SUBGROUPS OF SPIN(7) OR SO(7) WITH EACH ELEMENT CONJUGATE TO SOME ELEMENT OF G 2 AND APPLICATIONS TO AUTOMORPHIC FORMS arxiv: v2 [math.gr] 11 Oct 2016 GAËTAN CHENEVIER Laboratoire de Mathématiques

•0 views•20 pages

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.

Documenttranscript

SUBGROUPS OF SPIN(7) OR SO(7) WITH EACH ELEMENT CONJUGATE TO SOME ELEMENT OF G 2 AND APPLICATIONS TO AUTOMORPHIC FORMS arxiv: v2 [math.gr] 11 Oct 2016 GAËTAN CHENEVIER Laboratoire de Mathématiques d Orsay, Université Paris-Sud, CNRS, Université Paris-Saclay, Orsay, France, Abstract. As is well-known, the compact groups Spin(7) and SO(7) both have a single conjugacy class of compact subgroups of exceptional type G 2. We first show that if Γ is a subgroup of Spin(7), and if each element of Γ is conjugate to some element of G 2, then Γ itself is conjugate to a subgroup of G 2. The analogous statement for SO(7) turns out be false, and our main result is a classification of all the exceptions. They are the following groups, embedded in each case inso(7) in a very specific way: GL 2 (Z/3Z), SL 2 (Z/3Z), Z/4Z Z/2Z, as well as the nonabelian subgroups of GO 2 (C) with compact closure, similitude factors group {±1}, and which are not isomorphic to the dihedral group of order 8. More generally, we consider the analogous problems in which the Euclidean space is replaced by a quadratic space of dimension 7 over an arbitrary field. This type of questions naturally arises in some formulation of a converse statement of Langlands global functoriality conjecture, to which the results above have thus some applications. Moreover, we give necessary and sufficient local conditions on a cuspidal algebraic regular automorphic representation of GL 7 over a totally real number field so that its associated l-adic Galois representations can be conjugate into G 2 (Q l ). Date: November 1, Pendant la rédaction de ce travail, l auteur a été financé par le C.N.R.S. et a reçu le soutien du projet ANR-14-CE25 (PerCoLaTor). The author thanks Wee Teck Gan, Philippe Gille and Gordan Savin for their remarks. 1 2 Introduction For n 1, let Spin(n) denote the Spin group of the standard Euclidean space R n and SO(n) its special orthogonal group. As is well-known there is a unique conjugacy class of compact subgroups of Spin(7) (resp. SO(7)) which are connected, semisimple, and whose root system is of type G 2. We shall call such a subgroup a G 2 -subgroup. Theorem A. Let Γ Spin(7) be a subgroup such that each element of Γ is contained in a G 2 -subgroup. Then Γ is contained in a G 2 -subgroup. This perhaps surprising result has a very simple proof. Indeed, let W be a spin representation of Spin(7), an 8-dimensional real vector space, and let E be its standard 7-dimensional real representation, which factors through SO(7). If H is a G 2 -subgroup of Spin(7), then we have an isomorphism of R[H]-modules W 1 E. As a consequence, if Γ is as in the statement of Theorem A, we must have the equality det(t γ W ) = (t 1)det(t γ E ) for all γ Γ. But this implies that the (necessarily semisimple) R[Γ]-modules W and 1 E are isomorphic. In particular, Γ fixes a nonzero vector in W. We conclude as the G 2 -subgroups of Spin(7) are well-known to be exactly the stabilizers of the nonzero elements of W. Theorem A admits a generalization over an arbitrary field, that we prove in 3. Let k be a field and E a 7-dimensional nondegenerate (or even regular, see 1.1) quadratic space over k. If C is an octonion k- algebra whose quadratic subspace of pure octonions is isometric to E, then the automorphism group of C naturally embeds both in Spin(E) and in SO(E) ( 2.5). We call such a subgroup of Spin(E) or SO(E) a G 2 -subgroup. These subgroups may not exist, but in any case they are all conjugate (by the action of SO(E), see Proposition 2.11). When k = R and E is Euclidean they are exactly the subgroups introduced above. When k is algebraically closed, they also do exist and coincide with the closed connected algebraic subgroups which are simple of type G 2 (see Proposition 2.12). Theorem B. Let k be a field, E the quadratic space of pure octonions of some octonion k-algebra, W a spinor module for E, and Γ a subgroup of Spin(E). Assume that the k[γ]-module W is semisimple. Then the following assertions are equivalent: (a) each element of Γ is contained in a G 2 -subgroup of Spin(E), (b) for each γ Γ we have det(1 γ W ) = 0, (c) Γ is contained in a G 2 -subgroup of Spin(E). This theorem is Corollary 3.5 of Theorem 3.4, of which Theorem A is the special case k = R (see Remark 3.9). Note that Theorem A also follows from the case k = C of Theorem B by standard arguments from the theory of complexifications of compact Lie groups (see 5). The naive analogue of Theorem A to SO(7) turns out that to be false. Let us introduce three embeddings that will play some role in the correct statement. We fix a complex nondegenerate quadratic plane P, denote by GO(2) the unique maximal compact subgroup of the Lie group of orthogonal similitudes of P, and by O(2) ± GO(2) the subgroup 1 of elements with similitude factor ±1. Recall that E is the standard 7-dimensional real representation of SO(7). Then: there is a group homomorphism α : Z/4Z Z/2Z SO(7), unique up to conjugacy, such that the associated representation of Z/4Z Z/2Z on E C is the direct sum of its 7 nontrivial characters; there is a morphism β : GL 2 (Z/3Z) SO(7), unique up to conjugacy, such that the representation of GL 2 (Z/3Z) on E C is the direct sum of its 4 nontrivial irreducible representations of dimension 2; there is a group morphism γ : O(2) ± SO(7), unique up to conjugacy, such that the representation of O(2) ± on E C is isomorphic to the direct sum of P, P and of the three order 2 characters of O(2) ±. In each of these cases, if Γ SO(7) denotes the image of the given morphism, we prove in 4.6 that each element of Γ belongs to a G 2 - subgroup of SO(7). Let D 8 denote the dihedral group of order 8. Theorem C. Let Γ be a subgroup of SO(7) such that each element of Γ is contained in a G 2 -subgroup. Then exactly one of the following assertions holds: (i) Γ is contained in a G 2 -subgroup of SO(7), (ii) Γ is conjugate in SO(7) to one of the groups α(z/4z Z/2Z), β(gl 2 (Z/3Z)), β(sl 2 (Z/3Z)) or γ(s), where S O(2) ± is a nonabelian subgroup, nonisomorphic to D 8, and whose similitude factors group is {±1}. Let us discuss the main steps of the proof of this theorem, which turned out to be much harder than the one of Theorem A. Let Γ be a subgroup of SO(7). We view E as a (semisimple) representation of Γ. We first show in 4.1 that each element of Γ is contained in a G 2 -subgroup of SO(7) if, and only if, we have a R[Γ]-linear isomorphism 3 1 The group O(2) ± is isomorphic to the quotient of µ 4 O(2) by the diagonal order 2 subgroup, the similitude factor being the square of the factor µ 4. 4 ( ) Λ 3 E E Sym 2 E. It is an exercise to check that the images of α,β and γ do satisfy ( ). Note also that identity ( ) implies that Γ fixes a nonzero alternating trilinear form f on E. In the special case E C is irreducible, the classification of trilinear forms given in [CH88] shows that the stabilizer off is ag 2 -subgroup, and we are done. For reduciblee C, we proceed quite differently (and do not rely on any classification result). Our starting point is the following necessary and sufficient condition for Γ to belong to a G 2 -subgroup of SO(7). Let Γ Spin(7) denote the inverse image of Γ under the natural surjective morphism Spin(7) SO(7). We show that Γ is contained in a G 2 -subgroup of SO(7) if, and only if, the restriction to Γ of the spin representation W of Spin(7) contains an order 2 character (Corollary 3.8). Our strategy to prove Theorem C is then to analyze the simultaneous structure of the C[ Γ]- modules E C and W C, when the identity ( ) holds. We argue by descending induction on the maximal dimension of a Γ- stable isotropic subspace of the complex quadratic space E, an invariant that we call the Witt index of Γ (see 1.8). This invariant can only increase when we replace Γ by a subgroup. Moreover, for a given Witt index we also argue according to the possible dimensions (and selfduality) of the irreducible summands of E C. This forces us to study in details quite a number of special cases, which is somewhat unpleasant, but leads naturally to the discovery the counterexamples α, β and γ above. Several exceptional properties of low dimensional spinor modules play a role along the proof. Our arguments apply more generally to the special orthogonal group of a 7-dimensional regular quadratic space over an arbitrary algebraically closed field k, that we assume for this introduction to have characteristic 2. We define in 4.6 a natural analogue O ± 2 (k) of the group O(2) ±, as well as analogues of the morphisms α,β and γ above (the morphism β is actually only defined if the characteristic of k is 3). Our main theorem is then the following (see Theorem 4.18). Theorem D. Let k be an algebraically closed field, E a 7-dimensional nondegenerate quadratic space over k and Γ SO(E) a subgroup. Assume that each element of Γ has the same characteristic polynomial as some element of some G 2 -subgroup of SO(E), and that Γ satisfies the semisimplicity assumption denoted by (S) in Then the same conclusion as the one of Theorem C holds, with SO(7) replaced by SO(E), and O(2) ± replaced by O ± 2 (k). We refer to loc. cit. for a discussion of assumption (S). Let us simply say here that (S) holds if the k[γ]-module E is semisimple and if the characteristic of k is either 0 or 13. The special case k = C of Theorem D is actually equivalent to Theorem C (see 5). Let us end this group theoretic discussion by raising some natural questions. Let G be a group and H a subgroup of G. Denote by P(G,H) the following property of (G,H): for all subgroup Γ of G, if each element of Γ is conjugate to an element of H, then Γ is conjugate to a subgroup of H. We have explained that P(Spin(7),G 2 ) holds and that P(SO(7),G 2 ) does not. Questions: Can we classify the couples (G,H), with G a compact connected Lie group and H a closed connected subgroup, such that P(G,H) holds? Are there remarkable couples of finite groups (G, H) such that P(G, H) holds? For instance, one can show 2 that for integers a b 1, the property P(SO(a+b),SO(a) SO(b)) holds if, and only if, we have b = 1 and a {1,3}. As another example, if S n denotes the symmetric group of {1,...,n}, then 3 P(S n+1,s n ) holds if, and only if, we have n 3. Applications to automorphic and Galois representations Our original motivation for studying property P above is that it naturally arises in some formulation of a converse statement of Langlands functoriality conjecture. Before focusing on the specific case of our study, we first briefly give the general context, assuming some familiarity of the reader with Langlands philosophy [Lan70, Lan71] and automorphic forms [Cor77]. Let F be a number field, G (resp. H) a connected semisimple linear algebraic group defined and split over F, Ĝ (resp. Ĥ) a complex semisimple algebraic group dual to G (resp. H) in the sense of Langlands, G (resp. H) a maximal compact subgroup of Ĝ (resp. Ĥ), ρ : H G a continuous homomorphism, and π a cuspidal tempered automorphic representation of G(A F ). Assume that for all but finitely many places v of F, the Satake parameter c(π v ) of π v, viewed as a well-defined conjugacy class in G, is the conjugacy class some element in ρ(h). If P(G,ρ(H)) holds, then we may expect the existence of a cuspidal tempered automorphic representation π of H(A F ) such that for almost all finite places v of F, the G-conjugacy class of ρ(c(π v)) coincides with c(π v ). 2 Hints: consider subgroups Γ of the form SO(a+b c) SO(c) with c = 0,1,3 to show that P(SO(a+b),SO(a) SO(b)) implies b = 1 and a {1,3}. To check P(SO(4), SO(3)), observe that if V is the restriction to SO(3) of the tautological 4-dimensional real representation of SO(4), then we have an isomorphism V V 1 1 Λ 2 V (a similar observation proves P(SU(4),SU(3)) as well). 3 Hint: for n = 4,5 consider for Γ the subgroup of the alternating subgroup of S n+1 preserving {n,n+1} (we have Γ S n 1 ). 5 6 Indeed, here is a heuristical argument using the hypothetical Langlands group [Lan79, Kot84, Art02]. Fix a finite set S of places of F such that for v / S then v is finite and π v is unramified, and let L F,S be the Langlands group of F unramified outside S. Following Kottwitz, this is a compact topological group, equipped for each v / S with a distinguished conjugacy class Frob v. Langlands associates to π some continuous morphismφ : L F,S G, whose image has a finite centralizer in G, such that 4 the G-conjugacy class of φ(frob v ) is c(π v ) for each v / S. The expected density (even equidistribution!) of v/ S Frob v in L F,S implies that any element in φ(l F,S ) is conjugate to some element in ρ(h), so that we have φ = ρ φ for some continuous morphism φ : L F,S H, by property P(G,ρ(H)). In turn, φ is associated to some cuspidal tempered automorphic representation π of H(A F ). Of course, the properties P(Spin(7),G 2 ) and P(SO(7),G 2 ) studied before correspond to the special cases where H is of type G 2 and G is either PGSp 6 or Sp 6. As a first application, Theorems A & C thus provide interesting conjectures in this context. We postpone the discussion of the case G = PGSp 6, about which more can be said, to the end of this introduction (see Theorem G). Note that although P(SO(7),G 2 ) does not hold, the argument above still shows that the existence of π is expected to hold in the case G = Sp 6, because none of the exceptions in the statement of Theorem C has a finite centralizer in G SO(7) (their Witt index is nonzero). When one knows how to associate to π a compatible system of l-adic Galois representations, which requires at least some assumptions on the Archimedean components of π, the absolute Galois group of F may be used as a substitute of the hypothetical Langlands group. This allows us to prove the following theorem (Corollary 6.8). We denote by W M the Weil group of the local field M, G 2 a fixed split semisimple group over Q of type G 2, and if k is an algebraically closed field of characteristic 0 we fix an irreducible polynomial representation ρ : G 2 (k) GL 7 (k). Theorem E. Let F be a totally real number field and π a cuspidal automorphic representation of GL 7 (A F ) such that π v is algebraic regular for each real place v of F. The following properties are equivalent: (i) for all but finitely many places v of F, the Satake parameter of π v is the conjugacy class of an element in ρ(g 2 (C)), (ii) for any finite place v of F, there exists a continuous morphism φ v : W Fv SU(2) G 2 (C), unique up to G 2 (C)-conjugacy, such that ρ φ v is isomorphic to the Weil-Deligne representation attached to π v by the local Langlands correspondence [HT01]. 4 As pointed out by Langlands long ago, note that these properties of φ may however not determine it uniquely up to G-conjugacy in general (see e.g. [Lar94]). We refer to 6 for the unexplained terms of this statement. A key ingredient in the proof of this theorem is the following result, which is perhaps the main application of this paper (Corollary 6.5). Theorem F. Let F be a totally real number field and π a cuspidal automorphic representation of GL 7 (A F ) satisfying assumption (i) of Theorem E, and such that π v is algebraic regular for each real place v of F. Let E be a coefficient number field for π, l a prime and λ a place of E above l. Then there exists a continuous semisimple morphism r π,λ : Gal(F/F) G 2 (E λ ), unique up to G 2 (E λ )-conjugacy, satisfying the following property: for each finite place v of F which is prime to l, and such that π v is unramified, the morphism r π,λ is unramified at v and we have the relation det(t ρ( r π,λ (Frob v ))) = det(t c(π v )), where c(π) is the Satake parameter of π v viewed as a semisimple conjugacy class in GL 7 (E). The proof of this theorem uses the existence and properties of the compatible system of 7-dimensional l-adic Galois representations associated toπ [HT01, Shi11, CH13, BC11] and Theorem D. We show that we are not in the exceptional cases of Theorem D using the knowledge of the Hodge-Tate numbers of those representations. The uniqueness assertion is a consequence of a result of Griess [Gri95]. As promised earlier, we now go back to property P(G 2,Spin(7)). After the first version of this paper appeared on the arxiv, some discussions with Gan and Savin led to the conclusion that enough is known to prove unconditionally the conjecture mentioned above concerning the Langlands functorial lifting between G 2 and PGSp 6, using works of Arthur [Art13], Ginzburg-Jiang [GJ01] and Xu [Xu14]. We are grateful to them to let us include this result in this paper: see Theorem Recall that we may take Ĝ2 = G 2 (C) and PGSp 6 = Spin 7 (C). Theorem G. Let F be a number field and π a cuspidal automorphic representation of PGSp 6 (A F ). Assume that π is tempered, or more generally, that π is nearly generic (see 6.12). Then the following properties are equivalent: (a) for all but finitely many places v of F, the Satake parameter c(π v ) is the conjugacy class of an element of a G 2 -subgroup of Spin 7 (C), (b) there exists a cuspidal automorphic representation π of G 2 (A F ) such that for all but finitely many finite places v of F, the image in Spin 7 (C) of the Satake parameter c(π v ) is conjugate to c(π v). 7 8 Contents Introduction 2 1. Preliminaries on quadratic spaces and spinors 8 2. Octonion algebras and G 2 -subgroups Proof of Theorems A & B Proof of Theorem D Proof of Theorem C Automorphic and Galois representations 53 Appendix A. Some basic facts about restrictions and extensions of cusp forms 62 References Preliminaries on quadratic spaces and spinors 1.1. Regular quadratic spaces. Let k be a field. A quadratic space over k is a finite dimensional k-vector space V equipped with a quadratic map q : V k [Bou59, 3 no.4]. By definition, the map β V (x,y) := q(x + y) q(x) q(y), V V k, is a symmetric k- bilinear form, and we have q(λv) = λ 2 q(v) for all v V and λ k. We refer to Bourbaki [Bou59, 3, 4] and Knus [Knus88, Ch. 1] for the basic notions concerning quadratic spaces (isometries, orthogonal sums, base change, etc...). A quadratic space V over k will be called nondegenerate if the bilinear form β V is a perfect pairing. If the characteristic of k is 2, then β V is alternate, thus dim V has to be even if V is nondegenerate. It will be convenient to say that a quadratic space V over k is regular if either V is nondegenerate, or if we are in the following situation: dimv is odd, the characteristic of k is 2, the kernel of β V is one dimensional and q is not identically zero on it. 5 If I is a finite dimensional k-vector space, we denote by I its dual vector space and by H(I) the quadratic space I I with q(x+ϕ) = ϕ(x) ( hyperbolic quadratic space over I ). This is a nondegenerate quadratic space. The isometry group of a quadratic space V, denoted O(V), is the subgroup of elements g GL(V) such that q g = g. When V is regular, there is a well-defined subgroup SO(V) O(V) of proper isometries: when 2 k or dimv is odd, we set SO(V) = O(V) SL(V), and in the remaining case (dimv even and 2 k ) SO(V) is an index 2 subgroup of O(V) defined as the kernel of the Dickson invariant of V 5 We warn the reader that there does not seem to be any standard terminology for these classical notions. Several authors, such as Knus, use the term nonsingular for nondegenerate. Moreover, V is 1 2 regular is the sense of [Knus88, Knus91] if and only if dimv is odd and V is regular in our sense. [Knus88, Ch. 6 p.59]. We also denote by GO(V) GL(V) the subgroup of orthogonal similitudes of the quadratic space V, and when V is regular, by GSO(V) its subgroup of proper similitudes (see e.g. the end of II.1 in [CL14]) Clifford algebras, Spin groups, and their relatives. Let V be a quadratic space over k. Recall that the Clifford algebra C(V) of V is a k-superalgebra [Bou59, 9] [Knus88, Ch. 4] [Del99]. It is equipped with a canonical injective morphism k V C(V), that we shall always view as an inclusion. The Clifford group of

We Need Your Support

Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...Sign Now!

We are very appreciated for your Prompt Action!

x