BULLETIN (New Series) OF THE AMERICAN MATHEMATICAL SOCIETY Volume 4, Number 2, March 1981

ORDINARY i?O(G)-GRADED COHOMOLOGY

Let G be a compact Lie group. What is the appropriate generalization of singular cohomology to the category of G-spaces XI The simplest choice is the ordinary cohomology of EG xG X, where EG is the total space of a universal principal G-bundle. This Borel cohomology [1] is readily computable and has many applications, but is clearly inadequate for such basic parts of G-homotopy theory as obstruction theory. Another choice is Bredon cohomology [2], as generalized from finite to compact Lie groups by several authors. This gives groups HQ(X;M) for n > 0 and for a "coefficient system" M. Here M is a contravariant functor from the homotopy category of orbit spaces G/H and G-maps to the category Ab of Abelian groups. (Subgroups are understood to be closed.) Bredon cohomology is adequate for obstruction theory. For finite G, Triantafillou has used it to algebraicize rational G-homotopy theory [11], and she and two of us have used it to set up the foundations of the theory of localization of Gspaces for general G [8]. When M is constant at an Abelian group A, written M = A, we have (*) H"G(X;A) = Hn(X/G;A). In particular, we have H%(EG x X\ A) = Hn(EG xG X\A).

Thus Bredon cohomology generalizes Borel cohomology. Nevertheless, we feel that these theories do not comprise the full equivariant generalization of ordinary singular cohomology. The full theory should build in the interplay relating the Burnside ring ,4 (G), the real representation ringjRO(G), and G-homotopy theory. Any cohomology theory must be "stable". Bredon cohomology is only stable in the classical sense that HnG(X;M) = H^q(LqX;M)

for a based G-space X (with basepoint a fixed point). A fully equivariant theory E* should allow groups EV(X) for all G-representations V;En(X) should be the special case of the trivial representation Rn. If SV denotes the 1-point compactification of V and SUX denotes X A SV, we should have EV(X) = Ev(Bw(LwX).
1981 American Mathematical Society 0002-9904/81/0000-0109/$03.25

Received by the editors July 8, 1980. 1980 Mathematics Subject Classification. Primary 55N25, 55P42, 57T15. 208

ORDINARY #6>(G)-GRADED COHOMOLOGY

209

Such RO(G)-gmded cohomology theories have many advantages. They admit powerful splitting theorems that greatly simplify the task of computation; see e.g. [3]. They come equipped with transfers for G-bundles and G-fibrations with compact fibers [10], [13]. They are essential to the specification of a meaningful notion of orientability of G-bundles and spherical G-fibrations and are therefore essential to a meaningful formulation of even such a basic notion as Poincar duality [15]. Known examples of ./?<9(G)-graded theories include ^-theory, various cobordism theories, and theories produced by equivariant infinite loop space theory, a comprehensive treatment of which is given by Hauschild, May, and Waner [6]. The last example includes equivariant algebraic ^-theory of rings, which is defined and computed for finite fields by Fiedorowicz, Hauschild, and May [5]. It also includes spherical G-fibration theory, which is used by McClure [9] to prove the sharpest form of the equivariant Adams conjecture and to study the groups JOG(X). Ordinary cohomology is missing from this list. For finite G, Waner [14] has used the deepest form of equivariant infinite loop space theory (which also constructs jRG(G)-graded theories from Galois extensions with Galois group G [6]) to construct ordinary Z?0(G)-graded cohomology theories. However, at this writing all forms of equivariant infinite loop space theory are strictly limited to finite groups, and such a construction is obviously unsatisfactory for something as basic and presumably elementary as ordinary cohomology! For arbitrary compact Lie groups G and appropriate coefficient systems M, we have extended Bredon cohomology to an i?0(G)-graded theory H$(X;M). For finite G, M must extend to a Mackey functor as defined by Dress [4]. For general G, we have invented the appropriate notion of a Mackey functor. As a first application, we have the following simple proof of a deep unpublished theorem of Oliver. Let X be a G-space of the homotopy type of a G-CW complex and let H be a {closed) subgroup of G. Let R be any commutative ring. Then there exists a natural transfer homomorphism
THEOREM.

r:Hn(X/H;R)-^Hn(X/G;R)

forn>0

such that T 7T* is multiplication by the Euler characteristic x{G/H), where IT: X/H X/G is the map given by inclusion of orbits.
PROOF. By (*) above and by a change of groups isomorphism, we have a commutative diagram

HnG(X;R)^H"(X/G;R) HnG{G/H x X; R) = Hn(X/H; R).

**[

l*

210

G. LEWIS, J. P. MAY AND J. McCLURE

Here TT on the left is the projection G/H x X * X, which is obviously a G-bundle. This much precedes our theory, requiring only standard facts about Bredon cohomology. We have checked that R is the underlying coefficient system of a Mackey functor. Thus the groups on the left are terms in an RO(G)'graded cohomology theory. We therefore have a transfer homomorphism connecting them. On a technical note, we have a new treatment of transfer for G-bundles with compact fibres which requires no finiteness condition on the base space. The Euler characteristic formula drops out immediately from general properties of the transfer together with specific properties of the relevant Mackey functor. REMARKS (i). We are using singular cohomology, wheras Oliver uses Cech cohomology and totally different techniques. (ii) If H has finite index in G, the result is easy. Here Illman [7] already gave a transfer adequate for our proof. (iii) The G-CW homotopy type hypothesis on X can be weakened. By Waner's results [12], it is not very restrictive in any case. Our construction of H (X; M) is based on use of the equivariant stable category constructed by the first two authors. Let U = Vj, where Vj is the sum of countably many copies of Vt and {V.} runs through a set of representatives for the irreducible real representations of G. A G-spectrum E is a collection of based G-spaces EV indexed on the finite dimensional invariant sub spaces V of U together with G-homeomorphisms EV = SlwE(V + W) for V orthogonal to W. Maps E F are collections of G-maps EV FV compatible with the given > homeomorphisms. Homotopies are families of maps parametrized by the unit interval. A map is a weak equivalence if each fixed point map (EV/1 (FVf1 is a weak equivalence. The stable category HSG is obtained from the homotopy category of G-spectra by adjoining formal inverses to the weak equivalences. There is a notion of a G-CW spectrum, and HSG is equivalent to the homotopy category of G-CW spectra and cellular maps. HSG has all the good formal properties familiar from the nonequivariant context, and its suspension functor S is an equivalence. There is a stabilization functor S from based G-spaces to Gspectra, and DX X A S where S is the sphere G-spectrum 2 ^ . A G-spectrum E determines an i^O(G)-graded theory on G-spectra Y via E?{Y) = [Y, XVE]G. By restriction to Y = 2(X + ), there results a n i ^ G ) graded theory on G-spaces X, where X4" = X II {*}. Here V E RO(G), desuspension in HSG allowing the interpretation of XVE. Let 0 be the full subcategory of HSG with objects G/H+ A S. We define a Mackey functor to be a contravariant additive functor M: 0 Ab. This is (most unobviously!) equivalent to the usual notion when G is finite. By restriction to maps / + A 1 for G-maps : G/H - G/K9 a Mackey functor determines an underlying coefficient system. For a G-spectrum Y and integer n, define a Mackey

ORDINARY RO(G)-GRADED COHOMOLOGY

211

functor nnY by (nnY) {G/H+ A S) = [G/ff+ A Sw, Y]G,

where S = SS.

For example, if Y = S the right side is the Burnside ring A(H). For a G-CW spectrum Y with -skeleton Yn9 define C r = <nn(Yn/Yn-1). YnjYn~x is a wedge of G-spectra of the form G/H* A Sn. Then C^F is a complex in the Abelian category of Mackey functors. We can form the cellular cochain complex C*(7;M)=Hom (C*r,M) and pass to homology to obtain a Z-graded cohomology theory H*(Y; M) on Gspectra. Its Oth term is represented by an Eilenberg-Mac Lane spectrum K(M, 0). The RO(G)-gmded theory on G-spectra determined by K(Mf 0) extends the Zgraded cellular theory we started with and gives the desired i?0(G)-graded extension of Bredon cohomology with coefficients in M. We have a dual construction of i?0(G)-graded homology theories H*(Y; N) with coefficients in covariant functors N: 0 > Ab. For finite G, 0 is self-dual and the two kinds of coefficient systems are equivalent. For general G, 0 is not self-dual and a quite different kind of Eilenberg-Mac Lane spectrum K(N, 0) represents these homology theories. We have obtained change of groups isomorphisms, universal coefficients spectral sequences, Kunneth theorems, Green functors and products, AtiyahHirzebruch spectral sequences, and so forth for these new theories. Details will appear in due course. Computations are work in progress. Incidentally, we have also proven that, for finite G, the completed Burnside ring A(G) has defect set the p-Sylow subgroups of G. In particular, this implies that the Segal conjecture, which asserts that (G) is isomorphic to the 0th stable cohomotopy group of BG9 is true for all finite groups G if it is true for all pgroups.
REFERENCES
1. A. Borel, Seminar on transformation groups, Ann. of Math. Studies, no. 46, Princeton Univ. Press, Princeton, N. J., 1960. 2. G. Bredon, Equivariant cohomology theories, Lecture Notes in Math., vol. 34, Springer-Verlag, Berlin and New York, 1967. 3. T. torn Dieck, Transformation groups and representation theory, Lecture Notes in Math., vol. 766, Springer-Ver lag, Berlin and New York, 1979. 4. A. Dress, Contributions to the theory of induced representations, Lecture Notes in Math., vol. 342, pp. 183-240. Springer-Ver lag, Berlin and New York, 1973, pp. 183-240. 5. Z. Fiedorowicz, H. Hauschild, and J. P. May, Equivariant algebraic K-theory, Proc. Oberwolfach algebraic JC-theory conference, 1980 (to appear). 6. H. Hauschild, J. P. May, and S. Waner, Equivariant infinite loop space theory (to appear). 7. S. Illman, Equivariant singular homology and cohomology. I, Mem. Amer. Math. Soc. no. 156, 1975. 8. J. P. May, J. McClure, and G. Triantafillou, Equivariant localization (to appear).

212

G. LEWIS, J. P. MAY AND J. McCLURE

9. J. McClure, The groups JOCr(X) (to appear). 10. G. Nishida, The transfer homorphism in equivariant generalized cohomology theories (preprint). 11. G. Triantafillou, Minimal model for the G-rational homotopy type (preprint). 12. S. Waner, Equivariant homotopy theory and Milnor's theorem, Trans. Amer. Math. Soc. 258 (1980), 351-368. 13. , Equivariant fibrations and transfer, Trans. Amer. Math. Soc. 258 (1980), 369-384. 14. , Equivariant RO{Gygraded singular cohomology (preprint). 15. K. WirthmuUer, Equivariant homology and duality, Manuscripta Math. 11 (1974), 373-390. DEPARTMENT OF MATHEMATICS, UNIVERSITY OF CHICAGO, CHICAGO, ILLINOIS 60637