manuscripta math. 104, 173 – 200 (2001) © Springer-Verlag 2001
Mahdi Asgari · Ralf Schmidt Siegel modular forms and representations
Received: 28 March 2000 / Revised version: 25 October 2000
Abstract. This paper explicitly describes the procedure of associating an automorphic rep- resentation of PGSp(2n, A) with a Siegel modular form of degree n for the full modular group n = Sp(2n, Z), generalizing the well-known procedure for n = 1. This will show that the so-called “standard” and “spinor” L-functions associated with such forms are ob- tained as Langlands L-functions. The theory of Euler products, developed by Langlands, applied to a Levi subgroup of the exceptional group of type F4, is then used to establish meromorphic continuation for the spinor L-function when n = 3.
1. Introduction
Let f be a Siegel modular form of degree n for the full modular group n = Sp(2n, Z).Iff is an eigenfunction for the action of the Hecke algebra, then there are two L-functions associated with f . Let a0,a1,... ,an be the Satake parameters of f , and define the standard L-function n −1 −s −s −1 −s L1(s, f ) = (1 − p ) (1 − aip )(1 − ai p ) , (1) p i=1 and the spinor L-function n −1 L (s, f ) = ( − a a ...a p−s) . 2 1 0 i1 ik (2) p k=0 1≤i1<... M. Asgari: Department of Mathematics, The University of Michigan, Ann Arbor, MI 48109- 1109, USA. e-mail: [email protected] R. Schmidt: Universität des Saarlandes, Fachrichtung 6.1 Mathematik, Postfach 15 11 50, 66041 Saarbrücken, Germany. e-mail: [email protected] Mathematics Subject Classification (2000): 11F46 174 M. Asgari, R. Schmidt Theorem. The L-functions L1(s, f ) and L2(s, f ) have meromorphic continuation to all of C when n = 3. We should remark that Böcherer has proved stronger results for the standard L-functions (cf. [Bö]). Andrianov also gives all the analytic properties of spinor L-function when n = 2 in [An]. The procedure of associating an automorphic representation to a classical mod- ular form is well known in the case of elliptic modular forms, see [Ge] Chapter 3 or [Bu, Sect. 3.6]. There, one associates with an eigenform f an automorphic representation πf of GL(2, A), where A denotes the adeles of Q.Iff has no char- acter, then πf will have trivial central character, thus descends to a representation of PGL(2, A). In the higher dimensional case we will associate with an eigenform f of degree n an automorphic representation of the group GSp(2n, A) which is isomorphic to GL(2, A) if n = 1. Since f will be assumed to have no character, this is really a representation of PGSp(2n, A). To be more precise, utilizing a strong approximation theorem, we first lift f to a function f on G(A), where G = GSp(2n). We will assume f to be cuspidal, L2(G(Q)\G(A)). so that f lies in the cuspidal subspace 0 (Here we have to trans- late the classical cusp condition into the group theoretic one.) Then let π be the subrepresentation of this space generated by f . This π may not be irreducible, but if f is an eigenform, then all the irreducible components of π will turn out to be isomorphic. This isomorphism class is the automorphic representation πf associated with f . In Theorem 2 we shall describe its local components in terms of Satake parameters (at the finite places) and Harish-Chandra parameters (at the infinite place). We have to go through some Hecke algebra computations to identify the clas- sical Satake parameters of the eigenform f with the Satake parameters of the local components of πf which are spherical representations of the local groups GSp(2n, Qp). The group theoretic Satake parameters can be taken to be in the maximal torus in the dual group of G which is Gˆ = GSpin(2n + 1, C). Since we have a representation with trivial central character, the Satake parameters will in fact be elements of its derived group Spin(2n + 1, C). This latter group has two distinguished finite-dimensional representations, n namely the projection onto SO(2n + 1, C) which we denote by 1, and the 2 - dimensional spin representation 2, the smallest genuine representation (not factor- ing through 1) of this group.We use the weight structure of these finite-dimensional representations to determine the form of the standard Euler factor associated with 1 and 2. It turns out that these Euler factors are precisely the same as the ones in (1) and (2). In other words, the classical L-functions identify with standard Lang- lands L-functions coming from the representations 1 and 2 of the dual group (see Theorem 3). Having established this identification, one can apply results from representation theory to classical modular forms. For modular forms of degree 3, the underlying group GSp(6) can be embedded, as a Levi, in a Chevalley group of type F4. We then make use of Langlands’ method developed in [La] to prove the meromorphic continuation of the spin L-function for n = 3 (cf. Corollary 1). Siegel modular forms and representations 175 2. Notation The right group for handling Siegel modular forms (of degree n) in the context of automorphic representations is t G = GSp(2n) ={g ∈ GL(2n) :∃µ(g) ∈ GL(1)gJg = µ(g)J }, where 1n J = , 1n the n × n identity matrix. −1n AB g = If CD , then the conditions are equivalent to t t t t t t A D − B C = µ(g)1,AB = B A, C D = D C. The function µ is called the multiplier homomorphism. Its kernel is the group Sp(2n) and there is an exact sequence 1 −→ Sp(2n) −→ G −→ GL(1) −→ 1. The center Z of G consists of the scalar matrices, and the standard maximal torus is T ={diag(u1,... ,un,v1,... ,vn) : u1v1 = ...= unvn = 0}. We often write an element t ∈ T in the form t = (u ,... ,u ,u−1u ,... ,u−1u ), u ∈ ( ); diag 1 n 1 0 n 0 i GL 1 (3) then u0 = µ(t). We fix the following characters of the maximal torus T ⊂ G.If t ∈ T is written in the form (3), then let ei(t) = ui,i= 0, 1,... ,n. These characters are a basis for the character lattice of G, X = Ze0 ⊕ Ze1 ⊕ ...⊕ Zen. We also fix the following cocharacters of T : f (u) = ( ,... , ,u,...,u), 0 diag 1 1 n n f (u) = (u, ,... , ,u−1, ,... , ), 1 diag 1 1 1 1 n n . . f (u) = ( ,... , ,u, ,... , ,u−1). n diag 1 1 1 1 n n 176 M. Asgari, R. Schmidt Then these elements are a Z-basis for the cocharacter lattice of G, ∨ X = Zf0 ⊕ Zf1 ⊕ ...⊕ Zfn. With the natural pairing , : X × X∨ −→ Z, we have ei,fj =δij . We now choose the following set of simple roots: α (t) = u−1 u , ... ,α (t) = u−1u ,a(t) = u2u−1; 1 n−1 n n−1 1 2 n 1 0 (4) here t is written in the form (3). In other words, α1 = en − en−1, ... , αn−1 = e2 − e1,αn = 2e1 − e0. The numbering of the simple roots is such that the Dynkin diagram is ••··· •••< α1 α2 αn−2 αn−1 αn The corresponding coroots are α∨ = f − f , ... α∨ = f − f ,α∨ = f . 1 n n−1 n−1 2 1 n 1 (5) R ={α ,... ,α }⊂X R∨ ={α∨,... ,α∨}⊂X∨ If we let 1 n , 1 n , then (X,R,X∨,R∨) is the root datum of G = GSp(2n). The Cartan matrix is 2 −1 −12−1 −12−1 α ,α∨= ...... . i j . . . −12−1 −12−1 −22 With our choice of simple roots, we get the Borel subgroup B = TN, where A 0 1 B N consists of matrices of the form , with A ∈ GL(n) lower 0 tA−1 0 1 triangular unipotent and B symmetric. (Note that some authors use a different set of simple roots that results in A being upper triangular as was pointed out by the referee.) The torus T acts on the Lie-algebra n of N by the adjoint representation Ad. It is easy to compute the modular factor δB (t) = det(Adn(t)).Ift is given in the form (3), then δ (t) = u−n(n+1)/2u2u4 · ...· u2n. B 0 1 2 n (6) Siegel modular forms and representations 177 3. Local representations In Theorem 2 we shall describe the local components of an automorphic repre- sentation associated with a classical Siegel modular form. This is done in terms of Satake parameters at the finite places, and in terms of the weight (or Harish- Chandra parameters, if the weight is large enough) at the archimedean place. In the following sections we shall therefore collect all the required facts about local representations. Notation. In Sects. 3.1 through 3.4, F is a p-adic field, O its ring of integers, ω ∈ O a generator of the maximal ideal, and q = *O/ωO. The symbols G,T,... denote F -points of the underlying algebraic groups. We let K = GSp(2n, O). Section 3.5 deals with the archimedean place, i.e., the underlying local field is R. 3.1. The Satake isomorphism for GSp(2n) The Satake isomorphism for a reductive p-adic group is nicely described in [Ca]. Let H(G, K) be the unramified Hecke algebra of G, consisting of compactly supported functions f : G → C which are left and right K-invariant. The product in H(G, K) is given by convolution (f ∗ g)(x) = f (xy)g(y−1)dy. G Let ◦T := T(O). Then we also have the Hecke algebra H(T , ◦T). Special elements in this Hecke algebra are ∗ ∗ ∗ ∗ X := char diag(O ,... ,O ,ωO ,... ,ωO ) , 0 ∗ ∗ ∗ −1 ∗ ∗ ∗ X1 := char diag(ωO , O ,... ,O ,ω O , O ,... ,O ) , . . ∗ ∗ ∗ ∗ ∗ −1 ∗ Xn := char diag(O ,... ,O ,ωO , O ,... ,O ,ω O ) , where “char” stands for characteristic function. It is easily seen that Xk = (O∗,... ,O∗,ωkO∗,... ,ωkO∗) ,k∈ Z, 0 char diag and similarly for the other Xi. It is then clear that H(T , ◦T)= C[X±1,X±1,... ,X±1]. 0 1 n Now, for an element f ∈ H(G, K), the Satake transform is defined by 1/2 −1/2 (Sf )(t) =|δB (t)| f(tn)dn=|δB (t)| f (nt) dn. N N 178 M. Asgari, R. Schmidt Sf is an element of H(T , ◦T), and in fact S defines an isomorphism S : H(G, K) −→∼ H(T , ◦T)W . Here W denotes the Weyl group of G, which acts naturally on the torus and on ◦ H(T , T), and we take invariant elements. Assume now that f = char(KMK) with KMK = MiK. Because of the Iwasawa decomposition G = BK we may assume that d ω i1 0 A B M = i i A = .. . i di t −1 with i . (7) 0 ω 0 Ai ∗ ωdin Here ω is a prime element and the dij are integers. Note that di0 does not depend on i, since it equals the valuation of µ(M). Lemma 1. With f as above, we have n Sf = qδn(n+1)/4 Xδ (q−j X )dij . 0 j i j=1 where δ is the valuation of µ(M). Proof. We have to compute 1/2 (Sf )(t) =|δB (t)| 1Mi K (tn) dn, i N k k −k +k −k +k where we may assume t = diag(ω 1 ,... ,ω n ,ω 1 0 ,... ,ω n 0 ). Consider (tn) dn i tn ∈ M K n ∈ t−1M K N 1Mi K for fixed .Wehave i if and only if i .It − is clear that we can find such an n only if t 1Mi has units on the diagonal, i.e., − if kj = dij for all j. Assuming this is the case, then with n := t 1Mi ∈ N our integral equals 1nK (n) dn = 1K (n) dn = 1. N N We have proved that 1ifkj = dij for all j = 0, 1,... ,n, 1Mi K (tn) dn = . N 0 otherwise With the characteristic functions Xj defined above, this may be written as d d t −→ (tn) dn = X i0 X i1 · ...· Xdin. 1Mi K 0 1 n (8) N It follows from (6) that 1/2 k n(n+1)/4 −k −2k −...−nkn |δB (t)| = q 0 q 1 2 . Multiplying this by the function (8), we may replace kj by dij , and the assertion follows. ! Siegel modular forms and representations 179 3.2. Spherical representations An irreducible admissible representation of G is called spherical if it contains a non-zero vector fixed by K. All the spherical representations of G are obtained ∗ as follows. Let χ0,... ,χn be unramified characters of F (i.e., homomorphisms F ∗ → C∗ which are trivial on O∗). They define an unramified character of the Borel subgroup B = TN which is trivial on N and which, on T ,isgivenby t −→ χ0(u0)χ1(u1) · ...· χn(un), (9) with t ∈ T of the form (3). Normalized induction to G yields a representation which has a unique spherical constituent. We denote this spherical representation by π(χ0,χ1,... ,χn). The isomorphism class of this representation depends only on the unramified char- acters modulo the action of the Weyl group. It is further known that each spherical representation is obtained in this way. Thus there is a bijection between unramified characters of T modulo the action of the Weyl group, and isomorphism classes of spherical representations of G. Each unramified character of F ∗ is determined by its value on a prime ele- ment ω ∈ F . This value may be any non-zero complex number. With the Satake parameters bi := χi(ω), the character (9) is thus also determined by the vec- ∗ n+1 tor (b0,b1,... ,bn) ∈ (C ) . The Weyl group acts on this complex torus, and we see that unramified representations of G are parameterized by the orbit space (C∗)n+1/W. In fact, the Satake isomorphism S : H(G, K) −→∼ C[X±1,X±1,... ,X±1]W 0 1 n identifies the Hecke algebra with the coordinate ring of (C∗)n+1/W. Each point ∗ n+1 (b0,... ,bn) ∈ (C ) /W determines a character, i.e. an algebra homomorphism C[X±1,... ,X±1]W → C X b S 0 n , by mapping i to i.Via this also defines a character of H(G, K), which is nothing but the action of H(G, K) on the one-dimensional space of spherical vectors in π(χ0,... ,χn). These well-known facts may be summarized in the following commutative diagram, in which all the maps are bijections: {spherical representations}−−−−→ Hom (H(G, K), C) Alg {unramified characters}/W ←−−−− (C∗)n+1/W The left arrow is induction and then taking the spherical constituent, the top arrow is the action of H(G, K) on the space of spherical vectors, the map on the right H(G, K) # C[X±1,... ,X±1]W comes from the identification 0 n , and the bottom arrow assigns to the Satake parameters (b0,... ,bn) the unramified character with χi(ω) = bi. 180 M. Asgari, R. Schmidt Lemma 1 can now be restated in the following way. Let v be a spherical vector in π := π(χ0,... ,χn), and let f = 1KMK be the element from Lemma 1. Then n π(f)v = qδn(n+1)/4 bδ (q−j b )dij v, b = χ (ω). 0 j where j j (10) i j=1 Clearly the representation induced from the character (9) has central character χ 2χ ...χ 0 1 n. This character is trivial if and only if the Satake parameters satisfy b2b · ...· b = . 0 1 n 1 (11) Thus, exactly these parameters give spherical representations of the group PGSp(2n, F ). 3.3. Dual groups In Sect. 2 we described the root datum (X,R,X∨,R∨) of G = GSp(2n). By definition, the dual group Gˆ = GSpin(2n + 1, C) has root datum (X,R,X∨,R∨) = (X∨,R∨,X,R). e ,... ,e X Let 0 n be a basis of such that ei = fi X = X∨ f ,... ,f X∨ under the identification , and let 0 n be a basis of such that fi = ei ∨ under the identification X = X. The complex torus (C∗)n+1 of the previous section may be viewed as the maximal torus Tˆ ⊂ Gˆ . Elements of T/Wˆ are in one-one correspondence with semisimple conjugacy classes in Gˆ . By the previous section, we have a correspondence between spherical representations of G and those semisimple conjugacy classes. If χ is an unramified character of T defining a spherical representation, and tˆ ∈ Tˆ is the associated parameter, then χ and tˆ are related by the general relation ∨ χ(ϕ(ω)) = ϕ(t)ˆ for all ϕ ∈ X = X (12) (see [GS] p. 26). It is clear that each tˆ ∈ Tˆ can be written uniquely as n ∗ tˆ = fi (ti), ti ∈ C . (13) i=0 ˆ We have the pairing ei,fj =δij , and thus ti = ei(t). Therefore putting ϕ = ei = fi in (12) yields bi = ti. We proved the following. Siegel modular forms and representations 181 Lemma 2. Let π = π(χ) be a spherical principal series representation of G = GSp(2n, F ). Let b0,b1,... ,bn be the Satake parameters of π, as in 3.2. Then the associated semisimple conjugacy class in GSpin(2n + 1, C) is represented by n tˆ = fi (bi). i=0 Set G := PGSp(2n, F ) = G/C, where C is the center of G. The dual group of G is ˆ G = Spin(2n + 1, C) ⊂ GSpin(2n + 1, C), which is also the derived group of Gˆ . The element t ∈ Tˆ will correspond to a representation of G with trivial central character, i.e., to a representation of G,if and only if it lies in ˆ T := Tˆ ∩ Spin(2n + 1, C), the maximal torus of Spin(2n + 1, C). 3.4. Local Euler factors ˆ For a spherical representation π of G with parameter tˆ ∈ T and a finite-dimensional representation : Spin(2n + 1, C) −→ GL(m, C), − Langlands defines the local Euler factor L(s,π,)= det 1 − (t)qˆ −s 1. In this section we are going to compute this factor for the “projection representation” 1 : Spin(2n + 1, C) −→ SO(2n + 1, C), (14) n and for the 2 -dimensional spin representation 2. Similar computations can also be found in [As]. The projection representation. One can find the weight structure of the representa- tion 1 in [FH]. All the weight spaces of this (2n + 1)-dimensional representation are one-dimensional. In the notation of the previous section the weights are e , ... , e , , −e , ... , −e . 1 n 0 1 n The eigenvalues of an operator ρ(t)ˆ are therefore 1 and ±1 ei(t)ˆ ,i= 1,... ,n. ˆ ˆ With t as in Lemma 2, we get ei(t) = bi. This proves the following. Lemma 3. If π is a spherical representation of G¯ with Satake parameters b0,b1,... ,bn as in 3.2, then n −1 −s −s −1 −s L(s,π,1) = (1 − q ) (1 − biq )(1 − bi q ). i=1 182 M. Asgari, R. Schmidt The spin representation. Again from [FH], the 2n-dimensional spin representation ( n + , C) 1 (e + ...+ e ) 2 of Spin 2 1 has highest weight 2 1 n . Indeed, all the weight spaces are one-dimensional, and the weights of 2 are c e + ...+ c e 1 1 n n ,ci ∈{±1}. (15) 2 Let such a weight be given. Define a character η of Tˆ by n n ˆ ˆ η(t) = t0 ti, if t = fi (ti). i=1 i=0 ci =1 ˆ We claim that η coincides with the character (15) on T , the maximal torus of Spin(2n + 1, C). An element tˆ as above lies in this maximal torus if and only if t2t · ...· t = 0 1 n 1. Assuming this relation, we have n n (c e + ...+ c e )(t)ˆ = e (t)ˆ ci = tci 1 1 n n i i i=1 i=1 n n = t2 tci +1 = t2 t2 = ( η)(t),ˆ 0 i 0 i 2 i=1 i=1 ci =1 1 ˆ and it follows that indeed (c1e + ...+ cnen) = η on T . We can thus easily 2 ˆ 1 compute the eigenvalues of 2(t), and get the following lemma. Lemma 4. If π is a spherical representation of G¯ with Satake parameters b0,b1,... ,bn as in 3.2, then n L(s,π, )−1 = ( − b b ...b q−s). 2 1 0 i1 ik k=0 1≤i1<... 3.5. Lowest weight representations for PGSp(2n, R) In this section we shall construct a class of lowest weight representations of PGSp(2n, R), which appear as the infinite component in the automorphic rep- resentation attached to a holomorphic Siegel modular form. The Lie algebra of Sp(2n, R) is explicitly given by t g = X ∈ M(2n, R) : XJ + J X = 0 AB = ∈ M( n, R) : B = tB, C = tC, A =−tD . CD 2 (16) The standard maximal compact subgroup K∞ of Sp(2n, R) is AB K = ∈ ( n, R) : A tA + B tB = ,AtB = B tA . ∞ −BA GL 2 1 (17) Siegel modular forms and representations 183 AB K # U(n) −→ A + iB K We have ∞ via −BA . The Lie algebra of ∞ is AB k = ∈ M( n, R) : A =−tA, B = tB . −BA 2 This is also the 1-eigenspace of the Cartan involution θX =−tX. The (−1)- eigenspace is AB p = ∈ M( n, R) : A = tA, B = tB , B −A 2 + − so that g = k ⊕ p. It is easy to see that pC = pC ⊕ pC, with A ±iA p± = ∈ M( n, C) : A = tA . C ±iA −A 2 (18) ± ± ± Each of pC is an abelian subalgebra of gC,wehave[kC, pC]⊂pC, and + − gC = pC ⊕ kC ⊕ pC. In the complexified Lie-algebra kC of K let 0 Di Ti =−i , (19) −Di 0 where Di is the diagonal matrix with entry 1 at position (i, i), and zeros elsewhere. Then h = RT1 + ...+ RTn is a compact Cartan subalgebra of gC. Let ei be the linear form on hC which sends Ti to 1, and Tj to 0 for j = i. Then the following is a system of positive roots for (gC, hC): 2ej , 1 ≤ j ≤ n, ej + ek, 1 ≤ j