You are on page 1of 58

# EQUIVARIANT ALGEBRAIC K-THEORY

by Z. Fiedorowicz,

H. Hauschild, and J.P. May

There are many ways that group actions enter into algebraic K-theory and there are various theories that fit under the rubric of our title. To anyone familiar with

both equivariant topological K-theory and Quillen's original definition and calculations of algebraic K-theory, there is a perfectly obvious program for the definition

of the equivariant algebraic K-theory of rings and its calculation for finite fields. While this program surely must have occurred to others, there are no pubThat

lished accounts and the technical details have not been worked out before.

part of the program which pertains to the complex Adams conjecture was outlined in a letter to one of us from Graeme Segal, and the real analog was assumed without proof by tom Dieck [10,11.3.8]. by Loday [19]. From a topological point of view, one way of thinking about Quillen's original definition runs as follows. has a notion of a principal When H Let ~ be a topological group, perhaps discrete. space One Negatively indexed equivariant K-groups were introduced

K-bundle and a classifying

B~ for such bundles. (possibly with dis-

is discrete, a principal

~-bundle is just a covering ~.

connected total space) with fibre and group groups BH n N n with union

Given any increasing sequence of

H, we obtain an increasing sequence of classifying spaces We then think of B~ as a classifying space for stable

with union B~. When the ~ n

bundles.

are discrete,

B~ may have desirable homology groups but In the cases of interest, we can use the (BN) + with the same

will have trivial higher homotopy groups. plus construction to convert homology. When

B~ to a Hopf space (= H-space) ~q(B~) + = Kq(A) (B~) +.

~n = GL(n,A), we define

for q > 0. In practice, we have sum While this can

There is a more structured way of looking at maps {9 :~
m

x ~
n

÷ ~m+n B~

and a corresponding Whitney sum of bundles.

he used to give

a product, it is generally not a Hopf space, although it is so

24

in the classical bundle theory cases obtained from When ~ = ~ ~ n~0 n

~

n

= U(n)

and B~

~

n

= O(n). inherits a

is a permutative category under @ , B ~ ~BBQ,

= II n~0

n

structure of topological monoid and we can form ifying space of B ~ .

the loop space on the class~ :B~ ÷ ~ B B ~ (B~) +. Deeper ~BB~

Homological analysis of the natural map ~0BB~ is equivalent to

shows that the basepoint component analysis shows that B B ~ and thus (BH) +

also has a classifying space, and so on, so that

are actually infinite loop spaces. We mimic the outline just given. (G,~)-bundle p:E ÷ B. For a topological This is just a

Let G be a finite group.

group ~, one has a notion of a principal principal is a map of ~-bundle

and a G-map between G-spaces such that the action by each g c G That is, the actions of G and H on the total space commute One has a classifying G-

~-bundles.

(we think of ~ as acting on the right and G on the left). space B(G,~) for such (G,~)-bundles. sentations of G in ~. In particular,

This space carries information about the repreone has the following basic fact.

Proposition 0. I. ~O

Let H be a subgroup of G.

For a homomorphism

p :H + ~,

define

to be the centralizer of 0, namely {~Io(h)~ = ~p(h) for all h c H}.

Then the fixed point subspace B(G,~) H has the homotopy type of union runs over a set R+(H,~) of representatives

Jl m

B~ O '

where the of H in ~.

for the representations

That is, R+(H,~) consists of one p in each conjugacy class of homomorphisms H ÷ ~. Note that, with H = e, this says that the underlying nonequivariant homo-

topy type of B(G,~) is just B~. Again, we are interested in increasing sequences {~n } with union ~. While

the third author has conbtructed an equivariant plus construction for G-connected Gspaces X, namely those X for which all X H are path connected, we shall not use it. Instead, we shall prove that, in the cases of interest to us, equivalent to a topological G-monoid B ~(G). J~n B(G,~n ) is G-

This means that B ~ (G) is a G-space

and a topological monoid such that its unit is a fixed point and its product is a G-

28

map.

Since the standard

classifying

space functor takes G-monoids to G-spaces,

to G-spaces and

the loop space functor takes G-spaces the G-space ~BB~(G)

this will allow us to construct

together with a natural G-map ~:B ~(G)
+

flBB~ (G).

These ideas lead us to the following

definition.

Definition

0.2.

For a discrete ring A, let K(A,G) = ~ B B ~ ( A , G )

be the G-space obtained by setting ~n = GL(n,A) in the discussion above. algebraically closed field A of characteristic KO(A,G) unequal to 2, let

For an

= ~BBO(A,G) Let

be the G-space obtained by setting ~n = O(n,A). K(G) = ~ B B ~ ( G ) be the G-spaces obtained and

KO(G) = ~BB O(G)

by setting ~n = U(n) and Rn = O(n).

The appropriate use of more general

definition of KO(A,G) orthogonal

for general

commutative

rings A requires 2. 1 and 2, first

groups and will be given in section in sections

We shall give the details behind this definition giving precise models for the general classifying Proposition

spaces B(G,R) and verifying B(G,R n) relevant

0. I and then giving different models for the particular

to the definition.

The second model is needed to obtain the required monoid strucin H. The most important problems to nontool

tures since the first model is not product-preserving in equivariant equivariant the simplest homotopy

theory is the reduction of equivariant

ones by passage

to fixed point spaces, and we study the fixed points of introduced in Definition 0.2 in section 3.

examples of the G-spaces

For based G-spaces X and Y (with G-fixed basepoints), of G-homotopy classes of based G-maps X + Y. to define

let IX,Y] G denote the set for H C G, it is stan-

In particular,

dard and natural

26

H(y) q

= [(G/H)+ A sq,Y] G = [sq,Y H] = ~ (yH), q Here and henceforward, X+ denotes

where S q is the q-sphere with trivial G-action.

the union of a G-space X and a disjoint G-fixed basepoint. to be a weak G-equivalence have the homotopy consider,

A G-map f:X + Y is said If X and Y

if each fH:xH ÷ yH is a weak equivalence.

types of G-CW complexes,

as holds for all G-spaces we shall equivalence [7,53]. It is also

such an f is necessarily

a G-homotopy

natural to consider

the "homotopy groups" ~(Y) = [sV,Y]G , of a real representation topological V of G.

where Sv is the one-point

compactification

As we shall see in section 5, equivariant over compact G-spaces X is represented KG(X) = [X+,K(G)]G and

K-theory of G-bundles

in the form KOG(X) = [X+,KO(G)]G, K-groups when Y = K(G)

hence the homotopy groups above are all examples of (reduced) or Y = KO(G). equivariant We regard the corresponding We write invariants

of K(A,G) and K0(A,G) as

algebraic K-groups.

KGA = ~GK(A,G) q q and similarly G-homotopy in the orthogonal case.

and

KGA = ~GK(A,G) v v we are really more interested in the

However,

types K(A;G) than in these invariants. the orthogonal

While the general linear case is in applications from

the central one algebraically, algebra to topology. 4. In particular,

case is important

We develop formal properties

of these definitions

in section

we discuss the naturality

in A of K(A,G),

G prove that K,(A) is a formula. We

commutative

graded ring if A is commutative,

and prove the projection

also verify that [X,K(A,G)] G is naturally Algebraically, construction

a module over the Burnside ring A(G). the equivariant Qsecond

the obvious next step is to introduce give the equivariant

on exact categories,

version of Quillen's

definition of algebraic K-theory,

and prove the equivalence

of the two notions. [5]. With this

This can all be done, and an exposition will appear in Benioff

27

approach,

Quillen's

devissage

theorem applies directly to the computation algebraic K-groups. the appropriate

of fixed

point categories

and thus of equivariant

Another obvious step is to introduce

notion of a permutative We have carried out

G-

category and prove that K(A,G) is an infinite loop G-space. this step and will present it in [12].

We prefer to be elementary

in this paper and

so will only make a few remarks about this in passing. We shall concentrate here on another obvious step, namely the equivariant We shall first prove the

analogs of Quillen's basic calculations following result.

in [28] and [29].

Theorem 0.3.

Let ~

q

be the algebraic closure of the field of q elements,

where q

is a prime which does not divide the order of G.

Then there are Brauer lift G-maps

B:K(~q,G) ÷ K(G)

and

B:KO(~q,G)

÷ KO(G)

whose fixed point maps BH induce isomorphisms prime to q.

on mod n cohomology

for all integers n

The outline of the proof is obvious.

We simply use our study of classifying

spaces to write down an explicit map and check that on fixed point sets it reduces to a product of maps of the form studied by Quillen. section 6 after first developing representation We give the argument in

general facts about the relationship equivariant K-theory

between

rings and topological

in section 5. Dold

As we discuss briefly in section 7, this result and the equivariant theorem mod k of Hauschild and Waner equivariant

[13] can be used to prove the following

Theorem 0.4. k s E ~i

Let k be prime to the order of G and let s be minimal such that Then for any stable real G-vector bundle \$ over a

modulo the order of G. base space,

compact G-connected kes~k(~)

there exists an integer e > 0 such that kes~ and equivalent.

are stably fibre G-homotopy

28

That is,

kes(~k~ - ~) is in the kernel of the real equivariant J-homomorphism. Using different techniques,

The same assertion is also valid in the complex case.

McClure [25] has recently obtained a sharper result, identifying all of the kernel of the J-homomorphism. The factor s cannot be eliminated : it is already necessary

when the base space is a point and one is asking about stable G-homotopy equivalence of G-spheres associated to representations. Finally, in section 8, we obtain the expected relationship between the equivariant algebraic K-theory of finite fields and the Adams operations in equivariant topological K-theory.

Theorem 0.5.

Let k r be the field with r = qa elements, where q is a prime which Let F~r(G) and Four(G) be the homotopy fibres of Then Brauer lifting induces

does not divide the order of G. the G-maps ~r-I:K(G) + K(G) and

~r-I:KO(G) + KO(G).

G-homotopy equivalences

B:K(kr,G) ÷ F~r(G)

and

~ :KO(kr,G) ÷ Four(G).

By the five lemma, this allows application of the reservoir of known information about KG(X) to the calculation of the functor [X,K(kr,G)]G, and similarly in the (technically deeper) orthogonal ease. It is a pleasure to thank Martin Isaacs, Irving Kaplansky, and David Leep for helpful conversations.

§i.

The classifying spaces B(G,~)

As is well-known (e.g. [17,18]), an appropriate iterated join construction can be used to obtain classifying spaces B(G,~). peripherally in Waner his definition. A nicer construction appears

[34, §3.2], and we shall give a categorical reformulation of

29

Throughout groups.

this section,

both G and ~ can be essentially should be imposed;

arbitrary

topological or

(Some minor technical restraints

G and ~ Lie groups,

discrete groups, more than suffices.) By a G-category we understand morphism a small topological target, category ~ w h o s e identity, object and

spaces are G-spaces and whose source, We require the identity space, determined by ~

and composition The

functions are G-maps. nerve, or simplicial realization B~

function to be a G-cofibration. G-space.

is a simpllcial

Its geometric ~.

is therefore a G-space, where

called the classifying

G-space of

Observe that ( B ~ ) H = B ( ~ H ) , morphisms of ~.

O H denotes the category of H-fixed objects and

We have the concomitant transformation of G-functors B~

notions of a (continuous) (qgx = gqx

G-functor and of a natural A G-functor S:~ +

for objects x).

induces a G-map B S : B ~ + homotopy Bq:BS = BT.

and a natural transformation

n:S ÷ T induces a Gcategories or type of G-CW

If each SH is an equivalence and B ~

of topological

are of the homotopy

as holds if the object and morphism spaces of ~

and ~ are so, then BS is

a G-homotopy equivalence.

Definition

i.i.

Define B(G,H) to be the classifying The objects are all pairs

space of the G-category

~(G,K)

specified as follows.

(p,x) such that p is a continuous

homomorphism H ÷ ~ for some closed subgroup H of G and x is an element of the space G/H of right cosets. [ [ G/H. P This object set is regarded as the left G-space (p,x) + (~,y) are the morphisms of principal (G,~)-bundles

The morphisms

G x H II

B p

= G x K II

3O

such that a(x) = y. and similarly for

Here ~p denotes ~ with the left H-action induced by o:K + #; ~ is a G-map and B is a

p :H ÷ H

G × ~-map, where G and ~ act of G and ~.

on the left and right of

G x H Hpand G ×K ~o

via the multiplications

This morphism set is given trivial G-action, B as a subspace of the disjoint union over G XH lip + G XK ~O"

g(~,B) = (a,B), and is topologized via p and o of the function spaces of maps

Remark 1.2.

A morphism

(~,B):(p,x) + (o,y) and

is specified by

~(gH) = gfK

B(g,P) = (gf,qP)

for g ~ G and p c ~, where f e G and q ~ H are fixed elements such that fK ¢ (G/K) H for h e H. Two pairs and qp(H) = o(f-lhf)q

(f,q) and (fk,k-lq) define the same morphism (~,B) and, if

x = g0 H, then y = g0fK.

Waner bundles.

[34] proved that B(G,~) is a classifying G-space for principal Indeed, if one replaces the object space of ~(G,~) by ~

(G,~)the

G x H ~p,

same definition gives another G-category The evident projection ~(B,N) ÷ ~(G,~)

~ ( G , ~ ) with classifying G-space E(G,~). becomes a universal principal (G,~)-bundle

on passage to classifying spaces.

Although much of our motivation comes from bundle

theory, we shall work entirely on the classifying space level in this paper. We regard a group or monoid as a category with a single object. result gives a proof of Proposition 0.1. The following

Proposition

1.3.

There is an inclusion of categories

i:

11 p e R+(H,H)

~P ÷ ~ ( G , ~ ) H

such that i has a right adjoint.

Therefore Bi is a homotopy equivalence.

Proof.

Define i by sending the unique object of the category ~P to the object (1,B):(p,eH) + (p,eH), where

(p,eH) and sending an element q c~ p to the morphism

31

8(g,p) = (g,qp);

compare Remark 1.2.

let

~(H,G,H) be the subcategory of and all morphisms
P ~

d(G,N) H

which consists of all objects conjugate to some
P c

(o,eH), o :H + H,

(l,y). Since o is

R+(H,~),

it is easy to see that i maps ~(H,G,H).

I1 ~p R+(H,~ )

isomorphically onto a skeleton of

It therefore suffices to construct a ~(H,G,H) in C~(G,H) H. ~(G,N) H. Define

right adjoint k to the inclusion, j say, of Let (T,y), T:K ÷ ~ and y = fK E(G/K) H,

be a typical object of

k(T,y) = o, where o:H + ~ is specified by o(h) = T(f-lhf). map G/H + G/K, we see that the pullback of G xK K ÷ G/K

Regarding y as a Galong y is G x H ~o + G/H

and that the resulting pullback square is a morphism :(o,eH) + (T,y). For a morphism (~,8):(T,y) ÷ (V,z) in ~(G,A) H, ~:L ÷ A and z ~ (G/L) H, we obtain ~ = j(v,z):

the following commutative diagram by passage to pullbacks, where

/ / ' ~

G

xK ~[T

,,.

/ / ~ a r ~-

G x .uI~ ]

G XHg q . . . . . . . .

~'G x H K u

G/K

~

~=

G/L

G / H / / ~

G/H f

The dotted arrow gives

k(~,8).

With these specifications, k is a functor and Visibly kj = Id and \$ o j and k o ~ are

is a natural transformation jk + Id. identity transformations.

This proves the result.

We shall need the following naturality property of the categories

\$(G,A).

Lemma 1.4.

Let

o:~ ÷ ~

be a continuous homomorphism. o,: ~ ( G , ~ ) +
~(G,~)

Then o induces a G-functor

whose restriction to any ~P is

o:Np ÷ ~o o p

Moreover, conjugate homomorphisms

induce naturally isomorphic G-functors.

32

Proof.

On objects,

o,(0,x) = (o o O,x).

On morphisms

(e,8)

with

B(g,p) = (gf,qP)

as in Remark 1.2, o,8(g,p) = (gf, o(q)p).

o,(a,8) = (e,o,B), where

The requisite verifications are immediate from the definitions.

A defect of the construction is that inclusions ~(G,E) ÷ ~(H,E).

H C G

do not induce functors

A related defect is that the natural G-functor ~(G, n x Z) + ~(G,n) x C,(G,Z) However, it does restrict to the obvious

is not an isomorphism of G-categories. identification

(p,O) cR+(H)~ × Z)

p ~ R+(H,H)

O E R+(H,Z)

and therefore induces a G-homotopy equivalence B(G,n × Z) ÷ B(G,H) × B(G,Z). Due to this last defect, an associative sum on l l n n does not induce a G-monoid structure on J_[B(G,Hn). To remedy this, we introduce different categorical models

for the relevant G-homotopy types B(G,Hn).

§2.

The general linear and orthogonal G-categories

Except where explicitly specified otherwise, we assume henceforward that the ambient group G is finite. Let A be a topological ring and consider Aisomorphism classes of finite A representation is

representations of G, by which we understand dimensional

A-free left modules over the group ring A[G].

indecomposable

if it is not a proper direct sum of representations.

88

Definition

2.1.

Let U(A,G) be a direct sum of countably many copies of a of each indecomposable A-representation of G and let U(A,G) n be the to be the category whose

representative

direct sum of n copies of U(A,G). objects are the n-dlmensional are the A-linear isomorphisms given the discrete topology; set of morphisms M on ~(n,A,G) r gV and

Define~(n,A,G)

A-free sub A-modules of U(A,G) n and whose morphisms between such modules. Here the set of objects is topology and the Let G act that

free A-modules are given the product space topology.

÷ N is then given the function

by translation f ; > gfg -i

of submodules and conjugation Define G-functors x ~(n,A,G) ÷ ~(m+n,A,G)

of isomorphisms;

is, V~

~) : ~ ( m , A , G )

via direct sums of modules and isomorphisms; canonical identifications

this makes sense by virtue of the

U(A,G)m~)U(A,G) n = U(A,G) m+n. The unique object 0 c ~.~(0,A,G) is a unit for ~) , andS) is strictly associative. of G-functors. Define

It is commutative up to coherent natural

isomorphism

~.~(A,G)

= n J ~ 0 .~.~ (n,A,G).

Of course, we are primarily discrete variant fields. However,

interested

in the case of discrete rings and indeed real and complex numbers lead to equito

the topologized

topological

K-theory.

Here we could just as well restrict attention and to linear isometric isomorphisms, we

orthogonal

and unitary representations ~(G) and

obtaining G-categories obtain ~ ( G ) similarly,

~(G).

Since A need not be commutative,

by use of quaternions. is "weakly'" permutative. Genuine permutative G-

The G-category ~ ( A , G ) categories structure appropriate [5,12].

for equivariant

infinite loop space theory must have more is clearly sufficient to ensure

The structure we have prescribed commutative

that B ~ , ~ ( A , G )

is a G-homotopy

topological

G-monoid.

The following section.

result connects

this definition

to that given in the previous

34

Proposition

2.2.

There

is a G-functor v: C~(G,GL(n,A)) functor v H + .~.~(n,A,G) is an equivalence + B2~;~ (n,A,G) the following diagram commutes up to natural of categories. Therefore

such that each fixed point

By :B(G,GL(n,A)) is a G-homotopy isomorphism equivalence. Moreover,

of G-functors:

~(G,GL(m,A))

x

~(G,GL(n,A))

v

x

v

, .~(m,A,G)

x ~(n,A,G)

~.(G,GL(m,A)

T

x GL(n,A))

@

~(c,Ce(m+n~ A))

~

(re+n,A,G)

J_[ B(G,GL(n,A)) n)0 to the G-homotopy commutative Proof. module For each homomorphism

Therefore

is a Hopf G-space topological

and is equivalent B~(A,G).

as a Hopf G-space

G-monoid

p:H + GL(n,A),

let A n denote P

the corresponding

A[H]-

and choose an H-isomorphism i P :An + V C U(A,G) n. P P contains a copy of the A[G]-module by v(p,x) = XVp. G x H An . P For a

There exists For

such a Vp since U(A,G) xVp = gYp.

x = gH ~ G/H, let

Define v on objects = (gf,qp)

morphism v(=,B)

(e,B) :(p,x) + (o,y), with B(g,p) to be the following composite:

as in Remark

1.2, define

-I XVp = gYp g ~ Vp

.-I i P ~- Anp

i q ~- Ano¢ = AnO o ~ Vo gf ~'-gfVo = YVo "

Here ¢ :H + K is specified qp(h) = (o¢)(h)q; being posite the K-module

by ¢(h) = f-lhf

and

q ¢ GL(n,A)

satisfies An+
p

thus q may be viewed as an H-isomorphism A~ regarded as an H-module the total by pullback composite

An

o¢'

the target Since the comof the

along ¢.

fioqi-lp :V + fV p o

is an H-map,

is independent

35

choice of coset representative g E x. functor

It is easy to see that v is a well-defined G-

(continuity on morphisms requiring GL(n,A) to have the function space An + An). Clearly v H restricts to an isomorphism from H. For the last statement,

topology of maps ~L H p, p c R+(H,~)

~ = GL(n,A), to a skeleton of ~ ( n , A , G )

composites of the form

-i xV p ~o" g -~ V p ~o"

i -I P @°~

i Am+n = Am@A p ~o" p n o P

@ i °,-V p @ V g ~xV ~)xV o" p o

provide the required natural isomorphism.

There are variants of the discussion above in terms of the various classical subgroups of general linear groups. We single out the orthogonal case. that A is commutative. the group Thus assume then

If the scalar 2 is non-zero and is not a zero-divisor,

O(n,A) of matrices X c GL(n,A) such that X - I = X t is the group of Q(~ sic i) = ~ a~. However, We it is

automorphisms of An which fix the standard quadratic form

could repeat the discussion above with GL(n,A) replaced by O(n,A).

more appropriate to allow general quadratic forms and, to avoid restrictions concerning orthogonal 2, to work with bilinear forms rather than quadratic forms. A-module, we understand a finite dimensional symmetric bilinear form. Thus, by an

free A-module V together

with a nondegenerate orthogonal

We then write O(V) for the group of

isomorphisms V + V.

By an orthogonal A-representation of G, we A-module with a G-invariant

understand an A-representation of G and an orthogonal form.

Definition 2.3.

Let Q(A,G) be a direct sum of countably many copies of a repreA-representation of G. Define O(n,A,G)

sentative of each indecomposable orthogonal to be the G-category of n-dimensional orthogonal isomorphisms between them.

orthogonal Define

sub A-modules of Q(A,G) n and

O(A,G)

= J~. n~ 0

O(n,A,G).

36

Again

O(A,G)

is a weakly permutative topological G-monoid,

G-category,

hence B O (A,G) is a G-

homotopy commutative

and we define

KO(A,G)

= ~BBO(A,G).

When A is an algebraically modules are isomorphic Definiton 0.2.

closed field of characteristic

# 2, all orthogonal reduces to that of

A-

to standard ones and this definition

The following analog of Proposition

Proposition

2.4.

Let

{Vi}

be a set of representatives

for the isomorphism classes

of orthogonal

A-modules.

There is a G-functor

~:

U dim V. = n 1

~(G,O(Vi,A))

÷

O(n,A,G)

such that each fixed point functor ~H is an equivalence

of categories.

Therefore

B~: [ [
dimV, 1 =n

B(G,O(Vi,A))

+ BO(n,A,G)

is a G-homotopy

equivalence.

Moreover,

up to natural isomorphism ~i" B(G,O(Vi,A))

of G-functors, is equivalent

the as a

functors ~ are compatible wit~ sums. Hopf G-space to B O(A,G).

Therefore

Remark 2.5.

If

~ c A,

1

any orthogonal argument

A-module is a direct summand of a hyperbolic com-

one, hence a cofinality

(as in section 4) shows that the basepoint to the plus construction on

ponent of KO(A) = KO(A,e)

is equivalent

l~m O(hAn).

Thus the groups ~,KO(A) are examples of the higher Witt groups of Wall [32], Karoubi [15], and others; equivariant see Loday [20] for a survey. Clearly our procedures also yield

generalizations

of the other examples of such theories.

37

Remark 2.6.

We could just as well have used finitely generated projective A-modules The evident analogs of Propositions only ~0 would be altered in the

rather than finite dimensional free ones above. 2.2 and 2.4 would hold. nonequivariant case. Again, by cofinality,

Remark 2.7.

Consider H C G.

In situations where complete reducibility holds, Therefore, when regarded as J~(A,G) is isomorphic

U(A,G) is evidently isomorphic as an H-space to U(A,H). a weakly permutative H-category by neglect of structure, to ~ ( A , H ) . In particular, the fixed point category

J ~ ( A , G ) H is independent of

the ambient group G.

The same is true in the orthogonal case.

Remarks 2.8.

The definitions in this section apply perfectly well to general but the proof of Proposition Never2.2 and

topological groups G and continuous representations,

2.2 fails because the functor ~ fails to be continuous on object spaces. theless, for suitably restricted G and A, the equivalences of Propositions 2.4 can be recovered by use of equivarlant bundle theory. Define ~(n,A,G)

We sketch the idea.

to be the category whose objects consist of pairs (V,v), where V is

an A-free sub A-module of U(A,G) n and v is an element of V, and whose morphisms (V,v) + (W,w) are the A-linear isomorphisms f:V ÷ W such that f(v) = w. Here the

object set is topologized as i[ V and the morphism sets are topologized via the triples (v,f,w) as subspaces of V × W V x W; the factors V and W ensure continuity

of the source and target functions.

Again, G acts via translation of objects and

conjugation of morphisms, and there is an obvious projection of G-categories

: ~(n,A,G) + ~.~(n,A,G).

One checks first that B~ is a (G,A)-bundle with fibre An .

One then checks that if a

closed subgroup H of G acts through any continuous homomorphlsm p :H + GL(n,A) on the total space E of the principal and contractible. (G,A)-bundle associated to B~, then EH is non-empty

By consideration of bundles over fixed point spaces, it follows

38

that the classifying G-map B~ (n,A,G) ÷ B(G,GL(n,A))

induces an equivalence on fixed point subspaces and is therefore a G-equivalence; see [17,2.14]. In particular, such an argument works to prove

B ~ (n,G) = B(G,U(n))

and

B O (n,G) = B(G,O(n)).

Here again, since any unitary or orthogonal representation of a closed subgroup H of G is contained in a unitary or orthogonal representation of G and we have complete reducibility, ~(G) H and O(G) H are independent of the ambient group G.

§3.

Analysis of fixed point categories

The first step in the study of the G-spaces introduced in Definition 0.2 must be the analysis of their fixed points. have (~BM) H = ~B(MH). M H = B(~H) When M = B ~ For any topological G-monoid M, we evidently Q ,

for a weakly permutative G-category

has the monoid structure determined by the induced structure of a Thus we must analyze the permutative categories we have already observed that a skeleton of where p ranges through R+(H,~n ). By itself

permutative category on ~ H . ~ ( A , G ) H. With ~ n

= GL(n,A),

~ ( n , A , G ) H is isomorphic to

li~ p p n'

this tells us little about the permutative structure on ~ ( A , G ) the lack of complete reducibility obstructs easy analysis of the

H, and in general ~P. n By Remark

2.7, we may take H = G without loss of generality when complete reducibility holds. It will clarify the arguments of this section if we change our point of view slightly. Use of the ambient A[G]-spaces U(A,G) n in Definition 2.1 served to allow The non-

the precise specification of a small weakly permutative G-category. equivariant permutative category ~(A,G)

G may be viewed as a pedantically careful A-free

model for the large symmetric monoidal category of all finite dimensional A[G]-modules and all A[G]-isomorphisms between them.

We agree to use the same

3g

notation for the latter category. categories but draw conclusions

In general, we agree to work with such large small permutative categories. There

for equivalent

is no loss of rigor since passage to skeleta and then to permutative (e.g. via [21, 4.2; 23, VI.3.2]) allows functorial replacement

categories

of symmetric monoidal permutative

categories with sets of isomorphism categories. dimensional

classes of objects by equivalent

In this spirit, we let ~ Z ( A )

mean both the category of finite and its skeletal

free (right) A-modules and their isomorphisms lIGL(n,A) (e.g. [23, p.162]). We let

permutative model

K(A) = ~ B B ~ ( A )

= BGL(A) + × Z.

Proposition S = {V i}

3.1.

Let F be a field of characteristic

prime to the order of G. representations

Let

be a set of representatives

for the irreducible Then ~

of G over

F and let D i be the division ring HomF[G](Vi,Vi). a permutative category to

(F,G) G is equivalent as

x ~ 2 ( D i ) . Therefore K(F,G) G is equivalent to × K(Di). i i Proof. Define O i: ~ ( D i) ÷ ~ x ( A , G ) G by @i(M) = M ~D.Vi and pi(f) = f ® 1 i for a finite dimensional Di-module M and an isomorphism f:M ÷ M'. Certainly O i commutes with sums. the full subcategory It is immediate of ~ from Schur's lemma that pi is an equivalence F[G]-modules to of The

(A,G) G whose objects are the isotypical G is isomorphic

type V i and that ~ ( A , G ) conclusion

to the product of these subcategories. mutative equivalents.

follows on passage to small

Since

K~(F)C = ~qK(F,G)G,

we can K-groups 0

e our new equivariant "i" That is,

K-groups of F

in terms of the nonequivariant

KG(F)~ = ~)Kq(Di). Recall that, if char F # 0, each D i is necessarily a field since 1 V i is obtained by extension of scalars from an irreducible representation U i defined over some finite subfield k i of F and thus

D i = HomF[G](Vi,Vi)

~ F~ k i

H°mki[G](Ui,Ui)"

40

We need some preliminaries analog. Consider

(related

to [28, App]) to obtain V = HomF(V,F) ,

the orthogonal Let to V*

the duality operator,

on F[G]-modules.

S O C S consist and let

of one V from each pair

(V,V*) such that V is not isomorphic a self-dual F[G]-module V ~ S with

S O = {V*IV ¢S O }. division ring D. to a G-invariant

Now consider

Clearly an F[G]-isomorphism nondegenerate b0:V~F bilinear V + F,

V ÷ V* corresponds G-form

by

form, abbreviated

henceforward. in the forms

Given one G-form

any other G-form b can be written

b(v,w) = b0(dv,w) for non-zero elements d , ~ E D; the function

= b0(v,dw) d + d specifies an involution of D

(which depends

on the choice of the fixed initially given G-form b0). then b0(dv,w ) = b0(dw,v ) = b0(w,~v) = b0(dv,w)

If b 0 and b

are both symmetric,

and thus

d = d.

The same conclusion

holds if both b 0 and b are skew syrmmetric. then the involution then is trivial in

If all b are symmetric and D is necessarily particular, ~ = -d

or all b are skew symmetric, If b 0 is symmetric Let

a field.

and b is not,

d ~ ~;

if b is skew symmetric.

S+ = {VIv ~ V* and every b is symmetric} S_ = {VIv ~ V* and no b is symmetric} and ,

,

S± = {VIV ~ V* and some b is and some b is not symmetric}.

Visibly, we have a decomposition

s = so i l s

olis+U.s-ll,

s,.
Since c is sym-

If V ¢ S_ and b is a G-form on V, let c(v,w) metric~ it must be degenerate. Thus

= b(v,w) + b(w,v).

= {wlh(w,V)

= 0}

sub G-space of V and must be all of V, hence c must be identically and S_ is empty. If char F # 2,

If char F = 2, this is a contradiction

41

this implies

that S = {VIv ~ v* and every b is skew symmetric }.

If V ¢ S, and b is not s y ~ e t r i c , nondegenerate S

then a(v,w) = b(v,w) - b(w,v) must be If char F ~ 2, this gives G'forms}. orthogonal F[G]-

and is thus a skew symmetric G-form.

= {V I V admits both symmetric and skew symmetric

For an F[G]-module V, let hV = (V ~ V * , module. The G-form h is specified by h((v,f),(w,g)) Note that hV = hV*. prematurely = g(v) + f(w)

h) be the hyperbolic

for v,wE V and f,g~ V*. For simplicity, we assume

We need the following observations.

that the characteristic

of F is prime to the order of G. if b(v,v) = 0 for all v ~ V. basis [14, Thm 19].

Recall that a form b on V is said to be alternate It follows that V is even dimensional Of course, alternate

and admits a symplectic and conversely

forms are skew symmetric,

if char F # 2.

Lemma 3.2. (i) (ii) (iii) (iv) If

Let (V,b) be an irreducible

orthogonal

F[G]-module.

char F ~ 2, then (V,b) ~) (V,-b) ~ hV. and 3(V,b) ~ (V,b) ~ hV.

If char F = 2 and V e S+, then hV is irreducible If char F = 2 and V s S@, then 2(V,b) ~ hV. If char F = 2 and V is non-trivial,

then b is alternate and V e S± if and quadratic form.

only if V admits an invariant nonsingular Proof. Identify V with V* via V ++ b(v,?).

For (i), (v,v') + (v + v',v - v') runs through all sym-

specifies an isomorphism metric G-forms as b does.

1 1 hV ÷ (V, ~b ) ~) (V, - ~b), and ~

For (ii), hV ~ (V,b) + (V,b') is easily checked to be im-

possible and (v,v',v") + (v + v' + v", v + v', v + v') specifies an isomorphism (V,b) ~ hV + 3(V,b). 2(V,b') ÷ hV, where For (iii), (v,v') + (dv + dv,v') specifies an isomorphism and b' runs through all symof D. For (iv), the of V and is zero

d ~ ~ and b'(v,w) = (dv + ~v,w),

metric G-forms as d runs through all non-invariant kernel of the function v ÷ b(v,v) is an invariant only if V is trivial; q(v) = b'(v,v) since any nonsingular

elements

sub F[G]-module

form q can be written as

for some nonsymmetric

bilinear form b' and since b' can be made

invariant by averaging,

the last clause is clear.

42

Lemma

3.3.

Any orthogonal

F[G]-module

is an orthogonal forms:

direct

sum of orthogonal

F[G]-modules (i) (ii)

of one of the following

(V,b), where V ~ S+II S~ hV, where Proof.

(and b is a symmetric

G-form)

V ~ SO lIS - if char F # 2 and V ~ SO II s+ if char F = 2. on the dimension, forms. il: suffices Forgetting to show that any (W,b) has a that V c W

By induction

direct

summand

of one of the cited

b, we may assume

for some V c S. (W,b) = (V,b) ~ degenerate. and, (~,b) since

If blV is nondegenerate, (~,b), where ~

then V ~ S+II S± and = 0}. Thus assume that blV is C (~)i

= {wlb(w,V)

Then

= V and thus blV = 0.

Therefore

dim V = dim (Vi) i , where

these are equalities. bIW' is nondegenerate.

This implies Let U = (W') i C W. Then

= (V,0) ~ (W',b), (W',b).

(W,b) = (U,b) • sequences

Define T :W ÷ V* by T(w)(v)

= b(w,v).

We clearly have exact

0 ~ and, by restriction, 0 Let o :V* ÷ U split U. that blU is given by b((v,f) If (oV ,b)

~

~ W

T

V*

~ 0

, V

, U ~-l~V *

> 0 .

Regarding

U as V ~ V* via o, we see from blV = 0 and To = 1

+ (w,g)) = g(v) + f(w) + b(gf,og). then bloV* = 0 and and the splitting (U,b) = hV. If (oV*,b) is

is degenerate,

nondegenerate, gives

then V c S + ~ L S e

(U,b) = (oV ,b) ~

((gv*)l,b)

the conclusion.

For a field F, let F-spaces and orthogonal

O(F) be the symmetric isomorphisms 2.3.

monoidal

category

of orthogonal category, such

or some equivalent

permutative

as O (F,e) of Definition then II O(n,F)

If F is algebraically permutative

closed and

char F # 2, model in the

is the most efficient is given

model.

An efficient

case F = kr, r odd, of O(F) If finite

in [II, II.4.8]. = ~BBO(F).

See [14,

I-i0]

for the structure

when char F = 2. char F ~ 2, let

Let KO(F) ~(F)

be the category

of symplectic

F-spaces bilinear

(namely form) and

dimensional

F-spaces

with a nondegenerate

skew symmetric

43

symplectic isomorphisms

or some equivalent permutative category.

Since all is a

symplectic F-spaces are isomorphic to standard ones [14, p. 22], llSp(2n,F) suitable model [II, p.l13]. Let KSp(F) = ~BB let ~(F).

For a division ring D with involution,

~(D) be the category of unitary DHermitian form Let

spaces (namely finite dimensional right D-spaces with a nondegenerate [14]) and unitary isomorphisms or some equivalent permutative KU(D) = nBB~(D).

category.

Proposition 3.4.

Let F be a field of characteristic

prime to the order of G.

If

char F # 2, then ~ (F,G) G is equivalent as a permutative category to the product

[

× ~f(Di) ] × [ × C)(Di) ] x [ × +(Di) ] × [ × ~(Di) ]. V i ~ SO V i~ S+ V. ~ SV. ~ S/: 1 O(F,G) G is equivalent to the product

If char F = 2,

[ ×
Vi

~Z(Di)] × [ ×
V i c S+

O(Di) ] × [ ×
Vi ~ S

~(Di)].

~ SO

Therefore KO(F,G) G is equivalent KO(Di) , KSp(Di) , and KU(Di). Proof. Regard

to the corresponding

product of spaces K(Di),

O(F,G) G as the category of all orthogonal First consider V i ~ S O .

F[G]-modules and

orthogonal G-isomorphisms.

Passage to duals specifies an M ~)Do.P V i
1

isomorphism D~ p --- HomF[G](Vi,Vi) canonically isomorphic to M to the F[G]-module

and, for a right Di-space M,
,

is by sending

(M~) D. Vi) " 1

Define

pi:~£

(D i) ÷

O(F,G) G

(M ~DiVi) ~) (M % o p 1

Vi)

with the hyperbolic

symmetric

G-form and sending an isomorphism f :M ÷ M' to

(f ~ I) @ ((f-l)* @ i).

Observe that every orthogonal G-isomorphism

44

Pi(M) ÷ Pi(M')

is of the form pi(f) for some f.

Next, consider V i ~ S + R S _ l l S ~ =

and recall that D i is a field if V i c S+~i S_. form b i on Vi; otherwise, a symmetric b if V i ¢ S..

If V i ~ S_, fix a skew symmetric GLet M be a right Di-space with

fix a symmetric G-form b i.

form b if V i c S+, a skew symmetric Then, in all three cases,

form b if V i c S_, or a Hermitian form G-form

the tensorial

(b ~ bi)(m ~ v,m ~ ~ v') = bi(b(m,m')v,v') on M ~DiV i is symmetric. Moreover, if f :M + M' is an isometric isomorphism, is of this then

so is f ~ form.

i, and every orthogonal

G-isomorphism

M ~ D i V i + M' ~D'Vil

Thus we obtain functors Pi with the appropriate Pi (M) = M ~ D i V. l and pi(f) = f ~ i.

domain and with target of

(~(F'G)G via

Note that a decomposition

(M,b) as a sum of one-dimensional decomposition

forms (in the S+ or S± case) gives a corresponding symmetric G-forms, = hV i and

of Pi(M,b) as a sum of copies of V i with different

and all such sums are so realized. hD i is orthogonally isomorphic irreducible.

If char F = 2 and V i ~ S+, then Pi(hDi) By Schur's lemma and Lemma 3.3,

C>(F,G) G is

to a product of full subcategories

such that Pi maps via an equivalence

to the i th subcategory.

We shall need the corresponding

equivariant

K-theory.

The following pair of results do not appear in the literature Their consequences KG(X) =- R(G) ~ K(X) and KOG(X) ~ [R(G;R) (~ KO(X)] @ [R(G;C) ~ K(X)] @

in quite this form.

[R(G;U) 0 KSp(X)] [30]. Here and below,

for a compact space X with trivial G-action are due to Segal

G may be a compact Lie group and R(G;F) denotes the free Abelian group generated by those irreducible
F.

real orthogonal

representations

of G with associated

division ring

Let ~ ,

O , and

~denote

either the categories

of finite dimensional isomorphisms

complex,

real, and quaternionic

inner product

spaces and isometric

or their

45

permutative skeleta l ~ U ( n ) , KO = ~BB O , and KSp = ~ B B +

ilO(n),

and ~ S p ( n )

([23, p.163]).

Let K = ~ B B ~

,

; these spaces represent complex, real, and

quaternionic K-theory. × Xi i consisting of those points with all but finitely many coordinates the basepoint. Similarly, the weak product of based categories ~ i has all but finitely many Recall that the weak product of based spaces X i is the subspace of

coordinates of each object and morphism the base object and its identity morphism.

Proposition 3.5.

Let

S = {V i}

be a set of representatives for the irreducible Then ~(G) G is equivalent as a ~ for each V i. Therefore

unitary representations of a compact Lie group G.

permutative category to the weak product of one copy of

K(G) G is equivalent to the weak product of one copy of K for each V i. Proof. Regard ~(G) G as the category of unitary representations of G and pi: ~ + ~(G) G by Pi(M) = M ~ C Vi and pi(f) = f ~ i,

their isomorphisms, define

and argue as in the proofs above.

Let t:R(G) + R(G), r:R(G) ÷ RO(G), c':RSp(G) + R(G), c:RO(G) + R(G), and q:R(G) ÷ RSp(G) be conjugation and the evident neglect of structure and extension of scalars homomorphisms; of course, tV ~ V*. By [3, 3.57], we may choose sets

{Ui} , {Vj}, and (Wk} t i o n s of G such that

of ~rreducible o r t h o g o n a l , u n i t a r y ,

and s y ~ p l e c t i c r e p r e s e n t a -

s = s011 s0 II s+l[ s
where SO = {Vj}, SO = {tVj}, S+ = {cUi} , and S_ = {C'Wk}. Moreover, complete sets

of inequivalent irreducible orthogonal and symplectic representations of G are given by { U i } ~ i { r V j } i [ { r c ' W k} and {qcUi}~i{qVj} ~--i{W k} •

Proposition 3.6.

Let G be a compact Lie group.

Then

O(G) G is equivalent as a for each Ui, one copy

permutative category to the weak product of one copy of O

46

of ~

for each rVj, and one copy of

~

for each rc'W k.

Therefore KO(G) G is

equivalent Proof.

to the weak product of correspondingly Regard

indexed copies of KO, K, and KSp. representations of G and where from to throw

O(G) G as the category of orthogonal Via M + M ~F Z

their isomorphisms.

on objects and f + f ®

1 on morphisms, Pi' Pj' and Pk

Z is Ui, Vj, or W k and F is R, C, or H, we obtain functors O,~, and ~ to O(G) G. Alternatively, we can use Adams

[3, 3.50-3.57] 3.4.

the proof onto an argument

just like the case F = C of Proposition

Remark 3.7.

A similar argument applies

to equivariant

symplectlc K-theory,

where

the conclusion

is that KSp(G) G is equivalent

to the weak product

of one copy of KSp

for each qcUi, one copy of K for each qVj, and one copy of K0 for each Wk; compare Kawakubo [16].

§4.

Group completions;

naturality

and products

The natural map substitutes

~ :M ÷ ~BM

for G-homotopy plus construction properties.

commutative

topological

G-monoids M

for an equivariant

in our work, and we begin this We then use this information the spaces K(A,G). Even nonto

section by discussing construct

its universal

naturality maps and pairings relating this approach to products [19].

equivariantly,

seems cleaner than that based on use of the spectrum level

plus construction treatment,

We shall give a much more structured

like that of [24], in [12]. specified, G can be an arbitrary topological group. Consider

Until otherwise Hopf G-spaces,

namely G-spaces X with a G-map

X x X ÷ X

and a G-fixed basepoint

which acts as a two-sided unit up to G-homotopy. the product to be associative

For present purposes, we require We say that X is We say that a Hopf

and commutative up to G-homotopy.

grouplike if ~0(X H) is a group for each closed subgroup H of G. G-map f:X ÷ Y is a group completion fH ~ H + yH is a group completion.

if Y is grouplike and if each Hopf map This means that fH induces group completion on

47

n 0 and induces localization of H.(xH;R) at its submonoid ~0(X H) for each commutative coefficient ring R; see [21, §I] for discussion. If X itself is grouplike, then

each fH is an equivalence and thus f is a G-equivalence. For M as above, application of the "group completion theorem" fixed point spaces implies that ~ :M ÷ ~ I is a group completion. [22, 26, 27] to To e~plolt this

fact, we need the following notion.

Definition 4.1.

A cofinal sequence in a Hopf G-space X is a sequence of points

a t ~ X G, with a 0 the unit, such that at+ 1 is in the same path component of X G as bta t for some b t ~ X G and such that, for each H C G and c ~ X H, there exists d ~X H

and t ~ 0 such that dc and a t are in the same path component of X H.

Given such a sequence,

let X t be a copy of X and let bt :Xt ÷ Xt+ 1.

X

be the telescope of then each b t Indeed, if X t

the sequence of left translations

If X is grouplike,

is a G-equivalence and the natural map X = X 0 + X is a G-equivalence.

is given the basepoint a t and the multiplication obtained by composing the product on X with translation by a point in the component of ~0(X G) inverse to at, then each b t is an equivalence of Hopf G-spaces. If f:X ÷ Y is a group completion and we form X and ~ with respect to a

cofinal sequence in X and its image in Y, then f commutes up to G-homotopy with the translations and therefore factors up to G-homotopy as a composite X+~ ~ = Y.

The construction commutes with passage to fixed point spaces, and it is a direct consequence of the group completion property that each on homology. ~H induces an isomorphism this implies the

By an easy (but not obvious) homotopical argument,

following universal property of f; see [8] for details.

Proposition 4.2.

Let f :X + Y be a group completion, where X contains a cofinal be any Hopf G-map, where Z is grouplike. such that ~f Then there is a

sequence, and let g:X + Z Hopf G-map g:Y + Z,

unique up to G-homotopy,

is G-homotopic to g.

48

By a function space argument, is due to Caruso; see [8].

this result has the following consequence, which

Proposition 4.3.

Let f:X ÷ Y

and

f' :X' + Y' be group completions, where X and X'

contain cofinal sequences, and let g :XA X' + Z be any G-homotopy bilinear G-map, where Z is a grouplike Hopf G-space. ~AY' + Z, unique up to G-homotopy, Then there is a G-homotopy bilinear G-map such that g(f ^ f ' ) is G-homotopic to g.

Remarks 4.4.

We have deliberately misstated the previous two results: all

conclusions concerning G-homotopies really only hold up to "weak" G-homotopy in general. Here two G-maps f,f' :X + Y are weakly G-homotopic if their composites In other

fk,f'k:K ÷ Y are G-homotopic for any compact G-space K and G-map k:K + X.

words, the results are correct on the level of represented functors defined on compact G-spaces. The lim I exact sequences associated with the telescopes we have We leave it to the reader to insert the word

used are the source of the ambiguity.

"weak" where needed, doing so once in a while ourselves as a reminder.

The reader who would like to see the simpler nonequivariant to [7].

proofs is referred

While the results are true for any G, we shall only use them when G is T and T' are G-equlvalences.

finite and, peripherally, when G is compact Lie and

Assume that G is finite in the rest of this section.

We restrict attention to

the general linear case, but everything we say applies equally well in the orthogonal case. BQ We first verify our cofinality assumptions. Recall that components of ~ is a groupoid.

correspond bijectively to isomorphism classes of objects when

Lemma 4.5. Proof. cofinal. let

B~(A,G) Write

contains a cofinal sequence. r r U(A,G) = ~ V~. and let W t = i ~IV~ .

i=l
Certainly Wt+ I = W I ~)W t. Then Z ~ Y' ~)Y
H

1

We claim that C

{Wt} U(A,G),

is

For H C

G and an A-free H-fixed Y

Z = G ×.~ Y.

as an A[H]-space and Z' ~ Z ~ W t as an A[G]~ W t.

space for appropriate Y' and Z' in U(A,G) and some t, hence (Z' ~ Y') G Y

49

Remarks 4.6. ~IB(G,GL(n,A))

It would not be useful with respect

to form the telescope

of copies of

to the maps induced by the inclusions {W t} by the noncofinal telescope is G-equivalent

GL(n,A) + GL(n+I,A). sequence {A t }

This would amount to replacing

of trivial representations.

The resulting

to B(G,GLA) × Z, and its irrelevance our theory. telescope By cofinality, B~(A,G)

explains why B(G,GLA)

plays no explicit role in

we do have a natural G-map from B(G,GLA) x Z to the to the sequence {Wt}.

defined as above the respect

We illustrate maps.

the force of Proposition

4.2 by using it to construct Applying

naturality

Let f :B ÷ A be a ring homomorphism.

f to matrix entries, we obtain

homomorphisms f#:GL(n,B) + GL(n,A).

By Lemma 1.4 and the G-equivalences G-maps f By the evident compatibility its composite with ~.

of Proposition

2.2, these homomorphisms

induce

:B~Z(B,G)

+ B~(A,G).

with sums of the f#, f* is a Hopf G-map, hence so is 4.2, there results a Hopf G-map

By Proposition f

~(B,G) + K(A,G) free left B-module homomorphisms by pullback along

such that

f ~ = ~f .

If A is a d-dimensional

f, then a choice of basis for A over B determines

f#:GL(n,A) ÷ GL(nd,B).

Again, there results a Hopf G-map

f, :B~.~ (A,G) + B ~

(B,G)

and thus a Hopf G-map

f,:K(A,G) ÷ K(B,G)

such that

f,~ = ~f,.

50

These particular ambiguity, isomorphic

maps can also be obtained,

and with no weak homotopy Indeed, A®B U(B,G) is

by direct use of our categorical models.

to a direct summand of U(A,G) and, if A is B-free, to U(B,G).

U(A,G) is isomorphic of weakly per-

as a B[G]-module

Therefore we have evident morphisms

mutative G-categories

f#:~(B,G)

+~(A,G)

and

f#:~(A,G)

÷ ~(B,G).

The uniqueness

claim of Proposition

4.2 implies

that the resulting maps of K-theory paragraph. Of course,

spaces agree with those constructed Proposition argument

in the previous

4.2 may be applied in situations where no such precise categorical

is available. ring and write ~ n = GL(n,A). The tensor product

Now let A be a commutative homomorphisms ~ m × H n + ~ mn

induce G-maps

B(G,~m) x B(G,~n) = B(G, Hm × ~n ) + B(G, ~mn )

via Lemma

1.4.

since 9 0 is the trivial group,

C ( G , ~ 0) has object space Using the

J~H G/H

and terminal

object * = G/G, hence B(G,~ 0) is G-contractible. of ~ over ~ and Proposition bilinear G-map

distributivity

2.2, we find easily that these maps give

rise to a G-homotopy

@

:B~(A,G)

A B@~(A,G)

+ B~(A,G).

Composing with ~ and applying Proposition

4.3, we obtain a G-homotopy

bilinear G-map

:K(A,G)AK(A,G)

+ K(A,G).

Indeed, K(A,G) is a Hopf ring G-space in the sense that the axioms for a commutative ring hold up to (weak) G-homotopy. with ~ and the uniqueness o (f A f ) = f following B~(A,G) result, o ~ . For a ring homomorphism f :B ÷ A, f# commutes

claim of Proposition

4.3 implies that groups, this implies the

On passage to G-homotopy

its last statement

being a consequence

of the defintion of

and the group completion

property.

51

Let

R~A(G) denote the Grothendieck ring of finite dimensional A-free A-

representations of G under the additivity relations generated by A[G]-split short exact sequences.

Proposition 4.7. ring.

For commutative rings A, K~(A)

K~(A)

is naturally a commutative graded R?(G).

In particular,

is isomorphic to the ring

The appropriate relationship between products and f, is given by the following "projection formula".

Proposition 4.8.

Let f:B ÷ A

be a homomorphism of commutative rings such that A is Then

a finite dimensional free B-module.

f,(xf (y)) = f,(x)y in K~+r(B) for x~ KG(A) and y~ KG(B). q r Proof. The following diagrams commute up to conjugation, where d = dimBA:

GL(m,A) × GL(n,A)

1 x f#

GL(m,A) × GL(n,B)

f# × i

GL(md,B) × GL(n,B)

®

f#
GL (mn, A)

GL(mnd,B)

Therefore, by Lemma 1.4 and Proposition 2.2, the corresponding diagram with general linear groups replaced by classifying spaces B ~ ( A , G ) to G-homotopy. and B ~ ( B , G ) commutes up

Since both composites in the latter diagram are G-homotopy bilinear,

the uniqueness claim in Proposition 4.3 gives the G-homotopy commutativity of the diagram f,A i K(A,G) A K(A,G) 1 A f K(A,G)AK(B,G ~)
K(B,G) ^ K(B,G)

®

f#
K(A,G) The conclusion follows on passage to homotopy groups
G .-- K ( B , G )

52

Let A(G) denote the Burnside ring of G, namely the Grothendieck

ring associated

to the semi-ring of isomorphism classes of finite G-sets under disjoint union and Cartesian product. n:A(G) ÷ consequence Via permutation representations, there is a ring homomorphism f:B + A. We have the following

(G), and f ~ = ~

for a ring homomorphism 4.7.

of the last statement of Proposition

Corollary 4.9.

For commutative

rings A and compact based G-spaces X,

[X,K(A,G)]G is

naturally a module over A(G). Proof. x:X ÷ K(A,G), For aX aE A(G), n(~) may be viewed as a based G-map SO ÷ K(A,G). For

is defined to be the composite

X = S0^ X

~(~)AX~

K(A,G) ^ K(A,G)

®

~ K(A,G).

The commutativity

and compactness

hypotheses

are actually unnecessary,

and we

shall see in [12] that this is only a small part of the full algebraic our functors ticular, that is implied by equivariant infinite loop space theory.

structure on In par-

that theory will imply that K(A,G)

is a ring G-space up to all higher G-

homotopies

rather than merely up to weak G-homotopy.

§5.

Equivariant

K-theory and representation

rings

We first prove that K(G) and KO(G) represent

topological

equivariant

K-theory of

G-bundles over compact G-spaces and then construct natural ring homomorphisms a:R(~) ÷ KGB(G,~ ) and a:RO(~) ÷ KOGB(G,~) , rings of H. Throughout

where R(N) and RO(H) are the complex and real representation this section, H and G are to be compact Lie groups.

We shall work in the complex

case, but all results and proofs apply equally well in the real case. t Consider ~(G). Write U(C,G) = I V~ and let W t = I V~ and i t i>l i=l _t+l Z t = ( ~ Vi) ~ v t + I . Then {W t I t ~ O} is a eofinal sequence in B ~ ( G ) with i=l Z t ~ W t = Wt+ 1. Let ~(G) be t h e t e l e s c o p e o f t h e s e q u e n c e of t r a n s l a t i o n s Z t : B ~ ( G ) t * B ~ ( G ) t + I of copies of B~(G) = ~ B~(n,G) n~0 and let ~ : g ~ ( G ) + K(G)

53

be the resulting

factorization

of the group completion bundles.

~ :B~(G) ÷ K(G).

By Remarks

2.8, B ~ (G) classifies

(complex) G-vector

Proposition

5.1.

The map

~.'B~(G) + K(G)

is a G-homotopy

equivalence,

and these hold in

spaces represent the real case. Proof. G-vector

the functor K G on compact G-spaces X.

The same conclusions

KG(X )

is the Grothendieck group associated

to equivalence sum.

classes of

bundles over X under the addition given by Whitney :[X+, B ~ ( G ) ] G ÷ KG(X)

Define a function

as follows. say via ft"

By compactness, Let ft classify

a G-map

f:X ÷ ~ ( G )

factors through some B ~ ( G ) t , ~[f] = ~t - ~t" X x V ÷ X. Here, for a

the bundle ~t and define

unitary representation well-defined

V, V denotes the trivial G-bundle

Clearly ~ is a theorem V such trivial

injective function.

For a G-bundle ~ over X, the Peter-Weyl G-bundle n i and a representation

implies the existence that ~ On ±

of a complementary

is equivalent

to V; see Segal

[30, 2.4].

Adding on a further \$ - ~

bundle, we may take V = W t for some t. to ~ + nl - ~t"

Then a difference

in KG(X) is equal space for an

Therefore ~ is a surjection.

As the representing ~ ~(G)

Abelian group valued functor on compact G-spaces, G-space.

is a grouplike weak Hepf

Therefore its fixed point subspaces are grouplike weak Hopf spaces and Since each ~ is a homology isomorphism, each ~ is an

thus simple spaces. equivalence and ~

is a G-equivalence.

It is now clear from the discussion at the start of the previous is a weak Hopf G-map and thus from K(G). Note that ~ ~(G)

section that

inherits an actual Hopf G-space structure represents the natural map from G-bundles compact Gto

~ :B~(G) ÷ K(G)

virtual G-bundles. spaces X.

We define KG(X) = [X+, K(G)] G for not necessarily

From our present bundle theoretical complementary bundles that distinguishes

point of view, it is the existence of topological from algebraic K-theory. It is

54

the absence of complements

in the relevant bundle theories with discrete fibres that

causes the need for the plus construction and group completion. By use of Lemma 1.4 and Remarks 2.8, we see that the tensor products U(m) × U(n) ÷ U(mn) give rise to a map K(G) A K(G) ÷ K(G) which represents the tensor product operation on G-vector bundles over compact G-spaces and which gives K(G) a structure of weak Hopf ring G-space. space, with no lim I ambiguities, In fact, K(G) is an actual Hopf ring G-

and thus KG(X) is a commutative ring for any G-

space X by virtue of Proposition 5.1 and the following result.

Proposition 5.2.

KGB(G,H ) contains no lim I term.

That is, if B(G,H) is the union

of an expanding sequence of compact G-spaces Bn, then KGB(G,N) = lim KGB n.

This is not hard to prove directly when G and ~ are finite.

When G and ~ are

compact Lie groups, Haeberle and Hauschild have proven the result as a byproduct of a suitable generalization of the Atiyah-Segal completion theorem zation gives that KGB(G,~) [4]. The generali-

is naturally isomorphic to the completion of R(G x ~)

with respect to the topology generated by the kernels of the restriction homomorphisms R(G x H) ÷ R(H) for H C G.

We shall construct Adams operations ~r on K(G) and thus on KG(X) for any Gspace X in the course of the following proof.

Proposition 5.3.

There is a ring homomorphism ~:R(~) + KGB(G,N) which is natural in The same conclusions hold in the real

and commutes with the Adams operations ~r. case. Proof. For a representation

T:~ ÷ U(n), define ~(T)

to be the G-homotopy

class of the composite BT. B(G,N) ~ B(G,U(n)) = B ~ ( n , G ) C B~(G) ~ ~K(G)

obtained by use of Lemma 2.4 and Remarks 2.8. ~(o + T) = ~(o) + ~(T),

Trivial diagram chases show that Naturality is clear. By the very Consider

hence ~ extends over R(G).

the identity homomorphism of U(n) as an element construction of O :K(G) A K(G) ÷ K(G),

i n c R(U(n)).

we certainly have

55

~(i m @ in) = =(im) ® ~(in). representations Adams [i,§4], o and T.

It follows formally that

=(o @ T) = ~(o) ® ~(T)

for By

Therefore by bilinearity,

~ is a ring homomorphism. The maps ÷ K(G)

~r:R(~) + R(~) ÷ K(G)

is a natural ring homomorphism. determine a map ~r ~r:B~(G)

~r(in):B(G,U(n)) G-spaces.

therefore

of Hopf ring

By Propositions

4.2 and 4.3,

extends to a map of Hopf ring G-spaces Therefore r = ~r on

~r:K(G) + K(G), and ~r~(i n) = a~r(in) representations and thus, by linearity,

by construction. on R(~).

We need to understand H C G and a G-space X, let

the behavior

of ~ on passage to fixed points.

Thus, for

~H:KG(X) = [X+,K(G)]G ÷ [X~,K(G)H]
denote the restriction homomorphism. By Propositions 3.5 and 3.6 and Remarks 2.8,

K(G) H = x K and i are the weak products

KO(G) H = ( x K0) x ( x K) x ( x KSp) i j k irreducible of grouplike representations Hopf spaces, of H.

indexed on the appropriate × Y. i i

Note that, for a space X and weak product an inclusion [X,Y i] C i which is an isomorphism 0. I, we have

there is

[X, x Yi] i is finite. By Proposition

if X is compact or the product

p ~ R (H,K) and of course [V ~ , Y ] k = x [Xk,Y ] k for any spaces X k and Y.

Definition (i)

5.4.

Let p:H ÷ K be a morphism of compact Lie groups.

Define a homomorphism

uP:R(~)
as follows, where {V i}

÷

R(H) ® R(n p) : X R(~p)
i unitary representations of H.

is the set of irreducible

Regarding a representation T:E + U(n) as a C[H]-module V by pullback along p, we may n. write V : ~ C z ® C Vi and define Ti:~P + U(n i ) by c o m m u t a t i v i t y of t h e f o l l o w i n g

i d i a g r a m s f o r x E ~0:

56

cni @ C Vi i

Ti(x) @ 1
i
n.

x(x)

[ c ~®c vi
i

That is, x i is the composite of T:~ p + U(n) Tp and the projection

U(n) Tp ÷ U(ni)-

Define vP(T) = ~ T i . Additivity is easily checked, hence v 0 extends over R(R). i Visibly, if dimcV i = di, we have the character formula XT(x) = ~ diXx.(X) i x (ii) Define a homomorphism for x EH p

vP:R0(~) + [R(H;R) @ Ro(RP)] ~ [R(H;C) @ R(RP)] @ [R(H;H) @ RSp(RP)I

similarly, where R(H;F) is the free Abelian group generated by those irreducible real representations Proposition 3.6.) of H with associated division ring F. (Compare the proof of

Proposition 5.5. with I [B~,K] i

The following diagram commutes, where R(H) ~ K(BH p) is identified and the products run over p c R+(H,H):

R(K)

(vP)

~

x

R(H)

@

R(~ p)

0

H

KGB(G,E) '

v

• x [BH~, x K] 0 i

~

x R(H) @ K(B~ p) p

With the evident modifications,

the analogous diagram commutes in the real case.

57

Proof. definitions•

On representations,

this is immediate by inspection of the Of course, the real analog

requires use of the evident homomorphism :RSp(E p) ÷ KSp(ED).

We also need the following commutation relation between Adams operations and the
V0 .

Propositon 5.6.

Let H be finite and let r be prime to the order of H.

Then the

following diagram commutes for each homomorphism p :H + K :

R(n)

v0

, R(H) ~ R(n ~)

R(II)

v0

, R(H) @ R(II O)

The same conclusion holds in the real case except that, on the complex part R(H;C) ~ R(~ p) of the target of v p , @r ~ ~r ,,~r ~ ~=r,, which sends the jth summand @r if ~r(vj) c S O Proof. and via must be replaced by the homomorphism to the @r(j)th summand via

R(~ p)

~-r if ~2r(vj) c S O . X r T (x) = XT(xr) ~r and the standard

By the character formula

irreducibility criterion [31,p.29], irreducible representations of H.

acts as a permutation on the set of

Choosing a representative V in each orbit and ~r, we can arrange that ~r acts as a with

taking iterated translates of these V under permutation on {Vi} , say ~r(v i) = V . ~r(i) n TO = ~ vii, we have

For a homomorphism

T :~ ÷ U(n)

n.

~r(T) o p = ~r(Tp) = [ V i i ~r(i) We also have the character formula

~. d i X r (x) = ~ dixTi(x r) = XT(x r) = X (x) = [ diX@r(T)i(x) i ~ (T i) " *r(T)

58

~r(i ) 1 argument in the r e a l case is s i m i l a r , except that the isomorphism
*

for x¢ ~P , and of course

d

= d..

It follows that

~r(T) r(i ) = ~r(ri).

The

r t ~ r intervenes
with a linear {Vj,tVj};

when

V

~r(j)

~ SO.

The point is that the identification

of

tO = V

combination assigns a privileged role to one Vj in each of the pairs changing t, and the choice changes the corresponding ~-r = ~rt. component of ~P

by composition with

Corollary 5.7. following

Let G be finite and let r be prime to the order of G. for any subgroup H of G:

Then the

diagram commutes

K(G) H-

=

x
i

K

(~r)H !
K( )H =

I (X ~r)m i ×
i K

where

m

permutes

factors as

~r

permutes

the indexing representations

of H.

The

same conclusion x K, (x ~r)m j the ~ ( j ) t h Proof. that, for in

holds in the real case except that, on the complex part "(x ~i-r) . which sends the jth factor to J , ~r if ~r(vj) c S O and via ~-r if ~r(vj) c S O . 5.3, 5.5, and 5.6 with ~ = U(n), we conclude by the map

must be replaced factor via

Applying Propositions

R(U(n)),

[~r~(in)]H = [e~r(in)]H = × (i ~ a)~P~r(in) O

= x (~r ~ ~r )vP(in) " P

If p ~ R+(H,U(n)) U(n) p ÷ U(ni). for ~r :B~(G)

is

n. [ V i 1 , then the it h component of p as a sequence

of vP(in) is the projection terms, we see that,

Thinking

{n i} and collecting

+ K(G) as specified

in the proof of Propositon

5.3,

The conclusion

(~r)H = (x ~r)~: x B ~ + x KU. i i i follows on passage to group completions,

the proof in the real case

being identical.

59

§6.

Brauer lifting and the proof of Theorem 0.3

We revert to our standing assumption does not divide the order of G, let let k = ~ construct q denote the algebraic G-maps 8:K(k,G) ÷ K(G) and

that G is finite.

Fix a prime q which and

k r denote the field with r = qa elements, To prove Theorem 0.3, we must

closure of k . q

8:KO(k,G) ÷ KO(G)

whose fixed point maps 8H are products of the Brauer lift maps introduced Quillen. The construction with Proposition 5.3. of 8 is based on use of the following Note that the Frobenius automorphisms

by

result in composition ~r:GL(n,k) ÷ GL(n,k) ~r

obtained by raising matrix entries to the r th power induce ring automorphisms of the representation rings Rk(K ) and ROk(R) for any finite group ~.

Proposition fies

6.1.

There is a natural ring homomorphism for r = qa. The same conclusions

X:Rk(~) ÷ R(~)

which satiscase.

X~ r = ~rx Proof.

hold in the orthogonal

We sketch the argument.

One first fixes an embedding ~:k* ÷ C* T:H ÷ GL(n,k), one then defines

of the roots of unity.

For a representation

×~(~)(x) = ~ ~(~i ),
where {el,...,~ n} XX(T ) i are the eigenvalues of r(x).

xc~,
The Brauer induction theorem

is the character of a complex representation and the compatibility The orthogonal between

X(T). are checked

multiplicativity,

~r and ~r

by trivial character computations. For q # 2, Quillen homomorphism erroneously

case is obtained by restriction. to a

[28, App] gave details, Consider

and he also showed that X restricts Here [ii, p.174]

RSPk(K) + RSp(N). asserted

the case q = 2.

that X fails to carry orthogonal

representations

to orthogonal

representations.

To see that it does, let the exponent

of H be 2ah, where h is odd, By Quillen [22, on

choose b ~ a such that 2b - 1 is divisible by h, and let r = 2b. 5.4.2] and the explicit R(~) such that ~rx = X. choice for ~ given below, By Adams

~r is an idempotent

operator and

[3, 3.6.4], ~r carries both orthogonal

60

symplectic

representations

to orthogonal

representations.

In view of Lemma 3.3, it

suffices to show that if V is a k[G]-module self-dual [3,3.32],

with a symmetric G-form b, then X(V) is representations). By Adams

(and thus a sum of orthogonal and symplectic this will hold if p:H ÷ GL(n,k), XX(v)(X) = XX(v)(X -I)

for all x ¢ H.

Regarding V as a of the matrices to p(x)

homomorphism

we need only show that the eigenvalues

p(x) and p(x -I) are the same.

But this is clear since p(x -I) is adjoint

with respect to the given form b.

When q does not divide the order of H, k is an isomorphism. our work case. that k takes honest representations to choose

It is crucial to in this

to honest representations ~

To see this, it is convenient

so as to arrange precise comhomomorphism. The same

patibility

between Brauer lifting and the decomposition

point arose in work of May and Tornehave Let P0 be the ideal given

[23, p.222]; and we recall their solution. Inductively,

(q) in the ring R 0 = Z and let A 0 = Z(q).

Pj-I'

R j - I ' and Aj_I, l e t Aj be the l o c a l i z a t i o n of the ring of cyclotomic

integers R. z Z[exp(2~i/(qJ-l)(q j-I - l)...(q-l))] 3 at a chosen prime ideal Pj which contains fractions maximal of Aj and let I:A. ÷ G 3 Pj-I C Rj_ 1. Let Kj C C be the field of

be the inclusion.

The quotient of Aj by its k = lim kr(j) . Let

ideal is a field kr(j) of characteristic

q and

~:Aj + k r ( j )

be the quotient map.

Clearly Aj contains a group

~j

of

(qJ-l)...(q-l) th subgroup ~j of ~:k
*

roots of unity which ~ maps isomorphically kr(j) ,_
_

onto the corresponding as j varies. We specify

these isomorphisms

being compatible i o
-i

÷ C

*

by letting its restriction

to uj be

For j sufficiently

large that Kj contains

the m(H) th roots of unity, where m(~) of H, it is now obvious

is the least common multiple from Serre [31, 18.4] that

of the orders of the elements

%:Rkr(j)(II)-~ Rk(II) + R(]I)
coincides with

o:~ k r(j)

(n) + mK.(n) "--R(n), 3

61

where o is the canonical section of the decomposition homomorphism and the isomorphisms are given by extension of scalars.

Proposition 6.2.

If q does not divide the order of 9, then ~ ÷ R+(~)

X:Rk(~) + R(H)

is an

isomorphism and restricts to an isomorphism representations.

of semi-rings of honest

The same conclusions hold in the orthogonal case, where the three

types of irreducible representations (or two types if q = 2) are each mapped bijectively onto the corresponding type. Proof. The first statement follows from [31, 15.5]. For the last statement,

we agree to choose the respective sets SO C

S O jj S O so that XS 0 = S O and we note

that there are no irreducible representations of type S± over an algebraically closed field. For the case q = 2, we note that all self-dual complex represen-

tations of an odd order group are orthogonal since, with r as in the previous proof, ~rx = X for all complex characters X because all X are in the image of X.

We also need the following analog of Definition 5.4.

Definition 6.3. (i)

Let

o:H ÷ ~

be a homomorphism of finite groups.

Define a homomorphism v P : ~ ( H ) + ' ~ ( H ) ® Rk(~P) = ~ Rk(~P) i

as follows, where representation

{V i}

is the set of irreducible k-representations of H.
set vP(t)

For a

= ~ Ti, w h e r e Ti i s the composite of i T:~ p ÷ GL(n,k) ~p and the projection GL(n,k) Tp ÷ GL(ni,k). As in Definition 5.4, we n. regard Tp a s ~ k 1 ~ k Vi to specify the projec tions , and we h a v e t h e f o l l o w i n g i trace formula, where d i = dim k Vi:

T:H ÷ G L ( n , k ) ,

trr(x) = ~i d i trTi(x) (ii) Define a homomorphism

for

xc HP.

vP:ROk(~) + [R(H;S+) ~ ROk(~P) ] O [R(H;S 0) ~ Rk(HP)] O

[R(H;S ) ® RSPk(HP)]

similarly, where R(H;T) denotes the free Abelian group generated by those

62

irreducible orthogonal k-representations of H indexed on the set T. proof of Proposition 3.4.) We have the following analog of Proposition 5.5.

(Compare the

Proposition 6.4.

If q does not divide the order of H, then the following diagram p c R+(H,E):

commutes, where the products run over

Re(q)

(~P)
• x Rk(H ) O

P
(v ° )
~ X

Rk(IIP)

R(K)

R(H) ~ R(H P)
P

With the evident modifications, the analogous diagram commutes in the orthogonal case. Proof. We have that I:Rk(H ) + R(H) is the isomorphism induced by a bijection By Definitions 5.4 and 6.3,

of irreducibles, and of course I preserves dimension.

we thus have character formulas of the following form for each p:H + ~ :

i

dixl(Ti)(x) = Xl(~)(x ) = ~ d i XI(Ti)(x ) i

for x E N 0.

The conclusion follows.

With these preliminaries, it is now an easy matter to construct the G-maps B and prove Theorem 0.3. Let
a

ln,r:GL(n,kr) + GL(n,k) , be the natural inclusion. an element Regarding ln,r

r = q ,

as an element of Rk(GL(n,kr)), we obtain

~n,r = ~l(In,r) EKGB(G,GL(n,kr)), that is, a G-map Bn,r:B(G,GL(n,kr)) ÷ K(G). and I, these maps for fixed r specify a map ~r:BJ~(G,kr) of Hopf ring G-spaces. + K(G) By the additivity and multiplicity of

Moreover, up to G-homotopy, Br is the restriction of Br+ I.

63

By Propositions 4.2, 4.3, and 5.2, there result compatible maps 8r:K(G,k r) + K(G) of Hopf ring G-spaces. Since K(G,k) is clearly G-equivalent to the telescope of the Further, 8~ r is G-

K(G,kr) , there results a Hopf ring G-map p:K(G,k) + K(G). homotopic to l~r~.

The same assertions hold in the orthogonal case. ~n,r = ~ ( I n , r ) E KBGL(n'kr) and thus 8 r : B ~ f ( k r) + K

Taking G = e, we have and 8:K(k) + K.

By Quillen [28], this map and, when q ~ 2, its orthogonal By [ii, III §7], the symplectic

analog are homology isomorphisms away from q. analog and the orthogonal analog

with q = 2 are also homology isomorphisms away

from q (Proposition 6.1 leading to a cleaner construction but no substantive change of argument in the latter case). K(G,k) H = x K(k) and K(G) H ~ × K, where the i i products are indexed on the irreducible k-representations {V.} of H and their l images under %. ~H = × B, and similarly in the orthogonal case. i Clearly the claim will complete the proof of Theorem 0.3. By Propositions 5.5 and 6.4, we have Bn,rH = vH(Bn'r) = ix (~ ~ ~x)vP(in,r) , We claim that Consider H C G. We have

where p runs through R+(H,GL(n,kr)).

When k r contains the m(H) th roots of unity,

Rkr(H) E Rk(H) and V i ~ k xk W i for irreducible kr-representations W i of H. If r n. n° i p = ~ W i , then In,rP = ~ V i i and thus the ith component of vp (in,r) is just the composite of sequence {n i} lni,r and the projection GL(n,kr)P + GL(ni,kr). Thinking of p as a

and collecting terms, we see that this implies

B H = x Br: x B ~ ( k r ) r i

÷ x K, i

and our claim follows from the uniqueness clause of the nonequivariant version of Proposition 4.2.

§7.

The equivariant

A G-map

E:E ÷ X

is a G-fibration if each fibre

if it satisfies

the G-CHP.

It is a (linear)

spherical G-fibration compaetification of x.

E-l(x)

has the Gx-homotopy

type of the l-point

of a real representation

of Gx, where G x is the isotropy subgroup

(We require the basepoints There is a fibrewise In particular,

of fibres to specify a fibrewise cofibration X ÷ E smash product between spherical G-fibrations = EVE is the fibrewise (as

[22, 5.2]).

in [22, 5.6]).

~^~

suspension of E by V, X x Sv ÷ X. are

where V is a real representation Two spherical G-fibrations fibre G-homotopy equivalent

of G and V is the trivial G-fibration are stably equivalent if

E and E'

zV E and EVE ,

for some V. group of fibre

For a compact G-space X, define SphG(X) to be the Grothendieck G-homotopy equivalence classes of spherical G-fibrations of stable equivalence over X.

The elements of

SphG(X) are formal differences fibrations. Via fibrewise

classes of spherical Greal G-vector bundles over X

l-point compactification,

give rise to spherical G-fibrations, fibrewise smash products.

and this procedure converts Whitney sums to

The resulting natural homomorphism JG:KOG(X) ÷ SphG(X)

is the real equivariant

J-homomorphism.

SphG(X) is represented

by a Hopf G-space [34] and also

Sph(G) and JG is induced by a Hopf G-map j:K(G) + Sph(G);

see Waner

[12], where it will be shown that j is actually an infinite loop G-map. Theorem 0.4 asserts keSJG(~kE that, for any E E KOG(X), there exists e > 0 such that

- E) = 0, where k is prime to the order of G and s is minimal such that

k s E ±i modulo the order of G. A necessary representations multiplying condition for of the fibres of JG(~ - E) = 0 ~ and E is that the isotropy group The point of compare

be stably equivalent.

by s is that this suffices

to arrange this necessary condition;

[I0, 9.7] and [13, §2]. inequivalent canonical for all e.

It is easy to see that ke~kv and keV can he stably For example, representation. this happens when G = Z5, k = 3, and V is the Here there is a degree k G-map V + ~3V, so

2-dimenslonal

65

this example also shows that one obvious generalization fails. However,

because there can be other linear combinations fibre representations,

Dold theorem mod k ki ~ si(~ ~i - ~i ) Adams

with stably equivalent conjecture discussion.

our version of the equivariant See McClure

fails to detect all of the kernel of JG"

[25] for further

Theorem 0.4 was proven for one and two dimensional Hauschild and Waner appropriate [13], their procedure

G-vector bundles by

being to formulate and prove the of Adams [2]. It

equivariant

Dold theorem mod k and apply the arguments

follows that if f:Y + X is a finite G-cover and T:KOG(Y) ÷ KOG(X) is the associated transfer, then Theorem 0.4 holds for T(~) when ~ is one or two dimensional. the relation ~kT(~) = T~k(~) in KOG(X)[I/k], The

implication requires

which is proven

in [13, §5] and amounts loop G-map.

to the verification

that, away from k, ~k is an infinite [28, 2.3] to deduce

For the rest, one can either use Quillen's argument

the implication JG T = TJG'

from the explicit proofs in [13] or one can use the general relation [25] or deduced from the fact that

which can either be verified directly

j is an infinite loop G-map. Next, consider a finite group ~ and a principal a real representation construction elements V of n, E x V ÷ X (G,H)-bundle E ÷ X = E/H. This The For

is a G-vector bundle over X. RO(E) ÷ KO~(X).

in V and induces a homomorphism

in the image of any such homomorphism group to E.

are the elements of KOG(X) with a [28,§2], Theorem 0.4 holds

reduction of their structural for all such elements.

As in Quillen

To see this, let A be a subgroup of E, let Y = E/A, and let If V is induced from a representation W of V,

f:Y ÷ X be the associated G-cover. then

T(E ×A W) = E ×~V,

as is easily verified from the usual fibrewise description

~(E XAW) x = f(y)® = x (E x A W)y

86

Since every element of RO(~) is an integral linear combination induced from one and two dimensional implies the assertion. representations

of representations [28, 2.4], this

of subgroups

Thus it remains to reduce the general case of Theorem 0.4 to the finite structural equivariant say k = q. group case. setting. Again, Since Quillen's arguments = ~jk, in [28, §i] generalize to the

~j~k

we may as well assume that k is a prime, of the orthogonal case of Theorem 0.3 for

(Since we are in possession

q = 2 as well as for q > 2, we need not follow Quillen in handling the real case for q = 2 by reducing it to the complex case.)

The Brauer lift map B of Theorem 0.3 clearly factors through a G-map B:~O(k,G) such that each BH is a mod n homology elementary + BO(G) = KO(G) for n prime to q. This more

isomorphism

version of Theorem 0.3 is appropriate here since passage to group to bundle theory. Since ~ O(k,G) is the to

completion would obscure the relationship

telescope of copies of B O(k,G) under translations the telescope of the B ~ ( k r , G ) , structural ~O(k,G). We need some general facts about "G-connected Y be a based G-space. such that Then there is a G-connected is an isomorphism component of yH.

and B O(k,G)

is G-equlvalent

an element of KOG(X) admits a reduction of its G-map X + KO(G) lifts to

group to a finite group if its classifying

covers";

see [8] for proofs.

Let

G-space Y0 and a G-map Y:Y0 ÷ Y and q > 0. Thus, up

y,:~q(y~) ÷ ~q(yH)

for all H C G

to homotopy, natural.

Y~ is the basepoint

Y0 is a functor of Y and ¥ is

If Y is a Hopf G-space, G-space X,

then Y0 is a Hopf G-space and y is a Hopf G-map.

For a G-connected

y,:[X,Y0] G + [X,Y] G is a bijection. For a G-connected Hopf G-space Z, let Z[i/q] be the telescope of a sequence of be the inclusion of

copies of Z under the qth power map Z ÷ Z and let i:Z ÷ Z[I/q] the first copy of Z. Clearly Z[i/q] H is a localization

of Z H away from q.

67

Now consider O(k,G) 0 where s is minimal ~ull G-homotopic

the sequence of maps of G-connected

G-spaces i ~ Sph(G)0[i/q] ,

B0 s(~q-l)0 J0 ~KO(G) 0 • KO(G) 0 ~ Sph(G) 0 such that qS E ± 1

modulo the order of G.

The composite is for any G-

since it is so when precomposed with

f:X ÷ ~ ~<k,G) 0

map f defined on a compact G-connected here.

G-space X and since there are no lim I terms (see

The homotopy groups of the fixed point spaces Sph(G)~ are all finite obstruction theory

[12]), and it follows from Bredon's equivariant absence of lim I terms that

[6] and the

~0:[KO(G), is a bijection. follows. Therefore

Sph(G)0[I/q]] G ÷ [ ~ ( k , G )

0 Sph(G)0[I/q]] G and Theorem 0.4

iJ0s(~q - i) 0 is also null G-homotopic

§8.

The equivariant

algebraic K-theory of finite fields

We revert to the notation of section 6; k r denotes elements, where q is a prime not dividing algebraic closure of k r.

the field with r = qa

the order of G, and k denotes the

Let I denote both the extension of scalars functors spaces induced by these functors spaces.

induced by the inclusion of k r in k and the maps of K-theory functors (as in section 4). Similarly,

let ~r denote both the permutative

induced by the Frobenius automorphism

and the maps they induce on K-theory

We are heading towards the proof of Theorem 0.5, but we shall first prove the following analog.

Theorem 8.1.

The following are G-fibration

sequences:

K(kr,G ) and KO(k ,G)

I

~ K(k,G)

~r-1~ K(k,G)

l ~KO(k,G)

~r-l~Ko(k,G).

68

That is, K(kr,G)

is G-equivalent where

to the homotopy

theoretical

fibre FCr(G) of

the G-map Cr_ I = ~(¢r,x)A,

~ and X and A

are the standard loop product and is the diagonal map. Since Crl = I on

inverse map on K(k,G) = ~BB J ~ ( k , G ) the category level, the maps K(kr,G) + K(k,G). natural map struction~ equivalence Therefore ~r I

and i restrict to precisely

the same map of the

factors canonically as the composite y:K(kr,G ) + F~r(G).

F~r(G) + K(k,G)

and a lifting

With this con-

y is a Hopf G-map since ~r and I are. by verifying that yH is an equivalence of homotopy

We shall prove that y is a Gfor each subgroup H of G. Note

for this purpose that the construction

fibres (as the fibre over the commutes with passage to fixed

basepoint of the associated mapping path fibration) points and with products. component Of course, homotopy

fibres depend only on the basepoint

of the base space. Since J ~ ( F , G ) H = ~.~(F,H) H if char F is prime to the order of G, We shall

Thus fix H.

by Remark 2.7, we may quote the results of section 3 with G replaced by H. first analyze (~r)H and I H on the level of permutative categories. Let

s = SoU s0 II s+ii s
and
s r ° s0r r

be sets of representatives S_ and Sr

for the irreducible Propositions

representations

of H over k and kr~

being empty if q = 2. of permutative

3. I and 3.4 give the following

equivalences

categories.

(1)

~(k,G)

H

~ V~

x S

~,,~(k)

(2) O ( k , G ) H = [ x ~.~(k)] V ~S 0 (3) ~ (kr,G)H = × sr ~ ( k U E

x [ x O (k)] x [ x V ~S+ V E S_ u)

r

U E S0

r

U

S

U E Sr _

U ~ S±

r

69

Here, in (3) and (4), we have used the fact that any finite division ring is a field to write Hom k [G](U,U) = k u r r ~r for some u depending on U.

Clearly

acts as a permutation ~r

on the set of irreducible acts as a permutation

representations

of

H over k, hence we can arrange that representatives. invariant

on the set S of r_

We let [V] denote the orbit of V and let ~ denote the ~ W. E[v] TI~ r = For a ~r-invariant

representation

subset T of S, we let

w

{Vl[vl

c T}.
~r, and the

The duality operator on representations subsets

commutes with the action of r

S+, S_, and SOIl S ~ of S a r e

~ -invariant.

M o r e o v e r , we can w r i t e

soll so
as a ~r-set, where

-

oll 0 lis

T 0 = {vivc S O

and V* / [V]}

SO = { V * I v ~ and

0} = {VIV ES 0

and

V f [VI}

S± ={viv c S O

or V ¢S 0

and

V ~ [V]}.

Let

t: ~ ( k )

÷ ~(k)

be the functor which is the identity on objects and inverse. In analogy with topological K( r)H is

sends a nonsingular

matrix to its transpose ~-r = r t = t r.

theory, we agree to let precisely analogous

The following analysis of in Corollary 5.7.

to the analysis of

(~r)H

Proposition

8.2.

The following diagram of functors commutes:

~ (k,G) H (~r)H

x

,1~" ( k )

VES

~.~(k,G) H

=

×

@~:(k)

I

(

×

,r)~

v

VES

where m permutes

factors as ~r permutes

the indexing representations.

The same

conclusion holds in the orthogonal

case except that, on the general linear part

70

x ~(k), V c SO

(x ~r)m V

must be replaced by the functor

"(x }* r) ,, which sends the V and via ~-r if ~r(v) ¢S~. to k n O ~r(v) V

V th factor to the Proof.

~r(v)thfactor via ~r if ~r(v) c S O

The equivalence carries a k-space k n in the V th factor ~ ' ( k ) kn®

and carries a nonsingular matrix f to f O 1, while ~r carries and f ~ 1 to ~r(f) ~ i.

V to k n ~

This is the same as first shifting to the ~r(v)th factor, The argument in the orthogonal

next applying ~r, and then applying the equivalence.

case is the same except that, when V ~ SO, the equivalence carries k n to (kn ~ V) (kn O V*) with the hyperbolic G-form and carries f to (f ~ i) G) ((f-l)t ~ if ~r(v) i). Here

~ SO, then ~r(v*) = cr(v)* c S O and, after application of ~r, the two

summands appear in reverse order in the factor of ~.~(k) indexed on ~r(v*).

Clearly the product restrictions ( x

(x ~r)m breaks up into the product over V ~r)m, and similarly in the orthogonal case.

E S/~ r

of its

w c Iv]

To compute lH, we need the following general observation.

Lemma 8.3.

Let L be a Galois extension of a field K with Galois group ff and let M {ml,...,mn}. For o oH, define a K-map o:M ÷ M by

be an L-space with basis 0[~ yjmj) = ~ o(yj)mj L 0
K

for yj E L. Z (i 0 o)m

Then the composite L 0
K

(*)

M

~
o c H

M

Z ~ >
o

~
c ~

M

is an isomorphism of L-spaces, where ~ is the action map. regarded as an L-map M ÷ M via {m.} J and

If f c GL(n,L) is is the automorphism

o:GL(n,L) + GL(n,L)

induced entrywise by a, then the following diagram commutes.

L~ (l~f) IK

M

1 ~ °~-L~I M

~

~- M [ of

L O K

M

IO°mL~ K

M

~

---M

In particular, if a finite group H acts on M via

p:H ÷ GL(n,L) and oM

denotes M

71

with action o0, then (*) is an isomorphism L × K Proof. Clearly M m ~ o ~

of L[H]-modules oM . in general if it is so when M = L

(*) will be an isomorphism of dimensions, the trace

and here, by comparison monomorphism. b(y,y') form. Consider

it suffices

to show that (*) is a Give L the trace form [ L the sum

tr:L + K,

tr(y) = ~ oy.

= tr(yy'),

give L ~ k L the tensor product

form, and give

A simple calculation

shows that (*) is an isometry and thus a monomorphism.

Now consider 14.6],

the extension of scalars homomorphism Moreover,

I is a split monomorphism. ~ ( H ) ~r

I:R k (H) ÷ Rk(H). r its image is clearly contained

By [31, in the

subgroup

of representations

invariant

under #r, and S/~ r is a basis for V over k is a summand of t(U) for again,

the latter.

Since any irreducible representation

representation

some irreducible

U over kr, by [31,14.6] r

I:R k (H) + Rk(H)~ r is an isomorphism. In view of the natural isomorphisms

k~ r

Hom k (U,kr) ~ Hom k (U,k) ~ HOmk(k ~ U,k), r r r The following result analyzes the indexing

i commutes with the duality operators. sets and the fields k ru sets in (i) and (2). In particular,

in (3) and (4) in terms of the action of ~r on the indexing it shows that i restricts to an isomorphism

1:ROkr(H) ÷ ROk(H)~ r.

Lemma 8.4. bijection

S r and S can be so chosen that extension of scalars S r ÷ S/~ r which restricts to bijections

specifies

a

Sr+ ÷ S+/~ r

,

S r ÷ S_/ r

r -S 0 ÷ S0/~r

,

r * --* r (S O ) ÷ S0/~

72

and r S± + s e / r .

Moreover,

if U~ Sr with

1(U) = V

and if HOmkr[H](U,U)

= kru , then u is minimal

such that #rU(v) = V and is thus the number of elements in the orbit [V]. Further, t * = r if U ~ S~ and t is minimal such that V r (V), then u = 2t and we may therefore i choose [V] ~3 S 0 to be { r (V) I 0 ( i < t}. Proof. Fix U E S r. To simplify notation, over L. write K = k r and L = kru. We may since

regard U as a representation HomL[H](U,U ) ~ L.

As such, it is absolutely is irreducible over k.

irreducible

Therefore V = k ~ L U

Applying the functor

k ~ L (?) to the isomorphism u-I i

L~KU

.

i
i =0

+r U

of Lemma 8.3, we obtain an isomorphism of k[H]-modules u-I i ~ ~r (V) = m~, i =0 i of V in {\$r (V) I 0 ~ i < u}. ~ Regard U as L ~ L U C

k ~U

where m is the multiplicity

By our observations V. If V admits a G-

about I above, m = 1 and u is as stated. form b0' then nondegeneracy

implies the existence

of elements Uo,U ~ ~ U such that fb0(u0,u~) ~ 0 and It

b0(uO,u ~) ¢ 0. define b0(u,u')

Choose a K-linear functional = fb0(u,u') for u,u' ~U.

f:k + K such that

Then b 0 is a K-bilinear

G-form on U.

is nondegenerate

since it is not identically

zero and U is irreducible.

Any other G-

form on U is specified by bo(du,u' ) = fbo(du,u') for some d ~ L. Visibly, if b0 = f(dbo(u,u')) then all G-forms on Let t be
V, u

is symmetric or skew symmetric,

U are symmetric or skew symmetric. minimal such that V*
=

Now suppose that V c S O and V* ~ IV].

Certainly 0 < t < u. Since ~ r2t (V) = V ** = t-i i divides 2t. This implies u = 2t. Let W = [ sr (V). Then ~ = W ~ W i = 0 admits both the symmetric hyperbolic G-form and the non-symmetric G-form

~rt(v).

bx((W,~), (w',~')) =~(w') + x ~'(w)
for any x~ k, x # 0 and x # 1. Regard U as K~K U ~V (and observe that these

73

forms are not L-bilinear on U).

Choosing appropriate K-linear functionals k ÷ K, we V to G-forms on U with the

can compress the restrictions to U of these G-forms on same symmetry properties. Thus r U c Se.

The rest should be clear.

Proposition 8.5. morphism:

The following diagram of functors commutes up to natural iso-

~ Z ( k r ,G) H U ~S r ~(kru)

I

×(×,r
U i u-i × x .~(k) U ¢ S r i=0

I)

.~(k,G) H

~

The same conclusion holds in the orthogonal case except that, on the unitary factor ~ (k u ) indexed on U E S~, the right arrow must be replaced by r t-I i t-i x r I: ~ ( k u) + x Jk~(k), i=O r i=O where i is interpreted as neglect of form followed by extension of scalars. Proof. 0 ~ i < u, We regard S as the union of its orbits [V] and so index it on U ¢ Sr and i (U,i) corresponding to r (V) if I(U) = V . Fix U and again write The top equivalence carries the L-space Ln in the U th and carries f ~GL(n,L) to f ~ i. The functor iH carries

K = k r and L = kru. factor,~(L)
L n ( ~L U t o

to L n ~ L U

u-i k ~ K (Ln ~ L U) - - (k ~ L Ln) ~ k (k ~ K U) - - I (k ~ i=0 Here the first isomorphism is specified by x® (ej ~ u) ÷ ~ (x ~ ej) ~ (i ~ u)

i Ln) ~ r (V).

J
in terms of the canonical basis

J
{el,...,e n} of L n. Note that this is highly non-

invariant: we are using this basis to identify both sides with the sum of n copies of k ~ K U, and scalars from L in L n ~ L the isomorphism. previous proof. U must be pushed over to U before evaluating

The second isomorphism is derived from that displayed in the Writing f(ej) = ~ fjkek , f j k E L, and calculating, we see that

74

i O (f ~ I) corresponds under the composite isomorphism above to i (I ~ ~r f) ~ i. This implies the first statement. On the general linear, i orthogonal, and symplectic factors of O(kr,G)H , the orthogonal case is an obvious elaboration. Consider r U ~ Se with u = 2t. Then \$ rt specifies the involution on L the only Hermitian form

determined by a symmetric G-form on U.

Up to isomorphism,

on L n is the standard one [14, Thm 2.8], namely t ~ yj~r (y~).

b(~ yjej, ~ y ~ e j ) = J t Thus if f cU(n,L),

then

f - I = ~r ( f t ) ,

where f t i s the t r a n s p o s e of f .

Now

i
[vl ~ SO = {~r (V) I 0 ~ i < t}, copies of ~ ( k ) in O(k,G) H. and these are the elements of [V] which index i+t i We have ~r (V) = ~r (V)* and

~r

i+t

f = (~r

if_l)t.

W r i t i n g the r i g h t s i d e of the isomorphism above as
~r i L n) ~ (V)] ~) [(k ~ e n) ~ ~ri(v)*], i ~r (v)th

t-i

I [(k ~ i=O

we see that the last assertion follows from the specification of the funetor ~ ( k ) + O(k,G) H in the proof of Proposition 3.4.

Propositions

8.2 and 8.5 imply that the sequence
H

K(kr,G)H

I ~K(k,G)H

(~r)H-l_ K(k,G)H

breaks up into the product of sequences, i cr I K(k u ) r u-i × i=O

one for each U ¢ S r, [(~ cr)~]-i u-i × K(k) , i=O

K(k)

where m is the cyclic permutation of factors. orthogonal case except that, for i ,r I KU(k r) ~ t-i × i=O K(k) r U c S~,

A similar conclusion holds in the

the resulting sequence is of the form t-i × ,r)m]-i i=O r

[(,-r x

t-i x K(k) . i=O Using the

For Uc S r, let F~r(u) denote the homotopy fibre of the right-hand map.

commutation of homotopy fibres with passage to fixed point sets and with products,

75

we see that our original lifting 7:K(kr,G) ÷ F~r(G) of i restricts to a lift i y:K(k u ) + F#r(u) of x ~r I for each U. We thus have the following commutative r i diagrams, where i 0 is the inclusion of the 0 th factor and the commutativity of the right-hand square is a trivial computation:
u

~r K(k u) r I • K(k) - 1 ~ K(k)

F~r(u) r r For U ~ SO, S+, or Sr,_ K,KO, o r KSp t h r o u g h o u t .

u-I x i=O

K(k)

1'
x ~r i -r ~

[(~ ~r)m]-i ,

u-i x i=O

K(k)

1

i0

we have a precisely analogous diagram with K replaced by For U ~ s,,r the analogous diagram has the bottom right

map displayed above and the top row t - I K(k) .

KU(k u ) r

I

~ K(k)

By Quillen [29] in the general linear case and Fiedorowicz and Priddy [ii] in the orthogonal, symplectic, and unitary cases, the top rows are all fibration sequences. To prove Theorem 8.1, it suffices to show that the Hopf maps ~0' which is clear, and on it suffices ~ n of the basepolnt components the right-hand squares

induce isomorphisms on f o r n > O.

To show t h e l a t t e r ,

to check that

induce isomorphisms on kernels and cokernels of the maps of higher homotopy groups induced by the horizontal arrows. The required calculations are immediate from the The

following result, which determines the relevant maps of homotopy groups.

homotopy groups of K(k), KO(k), and KSp(k) are tabulated in [ii, p.246], the nontrivial groups being ~2n_iK(k) ~ ~4n_lKO(k) ~ ~4n_lKSp(k) ~ P # ~ q and, if q # 2, ~Sn+iKO(k) ~ ~Sn+2KO(k) ~ ~8n+5KSp(k) ~ ~8n+6KSp(k) ~ Z 2. (Note that, when q = 2, our space KO(k) has basepoint component equivalent to the space Zp=

rob ~ ( k )

of [ii]; see [11,11.7.6].)

Recall that

Zp~ ~ Z[I/q]/Z ~ Q/Z(q) . p ¢ q

76

Lemma 8.6. ~2n_iK(k).

The maps ~r and ~-r on K(k) induce multiplication

by rn and (-r) n on by r 2n on

The maps ~r on KO(k) and KSp(k) induce multiplication

~4n_iKO(k) and ~4n_iKSp(k) Z2• Proof.

and induce the identity on the homotopy groups equal to

By [23, VIII, 2.9] and [II, p.170-175],

Brauer lift gives rise to the

following commutative diagram of completions of basepoint components away from q, the map 8 being an equivalence:

~(k)0[i/q] S [ =

~r

; ~(k)0[i/q] = [ B

Ko[1/q]
The behavior of [23, V.2.9]). ~r

^

~r

~

Ko[1/q]

The same assertion holds with r replaced by -r and with K replaced by KO or KSp. on homotopy groups was computed by Adams [i, 5.2] (see also

By Bousfield and Kan [6, p.153], we have a natural exact sequence

0 + Ext I I Z .,~n X) + ~nX[I/q] + Hom I I Z ~,~n_l X) + 0 P#q P P#q P for simple spaces X. give Ext groups For X = KO(k) 0 amd X = KSp(k) 0, the homotopy groups
^

~n X = Z 2 since

Z 2 = WnX[I/q],

and these groups are mapped isomorphically ~r.

their images under 8 are mapped isomorphically by homotopy groups W2n_l X = [ Z ~

In the remaining cases, the

(n even for KO(k) or KSp(k)) give Hom groups ~r and ~-r

Z[I/q] = W2nX[I/q],

and +~#~ndP~ -r map these groups by rn or r -n since The conclusions follow.

so map their images under B.

We have now proven Theorem 8.1, and we turn to the proof of Theorem 0.5. Consider the following diagram:

F~r(c)

: K(G, k)

~r-l~ K(G,k)

1
F*r (G) ~

1
K(G) ~r-l>

B
K(G)

77

As noted in section 6, the right square is G-homotopy commutative. exists a lift equivalence 6.

Thus there ~ is a G-

In view of Theorem 8.1, it suffices to prove that

to conclude the complex case of Theorem 0.5. 6H is an equivalence.

For this, it suffices to 5.7 and Proposition for each

show that each 6.2, U E sr: 6H

In view of Corollary

restricts

to a llft ~ in the following diagram of fibrations

u-1
F~r(u) ~ x K(k)

[(~ +r)~]-1 u-1
~- x K(k)

I
F~r(u) diagram to one of the form u F~ r

u-1

°I

i

[(~ ,r)~]-1 u-1
• x i=0

°L
K However, com-

I- x K i=0

Because of the odd and even degrees in which the nontrivial homotopy groups occur on the right, we cannot conclude directly that ~ is an equivalence.

paring top rows to top rows and bottom rows to bottom rows, we can convert the

~r = K(k u)

u - 1 K(k)

~

Ir i
8
u

- K(k)

[
8
~. if

F~ r

~ K

~r u _

I

1

~

K

Here

6

is an equivalence

by Quillen's

results

[29].

The proof of the orthogonal highly non-trlvlal point.

case of Theorem 0.5 is exactly the same, modulo one there is a unique lift this is not true with K

In the diagram just given, by -r u. However, ~

This remains true with r u replaced

replaced by KSp or KO and, in the latter case, it is wrongly chosen. if it is a Hopf map. proving the existence Fiedorowlcz and Priddy

will fail to be an equivalence ~

[ii] proved that

is an equivalence

May [23] proved that there is a lift which is a Hopf map by of a lift which is an infinite loop map (by an argument

involving pulling back Bott perlodicty along the equivalence B:KO(k)o[I/q] ÷ KO0[I/q] and analyzing the relationship between space and spectrum In our situation,

level periodicity).

Presumably a more direct proof is possible.

78

we must prove that the original

lift ~H

~:FO~r(G)

÷ Four(G)

can be chosen as a Hopf

G-map in order to ensure that each

and thus each

r ~:FO~r(u) ÷ Four(u), U E S+,

is a Hopf map (passage to the last diagram above with K replaced by KO presenting no difficulty). Again, while a more direct proof should be possible, the only argument just cited;

we know is the equivariant see [12].

infinite loop space version of May's argument

79

Bibliography

I. 2. 3. 4.

Vector fields on spheres. On the groups J(X) - I. Lectures on Lie groups.

Annals of Math. Topology 2(1963), 1969.

75(1962), 181-195.

603-632.

Benjamin.

M.F. Atiyah and G.B. Segal. Geom. 3(1969), 1-18.

Equivariant

K-theory and completion.

J. Diff.

5. 6.

M. Benioff.

PhD Thesis.

Chicago,

In preparation. completions, and localiza1972.

A.K. Bousfield and D.M. Fan. tions.

Homotopy limits,

Lecture Notes in Mathematics Equivariant 1967. cohomology

Vol. 304. Springer.

7.

G. Bredon.

theories.

Lecture Notes in Mathematics

Vol 34. Springer. 8. J. Caruso,

F.R. Cohen, J.P. May, and L.R. Taylor. theorem. To appear. The equivariant

James maps,

Segal maps, and

the Kahn-Priddy 9.

J. Caruso and J.P. May. To appear.

splitting and Kahn-Priddy

theorems.

i0.

T. tom Dieck. in Mathematics

Transformation Vol. 766.

groups and representation 1979.

theory.

Lecture Notes

Springer.

11.

Z. Fiedorowicz

and S. Priddy.

Homology of classical groups over finite fields Lecture Notes in Mathematics

and their associated Vol 658. Springer. 12. H. Hauschild, To appear. 13.

infinite loop spaces.

1978. Equivariant infinite loop space theory.

J.P. May, and S. Waner.

H. Hauschild and S. Waner. jecture. Preprint.

Equivariant

Dold theorem mod k and the Adams con-

14. 15.

I. Kaplansky. M. Karoubi, Mathematics

Linear algebra and geometry. Periodicite

Allyn and Bacon.

1969.

de la K-theorie hermitienne. 1973, 301-411.

Lecture Notes in

Vol 343. Springer. AG-structure

16.

K. Kawakubo. JG(X).

of G-vector bundles and groups KOG(X), KSPG(X) , and

Preprint. Equivariant bundles. Preprint.

17.

R.K. Lashof.

80

18.

R.K. Lashof and M. Rothenberg. 32(1978), 211-265.

G-smoothing theory.

Proc. Symp. Pure Math.

Amer. Math. Soc. Ann. Sc.

19.

J.L. Loday.

K-theorie algebriques et representations des groupes.

Ec. Norm. Sup. Fasc. 3, t.9. 1976. 20. J.L. Loday. Higher Witt groups: a survey. Lecture Notes in Mathematics

Vol. 551. Springer. 1976, 311-335. 21. J.P. May. E spaces, group completions, and permutative categories.

London Math. Soc. Lecture Note Series II, 1974, 61-94. 22. J.P. May. Classifying spaces and fibrations. Memoirs Amer. Math. Soc.

No. 155. 1975. 23. J.P. May (with contributions by F. Quinn, N. Ray, and J. Tornehave). ring spaces and E Springer. 1977. 24. J.P. May. Pairings of categories and spectra. J. Pure and Applied Algebra. ring spectra. E

Lecture Notes in Mathematics Vol. 577.

19 (1980), 299-346. 25. 26. J. McClure. The groups JOG(X). To appear.

D. McDuff and G. Segal. theorem.

Homology fibrations and the "group completion"

Invent. Math. 31 (1976), 279-284. On the group completion of a simplicial monoid. The Adams conjecture. Topology 10(1970), 67-80. Preprint.

27. 28. 29.

D.G. Quillen. D.G. Quillen. D.G.Quillen.

On the cohomology and K-theory of the general linear groups Annals of Math 96(1972), 552-586. Publ. Math. IHES, 34(1968), 129-151. Deuxieme

over a finite field. 30. 31. G.B. Segal. J.P. Serre. edition. 32.

Equivariant K-theory.

Representations lineaires des groupes finis. 1967.

Hermann.

C.T.C. Wall. Vol. 343.

Foundations of algebraic L-theory.

Lecture Notes in Mathematics

Springer 1973, 166-200. Equivariant homotopy theory and Milnor's theorem. Trans. Amer.Math.

33.

S. Waner.

Soc. 258(1980), 351-368. 34. S. Waner. Equivariant classifying spaces fibrations. Trans. Amer. Math. Soc.

258(1980), 385-405.