Topology Vol. 27. No. I. pp. 7-21. 1988 Printed I Great Brtta~n.

w40

938348 s3.00+ on Perpamon Prcsa plc

A GENERALIZATION

OF THE SEGAL CONJECTURE

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

$1. INTRODUCTION is a generalization of the Segal conjecture about equivariant cohomotopy. It asserts an invariance property of the G-cohomology-theory S- xE(-); obtained from equivariant cohomotopy rrEby first localizing with respect to a general multiplicativelyclosed subset S in the Bumside ring A(G), and then completing with respect to a general ideal I c A(G). We first explain how we place previous localization theorems and completion theorems in one setting by formulating suitable invariance statements. Let G be a finite group; all our G-spaces will be G-C W complexes [23]. Let # be some class of subgroups and letf: X+ Y be a G-map. We will say thatfis an &-equivalence if the induced map of fixed-point-setsfH: XH + Y H is an ordinary homotopy equivalence for each HEX. (Thus we may assume without loss of generality that Z is closed under passing to conjugate subgroups.) Let h be a functor defined on G-spaces and G-maps; we will say that h is &?-invariant if it carries each x-equivalence to an isomorphism in the target category of h. The same property was previously introduced in [34] and studied further in [35]. In particular, let 2 be the class of all subgroups H c G; then an &?-equivalence is just a Ghomotopy-equivalence, and every G-cohomology-theory is Z-invariant. To place localization theorems in this setting, we assume that 2 is closed under passing to conjugate subgroups and larger subgroups. Then for any X we have an fl-fixed-point subcomplex Xx= u{X~:HEZJ, and the inclusion i:X-+X is an &?-equivalence.
THEOREM 1.4 below

Remark 1.1. In this case, h is Z-invariant iff h(i):h(X)+h(Xx) is iso for each X. (Only if is clear; and we will explain the converse in $7.) Localization theorems usually state that h(i) is iso when h is a functor obtained by localization, h =S k, and S,x are suitably related. Such theorems go back to Segal [29, Prop. 4.11. We place completion theorems in this setting. A class Z which is closed under passing to conjugate subgroups and smaller subgroups is called a family. A G-space Y qualifies as a universal space EF for the family B if YHis contractible for He9 and empty for H&F. For background on spaces E9, see [27, 10, 11 p. 175, 13-J.For any X the projection p:EF xX+X

is an 9-equivalence. 7

J. F. Adams, J.-P. Haeberly, S. Jackowski and J. P. May Remark 1.2. In this case, h is g-invariant

(Only if is clear; and ifJX+ G-homotopy-equivalence.) Completion theorems usually state that h(p) is iso when h is a functor obtained by completion, h = k(-);, and I, 9 are suitably related. Such theorems go back to Atiyah and Segal [6]. We will show that it makes sense to look for a best possible invariance result.

iff h(p):h(X)+h(EF x X) is iso for each X. Y is an F-equivalence, then 1 xj Ea x X+ E9 x Y is a

THEOREM1.3. For each G-cohomology-theory h* satisfying the axioms given in $7. there is

a unique minimal class 2

such that h* is #-invariant.

[Note that as J? decreases, the &-invariance property gets stronger, because less data on j-suffice to prove h(f) iso.] We seek specific invariance results (preferably best possible) for particular functors. The functors we consider are progroup-valued. The role of progroups in this subject has been recognized ever since the work of Atiyah and Segal [6]. Let h be a functor from finite G-C W complexes to R-modules. Then h yields a progroup-valued functor h defined on all G-C W complexes X; we define h(X) to be the inverse system (h(X,)), where X, runs over the finite GC W subcomplexes of X. Localization of promodules over R (with respect to a multiplicative set SC R) is done termwise: S-l{M,} = {S-M,}. To complete promodules (with respect to an ideal I c R) we define (M,}; be the inverse system {MJFM,}, where a runs as before and r runs over the non-negative integers. In particular, even if X is a finite complex, the completion h(X); is a progroup. We take h to be equivariant cohomotopy-see [l] or [30].

THEOREM 1.4. The theory S-x2(-& (progroup-valued equivariant cohomotopy localized at S c A(G) and completed at I c A(G)) is &-invariant, where

3V= u(Supp(P):PnS=@&P31}.

Here P runs over prime ideals of A(G), and Supp(P) is the support of P, which we define following Dress [ 121. [H&upp(P) if P comes from H via the restriction map A(G)-+ A(H) and P does not come from any K < H. Dress shows that Supp(P) is a single conjugacy class of subgroups H.] Our companion paper on K-theory [Z] shows that a theorem precisely analogous to (1.4) holds for equivariant K-theory; one just replaces the Burnside ring A(G) by the representation ring R(G), and supports in the sense of Dress [12] by supports in the sense of Segal[28]. Originally we sought the special case S= {l} of (1.4); this goes as follows.

THEOREM1.5. For any family 9

the theory @--)/(g,,

e q uivariant cohomotopy completed

at I(9) = ,?, is 9invariant.

Ker(A(G) --, A(H)),

A GENERALIZATION

OF THE SEGAL CONJECTURE

COROLLARY1.6. There is a pro-isomorphism

x;c;(X);,,,+m~(ES natural in the G-space X.

XX)

On the right of (1.6) we can omit the completion at I(s), because xf$(EF x X) is already complete (see $6). Given this, the result follows from (1.5) and (1.2). We refer to our companion paper [2] for the application of (1.6) to calculate the equivariant cohomotopy of equivariant classifying spaces. We may pass from the inverse systems in (1.6) to their inverse limits. We assume that X is a finite G-C W complex; then the inverse system r&X);,, is Mittag-Leffler; therefore the proisomorphic inverse system xE(Eg x X) is Mittag-Leffler; therefore its inverse limit is the representable G-cohomotopy of EF x X. All this goes back to [6]. The classical case is that in which 9 = (l}, EF becomes EG and the completion is done using the augmentation ideal Ker(.s:A(G)-,Z). In this case (1.6) becomes the Segal conjecture, which has been proved by the combined efforts of a number of mathematicians, by far the greatest contribution being due to Carlsson [8]. Compared with the special case F = {l}, the general case (1S), (1.6) has more flexibility, and (1.4) has more flexibility still. By adjusting S and I, we can obtain results about functors closer to cohomotopy, at the price of using stronger hypotheses on our spaces and maps. Conversely, (1.4) shows what price (in terms of S and I) will pay for a given level of invariance (every class _?Parises for suitable S and 1, usually for many). One of us [25] has obtained a further generalization of (1.4). In this he replaces the representing spectrum for cohomotopy, that is the sphere spectrum, by the suspension spectrum of a suitable classifying space. (See appendix.) As for history: completion theorems of the general form of (1.6) were proposed by one of us [17, IS]. For equivariant K-theory (over a compact Lie group G), such a theorem was proved independently, using different approaches, by two of us [16, 193. The analogy between K-theory and cohomotopy led to the starting-point of the present work, an attempt to prove (1.6). The statement (1.4) grew out of our attempts to explain our proof of (1.6); in order to prove completion theorems in cohomotopy, we were driven to use intermediate results which involved localization as well as completion, and involved classes 2 which were not families. The rest of this paper is organised as follows. Necessary preliminaries about progroups come in $2, and necessary preliminaries about the Burnside ring come in $3. $4 and $5 go to proving (1.4); 96 deduces (1.5) and (1.6); and finally, $7 covers (1.1) and (1.3). The proof of (1.4) may be summarized as follows. We assemble the result from information over the rationals, which is easy to come by, and p-adic information, which we derive ultimately from Carlsson [S]. The assembly job is done by (2.3), which is our main algebraic weapon. Carlsson proceeds from his p-adic result to the I-adic statement of the Segal conjecture by quoting the work of May and McClure [26]; our main proof, in $5, subsumes and generalises that part of the proof of the Segal conjecture. (Note that even for pgroups (1.4) gives some new information, because its proof builds in rational information.) The steps of our main argument prove special cases of (1.4) which grow successively more general. In the course of upgrading our information in $5, we need a relation between equivariant cohomotopy over a group G and equivariant cohomotopy over a quotient group G/H. We prepare this result in $4. The difference between the proof of (1.4) and that in [23 is explained by the fact that this relation works much better in cohomotopy than in K-theory, while the

10

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

Euler class is much more accessible in K-theory than in cohomotopy. Otherwise the only topological ingredient worth mentioning in $5 is the use of transfer in (5.4).

$2. PROGROUPS

In this section we will summarize what we need about progroups. The language of progroups is due to Grothendieck [15] and may be found in [4, 63 and later references. Inverse systems of Abelian groups, indexed on directed sets, qualify as progroups. The progroups which arise in the examples given in $1 are of this form. However, at the end of $7 we assume that h* carries any direct limit of G-spaces to an inverse limit in the category of progroups. To construct an inverse limit in the category of progroups, you take all the data contained in your inverse system of progroups, and interpret it as a single progroup [4]. To make this idea work as stated, one generalizes the allowable indexing systems to filtering categories. If {M,) and {NB) are progroups, one defines Prohom({M,),{iV~})=l~limHom(M,,N~), a where both limits are taken in the category of groups. There is a unique sensible definition for the composite of prohomomorphisms. The progroups and prohomomorphisms make up a category. A prohomomorphism (M,}+(N,} is a pro-isomorphism if it is an isomorphism in this category. In $1 we introduced a progroup-valued functor h, giving the definition on objects as h(X) = {h(Xd)}. It is easy to supply the definition of h on maps. The main use of the language of progroups is to make statements about inverse systems which cannot be expressed as statements about their limits. These are mostly statements about exactness. In fact, the category of progroups is an Abelian category, in which one can conduct exactness arguments. LEMMA2.1. The functor S - I@--);
sequences. of (1.4) carries pairs and cojiberings to pro-exact

Of course, the assertion about cofiberings assumes that one introduces the reduced theory it; and uses it in the usual way. It may be reassuring, and help in checking lemmas and details, if we make the definition of pro-exact utterly explicit. Let LLM$N be a sequence of two prohomomorphisms By definition, the element whose composite is the zero prohomomorphism.

kl$nlimHom(L,,MB) z is a system of compatible elements fa&mHom(L,,MB), ii and each fP is an equivalence class of representatives

A GENERALIZATION

OF THE SEGAL CONJECTURE

11

The sequence is pro-exact at M if for each such representative

there is a diagram
My% N,

L .&Ma

I
m

in which m is a map of M, gya is a representative for some component ga of g, and m(Rer gya)= Im_&. Cultural aside: inverse limits in the category of progroups preserve pro-exactness. Proofof(2.1). If X, is a finite G-C Wcomplex, then n,(X,) is a finitely generated Z-module [l] and therefore finitely generated over A(G). Thus S-x&(XJ is finitely generated over the Noetherian ring S-A(G). The Artin-Rees lemma [S] may now be used to show that if X,c Y, is a finite pair, the sequence . . . -+

s-G(Yg,XA
i (s-1z)'s-'7r"G(Y~,x,)

Ii

STr"G(Yg)

--, (s-'z)'s-'7r"c(Y~)

Ii

s - 7$(X,) -+ (s-z)s-7r;(x,)

+ ...

is proexact. Varying the finite pair, we get enough to prove the required proexactness statement for a general pair Xc Y. Similarly for cofiberings.
LEMMA 2.2. In order to prove that a G-cohomology theory h* is %-invariant, it is sujicient to verify the following special case: if Z is a pointed G-space such that ZH is contractible for HEY~, then fi*(Z) = 0.

The proof of (2.2) would be clear if h* were group-valued. We would assume given an xequivalencef:X+ Y, and apply the assumed property of h* to the mapping-cone Z = Yu,CX. We would then use the exact cohomology sequence of a cofibering (which is the only significant assumption on h* we need) to show that h*(f) is an isomorphism. Of course, this proof carries over to progroup-valued functors, and it is for this purpose that we have stated (2.1) explicitly. The equation ,fi*(Z)=O should now be read h;(Z) is prozero. Here a progroup (M,} is prozero if it is a zero object in the category of progroups, and this is equivalent to the following explicit condition: for each of its objects M,, the progroup has a zero map
M BAM,. Now we need a result for proving that progroups are prozero, and what follows is our main algebraic weapon. Let M = {M,} be a pro-object of finitely generated modules over a Noetherian ring R; let s be a multiplicative subset of R, and let I be an ideal in R. LEMMA2.3. S- M; is prozero &j-S; ME is prozero for each prime ideal P c R such that PnS=@ and PDI.

12

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

Here S; 1 means localization at P; that is, the multiplicative set S, is the complement of P.
Proof: It is immediate that if S-M; is prozero then so are all the other SF Mpn; we have to argue in the other direction. First we note that it is enough to prove the special case S = (l}, in which data are given for all P 3 I and the conclusion is M; = 0. For then to prove (2.3) in the generality given, we apply the special case to the promodule S-M over S- R; the primes Q of S- R for which we require data correspond to the primes P of R for which we have data. We will find a finite Assuming S = {l}, we take a typical term in M;, say T= MJIM,). P, containing Z and integers s(i) such that the map number of prime ideals P,, P,, . . . , T= M,/(ZM,)-+@S;il(MJ(Zr+ 1 P,)M,)

is mono. In fact, we take P,, P2, . . . , P, to be the associated prime ideals of T, which are finite in number by a standard result [22]. These prime ideals contain I, and therefore contain 1. Let LiC T be the submodule annihilated by Pi. By the Artin-Rees lemma [S] there exists s(i) such that LinP.TcP.L.=O . I 11 We will show that the kernel Ki of the map
T-+ S; ( T/P,scOT)

does not have Pi as an associated prime. For suppose it did, and for convenience write P, L, s, K instead of Pi, L,, s(i), Ki. Then we would have a monomorphism R/P+K, which must map into L. Since L+ T/PTis mono by the choice of s, we would get a monomorphism R/P+T/PT. Since localization preserves exactness, we would get the following commutative diagram.
RIP----+ T/PST

monoI S; (R/P)-

I S; (T/PST)

But the diagonal is zero because we assumed R/P mapped into K. This contradiction that Ki does not have Pi as an associated prime. But then the kernel of
T-+@S,(T/PfT) I

shows

has no associated primes, and must be zero as claimed. Given T= M/(ZM,), we now have the following commutative m: M,-+M, in M. MJZM,)
I @S,(MB/(Z 1 + Pf)M,)+@S,(MJ(Z+ I

diagram for any map

* MJZM,)
I

= T

mono P;)M,)

A GENERALIZATION OF THE SEGAL CONJECTURE

13

For each Pi our hypotheses allow us to choose rn so that S,M, maps to zero in S,(M,/PfM,). We can do this for a finite number of i, and so ensure that the lower horizontal arrow is zero; then m must be zero. This proves (2.3).

83. THE BURNSIDE RING

In this section we will say what we need about the Burnside ring. The Burnside ring A(G) is the Grothendieck group constructed from (finite) G-sets [ 123. For each subgroup H c G there is a homomorphism of rings

which carries a G-set W to ( WHj; bH depends only on the conjugacy class of H. With these maps as components, we obtain a map

where the product runs over all conjugacy classes (H); CD mono. By the going-up theorem is [S], each prime ideal P of A(G) is the restriction of a prime ideal of IIZ; that is, it may be written in the form for some H and some prime ideal (p) in 2. Here (p) is clearly determined by P; however, we may still get the same ideal q(H,p) for different choices of H. Fix a prime p>O; for each subgroup H c G, let H, be the smallest normal subgroup of H such that H/H, is a p-group. Then (H,,), is a characteristic subgroup of H,, and hence normal in H, so (HP)P= H,; thus H, is p-perfect, meaning that any quotient of it which is a p-group is trivial. Dress [12] says that H and K are p-equivalent, and writes H -&, if H, is conjugate to K,; he shows that q(H,p) = q(0) iffH N &. The support of q(H,p) is then the conjugacy class of H,. For p = 0 we can interpret this discussion in the same way as for any other prime which does not divide ICI; H, becomes H, and O-equivalence becomes conjugacy. In the rest of this paper we shall make free use of localization with respect to prime ideals in A(G). Integer denominators are sometimes more convenient than general elements of A(G), and we can reduce to that case. Let P be a prime in A(G), and let(p) be its counter-image under Z-+ A(G); we write S&,,for localization over Z at (p).
LEMMA

3.1. The map S,:A(G)+S;

A(G) is epi.

To prove this conveniently, we discuss the idempotents in S&rA(G). Such idempotents have been used by several authors [ll p8, 14,3,31]. We continue to write 4 after localizing at (p). If H - P/C then dH(x) = (PK(x)mod p for any x; in particular, if e is idempotent then 4H(e) must be constant at 0 or 1 as H runs over a p-equivalence class. By a standard result of commutative algebra [7] the Boolean algebra of idempotents in S,;:A(G) is canonically isomorphic to the Boolean algebra of open-and-closed sets in spec S&rA(G). This spectrum has been explicitly described by Dress [12]; it is the disjoint union of finitely many open sets, each containing just one of the ideals q(H,p). There is therefore just one primitive idempotent eH in S&r A(G) for each conjugacy class of p-perfect subgroups H, given by

hAed =

1 if Kw,H 0 if K+,H.

14

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

These idempotents can also be obtained by more elementary methods. For (p)=O we can interpret this discussion in the obvious way.
Proofof(3.1).

Let HE Supp(P) and let e=eH be the corresponding idempotent. Consider A(G)+S,:A(G)+eS,:A(G).

the map This is a map of rings which carries every element of S, to an invertible element. (The target eS(;:A(G) is a local ring because it has only one maximal ideal, and the counter-image of that maximal ideal is P.) It is also universal among such maps. [Any such map carries e to an
invertible the map is epi. It may be helpful to know that the localized cohomology theory S; h* is the same as that obtained by first localizing over Z to get SC;:h*, and then taking the summand e,S(;f h*; element and (1 - e) to zero.] This characterizes the target as S, A(G). But clearly

compare [20].

$4. TOPOLOGICAL

PRELIMINARIES

In this section we prove a topological result needed for the main proof. Let G be a
finite group q(G,p)=q(H,p) and (p) a given prime; let H = G, and let P be the corresponding in A(G), as in $3. prime ideal

PROPOSITION Then there is an isomorphism 4.1.

pointed G-spaces. [21] and probably others. We

Results of this sort were known to Araki [3] McClure separate off the first part of the proof. LEMMA 4.2. Restriction gives a natural isomorphism &1(X, Y}GS&?{XH, Y}C.

Here X runs over finite pointed G-spaces; Y runs over pointed G-spaces infinite; and (X, YjG means stable G-homotopy-classes of stable G-maps. Skerch proofof(4.2). all the G-cells S; {X, Y} Gis one group of a G-cohomology-theory (G/K) x E,(G/K) x S -

which may be

which is zero on

of X which are not in XH. See [20], Theorem

4.8. we can assume n 30. Restriction on H-fixed-point-sets

gives a

jinG(X)+ ir,,(XH).

A GENERALIZATION

OF THE SEGAL CONJECTURE

15

This is a map of A(G)-modules, provided we make A(G) act on i&(X) via the homomorphism 8: A(G)*A(G/H) which carries a G-set Wto WH. Notice now that G/H is a pgroup, .S,:A(G/H) is a local ring, and the counter-image of its unique maximal ideal in A(G) is P. Thus 0 carries an element of A(G) not in P to an element invertible in SC;:.4(G/H). So we get an induced map @S, it,(X)+S&i ?&(XH). We show that 4 is epi. The map s&i iiC(XH)+s,; ii& is split epi because any representative (G/H)-map is also a G-map. A fortiori, s, l iigX)-+S;; is epi. The map s; 1irc(X) + s, 1jinG( X) is epi by (4.2) applied to Y=s. We show that 4 is mono. Take an element of S, i 5$(X); using (3.1), we may write it m/d, where d is an integer prime to p andfis a representative G-map
7&(XH)

for a suitable representation V of G. Now assume Lfl/dgKer 4. Then after increasing both d and V, we may assume that the restriction off to SvH~ XH is G -nullhomotopic. Thus [fl maps to zero in s, l {YA XH, sv A Sy-T But then [fl maps to zero in

S;(S
by (4.2) applied to Y= Sv A
s.

X, Sv A

r}

Thus Lfl/d=O.

$5. THE MAIN PROOF

In this section we will prove Theorem 1.4. By (2.3) it is sufficient to consider S; n;;(X);; in this case the only relevant assumption is the contractibility of XH for one conjugacy class of H.
LEMMA5.1. Let G be a p-group. If X, the underlying space of X, is contractible, then iTE(X)& is prozero.

The result remains true in a trivial way if we take (p) to be the prime ideal (0), for we have to interpret it so that G= 1 is the only group which qualifies.
Proof of(5.1).

Carlsson [8] proves that the inverse limit of the inverse system

%W.)

PfCAX,)

is zero. Since the groups of this inverse system are finite groups, it follows that the inverse system is prozero.

16

J. F. Adams, J.-P. Haeberly, S. Jackowski and J. P. May Now let G be a finite group and (p) a given prime; let H = G, and P= q(G,p) = q(H,p) be as

in $4. LEMMA5.2. If X is contractible, then SF it:(X): Proof: By (4.1) we have an isomorphism is prozero.

of progroups

{p~~~~~~~)}eip~~~~~~)~. Since G/H is a p-group is, S;I?z(X)& result follows. Now let P be a general prime ideal q(H,p) in A(G), where H is p-perfect. is prozero. and XH is contractible, Since completion the right-hand side is prozero by (5.1). That at (p), the

is prozero.

at P is more drastic than completion

PROPOSITION If XH is contractible, then SF iio(X)$ 5.3.

The proof involves a construction. Let N be the normalizer of H in G, and let F/H be a Sylow p-subgroup of N/H, where F/H is interpreted as H/H if (p)=O. LEMMA5.4. SF lIzz(X); is a direct summand in (S; %((GIF) x X,, G/F)};. Here {SF $((G/F) complexes of X. x X,,G/F)} is a progroup in which X, runs over the finite sub-

LEMMA5.5. Zf XH is contractible, then (S; z,((G/F) x X,,G/F))p^ is prozero. Proof of5.4. For each finite G-space X, we have a G-covering-map (G/F) x X,,GfF+ natural in X,. This gives the following see, for example, [l] or [21]. commutative XvPt diagram, in which Tr means transfer-

n,((GIF) x X,, G/F) mY nnG(Xa,Pt),Wk %(X,7 Pt)

The horizontal arrow is multiplication by the class of G/F in A(G). Using the fact that H is p-perfect and F/H is a p-group, we find A,(GIF) = IN/F1 which is prime to p by the choice of F. Thus [G/F] does not lie in P, and on localizing get the following commutative diagram, which is natural for maps of X,. S; 1 G((W)
y_ s, l ax,, PG x X,, G/F) p s; l nc(X,, PO

at P we

The conclusion

follows.

A GENERALIZATION Proofof(5.5).

OF THE SEGAL CONJECTURE

17

Of course we use the natural isomorphism rMG/F) x X,, G/F) = ~;(X.,pr).

This isomorphism is a map of A(G)-modules, if we make A(G) act on n;-(X,,pt) via the restriction map i*: A(G)+A(F). It follows that (S; G((GIf3 x X,, G/O);
= {S - ncK,pt)};,

where on the right-hand side localization and completion are done over A(F), taking S = i*Sp,
z=(i*P)A(F).

We now wish to prove that S-%nf(X); is prozero. By (2.3) it is sufficient to prove that S, $(X); is prozero for each prime ideal Q of A(F) such that QnS= 0 and Q 131. Equivalently, we have to consider prime ideals Q whose counter-image in A(G) is P. We will show there is only one such ideal, namely the ideal q(H,p) of A(F). Any such Q has to be an ideal q(K,p) of A(F) for the same p and some K c F which is p-perfect and conjugate to H in G; it follows that K = H. Thus (5.2) applies and shows S; e;(X)& =O. This proves (5.5). Proposition 5.3 follows from (5.4) and (5.5). Theorem 1.4 follows immediately by assembling (2.2), (2.3) and (5.3).

In this section we deduce (1.5) and (1.6). Proofof( 1.5). We deduce (1.5) from (1.4) by taking the ideal I in (1.4) to be the ideal Z(9) in (1.5). For this it is enough to show that if 9 is a family, then the class ti = {Supp(P): P 1 Z(9)} in (1.4) is contained in 9. Since 9 is a family, the ideal Z(9)= n Ker(A(G)+A(H)) HE.9

is the intersection of the prime ideals q(H,O) over HER. If P 3 Z(9), then P must contain one of these ideals q(H,O). According to Dress [ 123, this means that P = q(H,p) for some H and some p. Since H,c HER and 9 is a family, we have Supp (P)c9. This holds for each
PIZ(d), so %CC.

[Of course, %=F since every HEN is the support of an ideal q(H,O).] We turn to (1.6). Let 9 be a family.
LEMMA 6.1. Let Y be aJinite G-C Wcomplex such that YH is emptyfor H&F. Then zE( Y) is annihilated by some power of Z(9). LEMMA 6.2. Let Y be a G-space such that Y is emptyfor H&F. Then the canonical pro-map is a pro-isomorphism. 6( Yk-43 Y),;,,

We omit the proof of (6.1) and the proof of (6.2) from (6.1); both are sufficiently well known, and the ideas go back to [6]. TOP 27:1-B

18

J. F. Adams, J.-P. Haeberly, S. Jackowski and J. P. May Lemma 6.2 applies to Y= EF x X and shows that in (1.6), the right-hand x X) is already complete. side

aE(EF

97

In this section we explain (1.1) and prove (1.3). We say that a G-CW complex X is an #-complex if its G-cells are all of the form G/H x E with HEY?. It is easy to prove the appropriate generalization of the theorem of J. H. C. Whitehead; this says that if X is an #*omplex andf: Y-+Z is an #-equivalence, then the induced map f*:CX,

ciG-CX,.qG

is a bijection. In particular, an x-equivalence between *-complexes is a G-homotopyequivalence; this was certainly known to previous authors [34, $11. Assume, as in (l.l), that x is closed under passing to larger subgroups. Then an xcequivalence f: X+ Y induces a map f: Xx-+ Y% which is an #-equivalence between &@-complexes, and therefore a G-homotopy-equivalence by the remarks above. Now (1.1) follows. One may also deduce the result from [21, II, 9.31. We turn to (1.3). Our assumptions on h* are as follows. It is Z-graded and satisfies Eilenberg-Steenrod Axioms 1-6, with the words exact and isomorphism interpreted as pro-exact and pro-isomorphism if h* is progroup-valued. No axiom of suspension with respect to arbitrary representations is required.

Proofof( 1.3). First we define the required class 2. For any subgroup K, let g(K) be the complement of the conjugacy class (K), i.e., the class of subgroups not conjugate to K. We lay down that K is not in # if and only if h* is invariant with respect to g(K). It follows that if h* if Y-invariant, then Y 12; for if K&5?, then %(K) 3 Y, so the y-invariance of h* implies the 55(K)-invariance, and K&V. This justifies the words unique minimal in (1.3). It remains to prove that h* is H-invariant. Let % be a family, and let %= F-u(H) be the adjacent family obtained by adjoining the conjugacy class of a subgroup H all of whose proper subgroups lie in % Consider the map i:X A (EFuP)+X
A @@UP).

Then iK is an equivalence for K + H, so if HQT it follows that h*(i) is iso..If HE%, the same conclusion follows trivially if we assume XH contractible. Suppose then that XH is contractible for all HE%. We can get from the empty family to the family of all subgroups by a finite number of the steps considered above; so h*(i) is iso for the map i:x A P+X A se. That is, K*(X)=O. Now (2.2) shows that h* is x-invariant. We remark that this proof carries over when G becomes a compact Lie group. We need one more assumption on h*: it carries any direct limit of G-spaces to an inverse limit in the category of progroups (see $2). The finite induction implicit in the proof above is replaced by an appeal to Zorns Lemma, using the class $7 of families % such that l*(X
A (E%uP)) =o.

A GENERALIZATION

OF

THE

SEGAL

CONJECTURE

19

The induction

starts because this class contains

the empty family. We have to show that for

any totally ordered bound

subset (.9:,} of families in V, the union 9 = u F_. will serve as an upper a the homotopy-limit Holim E9,. The construction of this zi over CT EF;, by induction over n, but x and the fact that we always so

in %7. For this we construct

limit involves extending G-maps (~3~9) EF:, + E9, x this is certainly possible in view of the properties of ETm, E9, have 9,~ FD. We observe that Holim a EFm qualifies

as EF,

X A (E4uP)=Holim(X of h* gives

A (ES,uP))

and the assumed

property

6*(X r\(ESuP))=O. Zorns lemma now shows that % has a maximal element 9. If 9 were not the family of all subgroups, then there would be a subgroup H minimal among subgroups not in 9, and the argument above applied to 9=9u(H) would yield a contradiction. Thus we conclude /i*(X) = 0, as before.

Acknowledgements-J.F.A. and S.J. thank similarly thanks also the Forschungsinstitut successive versions of this paper.

the University fiir Mathematik

of Chicago and Northwestern University, and S.J. ETH Zurich, for hospitality during discussions on the

REFERENCES 1. J. F. ADAMS: Prerequisites 483-532. 2. J. F. ADAMS, for Carlssons lecture. Lecrure Notes in Mathematics,

Springer, Berlin 1051(1984).

J.-P. HAEBERLY. S. JACKOWSKI~~~ J. P. MAY: A generalization

of the Atiyah-Segal

theorem, Topology 27 (1988). 1-6. 3. S. ARAKI: Equivariant stable homotopy theory and idempotents of Burnside rings, preprint (1983). 4. M. ARTIN and B. MAZUR: Etale homotopy, Lecture Notes in Mathematics, Springer, Berlin 100(1969), 154-166. 5. M. F. ATIKAH and I. G. MACDONALD: Introduction to commutative algebra, Addison-Wesley U.K. (1969), 62-107. 6. M. F. ATIYAH and G. B. SEGAL: Equivariant K-theory and completion, J. Difl Geom. 3 (1969), 1-18. 7. H. BASS: Algebraic K-theory, W. A. Benjamin U.S.A. (1968). p. 103. 8. G. CARLSSON: Equivariant stable homotopy and Segals Burnside ring conjecture, Ann. of Math. 120 (1984). 189-224. 9. J. CARUSO and J. P. MAY: Completions in equivariant cohomology theory. To appear. 10. T. TOM DIECK: Orbitypen und aquivariante Homologie 1, Arch. Moth. 23 (1972). 307-317. 11. T. TOM DIECK: Transformation groups and representation theory, Lecture Notes in Mathematics, Springer, Berlin 766 (1979). 12. A. DRESS: A characterisation of solvable groups, Math. Zeit. 110 (1969). 213-217. 13. A. D. ELMENDORFF: Systems of fixed point sets, Trans. Am. Math. Sot. 277 (1983), 275-284. .14. D. CLUCK: Idempotent formula for the Burnside algebra with applications to the p-subgroup simplicial complex, Illinois J. Math. 25 (198 1). 63-67. 15. A. GROTHENDIECK: Technique de descente et thtorems dexistence en gCometrie alg6brique II, Seminaire Bourbaki 12ieme annte, (1959-60). exp. 195. 16. J.-P. HAEBERLY: Completions in equivariant K-theory, Thesis, University of Chicago (1983). 17. S. JACKOWSKI: Localisation and completion theorems in equivariant cohomology theories, Thesis, University of Warsaw (1976). 18. S. JACKOWSKI: Equivariant K-theory and cyclic subgroups, Lecture Notes Lond. Math. Sot. C.U.P. 26 (1977).
7692.

20

J. F. Adams, J.-P. Haeberly,

S. Jackowski

and J. P. May

19. S. JACKOWSKI: Families of subgroups and completion, J. Pure and Appl. Algebra 37 (1985). 167-179. 20. C. KOSNIOWSKI: Equivariant cohomology and stable cohomotopy. Mnrh. Ann. 210 (1974), 83-104. 21. L. G. LEWIS.Jr, J. P. MAY and M. STEINBERGER (with contributions by J. E. MCCLURE):Equivariant stable homotopy theory, Lecture Notes in Mathemarics, Springer. Berlin, I213 (1986). 22. H. MATSUMARA: Commutative Algebra, W. A. Benjamin, U.S.A. (lY70), p. 52. 23. T. MATUMOTO: G-Cl+-complexes and a theorem of J. H. C. Whitehead. J. Far. Sci. Univ. Tokyo. Sect. IA on Marh. 18 (1971/72), 363-374. 24. J. P. MAY: The completion conjecture in equivariant cohomology, Lecture Notes in Mathemarics, Springer, Berlin 1051 (1984). 620-637. 25. J. P. MAY: A further generalization of the Segal conjecture. in preparation. 26. J. P. MAY and J. E. MCCLURE:A reduction of the Segal conjecture, Can. Math. Sot. Conierence Proceedings, Vol. 2 (I 982), 209-222. 27. R. S. PALAIS: The classification of G-spaces, Memoirs Am. Math. Sot. 36 (1960). 28. G. B. SEGAL: The representation ring of a compact Lie group, Inst. Hautes Etudes Sci. Pub/. Math. 34 (1968).
113-128.

29. G. B. SEGAL: Equivariant K-theory, Insf. Haures Etudes Sci. Puhl. Marh. 34 (1968), 129-151. 30. G. B. SEGAL: Equivariant stable homotopy theory. in Proceedings of the International Congress of Marhematicians 1970, Gauthier-Villars, Paris. 2 (1971). 59-63. 31. T. YOSHIDA: Idempotents in Burnside rings and Dress induction theorem, J. Algebra 80 (1983), 90-105. 32. L. G. LEWIS, J. P. MAYand J. E. MCCLURE: JR., Classifying G-spaces and the Segal conjecture, Can. Moth. Sot.
Conference Proceedings, 2 (1982), 165-179. 33. J. P. MAY:Stable maps between classifying spaces, Contemp. Math. 37 (1985). 121-129. 34. R. M. SEYMOUR: On G-cohomology theories and Kunneth formulae, Can. Math. Sot. Conference Proceedings, 2 (1982) 251-271. 35. R. M. SEYMOUR: A Kunneth formula for RO(G)-graded equivariant cohomology theories, preprint.

APPENDIX: BY J. P. MAY In [32] the Segal conjecture coefficients calculation was generalized to the assertion that equivariant cohomotopy with

in equivariant classifying spaces is { 1 }-invariant. This generalization specializes to give a of the stable maps between classifying spaces in the non-equivariant world [33]. In a later

paper [25], I will use the theorem below to prove that equivariant cohomotopy with coefficients in equivariant classifying spaces satisfies the X-invariance property analogous to (1.4) above. This generalization of (1.4) specializes to give a calculation of the equivariant stable maps between equivariant classifying spaces. The proof of (1.4) takes given information about p-groups and p-adic completion, namely that supplied by Carlsson and quoted as (5.1), and derives from it the strongest implications. This idea works in considerable generality. Let hf; be a Z-graded cohomology theory defined on G-C Wcomplexes. We want hE to take values in

modules over A(G), and the natural way to ensure this is to require hZ to be RO(G)-gradable, or equivalently, representable. We also require h; to be of finite type, in the sense that each @(G/H) is finitely generated, and this ensures that each h:(X) is finitely generated when X is a finite G-CW complex. We obtain a Z-graded progroup-valued cohomology theory on general G-CWcomplexes by setting h:(X)= (hg(X,)} as in $1.
C Wcomplexes

We need a relation like (4.1), and this requires us to construct representable theories h&K on H/Kfor subquotient groups H/K of G. There is a sensible way to do this [24, p. 626; 9;3; 21, II,

$93, the evident analog of (4.1) holds [3: 21, V, 963 and so does the analog of(5.4). When hz is stable Gcohomotopy with coefficients in a G-space Y, the theory associated to H/K is just stable H/Kcohomotopy with coefficients in the H/K-space YK.In this case. the proofs above of(4.1) and (5.4) apply with only notational changes. From here, one can argue exactly as in $5 to reach the following THEOREM. Suppose that h;,,
whenever SHJK is a p-group is o[jinite rypejbr each s&quotient

conditional
H/K

conclusion.

of G and that 6&x(X),i,=O Then the theory

and X is a nonequivariantly where

contractible

H/K-space.

h;(-);

is ;X-imariant,

.#=uujSupp(P)JPnS=@&P~I}.

A GENERALIZATION A comparable by Carlssons reduction

OF THE SEGAL to elementary

CONJECTURE Abelian p-groups

21 can be obtained

from general p-groups

methods

[24, 93.

Department of Pure Mathematics, University of Cambridge, 16 Mill Lane, Cambridge CB2 ISB, U.K. Dept. of Mathematics University of Washington Seattle, WA 98195 U.S.A. Mathematical Institute University qf Warsaw Palac Kultury i Nauki IXp 00-90 1 Warszawa Poland Dept. of Mathematics University of Chicago Chicago, IL 60637 U.S.A.