ebook img

Cohomology and generic cohomology of Specht modules for the symmetric group PDF

0.2 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Cohomology and generic cohomology of Specht modules for the symmetric group

COHOMOLOGY AND GENERIC COHOMOLOGY OF SPECHT MODULES FOR THE SYMMETRIC GROUP 9 DAVID J. HEMMER 0 0 2 Abstract. CohomologyofSpechtmodulesforthesymmetricgroupcanbeequatedinlowdegrees n with corresponding cohomology for the Borel subgroup B of the general linear group GL (k), but d a this has never been exploited to prove new symmetric group results. Using work of Doty on the J submodule structure of symmetric powers of the natural GL (k)-module together with work of d 8 Andersen on cohomology for B and its Frobenius kernels, we prove new results about Hi(Σ ,Sλ). d 2 WerecoverworkofJamesinthecasei=0. Thenweprovetwostabilitytheorems,oneofwhichisa “genericcohomology”resultforSpechtmodulesequatingcohomologyofSpλ withSp2λ.Thisisthe ] T firsttheoremweknowrelatingSpechtmodulesSλ andSpλ. Thesecondresultequatescohomology R of Sλ with Sλ+paµ for large a. . h t a m 1. Introduction [ 1.1. For an algebraic group G and a G-module M, the Frobenius morphism on G lets one define a 2 new G module M(1), the Frobenius twist of M. If G = GL (k) and L(λ) is an irreducible module n v then L(λ)(1) ∼= L(pλ). Thus theorems about G-modules which involve multiplying partitions by p 4 6 arise quite naturally and can be explained in terms of the Frobenius map. 7 In contrast, supposeSλ is a Specht module for the symmetric group Σ , where λ is a partition of d 3 d, denoted λ ⊢ d. Then pλ ⊢ pd, so Spλ is a Specht module for an entirely different group, Σ , and . pd 3 there is no evident relation between the two modules. Indeed we know of no theorems involving 0 both Sλ and Spλ. This paper will prove the first such theorem, although an explanation involving 8 0 only the symmetric group eludes us! : I would like to thank the anonymous referee for an extensive report which vastly improved the v i exposition, discovered an error in the proof of Lemma 6.5.4 and provided a correct version. X r 1.2. Let k be an algebraically closed field of characteristic p ≥ 3. For a partition λ ⊢ d let Sλ a denote the corresponding Specht module for the symmetric group Σ , and let S be its linear d λ dual. In characteristic two every Specht module is also a dual Specht module, and the problem of calculating cohomology is quite different and seems more difficult. Proposition 4.1.1 below, which equates symmetric group cohomology with that of the general linear group, only holds through degree 2p−4, so does not apply even to H1(Σ ,Sλ) when p = 2. Thus we will restrict to the odd d characteristic case for this paper. Much is known about the low degree cohomology groups Hi(Σ ,S ). In [3] (where again only d λ oddcharacteristic is considered), itis shownthatthiscohomology vanishesin degrees 1 ≤ i≤ p−3. Date: January 2009. 2000 Mathematics Subject Classification. Primary 20C30, Secondary 20G05, 20G10. Research of theauthor was supported in part byNSFgrant DMS-0556260. 1 SPECHT MODULE COHOMOLOGY 2 For p = 3 a complete description of the nonzero cohomology was given for i = 1,2 in [3, Thms. 2.4, 4.1]. In contrast little is known about the cohomology Hi(Σ ,Sλ) of Specht modules. For i = 0 this d was computed by James [12, Thm. 24.4]. It is at most one-dimensional, and explicit conditions are given on λ for it to be nonzero, see Theorem 5.1.1 below. In [11] the cohomology Hi(Σ ,Sλ) d for 0 ≤ i ≤ p−2 is shown to agree with certain B cohomology in degree i+ d , where B is the 2 Borel subgroup of GL (k). However this high degree B cohomology has not been computed, so no d (cid:0) (cid:1) new symmetric group results were obtained. Inourapproachtherelevantcohomology forB anditsFrobeniuskernelscanbecomputed,andso newresultsonthecohomology H1(Σ ,Sλ)ofSpechtmodulesareobtained. Weprovesomestability d theorems that generalize the results of James for H0(Σ ,Sλ) and are inspired by the known results d fortwo-partpartitions. Inparticularweprovea“genericcohomology” theoremthatH1(Σ ,Spaλ) pad stabilizes for a ≥ 1. There is a corresponding result for algebraic groups where one twists by the Frobenius automorphism. However for the symmetric group there is no Frobenius automorphism, and it is quite mysterious why any theorem should relate Specht modules Sλ and Spλ, which are apparently unrelated modules for two different groups! 2. Notation and preliminaries 2.1. Although our application is to symmetric group representation theory, the actual work will be done within the general linear group theory. A basic reference for key results and also for our notation is [13]. For information on the Schur algebra see [10]. We will also draw extensively from the paper [1] of Andersen. Recall k is an algebraically closed field of characteristic p ≥ 3 and let G = GL (k) bethe general n lineargroup. LetB (resp. B )betheBorelsubgroupoflower(resp. upper)triangularmatricesand + T the torus of diagonal matrices. Let R denote the root system with respect to T, with associated inner product h−,−i. Let X(T) ∼= Zn denote the weight lattice and S = {α1,α2,...,αn−1} a set of simple roots such that B corresponds to the negative roots. For α ∈ S the corresponding coroot is α∨ and the corresponding reflection on X(T) is s . Let ρ denote half the sum of the positive α roots. The set of dominant weights is X(T) = {λ ∈ X(T) | hα∨,λi ≥ 0 for all α ∈ S.} Let ≤ denote + the ordering on X(T) given by λ ≤ µ if and only if µ−λ is a linear combination of positive roots with nonnegative coefficients. The set of pr-restricted weights is denoted X (T) = {λ ∈ X(T) |0 ≤ r hα∨,λi < pr for all α ∈ S}. For a module M and a simple module S, the composition multiplicity of S in M will be denoted [M : S]. For a GL (k)-module M, the module M(r) will denote the rth Frobenius twist of M. n The rth Frobenius kernel of B is denoted B , see [13] for definitions. We will sometimes use the r fact that B/B is isomorphic to B and the modules can be identified using the rth power of the r Frobenius twist. 2.2. The Schur algebra S(n,d) is the finite-dimensional associative k-algebra End (V⊗d) where kΣd V ∼= kn is the natural representation of G. Excellent references for representation theory of the Schur algebra are the books [9] and [15]. The category M(n,d) of polynomial G-modules of a fixed degree d ≥ 0 is equivalent to the category of modules for S(n,d). SPECHT MODULE COHOMOLOGY 3 Simple S(n,d)-modules are in bijective correspondence with the set Λ+(n,d) of partitions of d with at most n parts, and are denoted by L(λ). Note that one can also identify Λ+(n,d) as the set of dominant polynomial weights of G of degree d. For λ ∈ Λ+(n,d), let H0(λ) = indGλ be B the induced module and let V(λ) be the Weyl module. Often we will write H0(d) rather than H0((d,0,...,0)) and similarly L(d). Note we are using λ here to denote the one-dimensional B module of weight λ. For λ,µ ∈ Λ+(n,d) let (cid:4) denote the usual dominance order [15, Def. 1.4.2]. 2.3. We recall two results about Ext groups for GL (k)-modules. The first is that for two S(n,d)- n modules M,N, the Exti groups are the same whether we work in the category of all G-modules or of S(n,d)-modules. The second is that for n > d the category of modules for S(n,d) is equivalent to that of S(d,d), where the equivalence “preserves” L(λ), H0(λ) and V(λ) for λ ⊢ d. (Although the modules certainly have different dimensions, for example V(1,1,1) is a one-dimensional S(3,3) module but a four-dimensional S(4,3) module.) Both results are special cases of a more general result regarding truncated categories (cf. [13, Chpt. A]). Proposition 2.3.1. [13, A.10, A.18] Let M and N be S(n,d)-modules, and hence also G-modules. Then for all i ≥ 0 Exti (M,N) ∼= Exti (M,N). S(n,d) G The next proposition will let us equate Σ cohomology with that of GL (k) for various n≥ d. d n Proposition 2.3.2. [10, 6.5g] Suppose n > d. Then there is an idempotent e ∈ S(n,d) such that eS(n,d)e ∼= S(d,d) and the functor taking M to eM is an equivalence of categories from mod-S(n,d) to mod-S(d,d) mapping L(λ) to L(λ) and similarly for H0(λ) and V(λ). 3. B-Cohomology and Spectral Sequences 3.1. As we will see in the next section, low degree Specht module cohomology can be equated with certain B cohomology, and it is in the setting of B cohomology that our work will be done. In this section we collect the results on B and B cohomology that we will need, and recall two important r spectral sequences that we will use. The first describes extensions between B modules that have been twisted. The second relates cohomology for B and B . r First we recall that L(λ) has simple head as a B-moduleand, if λ ∈ X (T), also as a B -module. r r Proposition 3.1.1. Let λ ∈ X(T) and ν ∈ X(T). Then: + k if λ = ν (a) Hom (L(λ),ν) ∼= B (0 otherwise. prν if ν =λ+prν (b) Hom (L(λ),ν) ∼= 1 1 where the isomorphism is as B-modules. Br (0 otherwise. Proof. Part (a) follows from Frobenius reciprocity (Proposition 4.1.3(a) below) and the fact [13, II.2.3] that H0(λ) has simple socle L(λ). Part (b) is Equation 3.1 in [1]. (cid:3) Calculating higher degree cohomology for B or B is a difficult problem and a subject of active r research. Fortunately we need only a few results in degree one. The first part of the following is immediate from Lemma 2.2 in [4], where a much stronger result is proven. We denote by St the r rth Steinberg module L((pr −1)ρ). SPECHT MODULE COHOMOLOGY 4 Proposition 3.1.2. Let λ ∈ X(T) and ν ∈ X(T). Then: + (a) If λ (cid:3) µ then Ext1 (L(λ),µ) = 0. B (b) [13, II.12.1] H1(B ,k) = 0. r (c) [1, Prop. 3.2] Suppose further that λ,ν ∈ X (T). Then the weights of Ext1 (L(λ),ν), as a r Br B-module, are contained in the set {prξ | λ−ν +prξ −(pr −1)ρ is a weight in St .} r 3.2. Wewillmakeuseoftwospectralsequences,bothofwhicharefirstquadrantspectralsequences of the form E∗,∗ converging to H∗. 2 ThefirstisaspectralsequencethatwewillapplytocompareExt∗ (M ,M )withExt∗(M(r),M(r)). B 1 2 B 1 2 Proposition 3.2.1. [13, I.6.10, II.10.14] Let M ,M be B-modules and B be the rth Frobenius 1 2 r kernel. There is a first quadrant spectral sequence: (3.2.1) Ei,j = Exti (M ,M ⊗Hj(B ,k)(−r)) ⇒ Exti+j(M(r),M(r)). 2 B 1 2 r B 1 2 The second is the Lyndon-Hochschild-Serre spectral sequence. Proposition 3.2.2. [13, I.6.6(1)] Let H denote either G or B. Let E and V be H-modules and M be an H/H -module. Then there is a first quadrant spectral sequence: r (3.2.2) Ei,j = Exti (M,Extj (E,V)) ⇒ Exti+j(M ⊗E,V). 2 H/Hr Hr H We will make use only of a few elementary properties, described below, of these spectral se- quences, so the reader does not need extensive familiarity with them. Specifically we need: Proposition 3.2.3. Suppose E∗,∗ converging to H∗ is one (3.2.1) or (3.2.2). 2 (a) There is a five-term exact sequence: (3.2.3) 0→ E1,0 → H1 → E0,1 → E2,0 → H2. 2 2 2 (b) Suppose Ei,j = 0 for all i+j = n. Then Hn = 0. 2 (c) E0,0 ∼= H0. 2 4. Specht module cohomology and symmetric powers 4.1. Although we will be studying symmetric group cohomology, we first transfer the problem to the general linear group. We recall below the result that allows this. For additional such results and explanations of how they are obtained using the Schur functor and the higher derived functors of its adjoint, see [14]. We use only the following. Proposition 4.1.1. [14, Corollary 6.3(b)(iii)] For 0≤ i ≤ 2p−4 we have Hi(Σ ,Sλ) ∼= Exti (H0(d),H0(λ)). d GLd(k) Remark 4.1.2. Proposition 4.1.1 is stated in [14] for 0 ≤ i ≤ 2p−3 but actually only holds up to i = 2p−4, cf. [11, 2.4] for an explanation. SPECHT MODULE COHOMOLOGY 5 We willactually doour computations usingB cohomology, so thefollowing is crucial, wherepart (b) is immediate from part (a) and Propositions 2.3.2 and 4.1.1. Proposition 4.1.3. (a) [13, 4.7a] Let V be a G-module and λ ⊢ d. Then for all i ≥ 0: Exti (V,H0(λ)) ∼= Exti (V,λ). G B (b) If 0 ≤ i≤ 2p−4 and G = GL (k) for n ≥ d then n Hi(Σ ,Sλ) ∼= Exti (H0(d),λ). d B 4.2. Structureofsymmetricpowers. Proposition4.1.3(b)suggestsonemustunderstandH0(d) in order to compute low degree Specht module cohomology. The module H0(d) is isomorphic to the dth symmetric power of the natural module V ∼= kn. In [7], Doty gave a complete description of the composition factors and submodule lattice of H0(d), which we recall below. Let B(d) := {β = (β ,β ,...,β ) |β ≥ 0 and β =d}. 1 2 n i i ThenH0(d)hasabasisofweightvectorsindexednaturallybyB(Xd)[7,2.1]. Forβ ∈B(d), associate a sequence of nonnegative integers c (β) as follows. First write each β out in base p. Then add i i them base p to get d. For i ≥ 1, let c (β) be the number that is “carried” to the top of the pi i column during the addition. For example if p = 3 and β = (5,5,2) then the addition 5+5+2=12 base three looks like: 1 2 1 2 1 2 + 0 2 1 1 0 so c(β) = (2,1). Doty calls this the carry pattern of β. Let C(d) be the set of all carry patterns which occur for some β ∈ B(d). Define a partial order on C(d) by declaring (c ,c ,...,c )≤ (c′,c′,...,c′ ) if c ≤ c′ for all i. Given a subset B ⊆ C(d), 1 2 m 1 2 m i i let T be the subspace of H0(d) spanned by all β with c(β) ∈ B. We can now state Doty’s result: B Theorem 4.2.1. [6] The correspondence B → T defines a lattice isomorphism between the lattice B of ideals in the poset C(d) and the lattice of submodules of H0(d). In particular the composition factors of H0(d) are in one-to-one correspondence with the subspaces T , for c ∈ C(d). If L(λ) c corresponds to T then λ is the maximal partition in Λ+(n,d) with carry pattern c. c Given λ ∈ Λ+(n,d), it is not difficult to determine if [H0(d) : L(λ)] 6= 0. The next lemma is an easy exercise working with carry patterns. Lemma 4.2.2. Let λ = (λ ,λ ,...,λ ) ∈ Λ+(n,d). Define numbers a with 0 ≤ a < p by 1 2 n ij ij λ = a pj. Then: i ij (a) [H0(d) :L(λ)] 6= 0 if and only if a 6= p−1 implies a = 0 for all l > i. P ij lj (b) [H0(pd) : L(pλ)] = [H0(d) :L(λ)]. Wewilllater useastrengthenedversion ofLemma4.2.2(b), namelythatH0(d)(1) is asubmodule of H0(pd) in a particularly nice way, cf. Proposition 6.4.1. SPECHT MODULE COHOMOLOGY 6 5. Previously known Specht module cohomology 5.1. Computing Hom (k,Sλ). The degree zero cohomology H0(Σ ,Sλ) ∼= Hom (k,Sλ) was Σd d Σd determinedbyJames. For aninteger tlet l (t)betheleast nonnegative integer satisfyingt < plp(t). p James proved: Theorem 5.1.1. [12, 24.4] The cohomology, H0(Σ ,Sλ) is zero unless λ ≡ −1 mod plp(λi+1) for d i all i, in which case it is one-dimensional. Remark 5.1.2. James’ result is proved entirely using symmetric group theory, applying the famous kernel intersection theorem. We observe that Doty’s description of H0(d) allows one to calculate precisely when Hom (H0(d),λ) is nonzero, and thus rederive James’ result in an entirely different B way. The space of homomorphisms can only be nonzero for λ ⊢ d with [H0(d) : L(λ)] = 1, so we mustdetermineforeachL(λ)aconstituentofH0(d), ifthereisanonzeroB modulehomomorphism from H0(d) to λ. There is such a map precisely when there does not exist a µ (cid:3) λ such that [H0(d) : L(µ)] = 1 and c(µ) > c(λ). This can easily be seen to be equivalent to the condition in Theorem 5.1.1. Since the result is already known, we leave the details to the reader. One must show that given such λ and µ (i.e. µ(cid:3)λ and c(µ) > c(λ)) the λ weight vector always lies in the B submodule generated by the µ weight vector, so there is no homomorphism in this case. Doty’s paper explicitly describes the B action, so this can be done. Example 5.1.3. In characteristic five, [H0(25) : L(20,5)] = 1 but H0(Σ ,S(20,5)) = 0. This is 25 because (24,1) (cid:3)(20,5) and c(24,1) = (1,1) > (1,0) = c(20,5). The (20,5) weight vector lies in the B submodule generated by the (24,1) weight vector, and so Hom (H0(25),(20,5)) = 0. B 5.2. Two-part partitions. The other known example of H1(Σ ,Sλ) is for λ = (λ ,λ ) a two- d 1 2 part partition, where the answer can be deduced from work of Erdmann [8] on Ext1 between Weyl modules for SL (k), as we describe below. 2 Forr > 0writer = r pi with0 ≤r ≤ p−1. For0 ≤ a≤ p−1defineasothat0 ≤ a≤ p−1 i≥0 i i and a+a≡ p−1 (mod p). Erdmann defined [8, p. 456] a collection of integers Ψ (r) by p P u−1 u (5.2.1) Ψ (r) = { r pi+pu+a : r 6= p−1,a ≥ 1,u ≥ 0}∪{ r pi : r 6= p−1,u ≥ 0}. p i u i u i=0 i=0 X X FromLemma3.2.1in[11]wededuceH1(Σ ,S(λ1,λ2))isatmostone-dimensional,anditisnonzero d precisely when λ ∈ Ψ (λ −λ ). The conditions on λ for this to occur were stated incorrectly in 2 p 1 2 [11], so we correct it here. First suppose λ lies in the second set in the description (5.2.1) of Ψ (λ −λ ). So we have: 2 p 1 2 (5.2.2) λ −λ = r +r p+r p2+··· 1 2 0 1 2 λ = r +r p+···+r pu 2 0 1 u λ = p−1+(p−1)p+(p−1)p2+···+(p−1)pu+r pu+1+··· 1 u+1 where r 6= 0. u Recall from Theorem 5.1.1 that H0(Σ ,S(λ1,λ2)) 6= 0, precisely when λ ≡ −1 mod plp(λ2). d 1 Equation 5.2.2 demonstrates that l (λ ) = u+1 and λ ≡ −1 mod pu+1. Thus we have the exactly p 2 1 James’ criterion and obtain Proposition 5.2.1(a) below. SPECHT MODULE COHOMOLOGY 7 Next we consider when λ lies in the first subset in (5.2.1). In this case we will have, for some 2 u≥ 0 and a≥ 1: (5.2.3) λ −λ = r +r p+r p2+···+r pu−1+r pu+··· 1 2 0 1 2 u−1 u λ = r +r p+···+r pu−1+0pu +pu+a 2 0 1 u−1 λ = p−1+(p−1)p+(p−1)p2 +···+(p−1)pu−1+r pu+··· 1 u where r 6= p−1. In (5.2.3), u is minimal such that λ ≡ −1 mod pu but not ≡ −1 mod pu+1. This u 1 proves the second part of: Proposition 5.2.1. Let λ = (λ ,λ ) ⊢ d with λ 6= 0. Then H1(Σ ,Sλ) is zero except in the two 1 2 2 d cases below, when it is one-dimensional: (i) If Hom (k,Sλ) 6= 0 or Σd (ii) If λ ≡ −1 mod pu but λ 6≡ −1 mod pu+1 for some u≥ 0 such that λ = c+pb for c < pu 1 1 2 and b > u. Example 5.2.2. In characteristic five, H1(Σ ,S(29,25)) ∼= k. Here choose u = 1,a = 2,c = 0 in 54 part (ii) above. Example 5.2.3. In characteristic p suppose a,b > 0 and pa ≥ pb. Then choosing u = 0 we get H1(Σ ,S(pa,pb)) 6= 0. d Case (i) in Proposition 5.2.1 is a special case of a more general result conjectured in [11], which can be proved using a suggestion of Andersen. Proposition 5.2.4 (Andersen). Suppose H0(Σ ,Sλ) 6= 0 and λ 6= (d). Then H1(Σ ,Sλ) 6= 0. d d Proof. This follows from the universal coefficient theorem. The key observation is that the Specht module is defined over the integers, but when λ 6= (d) then HomZ(Z,SZλ)= 0. (cid:3) 5.3. The following corollary is immediate from Proposition 5.2.1, and is the motivation for the generalizations proved in the next two sections. Corollary 5.3.1. Let λ = (λ ,λ )⊢ d < pa. Then 1 2 (a) H1(Σpd,Spλ)∼= H1(Σp2d,Sp2λ). (b) H1(Σd,Sλ)∼= H1(Σd+pa,S(λ1+pa,λ2)) 6. Generic cohomology for Specht modules? 6.1. For any G-module M there is a series of injections: (6.1.1) Hi(G,M) → Hi(G,M(1)) → Hi(G,M(2)) → ··· . This sequence is known [4] to stabilize, and the limit is called the generic cohomology of M. For example we get injections Hi(G,L(λ)) → Hi(G,L(pλ)) → Hi(G,L(p2λ)). It is natural then to have theorems for G which involve multiplying a partition by p, as this reflects what happens to the weights after a Frobenius twist. SPECHT MODULE COHOMOLOGY 8 For the symmetric group, we know of no theorems involving multiplying a partition by p. There is nothing that seems to play the role of the Frobenius twist. Moreover, the modules Sλ and Spλ are not even modules for the same symmetric group. However, for two-part partitions we observed in Corollary 5.3.1 that H1(Σpd,Spµ) ∼= H1(Σp2d,Sp2µ). In this section we generalize this stability result to arbitrary partitions, and show there is a generic cohomology for Specht modules in degree one. The main result is an isomorphism H1(Σpd,Spλ)∼= H1(Σp2d,Sp2λ). 6.2. Relating cohomology and Frobenius twists. We can use the spectral sequence (3.2.1) in our first step towards relating cohomology of Sλ and Spλ. Lemma 6.2.1. Let µ ⊢ d. Then Ext1(H0(d)(1),pµ) ∼= Ext1 (H0(d),µ). B B Proof. Take r = 1 and consider the five-term exact sequence (3.2.3) for the spectral sequence (3.2.1). Choosing M = H0(d) and M = µ, we obtain: 1 2 0 → Ext1 (H0(d),µ) → Ext1(H0(d)(1),pµ) → Hom (H0(d),H1(B ,k)(−1) ⊗µ)→ ··· . B B B 1 Recall from Proposition 3.1.2(b) that H1(B ,k) = 0, so we have the desired isomorphism. (cid:3) 1 Lemma 6.2.1 can be interpreted as the immediate stabilization of (6.1.1) when M = µ⊗H0(d)∗. Remark 6.2.2. If the left side in Lemma 6.2.1 had H0(pd) instead of H0(d)(1) we would have an isomorphism between H1(Σ ,Sλ) and H1(Σ ,Spλ), however this is false in general. d pd Next we prove a technical lemma: Lemma 6.2.3. Suppose λ ⊢p2d such that λ 6= pτ for any τ ⊢ pd. Suppose further that [H0(p2d) : L(λ)] 6= 0 and that µ ⊢d with λ(cid:3)p2µ. Then: hλ−p2µ,α∨i≥ p2 i for some i. Proof. Let λ = (λ ,λ ,...,λ ) and denote the p-adic expansion of λ by λ = a +a p+··· . 1 2 n i i i,0 i,1 Since p ∤ λ by Lemma 4.2.2(a), we must have λ > p2µ . Since λ and p2µ both partition p2d, 1 1 1 there must be some i ≥ 1 with λ ≥ p2µ and λ < p2µ . So let: i i i+1 i+1 λ = a +a p+p2t i i,0 i,1 λ = a +a p+p2s i+1 i+1,0 i+1,1 Our assumptions on λ and p2µ imply that t ≥ p2µ and s < p2µ . By Lemma 4.2.2(a) we have i i+1 a ≥ a for all l. Thus: i,l i+1,l hλ−p2µ,α∨i = (λ −p2µ )−(λ −p2µ ) i i i i+1 i+1 = a +a p+p2(t−p2µ )−a −a p−p2(s−p2µ ) i,0 i,1 i i+1,0 i+1,1 i+1 ≥ p2(t−p2µ )−p2(s−p2µ ) i i+1 ≥ p2 SPECHT MODULE COHOMOLOGY 9 (cid:3) Corollary 6.2.4. Let λ and p2µ be as in Lemma 6.2.3 above. Then λ−p2µ−(p−1)ρ is not a weight in the Steinberg module St = L((p−1)ρ). 1 Proof. Let γ = λ−p2µ−(p−1)ρ and choose i as in Lemma 6.2.3. Then h(p−1)ρ−γ,α∨i = h2(p−1)ρ−(λ−p2µ),α∨i i i = 2p−2−h(λ−p2µ),α∨i i ≤ 2p−2−p2 by Lemma 6.2.3 = −p2+2p−2< 0. Thus (p−1)ρ 6≥ γ so γ is not a weight in St . (cid:3) 1 Remark 6.2.5. We observe that Corollary 6.2.4 requires p2µ and the corresponding statement for pµ is false. For example if p = 5, λ = (9,1) and 5µ = (5,5) then λ−5µ−4ρ is a weight in the Steinberg module. This is why our stability theorem requires comparing pµ and p2µ. 6.3. Corollary 6.2.4 lets us prove a key vanishing result for B cohomology, which we state next. Proposition 6.3.1. Let λ ⊢ p2d with [H0(p2d) : L(λ)] 6= 0. Suppose λ is not of the form pτ and let µ ⊢ d. Then: Ext1(L(λ),p2µ)= 0. B Proof. We can assume that λ > p2µ without loss by Proposition 3.1.2(a). Write λ = λ +pτ with (0) 0 6= λ ∈ X (T) and, using the Steinberg tensor product theorem, we have: (0) 1 Ext1 (L(λ),p2µ) ∼= Ext1(L(λ )⊗L(pτ)⊗(−p2µ),k). B B (0) Now consider the spectral sequence (3.2.2) with B (cid:1)B. Set M = L(pτ)⊗−p2µ, E = L(λ ) and 1 (0) V = k to obtain: (6.3.1) Ei,j = Exti L(pτ)⊗(−p2µ),Extj (L(λ ),k) ⇒ Exti+j(L(λ),p2µ). 2 B/B1 B1 (0) B (cid:16) (cid:17) 1,0 By Proposition 3.1.1(b), we know Hom (L(λ ),k) = 0, and thus the E term in (6.3.1) is B1 (0) 2 0,1 zero. Now consider the E term. For any ξ ∈ X(T) we have: 2 Hom (L(τ)(1) ⊗(−p2µ),pξ)∼= Hom (L(τ)⊗(−pµ),ξ)∼= Hom (L(τ),ξ +pµ) B/B1 B B which, by Proposition 3.1.1(a), is zero unless τ = ξ+pµ. Applyingthis to the E0,1 term in (6.3.1), we see thatE0,1 is zero unless pξ = pτ−p2µis aweight 2 2 in Ext1 (L(λ ),k). By Proposition 3.1.2(c), this can only occur if λ +pτ −p2µ−(p−1)ρ is a B1 (0) (0) 0,1 weight in St . But this is ruled out by Corollary 6.2.4. Thus the E term vanishes as well, and 1 2 so Ext1(L(λ),p2µ)= 0 by Proposition 3.2.3(b). B (cid:3) SPECHT MODULE COHOMOLOGY 10 6.4. Twisted symmetric powers. Our next observation is that the twisted symmetric power H0(d)(1) embeds nicely in H0(pd) with cokernel containing no simple modules of the form L(pτ). Proposition 6.4.1. There is a short exact sequence (6.4.1) 0 → H0(d)(1) → H0(pd) → Q → 0 where for all τ ⊢ d, [Q : L(pτ)] = 0. Proof. Apply Hom (−,H0(pd)) to the sequence 0 → L(pd) → H0(d)(1) → U → 0 to obtain: G 0 → Hom (H0(d)(1),H0(pd)) → k → Ext1(U,H0(pd)). G G But Ext1(U,H0(pd)) = 0 since (pd) is not dominated by any weight in U (see [13, II.4.14]). Thus G Hom (H0(d)(1),H0(pd)) ∼= k and by comparing socles we see the map must be an injection. The G statement about Q is immediate by Lemma 4.2.2(b) since H0(pd) is multiplicity free. (cid:3) Now consider(6.4.1) withpdandp2d, andsupposep2µ ⊢ p2d. ApplyingHom (−,p2µ)to(6.4.1) B we obtain: (6.4.2) ··· → Ext1 (Q,p2µ)→ Ext1 (H0(p2d),p2µ)→ Ext1 (H0(pd)(1),p2µ) → Ext2 (Q,p2µ)→ ··· B B B B We will show, in Lemma 6.4.2 and Proposition 6.5.5 below that the first and last term in (6.4.2) are zero. Lemma 6.4.2. Let Q be as in (6.4.2). Then Ext1 (Q,p2µ)= 0. B Proof. Let [Q : L(λ)] 6= 0. By Proposition 6.4.1 we know λ is not of the form pτ. The result then follows by Proposition 6.3.1. (cid:3) 6.5. AnalyzingExt2 (Q,p2µ). NextweprovetheanalogueofProposition6.3.1forExt2(L(λ),p2µ), B B which will require even more intricate spectral sequence calculations. Let λ = λ +pτ with 0 6= (0) λ ∈ X (T) as before and consider the spectral sequence (6.3.1). We will prove Ext2 (L(λ),p2µ) (0) 1 B 0,2 1,1 2,0 is zero by showing the terms E ,E and E all vanish, and applying Proposition 3.2.3(b). 2 2 2 Lemma 6.5.1. The E2,0 term in (6.3.1) is zero. 2 Proof. This is immediate since λ 6= 0 implies Hom (L(λ ),k) = 0 by Proposition 3.1.1(b). (cid:3) (0) B1 (0) 1,1 To show the E term is zero we extend an argument of Andersen’s. For ǫ ∈ X (T) he defined 2 1 ([1, p.495]) R(ǫ) by the exact sequence (6.5.1) 0 → ǫ → St ⊗[(p−1)ρ+ǫ] → R(ǫ)→ 0. 1 We will use this to prove: Lemma 6.5.2. The E0,2 term in (6.3.1) is zero. 2

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.