DUKE MATHEMATICAL JOURNAL

Vol. 113, No. 1, c 2002

HIGHER ALGEBRAIC K -THEORY OF GROUP

ACTIONS WITH FINITE STABILIZERS

GABRIELE VEZZOSI and ANGELO VISTOLI

Abstract
We prove a decomposition theorem for the equivariant K -theory of actions of affine
group schemes G of finite type over a field on regular separated Noetherian alge-
braic spaces, under the hypothesis that the actions have finite geometric stabilizers
and satisfy a rationality condition together with a technical condition that holds, for
example, for G abelian or smooth. We reduce the problem to the case of a GLn -action
and finally to a split torus action.

Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2. General constructions . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1. Morphisms of actions and induced maps on K -theory . . . . . . . 6
2.2. The basic definitions and results . . . . . . . . . . . . . . . . . . 8
3. The main theorem: The split torus case . . . . . . . . . . . . . . . . . . 16
4. The main theorem: The case of G = GLn,k . . . . . . . . . . . . . . . . 23
5. The main theorem: The general case . . . . . . . . . . . . . . . . . . . 33
5.1. Proof of Theorem 5.4 . . . . . . . . . . . . . . . . . . . . . . . 36
5.2. Hypotheses on G . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.3. Final remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Appendix. Higher equivariant K -theory of Noetherian regular separated alge-
braic spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

1. Introduction
The purpose of this paper is to prove a decomposition theorem for the equivariant
K -theory of actions of affine group schemes of finite type over a field on regular sep-
arated Noetherian algebraic spaces. Let X be a regular connected separated Noethe-
rian scheme with an ample line bundle, and let K 0 (X ) be its Grothendieck ring of

DUKE MATHEMATICAL JOURNAL

Vol. 113, No. 1, c 2002
Received 27 April 2000. Revision received 13 June 2001.
2000 Mathematics Subject Classification. Primary 19E08, 14L30; Secondary 14A20.
Authors’ work partially supported by the University of Bologna funds for selected research topics.

1
2 VEZZOSI and VISTOLI

vector bundles. Then the kernel of the rank morphism K 0 (X ) → Z is nilpotent (see
[SGA6, Exp. VI, Th. 6.9]), so the ring K 0 (X ) is indecomposable and remains such
after tensoring with any indecomposable Z-algebra.
The situation is quite different when we consider the equivariant case. Let G be an
algebraic group acting on a Noetherian separated regular scheme, or algebraic space,
let X be over a field k, and consider the Grothendieck ring K 0 (X, G) of G-equivariant
perfect complexes. This is the same as the Grothendieck group of G-equivariant co-
herent sheaves on X , and it coincides with the Grothendieck ring of G-equivariant
vector bundles if all G-coherent sheaves are quotients of locally free coherent sheaves
(which is the case, e.g., when G is finite or smooth and X is a scheme). Assume that
the action of G on X is connected, that is, that there are no nontrivial invariant open
and closed subschemes of X . Still, K 0 (X, G) usually decomposes, after inverting
some primes; for example, if G is a finite group and X = Spec C, then K 0 (X, G)
is the ring of complex representations of G, which becomes a product of fields after
tensoring with Q.
In [Vi2] the second author analyzes the case where the action of G on X has
finite reduced geometric stabilizers. Consider the ring of representations R(G), and
consider the kernel m of the rank morphism rk : K 0 (X, G) → Z. Then K 0 (X, G) is
an R(G)-algebra; he shows that the localization morphism
K 0 (X, G) ⊗ Q −→ K 0 (X, G)m
is surjective and that the kernel of the rank morphism K 0 (X, G)m ⊗ Q → Q is
nilpotent. Furthermore, he conjectures that K 0 (X, G) ⊗ Q splits as a product of the
localization K 0 (X, G)m and some other ring, and he formulates a conjecture about
what the other factor ring should be when G is abelian and the field is algebraically
closed of characteristic zero. The proofs of the results in [Vi2] depend on an equivari-
ant Riemann-Roch theorem, whose proof was never published by the author; however,
all of the results have been proved and generalized in [EG].
The case where G is a finite group is studied in [Vi1]. Assume that k contains
all nth roots of 1, where n is the order of the group G. Then the author shows that,
after inverting the order of G, the K -theory ring K ∗ (X, G) of G-equivariant vector
bundles on X (which is assumed to be a scheme in that paper) is canonically the
product of a finite number of rings, expressible in terms of ordinary K -theory of
appropriate subschemes of fixed points of X . Here K ∗ (X, G) = i K i (X, G) is the
L

graded higher K -theory ring. The precise formula is as follows.

Let σ be a cyclic subgroup of G whose order is prime to the characteristic of k;
then the subscheme X σ of fixed points of X under the actions of σ is also regular. The
representation ring R(σ ) is isomorphic to the ring Z[t]/(t n −1), where t is a generator
of the group of characters hom(σ, k ∗ ). We call eR(σ ) the quotient of the ring R(σ ) by
the ideal generated by the element 8n (t), where 8n is the nth cyclotomic polynomial;
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 3

this is independent of t. The ring eR(σ ) is isomorphic to the ring of integers in the nth
cyclotomic field. Call NG (σ ) the normalizer of σ in G; the group NG (σ ) acts on
the scheme X σ and, by conjugation, on the group σ. Consider the induced actions of
NG (σ ) on the K -theory ring K ∗ (X σ ) and on the ring e
R(σ ).
Choose a set C (G) of representatives for the conjugacy classes of cyclic sub-
groups of G whose order is prime to the characteristic of the field. The statement of
the main result of [Vi1] is as follows.

THEOREM
There is a canonical ring isomorphism

NG (σ )
K ∗ (X σ ) ⊗ e
Y
K ∗ (X, G) ⊗ Z[1/|G|] ' R(σ ) ⊗ Z[1/|G|].
σ ∈C (G)

In the present paper we generalize this decomposition to the case in which G is an

algebraic group scheme of finite type over a field k, acting with finite geometric sta-
bilizers on a Noetherian regular separated algebraic space X over k. Of course, we
cannot expect a statement exactly like the one for finite groups, expressing equiv-
ariant K -theory simply in terms of ordinary K -theory of the fixed point sets. For
example, when X is the Stiefel variety of r -frames in n-space, then the quotient of
X by the natural free action of GLr is the Grassmannian of r -planes in n-space, and
K 0 (X, GLr ) = K 0 (X/ GLr ) is nontrivial, while K 0 (X ) = Z.
Let X be a Noetherian regular algebraic space over k with an action of an affine
group scheme G of finite type over k. We consider the Waldhausen K -theory group
K ∗ (X, G) of complexes of quasi-coherent G-equivariant sheaves on X with coherent
bounded cohomology. This coincides on the one hand with the Waldhausen K -theory
group K ∗ (X, G) of the subcategory of complexes of quasi-coherent G-equivariant
flat sheaves on X with coherent bounded cohomology (and hence has a natural ring
structure given by the total tensor product) and on the other hand with the Quillen
group K ∗0 (X, G) of coherent equivariant sheaves on X ; furthermore, if every coherent
equivariant sheaf on X is the quotient of a locally free equivariant coherent sheaf, it
also coincides with the Quillen group K ∗naive (X, G) of coherent locally free equiv-
ariant sheaves on X . These K -theories and their relationships are discussed in the
appendix.
Our result is as follows. First we have to see what plays the role of the cyclic
subgroups of a finite group. This is easy; the group schemes whose rings of represen-
tations are of the form Z[t]/(t n − 1) are not the cyclic groups, in general, but their
Cartier duals, that is, the group schemes that are isomorphic to the group scheme µn
of nth roots of 1 for some n. We call these group schemes dual cyclic. If σ is a dual
cyclic group, we can define e Rσ as before. A dual cyclic subgroup σ of G is called
4 VEZZOSI and VISTOLI

essential if X σ 6= ∅. The correct substitute for the ordinary K -theory of the subspaces
of invariants is the geometric equivariant K -theory K ∗ (X, G)geom , which is defined
as follows. Call N the least common multiple of the orders of all the essential dual
cyclic subgroups of G. Call S1 the multiplicative subset of the ring R(G) consisting
of elements whose virtual rank is a power of N ; then K ∗ (X, G)geom is the localiza-
tion S1−1 K ∗ (X, G). Notice that K ∗ (X, G)geom ⊗ Q = K ∗ (X, G)m , with the notation
above. Moreover, if every coherent equivariant sheaf on X is the quotient of a locally
free equivariant coherent sheaf, by [EG], we have an isomorphism of rings

K 0 (X, G)geom ⊗ Q = A∗G (X ) ⊗ Q,

where A∗G (X ) denotes the direct sum of G-equivariant Chow groups of X.

We prove the following. Assume that the action of G on X is connected. Then
the kernel of the rank morphism K 0 (X, G)geom → Z[1/N ] is nilpotent (see Cor. 5.2).
This is remarkable; we have made what might look like a small step toward making
the equivariant K -theory ring indecomposable, and we immediately get an indecom-
posable ring. Indeed, K ∗ (X, G)geom “feels like” the K -theory ring of a scheme; we
want to think of K ∗ (X, G)geom as what the K -theory of the quotient X/G should be,
if X/G were smooth, after inverting N (see Conj. 5.8).
Furthermore, consider the centralizer CG (σ ) and the normalizer NG (σ ) of σ in-
side G. The quotient wG (σ ) = NG (σ )/CG (σ ) is contained inside the group scheme
of automorphisms of σ , which is a discrete group, so it is also a discrete group.
It acts on e R(σ ), by conjugation, and it also acts on the equivariant K -theory ring
K ∗ (X , CG (σ )) and on the geometric equivariant K -theory ring K ∗ (X σ , CG (σ ))geom
σ

(see Cor. 2.5).

We say that the action of G on X is sufficiently rational when the following two
conditions are satisfied. Let k be the algebraic closure of k.
(1) Each essential dual cyclic subgroup σ ⊆ G k is conjugate by an element of
G( k ) to a dual cyclic subgroup of G.
(2) If two essential dual cyclic subgroups of G are conjugate by an element of
G( k ), they are also conjugate by an element of G(k).
Obviously, every action over an algebraically closed field is sufficiently rational.
Furthermore, if G is GLm , SLm , Spm , or a totally split torus, then any action of G is
sufficiently rational over an arbitrary base field (see Prop. 2.3). If G is a finite group,
then the action is sufficiently rational when k contains all nth roots of 1, where n is
the least common multiple of the orders of the cyclic subgroups of k of order prime
to the characteristic, whose fixed point subscheme is nonempty. Denote by C (G) a
set of representatives for essential dual cyclic subgroup schemes, under conjugation
by elements of the group G(k). Here is the statement of our result.
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 5

MAIN THEOREM
Let G be an affine group scheme of finite type over a field k, acting on a Noetherian
separated regular algebraic space X . Assume the following three conditions.
(a) The action has finite geometric stabilizers.
(b) The action is sufficiently rational.
(c) For any essential cyclic subgroup σ of G, the quotient G/CG (σ ) is smooth.
Then C (G) is finite, and there is a canonical isomorphism of R(G)-algebras

w (σ )
K ∗ (X σ , CG (σ ))geom ⊗ e
Y
K ∗ (X, G) ⊗ Z[1/N ] ' R(σ ) G .
σ ∈C (G)

Conditions (a) and (b) are clearly necessary for the theorem to hold. We are not sure
about (c). It is rather mild, as it is satisfied, for example, when G is smooth (this is
automatically true in characteristic zero) or when G is abelian. A weaker version of
condition (c) is given in Section 5.2.
In the case when G is abelian over an algebraically closed field of characteristic
zero, the main theorem implies [Vi2, Conj. 3.6]. When G is a finite group, and the base
field contains enough roots of 1, as in the statement of Theorem 1, then the conditions
of the main theorem are satisfied; since the natural maps K ∗ (X σ , CG (σ ))geom →
K ∗ (X σ )CG (σ ) become isomorphisms after inverting the order of G (see Prop. 5.7),
the main theorem implies [Vi1, Th. 1]. However, the proof of the main theorem here
is completely different from [Vi1, proof of Th. 1].
As B. Toen pointed out to us, a weaker version of our main theorem (with Q-
coefficients and assuming G smooth, acting with finite reduced stabilizers) follows
from his [To1, Th. 3.15]; the étale techniques he uses in proving this result make it
impossible to avoid tensoring with Q (see also [To2]).
Here is an outline of the paper. First we define the homomorphism (see Sec. 2.2).
Next, in Section 3, we prove the result when G is a totally split torus. Here the basic
tool is the result of R. Thomason, which gives a generic description of the action
of a torus on a Noetherian separated algebraic space, and we prove the result by
Noetherian induction, using the localization sequence for the K -theory of equivariant
coherent sheaves. As in [Vi1], the difficulty here is that the homomorphism is defined
via pullbacks, and thus it does not commute with the pushforwards intervening in the
localization sequence. This is solved by producing a different isomorphism between
the two groups in question, using pushforwards instead of pullbacks, and then relating
this to our map, via the self-intersection formula.
The next step is to prove the result in the case when G = GLn ; here the key point
is a result of A. Merkurjev which links the equivariant K -theory of a scheme with
a GLn -action to the equivariant K -theory of the action of a maximal torus. This is
carried out in Section 4. Finally (see Sec. 5), we reduce the general result to the case
6 VEZZOSI and VISTOLI

of GLn , by considering an embedding G ⊆ GLn , and the induced action of GLn on

Y = GLn ×G X . It is at this point that condition (c) enters, allowing a clear description
of Y σ where σ is an essential dual cyclic subgroup of G (see Prop. 5.6).

2. General constructions

Notation. If S is a separated Noetherian scheme, X is a Noetherian separated S-

algebraic space (which is most of the time assumed to be regular), and G is a flat
affine group scheme over S acting on X , we denote by K ∗ (X, G) (resp., K 0 (X, G)) the
Waldhausen K -theory of the complicial bi-Waldhausen (see [ThTr]) category W1,X,G
of complexes of quasi-coherent G-equivariant O X -modules with bounded coherent
cohomology (resp., the Quillen K -theory of G-equivariant coherent O X -modules). As
shown in the appendix, if X is regular, K ∗ (X, G) is isomorphic to K ∗0 (X, G) and has
a canonical graded ring structure. When X is regular, the isomorphism K ∗ (X, G) '
K ∗0 (X, G) then allows us to switch between the two theories when needed.

2.1. Morphisms of actions and induced maps on K -theory

Let S be a scheme. By an action over S we mean a triple (X, G, ρ) where X is an
S-algebraic space, G is a group scheme over S, and ρ : G × S X → X is an action of
G on X over S. A morphism of actions
( f, φ) : (X, G, ρ) −→ (X 0 , G 0 , ρ 0 )
is a pair of S-morphisms f : X → X 0 and φ : G → G 0 , where φ is a morphism of
S-group schemes, such that the following diagram commutes:
ρ
G × S X −−−−→ X
 
φ× f y
 f
y
G 0 × S X 0 −−−− → X0
0ρ

Equivalently, f is required to be G-equivariant with respect to the given G-action on

X and the G-action on X 0 induced by composition with φ.
A morphism of actions ( f, φ) : (X, G, ρ) → (X 0 , G 0 , ρ 0 ) induces an exact
functor ( f, φ)∗ : W3,X 0 ,G 0 → W3,X,G , where W3,Y,H denotes the complicial bi-
Waldhausen category of complexes of H -equivariant flat quasi-coherent modules
with bounded coherent cohomology on the H -algebraic space Y (see appendix). Let
(E ∗ , ε∗ ) be an object of W3,X 0 ,G 0 ; that is, E ∗ is a complex of G 0 -equivariant flat quasi-
coherent O X 0 -modules with bounded coherent cohomology, and for any i,
εi : pr0∗ i
g ρ 0∗ E i
2 E −→
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 7

is an isomorphism satisfying the usual cocycle condition. Here pr02 : G 0 × S X 0 → X 0

denotes the obvious projection, and similarly for pr2 : G × S X → X . Since

ρ ∗ f ∗ E ∗ = ( fρ)∗ E ∗ = (φ × f )∗ ρ 0∗ E ∗

and
pr∗2 f ∗ E ∗ =(φ × f )∗ pr0∗
2E ,
∗

.
we define ( f, φ)∗ (E ∗ , ε∗ ) = ( f ∗ E ∗ , (φ × f )∗ (ε∗ )) ∈ W3,X,G (the cocycle condition
for each (φ × f )∗ (εi ) following from the same condition for εi ); ( f, φ)∗ is then de-
fined on morphisms in the only natural way. Now, ( f, φ)∗ is an exact functor and, if
X and X 0 are regular so that the Waldhausen K -theory of W3,X,G (resp., of W3,X 0 ,G 0 )
coincides with K ∗ (X, G) (resp., K ∗ (X 0 , G 0 )) (see appendix), it defines a ring mor-
phism
( f, φ)∗ : K ∗ (X 0 , G 0 ) −→ K ∗ (X, G).
A similar argument shows that if f is flat, ( f, φ) induces a morphism

( f, φ)∗ : K ∗0 (X 0 , G 0 ) −→ K ∗0 (X, G).

Example 2.1
Let G and H be group schemes over S, and let X be an S-algebraic space. Moreover,
suppose that
(1) G and H act on X ;
(2) G acts on H by S-group scheme automorphisms (i.e., it is given a morphism
G → Aut(GrSch)/S (H ) of group functors over S);
(3) the two preceding actions are compatible; that is, for any S-scheme T , any
g ∈ G(T ), h ∈ H (T ), and x ∈ X (T ), we have

g · (h · x) = h g · (g · x),

where (g, h) 7→ h g denotes the action of G(T ) on H (T ).

If g ∈ G(S) and if gT denotes its image via G(S) → G(T ), let us define a morphism
of actions ( f g , φg ) : (X, H ) → (X, H ) as

f g (T ) : X (T ) −→ X (T ) : x 7 −→ gT · x,
φg (T ) : H (T ) −→ H (T ) : h 7−→ h gT .

This is an isomorphism of actions and induces an action of the group G(S) on K ∗0 (X,
H ) and on K ∗ (X, H ). This applies, in particular, to the case where X is an algebraic
space with a G action and G B H , G acting on H by conjugation.
8 VEZZOSI and VISTOLI

2.2. The basic definitions and results

Let G be a linear algebraic k-group scheme G acting with finite geometric stabilizers
on a regular Noetherian separated algebraic space X over k.
We denote by R(G) the representation ring of G.
A (Cartier) dual cyclic subgroup of G over k is a k-subgroup scheme σ ⊆ G
such that there exist an n > 0 and an isomorphism of k-groups σ ' µn,k . If σ, ρ
are dual cyclic subgroups of G and if L is an extension of k, we say that σ and ρ
are conjugate over L if there exists g ∈ G(L) such that gσ(L) g −1 = ρ(L) (where
.
H(L) = H ×Spec k Spec L, for any k-group scheme H ) as L-subgroup schemes of
G (L) .
A dual cyclic subgroup σ ⊆ G is said to be essential if X σ 6= ∅.
We say that the action of G on X is sufficiently rational if
(1) any two essential dual cyclic subgroups of G are conjugated over k if and only
if they are conjugated over an algebraic closure k of k;
(2) any essential dual cyclic subgroup ρ of G (k) is conjugated over k to a dual
cyclic subgroup of the form σ(k) where σ ⊆ G is (essential) dual cyclic.
We denote by C (G) a set of representatives for essential dual cyclic subgroups of G
with respect to the relation of conjugacy over k.

Remark 2.2
Note that if the action is sufficiently rational and if ρ, σ are essential dual cyclic
subgroups of G which are conjugate over an algebraically closed extension  of k,
then they are also conjugate over k.

PROPOSITION 2.3
Any action of GLn , SLn , Sp2n , or of a split torus is sufficiently rational.

Proof
If G is a split torus, condition (1) is clear because G is abelian, while it follows from
the rigidity of diagonalizable groups that any subgroup scheme of G k is in fact defined
over k. Let σ ⊆ GLm be a dual cyclic subgroup. Since σ is diagonalizable, we have
an eigenspace decomposition

Vχσ
M
V = km =
χ ∈b
σ

such that the χ with Vχ 6= 0 generate b

σ . Conversely, given a cyclic group C and a
decomposition M
V = Vχ
χ ∈C
b
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 9

such that the χ with Vχ 6= 0 generate C, b there is a corresponding embedding of

the Cartier dual σ of C into GLn with Vχ = Vχσ for each χ ∈ C = b σ . Now, if
σ ⊆ GLm,k is a dual cyclic subgroup defined over k, we can apply an element of
GLm (k) to make the Vχ defined over k, and then gσ g −1 is defined over k. If σ ⊆ GLm
and τ ⊆ GLm are dual cyclic subgroups that are conjugate over k, pick an element
of GLm (k) sending σ to τ . This induces an isomorphism φ : σk ' τk , which by
rigidity is defined over k. Then if χ and χ 0 are characters that correspond under the
isomorphism of b σ and b τ induced by φ, then the dimension of Vχσ is equal to the
dimension of Vχτ0 , so we can find an element g of GLm which carries each Vχσ onto the
corresponding Vχ0 τ ; conjugation by this element carries σ onto τ . For SL , the proof
m
is very similar if we remark that to give a dual cyclic subgroup σ ⊆ SLm ⊆ GLm
corresponds to giving a decomposition

Vχσ
M
V = km =
χ ∈b
σ

dim Vχσ
such that the χ with Vχσ 6 = 0 generate b
σ , with the condition χ∈b
σ χ σ.
Q
=1∈b
For Spm ⊆ GL2m , a dual cyclic subgroup σ ⊆ Spm gives a decomposition

Vχσ
M
V = k 2m =
χ∈b
σ

such that the χ with Vχσ 6= 0 generate b σ , with the condition that for v ∈ Vχσ and
σ the symplectic product of v and v 0 is always zero, unless χ χ 0 = 1 ∈ b
v 0 ∈ Vχ0 σ . Both
conditions then follow rather easily from the fact that any two symplectic forms over
a vector space are isomorphic.

Let N(G,X ) denote the least common multiple of the orders of essential dual cyclic
subgroups of G. Notice that N(G,X ) is finite: since the action has finite stabilizers,
the group scheme of stabilizers is quasi-finite over X ; therefore the orders of the
stabilizers of the geometric points of X are globally bounded.
.
We define 3(G,X ) = Z[1/N(G,X ) ].
If H ⊆ G is finite, we also write 3 H for Z[1/|H |]. Note that, if σ ⊆ G is dual
cyclic, then 3σ = 3(σ,Spec k) , and if, moreover, σ is essential, 3σ ⊆ 3(G,X ) .
If H ⊆ G is a subgroup scheme and if A is a ring, we write R(H ) A for R(H ) ⊗Z
A. We denote by rk H : R(H ) −→ Z and by rk H,3(G,X ) : R(H )3(G,X ) −→ 3(G,X ) the
rank ring homomorphisms.
We let
K ∗0 (X, G)3(G,X ) = K ∗0 (X, G) ⊗ 3G,X
and
K ∗ (X, G)3(G,X ) = K ∗ (X, G) ⊗ 3G,X
10 VEZZOSI and VISTOLI

(for the notation, see the beginning of this section). Recall that K ∗ (X, G)3(G,X ) is
an R(G)-algebra via the pullback R(G) ' K 0 (Spec k, G) → K 0 (X, G) and that
K ∗ (X, G) ' K ∗0 (X, G) since X is regular (see appendix).
If σ is a dual cyclic subgroup of G of order n, the choice of a generator t for the
.
dual group b σ = HomGrSch/k (σ, Gm,k ) determines an isomorphism
Z[t]
R(σ ) ' .
(t n − 1)
Let pσ be the canonical ring surjection
Z[t] Y Z[t],
−→ ,
(t n − 1) (8d )
d|n

and let f
pσ be the induced surjection
Z[t] Z[t]
−→ ,
(t n
− 1) (8n )
where 8d is the dth cyclotomic polynomial. If mσ is the kernel of the composition
Z[t] Z[t]
R(σ ) ' −→ ,
(t n − 1) (8n )
the quotient ring R(σ )/mσ does not depend on the choice of the generator t for b
σ.

Notation. We denote by e R(σ ) the quotient R(σ )/mσ . We remark that if σ is dual
σ , there are isomorphisms
cyclic of order n and if t is a generator of b
3σ [t] Y 3σ [t]
R(σ )3σ ' n ' . (1)
(t − 1) (8d )
d|n

Let π f σ : R(G)3(G,X ) → e R(σ )3(G,X ) be the canonical ring homomorphism. The

σ -localization K ∗ (X, G)σ of K ∗0 (X, G)3(G,X ) is the localization of the R(G)3(G,X ) -
0
.
module K ∗0 (X, G)3(G,X ) at the multiplicative subset Sσ = π f σ
−1 (1). The σ -localiza-

tion K ∗ (X, G)σ is defined in the same way. If H ⊆ G is a subgroup scheme, we also
write R(H )σ for Sσ−1 (R(H )3(G,X ) ).
If σ is the trivial group, we denote by K ∗0 (X, G)geom the σ -localization
of K ∗0 (X, G)3(G,X ) and call it the geometric part or geometric localization of
K ∗0 (X, G)3(G,X ) . Note that πe1 coincides with the rank morphism rkG,3(G,X ) :
R(G)3(G,X ) −→ 3(G,X ) . We have the same definition for K ∗ (X, G)geom .
Let NG (σ ) (resp., CG (σ ) ⊆ NG (σ )) be the normalizer (resp., the centralizer) of
σ in G; since Aut(σ ) is a finite constant group scheme,
. NG (σ )
WG (σ ) =
CG (σ )
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 11

is also a constant group scheme over k associated to a finite group wG (σ ).

LEMMA 2.4
Let H be a k-linear algebraic group, let σ ' µn,k be a normal subgroup, and let
Y be an algebraic space with an action of H/σ . Then there is a canonical action of
w H (σ ) on K ∗0 (Y, C H (σ )).

Proof
Let us first assume that the natural map

H (k) −→ w H (σ ) (2)

is surjective (which is true, e.g., if k is algebraically closed). Since C H (σ )(k) acts triv-
ially on K ∗0 (Y, C H (σ )) and, by Example 2.1, H (k) acts naturally on K ∗0 (Y, C H (σ )),
we may use (2) to define the desired action. In general, (2) is not surjective and we
proceed as follows. Suppose that we can find a closed immersion of k-linear algebraic
groups H ,→ H 0 such that
(i) σ is normal in H 0 ;
(ii) H 0 /C H 0 (σ ) ' W H (σ );
(iii) H 0 (k) → w H (σ ) is surjective.
Consider the open and closed immersion Y × C H (σ ) ,→ Y × H ; this induces an
open and closed immersion Y ×C H (σ ) C H 0 (σ ) ,→ Y ×C H (σ ) H 0 whose composition
with the étale cover Y ×C H (σ ) H 0 → Y × H H 0 is easily checked (e.g., on geometric
points) to be an isomorphism. Therefore,

K ∗0 Y × H H 0 , C H 0 (σ ) ' K ∗0 Y ×C H (σ ) C H 0 (σ ), C H 0 (σ ) ' K ∗0 Y, C H (σ ) ,

where the last isomorphism is given by the Morita equivalence theorem (see [Th3,
Prop. 6.2]). By (i) and (iii) we can apply the argument at the beginning of the proof
and get an action of w H (σ ) on K ∗0 (Y × H H 0 , C H 0 (σ )) and therefore on K ∗0 (Y, C H (σ )),
as desired. It is not difficult to check that this action does not depend on the chosen
immersion H ,→ H 0 .
Finally, let us prove that there exists a closed immersion H ,→ H 0 satisfying
conditions (i) – (iii) above. First choose a closed immersion j : H ,→ GLn,k for some
n. Clearly,
H/C H (σ ) ,→ GLn,k /CGLn,k (σ ),
and, embedding σ in a maximal torus of GLn,k , it is easy to check that

GLn,k (k) → GLn,k /CGLn,k (σ ) (k)

is surjective. Now define H 0 as the inverse image of H/C H (σ ) in the normalizer

NGLn,k (σ ).
12 VEZZOSI and VISTOLI

COROLLARY 2.5
There is a canonical action of wG (σ ) on K ∗0 (X σ , CG (σ )) which induces an action on
K ∗0 (X σ , CG (σ ))geom .

Proof
Since CG (σ ) = C NG (σ ) (σ ), Lemma 2.4, applied to Y = X σ (resp., Y = Spec k)
and H = N G (σ ), yields an action of wG (σ ) on K ∗0 X σ , CG (σ ) resp., on

K 0 (Spec k, CG (σ )) = R(CG (σ )) . The multiplicative system S1 = rk−1 (1) is pre-

served by this action so that there is an induced action on the ring S1−1 R(CG (σ )). The
pullback
K 0 Spec k, CG (σ ) → K 0 X σ , CG (σ )

is wG (σ )-equivariant, and then wG (σ ) acts on K ∗0 (X σ , CG (σ ))geom .

Remark 2.6
If Y is regular, Lemma 2.4 also gives an action of w H (σ ) on K ∗ (Y, C H (σ )) since
K ∗ (Y, C H (σ )) ' K ∗0 (Y, C H (σ )) (see appendix). In particular, since by [Th5, Prop.
3.1], X σ is regular, Corollary 2.5 still holds for K ∗ (X σ , CG (σ ))geom .

Note also that the embedding of k-group schemes WG (σ ) ,→ Autk (σ ) induces, by

Example 2.1, an action of wG (σ ) on K 0 (Spec k, σ ) = R(σ ). The product in σ induces
a morphism of k-groups,
σ × CG (σ ) −→ CG (σ ),
which in its turn induces a morphism

m σ : K ∗ X σ , CG (σ ) −→ K ∗ X σ , σ × CG (σ ) .

LEMMA 2.7
If H ⊆ G is a subgroup scheme and if σ is contained in the center of H , there is a
canonical ring isomorphism

K ∗ (X σ , σ × H ) ' K ∗ (X σ , H ) ⊗ R(σ ).

Proof
Since σ acts trivially on X σ , we have an equivalence (see [SGA3, Exp. I, par. 4.7.3])
M
(σ × H ) − Coh X σ ' (H − Coh X σ ) (3)
σ
b

σ is the character group of σ ) which induces an isomorphism

(where b

K ∗0 (X σ , σ × H ) ' K ∗0 (X σ , H ) ⊗ R(σ ).
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 13

We conclude since K ∗ (Y, H ) ' K ∗0 (Y, H ) and K ∗ (X σ , σ × H ) ' K ∗0 (X σ , σ × H )

(see appendix).

.
For any essential dual cyclic subgroup σ ⊆ G, let 3 = 3(G,X ) , and consider the
composition

K ∗ (X, G)3 → K ∗ X, CG (σ ) 3 → K ∗ X σ , CG (σ ) 3

mσ
−→ K ∗ X σ , CG (σ ) 3 ⊗3 R(σ )3

−→ K ∗ X σ , CG (σ ) geom ⊗3 eR(σ )3 ,


(4)

where the first map is induced by group restriction, the last one is the geometric
localization map tensored with the projection R(σ )3 → e R(σ )3 , and we have used
Lemma 2.7 with H = CG (σ ); the second map is induced by restriction along X σ ,→
X which is a regular closed immersion (see [Th5, Prop. 3.1]) and therefore has finite
Tor-dimension, so that the pullback on K -groups is well defined (see appendix). It
is not difficult to show that the image of (4) is actually contained in the invariant
submodule

w (σ )
K ∗ (X σ , CG (σ ))geom ⊗3 e
R(σ )3 G ,
so that we get a map

wG (σ )
ψσ,X : K ∗ (X, G)3 −→ K ∗ (X σ , CG (σ ))geom ⊗3 e
R(σ )3 .

Our basic map is

. Y
9 X,G = ψσ,X : K ∗ (X, G)3
σ ∈C (G)

wG (σ )
K ∗ (X σ , CG (σ ))geom ⊗3 e
Y
−→ R(σ )3 . (5)
σ ∈C (G)

Note that 9 X,G is a morphism of R(G)-algebras as a composition of morphisms of

R(G)-algebras.
The following technical lemma is used in Propositions 3.5 and 4.6.

LEMMA 2.8
Let G be a linear algebraic k-group acting with finite stabilizers on a Noetherian
.
separated k-algebraic space X , and let 3 = 3(G,X ) . Let H ⊆ G be a subgroup, and
let σ be an essential dual cyclic subgroup contained in the center of H . Consider the
composition

K ∗0 (Y σ , H )3 −→ K ∗0 (Y σ , H )3 ⊗3 R(σ )3 −→ K ∗0 (Y σ , H )geom ⊗3 e
R(σ )3 , (6)
14 VEZZOSI and VISTOLI

where the first morphism is induced by the product morphism σ × H → H (recall

Lem. 2.7) and the second is the tensor product of the geometric localization morphism
with the projection R(σ )3 → e R(σ )3 . Then (6) factors through K ∗0 (Y σ , H )3 →
K ∗0 (Y σ , H )σ , yielding a morphism
θ H,σ : K ∗0 (Y σ , H )σ −→ K ∗0 (Y σ , H )geom ⊗3 e
R(σ )3 . (7)

Proof
Let S1 (resp., Sσ ) be the multiplicative subset in R(H )3 consisting of elements going
to 1 via the homomorphism rk H,3 : R(H )3 → 3 (resp., R(H )3 → e R(σ )3 ). Ob-
0 σ 0 σ
serve that K ∗ (X , H )3 ⊗3 R(σ )3 (resp., K ∗ (X , H )geom ⊗3 R(σ )3 ) is canonically
e
an R(H )3 ⊗ R(σ )3 -module (resp., an S1−1 R(H )3 ⊗ e R(σ )3 -module) and therefore
an R(H )-module via the coproduct ring morphism
1σ : R(H )3 −→ R(H )3 ⊗ R(σ )3
resp., via the ring morphism
1σ
f σ : R(H )3 −→ R(H )3 ⊗ R(σ )3 −→ S1−1 R(H )3 ⊗ e
R(σ )3 .

If we denote by A0 the R(H )3 -algebra f σ : R(H )3 −→ S1−1 R(H )3 ⊗ e

R(σ )3 , it is
enough to show that the localization homomorphism
lσ0 : A0 −→ Sσ−1 (A0 )
is an isomorphism, because in this case the morphism (7) is induced by the Sσ -
localization of (6). Let A denote the R(H )3 -algebra
λ1 ⊗ 1 : R(H )3 −→ S1−1 R(H )3 ⊗ e
R(σ )3 ,
where λ1 : R(H )3 → S1−1 R(H )3 denotes the localization homomorphism. It is a
well-known fact that there is an isomorphism of R(H )3 -algebras ϕ : A0 → A; this is
exactly the dual assertion to “the action H × σ → σ is isomorphic to the projection
on the second factor H × σ → σ .” Therefore, we have a commutative diagram
ϕ
A0 −−−−→ A
 
0 
lσ y
l
yσ
Sσ−1 A0 −−−−→ Sσ−1 A
Sσ−1 ϕ

where lσ denotes the localization homomorphism, and it is enough to prove that lσ is

R(σ )3 is a free 3-module of finite rank
an isomorphism. To see this, note that the ring e
(equal to φ(|σ |), φ being the Euler function), and there is a norm homomorphism
R(σ )3 −→ 3
N:e
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 15

sending an element to the determinant of the 3-endomorphism of e

R(σ )3 induced by
multiplication by this element; obviously, we have

∗
N−1 (3∗ ) = eR(σ )3 .

Analogously, there is a norm homomorphism

N0 : A0 = S1−1 R(H )3 ⊗ e
R(σ )3 −→ S1−1 R(H )3 ,

and

∗
N−1 (S1−1 R(H )3 )∗ = S1−1 R(H )3 ⊗ e
R(σ )3 .

There is a commutative diagram

N0
S1−1 R(H )3 ⊗ eR(σ )3 −−−−→ S1−1 R(H )3
 
rk H,3 ⊗ idy
 rk
y H,3
R(σ )3
e −−−−→ 3
N

H,3 (3 ) = (S1 R(H )3 ) . Therefore, by definition

and, by definition of S1 , we get rk−1 ∗ −1 ∗

of Sσ , Sσ /1 consist of units in A, and we conclude the proof of the lemma.

The following lemma, which is an easy consequence of a result of Merkurjev, is the

main tool in reducing the proof of the main theorem from G = GLn,k to its maximal
torus T .

LEMMA 2.9
Let X be a Noetherian separated algebraic space over k with an action of a split
reductive group G over k such that π1 (G) (see [Me, Par. 1.1]) is torsion free. Then if
T denotes a maximal torus in G, the canonical morphism

K ∗0 (X, G) ⊗R(G) R(T ) −→ K ∗0 (X, T )

is an isomorphism.

Proof
Let B ⊇ T be a Borel subgroup of G. Since R(B) ' R(T ) and K ∗0 (X, B) '
K ∗0 (X, T ) (see [Th4, proof of Th. 1.13, p. 594]), by [Me, Prop. 4.1], the canonical
ring morphism
K ∗0 (X, G) ⊗R(G) R(T ) −→ K ∗0 (X, T )
is an isomorphism.
16 VEZZOSI and VISTOLI

Since Merkurjev states his theorem for a scheme, we briefly indicate how it ex-
tends to a Noetherian separated algebraic space X over k. By [Th1, Lem. 4.3], there
exists an open dense G-invariant separated subscheme U ⊂ X . Since Merkurjev’s
map commutes with localization, by the localization sequence and Noetherian induc-
tion it is enough to know the result for U . And this is given in [Me, Prop. 4.1]. Note
that by [Me, Prop. 1.22], R(T ) is flat (actually free) over R(G), and therefore the
localization sequence remains exact after tensoring with R(T ).

The following is [Vi1, Lem. 3.2]. It is used frequently in the rest of the paper, and it
is stated here for the convenience of the reader.

LEMMA 2.10
Let W be a finite group acting on the left on a set A , and let B ⊆ A be a set of
representatives for the orbits. Assume that W acts on the left on a product of abelian
Q
groups of the type α∈A Mα in such a way that

s Mα = Msα

for any s ∈ W .
For each α ∈ B , let us denote by Wα the stabilizer of α in W . Then the canonical
projection Y Y
Mα −→ Mα
α∈A α∈B
induces an isomorphism
 Y W Y
Mα −→ (Mα )Wα .
α∈A α∈B

3. The main theorem: The split torus case

In this section, T is a split torus over k.

PROPOSITION 3.1
Let T 0 ⊂ T be a closed subgroup scheme (diagonalizable, by [SGA3, Exp. IX, par.
8.1]), finite over k. Then the canonical morphism
Y
δ : R(T 0 )3T 0 −→ R(σ )3T 0
e
σ dual cyclic
σ ⊆T 0

is a ring isomorphism.

Proof
Since both R(T 0 )3T 0 and eR(σ )3T 0 are free 3T 0 -modules of finite rank, it is enough
Q
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 17

to prove that, for any nonzero prime p - |T 0 |, the induced morphism of F p -vector
spaces Y
R(T 0 )3T 0 ⊗Z F p −→ R(σ )3T 0 ⊗Z F p
e (8)
σ dual cyclic
σ ⊆T 0

is an isomorphism. Now, for any finite abelian group A, we have an equality |A| =
AC ϕ(|C|), where ϕ denotes the Euler function, |H | denotes the order of the
P

group H , and the sum is extended to all cyclic quotients of A; applying this to the
group of characters Tb0 (so that the corresponding cyclic quotients C are exactly the
group of characters b σ for σ dual cyclic subgroups of T 0 ), we see that the ranks of
both sides in (8) coincide with |T 0 |, and it is then enough to prove that (8) is injective.
Define a morphism
Y Y
f : R(τ )3T 0 −→
e R(σ )3T 0
τ dual cyclic σ dual cyclic
τ ⊆T 0 σ ⊆T 0

of R(T 0 )3T 0 -modules by requiring, for any dual cyclic subgroup σ ⊆ T 0 , the commu-
tativity of the following diagram:
f
R(τ )3T 0 σ dual cyclic R(σ )3T 0
Q Q
τ dual cyclic e −−−−→
τ ⊆T 0 σ ⊆T 0
 

Prσ y
pr
y σ

R(τ )3T 0 R(σ )3T 0

Q ]
τ ⊆σ
e ←−−−−
ϕ

where Prσ and prσ are the obvious projections and ϕ is the isomorphism
resστ (e
prτ )τ
Q
τ ⊆σ
Y Y
R(σ )3T 0 −−−−−−→ R(τ )3T 0 −−−→ R(τ )3T 0
e
τ ⊆σ τ ⊆σ

induced by (1). Obviously, f ◦ δ coincides with the map

Y 0 Y
resσT : R(T 0 )3T 0 −→ R(σ )3T 0 ,
σ dual cyclic σ dual cyclic
σ ⊆T 0 σ ⊆T 0

so we are reduced to proving that

Y
R(T 0 )3T 0 ⊗Z F p −→ R(σ )3T 0 ⊗Z F p
σ dual cyclic
σ ⊆T 0

is injective, that is, that if A is a finite abelian group and p - |A|, then
Y
ϕ : F p [A] −→ F p [C] (9)
C∈{cyclic quotients ofA}
18 VEZZOSI and VISTOLI

b = HomAbGrps (A, C∗ ) denotes the complex characters group of A,

is injective. If A
then R( A)
b ' Z[A] and
Y b Y
ϕ= A
resCb : R( A)
b −→ R(C).
b
C
b C∈{cyclic
b subgroups of A}
b

Since p - |A|, it is enough to prove that if ξ ∈ R( A)

b ⊗˙ Z Z[1/|A|] has image via

A
resCb ⊗ Z[1/|A|] : R( A)
b ⊗Z Z[1/|A|] −→ R(C) b ⊗Z Z[1/|A|]
b

b then ξ ∈


contained in p R(C) b ⊗Z Z[1/|A|] for each cyclic C b ⊆ A,


p R( A)
b ⊗ ˙ Z Z[1/|A|] .
By [Se, p. 73], there exists (θCb0 ) ∈
Q
C
b C∈{cyclic
b b R(C)
subgroups of A}
b ⊗Z Z[1/|A|]
such that X
indCAb ⊗ Z[1/|A|] (θCb );

0
1=
b

C
b

therefore
X
ξ= ξ indCAb ⊗ Z[1/|A|] (θCb )
b
0

C
b
X
indCAb ⊗ Z[1/|A|] θCb (resCb
A
⊗ Z[1/|A|])(ξ )

0

=
b b

C
b

(by the projection formula), and we conclude the proof of the proposition.

Remark 3.2
The proof of Proposition 3.1 is similar to [Vi1, proof of Prop. 1.5], which is, however,
incomplete; that is why we have decided to give all the details here.

COROLLARY 3.3
We have the following.
(i) If σ 6 = σ 0 are dual cyclic subgroups of T , we have e
R(σ )σ 0 = 0 and e
R(σ )σ =
R(σ ).
e
(ii) If T 0 ⊂ T is a closed subgroup scheme, finite over k, and if σ is a dual cyclic
subgroup of T , we have R(T 0 )σ = 0 if σ * T 0 .
(iii) If T 0 ⊂ T is a closed subgroup scheme, finite over k, the canonical morphism
of R(T )-algebras Y
R(T 0 )3T 0 −→ R(T 0 )σ
σ dual cyclic
σ ⊆T 0

is an isomorphism.
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 19

Proof
(i) Suppose σ 6= σ 0 , and let T 0 ⊂ T be the closed subgroup scheme of T generated
by σ and σ 0 . The obvious morphism π : R(T )3T 0 → e R(σ )3T 0 × e
R(σ 0 )3T 0 factors
through R(T 0 )3T 0 → e R(σ )3T 0 × e
R(σ 0 )3T 0 , which is an epimorphism by Proposition
3.1. If ξ ∈ R(T )3T 0 with π(ξ ) = (0, 1) ⊗ 1, we have
ξ ∈ Sσ 0 ∩ ker R(T )3T 0 → e R(σ )3T 0 .

R(σ )σ 0 = 0. The second assertion is obvious.

Then e
(ii) and (iii) These follow immediately from (i) and Proposition 3.1.

Now let X be a regular Noetherian separated k-algebraic space on which T acts with
.
finite stabilizers, and let 3 = 3(T,X ) . Obviously, C (T ) is just the set of essential dual
cyclic subgroups of T since T is abelian.

PROPOSITION 3.4
We have the following.
(i) If jσ : X σ ,→ X denotes the inclusion, the pushforward ( jσ )∗ induces an
isomorphism
K ∗0 (X σ , T )σ −→ K ∗0 (X, T )σ .
(ii) The canonical localization morphism
Y
K ∗0 (X, T )3 −→ K ∗0 (X, T )σ
σ ∈C (T )

is an isomorphism, and the product on the left is finite.

Proof
(i) The proof is the same as that of [Th5, Th. 2.1], but we substitute Corollary 3.3(ii)
for [Th5, Th. 2.1] since we use a localization different from Thomason’s.
(ii) By the generic slice theorem for torus actions (see [Th1, Prop. 4.10]), there
exist a T -invariant nonempty open subspace U ⊂ X , a closed (necessarily diagonal-
izable) subgroup T 0 of T , and a T -equivariant isomorphism
0
U ' T /T 0 × (U/T ) ' (U/T ) ×T T.
Since U is nonempty and T acts on X with finite stabilizers, T 0 is finite over k
and K ∗0 (U, T ) ' K ∗0 (U/T ) ⊗Z R(T 0 ), by Morita equivalence theorem (see [Th3,
Prop. 6.2]) and [Th1, Lem. 5.6]. By Corollary 3.3(ii), the proposition for X = U fol-
lows from Corollary 3.3(iii). By Noetherian induction and the localization sequence
for K 0 -groups (see [Th3, Th. 2.7]), the statement for U implies the same for X .
Again using Noetherian induction, Thomason’s generic slice theorem for torus
actions, and (i), one similarly shows that the product σ ∈C (T ) K ∗0 (X, T )σ is finite.
Q
20 VEZZOSI and VISTOLI

By Proposition 3.4, there is an induced isomorphism (of R(T )-modules, not a ring
isomorphism due to the composition with pushforwards)

K ∗0 (X σ , T )σ −→ K ∗0 (X, T )3 .
Y
(10)
σ ∈C (T )

As shown in Lemma 2.8, the product morphism σ × T → T induces a morphism

θT,σ : K ∗0 (X σ , T )σ −→ K ∗0 (X σ , T )geom ⊗ e
R(σ )3 . (11)

PROPOSITION 3.5
For any σ ∈ C (T ), θT,σ is an isomorphism.

Proof
We write θ X,σ for θT,σ in order to emphasize the dependence of the map on the space.
We proceed by Noetherian induction on X σ . Let X 0 ⊆ X σ be a T -invariant closed
subspace, and let us suppose that (11) is an isomorphism with X replaced by any T -
invariant proper closed subspace Z of X 0 . By Thomason’s generic slice theorem for
torus actions (see [Th1, Prop. 4.10]), there exist a T -invariant nonempty open sub-
scheme U ⊂ X 0 , a (necessarily diagonalizable) subgroup T 0 of T , and a T -equivariant
isomorphism
U σ ≡ U ' T /T 0 × (U/T ) ' (U/T ) ×T T.
0

Since U is nonempty and T acts on X with finite stabilizers, T 0 is finite over k and,
.
obviously, 3T 0 ⊆ 3. Let Z σ ≡ Z = X 0 U . Since
/ K ∗0 (Z σ , T )σ / K ∗0 (X 0σ , T )σ / K ∗0 (U σ , T )σ /

θ Z ,σ θY 0 ,σ θU,σ
  
/ K ∗0 (Z σ , T )geom e )3
⊗ R(σ / K ∗0 (X 0σ , T )geom e )3
⊗ R(σ / K ∗0 (U σ , T )geom e )3
⊗ R(σ /

is commutative, by the induction hypothesis and the five-lemma it is enough to show

that θU,σ is an isomorphism. By Morita equivalence theorem (see [Th3, Prop. 6.2])
and [Th1, Lem. 5.6], K ∗0 (U, T ) ' K ∗0 (U/T ) ⊗Z R(T 0 ), so it is enough to prove that

θSpec k,σ : K 00 (Spec k, T 0 )σ = R(T 0 )σ → K 00 (Spec k, T 0 )geom ⊗ e

R(σ )3
= R(T 0 )geom ⊗ e
R(σ )3

is an isomorphism. But this follows immediately from Corollary 3.3(i) and (iii).

Combining Proposition 3.5 with (10), we get an isomorphism

K ∗0 (X σ , T )geom ⊗ e
Y
8 X,T : R(σ )3 −→ K ∗0 (X, T )3 . (12)
σ ∈C (T )
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 21

The following lemma is a variant of [Th5, Lem. 3.2], which already proves it after
tensoring with Q.

LEMMA 3.6
Let X be a Noetherian regular separated algebraic space over k on which a split
k-torus acts with finite stabilizers, and let σ ∈ C (T ). Let X σ denote the regular
σ -fixed subscheme, let jσ : X σ ,→ X be the regular closed immersion (see [Th5,
Prop. 3.1]), and let N ( jσ ) be the corresponding locally free conormal sheaf. Then,
for any T -invariant algebraic subspace Y of X σ , the cap-product

λ−1 N ( jσ ) ∩ (−) : K ∗0 (Y, T )σ −→ K ∗0 (Y, T )σ

is an isomorphism.

Proof
We proceed by Noetherian induction on closed T -invariant subspaces Y of X σ . The
statement is trivial for Y = ∅, so let us suppose Y nonempty and

λ−1 N ( jσ ) ∩ (−) : K ∗0 (Z , T )σ −→ K ∗0 (Z , T )σ

an isomorphism for any proper T -invariant closed subspace Z of Y . By Thoma-

son’s generic slice theorem for torus actions (see [Th1, Prop. 4.10]), there exist a
T -invariant nonempty open subscheme U ⊂ Y , a closed (necessarily diagonalizable)
subgroup T 0 of T , and a T -equivariant isomorphism

U σ ≡ U ' T /T 0 × (U/T ).

Since U is nonempty and T acts on X with finite stabilizers, T 0 is finite over k. Using
the localization sequence and the five-lemma, we reduce ourselves to showing that

λ−1 N ( jσ ) ∩ (−) : K ∗0 (U, T )σ −→ K ∗0 (U, T )σ

is an isomorphism. For this, it is enough to show that (the restriction of) λ−1 (N ( jσ ))
is a unit in K 0 (U, T )σ ' K 0 (U/T )3 ⊗ R(T 0 )σ (see [Th3, Prop. 6.2]). Decomposing
N ( jσ ) according to the characters of T 0 , we may write, shrinking U if necessary,
M rρ
N ( jσ ) = OU/T ⊗ Lρ ,
ρ∈Tb0

where Lρ is the line bundle attached to the T 0 -character ρ and rρ ≥ 0, and there-
fore λ−1 (N ( jσ )) = ρ∈Tb0 (1 − ρ)rρ in K 0 (U/T ) ⊗ R(T 0 ). The localization map
Q

R(T 0 )3 → R(T 0 )σ ' e

R(σ )3 coincides with the composition
πσ pσ
R(T 0 )3 −→ R(σ )3 −→ e
R(σ )3
22 VEZZOSI and VISTOLI

of the restriction to σ followed by the projection (see Cor. 3.3), and then
M rχ
(id K 0 (U/T )3 ⊗πσ ) N ( jσ ) = OU/T ⊗ Lχ ,

χ∈b
σ \{0}

in K 0 (U/T )3 ⊗ R(σ )3 , where the summand omits the trivial character since the de-
composition of N ( jσ ) according to the characters of σ has vanishing fixed subsheaf
N ( jσ )0 (see, e.g., [Th5, Prop. 3.1]). Therefore,
Y
λ−1 (id K 0 (U/T )3 ⊗πσ )(N ( jσ )) = (1 − χ)rχ ,

χ ∈b
σ \{0}

and it is enough to show that the image of 1 − χ in e R(σ )3 via pσ is a unit for any
nontrivial character χ of σ . Now, the image of such a χ in
3[t]
R(σ )3 '
e
(8|σ | )

(8|σ | being the |σ |th cyclotomic polynomial) is of the form [t l ] for some 1 ≤ l < |σ |,
where [−] denotes the class mod8|σ | ; therefore the cokernel of the multiplication by
1 − [t l ] in 3[t](8|σ | ) is
3[t]
=0
(8|σ | , 1 − t l )
since 8|σ | and (1 − t l ) are relatively prime in 3[t] for 1 ≤ l < |σ |. Thus 1 − [t l ] is a
unit in 3[t](8|σ | ), and we conclude the proof of the lemma.

We are now able to prove our main theorem for G = T .

THEOREM 3.7
If X is a regular Noetherian separated k-algebraic space, then

K ∗ (X σ , T )geom ⊗ e
Y
9 X,T : K ∗ (X, T )3 −→ R(σ )3
σ ∈C (T )

is an isomorphism of R(T )-algebras.

Proof
Recall (see appendix) that K ∗ (X, T ) ' K ∗0 (X, T ) and K ∗ (X σ , T ) ' K ∗0 (X σ , T ),
since both X and X σ are regular (see [Th5, Prop. 3.1]). Since 8 X,T is an isomor-
phism of R(T )-modules, it is enough to show that the composition 9 X,T ◦ 8 X,T is an
isomorphism. A careful inspection of the definitions of 9 X,T and 8 X,T easily reduces
the problem to proving that, for any σ ∈ C (T ), the composition
jσ ∗ jσ∗
K ∗0 (X σ , T )σ −→ K ∗0 (X, T )σ −→ K ∗0 (X σ , T )σ
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 23

is an isomorphism, jσ : X σ ,→ X being the natural inclusion. Since jσ is regular,

there is a self-intersection formula

jσ∗ ◦ jσ ∗ (−) = λ−1 N ( jσ ) ∩ (−),

(13)

N ( jσ ) being the conormal sheaf associated to jσ , and we conclude by Lemma 3.6. To

prove the self-intersection formula (13), we adapt [Th5, proof of Lem. 3.3]. First, by
Proposition 3.4(i), jσ ∗ is an isomorphism, so it is enough to prove that jσ ∗ jσ∗ jσ ∗ (−) =
jσ ∗ (λ−1 (N ( jσ )) ∩ (−)). By the projection formula (see Prop. A.5), we have

jσ ∗ jσ∗ jσ ∗ (−) = jσ ∗ jσ∗ (1) ∩ jσ ∗ (−) = jσ ∗ jσ∗ (O X ) ∩ jσ ∗ (−)

= jσ ∗ (O X σ ) ∩ jσ ∗ (−) = jσ ∗ jσ∗ (O X σ ) ∩ (−) .

Now, as explained in the appendix, to compute jσ∗ (O X σ ) we choose a complex F ∗ of

flat quasi-coherent G-equivariant modules on X which is quasi-isomorphic to O X σ ,
and then
X
jσ∗ (O X σ ) = [ jσ∗ (F ∗ )] = [F ∗ ⊗ O X σ ] = (−1)i [H i (F ∗ ⊗ O X σ )].
i

But F ∗ is a flat resolution of O X σ , so H i (F ∗ ⊗ O X σ ) = ToriO X (O X σ , O X σ ) '

Vi
N ( jσ ), where the last isomorphism (see [SGA6, Exp. VII, par. 2.5]) is natural
and hence T -equivariant. Therefore, jσ∗ (O X σ ) = λ−1 (N ( jσ )), and we conclude the
proof of the theorem.

4. The main theorem: The case of G = GLn,k

In this section we use the result for 9 X,T to deduce the same result for 9 X,GLnk .

THEOREM 4.1
Let X be a Noetherian regular separated algebraic space over a field k on which
G = GLn,k acts with finite stabilizers. Then the map defined in (5),

9 X,G : K ∗ (X, G)3(G,X )

wG (σ )
K ∗ (X σ , C(σ ))geom ⊗3(G,X ) e
Y
−→ R(σ )3(G,X ) , (14)
σ ∈C (G)

is an isomorphism of R(G)-algebras and the product on the right is finite.

Throughout this section, entirely devoted to the proof of Theorem 4.1, we simply
write G for GLn,k , 3 for 3(G,X ) , and T for the maximal torus of diagonal matrices
in GLn,k . First, let us observe that we can choose each σ ∈ C (G) contained in T .
Moreover, 3(T,X ) = 3(G,X ) .
24 VEZZOSI and VISTOLI

We need the following three preliminary lemmas (Lems. 4.2, 4.3, 4.4).
If σ, τ ⊂ T are dual cyclic subgroups, they are conjugate under the G(k)-action
if and only if they are conjugated via an element in the Weyl group Sn . For any group
scheme H with a dual cyclic subgroup σ ⊆ H , we denote by mσH the kernel of
R(H )3 → e R(σ )3 and by R(H\ )3,σ the completion of R(H )3 with respect to the
H
ideal mσ .
The following lemma is essentially a variant of Lemma 2.9 for σ -localizations.

LEMMA 4.2
Let G = GLn,k , let T be the maximal torus of G consisting of diagonal matrices, and
let X be an algebraic space on which G acts with finite stabilizers.
(i) For any essential dual cyclic subgroup σ ⊆ T , the morphisms

ωσ,geom : K ∗0 X σ , CG (σ ) geom ⊗R(CG (σ ))3 R(T )3 −→ K ∗0 (X σ , T )geom ,

ωσ : K ∗0 X σ , CG (σ ) σ ⊗R(CG (σ ))3 R(T )3 −→ K ∗0 (X σ , T )σ

induced by T ,→ CG (σ ) are isomorphisms.

(ii) For any essential dual cyclic subgroup σ ⊆ T ,

(mC G (σ ) ) N · K 0 X σ , C (σ )


σ ∗ G σ
= 0, N  0,

and the morphism induced by T ,→ CG (σ ),

cσ : K ∗0 X σ , CG (σ ) σ ⊗R(C\
ω )3,σ −→ K ∗0 (X σ , T )σ ,


\
R(T
(σ ))
G 3,σ

is an isomorphism.

Proof
(i) Since CG (σ ) is isomorphic to a product of general linear groups over k and since
T is a maximal torus in CG (σ ), by Lemma 2.9 the canonical ring morphism

K ∗0 X, CG (σ ) ⊗R(CG (σ )) R(T ) −→ K ∗0 (X, T )

(15)

is an isomorphism. If H ⊆ G is a subgroup scheme, we denote by SσH the multi-

plicative subset of R(H )3 consisting of the elements sent to 1 by the canonical ring
homomorphism R(H )3 → e R(σ )3 . By (15), ωσ coincides with the composition

K ∗0 X σ , CG (σ ) σ ⊗R(CG (σ ))3 R(T )3

' K ∗0 (X σ , T ) ⊗R(CG (σ ))3 (SσCG (σ ) )−1 R(CG (σ ))3 ⊗R(CG (σ ))3 R(T )3

id ⊗νσ
−−−→ K ∗0 X σ , CG (σ ) ⊗R(CG (σ ))3 (SσT )−1 R(T )3 ' K ∗0 (X σ , T )σ ,


K -THEORY OF ACTIONS WITH FINITE STABILIZERS 25

where

νσ : (SσCG (σ ) )−1 R CG (σ ) ⊗R(CG (σ ))3 R(T )3 → (SσT )−1 R(T )3

3
(16)

is induced by T ,→ CG (σ ) and the last isomorphism follows from (15); the same is
true for ωσ,geom . Therefore, it is enough to prove that νσ and
C (σ ) −1
νσ,geom : (S1 G ) R CG (σ ) 3 ⊗R(CG (σ ))3 R(T )3 → (S1T )−1 R(T )3

C (σ )
are isomorphisms; that is, if Sτ denotes the image of S1 G via the restriction map

R CG (σ ) 3 −→ R(T )3 ,

then SτT /1 consists of units in (Sτ )−1 R(T )3 for τ = 1 and τ = σ . If 1σ denotes the
Weyl group of CG (σ ), which is a product of symmetric groups, we have R(CG (σ )) '
R(T )1σ and therefore

1
(SτCG (σ ) )−1 R CG (σ ) 3 ' (Sτ )−1 R(T )3 σ

since R(T ) is torsion free. Moreover, there is a commutative diagram

C (σ ) −1 /
(Sτ G ) R CG (σ ) 3 (Sτ )−1 R(T )3


jj
j jjjjjj
ψ
jjjj=ϕ .
 tjjjj 
R(τ )3 ' (Sτ ) R(τ )3
e T −1 o (Sτ )−1 R(T )3
T
T −1(Sτ )T resτ

where ψ is induced by eπτ and the isomorphism e R(τ )3 ' (SτT )−1 R(τ )3 is obtained
from Proposition 3.1 and Corollary 3.3. If we define the map

1
M : (Sτ )−1 R(T )3 −→ (Sτ )−1 R(T )3 σ ,
Y
ξ 7−→ g·ξ ,
g∈1σ

it is easily checked that for ξ ∈ (Sτ )−1 R(T )3 , ξ is a unit if M(ξ ) is a unit, and that
ψ(M(ξ )) = 1 implies that ξ is a unit in ((Sτ )−1 R(T )3 )1σ . But ϕ is 1σ -equivariant,
and therefore SτT /1 consists of units in (Sτ )−1 R(T )3 for τ = 1 or σ .
(ii) Since R(CG (σ )) → R(T ) is faithfully flat, by (i) it is enough to prove that

(mσT ) N K ∗0 (X σ , T )σ = 0 for N  0. (17)

But (17) can be proved using the same technique as in the proof of, for example,
Proposition 3.5, that is, Noetherian induction together with Thomason’s generic slice
26 VEZZOSI and VISTOLI

theorem for torus actions. The second part of (ii) follows, arguing as in (i), from the
fact that (16) is an isomorphism since

K ∗0 X σ , CG (σ )
R C\ (σ ) 0 σ
, (σ ) ,




σ
⊗R CG (σ ) G 3,σ
' K ∗ X C G σ
3

K ∗0 (X σ , T )σ ⊗R(T )3 R(T
\ )3,σ ' K ∗0 (X σ , T )σ .

If σ, τ ⊂ T are dual cyclic subgroups conjugated under G(k), they are conjugate
through an element of the Weyl group Sn and we write τ ≈ Sn σ ; moreover, we have
mσG = mτG because conjugation by an element in Sn (actually, by any element in G(k))
induces the identity morphism on K -theory and, in particular, on the representation
ring. Then there are canonical maps
Y
\
R(G) 3,σ ⊗ R(G)3 R(T )3 −→ \
R(T )3,τ , (18)
τ dual cyclicτ ≈ Sn σ

G (σ ) 3,σ ⊗ R(CG (σ ))3 R(T )3 −→ R(T

R C\ )3,σ .


\ (19)

LEMMA 4.3
Maps (18) and (19) are isomorphisms.

Proof
Since R(G) = R(T ) Sn → R(T ) is finite, the canonical map R(G)
\ 3,σ ⊗R(G)3
σ mG σ mG
R(T )3 → R(T
\ )3 (where R(T\ )3 denotes the mσG -adic completion of the
R(G)3 -module R(T )3 ) is an isomorphism. Moreover, R(G)3 = (R(T )3 ) Sn implies
that q \ q \
mσ R(T )3 =
G mτT = mτT ,
τ dual cyclic τ dual cyclic
τ ≈ Sn σ τ ≈ Sn σ

and, by Corollary 3.3(i), mτT + mτT0 = R(T )3 if τ 6= τ 0 . By the Chinese remainder

lemma, we conclude that (18) is an isomorphism.
Arguing in the same way, we get that the canonical map
Y
G (σ ) 3,σ ⊗R(CG (σ ))3 R(T )3 −→
R C\ )3,τ


\
R(T
τ dual cyclic
τ ≈1σ σ

is an isomorphism, where 1σ = Sn ∩ CG (σ ) is the Weyl group of CG (σ ) with respect

to T and we write τ ≈1σ σ to denote that τ and σ are conjugate through an element
of 1σ . But 1σ ⊂ CG (σ ), so that τ ≈1σ σ if and only if τ = σ , and we conclude
that (19) is an isomorphism.
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 27

LEMMA 4.4
For any essential dual cyclic subgroup σ ⊆ G, the canonical morphism

3,σ −→ R CG (σ ) 3,σ
\
R(G) \

is a finite étale Galois cover (see [SGA1, Exp. V]) with Galois group wG (σ ).

Proof
Since R(T ) is flat over R(G) = R(T ) Sn , we have

Sn
\
R(G) \
3,σ ' R(G)3,σ ⊗R(G)3 R(T )3

Sn
R(G)3 R(T )3
\ ⊗
' R(G)3,σ
 Y  Sn
' \
R(T )3·τ ,
τ dual cyclic
τ ≈ Sn σ

the last isomorphism being given in Lemma 4.3. By Lemma 2.10, we get

S
\σ ' R(T
R(G) \ )σ n,σ ,

where Sn acts on the set of dual cyclic subgroups of T which are Sn -conjugated to σ
and where Sn,σ denotes the stabilizer of σ . Analogously, denoting by 1σ the Weyl
group of CG (σ ), by Lemma 4.2(ii) we have

R C\ G (σ ) 3,σ ' R CG (σ ) 3,σ ⊗R C (σ )

\

R(T )3
1σ
G 3

1σ
' R(C\G (σ ))3,σ ⊗R(CG (σ ))3 R(T )3

1
\
' R(T )3,σ σ ,

where the last isomorphism is given by Lemma 4.3. From the exact sequence

1 → 1σ −→ Sn,σ −→ wG (σ ) → 1,

we conclude that

wG (σ )
\
R(G) G (σ ))3,σ
3,σ ' R(C\ . (20)
By [SGA1, Prop. 2.6, Exp. V], it is now enough to prove that the stabilizers of ge-
G (σ ))3,σ under the


ometric points (i.e., the inertia groups of points) of Spec R(C\
wG (σ )-action are trivial.
First, let us observe that Spec(e R(σ )3 ) is a closed subscheme of
Spec(R(C\ G (σ ))3,σ ). This can be seen as follows. It is obviously enough to
show that if s denotes the order of σ , the map
3[t]
πσ : R CG (σ ) −→ R(σ )3 =


3 (t s − 1)
28 VEZZOSI and VISTOLI

is surjective. First, consider the case where σ is contained in the center of G.

Since R(σ )3 is of finite type over 3, we show that for any prime∗ p - s the
induced map
πσ, p : R CG (σ ) 3 ⊗ F p −→ R(σ )3 ⊗ F p

is surjective. Note that if E denotes the standard n-dimensional representation of G,

πσ sends r E to nr t r . If p - n, then πσ, p is surjective (in fact, πσ (E) = nt and
V

n is invertible in F p ). If p | n, let us write n = qm, with q = pi and p - m. Since

(s, q) = 1, t q is a ring generator of R(σ )3 , and to prove πσ, p is injective, it is enough
to show that p - qn . This is elementary since the binomial expansion of

(1 + X )n = (1 + X q )m

in F p [X ] yields qn = m in F p . For a general σ ⊆ T , let CG (σ ) = li=1 GLdi ,k ,

Q

d = n, and let σi denote the image of σ in GLdi ,k , i = 1, . . . , l. Since

P
where
Ql i
σ ⊆ i=1 σi is an inclusion of diagonalizable groups, the induced map
l
Y  Ol
R σi = R(σi ) −→ R(σ )
i=1 i=1

is surjective (e.g., see [SGA3, Vol. II]). But R(CG (σ ))3 → R(σ )3 factors as
l
O l
O
R CG (σ ) R(GLdi ,k )3 −→ R(σi )3 −→ R(σ )3 ,


3
=
i=1 i=1

and also the first map is surjective (by the previous case, since σi is contained in the
center of GLdi ,k and |σi | divides |σ |). This proves that Spec(e R(σ )3 ) is a closed sub-
scheme of Spec(R(C\ G (σ ))3,σ ). Since R(C\G (σ ))3,σ is the completion of R(CG (σ ))3
along the ideal
ker R(CG (σ ))3 → e R(σ )3 ,

G (σ ))3,σ meets the closed sub-

any nonempty closed subscheme of Spec R(C\
scheme Spec(e R(σ )3 ). To prove that wG (σ ) acts freely on the geometric points of
G (σ ))3,σ , it is then enough to show that it acts freely on the geometric


Spec R(C\
points of Spec(e R(σ )3 ).
Actually, more is true: the map q : Spec(e R(σ )3 ) → Spec(3) is a (Z/sZ)∗ -
torsor† . In fact, if Spec() → Spec(3) is a geometric point, the corresponding geo-
metric fiber of q is isomorphic to the spectrum of
Y [t] Y
' 
(t − αi )
αi ∈e
µs () αi ∈e
µs ()

∗ Recall that σ is essential; hence s is invertible in 3.

† Recall that the constant group scheme associated to (Z/sZ)∗ is isomorphic to Autk (σ ).
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 29

and (Z/sZ)∗ acts by permutation on the primitive roots e

µs (), by α 7→ α k , (k, s) =
1. In particular, the action of the subgroup wG (σ ) ⊂ (Z/sZ)∗ on Spec(e R(σ )3 ) is
free.

PROPOSITION 4.5
The canonical morphism

wG (σ )
K ∗0 (X σ , CG (σ ))σ
Y
K ∗0 (X, G)3 −→
σ ∈C (G)

is an isomorphism.

Proof
By Lemma 2.9, the canonical ring morphism

K ∗0 (X, G) ⊗R(G) R(T ) −→ K ∗0 (X, T )

is an isomorphism. Since R(G) → R(T ) is faithfully flat, it is enough to show that

K ∗0 (X, T )3 ' K ∗0 (X, G)3 ⊗R(G)3 R(T )3

w (σ )
K ∗0 (X σ , CG (σ ))σ G ⊗R(G)3 R(T )3
Y
−→
σ ∈C (G)

is an isomorphism. By Proposition 3.4(ii), we are left to prove that

wG (σ )
K ∗0 (X σ , CG (σ ))σ
Y Y
⊗R(G)3 R(T )3 ' K ∗0 (X, T )σ . (21)
σ ∈C (G) σ dual cyclic
σ ⊂T

For any τ ∈ C (G) (τ ⊆ T , as usual), we have

K ∗0 X τ , CG (τ ) τ ⊗R(G)
\ R(T )3,τ


\
3,τ

' K ∗0 (X τ , CG (τ ))τ ⊗R(C\ G (τ ))3,τ ⊗R(G) )3,τ

(τ ))
R(C\ \
\
R(T
G 3,τ 3,τ

K ∗0 (X τ , CG (τ ))τ G (τ ))3,τ ⊗R(C\ )3,τ .

' ⊗ R(G)
\ R(C\ (τ ))
\
R(T
3,τ G 3,τ

\
By Lemma 4.4, for any R(G)3,τ -module M, we have

\ (τ ) ' wG (τ ) × M


M ⊗R(G)
\ R C G 3,τ
3,τ

since a torsor is trivial when base changed along itself. Therefore,

K ∗0 X τ , CG (τ ) τ ⊗R(G)
\ R(T )3,τ


\
3,τ

' wG (τ ) × K ∗0 (X τ , CG (τ ))τ ⊗R(C\ )3,τ

\
R(T (22)
(τ )) G 3,τ
30 VEZZOSI and VISTOLI

with wG (τ ) acting on left-hand side by left multiplication on wG (τ ). Applying

Lemma 4.2(ii) to the left-hand side, we get
K ∗0 X τ , CG (τ ) τ ⊗R(G) τ
\ R(T )3,τ ' wG (τ ) × K ∗ (X , T )τ ,
0


\
3,τ

and taking invariants with respect to wG (τ ),

wG (τ )
K ∗0 (X τ , CG (τ ))τ ⊗R(G)
\ R(T )3,τ
\ ' K ∗0 (X τ , T )τ . (23)
3,τ

Comparing (21) to (23), we are reduced to proving that for any σ ∈ C (G) there is an
isomorphism

w (σ )
K ∗0 (X σ , CG (σ ))σ G ⊗R(G)3 R(T )3

wG (τ )
K ∗0 (X τ , CG (τ ))τ ⊗R(G)
Y
' \
\
R(T )3,τ .
3,τ
τ dual cyclic
τ ≈ Sn σ

Since R(T\ )3,τ is flat over R(G)\ 3,τ and wG (τ ) acts trivially on it, we have (see
[SGA1])

w (τ )
K ∗0 (X τ , CG (τ ))τ ⊗R(G)
\
\
R(T )3,τ G
3,τ

w (τ )
' K ∗0 (X τ , CG (τ ))3,τ G ⊗R(G)\ R(T )3,τ .
\
3,τ

By Lemma 4.3, we have isomorphisms

w (σ )
K ∗0 (X σ , CG (σ ))σ G ⊗R(G)3 R(T )3

w (σ )
' K ∗0 (X σ , CG (σ ))σ G ⊗R(G) \
\
R(G)3,σ ⊗R(G)3 R(T )3
3,σ

w (σ )
K ∗0 (X σ , CG (σ ))σ G ⊗R(G)
Y
\ R(T )3,τ .
' \
3,σ
τ dual cyclic
τ ≈ Sn σ

3,τ for any τ ≈ Sn σ , since mσ = mτ .) For each

(Recall that R(G) \ \ G G
3,σ = R(G)
τ , choosing an element g ∈ Sn such that gσ g −1 = τ determines an isomorphism
K ∗0 (X σ , CG (σ ))σ ' K ∗0 (X τ , CG (τ ))τ whose restriction to invariants

w (σ )
w (τ )
K ∗0 (X σ , CG (σ ))σ G ' K ∗0 (X τ , CG (τ ))τ G
is independent of the choice of g. Therefore, we have a canonical isomorphism

w (σ )
K ∗0 (X σ , CG (σ ))σ G ⊗R(G)3 R(T )3

w (σ )
K ∗0 (X σ , CG (σ ))σ G ⊗R(G)
Y
\ R(T )3,τ
' \
3,σ
τ dual cyclic
τ ≈ Sn σ

wG (τ )
K ∗0 (X τ , CG (τ ))τ
Y
\ R(T )3,τ ,
' ⊗R(G) \
3,τ
τ dual cyclic
τ ≈ Sn σ
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 31

as desired.

Since K ∗ (X, G) ' K ∗0 (X, G) and K ∗ (X σ , CG (σ )) ' K ∗0 (X σ , CG (σ )), comparing

Proposition 4.5 with (14), we see that the proof of Theorem 4.1 can be completed by
the following.

PROPOSITION 4.6
For any σ ∈ C (G), the morphism given by Lemma 2.8 and induced by the product
CG (σ ) × σ → CG (σ ),

θCG (σ ),σ : K ∗0 X σ , CG (σ ) σ −→ K ∗0 X σ , CG (σ ) geom ⊗ e

R(σ )3 ,

is an isomorphism.

Proof
To simplify the notation, we write θσ for θCG (σ ),σ . As usual, we may suppose σ
contained in T . Since CG (σ ) is isomorphic to a product of general linear groups
over k, we can take T as its maximal torus, and by Lemma 2.9, the canonical ring
morphism
K ∗0 X, CG (σ ) ⊗R(CG (σ )) R(T ) −→ K ∗0 (X, T )

is an isomorphism. Moreover, R(CG (σ )) → R(T ) being faithfully flat, it is enough

to prove that θσ ⊗ idR(T ) is an isomorphism. To prove this, let us consider the com-
mutative diagram
θσ ⊗id
K ∗0 X σ , CG (σ ) σ ⊗R(CG (σ ))3 R(T )3


/ K ∗0 (X, CG (σ ))geom ⊗ e
R(σ )3 ⊗R(CG (σ ))3 R(T )3

ωσ γf
σ

 
K ∗0 (X σ , T )σ / K ∗0 (X σ , T )geom ⊗ e
R(σ )3
θT,σ

where
• K ∗ (X σ , CG (σ ))geom ⊗ eR(σ )3 is an R(CG (σ ))3 -module via the coproduct
ring morphism 1CG (σ ) : R(CG (σ ))3 → R(CG (σ ))3 ⊗ e R(σ )3 (induced by
the product CG (σ ) × σ → CG (σ ));
• ωσ is the canonical map induced by the inclusion T ,→ CG (σ ) and is an
isomorphism by Lemma 4.2;
• θT,σ is an isomorphism as shown in the proof of Theorem 3.7;
• γeσ sends (x ⊗ u) ⊗ t to (1T (t) · x|T ) ⊗ u, for x ∈ K ∗ (X σ , CG (σ ))geom ,
u ∈e R(σ )3 , t ∈ R(T )3 , 1T : R(T )3 → R(T )3 ⊗ e R(σ )3 being the coproduct
induced by the product T × σ → T .
So we are left to prove that γeσ is an isomorphism.
32 VEZZOSI and VISTOLI

First, let us observe that if R is a ring, A → A0 is a ring morphism, and M is an

A-module, there is a natural isomorphism

(M ⊗Z R) ⊗ A⊗Z B (A0 ⊗Z R) −→ (M ⊗ A A0 ) ⊗Z R,
(m ⊗ r1 ) ⊗ (a 0 ⊗ r2 ) 7 −→ (m ⊗ a 0 ) ⊗ r1r2 .

Applying this to M = K ∗ (X σ , CG (σ ))geom , R = e

R(σ )3 , A = R(CG (σ ))3 , A0 =
R(T )3 and using Lemma 4.2, we get a canonical isomorphism

f : K ∗0 (X σ , T )geom ⊗ e
R(σ )3
−→ K ∗0 (X σ , CG (σ ))geom ⊗ e

0
R(σ )3 ⊗R(CG (σ ))3 ⊗e ) ) , (24)


R(σ )3 R(T 3 ⊗ R(σ
e 3

where we have denoted by (R(T )3 ⊗ eR(σ )3 )0 the R(CG (σ ))3 ⊗ e

R(σ )3 -algebra

res ⊗ id : R CG (σ ) 3 ⊗ e
R(σ )3 −→ R(T )3 ⊗ e R(σ )3 .

It is an elementary fact that there are mutually inverse isomorphisms αCG (σ ) , βCG (σ ) ,
and αT , βT fitting into the commutative diagrams

R CG (σ ) 3 ⊗ e R(σ )3


(25)
mm6
1C G (σ ) mmm
m
mm mmm
m

m
R CG (σ ) 3 αC G (σ )
QQQ
QQQ
QQQ
id⊗1 QQQQ(


R CG (σ ) 3 ⊗ e R(σ )3

R(T 7)3 ⊗ e
R(σ )3 (26)
oo
1T ooo
ooo
ooo
R(T )3 O αT
OOO
OOO
O
id⊗1 OOO' 
R(T )3 ⊗ eR(σ )3
and compatible with restriction maps induced by T ,→ CG (σ ). This is exactly the
dual assertion to the general fact that “an action H × Y → Y is isomorphic over X to
the projection on the second factor pr2 : H × Y → Y ,” for any group scheme H and
any algebraic space Y . From (25) we get an isomorphism

0
α : R(CG (σ ))3 ⊗ e
e R(σ )3 ⊗R(CG (σ ))3 R(T )3 −→ R(T )3 ⊗ e R(σ )3 ,
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 33

0
where R(CG (σ ))3 ⊗ e
R(σ )3 denotes the R(CG (σ ))3 -algebra

1CG (σ ) : R CG (σ ) 3 → R CG (σ ) 3 ⊗ e
R(σ )3 .

Therefore, if we denote by (R(T )3 ⊗ e R(σ )3 )00 the R(CG (σ ))3 ⊗ e

R(σ )3 -algebra

(res ⊗ id) ◦ αT : R CG (σ ) 3 ⊗ eR(σ )3 −→ R(T )3 ⊗ e R(σ )3 ,

the composition

K ∗0 (X σ , CG (σ ))geom ⊗ e

0
R(σ )3 ⊗R(CG (σ ))3 R(T )3
= K ∗0 (X σ , CG (σ ))geom ⊗ eR(σ )3


O
0
R(CG (σ ))3 ⊗ e
R(σ )3 ⊗R(CG (σ ))3 R(T )3

R(CG (σ ))3 ⊗e
R(σ )3
α
id ⊗e
−→ K ∗0 (X σ , CG (σ ))geom ⊗ e
R(σ )3


O
00
R(T )3 ⊗ eR(σ )3
R(CG (σ ))3 ⊗e
R(σ )3
id ⊗βT
K ∗0 (X σ , CG (σ ))geom ⊗ e
R(σ )3


−→
O
0
R(T )3 ⊗ e
R(σ )3
R(CG (σ ))3 ⊗e
R(σ )3

' K ∗0 (X σ , T )geom ⊗ e
R(σ )3

is an isomorphism and it can be easily checked to coincide with γeσ .

5. The main theorem: The general case

In this section, we use Theorem 4.1 to deduce the same result for the action of a
linear algebraic k-group G, having finite stabilizers, on a regular separated Noetherian
k-algebraic space X . We write 3 for 3(G,X ) .
We start with a general fact.

PROPOSITION 5.1
Let X be a regular Noetherian separated k-algebraic space on which a linear alge-
braic k-group G acts with finite stabilizers. Then there exists an integer N > 0 such
that if a1 , . . . , a N ∈ K 0 (X, G)geom have rank zero on each connected component of
QN
X , then the multiplication by i=1 ai on K ∗0 (X, G)geom is zero.

In particular, we have the following.

34 VEZZOSI and VISTOLI

COROLLARY 5.2
Let X be a regular Noetherian separated k-algebraic space with a connected action of
a linear algebraic k-group G having finite stabilizers. Then the geometric localization

rk0,geom : K 0 (X, G)geom −→ 3

of the rank morphism has a nilpotent kernel.

Proof of Proposition 5.1

Let us choose a closed immersion G ,→ GLn,k (for some n > 0). By Morita equiva-
lence,
K ∗0 (X, G) ' K ∗0 (X ×G GLn,k , GLn,k )
and
K 0 (X, G) ' K 0 (X ×G GLn,k , GLn,k ).
Moreover, 3(X ×G GLn,k ,GLn,k ) = 3. Let ξ = x/s ∈ K ∗0 (X, G)geom with x ∈
K ∗0 (X, G)3 and s ∈ rk−1 (1) where rk : R(G) → 3 is the rank morphism,
and let ai = αi /si with αi ∈ K 0 (X, G)3 and si ∈ rk−1 (1) for i = 1, . . . , N .
Let us consider the elements x/1 in K ∗0 (X ×G GLn,k , GLn,k )geom , and αi /1 in
K 0 (X ×G GLn,k , GLn,k )geom for i = 1, . . . , N . Since the canonical homomorphism

K ∗0 (X ×G GLn,k , GLn,k )geom −→ K ∗0 (X, G)geom

is a morphism of modules over the ring morphism

K 0 (X ×G GLn,k , GLn,k )geom −→ K 0 (X, G)geom ,

if the theorem holds for G = GLn,k and if N is the corresponding integer, the product
i αi /1 in K 0 (X, G)geom annihilates x/1 ∈ K ∗ (X, G)geom and a fortiori
0
Q Q
i ai an-
nihilates ξ in K ∗0 (X, G)geom . So, we may assume G = GLn,k . Let T be the maximal
torus of diagonal matrices in G. By Lemma 4.2(i) with σ = 1, there are isomorphisms

ω1,geom : K 0 (X, GLn,k )geom ⊗R(GLn,k )3 R(T )3 ' K 0 (X, T )geom ,

K ∗0 (X, GLn,k )geom ⊗R(GLn,k )3 R(T )3 ' K ∗0 (X, T )geom .

Since R(GLn,k ) → R(T ) is faithfully flat and the diagram

rk0,geom ⊗ id
K 0 (X, GLn,k )geom ⊗R(GLn,k )3 R(T )3 −−−−−−−→ 3 ⊗R(GLn,k ) R(T )3
 
ω1,geom 
y
id ⊗ rk
y 3 T

K 0 (X, T )geom −−−−→ 3

rk0,geom
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 35

commutes, we reduce ourselves to proving the proposition for G = T , a split torus.

To handle this case, we proceed by Noetherian induction on X . By [Th1,
Prop. 4.10], there exist a T -invariant nonempty open subscheme j : U ,→ X , a
closed diagonalizable subgroup T 0 of T , and a T -equivariant isomorphism
U ' T /T 0 × (U/T ).
Since U is nonempty and T acts on X with finite stabilizers, T 0 is finite over k
and K ∗0 (U, T ) ' K ∗0 (U/T ) ⊗ R(T ) R(T 0 ), by Morita equivalence theorem (see [Th3,
Prop. 6.2]). Let i : Z ,→ X be the closed complement of U in X , and let N 0 be an in-
teger satisfying the proposition for both Z and U . Consider the geometric localization
sequence
i∗ j∗
K ∗0 (Z , T )geom −→ K ∗0 (X, T )geom −→ K ∗0 (U, T )geom ,
and let ξ ∈ K ∗0 (X, T )geom . Let a1 , . . . , a2N 0 ∈ K 0 (X, T )geom . By our choice of N 0 ,
j ∗ (a N 0 +1 · . . . · a2N 0 ∪ ξ ) = 0;
thus a N 0 +1 · . . . · a2N 0 ∩ ξ = i ∗ (η) for some η in K ∗0 (Z , T )geom . By the projection
formula, we get
a1 · . . . · a2N 0 ∪ ξ = i ∗ i ∗ (a1 ) · . . . · i ∗ (a N 0 ) ∪ η ,

which is zero by our choice of N 0 and by the fact that rank morphisms commute with
.
pullbacks. Thus, N = 2N 0 satisfies our proposition.

Remark 5.3
By Corollary 5.2, K ∗ (X, G)geom is isomorphic to the localization of K ∗ (X, G)3 at the
multiplicative subset (rk0 )−1 (1), where rk0 : K 0 (X, G)3 → 3 is the rank morphism.
Therefore, if X is regular, K ∗ (X, G)geom depends only on the quotient stack [X/G]
(see [LMB]) and not on its presentation as a quotient.

The main theorem of this paper is the following.

THEOREM 5.4
Let X be a Noetherian regular separated algebraic space over a field k, and let G
be a linear algebraic k-group with a sufficiently rational action on X having finite
stabilizers. Suppose, moreover, that for any essential dual cyclic k-subgroup scheme
σ ⊆ G, the quotient algebraic space G/CG (σ ) is smooth over k (which is the case if,
e.g., G is smooth or abelian). Then C (G) is finite, and the map defined in (5),

w (σ )
K ∗ (X σ , C(σ ))geom ⊗ e
Y
9 X,G : K ∗ (X, G)3 −→ R(σ )3 G ,
σ ∈C (G)

is an isomorphism of R(G)-algebras.
36 VEZZOSI and VISTOLI

Remark 5.5
In Section 5.1 we also give less restrictive hypotheses on G under which Theorem 5.4
still holds.
Note also that if X has the “G-equivariant resolution property” (i.e., any G-
equivariant coherent sheaf is the G-equivariant epimorphic image of a G-equivariant
locally free coherent sheaf), then in Theorem 5.4 one can replace our K ∗ with Quillen
K -theory of G-equivariant locally free coherent sheaves. This happens, for example,
if X is a scheme and G is smooth or finite (see [Th3]).

5.1. Proof of Theorem 5.4

Let us choose, for some n, a closed immersion G ,→ GLn,k and consider the algebraic
space quotient
.
Y = GLn,k ×G X.
We claim that if the theorem holds for Y with the induced GLn,k -action, then it holds
for X with the given G-action. First, let us note that Y is separated. The action map ψ :
G ×(GLn,k ×X ) → (GLn,k ×X )×(GLn,k ×X ) is proper (hence a closed immersion)
since its composition with the separated map
p123 : GLn,k ×X × GLn,k ×X → GLn,k ×X × GLn,k
(here we use that X is separated) is just id X ×a, where a is the action map of G on
GLn,k ; hence it is proper (see [EGAI, Rem. 5.1.7], which obviously carries over to
.
algebraic spaces). Let P = X × GLn,k . In the Cartesian diagram
j
P ×Y P −−−−→ P × P
 
 π×π
y y
Y −−−−→ Y × Y
1Y

P ×Y P ' G × P since π : P → Y is a G-torsor, and π is faithfully flat; therefore,

1Y is a closed immersion; that is, Y is separated.
Note that 3(Y,GLn,k ) = 3.
Consider the morphism defined in (5),

w (σ )
K ∗ (X σ , CG (σ ))geom ⊗ e
Y
9 X,G : K ∗ (X, G)3 −→ R(σ )3 G . (27)
σ ∈C (G)

By Theorem 4.1, the map

9Y,GLn,k : K ∗ (Y, GLn,k )3

wGL (ρ)
K ∗ (Y ρ , CGLn,k (ρ))geom ⊗ e
Y
−→ R(ρ)3 n,k

ρ∈C (GLn,k )
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 37

is an isomorphism, and by the Morita equivalence theorem (see [Th3, Prop. 6.2]),
K ∗ (Y, GLn,k )3 ' K ∗ (X, G)3 . We prove the theorem by constructing an isomor-
phism

w (ρ)
K ∗ (Y ρ , C(ρ))geom ⊗ e
Y
R(ρ) GLn,k
ρ∈C (GLn,k )

wG (σ )
K ∗ (X σ , CG (σ ))geom ⊗ e
Y
−→ R(σ )3 (28)
σ ∈C (G)

commuting with the 9’s and Morita isomorphisms.

Let α : C (G) → C (GLn,k ) be the natural map. If Y ρ 6= ∅, there exists a dual
cyclic subgroup σ ⊆ G, GLn,k -conjugate to ρ (and X σ 6= ∅); therefore, Y ρ = ∅
unless ρ ∈ im(α), and we may restrict the first product in (28) to those ρ in the image
of α and suppose im(α) ⊆ C (G) as well. The following proposition describes the
Y ρ ’s that appear.

PROPOSITION 5.6
Let X be a Noetherian regular separated algebraic space over a field k, and let G
be a linear algebraic k-group with a sufficiently rational action on X having finite
stabilizers. Suppose, moreover, that for any essential dual cyclic k-subgroup scheme
σ ⊆ G, the quotient algebraic space G/CG (σ ) is smooth over k. Let G ,→ GLn,k
be a closed embedding, let ρ ∈ im(α) be an essential dual cyclic subgroup, and let
.
Y = GLn,k ×G X be the algebraic space quotient for the left diagonal action of G. If
CGLn,k ,G (ρ) ⊆ C (G) denotes the fiber α −1 (ρ), then
(i) choosing for each σ ∈ CGLn,k ,G (ρ) an element u ρ,σ ∈ GLn,k (k) such that
u ρ,σ σ u −1
ρ,σ = ρ (in the obvious functor-theoretic sense) determines a unique
isomorphism of algebraic spaces over k,

NGLn,k (σ ) ×NG (σ ) X σ −→ Y ρ ;
a
jρ :
σ ∈CGLn,k ,G (ρ)

(ii) CGLn,k ,G (ρ) is finite.

Proof
Part (ii) follows from (i) since Y ρ is quasi-compact. The proof of (i) requires several
steps.
(a) Definition of jρ . If σ ∈ CGLn,k ,G (ρ), let Nσ be the presheaf on the category
Sch/k of k-schemes which associates to T → Spec k the set
σ
. NGLn,k (σ )(T ) × X (T )
Nσ (T ) = ;
NG (σ )(T )
38 VEZZOSI and VISTOLI

since NG (σ ) acts freely on NGLn,k (σ ) × X σ (on the left), the flat sheaf associated to
Nσ is NGLn,k (σ ) ×NG (σ ) X σ . Let Y
bρ be the presheaf on Sch/k which associates to
T → Spec k the set

ρ .
n GLn,k (T ) × X (T )
Y (T ) = [A, x] ∈
b ∀T 0
G(T )
o
→ T, ∀r ∈ ρ(T ), [r A T 0 , x T 0 ] = [A T 0 , x T 0 ] ;
0

the flat sheaf associated to Ybρ is Y ρ (e.g., see [DG, Chap. II, §1, n. 3]). If u ρ,σ ∈
GLn,k (k) is such that u ρ,σ σ u −1
ρ,σ = ρ (in the obvious functor-theoretic sense), the
presheaf map
bρ ,
jρ,σ : Nσ −→ Y
b
bρ (T )
jρ,σ (T ) : Nσ (T ) 3 [B, x] −→ [u ρ,σ B, x] ∈ Y
b

is easily checked to be well defined. Let jρ,σ : NGLn,k (σ ) ×NG (σ ) X σ → Y ρ denote

. `
the associated sheaf map, and define jρ = σ ∈CGL ,G (ρ) jρ,σ .
n,k
(b) The map jρ induces a bijection on geometric points. This is an elementary
check. Let ξ ∈ Y ρ () be a geometric point. Then there exist an fppf cover T0 →
Spec  and an element [A, x] ∈ Y bρ (T0 ) representing ξ . Therefore, for each T → T0
and each r ∈ ρ(T ) there exists g ∈ G(T ) such that
r A T g −1 = A T ,
gx T = x T .
Then A−1 ρ A defines (functorially over T0 ) a dual cyclic subgroup scheme σ00 of G (T0 )
over T0 . Since σ00 is isomorphic to some µn,T0 , it descends to a dual cyclic subgroup
σ 0 of G over k which is GLn,k -conjugate to ρ since T0 → Spec  has a section
and GLn,k satisfies our rationality condition (RC) (see Rem. 2.2(i)). By definition of
CGLn,k ,G (ρ), there exists a unique σ ∈ CGLn,k ,G (ρ) which is G-conjugated to σ 0 over
k; that is,
gσ 0 g −1 = σ
(functorially) for some g ∈ G(k). Since σ ∈ CGLn,k ,G (ρ), there is an element
u ∈ GLn,k (k) such that uσ u −1 = ρ. Therefore, u −1 Ag −1 restricted to T0 is
in NGLn,k (σ )(T0 ), gx ∈ X σ (T0 ), and if [u −1 Ag −1 , gx]∼ denotes the element in
(NGLn,k (σ ) ×NG (σ ) X σ )() represented by the element [u −1 Ag −1 , gx] in Nσ (T0 ),
we have jρ,σ ()([u −1 Ag −1 , gx]∼ ) = ξ by definition of jρ,σ . Thus, jρ () is surjec-
tive.
Now, let η ∈ (NGLn,k (σ ) ×NG (σ ) X σ )() resp., η0 ∈ (NGLn,k (σ 0 ) ×NG (σ )
0

X σ )() for σ and σ 0 in CGLn,k ,G (ρ). Choosing a common refinement, we can as-
0

sume that there exists an fppf cover T0 → Spec  such that η (resp., η0 ) is repre-
sented by an element [B, y] ∈ Nσ (T0 ) (resp., [B 0 , y 0 ] ∈ Nσ 0 (T0 )). If jρ ()(η) =
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 39

jρ ()(η0 ), there exists an fppf cover T1 → T0 such that [u ρ,σ B, y] = [u ρ,σ 0 B 0 , y 0 ]

in GLn,k (T1 ) × X (T1 )/G(T1 ); that is, there is an element g ∈ G(T1 ) such that

u ρ,σ Bg −1 = u ρ,σ 0 B 0 in GLn,k (T1 ),

gy = y 0 in X (T1 ).

Then it is easy to check that σ = g −1 σ 0 g over T1 and, as in the proof of surjectivity

of jρ (), since T1 → Spec  has a section and G satisfies our rationality condition
(RC), σ and σ 0 are G-conjugated over k as well, and therefore σ = σ 0 as elements
in CGLn,k ,G (ρ). In particular, g ∈ NG (σ )(T1 ) and [B, y] = [B 0 , y 0 ] in Nσ (T1 ). Since
T1 → Spec  is still an fppf cover, we have η = η0 and jρ () is injective.
(c) Each jρ,σ is a closed and open immersion. It is enough to show that each jρ,σ
is an open immersion because in this case it is also a closed immersion, Y ρ being
quasi-compact. Since NGLn,k (ρ) acts on both σ ∈CGL ,G (ρ) NGLn,k (σ ) ×NG (σ ) X σ
`
n,k
and Y ρ and since jρ is equivariant, it will be enough to prove that jρ,ρ is an open
immersion. We prove first that jρ,ρ is injective and unramified and then conclude
the proof by showing that it is also flat (in fact, an étale injective map is an open
immersion).
(c1 ) The map jρ,ρ is injective and unramified. It is enough to show that the inverse
image under jρ,ρ of a geometric point is a (geometric) point. Consider the commuta-
tive diagram
l
NGLn,k (ρ) × X ρ −−−−→ GLn,k ×X
 
py
 π
y
NGLn,k (ρ) ×NG (ρ) X ρ −−−−→ Y
i ρ ◦ jρ,ρ

where l and i ρ : Y ρ ,→ Y are the natural inclusions and where p, π are the natural
projections. Let y0 be a geometric point of Y in the image of i ρ ◦ jρ,ρ ; using the action
of NGLn,k (ρ) on NGLn,k (ρ) ×NG (ρ) X ρ and Y ρ , we may suppose that y0 is of the form
[1, x0 ] ∈ Y ρ (), with  an algebraically closed field and x0 ∈ X ρ (). Obviously,
(1, x0 ) ∈ NGLn,k (ρ) ×NG (ρ) X ρ () is contained in jρ,ρ
−1 (y ), and, by faithful flatness
0
of p, jρ,ρ (y0 ) = (1, x0 ) if
−1

p −1 (1, x0 ) = π −1 (y0 ) ∩ NGLn,k (ρ) ×NG (ρ) X ρ .

(29)

But G() ' π −1 (y0 ) via g 7 → (g −1 , gx0 ) and NG (ρ)() ' p −1 ((1, x0 )) via h 7→
(h −1 , hx0 ); therefore, (29) follows from NG (ρ) = NGLn,k (ρ) ∩ G.
(c2 ) The map jρ,ρ is flat. This fact is proved in Section 5.2, where we also sin-
gle out a more general technical hypothesis for the action of G on X under which
Proposition 5.6 still holds.
40 VEZZOSI and VISTOLI

The remaining part of this subsection is devoted to the conclusion of the proof of
Theorem 5.4 using Proposition 5.6. First we show that Proposition 5.6(ii) allows one
to define a canonical isomorphism

w (ρ)
K ∗ (Y ρ , CGLn,k (ρ))geom ⊗ e
Y
R(ρ)3 GLn,k
ρ∈C (GLn,k )

wGL (σ )
K ∗ (NGLn,k (σ ) ×NG (σ ) X σ , CGLn,k (σ ))geom ⊗ e
Y
' R(σ )3 n,k ;
σ ∈C (G)

next we show, using Lemma 2.10, that each factor in the right-hand side is isomorphic
to

w (σ )
K ∗ (CGLn,k (σ ) ×CG (σ ) X σ , CGLn,k (σ ))geom ⊗ e
R(σ )3 G .
The conclusion (i.e., the isomorphism (28)) is then accomplished by establishing,
for any regular Noetherian separated algebraic space Z on which G acts with finite
stabilizers, a “geometric” Morita equivalence

K ∗ (GLn,k ×G Z , GLn,k )geom ' K ∗ (Z , G)geom .

First, note that the choice of a family {u ρ,σ | σ ∈ CGLn,k ,G (ρ)} of elements
u ρ,σ ∈ GLn,k (k) such that u ρ,σ σ u −1 ρ,σ = ρ, which uniquely defines jρ in Proposi-
tion 5.6, also determines a unique family of isomorphisms

int(u ρ,σ ) : CGLn,k (ρ) → CGLn,k (σ ) σ ∈ CGLn,k ,G (ρ)



(where int(u ρ,σ ) denotes conjugation by u ρ,σ ), and this family gives us an action of
CGLn,k (ρ) on
NGLn,k (σ ) ×NG (σ ) X σ
a

σ ∈CGLn,k ,G (ρ)

(since NGLn,k (σ ), and then CGLn,k (σ ), acts naturally on NGLn,k (σ ) ×NG (σ ) X σ by left
multiplication on NGLn,k (σ )). With this action, jρ becomes a CGLn,k (ρ)-equivariant
isomorphism, and since int(u ρ,σ ) induces an isomorphism R(CGLn,k (ρ)) '
R(CGLn,k (σ )) commuting with rank morphisms, jσ induces an isomorphism

K ∗ Y ρ , CGLn,k (ρ) geom ⊗ e

R(ρ)3
K ∗ NGLn,k (σ ) ×NG (σ ) X σ , CGLn,k (σ ) geom ⊗ e
Y
R(σ )3


'
σ ∈CGLn,k ,G (ρ)

which, by definition of the action of NGLn,k (ρ) on each NGLn,k (σ )×NG (σ ) X σ , induces
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 41

an isomorphism

w (ρ)
K ∗ (Y ρ , CGLn,k (ρ))geom ⊗ eR(ρ)3 GLn,k

w (σ )
K ∗ (NGLn,k (σ ) ×NG (σ ) X σ , CGLn,k (σ ))geom ⊗ e
Y
' R(σ )3 GLn,k .
σ ∈CGLn,k ,G (ρ)

(30)
Now, if jρ0 is induced, as in Proposition 5.6, by another choice of a family {vρ,σ |
σ ∈ CGLn,k ,G (ρ)} of elements vρ,σ ∈ GLn,k (k) such that vρ,σ σ vρ,σ−1 = ρ, then

vρ,σ
−1 u
ρ,σ ∈ NGLn,k (σ )(k) and there is a commutative diagram

(vρ,σ
−1 u
ρ,σ )·
NGLn,k (σ ) ×NG (σ ) Xσ / NGLn,k (σ ) ×NG (σ ) X σ
PPP n
PPP
PPP n nnnn
n
jρ PPPP nnnnn jρ0
( v n
Yρ
Therefore, isomorphism (30) on the invariants is actually independent of the choice
of the family {u ρ,σ | σ ∈ CGLn,k ,G (ρ)}. Since CGLn,k ,G (ρ) = α −1 (ρ) and, as already
observed, Y ρ = ∅ unless ρ ∈ im(α), this gives us a canonical isomorphism

w (ρ)
K ∗ (Y ρ , CGLn,k (ρ))geom ⊗ e
Y
R(ρ)3 GLn,k
ρ∈C (GLn,k )

wGL (σ )
K ∗ (NGLn,k (σ ) ×NG (σ ) X σ , CGLn,k (σ ))geom ⊗ e
Y
' R(σ )3 n,k .
σ ∈C (G)

Now, let us fix σ ∈ C (G), and let us choose a set A ⊂ NGLn,k (σ )(k) such that
the classes in wGLn,k (σ ) of the elements in A constitute a set of representatives for
the wG (σ )-orbits in wGLn,k (σ ); A is a finite set. Since
CGLn,k (σ ) ×CG (σ ) X σ ,→ NGLn,k (σ ) ×NG (σ ) X σ
is an open and closed immersion, the morphism
CGLn,k (σ ) ×CG (σ ) X σ −→ NGLn,k (σ ) ×NG (σ ) X σ ,
a

A
[C, x] Ai ∈A 7−→ [Ai · C, x]
(in the obvious functor-theoretic sense), which is easily checked to induce an iso-
morphism on geometric points, is an isomorphism. Therefore, there is an induced
isomorphism
K ∗ CGLn,k (σ ) ×CG (σ ) X σ , CGLn,k (σ ) geom ⊗ e
Y
R(σ )3

A
' K ∗ NGLn,k (σ ) ×NG (σ ) X σ , CGLn,k (σ ) geom ⊗ e
R(σ )3 .


42 VEZZOSI and VISTOLI

Since wGLn,k (σ ) acts transitively on A with stabilizer wG (σ ), by Lemma 2.10 we get

a canonical isomorphism

w (σ )
K ∗ (NGLn,k (σ ) ×NG (σ ) X σ , CGLn,k (σ ))geom ⊗ e
R(σ )3 GLn,k

w (σ )
' K ∗ (CGLn,k (σ ) ×CG (σ ) X σ , CGLn,k (σ ))geom ⊗ e
R(σ )3 G .

Since, by Morita equivalence (see [Th3, Prop. 6.2]),

K ∗ CGLn,k (σ ) ×CG (σ ) X σ , CGLn,k (σ ) ' K ∗ X σ , CG (σ ) ,

(31)

to conclude the proof of Theorem 5.4 we need only show that the natural morphism

K ∗ CGLn,k (σ ) ×CG (σ ) X σ , CGLn,k (σ ) geom ' K ∗ X σ , CG (σ ) geom

(32)

induced by (31) is still an isomorphism. Since the diagram

α / K ∗ CGLn,k (σ ) ×CG (σ ) X σ , CGLn,k (σ )
K ∗ GLn,k ×CG (σ ) X σ , GLn,k geom



geom
WWWWW
WWWWW
WWWWW β
γ WWWWW
W+ 
K ∗ X σ , CG (σ ) geom

is commutative and, by Morita equivalence,

K ∗ CGLn,k (σ ) ×CG (σ ) X σ , CGLn,k (σ )

CGLn,k (σ )
CGLn,k (σ ) ×CG (σ ) X σ , GLn,k


' K ∗ GLn,k ×
(σ )
GLn,k × GLn,k X σ , GLn,k ,
C

' K∗

to show that β is an isomorphism it is enough to prove that for any regular separated
algebraic space Z on which G acts with finite stabilizers, Morita equivalence induces
an isomorphism

K ∗ (GLn,k ×G Z , GLn,k )geom ' K ∗ (Z , G)geom (33)

since in this case both α and γ are isomorphisms.

Let π : R(GLn,k ) → R(G) be the restriction morphism, let ρ : R(G) →
K 0 (Z , G) be the pullback along Z → Spec k, let rk0 : R(GLn,k ) → 3 and
. .
rk : R(G) → 3 be the rank morphisms, and let S 0 = (rk0 )−1 (1), S = (rk)−1 (1),
.
and T = π(S 0 ) ⊆ S; the following diagram commutes:
rk0,T
/
T −1 K 0 (Z , G)3
r r9 3
rr
rrr
rr rkgeom
 rr
K 0 (Z , G)geom
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 43

where rkgeom and rk0,T denote the localizations of the rank morphism rk0 :
K 0 (Z , G)3 → 3. By Morita equivalence, the natural map (which commutes with
the induced rank morphisms)

K 0 (GLn,k ×G Z , GLn,k )geom −→ T −1 K 0 (Z , G)3

is an isomorphism, and then, by Proposition 5.1, ker(rk0,T : T −1 K 0 (Z , G)3 → 3)

is nilpotent. Now, if s ∈ S, then rk0,T (ρ(s)/1) = rk(s) = 1, and therefore

T −1 K 0 (Z , G)3 → K 0 (Z , G)geom

and
K 0 (GLn,k ×G Z , GLn,k )geom → K 0 (Z , G)geom
are both isomorphisms. Since K ∗ (GLn,k ×G Z , GLn,k )3 is naturally a K 0 (GLn,k
×G Z , GLn,k )3 -module and an R(GLn,k )3 -module via the pullback ring morphism

ρ 0 : R(GLn,k )3 → K 0 (GLn,k ×G Z , GLn,k )3 ,

we have

K ∗0 (GLn,k ×G Z , GLn,k )geom

' K ∗0 (GLn,k ×G Z , GLn,k )3 ⊗ K 0 (GLn,k ×G Z ,GLn,k )3 K 0 (GLn,k ×G Z , GLn,k )geom
' K ∗0 (Z , G)3 ⊗ K 0 (Z ,G)3 K 0 (GLn,k ×G Z , GLn,k )geom
' K ∗0 (Z , G)3 ⊗ K 0 (Z ,G)3 K 0 (Z , G)geom
' K ∗0 (Z , G)geom ,

which proves (32), and we conclude the proof of Theorem 5.4.

5.2. Hypotheses on G
In this subsection we conclude the proof of Proposition 5.6, showing that (this is part
(c2 ) of the proof)
jρ,ρ : NGLn,k (ρ) ×NG (ρ) X ρ −→ Y ρ
is flat. This is the only step in the proof of Proposition 5.6 where we make use of the
hypothesis that the quotient algebraic space G/CG (ρ) is smooth over k. Actually, our
proof works under the following weaker hypothesis. Let S denote the spectrum of the
dual numbers over k,
S = Spec(k[ε]),
1
and for any k-group scheme H , let H (S, H ) denote the k-vector space of isomor-
phism classes of pairs (P → S, y), where P → S is an H -torsor and y is a k-rational
point on the closed fiber of P. Then Proposition 5.6, and hence Theorem 5.4, still
44 VEZZOSI and VISTOLI

holds with hypothesis

(S) for any essential dual cyclic subgroup scheme σ ⊆ G, the quotient G/CG (σ )
is smooth over k
replaced by the following:
(S0 ) for any essential dual cyclic k-subgroup scheme σ ⊆ G, we have
1 1
σ
dim H S, CG (σ ) = dim H (S, G) .

First we prove that jρ,ρ is flat assuming (S0 ) holds. Then we show that (S) implies
(S0 ). Since p : NGLn,k (ρ) × X ρ → NGLn,k (ρ) ×NG (ρ) X ρ is faithfully flat, it is enough
.
to prove that j ρ = jρ,ρ ◦ p is flat. Let π : GLn,k ×X → Y be the projection, and let

f : GLn,k ×X × G −→ GLn,k ×X,

(A, x, g) 7−→ (Ag −1 , gx).

Consider the following Cartesian squares:

/ π −1 (Y ρ )  
uρ
U / GLn,k ×X (34)
π
  
NGLn,k (ρ) × X ρ / Yρ  /Y
jρ

Since π is faithfully flat, it is enough to prove that u ρ is flat. But the squares

 / GLn,k ×X × G f
/ GLn,k ×X
U (35)
pr12 π
   
NGLn,k (ρ) × X ρ  / GLn,k ×X
π
/Y

are Cartesian and (in the obvious functor-theoretic sense)

U = (A, x, g) ∈ GLn,k ×X × G A−1 ρ A = ρ, x ∈ X ρ ' NGLn,k (ρ) × X ρ × G.



.
Moreover, if P = {A ∈ GLn,k | A−1 ρ A ⊆ G}, the map

π −1 (Y ρ ) = (A, x) ∈ NGLn,k (ρ) × X A−1 ρ A ⊆ G, x ∈ X A ρ A −→ P × X ρ ,

 −1

(A, x) 7−→ (A, Ax)

is an isomorphism. Therefore, we are reduced to proving that the map

vρ : NGLn,k (ρ) × X ρ × G −→ P × X ρ ,
(A, x, g) 7−→ (Ag −1 , Ax)
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 45

is flat. But since the diagram

vρ
NGLn,k (ρ) × X ρ × G −−−−→ P × X ρ
 
pr13 
y
pr
y 1
NGLn,k (ρ) × G −−−−→ P
2ρ
.
(where 2ρ (A, g) = (Ag −1 )) is easily checked to be Cartesian, it is enough to show
that zρ is flat. To do this, let us observe that ρ acts by conjugation on GLn,k /G
(quotient by the G-action on GLn,k by right multiplication), and we have a Cartesian
diagram
 / GLn,k
P
τ
  
(GLn,k /G)ρ  / GLn,k /G

Then τ is a G-torsor and 2ρ is G-equivariant. Thus, the following commutative dia-

gram, in which the vertical arrows are G-torsors,
2ρ
NGLn,k (ρ) × G −−−−→ P
 
pr1  τ
y y
NGLn,k (ρ) −−−−→ (GLn,k /G)ρ
χρ
.
(where χρ (A) = [A] ∈ GLn,k /G) is Cartesian, and we may reduce ourselves to prove
that χρ is flat. Now, observe that NGLn,k (ρ) acts on the left of both NGLn,k (ρ) and
(GLn,k /G)ρ in such a way that χρ is NGLn,k (ρ)-equivariant. Therefore, it is enough
to prove that χρ is flat when restricted to the connected component of the identity in
NGLn,k (ρ), that is, that the map

χρ0 : CGLn,k (ρ) −→ (GLn,k /G)ρ

is flat. Now, CGLn,k (ρ) = (GLn,k )ρ , where ρ acts by conjugation; both (GLn,k )ρ and
(GLn,k /G)ρ are smooth by [Th5, Prop. 3.1] (since GLn,k and GLn,k /G are smooth);
each fiber of χρ0 has dimension equal to dim(CG (ρ)) because χρ0 is CGLn,k (ρ)-
equivariant for the natural actions; and all the fibers are obtained from (χρ0 )−1 ([1]) =
CG (ρ) by the CGLn,k (ρ)-action. Therefore, χρ0 is flat if

dim CGLn,k (ρ) = dim CG (ρ) + dim (GLn,k /G)ρ .

(36)

Note that, in any case,

dim CGLn,k (ρ) ≤ dim CG (ρ) + dim (GLn,k /G)ρ .

(37)
46 VEZZOSI and VISTOLI

Since GLn,k is smooth, dim((GLn,k /G)ρ ) = dimk (T1 (GLn,k /G)ρ ), where T1 de-
1
notes the tangent space at the class of 1 ∈ GLn,k . Moreover, since H (S, GLn,k ) = 0,
there is an exact sequence of k-vector spaces
1
0 → Lie(G) −→ Lie(GLn,k ) −→ T1 (GLn,k /G) −→ H (S, G) → 0 (38)

which, ρ being linearly reductive over k, yields an exact sequence of ρ-invariants

0 → Lie(G)ρ −→ Lie(GLn,k )ρ −→ T1 (GLn,k /G)ρ −→ H (S, G)ρ → 0. (39)

1

But GLn,k is smooth, so

dimk Lie(GLn,k )ρ = dim(GLn,k )ρ = dim CGLn,k (ρ) ,

and, since Lie(CG (ρ)) = Lie(G) ∩ Lie(CGLn,k (ρ)), we get

dimk Lie(G)ρ = dimk Lie(CG (ρ)) .

By (39), we get

dimk H (S, G)ρ = dimk T1 (GLn,k /G)ρ − dim CGLn,k (ρ)

1

+ dimk Lie(CG (ρ))

= dim (GLn,k /G)ρ − dim CGLn,k (ρ) + dim CG (ρ)

+ dimk Lie(CG (ρ)) − dim CG (ρ) ;

(40)

hence (36) is satisfied if

dimk H (S, G)ρ = dimk Lie(CG (ρ)) − dim CG (ρ) .

1



(41)

But
1
dimk H (S, CG (ρ)) = dimk Lie(CG (ρ)) − dim CG (ρ)

by the exact sequence (analogous to (38) with G replaced by CG (ρ))

0 → Lie CG (ρ) −→ Lie(GLn,k ) −→ T1 GLn,k /CG (ρ)

1
−→ H S, CG (ρ) → 0;

hence (41) holds by hypothesis (S0 ). We complete the proof of Proposition 5.6 by
showing that (S) implies (S0 ). Since CG (ρ) ⊆ G, we have a natural map

 : H S, CG (ρ) −→ H (S, G)ρ ,

1
1

and by (40) and (37), we get

dimk H (S, G)ρ ≥ dimk H (S, CG (ρ)) .

1
1

(42)
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 47

Now, if (S) holds, that is, if G/CG (ρ) is smooth, and if [P → S, y] is a class in
1
H (S, G)ρ , P/CG (ρ) → S is smooth and y induces a point in the closed fiber of
P/CG (ρ) → S; we may reduce the structure group to CG (ρ), thus showing that  is
surjective. By (42), we conclude that  is an isomorphism, and this implies (S0 ).

5.3. Final remarks

PROPOSITION 5.7
Let X be a Noetherian regular separated algebraic space over k, and let G be a finite
group acting on X . There is a canonical isomorphism of R(G)-algebras

K ∗ (X, G)geom ⊗ Z 1/|G| ' K ∗ (X )G ⊗ Z 1/|G| .

   

Proof
Since ker rk : K 0 (X ) → H0 (X, Z[1/|G|]) is nilpotent by Corollary 5.2, the canon-

ical homomorphism
π ∗ : K ∗ (X, G) −→ K ∗ (X )G
induces a ring homomorphism (still denoted by π ∗ )

π ∗ : K ∗ (X, G)geom ⊗ Z 1/|G| −→ K ∗ (X )G ⊗ Z 1/|G| .

   

Moreover, the functor M

π∗ : F 7 −→ g∗F ,
g∈G

defined on coherent O X -modules, induces a homomorphism

π∗ : K ∗0 (X )G ⊗ Z[1/|G|] −→ K ∗0 (X, G)geom ⊗ Z[1/|G|],

and (recalling that K ∗ (X, G) ' K ∗0 (X, G)) we obviously get

π ∗ π∗ (F ) = |G| · F .

On the other hand, we have

π∗ π ∗ (F ) ' F ⊗ π∗ O X .

But rk(π∗ O X ) = |G|, and therefore π∗ π ∗ is an isomorphism, too, because of Corol-

lary 5.2.

As a corollary of this result and of Theorem 5.4, we recover [Vi1, Th. 1], which was
proved there in a completely different way.
48 VEZZOSI and VISTOLI

We conclude the paper with a conjecture expressing the fact that K ∗ (X, G)geom
should be the K -theory of the quotient X/G, if X/G is regular, after inverting the
orders of all the essential dual cyclic subgroups of G.

CONJECTURE 5.8
Let X be a Noetherian regular separated algebraic space over a field k, and let G
be a linear algebraic k-group acting on X with finite stabilizers in such a way that
the quotient X/G exists as a regular algebraic space. Let N denote the least common
multiple of the orders of all the essential dual cyclic subgroups of G, and let 3 =
Z[1/N ]. If p : X → X/G is the quotient map, the composition
p∗
K ∗ (X/G)3 −→ K ∗ (X, G)3 −→ K ∗ (X, G)geom

is an isomorphism.

Remark 5.9
Bertrand Toen pointed out to us that if X/G is smooth, it follows from the results of
[EG] that the composition

K 0 (X/G) ⊗ Q −→ K 0 (X, G) ⊗ Q −→ K 0 (X, G)geom ⊗ Q

is an isomorphism.

Appendix. Higher equivariant K -theory of Noetherian regular separated alge-

braic spaces
In this appendix we describe the K -theories we use in the paper and their relation-
ships. We essentially follow the example of [ThTr, Sec. 3]. We also adopt the language
of [ThTr].
Let us remark that it is strongly probable that there exist equivariant versions of
most of the results in [ThTr, Sec. 3]. In particular, there should exist a higher K -
theory of G-equivariant cohomologically bounded pseudocoherent complexes on Z
(resp., of G-equivariant perfect complexes on Z ) for any quasi-compact algebraic
space Z having most of the alternative models described in [ThTr, Pars. 3.5 – 3.12].
The arguments below can also be considered as a first step toward an extension of
[ThTr, Lems. 3.11, 3.12] to the equivariant case on algebraic spaces. However, to
keep the paper to a reasonable size, we have decided to give only the results we need,
and, moreover, we have made almost no attempt to optimize the hypotheses.
We would also like to mention the paper [J] (in particular, Section 1) in which,
among many other results, the general techniques of [ThTr] are used as guidelines for
the K -theory of arbitrary Artin stacks.
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 49

We work in a slightly more general situation than required in the rest of the paper.
Let S be a separated Noetherian scheme, and let G be a group scheme affine over
S which is finitely presented, separated, and flat over S. We denote by G-AlgSpreg the
category of regular Noetherian algebraic spaces separated over S with an action of G
over S and equivariant maps.

Definition A.1
If X ∈ G-AlgSpreg , we denote by K ∗ (X, G) (resp., K ∗0 (X, G), resp., K ∗naive (X, G))
the Waldhausen K -theory of the complicial bi-Waldhausen category (see [ThTr])
W1,X of complexes of quasi-coherent G-equivariant O X -modules with bounded co-
herent cohomology (resp., the Quillen K -theory of the abelian category of G-
equivariant coherent O X -modules, resp., the Quillen K -theory of the exact category
of G-equivariant locally free coherent O X -modules).

PROPOSITION A.2
Let Z → S be a morphism of Noetherian algebraic spaces such that the diagonal
Z → Z × S Z is affine, and let H → S be an affine group space acting on Z . Let
F be an equivariant quasi-coherent sheaf on Z of finite flat dimension; then there
exists a flat equivariant quasi-coherent sheaf F 0 on Z together with a surjective H -
equivariant homomorphism F 0 → F .
In particular, if Z is regular, this holds for all equivariant quasi-coherent sheaves
F on Z .

The hypotheses of Proposition A.2 ensure that the usual morphism Z × S H → Z × S Z

is affine. In fact, the projection Z × S H → Z is obviously affine, the projection
Z × S Z → Z has affine diagonal, so this follows from the elementary fact that if
Z → U → V are morphisms of algebraic spaces, Z → V is affine, and U → V has
affine diagonal, then Z → U is affine. Consider the quotient stack Z = [Z /H ] (see
[LMB]); the argument above implies that the diagonal Z → Z × S Z is affine. Since
an H -equivariant quasi-coherent O Z -module is the same as a quasi-coherent module
over Z , now Proposition 5.3 follows from the more general result below.

PROPOSITION A.3
Let S be a Noetherian algebraic space, and let X be a Noetherian algebraic stack
over S with affine diagonal. Let F be a quasi-coherent sheaf of finite flat dimension on
X ; then there exists a flat quasi-coherent sheaf F 0 on X together with a surjective
homomorphism F 0 → F .
50 VEZZOSI and VISTOLI

Proof
Take an affine scheme U with a flat morphism f : U → X ; then f is affine, and, in
particular, the pushforward f ∗ on quasi-coherent sheaves is exact. Consider a quasi-
coherent sheaf F on X of finite flat dimension, with the adjunction map F →
f ∗ f ∗ F . This map is injective; call Q its cokernel. Clearly, the flat dimension of
f ∗ f ∗ F is the same as the flat dimension of F ; we claim that the flat dimension of Q
is at most equal to the flat dimension of F . Now, if there were a section X → U of
f , then the sequence
0 → F → f∗ f ∗F → Q → 0
would split and this would be clear. However, to compute the flat dimension of Q , we
can pull back to any flat surjective map to X ; in particular, after pulling back to U ,
we see that f acquires a section, and the statement is checked. Now U is an affine
scheme, so we can take a flat quasi-coherent sheaf P on U with a surjective map
u : P → f ∗ F . Call F 0 the kernel of the composition f ∗ P → f ∗ f ∗ F → Q ; then
F 0 surjects onto F , and it fits into an exact sequence 0 → F 0 → f ∗ P → Q → 0.
But f ∗ P is flat over X , so the flat dimension of F 0 is less than the flat dimension
of Q , unless Q is flat. But since the flat dimension of Q is at most equal to the flat
dimension of F , we see that the flat dimension of F 0 is less than the flat dimension
of F , unless F is flat. The proof is completed with a straightforward induction on
the flat dimension of F .

THEOREM A.4
Let X be an object in G-AlgSpreg . The obvious inclusions of the following compli-
cial biWaldhausen categories induce homotopy equivalences on the Waldhausen K -
.
theory spectra K (i) (X ) = K (Wi,X ), i = 1, 2, 3. In particular, the corresponding
(i)
Waldhausen K -theories K ∗ (X, G) coincide.
(i) W1,X = (complexes of quasi-coherent G-equivariant O X -modules with
bounded coherent cohomology).
(ii) W2,X = (bounded complexes in G − Coh X ).
(iii) W3,X = (complexes of flat quasi-coherent G-equivariant O X -modules with
bounded coherent cohomology).
Moreover, the Waldhausen K -theory of any of the categories above coincides
with Quillen K -theory K ∗0 (X, G) of G-equivariant coherent O X -modules.

Proof
By [Th2, Par. 1.13], the inclusion of W2,X in W1,X induces an equivalence of K -theory
spectra. Proposition 5.3, together with [ThTr, Lem. 1.9.5] (applied to D = (flat G-
equivariant O X -modules) and A = (G-equivariant O X -modules)), implies that for
any object E ∗ in W1,X there exist an object F ∗ in W3,X and a quasi-isomorphism
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 51

g E ∗ . Therefore, by [ThTr, Par. 1.9.7 and Th. 1.9.8], the inclusion of W3,X in
F ∗ −→
W1,X induces an equivalence of K -theory spectra.
The last statement of the theorem follows immediately from [Th2, Par. 1.13, p.
518].

Since any complex in W3,X is degreewise flat and X is regular (hence boundedness of
cohomology is preserved under tensor product∗ ), the tensor product of complexes
makes the Waldhausen K -theory spectrum of W3,X into a functor K (3) from G-
AlgSpreg to ring spectra, with product

K (3) ∧ K (3) −→ K (3) ,

exactly as described in [ThTr, Par. 3.15]. In particular, by Theorem 5.3, K ∗ is a functor

from G-AlgSpreg to graded rings. In the same way, the tensor product with complexes
in W3,X gives a pairing
K (3) ∧ K (1) −→ K (1)
(1)
between the corresponding functors from G-AlgSpreg to spectra, so that K ∗ (X, G)
(3)
becomes a module over the ring K ∗ (X, G) functorially in (X, G) ∈ G-AlgSpreg .
We denote the corresponding cap-product by

∩ : K ∗(3) (X, G) ⊗ K ∗(1) (X, G) −→ K ∗(1) (X, G),

which becomes the ring product in K ∗ (X, G) with the identifications allowed by
Theorem 5.3. Note that there is an obvious ring morphism η : K ∗naive (X, G) →
(3)
K ∗ (X, G), and if
0
∩naive : K ∗naive (X, G) ⊗ K ∗ (X, G) −→ K ∗0 (X, G)

denotes the usual “naive” cap-product on Quillen K -theories, there is a commutative

diagram
∩naive
K ∗naive (X, G) ⊗ K ∗0 (X, G) −−−−→ K ∗0 (X, G)
 
η⊗u y
 u
y
(3) (1) (1)
K ∗ (X, G) ⊗ K ∗ (X, G) −−−−→ K ∗ (X, G)
∩
where u is the isomorphism of Theorem 5.3. Because of that, we simply write ∩ for
both the naive and nonnaive cap-products. Note that, as shown in [Th2, Par. 1.13, p.
519], K ∗0 (−, G) (and therefore K ∗ (X, G) under our hypotheses) is a covariant functor
∗ In
fact, this is a nonequivariant statement and a local property in the flat topology, so it reduces to the
same statement for regular affine schemes, which is elementary (see also [SGA6]).
52 VEZZOSI and VISTOLI

for proper maps in G-AlgSpreg ; on the other hand, since any map in G-AlgSpreg
has finite Tor-dimension, K ∗ (−, G) is a controvariant functor from G-AlgSpreg to
(graded) rings. In fact, if f : X −→ Y is a morphism in G-AlgSpreg , the same
argument in [ThTr, Par. 3.14.1] shows that there is an induced pullback exact functor
(3)
f ∗ : W3,Y → W3,X , and then we use Theorem A.4 to identify K ∗ (−, G) with
K ∗ (−, G).

PROPOSITION A.5 (Projection formula)

Let j : Z −→ X be a closed immersion in G-AlgSpreg . Then, if α is in K ∗ (X, G) and
β in K ∗0 (Z , G), we have

j∗ j ∗ (α) ∩ β = α ∩ j∗ (β)

in K ∗0 (X, G).

Proof
Since j is affine, j∗ is exact on quasi-coherent modules and therefore induces an exact
functor of complicial bi-Waldhausen categories j∗ : W1,Z → W1,X (the condition of
bounded coherent cohomology being preserved by regularity of Z and X ). Therefore,
the maps

(α, β) 7 −→ j∗ j ∗ (α) ∩ β ,

(α, β) 7−→ α ∩ j∗ (β)

from K ∗ (X, G) × K ∗ (Z , G) to K ∗0 (X, G) ' K ∗ (X, G) are induced by the exact

functors W3,X × W1,Z −→ W1,X ,

(F ∗ , E ∗ ) 7 −→ j∗ j ∗ (F ∗ ) ⊗ E ∗ ,

(F ∗ , E ∗ ) 7 −→ F ∗ ⊗ j∗ (E ∗ ). (43)

But for any equivariant quasi-coherent sheaf F on X and G on Z , there is a natural

(hence, equivariant) isomorphism

j∗ ( j ∗ F ⊗ G ) ' F ⊗ j∗ G

which, again by naturality, induces an isomorphism between the two functors in (43);
therefore, we conclude by [ThTr, Par. 1.5.4].

Remark A.6
Since we need the projection formula only for (regular) closed immersion in this pa-
per, we have decided to state the result only in this case. However, since, by [Th2,
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 53

Par. 1.13 p. 519], K ∗ (X, G) coincides also with Waldhausen K -theory of the cat-
egory W4,X of complexes of G-equivariant quasi-coherent injective modules on X
with bounded coherent cohomology, therefore, by Theorem 5.3, it also coincides
with Waldhausen K -theory of the category W5,X complexes of G-equivariant quasi-
coherent flasque modules on X with bounded coherent cohomology. For any proper
map f : X → Y in G-AlgSpreg , we have an exact functor f ∗ : W5,X → W5,Y , which
therefore gives a “model” for the pushforward f ∗ : K ∗ (X, G) → K ∗ (Y, G) (cf.
[ThTr, Par. 3.16]). Now, the proof of [ThTr, Prop. 3.17] should also give a proof of
Proposition 3 with j replaced by any proper map in G-AlgSpreg because it only uses
[ThTr, Th. 2.5.5], which obviously holds for X and Y Noetherian algebraic spaces,
and [SGA4, Exp. XVII, par. 4.2], which should give a canonical G-equivariant Gode-
ment flasque resolution of any complex of G-equivariant modules on any algebraic
space in G-AlgSpreg since it is developed in a general topos.

It is very probable that Theorem 5.3 and therefore the functoriality with respect to
morphisms of finite Tor-dimension still hold without the regularity assumption on the
algebraic spaces. On the other hand, it should also be true that with G and X as above
(therefore, X regular), the Waldhausen K -theory of the category of G-equivariant
perfect complexes on X coincides with K ∗0 (X, G). This last statement should fol-
low (with a bit of work to identify K ∗0 (X, G) with the Waldhausen K -theory of G-
equivariant pseudocoherent complexes with bounded cohomology on X ) from [J, Th.
1.6.2].

Acknowledgments. We wish to thank the referee for useful and precise remarks. We
also thank Bertrand Toen, who pointed out the content of Remark 5.9 to us.

References

[SGA4] M. ARTIN, A. GROTHENDIECK, and J. L. VERDIER, Théorie des topos et cohomologie

étale des schémas, Vol. 3, Séminaire de Géométrie Algébrique du Bois-Marie
(SGA 4), Lecture Notes in Math. 305, Springer, Berlin, 1973. MR 50:7132 53
[SGA6] P. BERTHELOT, A. GROTHENDIECK, and L. ILLUSIE, Théorie des intersections et
théorème de Riemann-Roch, Séminaire de Géométrie Algébrique du Bois-Marie
(SGA 6), Lecture Notes in Math. 225, Springer, Berlin, 1971. MR 50:7133 2, 23,
51
[DG] M. DEMAZURE and P. GABRIEL, Groupes algébriques, Vol. 1: Géométrie algébrique,
généralités, groupes commutatifs, Masson, Paris, 1970. MR 46:1800 38
[SGA3] M. DEMAZURE and A. GROTHENDIECK, Schémas en groupes, Séminaire de
Géométrie Algébrique du Bois-Marie (SGA 3), Lecture Notes in Math. 151, 152,
153, Springer, Berlin, 1970. MR 43:223a, MR 43:223b, MR 43:223c 12, 16, 28
[EG] D. EDIDIN and W. GRAHAM, Riemann-Roch for equivariant Chow groups, Duke
54 VEZZOSI and VISTOLI

Math. J. 102 (2000), 567 – 594. MR 2001f:14018 2, 4, 48

[SGA1] A. GROTHENDIECK, Revêtements étales et groupe fondamental, Séminaire de
Géométrie algébrique du Bois-Marie (SGA 1), Lecture Notes in Math. 224,
Springer, Berlin, 1971. 27, 30
[EGAI] A. GROTHENDIECK and J. A. DIEUDONNÉ, Eléments de géométrie algébrique, I: Le
langage des schémas, Springer, Berlin, 1971. 36
[J] R. JOSHUA, Higher intersection theory on algebraic stacks, II, preprint, 2000,
[Link] 48, 53
[LMB] G. LAUMON and L. MORET-BAILLY, Champs algébriques, Ergeb. Math. Grenzgeb. (3)
39, Springer, Berlin, 2000. MR 2001f:14006 35, 49
[Me] A. S. MERKURJEV, Comparison of the equivariant and the standard K -theory of
algebraic varieties, St. Petersburg Math. J. 9 (1998), 815 – 850. MR 99d:19003
15, 16
[Q] D. QUILLEN, “Higher algebraic K -theory, I” in Algebraic K -Theory, I: Higher
K -Theories (Seattle, 1972), Lecture Notes in Math. 341, Springer, Berlin, 1973.
MR 49:2895
[Se] J.-P. SERRE, Linear representations of finite groups, Grad. Texts in Math. 42, Springer,
New York, 1977. MR 56:8675 18
[Sr] V. SRINIVAS, Algebraic K -Theory, 2d ed., Progr. Math. 90, Birkhäuser, Boston, 1996.
MR 97c:19001
[Th1] R. W. THOMASON, Comparison of equivariant algebraic and topological K -theory,
Duke Math. J. 53 (1986), 795 – 825. MR 88h:18011 16, 19, 20, 21, 35
[Th2] , Lefschetz-Riemann-Roch theorem and coherent trace formula, Invent. Math.
85 (1986), 515 – 543. MR 87j:14028 50, 51, 53
[Th3] , “Algebraic K -theory of group scheme actions” in Algebraic Topology and
Algebraic K -Theory (Princeton, 1983), Ann. of Math. Stud. 113, Princeton Univ.
Press, Princeton, 1987, 539 – 563. MR 89c:18016 11, 19, 20, 21, 35, 36, 37, 42
[Th4] , Equivariant algebraic vs. topological K -homology Atiyah-Segal-style, Duke
Math. J. 56 (1988), 589 – 636. MR 89f:14015 15
[Th5] , Une formule de Lefschetz en K -théorie équivariante algébrique, Duke Math.
J. 68 (1992), 447 – 462. MR 93m:19007 12, 13, 19, 21, 22, 23, 45
[Th6] , Les K -groupes d’un schéma éclaté et une formule d’intersection
excédentaire, Invent. Math. 112 (1993), 195 – 215. MR 93k:19005
[ThTr] R. W. THOMASON and T. TROBAUGH, “Higher algebraic K -theory of schemes and of
derived categories” in The Grothendieck Festschrift, Vol. III, Progr. Math. 88,
Birkhäuser, Boston, 1990, 247 – 435. MR 92f:19001 6, 48, 49, 50, 51, 52, 53
[To1] B. TOEN, Théorèmes de Riemann-Roch pour les champs de Deligne-Mumford,
K -Theory 18 (1999), 33 – 76. MR 2000h:14010 5
[To2] , Notes on G-theory of Deligne-Mumford stacks, preprint,
arXiv:[Link]/9912172 5
[Vi1] A. VISTOLI, Higher equivariant K -theory for finite group actions, Duke Math. J. 63
(1991), 399 – 419. MR 92d:19005 2, 3, 5, 16, 18, 47
[Vi2] , “Equivariant Grothendieck groups and equivariant Chow groups” in
Classification of Irregular Varieties (Trento, Italy, 1990), ed. E. Ballico,
K -THEORY OF ACTIONS WITH FINITE STABILIZERS 55

F. Catanese, and C. Ciliberto, Lecture Notes in Math. 1515, Springer, Berlin,

1992, 112 – 133. MR 93j:14008 2, 5

Vezzosi
Dipartimento di Matematica, Università di Bologna, Piazza di Porta San Donato 5, Bologna
40127, Italy; vezzosi@[Link]

Vistoli
Dipartimento di Matematica, Università di Bologna, Piazza di Porta San Donato 5, Bologna
40127, Italy; vistoli@[Link]