7opolog~ Vol 27, No. I. pp. I-6. Prmted m Great Bntain.

1988

0040 9383,X8 S300+ 00 Pcrgamon Press plc

A GENERALIZATION

OF THE ATIYAH-SEGAL THEOREM

COMPLETION

J. F. ADAMS, J.-P. HAEBERLY, JACKOWSKI J. P. MAY S. and

(Received 9 June 1986)

$1. INTRODUCTION

WE SHALL give a geodesic path from the equivariant Bott periodicity theorem to the generalized completion theorem in equivariant K-theory. Let G be a compact Lie group. Our G-spaces are understood to be G-C W complexes and we let KE denote the progroup valued G-cohomology theory specified by KnG = {KnG(X.)}, (X) where X, runs over the finite subcomplexes of X. For a subgroup H of G, we have a restriction homomorphism I~. R(G)-+R(H) and we let 1; be its kernel. (Subgroups are understood to be G* closed.) A set # of subgroups of G closed under subconjugacy is called a family. We let (Kg);, the y-adic completion of Kg, denote the progroup valued G-cohomology theory specified by K:(X)? = {K;(X,)/JKG(X,)}, where J runs over the finite products of ideals Z$ with HEY. (The relevant information about progroups is summarized in [2, 923.)
THEOREM1.1. If a G-mapf: X+

HE%, then (j*)>:

(KZ_Y)>+(KzX)>

Y restricts to a homotopy equivalencef u: XH+ YHfor each is an isomorphism.

The same assertion holds with Ko and R(G) replaced by KOo and RO(G). Theorem 1.l was first conjectured in 1976 [S] and was first proven, independently, by two of us in 1983 [7,9]. The case 2 = {11 is the Atiyah-Segal completion theorem of [4], and the proof in [7] follows [4] in outline. The proof in [9] contained the key idea of proceeding by direct induction rather than giving unitary groups and tori a privileged role. Our variant of this idea exploits an argument due to Carlsson [5] in cohomotopy to obtain an immediate reduction to quotation of Bott periodicity for the equivariant K-theory of G-spheres. It is to be emphasized that our argument, like that of [9], includes a new proof of the original Atiyah-Segal theorem. We use (1.1) to compute equivariant K-theory characteristic classes in 52. We prove (1.1) in 53 and make a few remarks on it in 45. In $4, we use (1.1) to prove the following mixed localization and completion theorem. Its cohomotopy analog was the main result of our paper [2], and more discussion of such invariance theorems may be found there. Pro-R(G)modules are localized termwise, S- {M,} = {S-M,}.
THEOREM 1.2. Let S c R(G) be a multiplicative set, let I c R(G) be an ideal, and define

Z=u{Supp(P)IPnS=b

and PzI}.

J. F. Adams, J.-P. Haeberly, S. Jackowski and J. P. May

!f a G-map f: X-, Y restricts to a homotopy equivalence f: XH+ YH for all HE_@, then s-(f*);: S_KE(Y-)/ +s- K;(X); is an isomorphism. The same assertion holds with K, and R(G) replaced by KO, and RO(G). Here P runs over prime ideals of R(G) and Supp(P) not come from any Kc H. Segal shows that Supp(P) is the support is a single of P as defined conjugacy by

Segal [ 131: HE Supp(P) if P comes from H via the restriction (topologically) cyclic subgroups H. The theorem has content

map R(G)+R(H)

and P does class of

even when S = { I} and I = 0.

COROLLARY 1.3. If

a G-mapf X+ Y restricts to a homotopy equivalencef H: XH --+YH.for all (X) is an isomorphism.

cyclic subgroups H, then f *: KE (Y)+Ka For finite groups

G, this result goes back to [8].

$2. EQUIVARIANT

K-THEORY

OF CLASSIFYING

SPACES

The main motivation for Theorem actually equivalent to the theorem).

1.1 comes from the following

consequence

(which is

Let E% be a universal j-free G-space, so that (EJ?)~ is contractible if HE% and is empty if H&F. For any G-space X, the projection Ey xX+X restricts to a homotopy equivalence (Ey x X)+ X for each HE$, so (1.1) gives an isomorphism K E(X); +K E(E# x X)2. For a G-space Y, such as E% x X, all of whose isotropy groups are in 2, the groups of the inverse system KE( Y) are j-adically complete. For a finite G-C W complex X, the inverse system K:(X)> satisfies the Mittag-Leffler condition. These facts imply that the algebraic completion KE(X)a is isomorphic to the topological completion KE(Ey x X).
COROLLARY 2.1. If X is ajinite G-C Wcomplex, then the projection E# x X-+X induces an isomorphism Kg (X);+KE (Ey x X).

McClure has obtained interesting applications [12]. For example, he has shown that K:(X) is detected by the family of finite subgroups of G, so that a G-vector bundle is stably trivial if it is stably trivial when regarded as an H-vector bundle for each finite subgroup H of G. With X a point, the calculation of the K-theory original Atiyah-Segal completion theorem of classifying spaces in terms of completions specializes to a of representation

rings. There is an analogous specialization of (2.1) to the calculation of the KG-theory of classifying G-spaces. To see this, let lT be a normal subgroup of a compact Lie group r with quotient group G. The orbit projection q: Y+ Y/II of a T&free r-space is a kind of equivariant bundle, and there is a universal bundle E(lI; T)+B(lI; r) of this sort. Classically, r = G x IT, and q is then called a principal (G, lT)-bundle. For example, a smooth G-n-plane bundle has an associated principal (G, O(n))-bundle. The universal n-free r-space E(II; r) is just E#, where # = #(IT; f) is the family of subgroups A of r such that AnlT = e, and we have the following calculation of K,(B(II; r)).
COROLLARY 2.2.

The projection E(II; T)+(pt} R(r);,,; r,SK,(E(l3;

induces an isomorphism r))rK,(B(II; r)). of [2, (1.6)]. The case TJ = 1 is

Parenthetically, we insert the analogous specialization Segals original version of the Segal conjecture.

ATIYAH-SEGAL COROLLARY2.3. Let r be

COMPLETION

THEOREM

jinite. The projection E(Il; T)+{pt}

induces an isomorphism

~(r)j,,:

r,%;(~(n; I-)) z ~c:(B(II; r)).

The right-hand change of groups isomorphisms in (2.2) and (2.3) are standard; see [ 14, 2.11 for K-theory, [l, 5.33 for cohomotopy, and [ll, II@] for general theories.

43. PROOF OF THEOREM 1.1

Since (Ka)i is a progroup valued cohomology theory (as explained in [2]), exact sequences derived from cofibre sequences imply that (1.1) is equivalent to the following vanishing theorem.
THEOREM3.1. R E (X)2 is pro-zero for every based G-space such that XH is contractible for

each HE%, and similarly for (KOE)j.

We deduce this from a special case. Let U be the sum of countably many copies of each of a countable set of non-trivial representations Viof G such that each VP = 0 and some V/ # 0 if H is a proper subgroup of G. For K& we restrict attention to complex representations. For KOE, we restrict attention to Spin representations with dimension divisible by eight. Since the arguments are otherwise identical, we concentrate on the complex case from now on. Let Y be the colimit of the one-point compactifications S of the finite dimensional subrepresentations V of U. Since each v = 0, YG= So. If Vc W and ( W- V) #O, where W- V is the complement of V in W, then the inclusion S +Sw is null H-homotopic. It follows that YHis contractible and Y is H-contractible for H # G.
LEMMA 3.2. If # is proper (G&9), then R;(Y);

is pro-zero.

To deduce (3.1) from (3.2), we need to know the behavior of (R;F;)j with respect to restriction to subgroups. For H c G, let 2/H be the family of subgroups of H which are in $.
LEMMA 3.3. For any based G-space X,

@(G/H)+ A X,3 z RQ(X)&

Proof of (3.1). Since .j is a family, the equivariant Whitehead theorem shows that X is Hcontractible and thus I?t(X)is pro-zero for HE/. We must show that R:(X)? is pro-zero. To avoid triviality, we assume that G$%. Since the descending chain condition on subgroups allows induction, we may assume that Rs(X)^blH is pro-zero for all proper subgroups H c G. Since YG= So, we have a cofibre sequence

so + Y+ Y/SO. Taking smash products with X, we obtain a cofibre sequence X-bXA Y-+XA(Y/SO). It suffices to prove that Z?E(X A r); and RE(X A ( Y/So))3 are both pro-zero. We claim first that Z?E(WA Y)? is pro-zero for any G-C Wcomplex W. Since the zero skeleton W and the skeletal quotients Wn/ W- for n >O are wedges of G-spaces of the form (G/H) + A s and

J. F. Adams, J.-P. Haeberly, S. Jackowski and J. P. May A S and

since we may as well assume that Wis finite, we need only verify this for W=(G/H)+

thus, by suspension, for W= (G/H)+. Here (3.2) gives the conclusion if H = G and (3.3) and the H-contractibility of Y give the conclusion of H # G. We claim next that KE(X A Z)j is prozero for any G-C W complex Z, such as Y/So, such that Z is a point. Arguing subgroup as above, we need only verify this when Z =(G/H)+ for a proper holds by (3.3) and the induction hypothesis. The argument just given is an adaptation [S]. [3, 141, RE(S) is the free RE(S) -module E., restricts generated by the Segal conjecture Proofof(3.2). H, and here the conclusion steps in Carlssons proof of

of the preliminary

By Bott periodicity Moreover,

the Bott class &&g(S).

to the Bott class in Z?g(Sj for each H c G.

The Euler class x+R(G) = if g(S) is e*(i,), where e: So -+S is the evident inclusion. If H # G and VH# 0, then e is null H-homotopic and XV&. If Vc W, then the inclusion i: S+S is IAe, e: S+Sw-. Since i.,=i.,_,i.,, the homomorphism i*: Rg(SW)+r?E(S) is given by the formula

for x& F(S); that is, i* is multiplication by xw _ . We may view Z?,$(r); as the inverse limit in the category of progroups of the R E( Y)/JR E( I), where J runs over the finite products of ideals lg with HEY (see [2, $21). So it suffices to prove that Z?g( Y)/JI?E( Y) is pro-zero for each such J. This means that, for each V, there exists WI V such that i*: R~(SW)/JRE(Sw)-RE(S)/JRE(S) is zero. If J=Ig, that .. 1: and we choose W- V to be the sum of representations Wi such

Wz#O, then i* is zero since it is multiplication by xw, . . . x,,,,EJ. Since If E((G/H) + A X) z R;;(X) as pro-R(G)-modules, where R(G) acts on R g(X) through r$ R(G)+R(H), the following algebraic fact implies (3.3). LEMMA3.4. The ,/-adic and (f(H)-adic topologies coincide on R(H).

This follows from Segals results on R(G) [13, $31. The key point is the following observation about supports of prime ideals, which can be derived from [13, 3.5 or 3.73. LEMMA 3.5. 1fSc S is a support of P. Proofqf(3.4). If LEjlH, then rg(Z,) R (H) c I, since rfr$ = rf. Conversely, if KEY and if I = r$(l$)R(H), then I contains some product of ideals 1f with LE,$IH. To see this, note that some product of prime ideals Q 11 is contained in I and that any prime ideal QcR(H) contains I,, where SC H is a support of Q. So it suffices to check that S is in 9 when Q contains I. If P = (r$)- l(Q) c R(G), then S is a support of P and P contains I;. Since R(K) is finitely generated and thus integral over R(G)/Zg [ 13,3.2], P = (r$)- (P) for some prime ideal Pc R(K). Therefore P has a support Sc K. Since any two supports of a given prime ideal are conjugate [13, 3.73 and S is in ,/, S is in 3. Remark 3.6. The previous two lemmas remain valid for RO(G). The essential points are that any prime ideal Q of RO(G) is the restriction of a prime ideal P of R(G) and that if P is also the restriction of P #P, then P is the complex conjugate of P. H isa support ofaprime ideal QcR(H)andifP=(r$-(Q)cR(G), then

ATIYAH-SEGAL COMPLETION THEOREM

$4. PROOF OF THEOREM 1.2

Again, (1.2) is equivalent to the following vanishing theorem. THEOREM 4.1. S- Z? F(X); is pro-zero for every based G-space such that XH is contractible for each HEX, and similarly for S- (KOE),^.

As a matter of algebra [2,2.3], it suffices to prove that S,RE(X), is pro-zero for each prime ideal PC R(G) such that PnS = 4 and P 1 I. Here SF means localization at P: that is, the multiplicative set S, is the complement of P. Let H&upp(P) and let # be the family of subgroups of G subconjugate to H. By (3.1) R$( ~~ is pro-zero if Y K is contractible for all is KEY. Since P contains I $, it follows that Rz( Y)p^ pro-zero, and a fortiori S; RE( Y); is pro-zero. For X as in (4.1), XH is contractible but XK need not be contractible for KC H. However, we can embed X as a subcomplex of a G-C W complex Y such that YK= X for all K which contain a conjugate of H and Y is contractible for all other K. For example, we can take Y= X A l?g, where 3 is the family of subgroups of G which do not contain a conjugate of H and l?g is the unreduced suspension of E9 with one of the cone points as basepoint; the inclusion of So in ES induces the inclusion of X in Y. The classical localization theorem [14,4.1-j implies that S,KE(Y)+S,RE(X) is a pro-isomorphism; a fortiori S,iC~(Y)p^ +s, I?:(X), is a pro-isomorphism and SF R:(X); is pro-zero. In more detail, let {Y,} run over the finite subcomplexes of Y and let X,=Xn Y,. Then SF I?:( Y&S, R E(X,) is an isomorphism for each c1 induction up the finitely many cells by of Y, not in X, since these cells are of orbit type G/K with K&J and since S; R(K) =O for such K by C13.3.71.
Remark 4.2. Every collection x

of cyclic subgroups of G can be realized, generally in and PxI}

several ways, as
31C=u{Supp(P)jPnS=4

for some multiplicative set S and ideal I.

$5. REMARKS

We conclude with three unrelated comments.

Remark 5.1. Clearly (2.1) remains valid when X is finitely dominated but not necessarily finite. The extra generality is significant because locally linear compact topological Gmanifolds are compact G-ENRs and are therefore finitely dominated, but they need not have the homotopy types of finite G-C Wcomplexes (even stably). It is also possible to rework our proofs for general compact G-spaces such that each K,(X) is finitely R(G)-generated. Remark 5.2. Let % be a family in G. As one of us observed [9], (3.4) and Segals results in [13, p. 1213 imply that f contains all topologically cyclic subgroups of G itI the j-adic topology on R(G) is complete. McClure [12] has proven that $J contains all finite cyclic subgroups of G iff the #-adic topology is Hausdorff on all finitely generated R(G)-modules. (This fact is the key to the proof of his result cited in $2.) Remark 5.3. One might expect the Real K-theory case to work equally well. However, in the presence of a non-trivial involution on G, use of families other than (1) leads to difficulty. A Real G-space is the same thing as a G-space, where G is the semi-direct product of G and Z,

J.

F. Adams, J.-P. Haeberly, S. Jackowski and J. P. May

determined by the involution, and KRE is a cohomology theory on C-spaces. For a general subgroup L of G, we do not have a good description of KR ,?JG/L)+ A X); if L = I? for a Real subgroup H of G, then this is KR$(X). This suggests that we should restrict attention to Real families in G, but some of our arguments require use of actual families in G.

REFERENCES

1. J. F. ADAMS: Prerequisites for Carlssons lecture. Lecture Notes in Mathematics, Springer 1051 (1984) 483-532. 2. J. F. ADAMS, J.-P. HAEBERLY, JACKOWSKI J. P. MAY:A generalization of the Segal conjecture. Topology S. and 27 (1988) 7-21. 3. M. F. ATIYAH: Bott periodicity and the index ofelliptic operators. Quart. J. Math. Oxford Ser. 19(1983), 113-140. 4. M. F. ATIYAH and G. 9. SEGAL:Equivariant K-theory and completion. J. Di# Geom. 3 (1969), I-18. 5. G. CARLSSON: Equivariant stable homotopy and Segals Burnside ring conjecture. Ann. Math. 120 (1984).
Archiu der Math. 20 (1969). 136-143. 7. J.-P: HAEBERLY: Completions in equivariant K-theory. Thesis. University of Chicago (1983). 8. S. JACKOWSKI: Equivariant K-theory and cyclic subgroups. London Math. Sot. Lecrure Notes Cambridge Unio. Press 26 (1977), 76-92. 9. S. JACKOWSKI: Families of subgroups and completions. J. Pure Appt.Agebru 37 (1985), 167-179. 10. R. LASHOF:Equivariant bundles. Illinois J. Math. 26 (1982), 257-271. 11. L. G. LEWIS,Jr, J. P. MAY, and M. STEINBERGER: Equivariant stable homotopy theory. Lecture Notes in Mathematics. Springer, in press. 12. J. E. MCCLURE,Jr: Restriction maps in equivariant K-theory. Topology 25 (1986), 399409. 13. G. 9. SEGAL: The representation ring of a compact Lie group. Inst. Haures Etudes Sci. Publ. Math. 34 (1968), 113-128. 14. G. 9. SEGAL:Equivariant K-theory. Inst. Hautes Etudes Sci. Publ. Math. 34 (1968). 129-151. 189-224. 6. T. TOM DIECK: Faserbundel rnit Gruppenoperation.

D.P.M.M.S., 16 Mill Lane, Cambridge CB2 lSB, England. Dept. of Mathematics, University of Washington, Seattle, WA 98195, U.S.A. Dept. of Mathematics, University of Warsaw, Palac Kultury i Nauki IXp, 00-901 Warsaw, Poland. Dept. of Mathematics, University of Chicago, Chicago, IL 60637, U.S.A.