Divisibility of the stable Miller-Morita-Mumford classes Soren Galatius, Ib Madsen, Ulrike Tillmann* 7 0 0 2 n Abstract. We determine the sublattice generated by the Miller-Morita-Mumford a classes κ in the torsion free quotient of the integral cohomology ring of the sta- J i ble mapping class group. We further decide when the mod p reductions κ ∈ 9 i H∗(BΓ∞;Fp) vanish. ] T A Keywords. Mapping class group, characteristic classes, surface bundles. . h t a MR classification. 57R20, 55P47. m [ 1 v 7 1. Introduction and results. 4 2 1 Let Γs denote the mapping class group of a surface of genus g with b ordered g,b 0 boundary components and s marked points. We will supress s or b when their 7 value is zero. Gluing a disk or a torus with two boundary components to one of 0 / the boundary components induces homomorphisms h t a m (1.1) Γs ←− Γs −→ Γs . g,b−1 g,b g+1,b : v i Recall that by Harer-Ivanov’s stability theory both homomorphisms induce a ho- X mology isomorphism in dimensions ∗ with 2 ∗ +1 < g, cf. [H2], [I]. Let Γ := r ∞ a lim Γ be the stable mapping class group. g→∞ g,2 Mumford in [Mu] introduced certain tautological classes in the cohomology of moduli spaces of Riemann surfaces. Miller [Mi] and Morita [Mo] studied topological analogues: Let e ∈ H2(BΓ1 ;Z) be the Euler class of the central extension g,b (1.2) Z −→ Γ −→ Γ1 g,b+1 g,b which is induced by gluing a disk with a marked point to one of the boundary components. Define κ := π (ei+1) ∈ H2i(BΓ ;Z) i ! g,b where π is the Umkehr (or integration along the fibre) map associated to the ! forgetful map Γ1 → Γ . These correspond under the maps of (1.1) when i > 0 g,b g,b and hence define classes in H∗(BΓ ;Z). We will only be concerned with these ∞ stable classes in this paper. *The third author was supported by an Advanced Fellowship of the EPSRC. 2 SOREN GALATIUS, IB MADSEN, ULRIKE TILLMANN By the proof of the Mumford conjecture [MW], H∗(BΓ ;Q) ≃ Q[κ ,κ ,...]. ∞ 1 2 In contrast, little is known about κ in integral cohomology though it follows from i [H1] that κ is precisely divisible by 12 (cf. [MT, p. 537]). We write 1 H∗ (BΓ ) := H∗(BΓ ;Z)/Torsion free ∞ ∞ for the integral lattice in H∗(BΓ ;Q). ∞ Theorem 1.1. Let D be the maximal divisor of κ in H∗ (BΓ ). Then for all i i free ∞ i ≥ 1 B i D = 2 and D = den( ). 2i 2i−1 2i Here B denotes the i-th Bernoulli number and den is the function that takes a i rational number when expressed as a fraction inits lowest terms to its denominator. It is well-known, cf. [MSt; Appendix B], that den(B ) is the product of all primes i p such that p−1 divides 2i, and that a prime divides den(B /2i) if and only if it i divides den(B ). So in terms of their p-adic valuation the D are determined by the i i formula 1+ν (i+1) if i+1 ≡ 0 mod (p−1) p (1.3) ν (D ) = p i 0 if i+1 6≡ 0 mod (p−1), ( and D = 22 ·3, D = 23 ·3·5, D = 22 ·32 ·7 .... 1 3 5 Our Theorem 1.1 is inspired by a conjecture of T. Akita [Ak] which we also prove: Theorem 1.2. The element κ in H2i(BΓ ;F ) vanishes if and only if i + 1 ≡ i ∞ p 0mod (p−1). Remark 1.3. The divisor D of κ in H∗ (BΓ ) is not necessarily equal the i i free ∞ maximal divisor DZ of κ in integral cohomology H∗(BΓ ;Z) but only provides i i ∞ an upper bound for it. However, Theorem 1.2 gives Z p divides D ⇐⇒ p divides D , i i which was strengthened by the first author in [G2] to p2 divides DZ ⇐⇒ p2 divides D . i i It follows that for all even i and for many odd i (i = 1,5,9,13,...), D is indeed i Z Z equal to D , and one may expect that D = D for all i ≥ 1. i i i 3 Remark 1.4. The integral lattice H∗ (BΓ ) inherits a Hopf algebra structure. free ∞ The graded module of primitive elements P(H∗ (BΓ )) is a copy of Z in each free ∞ even degree, and κ is a primitive element of H∗ (BΓ ). The structure of the i free ∞ Hopf algebra H∗ (BΓ ) is not completely understood at present, but we have free ∞ the following partial results. Theorem 1.5. For odd primes p there is an isomorphism of Hopf algebras over the p-local integers Z (p) H∗ (BΓ ;Z ) ≃ H∗(BU;Z ). free ∞ (p) (p) This fails for p = 2. In fact, the squaring map ξ : H2 (BΓ )⊗F → H4 (BΓ )⊗F free ∞ 2 free ∞ 2 is not injective, so the algebra H∗ (BΓ ;Z ) is not polynomial. free ∞ (2) In outline, the proofs of the above theorems depend on previous results as fol- lows. For Theorem 1.2, the proof of the “if” part in Section 3.3 is a calculation of characteristic classes which relies on the fact that there is a map of infinite loop spaces α : Z × BΓ+ → Ω∞CP∞ (compare [T] and [MT] or Theorems 2.1 and ∞ −1 2.2 below). The “only if” part is implied by Theorem 1.1: if p divides κ then in i particular it must divide its reduction to the free part. For Theorem 1.1, we first establish a lower bound: D ≥ 2 by the “if ” part 2i of Theorem 1.2 and D ≥ den(B /2i) by a well-known relation between the κ 2i−1 i i classes and the symplectic characteristic classes for surface bundles, (here stated as Theorem 4.2). The main theorem of [MT] provides an upper bound which is tight for i even and precisely twice the lower bound for i odd. To eliminate the indeterminacy of the factor 2, the main theorem of [MW] (Theorem 2.4 below), as well as calculations from [G1] (Proposition 4.4)) and a stronger version of the main result of [MT] (given in Theorem 2.2 and proved in Section 5) are used. Theorem 1.5 is proved in Section 5.3. Given the interest in the mapping class groups also outside the topology commu- nity we have strived to make this paper as self contained as possible. In particular we have spelled out some of the more obscure parts of [MT]. 2. Spectrum cohomology and earlier results. 2.1. Spectra and spectrum cohomology. Let E = {E ,ǫ } be a CW-spec- n n trum1 in the sense of [A1]: E is a sequence of pointed CW-complexes and ǫ : n n SE → E a (pointed) isomorphism onto a subcomplex, where S(−) denotes n n+1 suspension. The associated infinite loop space is the direct limit Ω∞E = colim ΩnE n 1If one does not assume the spaces to be CW-complexes then one should assume that ǫn is a closed cofibration. 4 SOREN GALATIUS, IB MADSEN, ULRIKE TILLMANN of the n-th loop space of E ; the limit is taken over the adjoint maps ǫ′ : E → n n n ΩE . n+1 The k-th homotopy group of E is defined to be the direct limit of π (E ). It n+k n is equal to the k-th homotopy group of the space Ω∞E. In particular, the group of components of Ω∞E is the direct limit of π (E ). For α ∈ π (Ω∞E) we let Ω∞E n n 0 α be the component determined by α. In particular we write Ω∞E for the component 0 of the zero element. The homology and cohomology groups of E are Hk(E) = lim H˜k+n(E ), H (E) = colim H˜ (E ) n→∞ n k n→∞ k+n n where the limits are induced from the maps ǫ together with the suspension iso- n morphisms.2 In contrast to homotopy groups the cohomology groups of a spectrum are usually much simpler than the cohomology groups of Ω∞E. The evident evaluation map from SnΩnE to E induces maps n n (2.1) σ∗ : H∗(E) −→ H˜∗(Ω∞E), σ : H˜ (Ω∞E) −→ H (E). 0 ∗ ∗ 0 ∗ If we use field coefficients in the cohomology groups then H∗(Ω∞E) is a con- 0 nected Hopf algebra and the image of σ∗ is contained in the graded vector space PH∗(Ω∞E) of primitive elements. We shall be particularly concerned with the 0 torsion free integral homology and cohomology groups H∗ (Ω∞E) = H∗(Ω∞E;Z)/Torsion, Hfree(Ω∞E) = H (Ω∞E;Z)/Torsion. free 0 0 ∗ 0 ∗ 0 They are lattices in H∗(Ω∞E;Q) and H (Ω∞E;Q) and are dual Hopf algebras. 0 ∗ 0 Moreover, the image of σ∗ is contained in the module of primitive elements σ∗ : H∗ (E) −→ P(H∗ (Ω∞E)), free free 0 and dually σ factors over the indecomposable elements of Hfree(Ω∞E). ∗ ∗ 0 Given a pointed space X we have the associated suspension spectrum S∞X whose n-th term is SnX with infinite loop space Ω∞S∞X. There is an obvious inclusion i : X → Ω∞S∞X inducing a splitting of σ∗ (and σ ): ∗ ∗ ∗ (2.2) H∗(S∞X) −σ→ H˜∗(Ω∞S∞X) −i→ H˜∗(X) 0 is the suspension isomorphism. Ω∞S∞X is the free infinite loop space on X and satisfies the universal property that any pointed map from X to some infinite loop space Y can be extended in a unique way up to homotopy to a map of infinite loop spaces from Ω∞S∞X to Y. The spectra of most relevance to us are CP∞ and the suspension spectrum −1 S∞CP∞ of CP∞ ⊔ {+}. We recall the definition of the former. There are two + 2The k-th spectrum cohomology of E is normally defined as the group of homotopy classes of degree k spectrum maps from E to the Eilenberg-MacLane spectrum K(Z). This coincides with the above formula for Hk(E) whenever for all k the inverse system lim n→∞H˜k+n(En) satisfies theMittag-Lo¨fflercondition(see[A1]). Inparticularthiswillbethecasewhenthestructuremaps ǫ:SE →E are c(n)-connected for some function c(n) with c(n)→∞ as n→∞. 5 complex vector bundles over the complex projective n-space CPn, namely the tau- tological line bundle L and its n-dimensional complement L⊥ in CPn × Cn+1. n n Its Thom space (or one point compactification) is denoted by Th(L⊥). Since the n restriction of L⊥ to CPn−1 ⊂ CPn is equal to L⊥ ⊕C where C denotes the trivial n n−1 line bundle over CPn−1 we get a map ǫ : S2Th(L⊥ ) −→ Th(L⊥). n−1 n The spectrum CP∞ has −1 (CP∞) = Th(L⊥ ), (CP∞) = STh(L⊥ ) −1 2n n−1 −1 2n+1 n−1 and the structure map ǫ is given by the above ǫ. The associated infinite loop 2n+1 space is Ω∞CP∞ = colim Ω2nTh(L⊥ ). −1 n→∞ n−1 The inclusion of L⊥ into L⊥ ⊕L = CPn−1×Cn via the zero section of L n−1 n−1 n−1 n−1 induces a map from Th(L⊥ ) into S2n(CPn−1) and hence a map n−1 + ω : Ω∞CP∞ −→ Ω∞S∞(CP∞). −1 + This map fits into a fibration sequence (2.3) Ω∞CP∞ −ω→ Ω∞S∞(CP∞) −∂→ Ω∞S∞−1 −1 + where the right-hand term is the direct limit of ΩnSn−1 [R]. Indeed the inclusion of a fibre Cn → L⊥ induces a map S2n → Th(L⊥) and gives rise to a cofibre sequence n n of spectra S∞(S−2) → CP∞ → S∞CP∞ → S∞(S−1). (2.3) is the associated −1 + fibration sequence of infinite loop spaces. The component groups of (2.3) are 0 −→ Z π−0→(ω) Z π−0→(∂) Z/2 −→ 0 so π (ω) is multiplication by ±2, depending on the choice of generators. There is 0 a canonical splitting of infinite loop spaces (2.4) Ω∞S∞(CP∞) ≃ Ω∞S∞(CP∞)×Ω∞S∞. + We fix the generator of π Ω∞S∞(CP∞) to be the element that maps to +1 under 0 + the isomorphisms π (Ω∞S∞(CP∞)) π−0→(c) π (Ω∞S∞) d−eg→ree Z, 0 + 0 where c collapses CP∞ to the non-base point of S0. We fix the generator of π (Ω∞CP∞) so that π (ω) is multiplication by −2. 0 −1 0 2.2. Review of results used. Our divisibility result of Theorem 1.1 is based upon the following three theorems. 6 SOREN GALATIUS, IB MADSEN, ULRIKE TILLMANN Theorem 2.1. [T]. The spaces Z× BΓ+ and BΓ+(= {0} × BΓ+) are infinite ∞ ∞ ∞ loop spaces. Herethesuperscript(+)denotesQuillen’splusconstruction, cf. [B].Theproduct structure can be described as follows: We may view Γ as the mapping class group g,2 of surfaces with one incoming and one outgoing boundary component. Gluing the incoming boundary component of one surface to the outgoing component of the other defines a map Γ ×Γ −→ Γ g,2 h,2 g+h,2 andacorrespondingmapofclassifyingspacesthatmakesthedisjointunion BΓ g,2 over all g ≥ 0 into a topological monoid. Consider the map F BΓ −→ Z×BΓ+ g,2 ∞ g≥0 G that sends BΓ into the component {g} × BΓ+ by the stabilization map (1.1) g,2 ∞ followed by the map into the plus construction. The infinite loopspace structure on Z×BΓ+ is compatible with the monoidal structure on BΓ , and the induced ∞ g,2 map F ΩB( BΓ ) −→ Z×BΓ+ g,2 ∞ g≥0 G is a homotopy equivalence of loop spaces. We refer to [T] for details. To state the next result, for each prime p we pick a positive integer k = k(p) so that −k reduces to a generator of the units (Z/p2)× when p is odd. We pick k = 5 when p = 2. Write ψ−k for the self map of CP∞ that multiplies by −k on the second cohomology group. Composing with the inclusion into Ω∞S∞(CP∞) and using the loop sum we have a map 1+kψ−k : CP∞ −→ Ω∞S∞(CP∞), and, using the universal property of free infinite loop spaces, a unique extension to a self map of Ω∞S∞(CP∞), again denoted 1+kψ−k. Theorem 2.2. There are infinite loop maps α : Z×BΓ+ −→ Ω∞CP∞, µ : Ω∞S∞(CP∞) −→ (Z×BΓ+)∧ ∞ −1 p + ∞ p such that the composition ω ◦α◦µ and the self map p 1+kψ−k 0 : Ω∞S∞CP∞ ×Ω∞S∞ −→ Ω∞S∞CP∞ ×Ω∞S∞ 0 −2 (cid:18) (cid:19) become homotopic after p-adic completion. This is an improvement on the main theorem of [MT] where the map in the lower left corner had been left undetermined. For our calculations in Section 4.3 we need this map however to be zero. A proof of Theorem 2.2 is given in the final Section 5. 7 Remark 2.3. The reader is referred to [BK] for the notion of p-adic completion (also called F -completion). For connected, compact CW-complexes X and infinite p loop spaces Ω∞E of finite type one has [X,(Ω∞E)∧] = [X,Ω∞E]⊗Z , H∗((Ω∞E)∧;Z) = H∗(Ω∞E;Z ). p p p p Furthermore, note that the homotopy class of the map α in Theorem 2.2 is uniquely determined by its composition with Z×BΓ → Z×BΓ+. Indeed since Ω∞CP∞ ∞ ∞ −1 is an infinite loop space the induced map [Z×BΓ+, Ω∞CP∞] −→ [Z×BΓ , Ω∞CP∞] ∞ −1 ∞ −1 is an isomorphism. This is a standard property of the plus construction, cf. [B]. Before we give a detailed description of α in the next section, we state here the third result. Theorem 2.4. ([MW]). The map α is a homotopy equivalence. 3. Characteristic classes of surface bundles. 3.1. Universal surface bundles. The methods used in this and the surrounding papers do not use the mapping class groups directly but rather the topological groups of orientation preserving diffeomorphisms of surfaces. We briefly review the correspondence. Let F be a connected surface of genus g with b boundary circles. We write g,b Diff(F ;∂) for the topological group of orientation preserving diffeomorphisms g,b that keep (a neighborhood of) the boundary pointwise fixed. For g ≥ 2, results from [EE] and [ES] yield BΓ ≃ BDiff(F ;∂) g,b g,b so that BΓ classifies diffeomorphism classes of smooth fibre bundles π : E → X g,b with fibre F and standard boundary behavior: g,b ∂E = X ×⊔bS1, π|∂E = proj . 1 X Similarly, BΓs ≃ BDiff(F ;∂ ⊔{x ,...,x }) g,b g,b 1 s where x ,...,x are distinct interior points of F . Take s = 1. Since Diff(F ;∂) 1 s g,b g,b acts transitively on the interior of F , g,b E(F ) := EDiff(F ;∂)× F ≃ BDiff(F ;∂ ⊔{x}) ≃ BΓ1 . g,b g,b Diff(Fg,b;∂) g,b g,b g,b The forgetful map π : BΓ1 → BΓ corresponds to the universal smooth F g,b g,b g,b bundle (3.1) F −→ E(F ) −→ BDiff(F ;∂). g,b g,b g,b The central extension (1.2) is classified by “the differential at x”, Diff(F ;∂ ⊔{x}) −→ GL+(T F ) ≃ SO(2). g,b x g,b Hence the circle bundle induced from (1.2) by applying the classifying space functor corresponds tothecirclebundle oftheverticaltangent bundle associatedwith(3.1). 8 SOREN GALATIUS, IB MADSEN, ULRIKE TILLMANN 3.2. The map α and the kappa classes. The map of infinite loop spaces α : Z × BΓ+ → Ω∞CP∞ constructed in section 2 of [MT] restricts to a map ∞ −1 α : BΓ → Ω∞CP∞ that is homotopic to the composition g,2 g,2 g −1 (3.2) α : BΓ −→ BΓ α−g→+1 Ω∞CP∞, g,2 g,2 g+1 g −1 where the left hand map is induced from gluing the two parametrized boundary circles together. These maps are up to homotopy compatible with the monoidal structure on BΓ . g,2 We next recall a description of α which is well-suited for identifying the g+1 F kappa classes. Let π : E → X be a smooth surface bundle with closed fiber F. Thus E = P × F where P is a principal Diff(F) bundle over X. We do not Diff(F) assume that X is smooth or finite dimensional, only that X is paracompact (or a CW-complex). We denote by Emb(F,Rn) the space of smooth embeddings in the C∞-topology, and let R∞ and Emb(F,R∞) be the colimits of Rn and Emb(F,Rn), respectively. We shall consider fiberwise embeddings ι : E → X × R∞, that is, fiberwise maps such that each ι : E → {x} × R∞ is an embedding and such that the adjoint x x Diff(F)-equivariant map P → Emb(F,R∞) is continuous. Such an ι is equivalent to a section of P × Emb(F,R∞). Note that Emb(F,R∞) is contractible so Diff(F) that such a section always exists. An embedding ι : F ֒→ Rn+2 extends to a map from the normal bundle x Nnι = {(p,v)|v⊥T F} into Rn+2 by sending (p,v) to p+v. (Here we have identi- x p fied F withits imageunder ι .) We call the embedding ι fat if this map restricts to x x an embedding of the unit disk bundle D(Nnι ). The subspace of fat embeddings x Embf(F,R∞) ⊂ Emb(F,R∞) is contractible, since the inclusion is a homotopy equivalence by the tubular neighborhood theorem and since Emb(F,R∞) is con- tractible by Whitney’s embedding theorem. A fibrewise fat embedding ι : E → X ×R∞ is then a section of the fibre bundle P × Embf(F,R∞). Diff(F) Suppose first that ι : E → X×Rn+2 is a fibrewise fat embedding of codimension n. The Pontryagin-Thom construction associates a “collapse ” map onto the Thom space of the fibrewise normal bundle, c : X ∧Sn+2 −→ D(Nnι)/S(Nnι) = Th(Nnι). π,ι + π π π We are particularly interested in its adjoint map X → Ωn+2Th(Nnι). π Let G(2,n) be the Grassmann manifold of oriented 2-dimensional subspaces of Rn+2, and let U and U⊥ be the two complementary universal bundles over it of n n dimension 2 and n, respectively. The fat embedding ι induces bundle maps T E −→ U , Nnι −→ U⊥ π n π n and a commutative diagram X ∧Sn+2 −−c−π−,ι→ Th(Nnι) −−−s−→ Th(T E ⊕Nnι) + π π π (3.3) X ∧(cid:13)(cid:13)Sn+2 −−−cn−→ Th(U⊥) −−−s−→ Th(U ⊕U⊥). (cid:13) y y 9 In the general case of a fiberwise fat embedding ι : E → X ×R∞, the base space X is the colimit of the subspaces X := {x ∈ X|ι (E ) ⊂ {x}×Rn+2}, n x x and the diagram (X ) ∧Sn+2 −−−−→ Th(U⊥) n + n   (Xn+1)+∧Sn+3 −−−−→ Th(Un⊥+1) y y is commutative since U⊥ | = U⊥. Taking adjoints we get n+1 G(2,n) n α : X −→ colim Ωn+2Th(U⊥). π,ι n→∞ n Since Embf(F,R∞) is contractible, all sections of P × Embf(F,R∞) are ho- Diff(F) motopic, and consequently the homotopy class [α ] is independent of the choice π,ι of ι. We will therefore from now on suppress the subscript ι. Realification gives a (2n−1) connected map from CPn into the oriented Grass- mannian G(2,2n) covered by a bundle map L⊥ → U⊥. Thus G(2,∞) ≃ CP∞ n 2n and Ω∞CP∞ = colim Ω2n+2Th(L⊥) −≃→ colim Ω2n+2Th(U⊥) −1 n 2n is a homotopy equivalence. Altogether we have a well-defined homotopy class α : X −→ Ω∞CP∞. π −1 For X = BDiff(F ) ≃ BΓ this is the map α of (3.2). g+1 g+1 g+1 Let us check that the image of α , and hence the image of α , lie in the g+1 g,2 g-component of Ω∞CP∞, or equivalently that the composition −1 proj◦ω ◦α : BΓ −→ Ω∞S∞(CP∞) −→ Ω∞S∞ g+1 g+1 + lands in the −2g component (with identification of components chosen at the end of Section 2.1). Consider (3.3) with X a single point and E = F . The bottom g+1 row in (3.3) is thus s◦c : Sn+2 → G(2,2n) ∧Sn+2 and we need to compute the n + degree of the composition of this map with the projection onto Sn+2. This degree is given by the evaluation of the pullback of the generator of Hn+2(Sn+2) on the fundamental class [Sn+2]. Under the projection the fundamental class is pulled back to the Thom class of the trivial bundle U ⊕U⊥. Writing λ for the Thom n n U class of the vector bundle U, the degree is thus given by < c∗s∗(λ ·λ ),[S2n+2] > =< c∗(e(TF )·λ ),[S2n+2] > π Un Un⊥ π g+1 Un⊥ =< e(TF ),[F ] >= −2g, g+1 g+1 as claimed. Next we compute the maps α and α under the map BΓ → BΓ g,2 g+1,2 g,2 g+1,2 induced from gluing a torus with two boundary circles F to F . Considering 1,2 g,2 F as a fibre bundle over a point the construction above gives an element [1] ∈ 1,2 10 SOREN GALATIUS, IB MADSEN, ULRIKE TILLMANN Ω∞CP∞. Loop sum with [1] in Ω∞CP∞ translates the g component into the 1 −1 −1 (g +1) component and BΓ −−α−g−,2→ Ω∞CP∞ g,2 g −1 (3.4) ∗[1] BΓg+1,2 −α−g−+−1→,2 Ω∞g+1CP−∞1 y y is homotopy commutative. To see this observe that the left vertical map is multi- plication by the basepoint of BΓ in the monoid BΓ and the right vertical 1,2 g≥0 g,2 map is multiplication by its image in Ω∞CP∞. Homotopy commutativity of dia- 1 −1 F gram (3.4) now follows because the α induce a map of monoids up to homotopy g,2 for: α is a map of infinite loop spaces, α restricts to α on BΓ , and the infi- g,2 g,2 nite loop space structure on Z×BΓ+ is compatible with the monoidal structure. ∞ (Alternatively, homotopy commutativity of (3.4) follows from a calculations of pre- transfers, cf. [G2].) Let α˜ denote the restriction of α in Theorem 2.2 to the zero component. Re- stricted to BΓ it is homotopic to α˜ = (∗[−g])◦ α . We can now relate the g,2 g,2 g,2 kappa classes to spectrum cohomology. Consider BΓ −α˜g→,2 Ω∞CP∞ −ω→ Ω∞S∞(CP∞) g,2 0 −1 0 + and recall the cohomology suspension from Section 2.1 σ∗ : H2i(CP∞;Z) ≃ H2i(S∞CP∞) −→ H2i(Ω∞S∞(CP∞)). + 0 + Theorem3.1. The Miller-Morita-Mumford class κ is equal to(ω◦α˜)∗(σ∗ei) where i e ∈ H2(CP∞;Z) is the Euler class of the canonical line bundle. Proof: Let π : E → X be a smooth fibre bundle with fibre F , classified by g+1 f : X → BΓ . By definition π g+1 f∗(κ ) = π (e(T E)i+1) ∈ H2i(X;Z) π i ! π where π is the composition of the Thom isomorphism and the Pontrjagin-Thom ! collapse map ∗ H˜2i+2(E;Z) −≃→ H˜2i+2n+2(Th(Nnι);Z) −c→π H2i+2n+2(S2n+2 ∧X ;Z) π + followed by the (2n+2)-nd desuspension; here the notation is as in (3.3). Let x ∈ Hk(CPN;Z) ≃ Hk(G(2,2N);Z) for N ≫ k. The 2n+2-fold suspension Σ2n+2(x) ∈ Hk+2n+2(S2n+2 ∧ CPN,Z) is x times the Thom class of the trivial 2n+2 real bundle L⊥ ⊕L . Thus n n s∗(Σ2n+2(x)) = s∗(λ ·λ ·x) = λ ·e(L )·x. ⊥ L ⊥ n