreover, the pro-left adjoint (on eventually coconnective subcategories):

coact𝐾,𝑤,`,𝐿 : pIndCoh˚ pZq b p

g𝑐𝑟𝑖𝑡 –mod𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒 q` Propp
g𝑐𝑟𝑖𝑡 –mod𝐾,𝑤,𝑛𝑎𝑖𝑣𝑒,` q
sends z𝑛 b V𝑐𝑟𝑖𝑡,𝑚 to lim z𝑛 b V𝑐𝑟𝑖𝑡,𝑚 (the limit being formed in the pro-category), which is
𝑟ě𝑛,𝑚 z𝑟
again seen using Koszul/Chevalley resolutions. As each z𝑛 b V𝑐𝑟𝑖𝑡,𝑚 is connective and compact in
z𝑟
g𝑐𝑟𝑖𝑡 –mod𝐾,𝑤 , we obtain the claim.
p

Ñ ! Ñ
b b
The above result implies that Z P ComAlgpAlg𝑟𝑒𝑛 , bq on 𝑈 pp
g𝑐𝑟𝑖𝑡 q P Alg𝑔𝑒𝑛 .
Remark 11.16.2. As with Remarks 11.13.2 and 11.14.1, the above discussion implies that IndCoh˚ pZq
coacts on p
g–mod𝑐𝑟𝑖𝑡 P 𝐺p𝐾q–mod𝑤𝑒𝑎𝑘 .
11.17. We now include Harish-Chandra data to extend to strong actions.
Suppose 𝐻 is a Tate group indscheme satisfying the hypotheses of Theorem 10.8.1.
Ñ
b,𝐻ñ
Recall that in S10.20, we defined a monad 𝐴 ÞÑ 𝐴#𝑈 phq on Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 . We claim that this monad
Ñ !
b
canonically upgrades to a pAlg𝑐𝑜𝑛𝑣,𝑟𝑒𝑛 , bq-linear monad, i.e., it canonically lifts along the forgetful
map:

´ Ñ ¯ ´ Ñ ¯ Ñ
b,𝐻ñ b,𝐻ñ b,𝐻ñ
Alg End Ñ
b
pAlg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 q Ñ Alg EndpAlg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 q “ tmonads on Alg𝑐𝑜𝑛𝑣,𝑔𝑒𝑛 u.
Alg𝑐𝑜𝑛𝑣,𝑟𝑒𝑛 –mod

Indeed, this follows immediately from the constructions and the observation that the functor
p´qexpphq,𝑤 is DGCat𝑐𝑜𝑛𝑡 -linear for 𝐻, as is clear from Proposition 8.21.1.
Ñ
b,𝐺p𝐾qñ Ñ
11.18. Let Alg𝐻𝐶,𝑐𝑟𝑖𝑡 be defined as the category of b-algebras with genuine 𝐺p𝐾q-actions and
Ñ
b,𝐺p𝐾qñ
critical level Harish-Chandra data (as in S10.21). By S11.17, Alg𝐻𝐶,𝑐𝑟𝑖𝑡 is canonically a module
Ñ !
b
category for pAlg𝑐𝑜𝑛𝑣,𝑟𝑒𝑛 , bq.
Ñ !
b
We claim that our earlier constructions upgrade to an action of Z P ComAlgpAlg𝑐𝑜𝑛𝑣,𝑟𝑒𝑛 , bq on
Ñ
b,𝐺p𝐾qñ
𝑈 pp
g𝑐𝑟𝑖𝑡 q P Alg𝐻𝐶,𝑐𝑟𝑖𝑡 (where 𝑈 pp
g𝑐𝑟𝑖𝑡 q is equipped with its tautological critical level Harish-Chandra
datum).
Ñ
Indeed, as all of the b-algebras here are classical, by Corollary 10.23.5, this amounts to the
evident commutativity of the diagram:
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 135

g𝑐𝑟𝑖𝑡
p
𝜉ÞÑ1b𝜉
y
! #
Z b 𝑈 pp
g𝑐𝑟𝑖𝑡 q
act / 𝑈 pp
g𝑐𝑟𝑖𝑡 q.
Therefore, we obtain:
Theorem 11.18.1. There is a canonical coaction of IndCoh˚ pZq on p g𝑐𝑟𝑖𝑡 –mod considered as an
object of the (DGCat𝑐𝑜𝑛𝑡 -enriched category) 𝐺p𝐾q–mod𝑐𝑟𝑖𝑡 .
Recalling the Feigin-Frenkel isomorphism Z » Op𝐺ˇ (see [BD1] S3.7 for this form of the isomor-
phism), we in particular obtain a coaction of IndCoh˚ pOp𝐺ˇ q on p
g𝑐𝑟𝑖𝑡 –mod P 𝐺p𝐾q–mod𝑐𝑟𝑖𝑡 .
11.19. A compatibility. We now establish a compatibility with a related result from [Ras6]. We
use the notation from loc. cit. without further mention.
Theorem 11.19.1. The equivalence:
»
Whitpp Ñ IndCoh˚ pOp𝐺ˇ q
g𝑐𝑟𝑖𝑡 –modq Ý
from [Ras6] Corollary 7.8.1 canonically upgrades to an equivalence of IndCoh˚ pOp𝐺ˇ q-comodules,
where the comodule structure on the left hand side comes from Theorem 11.18.1 and the comodule
structure on the right hand side is the tautological one.
Remark 11.19.2. Despite the appearance of the Langlands dual group in the statement, this is
essentially notational: we are just choosing to write Op𝐺ˇ instead of Z.
Proof of Theorem 11.19.1. In [Ras6] S5, we showed that Whitpp g𝑐𝑟𝑖𝑡 –modq is compactly generated,
and we equipped it with a certain canonical 𝑡-structure for which compact objects are bounded. By
loc. cit. Corollary 7.8.1, compact objects are even closed under truncation functors in Whitppg𝑐𝑟𝑖𝑡 –modq
(this is special to critical level).
Moreover, loc. cit. shows that the natural functor Ψ : Whitpp g𝑐𝑟𝑖𝑡 –modq Ñ Vect (induced by
Drinfeld-Sokolov reduction) is 𝑡-exact, and Ψ|Whitppg𝑐𝑟𝑖𝑡 –modq` is shown to be conservative.
Ñ »
By Proposition 3.7.1, there is a canonical convergent, connective b-algebra W𝑐𝑟𝑖𝑡 with Whitppg𝑐𝑟𝑖𝑡 –modq` Ý
Ñ
W𝑐𝑟𝑖𝑡 –mod . By [Ras6] S5, this algebra identifies with the usual critical level affine W-algebra (and
`

in particular, it is classical); this isomorphism is canonical, and has to do with the description of
the functor Ψ via Drinfeld-Sokolov reduction.
Now by Theorem 11.18.1 and functoriality, Whitpp g𝑐𝑟𝑖𝑡 –modq is canonically a IndCoh˚ pOp𝐺ˇ q-
comodule. It follows that the functor Ψ factors through a unique morphism:

Ψ𝑒𝑛ℎ / IndCoh˚ pOp ˇ q

Whitpp
g𝑐𝑟𝑖𝑡 –modq 𝐺
Ψ

' x ΓIndCoh pOp𝐺

ˇ ,´q
Vect
with Ψ𝑒𝑛ℎ a morphism of IndCoh pOp𝐺ˇ q-comodule categories. Note that Ψ𝑒𝑛ℎ is 𝑡-exact: as compact
˚

objects in Whitppg𝑐𝑟𝑖𝑡 –modq are closed under truncations, this follows as Ψ and ΓIndCoh pOp𝐺ˇ , ´q are
𝑡-exact and conservative on eventually coconnective subcategories.
Therefore, there is a canonical map 𝛼 : Z Ñ W𝑐𝑟𝑖𝑡 corresponding to the functor Ψ𝑒𝑛ℎ . By
construction, 𝛼 is the standard map arising by functoriality of Drinfeld-Sokolov reduction. As in
[FF], 𝛼 is an isomorphism. Moreover, the equivalence Z » FunpOp𝐺ˇ q from [BD1] S3.7 corresponds
136 SAM RASKIN

»
Ñ W𝑐𝑟𝑖𝑡 and the identification W𝑐𝑟𝑖𝑡 “ FunpOp𝐺ˇ q from Feigin-Frenkel
to the identification 𝛼 : Z Ý
duality [FF].
It follows that Ψ𝑒𝑛ℎ is an equivalence on bounded below subcategories. As compact objects on
both sides are exactly almost compact objects, Ψ𝑒𝑛ℎ is actually an equivalence. Moreover, by the
above discussion, Ψ𝑒𝑛ℎ is canonically isomorphic to the equivalence [Ras6] Corollary 7.8.1.
As Ψ𝑒𝑛ℎ was a morphism of IndCoh˚ pOp𝐺ˇ q-comodules by construction, we obtain the claim.


References
[AB] Maurice Auslander and David A Buchsbaum. Homological dimension in noetherian rings. Transactions of
the American Mathematical Society, 88(1):194–206, 1958.
[ABC` ] Dima Arinkin, Dario Beraldo, Justin Campbell, Lin Chen, Yuchen Fu, Dennis Gaitsgory, Quoc Ho, Sergey
Lysenko, Sam Raskin, Simon Riche, Nick Rozenblyum, James Tao, David Yang, and Yifei Zhao. Notes
from the winter school on local geometric Langlands, 2018. Available at [Link]
winterlanglands2018/notes-of-talks.
[AG1] Dima Arinkin and Dennis Gaitsgory. Singular support of coherent sheaves and the geometric Langlands
conjecture. Selecta Mathematica, 21(1):1–199, 2015.
[AG2] S. Arkhipov and D. Gaitsgory. Differential operators on the loop group via chiral algebras. Int. Math. Res.
Not., (4):165–210, 2002.
[AG3] S Arkhipov and D Gaitsgory. Localization and the long intertwining operator for representations of affine
Kac-Moody algebras. Preprint, [Link] math. harvard. edu/˜ gaitsgde/GL/Arkh. pdf, 2015.
[Ark1] Sergei Arkhipov. Semiinfinite cohomology of Tate Lie algebras. Moscow Mathematical Journal, 2(1):35–40,
2002.
[Ark2] Sergey Arkhipov. Algebraic construction of contragradient quasi-Verma modules in positive characteristic.
In Representation Theory of Algebraic Groups and Quantum Groups, pages 27–68. Mathematical Society of
Japan, 2004.
[Ati] MF Atiyah. Algebraic topology and operators in Hilbert space. In Lectures in Modern Analysis and appli-
cations I, pages 101–121. Springer, 1969.
[BBE] Alexander Beilinson, Spencer Bloch, and Hélène Esnault. 𝜀-factors for Gauss-Manin determinants. Mosc.
Math. J., 2(3):477–532, 2002. Dedicated to Yuri I. Manin on the occasion of his 65th birthday.
[BD1] Sasha Beilinson and Vladimir Drinfeld. Quantization of Hitchin’s integrable system and Hecke eigensheaves.
Available at [Link]
[BD2] Sasha Beilinson and Vladimir Drinfeld. Chiral algebras, volume 51 of American Mathematical Society Col-
loquium Publications. American Mathematical Society, Providence, RI, 2004.
[BDGH] Mathew Bullimore, Tudor Dimofte, Davide Gaiotto, and Justin Hilburn. Boundaries, mirror symmetry, and
symplectic duality in 3𝑑 N “ 4 gauge theory. Journal of High Energy Physics, 2016(10):108, 2016.
[Bei] Sasha Beilinson. Remarks on topological algebras. Moscow Math. Journ, 8(1):1–20, 2008.
[Ber] Dario Beraldo. Loop group actions on categories and Whittaker invariants. arXiv preprint arXiv:1310.5127,
2013.
[Bez] Roman Bezrukavnikov. On two geometric realizations of an affine Hecke algebra. Publ. Math. Inst. Hautes
Études Sci., 123:1–67, 2016.
[BF1] Roman Bezrukavnikov and Michael Finkelberg. Equivariant Satake category and Kostant–Whittaker reduc-
tion. Mosc. Math. J, 8(1):39–72, 2008.
[BF2] Alexander Braverman and Michael Finkelberg. Coulomb branches of 3-dimensional gauge theories and
related structures. In Geometric Representation Theory and Gauge Theory, pages 1–52. Springer, 2019.
[BFN] Alexander Braverman, Michael Finkelberg, and Hiraku Nakajima. Towards a mathematical definition of
Coulomb branches of 3-dimensional N “ 4 gauge theories, II. Advances in Theoretical and Mathematical
Physics, 22(5):1071–1147, 2018.
[BGS] Alexander Beilinson, Victor Ginzburg, and Wolfgang Soergel. Koszul duality patterns in representation
theory. Journal of the American Mathematical Society, 9(2):473–527, 1996.
[BLL` ] Christopher Beem, Madalena Lemos, Pedro Liendo, Wolfger Peelaers, Leonardo Rastelli, and Balt C van
Rees. Infinite chiral symmetry in four dimensions. Communications in Mathematical Physics, 336(3):1359–
1433, 2015.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 137

[Bot1] Raoul Bott. An application of the Morse theory to the topology of Lie-groups. Bulletin de la Société
Mathématique de France, 84:251–281, 1956.
[Bot2] Raoul Bott. The stable homotopy of the classical groups. Annals of Mathematics, pages 313–337, 1959.
[BTCZ] Roman Bezrukavnikov, Roman Travkin, Tsao-Hsien Chen, and Xinwen Zhu. Quantization of Hitchin inte-
grable system via positive characteristic. arXiv preprint arXiv:1603.01327, 2016.
[BZN] David Ben-Zvi and David Nadler. Loop spaces and connections. Journal of Topology, 5(2):377–430, 2012.
[CDR] Justin Campbell, Gurbir Dhillon, and Sam Raskin. Fundamental local equivalences in quantum geometric
Langlands. arXiv preprint arXiv:1907.03204, 2019.
[CG] Kevin Costello and Davide Gaiotto. Vertex Operator Algebras and 3d N “ 4 gauge theories.
[CR] Justin Campbell and Sam Raskin. Derived Satake via factorization. 2020. Forthcoming.
[CW] Sabin Cautis and Harold Williams. Canonical bases for Coulomb branches of 4𝑑 𝒩 “ 2 gauge theories.
Forthcoming.
[DGGH] Tudor Dimofte, Niklas Garner, Michael Geracie, and Justin Hilburn. Mirror symmetry and line operators.
Journal of High Energy Physics, 2020(1908.00013):1–150, 2020.
[Dhi] Gurbir Dhillon. Semi-infinite cohomology and the linkage principle for W-algebras. arXiv preprint
arXiv:1905.06477, 2019.
[Die] Jean Alexandre Dieudonné. History of algebraic geometry: an outline of the history and development of
algebraic geometry. 1985.
[Dri] Vladimir Drinfeld. Infinite-dimensional vector bundles in algebraic geometry. In The unity of mathematics,
pages 263–304. Springer, 2006.
[Fei] Boris Lvovich Feigin. The semi-infinite homology of Kac-Moody and Virasoro Lie algebras. Russian Math-
ematical Surveys, 39(2):155, 1984.
[FF] Boris Feigin and Edward Frenkel. Duality in W-algebras. International Mathematics Research Notices,
1991(6):75–82, 1991.
[FG1] Edward Frenkel and Dennis Gaitsgory. Local geometric Langlands correspondence and affine Kac-Moody
algebras. In Algebraic geometry and number theory, pages 69–260. Springer, 2006.
[FG2] Edward Frenkel and Dennis Gaitsgory. 𝐷-modules on the affine flag variety and representations of affine
Kac-Moody algebras. Represent. Theory, 13:470–608, 2009.
[Flo] Andreas Floer. Morse theory for Lagrangian intersections. Journal of differential geometry, 28(3):513–547,
1988.
[Gai1] Dennis Gaitsgory. Quasi-coherent sheaves on stacks. Available at [Link]
GL/[Link].
[Gai2] Dennis Gaitsgory. The notion of category over an algebraic stack. arXiv preprint math/0507192, 2005.
[Gai3] Dennis Gaitsgory. Whittaker categories. Available at [Link]
[Link]. Formerly titled Les jours et les travaux, 2008.
[Gai4] Dennis Gaitsgory. Generalities on DG categories. Available at [Link]
[Link], 2012.
[Gai5] Dennis Gaitsgory. Ind-coherent sheaves. Moscow Mathematical Journal, 13(3):399–528, 2013.
[Gai6] Dennis Gaitsgory. Outline of the proof of the geometric Langlands conjecture for GL (2). arXiv preprint
arXiv:1302.2506, 2013.
[Gai7] Dennis Gaitsgory. Kac-Moody representations (Day IV, Talk 3). Available at [Link]
site/geometriclanglands2014/notes. Unpublished conference notes, 2014.
[Gai8] Dennis Gaitsgory. Sheaves of categories and the notion of 1-affineness. In T. Pantev, C. Simpson, B. Toën,
M. Vaquié, and G. Vezzosi, editors, Stacks and Categories in Geometry, Topology, and Algebra:, Contem-
porary Mathematics. American Mathematical Society, 2015.
[Gai9] Dennis Gaitsgory. A conjectural extension of the Kazhdan-Lusztig equivalence. arXiv preprint
arXiv:1810.09054, 2018.
[GK] Dennis Gaitsgory and David Kazhdan. Representations of algebraic groups over a 2-dimensional local field.
Geometric & Functional Analysis GAFA, 14(3):535–574, 2004.
[GR1] Davide Gaiotto and Miroslav Rapčák. Vertex algebras at the corner. Journal of High Energy Physics,
2019(1):160, 2019.
[GR2] Dennis Gaitsgory and Nick Rozenblyum. Crystals and D-modules. Pure and Applied Mathematics Quarterly,
10(1):57–154, 2014.
[GR3] Dennis Gaitsgory and Nick Rozenblyum. DG indschemes. Contemp. Math, 610:139–251, 2014.
[GR4] Dennis Gaitsgory and Nick Rozenblyum. A study in derived algebraic geometry. American Mathematical
Soc., 2017.
138 SAM RASKIN

[Gro] Alexander Grothendieck. Éléments de géométrie algébrique. IV. Étude locale des schémas et des morphismes
de schémas IV. Inst. Hautes Études Sci. Publ. Math., page 361, 1967.
[GW] Davide Gaiotto and Edward Witten. 𝑆-duality of boundary conditions in N “ 4 super Yang-Mills theory.
Advances in Theoretical and Mathematical Physics, 13(3):721–896, 2009.
[HR] Justin Hilburn and Sam Raskin. Tate’s thesis in the de Rham setting. 2020. Forthcoming.
[Hum] James E. Humphreys. Representations of Semisimple Lie Algebras in the BGG Category O.
[HW] Amihay Hanany and Edward Witten. Type IIB superstrings, BPS monopoles, and three-dimensional gauge
dynamics. Nuclear Physics B, 492(1-2):152–190, 1997.
[IS] Kenneth Intriligator and Nathan Seiberg. Mirror symmetry in three dimensional gauge theories. Physics
Letters B, 387(3):513–519, 1996.
[KV] Mikhail Kapranov and Eric Vasserot. Vertex algebras and the formal loop space. Publications mathématiques
de l’IHÉS, 100:209–269, 2004.
[Liu] Chia-Cheng Liu. Semi-infinite cohomology and kazhdan-lusztig equivalence at positive level. arXiv preprint
arXiv:1807.01773, 2018.
[Lur1] Jacob Lurie. A survey of elliptic cohomology. In Algebraic topology, pages 219–277. Springer, 2009.
[Lur2] Jacob Lurie. Higher Topos Theory, volume 189. Princeton University Press, 2009.
[Lur3] Jacob Lurie. Higher algebra. Available at [Link]
2012.
[Lur4] Jacob Lurie. Spectral algebraic geometry. Preprint, available at www. math. harvard. edu/˜
lurie/papers/SAG-rootfile. pdf, 2016.
[Neg] Andrei Negut. Groups acting on categories. Notes from a talk, available at [Link]
edu/~gaitsgde/grad_2009/SeminarNotes/April6(GrpActCat).pdf., 2010.
[Pos1] Leonid Positselski. Homological algebra of semimodules and semicontramodules: Semi-infinite homological
algebra of associative algebraic structures, volume 70. Springer Science & Business Media, 2010.
[Pos2] Leonid Positselski. Two kinds of derived categories, Koszul duality, and comodule-contramodule correspon-
dence. American Mathematical Soc., 2011.
[PS] AN Pressley and AN Segal. Loop groups and their representations. Oxford University Press, Oxford, 1985.
[Ras1] Sam Raskin. A geometric proof of the Feigin-Frenkel theorem. Representation Theory of the American
Mathematical Society, 16(13):489–512, 2012.
[Ras2] Sam Raskin. Chiral categories. Available at [Link]
2015.
[Ras3] Sam Raskin. 𝐷-modules on infinite dimensional varieties. Available at [Link]
sraskin/[Link], 2015.
[Ras4] Sam Raskin. On the notion of spectral decomposition in local geometric Langlands. arXiv preprint
arXiv:1511.01378, 2015.
[Ras5] Sam Raskin. Chiral principal series categories II: the factorizable Whittaker category. Available at https:
//[Link]/users/sraskin/[Link], 2016.
[Ras6] Sam Raskin. W-algebras and Whittaker categories. Available at [Link]
sraskin/[Link], 2016.
[Ras7] Sam Raskin. Affine Beilinson-Bernstein localization at the critical level for 𝐺𝐿2 . 2020.
[Ras8] Sam Raskin. Categorical Weil representations. In preparation, 2020.
[Roz] Nick Rozenblyum. Localization of Kac-Moody representations (Day V, Talk 4). Available at https://
[Link]/site/geometriclanglands2014/notes. Unpublished conference notes, 2014.
[Sch] Peter Scholze. Lectures on condensed mathematics. Notes available at https: // www. math. unibonn. de/
people/ scholze/ Condensed. pdf , 2019.
[Seg] Graeme Segal. Elliptic cohomology. Séminaire Bourbaki, 88:161–162, 1987.
[Ser1] Jean-Pierre Serre. Faisceaux algébriques cohérents. Annals of Mathematics, pages 197–278, 1955.
[Ser2] Jean-Pierre Serre. Sur la dimension homologique des anneaux et des modules noethériens. In Proceedings
of the international symposium on algebraic number theory, Tokyo & Nikko, volume 1956, pages 175–189,
1955.
[Vor] Alexander A Voronov. Semi-infinite homological algebra. Inventiones mathematicae, 113(1):103–146, 1993.
[Wit1] Edward Witten. Elliptic genera and quantum field theory. Communications in Mathematical Physics,
109(4):525–536, 1987.
[Wit2] Edward Witten. An 𝑆𝑈 p2q anomaly. In Instantons In Gauge Theories, pages 230–234. World Scientific,
1994.
 HOMOLOGICAL METHODS IN SEMI-INFINITE CONTEXTS 139

The University of Texas at Austin, Department of Mathematics, RLM 8.100, 2515 Speedway Stop
C1200, Austin, TX 78712
E-mail address: sraskin@[Link]