Subalgebras of group cohomology defined by infinite loop spaces John Robert Hunton & Björn Schuster 30th November 2001 Abstract We study natural subalgebras ChE (G) of group cohomology defined in terms of infinite loop spaces E and give representation theoretic de- scriptions of those based on QS0 and the Johnson-Wilson theories E(n). We describe the subalgebras arising from the Brown-Peterson spectra BP and as a result give a simple reproof of Yagita's theorem that the image of BP *(BG) in H*(BG; Fp) is F -isomorphic to the whole coho- mology ring; the same result is shown to hold with BP replaced by any complex oriented theory E with a map of ring spectra E ! HFp which is non-trivial in homotopy. We also extend the constructions to define subalgebras of H*(X; Fp) for any space X; when X is finite we show that the subalgebras ChE(n)(X) give a natural unstable chromatic filtration of H*(X; Fp). 1 Introduction The main aim of this paper is to study certain natural subalgebras ChE (G) of H*(G), the cohomology of a finite group G with coefficients in Fp, which are defined using infinite loop spaces associated to a spectrum E. Though not originally defined in this manner, perhaps the first example of such a subal- gebra is that of the Chern subring, Ch(G), defined and studied in detail by Thomas [29 ] and others. The Chern subring is the subring of H*(G) gener- ated by the Chern classes of its irreducible complex representations; it is in some sense that part of group cohomology most accessible using the tools of representation theory. 1 In [12 ] the construction ChE (G) was introduced for a group G and a spec- trum E and was shown to give a subalgebra of H*(G), natural in G and closed under the action of the Steenrod algebra. We review this construction and de- velop further some of its basic properties in Section 2. Taking E to be the spectrum representing complex K-theory, ChK (G) was shown in [12 ] to coin- cide with Thomas' Chern subring Ch(G). The main goal of [12 ] however was to study the subalgebras Ch[E(n)(G) based on Baker and Würgler's spectra [E(n), n = 1, 2, . .,.see [4, 5], and to give these subalgebras a representation theoretic description as well. The main tools used were a `Hopf ring' analysis of the spaces in the -spectra for [E(n) [18 , 20, 33], the generalised charact* *er theory describing (E[(n))*(BG) of [17 ] and the work of Green and Leary [11 ] on the varieties associated to subrings of H*(G). The reinterpretation of Ch(G) as ChK (G) was similar, though easier, using the well known descriptions of the cohomology of BU and Atiyah's theorem on K*(BG) [2]. Despite the work of [12 ], a number of basic aspects of the subalgebras ChE (G) remained mysterious. In particular, it was not clear exactly how the subalgebras were related when the spectrum E was changed, even to a closely related spectrum, nor was it clear what subalgebras were produced when some very basic examples of spectra were taken, such as the Brown- Peterson spectra BP , the (uncomplete) Johnson-Wilson theories E(n) [21 ] or the universal example QS0. When a spectrum E has an associated theorem describing E*(BG) as a functor from groups to algebras, just as Atiyah's theorem [2] identifies K*(BG) with a certain completion R[(G) of the complex representation ring, it is rea- sonable to suppose there is a related theorem describing ChE (G), provided a good enough understanding of the spaces in the E-theory -spectrum can be established. This was the case for the [E(n) spectra in [12 ] using [17 ] and [18 ]. In a similar way, we show below that a description of ChQS0 (G) can be made using the Segal conjecture [9] and classical work on the homology of QS0 [25 ]. Perhaps of more interest though are the cases of subalgebras based on BP or E(n) as these spectra have at present no associated character the- ory or analogue of Atiyah's, Carlsson's or Hopkins-Kuhn-Ravenel's theorems [2, 9, 17]. 2 The case of BP -theory turns out to yield results through one further prop- erty of the ChE (G) construction, not considered in [12 ]. By its nature, ChE (* *G) is essentially a construction in unstable homotopy, and as such there is an ass* *o- ciated notion of stabilisation. Using this idea we show that ChBP (G) is in fact F -isomorphic to H*(BG). Moreover, the same is true for ChE (G) whenever E is a complex oriented spectrum with a map of ring spectra :E ! HFp which is onto in homotopy. A corollary of this is a very simple reproof of a theorem of Yagita [34 ] that says that the map *: BP *(BG) ! H*(BG; Fp) is an F -epimorphism. This work is described in Section 3. Our results also show that the more interesting subalgebras ChE (G) are going to be those arising from spectra E which are either non-complex oriented (such as QS0) or else are complex oriented but periodic, such as K-theory or the Johnson-Wilson theories E(n). The case of QS0, already indicated, is detailed in Section 4. In Section 5 we examine how ChE (-) varies as a functor when the spectrum E is changed, con- centrating in particular on changing from E to an exact E-module spectrum, though even this simple case is more complicated than might be thought. Suf- ficient is proved to show that, for finite groups G, the subalgebras ChE(n)(G) and Ch[E(n)(G) are equal, and so the representation theoretic description of Ch[E(n)(G) in [12 ] applies also to the subalgebras ChE(n)(G). It should be noted that the generalised character theory of [17 ] applies only to the com- pleted spectra; this suggests that the results of [12 ] ought perhaps be indepe* *n- dent of the work of Hopkins, Kuhn and Ravenel [17 ]. Much of the work of this paper applies to define subalgebras ChE (X) of H*(X) for an arbitrary space X, not just for spaces of the form X = BG for a finite group G. It is useful and in fact necessary in some sections to work in such generality: we do so as far as possible throughout. In the final section we concentrate on the case where X has the homotopy type of a finite CW complex. Here the spectra E(n) allow the definition of a new `unstable' chromatic filtration of H*(X), compatible with the standard stable chromatic filtration of Ravenel and others (see, for example, [27 , Def. 7.5.3]) defined * *in terms of Bousfield localisation. For X the classifying space of a finite group, a residue of this result exists at the level of varieties. 3 Notation conventions Throughout p will denote a fixed prime and H*(G) or H*(X) denotes the group or singular cohomology of G or X respectively with constant coefficients in the field Fp of p elements. When E is a p-local spec- trum, such as BP , E(n) and so on, the prime considered will be assumed to be the same one. We identify H*(G) (group cohomology of G) with H*(BG) (singular cohomology of its classifying space), generally using the former nota- tion except when desiring to stress the topological nature of the construction. By the variety of an Fp-algebra R, denoted var(R), we mean the set of algebra homomorphisms from R to an algebraically closed field k of characteristic p, topologised with the Zariski topology. In general R will be graded, and as discussed in [11 ] and [12 ] var(R) will be homeomorphic to var(Reven) (even when p = 2). Acknowledgements Both authors are pleased to acknowledge the support of the London Mathematical Society who, through a Scheme 4 grant, enabled the second author to visit the University of Leicester where much of this research was carried out. The first author also gratefully acknowledges the support of a Leverhulme Research Fellowship. 2 The basic construction We shall suppose throughout that E denotes a spectrum [1] representing a generalised cohomology theory E*(-) and for convenience shall suppose the coefficients E* = E*(point) are concentrated in even degrees. Moreover, we shall follow [20 , 28 ] and write E_r for the rth space in the spectrum for E, so E_r = 1-r E and E_r represents E-cohomology in degree r, that is Er(X) = [X, E_r] for any space X. For reasons discussed further in [12 ] it is convenient to restrict attention to the even graded spaces E_2r, and we shall do so throughout. (In Section 4 we restrict further to the single space E_0.) Note that each E_ris an H-space with product E_rx E_r! E_rrepresenting addition in E-cohomology and arising from the loop space structures E_r+1 = E_r; moreover, each E_r is an infinite loop space. 4 Definition 2.1 For a spectrum E and space X, define ChE (X) to be the subalgebra of H*(X) = H*(X; Fp) generated, as an Fp-algebra, by elements of the form f*(x) where f 2 E2r(X) (so f may be considered as a map f :X ! E_2r) and x a homogeneous element in H*(E_2r). Remark 2.2 An element of Hk(X) may of course be represented by a (ho- motopy class of a) map X ! H_k where H_k denotes an Eilenberg-Mac Lane space of type K(Fp, k). It is thus equivalent to declare that ChE (X) is the subalgebra of H*(X) generated by maps X ! H_k which have a factorisation X -f! E_2r -x! H_k for some r. Remark 2.3 It is immediate that this construction behaves well with respect ` Q to disjoint unions of spaces and ChE ( iXi) = iChE (Xi). With this obser- vation we shall largely restrict ourselves to connected spaces X. Remark 2.4 With the aid of the Cartan formula and the fact that ChE (X) by definition is closed under ring operations, it is not hard to check that the subalgebra ChE (X) is closed under the action of the Steenrod algebra. Example 2.5 ([12 ] Prop. 1.6) Let G be a compact Lie group and write ChE (G) for ChE (X) with X = BG. Write K for the complex K-theory spectrum. Then ChK (G) is equivalent to the classical Chern subring [29 ] of H*(G) generated by Chern classes of irreducible representations. Corollary 2.6 For G a finite group and E a Landweber exact theory [23 ], ChE (G) is a finitely generated Fp-algebra and H*(G) is a finitely generated ChE (G) module. Proof We see from [12 , Prop. 2.8] that if E is Landweber exact and X is any space, there is an inclusion ChK (X) ChE (X). As we can identify (2.5) ChK (G) with the Chern subring, over which we know [29 ] H*(G) is finitely generated as a module, the inclusion ChK (X) ChE (X) immediately tells us 5 that H*(G) is a finitely generated module over ChE (G). The finite generation of ChE (G) as an Fp-algebra now follows from basic commutative algebra [3, Prop. 7.8] and the finite generation of H*(G) as an Fp-algebra [10 , 26, 30, 31* *]. In general however, there seem at first sight to be potentially a lot of maps f :X ! E_2rthat need to be considered in the construction of ChE (X). The following technical lemma usefully cuts down on othe quantity of the maps needed. For a CW complex X we write X(m) for its m-skeleton and recall Milnor's short exact sequence 0 -! lim1E2r-1(X(m)) -! E2r(X) -! limE2r(X(m)) -! 0 . Recall also that the skeleta {X(m)} define a topology (which we shall refer to as the skeletal topology on the group lim E2r(X(m)) where the open neigh- bourhoods of zero are given by i j Fm E2r(X) = ker E2r(X) -! E2r(X(m)) . Lemma 2.7 Suppose X is a CW complex. Then ChE (X) is generated by elements f*(x) where f :X ! E_2rrun over a set of topological generators of the E*-module Eeven(X) modulo phantom maps. Proof First note that we can disregard any phantom map f :X ! E_2r. Suppose such a map gave rise to an element f*(x) 2 Hk(X) via some element x 2 Hk(E_2r). As Hk(X) ~=Hk(X(m)) for any m > k+1 with the isomorphism induced by restriction æ :X(m) ! X, the element f*(x) is non-zero if and only if the composite X(m) -fj!E_2r -x! H_k is non-zero. However, by definition of f being a phantom map, fæ is trivial and we are just picking up the zero element. We can thus restrict the elements f we need to those having non-zero image in lim Eeven(X(m)). 6 Next observe that is is sufficient to consider maps f :X ! E_2r which are E*-module generators of Eeven(X). The E*-action on E*(X) may be represented by maps ~ :E2t x E_2r! E_2(r+t) where the group E2t is regarded as a space with the discrete topology. Now consider an element of ChE (X) given by (fff)*(x) where f 2 E2r(X) and ff 2 E2t. The map fff is represented by the composite X -! X x X ffxf-!E2t x E_2r-~! E_2(r+t) but this is equivalent to a composite X -g! {ff} x E_2r~=E_2r -~! E_2(r+t) where we write ~ also for the restriction of ~ to this specific component. Thus we have the equality (fff)*(x) = g*(x~) . Meanwhile, addition in E2r(X) is represented by a map oe :E_2rx E_2r-! E_2r. Consider an element of ChE (X) given by (f1 + f2)*(x) with fi 2 E_2r and x 2 Ht(X). Then this element can be written as the composite X -! X x X f1xf2-!E_2rx E_2r-ff!E_2r -x! H_t. P If we write oe*(x) as x0 x002 H*(E_2r) H*(E_2r) this shows that (f1 + P f2)*(x) = f*1(x0)f*2(x00), an element in the ring generated using just f1 and f2. Finally, we show that we need only consider topological generators of Eeven(X) mod phantom maps. Once again consider a generator f*(x) 2 ChE (X) and let x 2 Hk(E_2r). Suppose f 2 E2r(X) is represented by a sequence of elements fm 2 E2r(X), m = 1, 2, . .,.where f - fm 2 Fm E2r(X) and fm is in the (algebraic) E*-span of a given set of topological generators. As Hk(X) ~= Hk(X(m)) for all m > k + 1, the elements f*(x) and (fm )*(x) are identical for all m > k + 1. This concludes the proof. 7 Remark 2.8 If E is a ring spectrum and so E*(X) is an E*-algebra, a similar argument to the one above for sums shows that we can reduce the maps f 2 E*(X) further to include only topological ring generators of E*(X) mod phantoms. Warning 2.9 The result (2.7) does not imply that the ChE (X) may be com- puted for a CW complex X in terms of ChE (X(m)). While an element of ChE (X) certainly gives rise, by restriction, to an element of lim ChE (X(m)), it is possible to have a tower of elements in lim ChE (X(m)) which do not lift to ChE (X). We conclude this section with some remarks about the relation of the subalgebra ChE (X) to the Bousfield E-localisation of X. Recall [7] defines for a generalised homology theory E*(-) a localisation functor LE with a natural transformation jX : X 7! LE X satisfying certain universal properties. Proposition 2.10 For any space X, the subalgebra ChE (X) H*(X) is contained in the image of j*X: H*(LE X) ! H*(X). Proof The spaces E_2r are themselves E-local. Hence any map f :X ! E_2r factors through the localisation jX : X 7! LE X. Thus any generator of ChE (X) lies in j*X. As j*X is an algebra homomorphism it is closed under sums and products and the result follows. Remark 2.11 The inclusion ChE (X) Im (j*X) of (2.10) will usually be strict. For example, if E is any Landweber exact spectrum, H*(E_2r) lies entirely in even dimensions [6] and hence ChE (X) Heven(X); there are many examples where Im (j*X) contains odd dimensional elements, for example if we take X = BZ=p and E = BP . However, even the inclusion ChE (X) Im (jevenX) for such spectra will also likely be strict. For an example here, let X be the classifying space of an elementary abelian group of rank 3 then (as X is then a nilpotent space) [7] tells us that LBP X is just the p-localisation of X and so Im (jevenX) is the w* *hole of Heven(X). We shall see by Corollary 3.5 that ChBP (X) can be identified with the image of BP *(X) under the Thom map, and for this particular X this image is not the whole of Heven(X). 8 3 Stabilisation and BP -theory There is no reason to suppose that the maps x :E_2r! H_k used in the defini- tion of ChE (X) commute with the loop space structure on these spaces. This observation allows us to make the following definition. Definition 3.1 Define the subalgebra dChE (X) of ChE (X) H*(X) to be that generated by elements f*(x) where x :E_2r ! H_k is a d-fold loop map, i.e., there is map y :E_2r+d = -d E_2r! H_k+d = -d H_k with x = dy. Likewise, define ChsE(X), the stable E-Chern subalgebra, to be the intersection of all dChE (X), that is the subalgebra generated by elements f*(x) where x :E_2r! H_k is a map of infinite loop spaces. Remark 3.2 The process of stabilisation may yield only the trivial subalge- bra. For example, the K-theory subalgebra ChK (X) is in general non-trivial (for example, taking X to be BZ=p, we have ChK (BZ=p) = Heven(BZ=p)). However, there are no non-trivial stable maps K ! HFp and thus ChsK(X) = 0 for all X. Similar remarks apply with K replaced by any of the E(n) spectra. Now suppose the spectrum E is a ring spectrum and has a Thom map, by which for the moment we just mean a distinguished map of ring spectra :E ! HFp which is onto in homotopy. Our principal examples are E = BP or MU, but we also include spectra such as k(n), P (n) and BP ; the periodic spectra such as K(n) and E(n) are obviously not included. We write also for the corresponding individual infinite loop maps E_2r! H_2r. Definition 3.3 Define the subalgebra ChE (X) of H*(X) to be that generated by elements of the form f*( ) where f 2 E2r(X) and is the appropriate infinite loop map, as just given. Of course this is just the image of E*(X) in H*(X) under the ring homomorphism *. In this situation we have natural inclusions ChE (X) ChsE(X) ChE (X) H*(X). 9 Before the next result let us remark again that until otherwised mentioned, all spectra such as the E and T following are supposed to have coefficients concentrated in even degrees. We recall the notion of Landweber exact spectra [23 ]. This class of spectra include complex cobordism MU and the Brown- Peterson theories BP , the Johnson-Wilson spectra E(n) and their various completions such as [E(n) and Morava E-theory, it also contains complex K- theory and many examples of elliptic spectra. The topology of the spaces in the spectrum for a Landweber exact theory is well understood [6, 16, 18, 20, 28]; in particular, H*(E_2r; Z) is torsion free and concentrated in even dimensions. Proposition 3.4 Let E be any Landweber exact spectrum and suppose T to be a ring spectrum with Thom map. Then for any space X there is an inclusion of subalgebras ChE (X) ChT (X) . Proof The Thom map :T ! HFp induces a map of Atiyah-Hirzebruch spectral sequences for each space E_2rwhich on E2-pages is a surjection H*(E_2r; T *) -! H*(E_2r; Fp) . By the properties of H*(E_2r; Z) for a Landweber exact spectrum E mentioned above, both these spectral sequences collapse and we conclude that any map x :E_2r! H_k can be lifted through the Thom map E_2r-!l T_k -! H_k. (Note that only even values of k arise as H*(E_2r) is concentrated in even dimensions* *.) Thus any element f*(x) 2 ChE (X) with f :X ! E_2rand x 2 Hk(E_2r) may be written as (lf)*( ) 2 ChT (X). The spectra MU and BP each satisfy the hypotheses on both E and T in Proposition 3.4 (necessarily assuming the prime p of the underlying field Fp is the same as that for the version of BP considered). As ChT (X) ChT (X), by choosing both E and T to be BP (and likewise choosing both to be MU), (3.4) gives us 10 Corollary 3.5 For any space X we have equalities ChBP (X) = ChsBP(X) = ChBP (X) ChMU (X) = ChsMU (X) = ChMU (X) . Remark 3.6 This result should be contrasted with Remark 3.2 which, with Corollary 2.6 observes that for E any of the periodic Landweber exact spectra listed above, the subalgebras ChE (X) and ChsE(X) are quite different - the former, for X = BG, containing the whole Chern subring, the latter being trivial. Choosing one of E or T to be BP and the other to be MU, Proposition 3.4 tells us Corollary 3.7 For X any space, ChBP (X) = ChMU (X). Theorem 3.8 Let T be a complex oriented ring spectrum with a Thom map and suppose G to be a finite group. Then the inclusions ChT (G) H*(G) and ChT (G) H*(G) are F -isomorphisms. Hence, by the first observation in (3.3), for such T the map *: E*(BG) ! H*(G) is an F -epimorphism. We offer two proofs of this result, both resting on the stabilisation results j* *ust proved. The first proof uses also the main theorem of [12 ], while the second uses instead work of Carlson [8] or of Green and Leary [11 ] on corestriction of Chern classes. First Proof By Proposition 3.4, for every n = 1, 2, . .,.there are inclusions Ch[E(n)(G) ChT (G) ChT (G) H*(G) . However, by [12 , Theorem 0.2] the inclusion Ch[E(n)(G) H*(G) is an F - isomorphism if n is not less than the p-rank of G. Second Proof By Proposition 3.4 there are inclusions ChK (G) ChT (G) ChT (G) H*(G) . 11 ______ We follow [11 ] and write R(G) for the closure of a functorial subring R(G) H*(G) under corestriction of elements from R(H) for all subgroups H of G. _________ By [11 , x8] or [8] (see [11 ] for detailed discussion), ChK (G) is F -isomorp* *hic _________ to H*(G), implying ChT (G) is also F -isomorphic to H*(G). However, as _________ corestriction is a stable construction, ChT (G) = ChT (G) and so ChT (G) and hence ChT (G) is F -isomorphic to H*(G). Remark 3.9 If we take T = BP or k(n) in Theorem 3.8 we recover Yagita's results [34 , (4.2), (4.3)]. As noted at the end of the previous section, the inclusions ChBP (G) Heven(G) can however be strict, for example when G is elementary abelian of rank 3. In the next section we shall show that the result (3.8) can fail when T is not complex oriented. Corollary 3.10 Suppose T is a complex oriented ring spectrum with a Thom map and G is a finite group. Let A(G) be the category of elementary abelian subgroups of G with morphisms those inclusions V ! W given by conjugation by an element of G. Then there is a homeomorphism of varieties var(ChT (G)) ~= colim var(H*(V )) . V 2A(G) Proof This follows from Theorem 3.8 which identifies var (ChT (G)) with var(H*(G)) and Quillen's theorem [26 ] which describes var(H*(G)) as the colimit of the H*(V ) over the category indicated. 4 Subalgebras defined by QS0 Suppose throughout this section that G is a finite group. The main result below identifies the subalgebra ChQS0 (G) of H*(G) by which we mean the subalgebra of H*(G) generated by the images in cohomology of maps BG ! QS0. Recall that QS0 = nlim!1 n nS0 and is the infinite loop space that represents stable cohomotopy in degree zero, i.e., ß0s(X) = [X, QS0]. We restrict attention here to the zero space QS0 as, by the Segal conjecture [9], there are no non-trivial maps BG ! QSn for n > 0. The cases n < 0 present formidable difficulties. 12 We begin by recalling the algebra Sh = Sh(G) H*(G) of [13 , Def. 1.2]. For a finite G-set X of cardinality n, a choice of bijection between X and the set {1, . .,.n} induces a homomorphism æX : G ! n where n is the symmetric group on n letters. It thus induces an algebra homomorphism æ*X: H*( n) ! H*(G). For a fixed space X two choices of æX differ only by an inner automorphism of n and so the homomorphism æ*X: H*( n) ! H*(G) depends only on X and not on the choice of bijection. Then Sh = Sh(G) is defined as the subalgebra of H*(G) generated by elements of the form æ*X(x) as X runs over all finite G sets and where the x are homogeneous elements of H*( n). Theorem 4.1 For G finite there is an algebra isomorphism ChQS0 (G) ~= Sh(G). Proof First we show that Sh(G) is a subalgebra of ChQS0 (G). By definition, Sh(G) is generated by elements of the form f*(x) where f is drawn from some particular class of maps BG ! B n and x 2 H*( n). Given such an element, consider the composite BG -f! B n -in!B 1 -D! QS0 where in is induced by the incusion of n in the infinite symmetric group, and D is the Dyer-Lashof map. By [25 ] D induces an isomorphism in cohomology, while by Nakaoka [24 , x7] the induced map i*n:H*(B 1 ) ! H*(B n) is a surjection. Thus there is an element y 2 H*(QS0) with x = (Din)*(y). Hence f*(x) = (Dinf)*(y) 2 ChQS0 (G). To show the converse we use the Segal conjecture [9]. Recall that this identifies the homotopy classes of maps BG ! QS0 with the elements of the Burnside ring, completed at the augmentation ideal. Specifically, to an element of the Burnside ring A(G) we can associate a finite G-set X; the Segal conjecture tells us that the map which sends this element to the resulting composite BG -f! B n -in!B 1 -D! QS0 13 (where again, n is the cardinality of X) is continuous (with respect to the topologies induced by powers of the augmentation ideal in A(G) and the skeleta in ß0s(BG)) and an isomorphism after completion; the ring ß0s(BG) is already complete under the skeletal topology. Thus we can take topological generators of ß0s(BG) from the ring A(G) itself and any element g*(y) 2 ChQS0 (G) rep- resented by such an element g 2 ß0s(BG) and y 2 H*(QS0) can be realised as an element f*(x) 2 Sh(G) where x = (Din)*(y). Lemma 2.7 now shows that this is sufficient to prove that ChQS0 (G) is contained in Sh(G). Remark 4.2 A representation theoretic description of the variety associated to Sh(G) is given in [13 ]. The work there combined with Theorem 4.1 shows there is a natural homeomorphism var(ChQS0 (G)) ~= colim var(H*(V )) V 2Ah(G) where Ah(G) denotes the category whose objects are elementary abelian sub- groups of G and whose morphisms are those injective group homomorphisms f :V ! W for which f(U) is conjugate in G to U for every subgroup U of V . Remark 4.3 It is also noted in [13 ] that in general neither Sh(G) nor Ch(G), the classical Chern subring, are contained in the other. See [13 ] for worked examples. 5 Subalgebras defined by exact spectra In this section we examine the relation between ChE (X) and ChF (X) for spectra E and F that are themselves related; specifically, we are interested in the case where F is exact over E, by which we mean that E is a ring spectrum, F an E-module spectrum and the corresponding homology theories are related by a natural equivalence F*(X) ~=F* E*(X) E* for any space X. 14 It is of course not true that if F is exact over E then ChE (X) and ChF (X) are identical: for example, complex K-theory is exact over BP -theory, but the inclusion ChK (X) ChBP (X) is certainly strict in many cases. However, exactness of F over E is sufficient, under suitable finiteness conditions, to g* *ive in general an inclusion ChF (X) ChE (X); see the results (5.2), (5.3), (5.4) and (5.8) below. Our main goal here though, Theorem 5.7, is to show that the complete theories [E(n) used in [12 ] are not the only ones that can give the topological interpretation of the subgroup categories A(n) of [11 , 12], and that the more classical Johnson-Wilson theories E(n) [21 ] satisfy a similar theorem. The Baker-Würgler completions [E(n), or equivalently the Morava E-theories, were necessary in [12 ] so as to apply the generalised character theory of Hopkins, Kuhn and Ravenel [17 ] and the completeness assumptions seem integral to that character theory. Thus Theorem 5.7 below has apparently by-passed the subtleties of [17 ] and suggests it ought to be possible to find a proof of Theorem 5.7 without reference to [17 ]; it would be interesting to see such a direct argument. Throughout this section we shall not only continue the convention that any subalgebra ChE (X) or ChF (X) is defined using just the even graded spaces in the E or F -spectra, but we shall further restrict these subalgebras to those lying in even dimensional cohomology, that is, by ChE (X) we shall really mean ChE (X) \ Heven(X), etc. This convention is made for notational simplicity and could be avoided if one wanted. However, as the main examples we wish to discuss here are of Landweber exact spectra, we see by [6] or by [20 , 28] t* *hat this is not actually a restriction at all: there are no odd dimensional elements in the original subalgebras ChE (X) for such E. We begin with the following general observation about the subalgebras ChE (X) and ChF (X) for arbitrary spectra E and F (not necessarily exact or even one a module spectrum over the other) arrived at by considering the universal example of X = F_2*. Proposition 5.1 Suppose the cohomology of each space Heven(F_2r) is of fi- nite type and suppose ChE (F_2r) = Heven(F_2r) for each r. Then ChF (X) ChE (X) for every space X. 15 Proof Consider a generator f*(x) of ChF (X), where f :X ! F_2r and x 2 H2t(F_2r). As ChE (F_2r) = Heven(F_2r), the finiteness assumption means that we can write x as a finite sum of products of elements g*i(yi) 2 ChE (F_2r), wi* *th gi: F_2r! E_2siand yi 2 H*(E_2si). The element f*(x) may thus be written as the composite Y Q gi Y Y X -f! F_2r -! F_2r- ! E_2si-! H_2ti-! H_2t i i i where the last arrow denotes the summing and multiplication operations. The Q composite of the final two arrows denotes an element of H2t( iE_2si) and as such we can write, using the Künneth theorem for mod p cohomology, in the P form y(1) . . .y(q)(where the sum is potentially infinite). If we write fi Q for the projection of the composite X -f! F_2r-! iE_2sito the ith factor, we see that X f*(x) = f*1(y(1)) . .f.*q(y(q)) , an element of ChE (X): note that this sum will contain only a finite number of non-zero terms by virtue of the finite type assumption on Heven(F_2r), even P Q if the first sum y(1) . . .y(q)2 H*( iE_2si) is infinite. Proposition 5.2 Suppose F is an exact E-module spectrum and that the space X is of finite type. Then ChF (X) ChE (X). Proof Assume X is connected, for else we can argue by connected compo- nents. If X had the homotopy type of a finite CW complex then F *(X) = F * E*(X); if X is only of finite type then this statement is only true after E* completion of the tensor product. The finite type hypothesis however tells us that the completion is with respect to the skeletal topology (as in Section 2): ` ' F *(X) = F *c E*(X) = lim F * E*(X(m)) . E* m!1 E* Thus topological generators of F *(X) may be taken from elements of the (uncompleted) tensor product F * E*(X). Such an element, considered as E* a map X ! F_2r, may thus be lifted as a finite sum of maps through spaces 16 F *x E_2si, where F *is regarded as a discrete space, and thus we can represent f as a composite Y X -! (F *x E_2si)- ! F_2r. i By the connectedness assumption, such a map actually lifts through a finite product of spaces E_2si. Now consider a generator f*(x) 2 ChF (X) as usual. Factoring f as above Q as a map X -! iE_2si-! F_2rtogether with the finite type assumption on X now shows f*(x) to be in the subalgebra ChE (X). Corollary 5.3 If F is an exact E-module spectrum and the spaces F_2r rep- resenting F are of finite type, then ChF (X) ChE (X) for every space X. Proof We apply Proposition 5.2 with X = F_2rand deduce that ChF (F_2r) ChE (F_2r) H*(F_2r). But ChF (F_2r) = H*(F_2r) by definition of ChF (-) and hence ChE (F_2r) = H*(F_2r). The result now follows from Proposition 5.1. Remark 5.4 There are a number of examples to which this corollary may be applied. First of all, we could take E = MU and F = K and we recover the inclusion ChK (X) ChMU (-) of Propositions 2.5 and 3.4. New results however include the inclusions ChKO (X) ChMSp (X) ChKO (X) ChMSpin (X) ChKO (X) ChEl(X) ChK (X) ChMSpinc (X) ChEl(X) ChMSpin (X) where MSpinc is the self-conjugate spin cobordism of [14 ] and El is the (non- complex oriented) integral elliptic theory of Kreck and Stolz [22 ]; see [15 ]. With care, the arguments above may be extended to prove also the inclusion ChK(n)(X) ChP(n)(X) using Yagita's In version of the Landweber exact functor theorem [35 ]. Lemma 5.5 Suppose E and F are spectra equipped with a map of spectra OE :E ! F inducing an inclusion in mod p homology of the corresponding -spectra, H*(E_*) ,! H*(F_*). Then ChE (X) ChF (X) for any space X. 17 Proof If OE induces an injection in homology, it induces a surjection H*(F_*) i H*(E_*) in cohomology. Thus, given a generator f*(x) of ChE (X) with f :X ! E_2rand x 2 H2t(E_2r), the element x may be lifted as x = OE*(y) for some y 2 H2t(F_2r) and f*(x) can be represented as a composite X -f! E_2r -ffi!F_2r-! H_2t, an element of ChF (X). This result will give us a converse inclusion to that offered by the previous results. Of course there is no reason to suppose that even if F is exact over E there should be an inclusion of the homology of theory -spectra. For example, K is an exact BP -module spectrum but the map H*(BP__*) ! H*(K_*) is not an inclusion. Thus we cannot conclude from (5.4) the equality of ChBP (-) and ChK (-), but then, as noted above, they are not equivalent. We can however apply these results to the case of E = E(n) and F = [E(n): we know [4, 5] that [E(n) is an exact spectrum over E(n) and it is shown in [18 ] that the map H*(E(n)_*) ! H*(E[(n)_*) induced by the completion is an inclusion. We thus obtain by (5.2) and (5.5) Corollary 5.6 For X with the homotopy type of a finite type CW complex, there is an equivalence ChE(n)(X) = Ch[E(n)(X) . Recall the categories A(n) = A(n)(G) of elementary abelian subgroups of a group G defined in [11 ]: the objects of A(n)(G) are the elementary abelian subgroups of G, denoted W , V , etc., and the morphisms from W to V are those injective group homomorphisms f :W ! V with the property that if U is any subgroup of W of rank at most n, the restriction of f to U is given by conjugation by some element of G. Corollary 5.6, together with the main result of [12 ], now gives 18 Theorem 5.7 For a finite group G there is a homeomorphism of varieties var(ChE(n)(G)) ~= colim var(H*(V )) . V 2A(n)(G) We can now also explain the relationship between ChE(1)(X) and ChK (X). For X the classifying space of a finite group it was shown in [12 ] that Ch[E(1* *)(X) and ChK (X) are F -isomorphic, the latter being the classical Chern subring. The former, by (5.6), is equivalent to ChE(1)(X). Proposition 5.8 For any space X the subalgebras ChE(1)(X) and ChK (X) of H*(X) are equal. Proof Write KZ(p)for p-local complex K-theory; the localisation map Z ! Z(p)induces a map of ring spectra K -!l KZ(p)and in fact KZ(p)is exact over K via this map. Moreover, a routine calculation shows that the induced map H*(K_*) -l*!H*(KZ(p)_*) is an inclusion (for example, the calculations of the form of [19 , x3] show that in positive dimensions this is an isomorphism). Thus it follows from Corollary 5.3 and Lemma 5.5 that ChK (X) = ChKZ(p)(X) for all X. However, KZ(p) splits into a sum of suspensions of copies of E(1) and in fact is exact over E(1); moreover, the splitting gives the necessary inclusion of homology to allow us to use (5.3) and (5.5) again to prove that ChE(1)(X) = ChKZ(p)(X). 6 Unstable chromatic filtrations In [12 ] it was shown for a finite group G that the set of subalgebras Ch[E(n)(* *G) (which by Corollary 5.6 we can now identify with ChE(n)(G)) formed an "F - filtrationö f H*(G), in the sense that their associated varieties formed a se- quence of quotient spaces var(H*(G)) i . .i.var(Ch E"(n+1)(G)) i var(Ch [E(n)(G)) i . .i.k . 19 The exact relation between the subalgebras ChE(n)(G) and ChE(n+1)(G) is however uncertain: there is no reason to suppose that there is an actual in- clusion ChE(n)(G) ChE(n+1)(G). In this section we examine the subalgebras ChE(n)(X) for a space X with the homotopy type of a finite CW complex. Here we have a stronger result which shows the subalgebras ChE(n)(X) form an actual filtration of the algebra Heven(X); by the work of Section 2 this is as subalgebras over the Steenrod algebra and compatible with the (unstable) chromatic filtration of the space X (cf. [27 , Def. 7.5.3] and see Remark 6.2 below). Our main result is as follows. Theorem 6.1 Let X be a finite CW complex. Then for every n = 1, 2, . . . there is an inclusion ChE(n)(X) ChE(n+1)(X) as subalgebras of Heven(X). Proof We use the associated spectra BP and Wilson's Splitting Theorem of [32 ]. Specifically, the spectrum BP has homotopy Z(p)[v1, . .,.vn] with vi of dimension |vi| = 2(pi- 1), and for r < (pn + . .+.p + 1) BP__2r' BP__2rx BP__2r+|vn+1| as H-spaces [32 , Corollary 5.5]. Recall also that we can define the spectrum E(n) as v-1nBP and the space E(n)_2rcan be constructed as the homotopy colimit of BP__2r-vn!BP__2r-|vn|vn-!BP__2r-2|vn|-! . . . where the maps, as indicated, represent in homotopy multiplication by vn. Now consider a generator of ChE(n)(X) which we take to be represented by f*(x) for a map f :X ! E(n)_2rand an element x 2 H*(E(n)_2r). It suffices to show that f*(x) 2 ChE(n+1)(X). Viewing E(n)_2r as the homotopy colimit as mentioned, compactness of X means that the map f factors through some in- termediate space BP__2r-k|vn|for k sufficiently large. Perhaps taking k to be even larger, view this space as a factor in BP__2r-k|vn|= BP__2r-k|vn* *|x BP__2r-k|vn|+|vn+1|by the Wilson splitting theorem. Then we have a fac- torisation of f as X -f1!BP__2r-k|vn|-f2!E(n)_2r 20 where f2 is projection on the first factor followed by localisation with respect to powers of vn. Note that f*2(x) will be an element of the form y 1 in H*(BP__2r-k|vn|) = H*(BP__2r-k|vn|) H*(BP__2r-k|vn|+|vn+1|) . Now consider the localisation of BP__2r-k|vn|with respect to powers of vn+1 . For s sufficiently small, the map BP__s = BP__s x BP__s+|vn+1| ?? ?yvn+1 BP__s-|vn+1| = BP__s-|vn+1|x BP__sx BP__s+|vn+1| is the inclusion as the last two factors. In particular (again assuming s suffi- ciently small), the vn+1 -localisation map BP__s- ! E(n_+_1)_s gives an epimorphism in cohomology. We thus have a diagram X -f1!BP__2r-k|vn|-f2!E(n)_2r, ?? ?yg E(n_+_1)_2r-k|vn| where g is the vn+1 localisation map, and an element z 2 H*(E(n_+_1)_2r-k|vn|) with g*(z) = f*2(x). Then f*(x) = (gf1)*(z) 2 ChE(n+1)(X), showing ChE(n)(X) to be contained in ChE(n+1)(X) as claimed. Remark 6.2 Recall the ä lgebraic" chromatic filtration of a finite spectrum X (for example, [27 , Def. 7.5.3]). This is given by a tower of p-local spectra L0X - L1X - L2X - . . .- X and is defined as the corresponding filtration of ß*(X) by the kernels of the h* *o- momorphisms (jX )*: ß*(X) ! ß*(LnX). Here Ln is localisation with respect to the spectra E(n). The corresponding tower in the category of spaces gives the filtration of H*(X) by images of the j*X. Proposition 2.10 shows the set 21 of subalgebras ChE(n)(X) to be compatible with this filtration, but what is not immediate for general spaces X is that the subalgebras ChE(n)(X) are nested. Theorem 6.1 shows that, under suitable finiteness conditions, there is a filtration ChE(1)(X) . . .ChE(n)(X) ChE(n+1)(X) . . .H*(X) with a levelwise inclusion in the unstable algebraic chromatic filtration. Remark 6.3 In the light of the results of this section, the equivalences pro* *ved in (5.6) and (5.8), the inclusion of (3.4) and the non-equivalences in general * *of ChE(n)(-) and ChE(n+1)(-) noted in [12 ], we are tempted to wonder whether, for suitable E and X, the subalgebra ChE (X) in fact only depends on the Bousfield equivalence class of E. In the first instance, `suitable' might inclu* *de complex oriented E and spaces X which are either finite CW complexes or else are models of BG for a finite group G References [1]J. F. Adams. Stable homotopy and generalised homology. Chicago Lec- tures in Math., Chicago Univ. Press, Chicago, 1974. [2]M. F. Atiyah. Characters and the cohomology of finite groups. Inst. Hautes 'Etudes Sci. Publ. Math. 9 (1961), 23-64. [3]M. F. Atiyah and I. G. MacDonald. Introduction to commutative algebra. Addison-Wesley, 1969. [4]A. J. Baker and U. Würgler. Liftings of formal groups and the Artinian completion of v-1nBP . Math. Proc. Camb. Phil. Soc. 106 (1989), 511-530. [5]A. J. Baker and U. Würgler. Bockstein operations in Morava K-theories. Forum Math. 3 (1991), 543-560. [6]M. Bendersky and J. R. Hunton. On the coalgebraic ring and Bousfield- Kan spectral sequence for a Landweber exact spectrum. In preparation. 22 [7]A. K. Bousfield. The localisation of spaces with respect to homology. Topology 14 (1975), 133-150. [8]Jon F. Carlson. Varieties and transfers. In: Proceedings of the Northwest- ern conference on cohomology of groups (Evanston, Ill., 1985). J. Pure Appl. Algebra 44 (1987), 99-105. [9]G. Carlsson. Equivariant stable homotopy and Segal's Burnside ring con- jecture. Ann. of Math. 120 (1984), 189-224. [10]L. Evens. The cohomology ring of a finite group. Trans. Amer. Math. Soc. 101 (1961), 224-239. [11]D. J. Green and I. J. Leary. The spectrum of the Chern subring. Com- ment. Math. Helv. 73 (1998), 406-426. [12]D. J. Green, J. R. Hunton and B. Schuster. Chromatic characteristic classes in ordinary group cohomology. Preprint at math.AT/0109019. [13]D. J. Green, I. J. Leary and B. Schuster. The subring of group coho- mology constructed by permutation representations. To appear in Proc. Edinburgh Math. Soc. [14]M. J. Hopkins and M. Hovey. Spin cobordism determines real K-theory. Math. Z. 210 (1992), 181-196. [15]M. Hovey. Spin bordism and elliptic homology. Math. Z. 219 (1995), 163-170. [16]M. J. Hopkins and J. R. Hunton. On the structure of spaces representing a Landweber exact cohomology theory. Topology 34 (1995), 29-36. [17]M. J. Hopkins, N. J. Kuhn and D. C. Ravenel. Generalized group charac- ters and complex oriented cohomology theories. J. Amer. Math. Soc. 13 (2000), 553-594. [18]J. R. Hunton. Complete cohomology theories and the homology of their omega spectra. To appear in Topology. 23 [19]J. R. Hunton and N. Ray. A rational approach to Hopf rings. J. Pure Appl. Algebra 101 (1995), 313-333. [20]J. R. Hunton and P. R. Turner. The homology of spaces representing exact pairs of homotopy functors. Topology 38 (1999), 621-634. [21]D. C. Johnson and W. S. Wilson. Projective dimension and Brown- Peterson homology. Topology 12 (1973), 327-353. [22]M. Kreck and S. Stolz. HP 2-bundles and elliptic homology. Acta Math. 171 (1993), 231-261. [23]P. S. Landweber. Cobordism and classifying spaces. In: A. Liulevi- cius, Ed. Algebraic Topology (Madison, 1970). Proceedings of Symposia in Pure Mathematics 22, 125-129, Amer. Math. Soc., Providence, R.I., 1971. [24]M. Nakaoka. Homology of the infinite symmetric group. Ann. of Math. 73 (1961), 229-257. [25]S. B. Priddy. On 1 S1 and the infinite symmetric group. In: A. Liulevi- cius, Ed. Algebraic Topology (Madison, 1970). Proceedings of Symposia in Pure Mathematics 22, 217-220, Amer. Math. Soc., Providence, R.I., 1971. [26]D. Quillen. The spectrum of an equivariant cohomology ring I, II. Ann. of Math. 94 (1971), 573-602. [27]D. C. Ravenel. Nilpotence and periodicity in stable homotopy theory. Ann. of Math. Stud. 128, Princeton University Press, 1992. [28]D. C. Ravenel and W. S. Wilson. The Hopf ring for complex cobordism. J. Pure Appl. Algebra 9 (1977), 241-280. [29]C. B. Thomas. Characteristic classes and the cohomology of finite groups. Cambridge Studies in Advanced Mathematics, Cambridge University Press, 1986. 24 [30]B. B. Venkov. Cohomology algebras for some classifying spaces. Dokl. Akad. Nauk. SSSR 127 (1959), 943-944. [31]B. B. Venkov. Characteristic classes for finite groups. Dokl. Akad. Nauk. SSSR 137 (1961), 1274-1277. [32]W. S. Wilson. The -spectrum for Brown-Peterson cohomology part II. Amer. J. Math. 97 (1975), 101-123. [33]W. S. Wilson. The Hopf ring for Morava K-theory. Publ. Res. Inst. Math. Sci., Kyoto University 20 (1984), 1025-1036. [34]N. Yagita. Equivariant BP -cohomology for finite groups. Trans. Amer. Math. Soc. 317 (1990), 485-499. [35]N. Yagita. The exact functor theorem for BP*=In-theory. Proc. Japan Acad. 52 (1976), 1-3. The Department of Mathematics and Computer Science, University of Leicester, Un* *i- versity Road, Leicester, LE1 7RH, England. Email: J.Hunton@mcs.le.ac.uk Department of Mathematics, University of Wuppertal, Gaußstraße 20, D-42097 Wup- pertal, Germany. Email: schuster@math.uni-wuppertal.de 25