Invent. math.

153, 1–44 (2003)

DOI: 10.1007/s00222-002-0275-2

Higher algebraic K-theory for actions

of diagonalizable groups
Gabriele Vezzosi
, Angelo Vistoli

Dipartimento di Matematica, Università di Bologna, 40126 Bologna, Italy
(e-mail: vezzosi@[Link], vistoli@[Link])

Oblatum 23-IV-2002 & 24-X-2002

Published online: 10 April 2003 –  Springer-Verlag 2003

Abstract. We study the K-theory of actions of diagonalizable group schemes

on noetherian regular separated algebraic spaces: our main result shows
how to reconstruct the K-theory ring of such an action from the K-theory
rings of the loci where the stabilizers have constant dimension. We apply
this to the calculation of the equivariant K-theory of toric varieties, and give
conditions under which the Merkurjev spectral sequence degenerates, so
that the equivariant K-theory ring determines the ordinary K-theory ring.
We also prove a very refined localization theorem for actions of this type.

Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 Notations and conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Preliminary results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3 Specializations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4 Reconstruction from the strata . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5 Actions with enough limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
6 The K-theory of smooth toric varieties . . . . . . . . . . . . . . . . . . . . . . . 27
7 The refined decomposition theorem . . . . . . . . . . . . . . . . . . . . . . . . . 34
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

Introduction
Fix a base noetherian separated connected scheme S, and let G be a diag-
onalizable group scheme of finite type over S (see [SGA3, Exposé VII]);
recall that this means that G is the product of finitely many multiplicative

Partially supported by the University of Bologna, funds for selected research topics.
Mathematics Subject Classification (2000): 19E08, 14L30
2 G. Vezzosi, A. Vistoli

groups Gm,S and group schemes µn,S of n th roots of 1 for various values
of n. Suppose that G acts on a separated noetherian regular algebraic space
X over S.
If G acts on X with finite stabilizers, then [Ve-Vi] gives a decomposition
theorem for the equivariant higher K-theory ring K∗ (X, G); it says that,
after inverting some primes, K∗ (X, G) is a product of certain factor rings
K∗ (X σ , G)σ for each subgroup scheme σ ⊆ G with σ  µn for some n and
X σ  = ∅ (the primes to be inverted are precisely the ones dividing the orders
of the σ ). A slightly weaker version of this theorem was given in [Toen].
From this one can prove analogous formulas assuming that the stabilizers
are of constant dimension (Theorem 7.4).
This paper deals with the general case, when the dimensions of the sta-
bilizers are allowed to jump. In this case one sees already in the simplest
examples that K∗ (X, G) will not decompose as a product, not even after
tensoring with Q; for example, if S is a field, G is a torus and X is a repre-
sentation of G, then K0 (X, G) is the ring of representations RG, which is
a ring of Laurent polynomials over Z.
However, we show that the ring K∗ (X, G) has a canonical structure of
fibered product. More precisely, for each integer s we consider the locus
X s of X where the stabilizers have dimension precisely equal to s; this
is a locally closed regular subspace of X. For each s consider the nor-
mal bundle Ns of X s in X, and the subspace Ns,s−1 where the stabilizers
have dimension precisely s − 1. There is a pullback map K∗ (X s , G) →
K∗ (Ns,s−1, G); furthermore in Sect. 3 we define a specialization homo-
morphism Sps−1 X,s : K∗ (X s−1 , G) → K∗ (Ns,s−1 , G), via deformation to the
normal bundle. Our first main result (Theorem 4.5) shows that these spe-
cializations homomorphisms are precisely what is needed to reconstruct the
equivariant K-theory of X from the equivariant K-theory of the strata.

Theorem 1 (The theorem of reconstruction from the strata). Let n be

the dimension of G. The restriction homomorphisms

K∗ (X, G) −→ K∗ (X s , G)

induce an isomorphism

K∗ (X, G)  K∗ (X n , G) × K∗ (X n−1 , G) ×
K∗ (Nn,n−1 ,G) K∗ (Nn−1,n−2 ,G)

... × K∗ (X 1 , G) × K∗ (X 0 , G).
K∗ (N2,1 ,G) K∗ (N1,0 ,G)

In other words: the restrictions
K∗ (X, G) → K∗ (X s , G) induce an injective

homomorphism
K∗ (X, G) → s K∗ (X s , G), and an element (αn , . . . , α0 )
of the product s K∗ (X s , G) is in the image of K∗ (X, G) if and only if the
pullback of αs ∈ K∗ (X s , G) to K∗ (Ns,s−1, G) coincides with Sps−1
X,s (αs−1 ) ∈
K∗ (Ns,s−1, G) for all s = 1, . . . , n.
Higher algebraic K-theory for actions of diagonalizable groups 3

This theorem is a powerful tool in studying the K-theory of diagonaliz-

able group actions. From it one gets easily a description of the higher equiv-
ariant K-theory of regular toric varieties (Theorem 6.2). This is analogous
to the description of their equivariant Chow ring in [Bri97, Theorem 5.4].
One can put Theorem 1 above together with the main result of [Ve-Vi]
to give a very refined description of K∗ (X, G); this is Theorem 7.12, which
can be considered the ultimate localization theorem for actions of diago-
nalizable groups. However, notice that it does not supersede Theorem 1,
because Theorem 1 holds with integral coefficients, while for the formula
of Theorem 7.12 to be correct we have to invert some primes.
Many results are known for equivariant intersection theory, or for equiv-
ariant cohomology; often one can use our theorem to prove their K-theoretic
analogues. For example, consider the following theorem of Brion, inspired
in turn by results in equivariant cohomology due to Atiyah ([Ati74]), Bre-
don ([Bre74]), Hsiang ([Hsi75]), Chang and Skjelbred ([Ch-Sk74]), Kirwan
([Kir84]), Goresky, Kottwitz and MacPherson ([G-K-MP98]); see also the
very useful discussion in [Bri98].

Theorem ([Bri97, 3.2, 3.3]). Suppose that X is a smooth projective al-

gebraic variety over an algebraically closed field with an action of an
algebraic torus G.
(i) The rational equivariant Chow ring A∗G (X )Q is free as a module over
A∗G (pt)Q .
(ii) The restriction homomorphism

A∗G (X )Q −→ A∗G (X G )Q = A∗ (X G )Q ⊗ A∗G (pt)Q

is injective, and its image is the intersection of all the images of the
restriction homomorphisms A∗G (X T )Q → A∗G (X G )Q , where T ranges
over all the subtori of codimension 1.

From this one gets a very simple description of the rational equivariant
Chow ring when the fixed point locus X G is zero dimensional, and the
fixed point set X T is at most 1-dimensional for any subtorus T ⊆ G of
codimension 1 ([Bri97, Theorem 3.4]).
In this paper we prove a version of Brion’s theorem for algebraic K-
theory. Remarkably, it holds with integral coefficients: we do not need to
tensor with Q. This confirms the authors’ impression that when it comes to
torsion, K-theory tends to be better behaved than cohomology, or intersec-
tion theory.
The following is a particular case of Corollary 5.11; when G is a torus,
it is an an analogue of part (2) of Brion’s theorem.

Theorem 2. Suppose that G is a diagonalizable group acting a smooth

proper scheme X over a perfect field; denote by G 0 the toral component
of G, that is, the largest subtorus contained in G.
4 G. Vezzosi, A. Vistoli

Then the restriction homomorphism K∗ (X, G) → K∗ (X G 0 , G) is injec-

tive, and its image equals the intersection of all the images of the restriction
homomorphisms K∗ (X T , G) → K∗ (X G 0 , G) for all the subtori T ⊆ G of
codimension 1.

From this one gets a very complete description of K∗ (X, G) when G is

a torus and X is smooth and proper over an algebraically closed field, in the
“generic” situation when X contains only finitely many invariant points and
finitely many invariant curves (Corollary 5.12).
We also analyze the case of smooth toric varieties in detail in Sect. 6.
The analogue of Theorem 2 should hold for the integral equivariant
topological K-theory of a compact differentable manifold with the action of
a compact torus. Some related topological results are contained in [Ro-Kn].

Description of contents. Section 1 contains the setup that will be used

throughout this paper. The K-theory that we use is the one described
in [Ve-Vi]: see the discussion in Subsect. 1.1.
Section 2 contains some preliminary technical results; the most sub-
stantial of these is a very general self-intersection formula, proved fol-
lowing closely the proof of Thomason of the analogous formula in the
non-equivariant case ([Tho93, Théorème 3.1]). Here we also discuss the
stratification by dimensions of stabilizers, which is our basic object of
study.
In Sect. 3 we define various types of specializations to the normal bundle
in equivariant K-theory. This is easy for K0 , but for the whole higher K-
theory ring we do not know how to give a definition in general without using
the language of spectra.
Section 4 contains the proof of Theorem 1.
Section 5 is dedicated to the analysis of the case when X is complete,
or, more generally, admits enough limits (Definition 5.8). The condition
that K∗ (X, G) be free as a module over the representation ring RG is not
adequate when working with integral coefficients: here we analyze a rather
subtle condition on the RG-module K∗ (X, G) that ensures that the analogue
of Brion’s theorem above holds, then we show, using a Białynicki-Birula
stratification, that this condition is in fact satisfied when X admits enough
limits over a perfect field.
We also apply our machinery to show that the degeneracy of the Merkur-
jev spectral sequence in [Mer97], that he proves when X is smooth and
projective, in fact happens for torus actions with enough limits.
Section 6 is dedicated to the K-theory of smooth toric varieties. For
any smooth toric variety X for a torus T , we give two descriptions of
K∗ (X, T ). First of all, we show how Theorem 4.5 in this case gives a simple
description of it as a subring of a product of representation rings, analogous
to the description of its equivariant Chow ring in [Bri97, Theorem 5.4].
Furthermore, we give a presentation of K∗ (X, T ) by generators and relations
over the K-theory ring of the base field (Theorem 6.4), analogous to the
Higher algebraic K-theory for actions of diagonalizable groups 5

classical Stanley–Reisner presentation for its equivariant cohomology first

obtained in [B-DC-P90]. For K0 the result is essentially stated in [Kly83].
In Sect. 7 we generalize the result of [Ve-Vi] by giving a formula that
holds for all actions of diagonalizable groups on regular noetherian algebraic
spaces, irrespective of the dimensions of the stabilizers (Theorem 7.12).

Acknowledgements. The first author is thankful for the hospitality at the University of Utah
and the Université de Grenoble, where some of the work on this paper has been done.
During the writing of the first draft of this paper the second author held a visiting
position at the Department of Mathematics of the University of Utah: he is very grateful for
the hospitality. He would like to thank Jim Carlson, Steve Gersten, Dragan Milicic, Anne
and Paul Roberts, Paula and Domingo Toledo, and, most particularly, Aaron Bertram, Herb
Clemens and their families for making his stay such an enjoyable experience.
Both of us would like to thank Michel Brion for his help, and for many very helpful
conversations; among other things, he suggested the possibility that we might be able to
prove Corollary 5.11. Also, his excellent articles [Bri97] and [Bri98] were a source of
inspiration.
The results of Subsect. 6.2 were obtained after stimulating conversations with Bernd
Sturmfels, Howard Thompson, Allen Knutson and Corrado De Concini, that took place
while the second author was a visitor at the Mathematical Sciences Research Institute in
Berkeley. Knutson also pointed out the references [Ati74] and [Bre74] to us. We are grateful
to all of them, and to MSRI for the hospitality.
We also want to thank the referee, who did an unusually careful and fair job.
Finally, we would like to remark how much we owe to the articles of the late Robert
W. Thomason: without his groundbreaking work on K-theory, equivariant and non, this
paper might never have been written. His premature death has been a great blow to the
mathematical community.

1. Notations and conventions

Throughout the paper we fix a base scheme S, that is assumed to be con-

nected, separated and noetherian.
We will denote by G a diagonalizable group scheme of finite type over
S (see [SGA3]), except when otherwise mentioned. Its groups of characters
 def
is G = Hom S (G, Gm,S ); the contravariant functor from the category of
diagonalizable groups schemes of finite type over S to the category of
finitely generated abelian groups given by G → G  is an antiequivalence of
categories. The ring of representations of G is, by definition, RG = ZG, 
and furthermore G = Spec RG ×Spec Z S.
We will denote by G 0 the toral part of G, that is, the largest subtorus
of G. The group of characters G0 is the quotient of G
 by its torsion subgroup.
A G-space will always be a regular separated noetherian algebraic space
over S over which G acts; sometimes we will talk about a regular G-space,
for emphasis.
We notice explicitly that if S → S is a morphism of schemes, with
S connected, then every diagonalizable subgroup scheme of G ×S S is
obtained by base change from a unique diagonalizable subgroup scheme
of G. This will be used as follows: if p : Spec Ω → X is a geometric point,
6 G. Vezzosi, A. Vistoli

then we will refer to its stabilizer, which is a priori a subgroup scheme of

G ×S Spec Ω, as a subgroup scheme of G.
If Y → X is a regular embedding, we denote by NY X the normal bundle.

1.1. Equivariant K-theory. In this subsection G will be a group scheme

over S that is flat, affine and of finite type. We use the same K-theoretic
setup as in [Ve-Vi], that uses the language of [Th-Tr90]. The following is
a slight extension of [Ve-Vi, Theorem 6.4].
Proposition 1.1. Let G be flat affine separated group scheme of finite type
over S, acting over a noetherian regular separated scheme X over S. Con-
sider the following complicial bi-Waldhausen categories:
(i) the category W1 (X, G) of complexes of quasicoherent G-equivariant
O X -modules with bounded coherent cohomology;
(ii) the category W2 (X, G) of bounded complexes of coherent G-equi-
variant O X -modules;
(iii) the category W3 (X, G) of complexes of flat quasicoherent G-equi-
variant O X -modules with bounded coherent cohomology, and
(iv) the category W4 (X, G) of bounded above complexes of G-equivariant
quasi-coherent flat O X -Modules with bounded coherent cohomology.
Then the inclusions
W2 (X, G) ⊆ W1 (X, G) and W4 (X, G) ⊆ W3 (X, G) ⊆ W1 (X, G)
induce isomorphisms on the corresponding Waldhausen K-theories. Fur-
thermore the K-theory of any of the categories above coincides with the
Quillen K-theory K∗ (X, G) of the category of G-equivariant coherent O X -
modules.
Proof. For the first three categories, and the Quillen K-theory, the statement
is precisely [Ve-Vi, Theorem 6.4].
Let us check that the inclusion W4 (X, G) ⊆ W1 (X, G) induces an
isomorphism in K-theory. By [Ve-Vi, Proposition 6.2], which shows that
hypothesis [Link] is satisfied, we can apply [Th-Tr90, Lemma 1.9.5], in
the situation where A is the category of G-equivariant quasicoherent O X -
Modules, C the category of cohomologically bounded complexes in A, D
the category of G-equivariant quasicoherent flat O X -Modules, F : D → A
is the natural inclusion. In particular, any complex in W1 (X, G) receives
a quasi-isomorphism from a complex in W4 (X, G). That is, [Th-Tr90,
[Link]], applied to the inclusion W4 (X, G) → W1 (X, G), is satisfied;
since the other hypothesis [Link] of [Th-Tr90, 1.9.7] is obviously satisfied,
we conclude by [Th-Tr90, Theorem 1.9.8].
We will denote by K(X, G) the Waldhausen K-theory spectrum and by
K∗ (X, G) the Waldhausen K-theory group of any of the categories above.
As observed in [Ve-Vi, p. 39], it follows from results of Thomason that
Higher algebraic K-theory for actions of diagonalizable groups 7

K∗ (−, G) is a covariant functor for proper maps of noetherian regular sepa-

rated G-algebraic spaces over S; furthermore, each K∗ (X, G) has a natural
structure of a graded ring, and each equivariant morphism f : X → Y of
noetherian regular separated G-algebraic spaces over S induces a pullback
f ∗ : K∗ (Y, G) → K∗ (X, G), making K∗ (−, G) into a contravariant functor
from the category of noetherian regular separated G-algebraic spaces over S
to graded-commutative rings. Furthermore, if i : Y → X is a closed embed-
ding of noetherian regular separated G-algebraic spaces and j : X \ Y → X
is the open embedding, then K(X \ Y, G) is the cone of the pushforward
map i ∗ : K(Y, G) → K(X, G) ([Tho87, Theorem 2.7]), so there is an exact
localization sequence
i∗
· · · −→ Kn (Y, G) −→ Kn (X, G)
j∗ ∂
−→ Kn (X \ Y, G) −→ Kn−1 (Y, G) −→ · · ·
Furthermore, if π : E → X is a G-equivariant vector bundle, the pull-
back
π ∗ : K∗ (X, G) −→ K∗ (E, G)
is an isomorphism ([Tho87, Theorem 4.1]).

2. Preliminary results

2.1. The self-intersection formula. Here we generalize Thomason’s self-

intersection formula ([Tho93, Théorème 3.1]) to the equivariant case.
Theorem 2.1 (The self-intersection formula). Suppose that a flat group
scheme G separated and of finite type over S acts over a noetherian regular
separated algebraic space X. Let i : Z → X be a regular G-invariant
closed subspace of X. Then
i ∗ i ∗ : K(Z, G) −→ K(Z, G),
coincides up to homotopy with the cup product
 
λ−1 N∨Z X  (−) : K(Z, G) −→ K(Z, G),
where N∨Z X is the conormal sheaf of Z in X.
In particular, we have the equality
 
i ∗ i ∗ = λ−1 N∨Z X  (−) : K∗ (Z, G) −→ K∗ (Z, G).
Proof. The proof follows closely Thomason’s proof of [Tho93, Théo-
rème 3.1], therefore we will only indicate the changes we need for that
proof to adapt to our situation.
Let us denote by W (Z, G) the Waldhausen category consisting of
pairs (E • , λ : L •• → i ∗ E • ) where E • is a bounded above complex of
8 G. Vezzosi, A. Vistoli

G-equivariant quasi-coherent flat O Z -Modules with bounded coherent co-

homology, L •• is a bicomplex of G-equivariant quasi-coherent flat O X -
Modules with bounded coherent total cohomology such that L ij = 0 for
j ≤ 0, any i and also L ij = 0 for i > N, for some integer N, any i;
finally λ : L •• → i ∗ E • is an exact augmentation of the bicomplex L •• .
In particular, for any i, the horizontal complex L i• is a flat resolution of
i ∗ E i . The morphisms, cofibrations and weak equivalences in W (Z, G) are
as in [Tho93, 3.3, p. 209]. Thomason [Tho93, 3.3] shows that the forget-
ful functor (E • , λ : L •• → i ∗ E • ) → E • from W (Z, G) to the category
W4 (Z, G) of bounded above complexes of G-equivariant quasi-coherent
flat O Z -Modules with bounded coherent cohomology induces a homotopy
equivalence between the associated Waldhausen K -theory spectra. In other
words, by Proposition 1.1, we can (and will) use W (Z, G) as a “model”
for K(Z, G).
With these choices, the morphism of spectra i ∗ i ∗ : K(Z, G) −→
K(Z, G) can be represented by the exact functor W (Z, G) −→ W1 (Z, G)
which sends (E • , λ : L •• → i ∗ E • ) to the total complex of the bicomplex
i ∗ (L •• ).
The rest of the proof is exactly the same as in [Tho93, 3.3, pp. 210–212].
One first consider functors Tk : W (Z, G) → W1 (Z, G) sending an object
(E • , λ : L •• → i ∗ E • ) to the total complex of the bicomplex

.. .. .. ..
. . . .

0 /

im ∂hi,−k−1 /

i ∗ L i,−k /

··· /

i ∗ L i,−1 /

i ∗ L i,0 /

0 /

im ∂hi+1,−k−1 /

i ∗ L i+1,−k /

··· /

i ∗ L i+1,−1 /

i ∗ L i+1,0 /

.. .. .. ..
. . . .

which results from truncating all the horizontal complexes of i ∗ L •• at the

k-th level.
The functors Tk are zero for k < 0 and come naturally equipped with
functorial epimorphisms Tk
Tk−1 whose kernel h k has the property that
h k (E • , λ : L •• → i ∗ E • ) is quasi isomorphic to Λk (N∨Z X ) ⊗O Z i ∗ E • [k]
([Tho93, 3.4.4], essentially because each horizontal complex in L i• is
a flat resolution of i ∗ E i . Therefore, by induction on k ≥ −1, starting
from T−1 = 0, each Tk has values in W1 (Z, G) and preserves quasi-
isomorphisms. Moreover, the arguments in [Tho93, 3.4, pp. 211–212],
show that Tk actually preserves cofibrations and pushouts along cofibra-
tions; hence each Tk : W (Z, G) −→ W1 (Z, G) is an exact functor of
Waldhausen categories.
Higher algebraic K-theory for actions of diagonalizable groups 9

As in [Tho93, 3.4, p. 212], the quasi-isomorphism

 
h k (E • , λ : L •• → i ∗ E • )  Λk N∨Z X ⊗O Z i ∗ E • [k]
shows that the canonical truncation morphism i ∗ (L •• ) → Td (E • , λ : L •• →
i ∗ E • ), d being the codimension of Z in X, is a quasi-isomorphism, i.e. the
morphism of spectra i ∗ i ∗ : K (Z, G) −→ K (Z, G) can also be represented
by the exact functor Td : W (Z, G) −→ W1 (Z, G). Now, the Additivity
Theorem ([Th-Tr90, 1.7.3 and 1.7.4]) shows that the canonical exact se-
quences of functors h k → Tk
Tk−1 yield up-to-homotopy equalities
Tk = Tk−1 + h k between the induced map of spectra. And finally, recalling
that a shift [k] induces multiplication by (−1)k at the level of spectra, by
induction on k ≥ −1 we get equalities up to homotopy
i ∗ i ∗ = Td (−)

= h k (−)
k
  
= Λk N∨Z X ⊗O Z i ∗ (−)[k]
k
 
= λ−1 N∨Z X  (−)
of morphisms of spectra K(Z, G) → K(Z, G) ([Tho93, p. 212]).

2.2. Stratification by dimensions of stabilizers. Let G be a diagonaliz-

able group scheme of finite type acting on X as usual. Consider the group
scheme H → X of stabilizers of the action. Since for a point x ∈ X
the dimension of the fiber Hx equals its dimension at the point γ(x),
where γ : X → H is the unit section, it follows from Chevalley’s theo-
rem ([EGAIV, 13.1.3]) that there is an open subset X ≤s where the fibers of
H have dimension at most s. We will use also X <s with a similar meaning.
We denote by X s the locally closed subset X ≤s \ X <s ; we will think of it as
a subspace of X with the reduced scheme structure. Finally, we call Ns the
normal bundle of X s in X, and N0s the complement of the 0-section in Ns .
Notice that G acts on Ns , so we may consider the subscheme (Ns )<s ⊆ Ns .
Proposition 2.2. Let s be a nonzero integer.
(i) There exists a finite number of s-dimensional subtori T1 , . . . , Tr in G
T
such that X s is the disjoint union of the X ≤sj .
(ii) X s is a regular locally closed subspace of X.
(iii) N0s = (Ns )<s .
Proof. To prove part (i) we may restrict the action of G to its toral com-
ponent. By Thomason’s generic slice theorem ([Tho86a, Proposition 4.10])
there are only finitely many possible diagonalizable subgroup schemes of
G that appear as stabilizers of a geometric point of X. Then we can take the
T j to be the toral components of the s-dimensional stabilizers.
Parts (ii) and (iii) follow from (i) and [Tho92, Proposition 3.1].
10 G. Vezzosi, A. Vistoli

3. Specializations

In this section G will be a flat, affine and separated group scheme of finite
type over S, acting on a noetherian regular separated algebraic space Y
over S.
Definition 3.1. A G-invariant morphism Y → P1S is regular at infinity if
the inverse image Y∞ of the section at infinity in P1S is a regular effective
Cartier divisor on Y .
Theorem 3.2. Let π : Y → P1S be a G-invariant morphism over S that is
regular at infinity. Denote by i ∞ : Y∞ → Y the inclusion of the fiber at
infinity, j∞ : Y \ Y∞ → Y the inclusion of the complement. Then there
exists a specialization homomorphism of graded groups
SpY : K∗ (Y \ Y∞ , G) −→ K∗ (Y∞ , G)
such that the composition
∗
j∞ SpY
K∗ (Y, G) −→ K∗ (Y \ Y∞ ) −→ K∗ (Y∞ , G)
∗
coincides with i ∞ : K∗ (Y, G) → K∗ (Y∞ , G).
Furthermore, if Y is another noetherian separated regular algebraic
space over S and f : Y → Y is a G-equivariant morphism over S such that
the composition π f : Y → P1S is regular at infinity, then the diagram
SpY
K∗ (Y \ Y∞ , G) /

K∗ (Y∞ , G)
∗
f∗ f∞
SpY
K∗ (Y \ Y∞ , G) /

K∗ (Y∞ , G)

commutes; here f ∞ is the restriction of f to Y∞ → Y∞ .

We refer to this last property as the compatibility of specializations.
Proof. Let us denote by K(X, G) the Quillen K-theory spectrum associ-
ated with the category of coherent equivariant G-sheaves on a noetherian
separated algebraic space X. There is a homotopy equivalence
 i∞∗ 
Cone K(Y∞ , G) −→ K(Y, G)  K(Y \ Y∞ , G).
The commutative diagram
i∞∗
K(Y∞ , G) /

K(Y, G)
∗
i∞
∗ i
i∞ ∞∗
K(Y∞ , G) /

K(Y∞ , G)
Higher algebraic K-theory for actions of diagonalizable groups 11

induces a morphism of spectra

 i∞∗ 
K(Y \ Y∞ , G)  Cone K(Y∞ , G) −→ K(Y, G)
(3.1)  ∗ i
i∞ 
∞∗
−→ Cone K(Y∞ , G) −→ K(Y∞ , G) .
By the self-intersection formula (Theorem 2.1) there is a homotopy
∗
 
i∞ i ∞∗  λ−1 N∨Y∞  (−) : K(Y∞ , G) → K(Y∞ , G);
on the other hand λ−1 (NY∨∞ )  (−) is homotopic to zero, because N∨Y∞ is
trivial. So we have that
 ∗ i
i∞ ∞∗ 
Cone K(Y∞ , G) −−−→ K(Y∞ , G)  K(Y∞ , G)[1] ⊕ K(Y∞ , G),
where (−)[1] is the suspension of (−). We define the specialization mor-
phism of spectra
SY : K(Y \ Y∞ , G) −→ K(Y∞ , G)
by composing the morphism (3.1) with the canonical projection
K(Y∞ , G)[1] ⊕ K(Y∞ , G) −→ K(Y∞ , G).
Finally, SpY is defined to be the homomorphism induced by SY on homotopy
groups.
Let us check compatibility; it suffices to show that the diagram of spectra
i∞∗
K(Y∞ , G) /

K(Y , G)
∗
f∞ f∗
i∞∗
K(Y∞ , G) /

K(Y, G)

commutes up to homotopy. The essential point is that the diagram of alge-

braic spaces
i∞
Y∞ /

Y
f∞ f
i∞
Y∞ /

is Tor-independent, that is, TorO

i (OY , OY∞ ) = 0 for all i > 0. Write
Y

W(Y , G) and W(Y∞ , G) for the Waldhausen categories of G-equivariant

complexes of quasicoherent OY -modules and OY∞ -modules with bounded
coherent cohomology, while W(Y, G) and W(Y∞ , G) will denote the Wald-
hausen categories of complexes of G-equivariant quasicoherent OY -modules
and OY∞ bounded coherent cohomology, that are respectively degreewise
12 G. Vezzosi, A. Vistoli

f ∗ -acyclic and degreewise f ∞

∗
-acyclic. By the Tor-independence of the di-
agram above, we have that i ∞∗ gives a functor W(Y∞ , G) → W(Y, G), and
the diagram

i∞∗
W(Y∞ , G) /

W(Y, G)
∗
f∞ f∗
i∞∗
W(Y∞ , G) /

W(Y , G)

commutes. By [Th-Tr90, 1.5.4] this concludes the proof of the theorem.

Remark 3.3. For the projection pr2 : X ×S P1S → P1S the specialization
homomorphism
 
Sp X×S P1S : K∗ X ×S A1S , G −→ K∗ (X, G)

coincides with the pullback s0∗ : K∗ (X ×S A1S , G) −→ K∗ (X, G) via the

zero-section s0 : X → X ×S A1S .
∗
In fact, the pullback j∞ : K∗ (X ×S P1S ) → K∗ (X ×S A1S ) is surjective,
and we have
∗
 
Sp X×S P1S ◦ j∞ = s0∗ ◦ j∞
∗ ∗
= i∞ : K∗ X ×S P1S −→ K∗ (X, G).

Remark 3.4. Since the restriction homomorphism K0 (Y, G) → K0 (Y \Y∞ ,

G) is a surjective ring homomorphism, and its composition with SpY :
K0 (Y \ Y∞ , G) → K0 (Y∞ , G) is a ring homomorphism, it follows that
the specialization in degree 0 SpY : K0 (Y \ Y∞ , G) → K0 (Y∞ , G) is also
a ring homomorphism. This should be true for the whole specialization
homomorphism SpY : K∗ (Y \Y∞ , G) → K∗ (Y∞ , G), but this is not obvious
from the construction, and we do not know how to prove it.

Remark 3.5. From the proof of Theorem 3.2 we see that one can define
a specialization homomorphism K∗ (Y \ Z, G) → K∗ (Z, G) if Z is a regular
effective G-invariant divisor on Y whose normal sheaf is G-equivariantly
trivial.

3.1. Specializations to the normal bundle. Let us go back to our standard

situation, in which G is a diagonalizable group scheme of finite type acting
on a regular separated noetherian algebraic space X. Fix a nonnegative
integer s, and consider the closed immersion X s → X ≤s ; denote by Ns its
normal bundle. Consider the deformation to the normal cone π : Ms → P1S ,
the one denoted by M0X s X ≤s in [Ful93a, Chapter 5]. The morphism π : Ms →
P1S is flat and G-invariant. Furthermore π −1 (A1S ) = X ≤s ×S (A1S ), while the
fiber at infinity of π is Ns . Consider the restriction π 0 : M0s → P1S to the
Higher algebraic K-theory for actions of diagonalizable groups 13

open subset M0s = (Ms )<s ; then (π 0 )−1 (A1S ) = X <s ×S A1S , while the fiber
at infinity of π 0 is N0s = (Ns )<s . We define a specialization homomorphism
 
Sp X,s : K∗ (X <s , G) −→ K∗ N0s , G

by composing the pullback

   
K∗ (X <s , G) −→ K∗ X <s ×S A1S , G = K∗ M0s \ N0s , G

with the specialization homomorphism

   
SpM0s : K∗ M0s \ N0s , G −→ K∗ N0s , G

defined in the previous subsection.

We can also define more refined specializations.

Proposition 3.6. Let Y → P1S be regular at infinity.

(i) If H ⊆ G is a diagonalizable subgroup scheme, then the restriction
Y H → P1 is regular at infinity.
(ii) If s is a nonnegative integer, the restriction Ys → P1S is also regular at
infinity.

Proof. Part (i) and Proposition 2.2 (i) imply part (ii).
To prove (i), notice that, by [Tho92, Prop. 3.1], Y H is regular, and so is
H
Y∞ . Let f be the pullback to Y of a local equation for the section at infinity
of P1S → S, and let p be a point of Y∞ H
. Since the conormal space to Y H in
Y has no nontrivial H-invariants, clearly the differential of f at p can not
lie in this conormal space, hence f is not zero in any neighborhood of p in
Y H . This implies that Y∞H
is a regular Cartier divisor on Y H , as claimed.
def
If t is an integer with t < s, let us set Ns,t = (Ns )t . We have that the
restriction (Ms )t → P1S is still regular at infinity, by Proposition 3.6 (ii); so
we can also define a specialization homomorphism
 
SptX,s : K∗ (X t , G) −→ K∗ Ns,t , G

by composing the pullback

    
K∗ (X t , G) −→ K∗ X t ×S A1S , G = K∗ M0s t \ Ns,t , G

with the specialization homomorphism

    
Sp(M0s )t : K∗ M0s t \ Ns,t , G −→ K∗ Ns,t , G .

The specializations above are compatible, in the following sense.

14 G. Vezzosi, A. Vistoli

Proposition 3.7. In the situation above, the diagram

K∗ (X <s , G) /

K∗ (X t , G) ,
Sp X,s SptX,s
 
K∗ N0s , G /

K∗ (Ns,t , G)

where the rows are restriction homomorphisms, commutes.

Proof. This follows immediately from the compatibility of specializations
(Theorem 3.2).

4. Reconstruction from the strata

4.1. K-rigidity.
Definition 4.1. Let Y be a G-invariant regular locally closed subspace
of X. We say that Y is K-rigid inside X if Y is regular and λ−1 (N∨Y X ) is not
a zero-divisor in the ring K∗ (Y, G).
This condition may seem unlikely to ever be verified: in the non-
equivariant case λ−1 (N∨Y X ) is always a nilpotent element, since it has rank
zero over each component of X. However, in the equivariant case this is not
necessary true. Here is the basic criterion that we will use use to check that
a subspace is K-rigid.
Lemma 4.2. Let Y be a G-space, E an equivariant vector bundle on Y .
Suppose that there is a subtorus T of G acting trivially on Y , such that
in the eigenspace decomposition of E with respect to T the subbundle
corresponding to the trivial character is 0. Then λ−1 (E) is not a zero-
divisor in K∗ (Y, G).
Proof. Choose a splitting G  D × T ; by [Tho86b, Lemme 5.6], we have
 
K∗ (Y, G) = K∗ (Y, D) ⊗ RT = K∗ (Y, D) ⊗ Z t1±1 , . . . , tn±1 .

If E =  E χ is the eigenspace decomposition of E, we have that
χ∈T

λ−1 (E) corresponds to the element χ∈T λ−1 (E χ ⊗ χ) of K∗ (Y, D) ⊗ RT ,
so it enough to show that λ−1 (E χ ⊗χ) is not a zero-divisor in K∗ (Y, D)⊗RT .
But we can write
λ−1 (E χ ⊗ χ) = 1 + r1 χ + r2 χ 2 + · · · + rn χ n ∈ K∗ (Y, D) ⊗ RT,
where rn = (−1)n [det E χ ] is a unit in K∗ (Y, D).
Now we can apply the following elementary fact: suppose that A is a ring,
r1 , . . . , rn central elements of A such that rn is a unit, χ ∈ A[t1±1 , . . . , tn±1 ]
a monomial different from 1. Then the element 1 + r1 χ + r2 χ 2 + · · · + rn χ n
is not a zero-divisor in A[t1±1 , . . . , tn±1 ].
Higher algebraic K-theory for actions of diagonalizable groups 15

The next proposition is a K-theoretic variant of [Bri98, Proposition 3.2].

Proposition 4.3. Let Y be a closed K-rigid subspace of X, and set U =
X \ Y . Call i : Y → X and j : U → X the inclusions.
(i) The sequence
i∗ j∗
0 −→ K∗ (Y, G) −→ K∗ (X, G) −→ K∗ (U, G) −→ 0
is exact.
(ii) The two restriction maps
i ∗ : K∗ (X, G) −→ K∗ (Y, G) and j ∗ : K∗ (X, G) −→ K∗ (U, G)
induce a ring isomorphism
∼
(i ∗ , j ∗ ) : K∗ (X, G) −→ K∗ (Y, G) × K∗ (U, G)
K∗ (Y,G)/(λ−1 (N∨
Y X ))

where N∨Y X is the conormal bundle of Y in X, the homomorphism

  
K∗ (Y, G) −→ K∗ (Y, G)/ λ−1 N∨Y X
is the projection, while the homomorphism
  
K∗ (U, G)  K∗ (X, G)/i ∗ K∗ (Y, G) −→ K∗ (Y, G)/ λ−1 N∨Y X
is induced by i ∗ : K∗ (X, G) → K∗ (Y, G).
Proof. From the self-intersection formula (Theorem 2.1) we see that the
composition i ∗ i ∗ : K∗ (Y, G) → K∗ (Y, G) is multiplication by λ−1 (N∨Y X ),
so i ∗ is injective. We get part (i) from this and from the localization sequence.
Part (ii) follows easily from part (i), together with the following elemen-
tary fact.
Lemma 4.4. Let A, B and C be rings, f : B → A and g : B → C ring ho-
momorphisms. Suppose that there exist a homomorphism of abelian groups
φ : A → B such that:
(i) The sequence
φ g
0 −→ A −→ B −→ C −→ 0.
is exact;
(ii) the composition f ◦φ : A → A is the multiplication by a central element
a ∈ A which is not a zero divisor.
Then f and g induce an isomorphism of rings
( f, g) : B → A × C,
A/(a)

where the homomorphism A → A/(a) is the projection, and the one

C → A/(a) is induced by the isomorphism C  B/ im φ and the ring
homomorphism f : B → A.
16 G. Vezzosi, A. Vistoli

4.2. The theorem of reconstruction from the strata. This section is en-
tirely dedicated to the proof of our main theorem. Let us recall what it says.
Let G act on X with the usual hypotheses. Consider the strata X s defined in
Subsect. 2.2, and the specialization homomorphisms
 
SptX,s : K∗ (X t , G) −→ K∗ Ns,t , G
defined in Subsect. 3.1.
Theorem 4.5 (The theorem of reconstruction from the strata). The
homomorphism
n
K∗ (X, G) −→ K∗ (X s , G)
s=0

obtained from the restrictions K∗ (X, G)
n→ K∗ (X s , G) is injective. Its

image consists of the sequences (αs ) ∈ s=0 K∗ (X s , G) with the  property
that for each s = 1, . . . , n the pullback of α ∈ K∗ (X s , G) to K∗ Ns,s−1 , G
 
X,s (αs−1 ) ∈ K∗ Ns,s−1 , G .
coincides with Sps−1
In other words, we can view K∗ (X, G) as a fiber product
K∗ (X, G)  K∗ (X n , G) × K∗ (X n−1 , G) ×
K∗ (Nn,n−1 ,G) K∗ (Nn−1,n−2 ,G)

... × K∗ (X 1 , G) × K∗ (X 0 , G).
K∗ (N2,1 ,G) K∗ (N1,0 ,G)

Here is our starting point.

Proposition 4.6. X s is K-rigid in X.
Proof. This follows from Proposition 2.2 (iii), and Lemma 4.2.
So from Proposition 4.3 (ii) applied to the closed embedding i s : X s →
X ≤s , we get an isomorphism
K∗ (X ≤s , G)  K∗ (X s , G) × K∗ (X <s , G).
K∗ (X s ,G)/(λ−1 (N∨
s ))

We can improve on this.

Proposition 4.7. The restrictions
K∗ (X ≤s , G) −→ K∗ (X s , G) and K∗ (X ≤s , G) −→ K∗ (X <s , G)
induce an isomorphism
∼
K∗ (X ≤s , G) −→ K∗ (X s , G) × K∗ (X <s , G),
K∗ (N0s ,G)

where the homomorphism K∗ (X s , G) → K∗ (N0s , G) is the pullback, while

 
SpY,s : K∗ (X <s , G) → K∗ N0s , G
is the specialization.
Higher algebraic K-theory for actions of diagonalizable groups 17

Proof. Let us start with a lemma.

∗ (X s , G) → K∗ (Ns , G) is
0
Lemma 4.8. The restriction homomorphism K
surjective, and its kernel is the ideal λ−1 (N∨s ) ⊆ K∗ (X s , G).
Proof. Since the complement of the zero section s0 : X s → Ns coincides
with (Ns )<s (Proposition 2.2 (iii)), we can apply Proposition 4.3 (i) to the
normal bundle Ns , and conclude that there is an exact sequence
s0∗  
0 −→ K∗ (X s , G) −→ K∗ (Ns , G) −→ K∗ N0s , G −→ 0.
Now, s0∗ : K∗ (Ns , G) → K∗ (X s , G) is an isomorphism, and the com-
position s0∗ s0∗ : K∗ (X s , G) → K∗ (X s , G) is multiplication by λ−1 (N∨s ),
because of the self-intersection formula 2.1, and this implies the thesis.
Therefore the restriction homomorphism
 K∗ (X s , G) → K∗ (N0s , G) in-
duces an isomorphism of K ∗ (X s , G)/ λ−1 (N∨s ) with K∗ (N0s , G); the propo-
sition follows from this, and from Proposition 3.7.
Now we proceed by induction on the largest integer s such that X s  = ∅.
If s = 0 there is nothing to prove. If s > 0, by induction hypothesis the
homomorphism
K∗ (X <s , G) −→ K∗ (X s−1 , G) × ··· × K∗ (X 0 , G)
K∗ (Ns,s−1 ,G) K∗ (N1,0 ,G)

induced by restrictions is an isomorphism; so from Proposition 4.7 we see

that to prove Theorem 4.5 it is sufficient to show that if αs ∈ K∗ (X s , G),
α<s ∈ K∗ (X < s, G), αs−1 is the restriction of α<s to K∗ (X s−1 , G), αs0 is the
pullback of αs to K∗ (N0s , G) and αs,s−1 is the pullback of αs to K∗ (Ns,s−1, G),
then Sp X,s (α<s ) = αs0 if and only if Sps−1
X,s (αs−1 ) = αs,s−1 . But the diagram

Sp X,s  
K∗ (X <s , G) /

K∗ N0s , G ,

Sps−1
X,s
K∗ (X s−1 , G) /

K∗ (Ns,s−1, G)

where the colums are restriction homomorphisms, is commutative (Propo-

sition 3.7); hence it suffices to show that the restriction homomorphism
 
K∗ N0s , G −→ K∗ (Ns,s−1, G)
is injective. To prove this we may suppose that the action of G on X s is
connected, that is, X s is not a nontrivial disjoint union of open invariant
subspaces. In this case the toral component of the isotropy group of a point
of X s is constant. 
Set E = Ns , and consider the eigenspace decomposition E = χ∈T E χ .

We obtain a decomposition E = i E i by grouping together E χ and E χ
18 G. Vezzosi, A. Vistoli

when the characters χ and χ are multiple of a common primitive character

. Then clearly a geometric point of E is in E s−1 if and only if exactly
in T 
one of its components according to the decomposition E = i E i above
is nonzero. In other words, Ns,s−1 is the disjoint union i E i0 , where E i0
is embedded in E by setting all the other components equal to 0. The
same argument as in the proof of Lemma 4.8 shows that the kernel of the
pullback K∗ (X s , G) → K∗ (E i0 , G) is generated by λ−1 E i , so the kernel of
the pullback
 
K∗ (X s , G) −→ K∗ (Ns,s−1 , G) = K∗ E i0 , G
i

equals ∩i (λ−1 E i ); hence we need to show that (λ−1 E) = ∩i (λ−1 E i ).

This is done as follows. Choose a splitting G = D × T : we have
K∗ (X s , G) = K∗ (X s , D) ⊗ RT , as in the
beginning of the proof of
Lemma 4.2. First of all, we have λ−1 E = i λ−1 E i . Furthermore, for
each i we can choose a primitive character χi in T  such that all the charac-
ters which appear in the decomposition of E i are multiples of χi ; from this
we see that λ−1 E i is of the form nk=m
i
r χ k , where m i and n i are (possibly
i i,k i
negative) integers, ri,k ∈ K0 (X s , G), ri,m i and ri,ni are invertible. Then the
conclusion of the proof follows from the following fact.
Lemma 4.9. Let A be a ring, H a free finitely generated abelian group,
χ1 , . . . , χr linearly independent elements of H. Let γ1 , . . . , γr be elements
of the group ring AH of the form γi = nk=m i
r χ k , where the ri,k are
i i,k i
central elements of A such that ri,m i and ri,ni are invertible. Then we have
an equality of ideals (γ1 . . . γr ) = (γ1 ) ∩ . . . ∩ (γr ) in AH.
−1 −m i
Proof. By multiplying each γi by ri,m χ
i i
we may assume that γi has the
si
form 1 + ai,1 χi + · · · + ai,si χi with si ≥ 0 and ai,si a central unit in A.
We will show that for any i  = j the relation γi | qγ j , q ∈ AH implies
γi | q; from this the thesis follows with a straightforward induction. We
may assume that r = 2, i = 1 and j = 2.
Since χ1 , χ2 are Z-linearly independent elements of H, we may complete
them to a maximal Z-independent sequence χ1 , . . . , χn of H; this sequence
generates a subgroup H ⊆ H of finite index.  
Suppose at first that H = H, so that AH = A χ1±1 , ..., χn±1 . Replacing
   
A by A χ3±1 , . . . , χn±1 , we may assume that AH = A χ1±1 , χ2±1 .
If pγ1 = qγ2 , we can multiply this equality by a sufficiently high power
of χ1 χ2 and assume that p and q are polynomials in A[χ1 , χ2 ]. Since γ2
is a polynomial in A[χ2 ] with central coefficients and invertible leading
coefficient, the usual division algorithm allows us to write p = sγ2 + r ∈
A[χ1 , χ2 ], where r is a polynomial whose degree in χ2 is less than s2 =
degχ2 γ2 . By comparing the degrees in χ2 in the equality rγ1 = (q − sγ1 )γ2
we see that q − sγ1 must be zero, and this proves the result.
In the general case, choose representatives u 1 , . . . , u r for the cosets of
H in H; then any element f of AH can be written uniquely as ri=1 u i f i
Higher algebraic K-theory for actions of diagonalizable groups 19

 
with f i ∈ AH . Then from the equality i u i pi γ1 = i u i qi we get
pi γ1 = qi γ2 for all i, because γ1 and γ2 are in AH ; hence the thesis follows
from the previous case.

5. Actions with enough limits

Let us start with some preliminaries in commutative algebra.

5.1. Sufficiently deep modules. Let A be a finitely generated flat Cohen–

Macaulay Z-algebra, such that each of the fibers of the morphism Spec A →
Spec Z has pure dimension n. If V is a closed subset of Spec A, we define
the fiber dimension of V to be the largest of the dimensions of the fibers of
V over Spec Z, and its fiber codimension to be n minus its fiber dimension.
We say that V has pure fiber dimension if all the fibers of V have constant
dimension at all points of V (of course some of the fibers may be empty).
The fiber dimension and codimension of an ideal in A will be the fiber
dimension and codimension of the corresponding closed subset of Spec A.

Definition 5.1. Let M be an A-module. Then we say that M is sufficiently

deep if the following two conditions are satisfied.
(i) All associated primes of M have fiber codimension 0.
(ii) Ext1A (A/p, M) = 0 for all primes p in A of fiber codimension at least 2.

Here are the properties that we need.

Proposition 5.2.
(i) If 0 → M → M → M → 0 is an exact sequence of A-modules,
M and M are sufficiently deep, then M is sufficiently deep.
(ii) Direct limits and direct sums of sufficiently deep modules are suffi-
ciently deep.
(iii) If N is an abelian group, then N ⊗Z A is sufficiently deep.
(iv) If M is a sufficiently deep A-module, then Ext1A (N, M) = 0 for all
A-modules N whose support has fiber codimension at least 2.

Proof. Part (i) is obvious.

Part (ii) follows from the fact that A is noetherian, so formation of
Ext1A (A/p, −) commutes with direct sums and direct limits.
Let us prove part (iii). From part (ii) we see that we may assume that N
is cyclic. If N = Z, then M = A, and the statement follows from the facts
that A is Cohen–Macaulay, and that the height of a prime ideal is at least
equal to its fiber codimension.
Assume that N = Z/m Z, so that M = A/m A. The associated primes of
M are the generic points of the fibers of A over the primes dividing m, so
condition 5.1 (i) is satisfied.
20 G. Vezzosi, A. Vistoli

Take a prime p of A of fiber codimension at least 2, and consider the

exact sequence
0 = Ext1A (A/p, A) → Ext1A (A/p, A/m A) → Ext2A (A/p, A)
m
→ Ext2A (A/p, A).
If the characteristic of A/p is positive, then the height of p is at least 3,
so Ext2A (A/p, A) = 0, because A is Cohen–Macaulay, and we are done.
Otherwise, we have an exact sequence
m
0 −→ A/p −→ A/p −→ A/((m) + p) −→ 0;
but the height of (m) + p is at least 3, so Ext2A (A/((m) + p), A) = 0. From
this we deduce that multiplication by m is injective on Ext2A (A/p, A), and
this concludes the proof of part (iii).
For part (iv), notice first of all that if N is a finitely generated A-
module of fiber codimension at least 2 then we can filter N with successive
quotients of type A/p, where p is a prime of fiber codimension at least 2, so
Ext1A (N, M) = 0.
If N is not finitely generated and 0 → M → E → N → 0 is an exact
sequence of A-modules, N is a finitely generated submodule of N, and E is
the pullback of E to N , then the sequence 0 → M → E → N splits; but
because of part (i) of the definition we have Hom A (N , M) = 0, hence there
is a unique copy of N inside E . Hence there is a unique copy of N inside E,
and the sequence splits. This completes the proof of the proposition.

5.2. Sufficiently deep actions. Let G be a diagonalizable group scheme of

finite type over S; all the actions will be upon noetherian separated regular
algebraic spaces, as in our setup. The ring of representations RG = ZG 
is a finitely generated flat Cohen–Macaulay Z-algebra, and each of the
fibers of the morphism Spec RG → Spec Z has pure dimension equal to the
dimension of G.
Definition 5.3. We say that the action of G on X is sufficiently deep when
the RG-module K∗ (X, G) is sufficiently deep.
Theorem 5.4. Suppose that a diagonalizable group scheme of finite type
G acts on a noetherian regular separated algebraic space X, and that the
action is sufficiently deep. Then the restriction homomorphism K ∗ (X, G) →
K∗ (X G 0 , G) is injective, and its image is the intersection of the images of the
restriction homomorphisms K∗ (X T , G) → K∗ (X G 0 , G), where T ranges
over all subtori of G of codimension 1.
Proof. We need some preliminaries.
Lemma 5.5. Suppose that G acts on X with stabilizers of constant dimen-
sion s. Then the support of K∗ (X, G) as an RG-module has pure fiber
dimension s, and any associated prime of K∗ (X, G) has pure fiber dimen-
sion s.
Higher algebraic K-theory for actions of diagonalizable groups 21

Proof. Suppose first of all that s is 0. Then it follows easily from Thoma-
son’s localization theorem ([Tho92]) that the support of K∗ (X, G) has fiber
dimension 0, and from this we see that every associated prime must have
fiber dimension 0.
In the general case, we may assume that the action is connected (that is,
X is not a nontrivial disjoint union of open invariant subspaces); then there
will be a splitting G = H ×S T , where H is a diagonalizable group scheme
of finite type acting on X with finite stabilizers, and T is a totally split torus
that acts trivially on X. In this case
K∗ (X, G) = K∗ (X, H ) ⊗Z RT = K∗ (X, H ) ⊗RH RG.
The proof is concluded by applying the following lemma.
Lemma 5.6. Let A be a flat Cohen–Macaulay Z-algebra of finite type,
A → B a smooth homomorphism of finite type with fibers of pure dimen-
sion s. Suppose that M is an A-module whose support has fiber dimension 0.
Then M ⊗ A B has support of pure dimension s, and each of its associated
primes has fiber dimension s.
Proof. Since tensor product commutes with taking direct limits and B is
flat over A, we may assume that M is of finite type over A. By an obvious
filtration argument, we may assume that M is of the form A/p, where p is
a prime ideal of fiber dimension 0. In this case the only associated primes
of M ⊗ A B are the generic components of the fiber of Spec B over p, and
this proves the result.
Lemma 5.7. Suppose that X and Y are algebraic spaces on which G acts
with stabilizers of constant dimension respectively s and t. If N is an RG-
submodule of K ∗ (Y, G) and t < s, then there is no nontrivial homomorphism
of RG-modules from N to K∗ (X, G).
Proof. Given such a nontrivial homomorphism N → K∗ (X, G), call I
its image. The support of I has fiber dimension at most t, so there is an
associated prime of fiber dimension at most t in K∗ (X, G), contradicting
Lemma 5.5.
Now we prove Theorem 5.4. Let n be the dimension of G, so that
X n = X G 0 . First of all, let us show that the natural projection
K∗ (X, G) −→ K∗ (X n , G) × K∗ (X n−1 , G)
K∗ (Nn,n−1 ,G)

is an isomorphism. This will be achieved by showing that for all s with

0 ≤ s ≤ n − 1 the natural projection K∗ (X, G) −→ Ps is an isomorphism,
where we have set
Ps = K∗ (X n , G) × K∗ (X n−1 , G) ×
K∗ (Nn,n−1 ,G) K∗ (Nn−1,n−2 ,G)

··· × K∗ (X s , G).
K∗ (Ns+1,s ,G)
22 G. Vezzosi, A. Vistoli

For s = 0 this is our main Theorem 4.5, so we proceed by induction. If

s < n − 1 and the projection above is an isomorphism, we have an exact
sequence
0 −→ K∗ (X, G) −→ Ps+1 × K∗ (X s , G) −→ K∗ (Ns+1,s , G),
where the last arrow is the difference of the composition of the projection
Ps+1 → K∗ (X s+1 , G) with the pullback K∗ (X s+1 , G) → K∗ (Ns+1,s , G),
and of the specialization homomorphism K∗ (X s , G) → K∗ (Ns+1,s , G).
If we call N the image of this difference, we have an exact sequence of
RG-modules
0 −→ K∗ (X, G) −→ Ps+1 × K∗ (X s , G) −→ N → 0,
and the support of N is of fiber dimension at most s ≤ n − 2 by Lemma 5.5,
hence it is of fiber codimension at least 2. It follows from the fact that
K∗ (X, G) is sufficiently deep and from Proposition 5.2 (iv) that this se-
quence splits. From the fact that K∗ (X, G) has only associated primes of
fiber dimension 0 we see that the pullback map K∗ (X s , G) → N must be in-
jective, and from Lemma 5.7 that a copy of N living inside Ps+1 ×K∗ (X s , G)
must in fact be contained in K∗ (X s , G); this implies that the projection
K∗ (X, G) → Ps+1 is an isomorphism.
So the projection
K∗ (X, G) −→ K∗ (X n , G) ×K∗ (Nn,n−1 ,G) K∗ (X n−1 , G)
is an isomorphism. Then the kernel of the specialization homomorphism
from K∗ (X n−1 , G) to K∗ (Nn,n−1 , G) maps injectively in K∗ (X, G), so it
must be 0, again because K∗ (X, G) has only associated primes of fiber
T
dimension 0. Furthermore X n−1 is the disjoint union of the X n−1 when
T ranges over all finite subtori of G of codimension 1, and similarly
for Nn,n−1 . On the other hand, because of our main theorem applied to
the action of G on X T , we have the natural isomorphism K∗ (X T , G) →
K∗ (X G 0 , G) ×K∗ (Nn,n−1
T ,G) K∗ (X n−1 , G), and this completes the proof of
T

Theorem 5.4.

5.3. Actions with enough limits are sufficiently deep. For the rest of this
section S will be the spectrum of a field k, G is a diagonalizable group
scheme of finite type acting on a smooth separated scheme X of finite type
over k; call M the group of one-parameter subgroups Gm,k → G of G. There
is a natural Zariski topology on M  Zn in which the closed subsets are the
loci of zeros of sets of polynomials in the symmetric algebra Sym•Z M ∨ ; we
refer to this as the Zariski topology on M.
We will denote, as usual, by G 0 the toral component of G. If n is the
dimension of G, then X n = X G 0 . Furthermore, if we choose a splitting
G  G 0 × G/G 0 we obtain an isomorphism of rings
 
K∗ (X G 0 , G)  K∗ X G 0 , G/G 0 ⊗ RG 0
([Tho86b, Lemme 5.6]).
Higher algebraic K-theory for actions of diagonalizable groups 23

Definition 5.8. Suppose that k is algebraically closed. Consider a one pa-

rameter subgroup H = Gm,k → G, with the corresponding action of Gm,k
on X. We say that this one parameter subgroup admits limits if for every
closed point x ∈ X, the morphism Gm,k → X which sends t ∈ G to tx
extends to a morphism A1 → X. The image of 0 ∈ A1 (k) in X is called the
limit of x for the one parameter subgroup H.
We say that the action of G on X admits enough limits if the one
parameter subgroups of G which admit limits form a Zariski-dense subset
of M.
If k is not algebraically closed, then we say that the action admits enough
limits if the action obtained after base change to the algebraic closure of k
does.
Remark 5.9. One can show that the locus of 1-parameter subgroups of G
admitting limits is defined by linear inequalities, so the definition can be
stated in more down to earth terms (we are grateful to the referee for pointing
this out).
The notion of action with enough limits is a weakening of the notion of
filtrable action due to Brion. More precisely, an action has enough limits if it
satisfies condition (i) in [Bri97, Definition 3.2]; there is also a condition (ii)
on closures of strata.
The main case when the action admits enough limits is when X is
complete; in this case of course every one-parameter subgroup admits limits.
Another case is when the action of G 0 = Gnm,k on X extends to an action
of the multiplicative monoid An . Also, we give a characterization of toric
varieties with enough limits in Proposition 6.7.
Theorem 5.10. Suppose that a diagonalizable group scheme of finite type
G over a perfect field k acts on a smooth separated scheme of finite type X
over k. If the action of G admits enough limits, then it is sufficiently deep.
By putting this together with Theorem 5.4 we get the following.
Corollary 5.11. Suppose that a diagonalizable group scheme of finite type
G over a perfect field k acts on a smooth separated scheme of finite type
X over k. If the action of G admits enough limits, then the restriction
homomorphism
K∗ (X, G) −→ K∗ (X G 0 , G)
is injective, and its image is the intersection of the images of the restric-
tion homomorphisms K∗ (X T , G) → K∗ (X G 0 , G), where T ranges over all
subtori of G of codimension 1.
For example, consider the following situation, completely analogous
to the one considered in [Bri98, Corollary 7] and in [G-K-MP98]. Let G
be an n-dimensional torus acting on a smooth complete variety X over an
24 G. Vezzosi, A. Vistoli

algebraically closed field k. Assume that the fixed point set X G 0 = X n is

zero-dimensional, while X n−1 is 1-dimensional. Set X G 0 = {x1 , . . . , xt },
and call P1 , . . . , Pr the closures in X of the connected components of X n−1 .
Then each P j is isomorphic to P1 , and contains precisely two of the fixed
points, say xi and xi . Call D j the kernel of the action of G on P j ; then the
image of the restriction homomorphism
K∗ (P j , G) → K∗ (xi , G) × K∗ (xi , G) = K∗ (k) ⊗ RG × K∗ (k) ⊗ RG
consists of the pairs of elements
(α, β) ∈ K∗ (k) ⊗ RG × K∗ (k) ⊗ RG
whose images in K∗ (k) ⊗ RD j coincide (this follows immediately from
Theorem 4.5). From this and from Corollary 5.11 we get the following.
Corollary 5.12. In the situation above, the restriction map
t n
K∗ (X, G) −→ K∗ (xi , G) = K∗ (k) ⊗ RG
i=1 i=1

is injective. Its image consist of all elements (αi ) such that if xi and xi are
contained in some P j , then the restrictions of αi and αi to K∗ (k) ⊗ RD j
coincide.
Theorem 5.10 is proved in the next subsection.

5.4. Białynicki-Birula stratifications. Let us prove Theorem 5.10: like

in [Bri97], the idea is to use a Białynicki-Birula stratification. We will prove
the following.
Proposition 5.13. Suppose that a diagonalizable group scheme of finite
type G over a perfect field k acts with enough limits on a smooth separated
scheme of finite type X over k. Then the RG-module K∗ (X, G) is obtained
by taking finitely many successive extensions of RG-modules of the form
N ⊗Z RG, where N is an abelian group.
Theorem 5.10 follows from this, in view of Proposition 5.2, parts (i) and
(iii). Let us prove the proposition.
First of all, let us assume that k is algebraically closed. We will only
consider closed points, and write X for X(k).
It is a standard fact that the one-parameter subgroups H = Gm,k → G 0
with the property that X G 0 = X H form a nonempty Zariski open subset of
M, so we can choose one with this property that admits enough limits. There
is a (discontinuous) function X → X G 0 sending each point to its limit. Let
T1 , . . . , Ts be the connected components of X G 0 ; call E i the inverse image
of Ti in X, and π j : E j → T j the restriction of the limit function. The
following is a fundamental result of Białynicki-Birula.
Higher algebraic K-theory for actions of diagonalizable groups 25

Theorem 5.14 (Białynicki-Birula). In the situation above:

(i) The E j are smooth locally closed G-invariant subvarieties of X.
(ii) The functions π j : E j → T j are G-invariant morphisms.
(iii) For each j there is a representation V j of H and an open cover {Uα } of
T j , together with equivariant isomorphisms π −1
j (Uα )  Uα × V j , such
−1
that the restriction π j : π j (Uα ) → Uα corresponds to the projection
Uα × V j → Uα .
(iv) If x is a point of T j , then the normal space to E j in X at x is the sum
of the negative eigenspaces in the tangent space to X at x under the
action of H.

Of course in part (iii) we may take V j to be the normal bundle to E j in

X at any point of T j .
This theorem is proved in [Bia73]; we should notice that the condition
that X is covered by open invariant quasiaffine subsets is always verified,
thanks to a result of Sumihiro ([Sum75]).
Now we remove the hypothesis that k is algebraically closed: here is the
variant of Białynicki-Birula’s theorem that we need.

Theorem 5.15. Suppose that a diagonalizable group scheme of finite type

G over a perfect field k acts with enough limits on a smooth separated
scheme of finite type X over k. Let Y1 , . . . , Yr be the connected components
of X G 0 ; there exists a stratification X 1 , . . . , X r of X in locally closed
G-invariant smooth subvarieties, together with G-equivariant morphisms
ρi : X i → Yi , such that:
(i) X i contains Yi for all i, and the restriction of ρi to Yi is the identity.
(ii) If U is an open affine subset of Yi and NU is the restriction of the
normal bundle NYi X i to U, then there is a G-equivariant isomorphism
ρi−1 (U )  NU of schemes over U.
(iii) In the eigenspace decomposition of the restriction of N X i X to Yi , the
subbundle corresponding to the trivial character of G 0 is 0.

Proof. Let X = X ×Spec k Spec k, and call Γ the Galois group of k over k.
Choose a one parameter subgroup H = Gm,k → G as before. Let T1 , . . . , Ts
G0
be the connected components of X , π j : E j → T j as in Białynicki-
Birula’s theorem. The Yi correspond to the orbits of the action of Γ on
{T1 , . . . , Ts }; obviously Γ also permutes the E j , so we let X 1 , . . . , X r
be the smooth subvarieties of X corresponding to the orbits of Γ on
{E 1 , . . . , E s }. The π j : E j → T j descend to morphisms X i → Yi . Prop-
erties (i) and (iii) are obviously satisfied, because they are satisfied after
passing to k.
We have to prove (ii). Let E be the inverse image of U in X i , I the
ideal of U in the algebra k[E]. Because U is affine, I/I 2 is a projec-
tive k[U]-module, and G is diagonalizable, the projection I → I/I 2 has
26 G. Vezzosi, A. Vistoli

a k[U]-linear and G-equivariant section I/I 2 → I . This induces a G-

equivariant morphism of U-schemes E → NU , sending U to the 0-section,
whose differential at the zero section is the identity (notice that NU is
also the restriction to U of the relative tangent bundle T X/U ). We want
to show that this is an isomorphism; it is enough to check that this is
true on the fibers, so, let V be one of the fibers of NU on some point
p ∈ U. According to part (ii) of the theorem of Białynicki-Birula, the
fiber of X on p is H-equivariantly isomorphic to V ; hence an appli-
cation of the following elementary lemma concludes the proof of Theo-
rem 5.15.
Lemma 5.16. Suppose that Gm,k acts linearly with positive weights on
a finite dimensional vector space V over a field k. If f : V → V is an equiv-
ariant polynomial map whose differential at the origin is an isomorphism,
then f is also an isomorphism.
Proof. First all, notice that because of the positivity of the weights, we
have f(0) = 0. By composing f with the inverse of the differential of
f at the origin, we may assume that this differential of f is the identity.
Consider the eigenspace decomposition V = V1 ⊕ V2 ⊕ · · · ⊕ Vr , where
Gm,k acts on Vi with a character t → t m i , and 0 < m 1 < m 2 < · · · < m r .
Choose a basis of eigenvectors of V ; we will use groups of coordinates
x1 , . . . , xr , where xi represents the group of elements of the dual basis
corresponding to basis elements in Vi , so that the action of Gm,k is described
by t · (x1 , . . . , xr ) = (t m 1 x1 , . . . , t mr xr ). Then it is a simple matter to verify
that f is given by a formula of the type
f(x1 , . . . , xr )
= (x1 , x2 + f 2 (x1 ), x3 + f 3 (x1 , x2 ), . . . , xr + fr (x1 , . . . , xr−1 ))
and that every polynomial map of this form is an isomorphism.
Now let us show that Theorem 5.15 implies Proposition 5.13. First of all,
Theorem 5.15 (ii) and a standard argument with the localization sequence
imply that the pullback map K∗ (Yi ) ⊗Z RG = K∗ (Yi , G) → K(X i , G) is
an isomorphism.
Now, let us order the strata X 1 , . . . , X r by decreasing dimension, and
let us set Ui = X 1 ∪ . . . ∪ X i . Clearly X i is closed in Ui . We claim that X i
is K-rigid in Ui for all i. In fact, it is enough to show that the restriction of
λ−1 (N∨X i X ) to Yi is not a zero-divisor, and this follows from Lemma 4.2 and
Theorem 5.15 (iii).
Then by Proposition 4.3 (i) we have an exact sequence
0 −→ K∗ (X i , G) −→ K∗ (Ui , G) −→ K∗ (Ui−1 , G) −→ 0;
so each K∗ (Ui , G) is obtained by finitely many successive extensions of
RG-modules of the form N ⊗Z RG, and X = Ur . This concludes the proof
of Proposition 5.13, and of Theorem 5.10.
Higher algebraic K-theory for actions of diagonalizable groups 27

5.5. Comparison with ordinary K-theory for torus actions with enough
limits. Assume that T is a totally split torus over a perfect field k, acting
on a separated scheme X of finite type over k. We write T instead of G for
conformity with the standard notation.
The following is a consequence of Proposition 5.13.
Corollary 5.17. If X is smooth and the action has enough limits, we have
 
TorRT
p K ∗ (X, T ), Z = 0 for all p > 0.
The interest of this comes from the following result of Merkurjev.
Theorem 5.18 ([Mer97, Theorem 4.3]). There is a homology spectral se-
quence
 
E 2pq = TorRT
p Z, K q (X, T ) ⇒ K p+q (X )
such that the edge homomorphisms
Z ⊗RT Kq (X, T ) −→ Kq (X )
are induced by the forgetful homomorphism K∗ (X, T ) → K∗ (X ).
In particular the ring homomorphism Z ⊗RT K0 (X, T ) → K0 (X ) is an
isomorphism.
Furthemore, if X is smooth and projective we have E 2pq = 0 for all p > 0,
so the homomorphism Z ⊗RT K∗ (X, T ) → K∗ (X ) is an isomorphism.
More generally, Merkurjev produces his spectral sequence for actions
of reductive groups whose fundamental group is torsion-free.
From Corollary 5.17 we get the following extension of Merkurjev’s
degeneracy result.
Theorem 5.19. Suppose that T is a totally split torus over a perfect field k,
acting with enough limits on a smooth scheme separated and of finite type
over k. Then the forgertful homomorphism K∗ (X, T ) → K∗ (X ) induces an
isomorphism
∼
Z ⊗RT K∗ (X, T ) −→ K∗ (X ).

6. The K-theory of smooth toric varieties

Our reference for the theory of toric varieties will be [Ful93b].

In this section we take T to be a totally split torus over a fixed field k,
∨
N = Hom(Gm,k , T ) = T
its lattice of one-parameter subgroups, ∆ a fan in N ⊗ R, X = X(∆) the
associated toric variety. We will always assume that X is smooth; this is
equivalent to saying that every cone in ∆ is generated by a subset of a basis
of N.
28 G. Vezzosi, A. Vistoli

We will give two different descriptions of the equivariant K-theory ring

of X, one as a subring of a product of representation rings, and the second by
generators and relations, analogously to what has been done for equivariant
cohomology in [B-DC-P90].

6.1. The equivariant K-theory ring as a subring of a product of rings

of representations. There is one orbit Oσ of T on X for each cone σ ∈ ∆,
containing a canonical rational point xσ ∈ Oσ (k). The dimension of Oσ is
def
the codimension codim σ = dim T − dim σ , and the stabilizer of any of
its geometric points is the subtorus Tσ ⊆ T whose group of one-parameter
subgroups is precisely the subgroup σ  = σ + (−σ) ⊆ N; the dimension
of Tσ is equal to the dimension of σ (see [Ful93b, 3.1]). Hence X s is the
disjoint union of the orbits Oσ = T/Tσ with dim σ = s.
Given a cone σ in N ⊗ R, we denote by ∂σ the union of all of its faces
of codimension 1.
Since
K∗ (Oσ , T ) = K∗ (T/Tσ , T ) = K∗ (Spec k, Tσ ) = K∗ (k) ⊗ RTσ
we have that
K∗ (X s , T ) = K∗ (Oσ , T ) = K∗ (k) ⊗ RTσ .
s σ∈∆ σ∈∆

Lemma 6.1. Fix an positive integer s. Then there is a canonical isomorph-

ism
Ns,s−1  Oτ .
σ∈∆ τ∈∂σ
dim σ=s

Furthermore, for each pair σ , τ such that σ has dimension s, τ has

dimension s − 1, and τ is a face of σ , the composition of the specialization
homomorphism

X,s : K∗ (X s−1 , T ) =
Sps−1 K∗ (Oτ , T )
τ∈∆
dim τ=s−1

−→ K∗ (Ns,s−1, T ) = K∗ (Oτ , T ),
σ∈∆ τ∈∂σ
dim σ=s

with the projection

prσ,τ : K∗ (Oτ , T ) −→ K∗ (Oτ , T )
σ∈∆ τ∈∂σ
dim σ=s

is the projection
prτ : K∗ (Oτ , T ) −→ K∗ (Oτ , T ).
τ∈∆
dim τ=s−1
Higher algebraic K-theory for actions of diagonalizable groups 29

Proof. We use the same notation as in [Ful93b]; in particular, for each cone
σ of the fan of X, we denote by Uσ the corresponding affine open subscheme
of X.
First of all, assume that the fan ∆ consists of all the faces of an s-
dimensional cone σ . Call B a part of a basis of N that spans σ : we have
an action of Tσ on the k-vector space Vσ generated by B, by letting each
1-parameter subgroup Gm → G in B act by multiplication on the corres-
ponding line in Vσ , and an equivariant embedding Tσ ⊆ Vσ . Then X = Uσ
is T -equivariantly isomorphic to the T -equivariant vector bundle
T ×Tσ Vσ = (T × Vσ )/Tσ −→ Oσ = T/Tσ
in such a way that the zero section corresponds to Oσ ⊆ Uσ . Since X s =
Oσ and Uσ is a vector bundle over Oσ , we get a canonical isomorphism
Uσ  Ns , and from this a canonical isomorphism
Ns,s−1  (Uσ )s−1 = Oτ .
τ∈∂σ
It follows that the deformation to the normal bundle of Oσ in Uσ is also
isomorphic to the product Uσ ×k P1 , and from this we get the second part
of the statement.
In the general case we have X s = dim σ=s Oσ , and if σ is a cone of
dimension s in ∆, the intersection of X s with Uσ is precisely Oσ . From this
we get the first part of the statement in general.
The second part follows by applying the compatibility of specializa-
tions to the morphism of deformation to the normal bundle induced by the
equivariant morphism dim σ=s Uσ → X s .
Using this lemma together with Theorem 4.5 we get that K∗ (X, T ) is
the subring of
K∗ (Oσ , T ) = K∗ (k) ⊗ RTσ
σ∈∆ σ∈∆
consisting of elements (aσ ) with the property that the restriction of aσ ∈
K∗ (k) ⊗ RTσ to K∗ (k) ⊗ RTτ coincides with aτ ∈ K∗ (k) ⊗ RTτ every
time τ is a face of codimension 1 in σ . Since every face of a cone is
contained in
a face of codimension 1, this can also be described as the
subring of σ K∗ (k) ⊗ RTσ consisting of elements (aσ ) with the property
that the restriction of aσ ∈ K∗ (k) ⊗ RTσ to K∗ (k) ⊗ RTτ coincides with
aτ ∈ K∗ (k) ⊗ RTτ every time τ is a face of σ . But every cone in ∆ is
contained in a maximal cone in ∆, so we get the following description of
the equivariant K-theory of a smooth toric variety.
Theorem 6.2. If X(∆) is a smooth toric variety associated with a fan ∆ in
N ⊗ R, there is an injective homomorphism of RT -algebras
K∗ (X(∆), T ) → K∗ (k) ⊗ RTσ ,
σ∈∆max

where ∆max is the set of maximal cones in ∆.

30 G. Vezzosi, A. Vistoli

An element (aσ ) ∈ σ K∗ (k)⊗RTσ is in the image of this homomorphism
if and only if for any two maximal cones σ1 and σ2 , the restrictions of aσ1
and aσ2 to K∗ (k) ⊗ RTσ1 ∩σ2 coincide.
This description of the equivariant K-theory ring of X is analogous to
the description of its equivariant cohomology in [B-DC-P90], and of its
equivariant Chow ring in [Bri97, Theorem 5.4].

6.2. The multiplicative Stanley–Reisner presentation. From the descrip-

tion above it is easy to get a presentation of the equivariant K-theory ring
of the smooth toric variety X(∆), analogous to the Stanley–Reisner presen-
tation of its equivariant cohomology ring obtained in [B-DC-P90]. Denote
by ∆1 the subset of ∆ consisting of 1-dimensional cones. We will use the
following notation: if σ ∈ ∆max , call Nσ ⊆ N the group of 1-parameter
subgroups of Tσ , so that T σ = (Nσ )∨ . We will use multiplicative notation
for Tσ . Furthermore, for any ρ ∈ ∆1 we call vρ ∈ N the generator for the
monoid ρ ∩ N.

For each ρ ∈ ∆1 we define an element u ρ of the product σ∈∆max T σ ,
as follows. If ρ is not a face of σ , we set (u ρ )σ = 1. If ρ ∈ ∆1 is a face
of σ ∈ ∆max and {ρ = ρ1 , ρ2 . . . , ρt } is the set of 1-dimensional faces of
σ , then, since the variety X(∆) is smooth, we have that vρ1 , . . . , vρt form
σ = Nσ∨ , we set u ρ = vρ∨ .
a basis for Nσ . If vρ∨1 , . . . , vρ∨t is the dual basis in T

1
Denote by V∆ ⊆ σ∈∆max T σ the subgroup consisting of the elements
(xσ ) with the property that for all σ1 , σ2 in ∆max the restrictions of xσ1 ∈ T σ1
and xσ2 ∈ T σ2 to Tσ1 ∩σ2 coincide. Then we have the following fact.
Proposition 6.3. The elements u ρ form a basis of V∆ .
The proof is straightforward.
We have the inclusions
σ ⊆
T RTσ ⊆ K∗ (k) ⊗ RTσ ;
σ∈∆max σ∈∆max σ∈∆max
 
because

of the description of the ring K ∗ X(∆), T as a subring of
σ∈∆max K∗ (k) ⊗ RTσ given in Theorem 6.2, we see that we can consider
the u ρ as elements of K∗ (X(∆), T ).
There are some obvious relations that the u ρ satisfy in

K∗ (X(∆), T ) ⊆ K∗ (k) ⊗ RTσ .

σ∈∆max

Suppose that S is a subset of ∆1 not contained in any maximal cone of ∆.

σ ;
Then for all σ in ∆max there will be some ρ ∈ S such that (u ρ )σ = 1 in T
hence we have the relation
(u ρ − 1) = 0 in K∗ (X(∆), T ) ⊆ K∗ (k) ⊗ RTσ .
ρ∈S σ∈∆max
Higher algebraic K-theory for actions of diagonalizable groups 31

From this we get a homomorphism of K∗ (k)-algebras

 
K∗ (k) xρ±1
(6.1) 
 −→ K∗ (X(∆), T )
ρ∈S (x ρ − 1)

where the xρ are indeterminates, where ρ varies over ∆1 , and S over the
subsets of ∆1 whose elements are not all contained in a maximal cone in ∆,
by sending each xρ to u ρ .
Theorem 6.4. Suppose that X(∆) is a smooth toric variety associated with
a fan ∆ in N ⊗ R. Then the homomorphism (6.1) above is an isomorphism.
Proof. First of all, let us show that the u ρ and their inverses generate
K∗ (X(∆), T ) ⊆ K∗ (k) ⊗ RTσ .
σ∈∆max

Set ∆max = {σ1 , . . . , σr }, and let α be an element of K∗ (X(∆), T ); we want

to show that α can be expressed as a Laurent polynomial in the xρ evaluated
in the u ρ . The ring K∗ (k) ⊗ RTσ1 is a ring of Laurent polynomials in the
images of the u ρ with ρ ⊆ σ1 , so we can find a Laurent polynomial p1 (xρ ),
in which only the xρ with ρ ⊆ σ1 appear, such that the image of p1 (u ρ ) in
K∗ (k) ⊗ RTσ1 equals the image of α in the same ring. By subtracting p1 (u ρ ),
we see that we may assume that the projection of α into K∗ (k) ⊗ RTσ1 is
zero.
Now, let us repeat the procedure for the maximal cone σ2 : find a poly-
nomial p1 (xρ ), in which only the xρ with ρ ⊆ σ2 appear, such that the image
of p2 (u ρ ) in K∗ (k) ⊗ RTσ2 equals the image of α in the same ring. The key
point is that the restriction of α to K∗ (k) ⊗ RTσ1 ∩σ2 is zero, so in fact p2 (xρ )
can only contain the variables xρ with ρ not in σ1 . Hence the restriction
of p2 (xρ ) to K∗ (k) ⊗ RTσ1 is also zero, and after having subtracted p2 (xρ )
from α we may assume that the restriction of α to both K∗ (k) ⊗ RTσ1 and
K∗ (k) ⊗ RTσ2 is zero. We can continue this process for the remaining cones
σ3 , . . . , σr ; at the end all the projections will be zero, and therefore α will
be zero too.
Now we have to show that the kernel of the homomorphism
 
(6.2) K∗ (k) xρ±1 −→ K∗ (k) ⊗ RTσ
σ∈∆max


sending each xρ to u ρ equals the ideal ρ∈S (x ρ − 1) , where S varies over
all subsets of ∆1 not contained in any maximal cone. The kernel of the
projection
 
K∗ (k) xρ±1 −→ K∗ (k) ⊗ RTσ
equals the ideal Iσ generated by the xρ − 1, where ρ varies over the set
of 1-dimensional cones in ∆1 not contained in σ , hence the kernel of
the homomorphism 6.2 is the intersection of the Iσ . The result is then
a consequence of the following lemma.
32 G. Vezzosi, A. Vistoli

Lemma 6.5. Let R be a (not necessarily commutative) ring, {xρ }ρ∈E a finite
set of central indeterminates; consider the ring of Laurent polynomials
R xρ±1 . Let A1 , . . . , Ar be subsets of E; for each j = 1, . . . , r call I j the
 
ideal of R xρ±1 generated by the elements xρ − 1 with ρ ∈ A j .
 ±1 
Then the intersection

I 1 ∩ · · · ∩ I r is the ideal of R xρ generated by
the elements ρ∈S (xρ − 1), where S varies over all subsets of E that meet
each A j .
Remark 6.6. When each A j contains a single element this is a particular
case of Lemma 4.9. The obvious common generalization should also hold.
Proof. We proceed by induction on r; the case r = 1 is clear. In general,
take p ∈ I1 ∩ · · · ∩ Ir ⊆ I2 ∩ · · · ∩ Ir ; by induction hypothesis, we can write
 
p= (xρ − 1) qS ,
S ρ∈S

where S varies over all subsets of E whose intersection with each of the
A2 , . . . , Ar is not empty. We can split the sum as
     
p= (xρ − 1) qS + (xρ − 1) qS ;
S∩A1  =∅ ρ∈S S∩A1 =∅ ρ∈S

the first summand is in I1 ∩ · · · ∩ Ir and is of the desired form, so we may

subtract it from p and suppose that p is of the type
  
p= (xρ − 1) qS .
S∩A1 =∅ ρ∈S
 
Now, consider the ring R xρ±1 ρ∈A of Laurent polynomials not involving
/ 1  
the variables in A1 ; it is a subring of R xρ±1 , and there is also a retraction
   
π : R xρ±1 −→ R xρ±1 ρ∈A / 1
,
sending

each xρ with ρ ∈ A1 to 1, whose kernel is precisely I1 . The elements
±1
ρ∈S (x ρ − 1) are in R x ρ ρ∈A / 1
, and
  
 
πp = (xρ − 1) πqS = 0 ∈ R xρ±1 ρ∈A / 1
S∩A1 =∅ ρ∈S

so we can write
  
p= (xρ − 1) (qS − πqS ).
S∩A1 =∅ ρ∈S

Then we write each qS −πqS ∈ I1 as a linear combination of the polynomials

xρ − 1 with ρ ∈ A1 ; this concludes the proofs of the lemma and of the
theorem.
Higher algebraic K-theory for actions of diagonalizable groups 33

6.3. Ordinary K-theory of smooth toric varieties. From Merkurjev’s

theorem (5.18) we get that K0 (X ) = Z ⊗RT K0 (X, T ). If the Merkurjev
spectral sequence degenerates then we also have K∗ (X ) = Z ⊗RT K∗ (X, T );
this gives a way to compute the whole K-theory ring of X.
In general, the spectral sequence will not degenerate, and the ring K∗ (X )
tends to be rather complicated (for example, when X = T ). When the toric
variety has enough limits we can apply Theorem 5.19. Using the description
of closures
  of orbits given in [Ful93b, 3.1], one shows that given a point
x ∈ X k lying in an orbit Oτ , and a one-parameter subgroup corresponding
to an element v ∈ N, then the point has a limit under the one parameter
subgroup if and only if v lies in the subset

(σ + τ) ⊆ N ⊗ R,
σ∈Star τ

where τ denotes the subvector space τ + (−τ) ⊆ N ⊗ R, and Star τ is the
set of cones in ∆ containing τ as a face. From this we obtain the following.

Proposition 6.7. X has enough limits if and only if the subset

 
(σ + τ) ⊆ N ⊗ R
τ∈∆ σ∈Star τ

has nonempty interior.

Remark 6.8. If X = X(∆) is a smooth toric variety with enough limits, the
K-theory ring of X can be described in a slightly more efficient fashion:
there is an injective homomorphism of RT -algebras

K∗ (X, T ) → K∗ (k) ⊗ RT,

σ∈∆
dim σ=dim T


and an element (aσ ) ∈ σ K∗ (k)⊗RT is in the image of this homomorphism
if and only if for any two adjacent maximal cones σ1 and σ2 , the restrictions
of aσ1 and aσ2 to K∗ (k) ⊗ RTσ1 ∩σ2 coincide.

Theorem 6.9. If X is a smooth smooth toric variety with enough limits,

then K0 (X, T ) is a projective module over RT of rank equal to the number
of maximal cones in its fan; furthermore the natural ring homomorphism
K∗ (k) ⊗ K0 (X, T ) → K∗ (X, T ) is an isomorphism.

In particular we have

p (K∗ (X, T ), Z) = 0 for all p > 0:

TorRT

so from Merkurjev’s theorem (5.18) we get the following.

34 G. Vezzosi, A. Vistoli

Corollary 6.10. Let X be a smooth toric variety with enough limits.

(i) The natural homomorphism of rings

Z ⊗RT K∗ (X, T ) −→ K∗ (X )

is an isomorphism of RT -algebras.
(ii) K0 (X ) is a free abelian group of rank equal to the number of maximal
cones in ∆.
(iii) The natural homomorphism K∗ (k)⊗K0 (X ) → K∗ (X ) is an isomorph-
ism.

Remark 6.11. Suppose that the base field is the field C of complex numbers.
The Merkurjev spectral sequence is an analogue of the Eilenberg–Moore
spectral sequence ([Ei-Mo62])
∗  
E 2 = Tor Hp,q(BT,Z) HT∗ (X, Z), Z ⇒ H p+q (X, Z).
p,q

Then [B-B-F-K02] contains a description of the fans of the simplicial toric

varieties for which this spectral sequence degenerates after tensoring with Q.
Presumably there should be a similar description for the case considered
here.

7. The refined decomposition theorem

The main result of [Ve-Vi] shows that if G is an algebraic group acting

with finite stabilizers on a noetherian regular algebraic space X over a field,
the equivariant K-theory ring of X, after inverting certain primes, splits as
a direct product of rings related with the K-theory of certain fixed points
subsets. For actions of diagonalizable groups it is not hard to extend this
decomposition to the case when the stabilizers have constant dimension.
So, in the general case when we do not assume anything about the
dimension of the stabilizers, this theorem gives a description of the K-
theory of each stratum X s ; it should clearly be possible to mix this with
Theorem 4.5 to give a result that expresses K∗ (X, G), after inverting certain
primes, as a fibered product. This is carried out in this section.
The material in this section is organized as follows. We first extend the
main result of [Ve-Vi] over an arbitrary noetherian separated base scheme
S (Theorem 7.1) by giving a decomposition theorem in the case of an
action with finite stabilizers of a diagonalizable group scheme G of finite
type over S on a noetherian regular separated algebraic space X over S.
Next, we deduce from this a decomposition theorem in the case where the
action of G on X has stabilizers of fixed constant dimension (Theorem 7.4).
Finally, we combine the analysis carried out in Sect. 4, and culminating with
Theorem 4.5, together with Theorem 7.4 to prove a general decomposition
theorem (Theorem 7.12) where no restriction is imposed on the stabilizers.
Higher algebraic K-theory for actions of diagonalizable groups 35

7.1. Actions with finite stabilizers. Here we recall the main result of
[Ve-Vi] for actions of diagonalizable groups, extending it over any noethe-
rian separated base scheme S.
Suppose that G is a diagonalizable group scheme of finite type acting
with finite stabilizers on a noetherian regular separated algebraic space over
a noetherian separated scheme S. A diagonalizable group scheme of finite
type σ over S is called dual cyclic if its Cartier dual is finite cyclic, that is,
if σ is isomorphic to a group scheme of the form µn,S for some positive
integer n.
A subgroup scheme σ ⊆ G is called essential if it is dual cyclic, and
X σ  = ∅.
There are only finitely many essential subgroups of G; we will fix
a positive integer N which is divisible by the least common multiple of their
orders.
Suppose that σ is a dual cyclic group of order n. The ring of representa-
tions Rσ is of the form Z[t]/(t n − 1), where t corresponds to a generator of
the group of characters  σ . Denote by 
Rσ the quotient of Rσ corresponding
to the quotient
 
Rσ = Z[t]/(t n − 1)
Z[t]/ Φn (t) ,

where Φn is the n th cyclotomic polynomial. This quotient 

Rσ is independent
of the choice of a generator for  σ . We have a canonical homomorphism
RG → Rσ
 Rσ induced by the embedding σ ⊆ G.
We also define a multiplicative system
Sσ ⊆ RG
as follows: an element of RG is in Sσ if its image in 
Rσ is a power of N.
For any RG-module M, we define the σ -localization Mσ of M to be
−1
Sσ M. Consider the σ -localization

K∗ (X σ , G)σ = S−1 σ
σ K∗ (X , G)

of the RG-algebra K∗ (X σ , G). The tensor product K∗ (X σ , G)σ ⊗ Q is the

localization
 
K∗ (X σ , G) ⊗ Q mσ

of the Rσ -algebra K∗ (X σ , G) ⊗ Q at the maximal ideal

mσ = ker(Rσ ⊗ Q

Rσ ⊗ Q).
We are particularly interested in the σ -localization when σ is the trivial
subgroup of G; in this case we denote it by K∗ (X, G)geom , and call it the
geometric equivariant K-theory of X. The localization homomorphism
K∗ (X, G) ⊗ Z[1/N] −→ K∗ (X, G)geom
36 G. Vezzosi, A. Vistoli

is surjective, and its kernel can be described as follows. Consider the kernel
p = ker rk of the localized rank homomorphism

rk : K∗ (X, G) ⊗ Z[1/N] −→ Z[1/N];

then the power pk is independent of k if k is large, and this power coincides

with the kernel of the localization homomorphism.
For each essential subgroup σ ⊆ G, consider the compositions

locσ : K∗ (X, G) ⊗ Z[1/N] −→ K∗ (X σ , G) ⊗ Z[1/N] −→ K∗ (X σ , G)σ ,

where the first arrow is a restriction homomorphism, and the second one is
the localization.
There is also a homomorphism of RG-algebras K∗ (X σ , G) →
K∗ (X σ , G)geom ⊗ 
Rσ , defined as the composition

K∗ (X σ , G) −→ K∗ (X σ , G × σ)  K∗ (X σ , G) ⊗ Rσ
−→ K∗ (X σ , G)geom ⊗ 
Rσ,

where the first morphism is induced by the multiplication G × σ → G, the

second one is a natural isomorphism coming from the fact that σ acts trivially
on X σ ([Ve-Vi, Lemma 2.7]), and the third one is obtained from the local-
ization homomorphism K∗ (X σ , G) → K∗ (X σ , G)geom and the projection
RG →  Rσ . Then the homomorphism K∗ (X σ , G) → K∗ (X σ , G)geom ⊗  Rσ
σ
factors through K∗ (X , G)σ ([Ve-Vi, Lemma 2.8]), inducing a homomorph-
ism

θσ : K∗ (X σ , G)σ −→ K∗ (X σ , G)geom ⊗ 
Rσ.

Theorem 7.1.

(i) There are finitely many essential subgroup schemes in G, and the homo-
morphism

locσ : K∗ (X, G) ⊗ Z[1/N] −→ K∗ (X σ , G)σ ,

σ σ

where the product runs over all the essential subgroup schemes of G,
is an isomorphism.
(ii) The homomorphism

θσ : K∗ (X σ , G)σ −→ K∗ (X σ , G)geom ⊗ 
Rσ

is an isomorphism of RG-algebras.
Higher algebraic K-theory for actions of diagonalizable groups 37

Proof. If the base scheme S is the spectrum of a field, this is a particular

case of the main theorem of [Ve-Vi].
If G is a torus, the proof of this statement given in [Ve-Vi] goes through
without changes, because it only relies on Thomason’s generic slice theorem
for torus actions ([Tho86a, Proposition 4.10]).
In the general case, choose an embedding G → T into some totally
split torus T over S, and consider the quotient space
def
Y = X ×G T = (X × T )/G
by the customary diagonal action of G; this exists as an algebraic space
thanks to a result of Artin ([La-MB00, Corollaire 10.4]). The same argument
as in the beginning of Sect. 5.1 of [Ve-Vi] shows that Y is separated.
Now observe that if σ ⊆ T is an essential subgroup relative to the action
of T on Y , we have Y σ = ∅ unless σ ⊆ G is an essential subgroup; hence
the least common multiples of the orders of all essential subgroups are the
same for the action of G on X and the action of T on Y .
Also, if σ ⊆ G is an essential subgroup we have Y σ = X σ ×G T ,
and therefore, by Morita equivalence ([Tho87, Proposition 6.2]), we get an
isomorphism
K∗ (Y σ , T )  K∗ (X σ , G)
which is an isomorphism of RT -algebras, if we view K∗ (X σ , G) as an
RT -algebra via the restriction homomorphism RT → RG.
Moreover, SσT ⊆ R(T ) is exactly the preimage of SσG ⊆ R(G) under the
natural surjection RT → RG; therefore we have compatible σ -localized
Morita isomorphisms
 T −1  −1
Sσ K∗ (Y σ , T )  SσG K∗ (X σ , G)
and (for σ equal to the trivial subgroup)
K∗ (Y σ , T )geom  K∗ (X σ , G)geom ;
hence the theorem for the action of G on X follows from the theorem for
the action of T on Y .

7.2. Actions with stabilizers of constant dimension. From the theorem

on actions with finite stabilizers we can easily get a decomposition result
when we assume that the stabilizers have constant dimension. Assume
that G is a diagonalizable group scheme of finite type over S, acting on
a noetherian regular separated algebraic space X over S with stabilizers of
constant dimension equal to s.
Definition 7.2. A diagonalizable subgroup scheme σ ⊆ G is dual semi-
cyclic if σ/σ0 is dual cyclic, where σ0 is the toral component of σ .
The order of a dual semi-cyclic group σ is by definition equal the order
of σ/σ0 .
38 G. Vezzosi, A. Vistoli

Equivalently, σ ⊆ G is dual semicyclic if it is isomorphic to Grm,S × µn,S

for some r ≥ 0 and n > 0.
Definition 7.3. A subgroup scheme σ ⊆ G is called essential if it is dual
semicyclic and s-dimensional, and X σ  = ∅.
There are finitely many subtori T j ⊆ G of dimension s in G with
X  = ∅, and X is the disjoint union of the X T j . The toral part of an
Tj

essential subgroup of G coincides with one of the T j ; hence there are only
finitely many essential subgroups of G.
We fix a positive integer N which is divisible by the least common
multiple of the orders of the essential subgroups of G.
For each dual semicyclic subgroup σ ⊆ G, we define a multiplicative
system
def
Sσ = Sσ/σ0 ⊆ R(G/σ0 ) ⊆ RG.
as the set of those elements of R(G/σ0) whose image in R(σ/σ0 ) is a power
of N. If M is a module over RG, we define, as before, the σ -localization of
M to be Mσ = S−1 σ M.
If σ ⊆ G is an essential subgroup, we can choose a splitting G 
(G/σ0 ) × σ0 ; according to [Tho86b, Lemme 5.6] this splitting induces an
isomorphism
K∗ (X σ , G)  K∗ (X σ , G/σ0 ) ⊗ Rσ0
and also an isomorphism of σ -localizations
K∗ (X σ , G)σ  K∗ (X σ , G/σ0 )σ/σ0 ⊗ Rσ0 .
Fix one of the T j , and choose a splitting G  G/T j × T j . We have a com-
mutative diagram


K∗ (X σ , G)σ
K∗ (X , G) ⊗ Z[1/N]
Tj
σ essential
/

σ0 =T j

∼ ∼


  ∼ K∗ (X σ , G/σ0 )σ/σ0 ⊗ Rσ0
K∗ X , G/T j ⊗ RT j ⊗ Z[1/N]
Tj /

σ essential
σ0 =T j

where the two columns are isomorphisms induced by the choice of a splitting
G  G/T j × T j , and the rows are induced by composing the restriction
homomorphism from X T j to X σ with the localization homomorphism. The
bottom row is in an isomorphism because of Theorem 7.1 (i).
Since the product of the restriction homomorphisms
K∗ (X, G) −→ K∗ (X T j , G)
j

is an isomorphism, we obtain the following generalization of Theorem 7.1.

Higher algebraic K-theory for actions of diagonalizable groups 39

Theorem 7.4. Suppose that a diagonalizable group scheme G of finite type

over S acts with stabilizers of constant dimension on a noetherian regular
separated algebraic space X over S.
(i) There are finitely many essential subgroup schemes in G, and the homo-
morphism

locσ : K∗ (X, G) ⊗ Z[1/N] −→ K∗ (X σ , G)σ ,

σ σ

where the product runs over all the essential subgroup schemes of G,
is an isomorphism.
(ii) For any essential subgroup scheme σ ⊆ G, a choice of a splitting
G  (G/σ0 ) × σ0 gives an isomorphism
K∗ (X σ , G)σ −→ K∗ (X σ , G/σ0 )geom ⊗ 
Rσ ⊗ Rσ0 .
Remark 7.5. If s = 0, then σ0 = 1 for each essential subgroup σ ⊆ G, so
there is a unique splitting G  (G/σ0 ) × σ0 , and the isomorphism in (ii) is
canonical.

7.3. More specializations. For the refined decomposition theorem we need

more specialization homomorphisms. Let a diagonalizable group scheme
G of finite type over S act on a noetherian regular separated algebrac space
X over S, with no restriction on the dimensions of the stabilizers.
Notation 7.6. Given a diagonalizable subgroup scheme σ ⊆ G, we set
X (σ) = X σ ∩ X ≤dim σ .
Equivalently, X (σ) = (X dim σ )σ .
Obviously X (σ) is a locally closed regular subspace of X.
Definition 7.7. Let σ and τ be two diagonalizable subgroup schemes of G.
We say that τ is subordinate to σ , and we write τ ≺ σ , if τ is contained in σ ,
and the induced morphism τ → σ/σ0 is surjective.
Suppose that σ and τ are diagonalizable subgroup schemes of G of
dimension s and t respectively, and that τ is subordinate to σ . Consider
the deformation to the normal cone Ms → P1S of X s in X ≤s , considered in
Subsect. 3.1. By Proposition 3.6, the restriction M(τ) s → P S is regular at
1

infinity, so we can define a specialization homomorphism

 
K∗ (X (τ) , G) −→ K∗ N(τ)
s , G .

Denote by Nσ the restriction of Ns to X (σ) . We define the specialization

homomorphism
 
SpτX,σ : K∗ (X (τ) , G) −→ K∗ N(τ)
σ , G
40 G. Vezzosi, A. Vistoli

as the composition of the homomorphism K∗ (X (τ) , G) → K∗ (N(τ)

s , G)
(τ) (τ)
above with the restriction homomorphism K∗ (Ns , G) → K∗ (Nσ , G).
We also denote by
 
SpτX,σ : K∗ (X (τ) , G)τ −→ K∗ N(τ)
σ ,G τ

the τ-localization of this specialization homomorphism.

Remark 7.8. Since τ is subordinate to σ , it is easy to see that N(τ)
σ is a union
(τ)
of connected components of Ns .

7.4. The general case. The hypotheses are the same as in the previous
subsection: G is a diagonalizable group scheme of finite type over S, acting
on a noetherian regular separated algebraic space X over S.
Definition 7.9. An essential subgroup of G is a dual semicyclic subgroup
scheme σ ⊆ G such that X (σ)  = ∅.
A semicyclic subgroup scheme of G is essential if and only if it is
essential for the action of G on X s for some s; hence there are only finitely
many essential subgroups of G. We will fix a positive integer N that is
divisible by the orders of all the essential subgroups of G.
If σ is a dual semicyclic subgroup of G, we define the multiplicative
system Sσ ⊆ R(G/σ0 ) ⊆ RG as before, as the subset of R(G/σ0 ) ⊆ RG
consisting of elements whose image in  R(σ/σ0) is a power of N. Also,

K∗ (X (σ) , G)σ = S−1 (σ)

def
σ K∗ (X , G),

as before.
Proposition 7.10. Let σ and τ be two semi-cyclic subgroups of G. If τ ≺ σ ,
then Sσ ⊆ Sτ .
Proof. Consider the commutative diagram of group schemes
G/τ0
O
/ /

G/σ0
O

O
O

τ/τ0 / /

σ/σ0 ;
by taking representation rings we get a commutative diagram of rings
R(G/τ0 ) o o

R(G/σ0 )

R(τ/τ0 ) o o

R(σ/σ0)


R(τ/τ0 ) o


R(σ/σ0)
Higher algebraic K-theory for actions of diagonalizable groups 41

(without the dotted arrow). But it is easy to see that in fact the composition
R(σ/σ0) → R(τ/τ0 ) →  R(τ/τ0 ) factors through  R(σ/σ0), so in fact the
dotted arrows exists; and this proves the thesis.
Now, consider the restriction of the projection πσ,τ : N(τ) (σ)
σ → X , where
σ and τ are dual semicyclic subgroups of G, and τ is subordinate to σ .
Because of Proposition 7.10, we can consider the composition of the
pullback K∗ (X (σ) , G)σ → K∗ (N(τ) σ , G)σ with the natural homomorphism
(τ) (τ)
K∗ (Nσ , G)σ → K∗ (Nσ , G)τ coming from the inclusion Sσ ⊆ Sτ ; we
denote this homomorphism by
∗
 
πσ,τ : K∗ (X (σ) , G)σ −→ K∗ N(τ) σ , G τ
.
Definition 7.11. Suppose that σ and τ are dual semicyclic subgroups of
G and that τ is subordinate to σ . Two elements aσ ∈ K∗ (X (σ) , G)σ and
aτ ∈ K∗ (X τ , G)τ are compatible if
∗
 
πσ,τ aσ = SpτX,σ aτ ∈ K∗ N(τ)
σ , G τ
.

For each essential dual semicyclic subgroup σ ⊆ G we denote by

locσ : K∗ (X, G) ⊗ Z[1/N] → K∗ (X (σ) , G)σ
the composition of the restriction homomorphism
K∗ (X, G) ⊗ Z[1/N] → K∗ (X (σ) , G) ⊗ Z[1/N]
with the localization homomorphism
K∗ (X (σ) , G) ⊗ Z[1/N] → K∗ (X (σ) , G)σ .
The following is the main result of this section.
Theorem 7.12. The ring homomorphism
locσ : K∗ (X, G) ⊗ Z[1/N] −→ K∗ (X (σ) , G)σ
σ σ⊆G
σ essential


is injective. Its image consists of the elements (aσ ) of σ K∗ (X (σ) , G)σ with
the property that if σ and τ are essential, τ ≺ σ , and dim σ = dim τ + 1,
then aτ and aσ are compatible.
Notice that in the particular case that the action has stabilizers of constant
dimension, all essential subgroups of G have the same dimension, and this
reduces to Theorem 7.4.
Also, if σ is an essential subgroup of G then X (σ) = X σdim σ , so it follows
from Theorem 7.4 (ii) that a splitting G  (G/σ0 )×σ0 gives an isomorphism
of rings
K∗ (X (σ) , G)σ  K∗ (X (σ) , G/σ0 )geom ⊗ 
Rσ ⊗ Rσ0 .
However, this isomorphism is not canonical in general, as it depends on the
choice of a splitting.
42 G. Vezzosi, A. Vistoli

Proof. To simplify the notation, we will implicitly assume that everything

has been tensored with Z[1/N].
We apply Theorem 4.5 together with Theorem
7.4. According to Theo-
rem 4.5 we have an injection K∗ (X,
G) → s K∗ (X s , G) whose image
is the subring of sequences (αs ) ∈ ns=0 K∗ (X s , G) with the property that
for each s = 1, . . . , n the pullback of αs ∈ K∗ (X s , G) to K∗ (Ns,s−1 , G)
coincides with Sps−1X,s (αs−1 ) ∈ K∗ (Ns,s−1 , G). Moreover,


by Theorem 7.4,
(σ)
we can decompose further each K∗ (X s , G) as σ K∗ (X , G)σ , where σ
varies in the (finite) set of essential subgroups of G of dimension s. By
compatibility of specializations, for any s ≥ 0, the following diagram is
commutative


∼ K∗ (X (τ) , G)τ
K∗ (X s−1 , G) /

τ essential
dim τ=s−1
QQQ

QQQτ,σ SpτX,σ

s−1 SpτX,s QQQ
τ Sp X,s QQQ
Q



 
(

∼ K∗ (N(τ) , G)τ φ K∗ N(τ)

σ ,G
K∗ (Ns,s−1 , G) /

τ essential
/

τ essential σ τ
dim τ=s−1 dim τ=s−1 dim σ =s

where φ is induced by the obvious pullbacks. On the other hand, the fol-
∗
lowing diagram commutes by definition of πσ,τ


∼ K∗ (X (σ) , G)σ
K∗ (X s , G) /

σ essential
dim σ =s
QQQ
∗
QQQ σ,τ πσ,τ
QQQ
QQQ
Q



 
(

∼ K∗ (N(τ) , G)τ φ K∗ N(τ)

σ ,G .
K∗ (Ns,s−1 , G) /

τ essential
/

τ essential σ τ
dim τ=s−1 dim τ=s−1 dim σ =s

Then the theorem will immediately follow if we show that φ is an isomorph-

ism. This is true because of the following lemma.
Lemma 7.13. Fix an essential subgroup τ of dimension s − 1. Then for any
σ  τ with dim σ = s, the scheme N(τ) (τ) (τ)
σ is open in Ns ; furthermore, Ns is
(τ)
the disjoint union of the Nσ for all essential σ with σ  τ, dim σ = s.
Proof. We will show that N(τ) (τ)
s is the disjoint union of the Nσ ; since each
N(τ) (τ)
σ is closed in Ns , and there are only finitely many possible σ , it follows
that each N(τ)σ is also open in Ns .
(τ)

Let us first observe that if σ and σ are essential subgroups in G of dimen-

 (τ)  (σ )
sion s to which τ is subordinate, and N(τ) σ N σ  = ∅, then X (σ)
X  = ∅,
therefore σ0 is equal to σ0 . But this implies that σ = σ , since σ and σ are
both equal to the inverse image in G of the image of τ → G/σ0 = G/σ0 .
According to Proposition 2.2 (i), if T1 , . . . , Tr are the essential s-
dimensional subtori of G, then Ns is the disjoint union of the NT j . Clearly,
if τ0 is not contained in T j , then N(τ) (τ)
T j is empty, so Ns is the disjoint union of
Higher algebraic K-theory for actions of diagonalizable groups 43

the N(τ)
T , where T ranges over the essential s-dimensional subtori of G with
τ0 ⊆ T . But there is a bijective correspondence between s-dimensional dual
semi-cyclic subgroups σ ⊆ G with σ  τ and s-dimensional subtori of G
with τ0 ⊆ T : in one direction we associate with each σ its toral part σ0 , in
the other we associate with each T the subgroup scheme τ + T ⊆ G.
The proof is concluded by noticing that if σ and τ are as above, with
σ0 = T , then N(τ)
T = Nσ .
(τ)

References
[Ati74] Atiyah, M.F.: Elliptic operators and compact groups. Lect. Notes Math. 401.
Berlin, New York: Springer 1974
[B-B-F-K02] Barthel, G., Brasselet, J.P., Fieseler, K.H., Kaup, L.: Combinatorial intersec-
tion cohomology for fans. Tohoku Math. J., II. Ser. 54, 1–41 (2002)
[Bia73] Białynicki-Birula, A.: Some theorems on actions of algebraic groups. Ann.
Math. (2) 98, 480–497 (1973)
[B-DC-P90] Bifet, E., De Concini, C., Procesi, C.: Cohomology of regular embeddings.
Adv. Math. 82, 1–34 (1990)
[Bre74] Bredon, G.E.: The free part of a torus action and related numerical equalities.
Duke Math. J. 41, 843–854 (1974)
[Bri97] Brion, M.: Equivariant Chow groups for torus actions. Transform. Groups 2,
225–267 (1997)
[Bri98] Brion, M.: Equivariant cohomology and equivariant intersection theory. In:
Representation theories and algebraic geometry, pp. 1–37. Dordrecht: Kluwer
Academic Publishers 1998
[Ch-Sk74] Chang, T., Skjelbred, T.: The topological Shur lemma and related results.
Ann. Math. (2) 100, 307–321 (1974)
[Dan78] Danilov, V.: The geometry of toric varieties. Russ. Math. Surv. 33, 97–154
(1978)
[SGA3] Demazure, M.: Schémas en groupes. Lect. Notes Math. 151, 152 and 153.
Springer 1970
[Ed-Gr98] Edidin, D., Graham, W.: Equivariant intersection theory. Invent. Math. 131,
595–634 (1998)
[Ei-Mo62] Eilenberg, S., Moore, J.C.: Limits and spectral sequences. Topology 1, 1–23
(1962)
[Ful93a] Fulton, W.: Intersection Theory. Springer 1993
[Ful93b] Fulton, W.: Introduction to toric varieties. Princeton: Princeton University
Press 1993
[G-K-MP98] Goresky, M., Kottwitz, R., MacPherson, R.: Equivariant cohomology, Koszul
duality, and the localization theorem. Invent. Math. 131, 25–84 (1998)
[EGAIV] Grothendieck, A., Dieudonne, J.: Elements de Gèomètrie Algèbrique IV.
Étude locale des schémas et des morphismes de schémas. Publ. Math., Inst.
Hautes Étud. Sci. 28, 1–255 (1966)
[Hsi75] Hsiang, W.Y.: Cohomology theory of Topological Transformation Groups.
Springer 1975
[Kir84] Kirwan, F.: Cohomology of quotients in symplectic and algebraic geometry.
Math. Notes 31. Princeton: Princeton University Press 1984
[Kly83] Klyachko, A.A.: Vector bundles on Demazure models. Sel. Math. Sov. 3,
41–44 (1983/84)
[La-MB00] Laumon, G., Moret-Bailly, L.: Champs Algébriques. Springer 2000
[Mer97] Merkurjev, A.: Comparison of the equivariant and the standard K-theory of
algebraic varieties. Algebra Anal. 9, 175–214 (1997). Translation in: St. Pe-
tersbg. Math. J. 9, 815–850 (1998)
44 G. Vezzosi, A. Vistoli

[Ro-Kn] Rosu, I., Knutson, A.: Equivariant K-theory and Equivariant Cohomology.
Math. Z. 243, 423–448 (2003)
[Sum75] Sumihiro, H.: Equivariant completion II. J. Math. Kyoto Univ. 15, 573–605
(1975)
[Tho86a] Thomason, R.W.: Comparison of equivariant algebraic and topological K-
theory. Duke Math. J. 68, 447–462 (1986)
[Tho86b] Thomason, R.W.: Lefschetz–Riemann–Roch theorem and coherent trace for-
mula. Invent. Math. 85, 515–543 (1986)
[Tho87] Thomason, R.W.: Algebraic K-theory of group scheme actions. In: Algebraic
Topology and Algebraic K-theory, Ann. Math. Stud. 113. Princeton: Princeton
University Press 1987
[Tho92] Thomason, R.W.: Un formule de Lefschetz en K-thèorie equivariante al-
gèbrique. Duke Math. J. 68, 447–462 (1992)
[Tho93] Thomason, R.W.: Les K-groupes d’un schéma éclaté et un formule d’inter-
section excédentaire. Invent. Math. 112, 195–216 (1993)
[Th-Tr90] Thomason, R.W., Trobaugh, T.: Higher algebraic K-theory of schemes and
of derived categories. Grothendieck Festschrift vol. III, 247–435. Basel:
Birkhäuser 1990
[Toen] Toen, B.: Notes on G-theory of Deligne–Mumford stacks.
Preprint [Link]/9912172
[Ve-Vi] Vezzosi, G., Vistoli, A.: Higher algebraic K-theory of group actions with
finite stabilizers. Duke Math. J. 113, 1–55 (2002)