CONSTRUCTION OF SERIES OF DEGENERATE REPRESENTATIONS FOR GSP(2) AND PGL(N)
DISSERTATION
Presented in Partial Fulfillment of the Requirements for
the Degree Doctor of Philosophy in the Graduate
School of the Ohio State University
By
Martin Bozhidarov Nikolov, M.Sc.
*****
The Ohio State University 2008
Dissertation Committee: Approved by Professor Yuval Flicker, Advisor
Professor James Cogdell Advisor Professor Wenzhi Luo Graduate Program in Mathematics
ABSTRACT
The relative trace formula is a tool for studying automorphic representations on symmetric spaces. In this work two relative trace formulas are studied: One for
GSp(2) relative to a form of SO(4) and one for PGL(n) relative to GL(n − 1).
The first trace formula is used to study the Saito-Kurokawa lifting of automor- phic representations from PGL(2) to PGSp(2), thus characterising the image as the representations with a nonzero period for the special orthogonal group SO(4, E/F ) associated to a quadratic extension E of the base field F , and a nonzero Fourier coefficient for a generic character of the unipotent radical of the Siegel parabolic subgroup.
The second trace formula is used to study the representations π of PGL(n) which have a nonzero linear form invariant under the Levi subgroup GL(n−1), and a certain nonvanishing degenerate Fourier coefficient, over local and global fields. The result
0 is that these π are of the form I(1n−2, π ), namely normalizedly induced from the parabolic subgroup of type (n − 2, 2), trivial representation on the GL(n − 2)-factor and cuspidal π0 on the PGL(2)-factor. The analysis of the formula is intricate our main achievement is rigorously establishing the convergence of the formula. New in this case is that none of the representations of PGL(n) involved are cuspidal, they rather occur in the continuous spectrum, but discretely in our formula.
ii To Lore Thaler
iii ACKNOWLEDGMENTS
First and foremost, I would like to thank my advisor Professor Yuval Flicker for allowing me to work on this thesis, and for all of his time, encouragement, concern, and patience with me. I have appreciated the many conversations and discussions we have had, both mathematical and otherwise.
I would also like to thank all my committee members, my advisor Professor Yuval
Flicker, Professor James Cogdell, and Professor Wenzhi Luo, for agreeing to serve on my committee.
I would like to thank the Department of Mathematics for their support expressed through several teaching and research fellowships that I enjoyed.
iv VITA
2002-Present ...... Ph.D. Candidate, The Ohio State Univer- sity
2002 ...... M.Sc. Mathematics, Sofia University
2001-2002 ...... Researcher, Bulgarian Academy of Sci- ences, Institute of Mathematics and In- formatics,
2000-2001 ...... University of Electro Communications, Tokyo Japan
1997-2002 ...... Sofia University “St. Kliment Ohirdski”
FIELDS OF STUDY
Major Field: Mathematics
Specialization: Number Theory
v TABLE OF CONTENTS
Abstract ...... ii
Dedication ...... iii
Acknowledgments ...... iv
Vita...... v
CHAPTER PAGE
1 Introduction ...... 1
2 Cusp Forms on GSp(2) with SO(4)-Periods ...... 10
2.1 Definitions and Notations ...... 10 2.2 The group ...... 20 2.3 Geometric side ...... 25 2.4 Asymptotes ...... 32 2.5 Matching ...... 39 2.6 Corresponding ...... 45 2.7 Spectral analysis ...... 51 2.8 Computing the integrals of truncated Eisenstein series . . 57 2.9 The contribution from the maximal parabolics ...... 64 2.10 The contribution from the minimal parabolic ...... 70 2.11 Comparison ...... 79 2.12 Converse ...... 89 2.13 Split cycles ...... 93 2.14 Invariance of Fourier coefficients of cyclic automorphic forms 96
3 A Trace Formula for the Symmetric Space PGL(n)/ GL(n − 1) . . . . . 109
3.1 Notations ...... 109 3.2 Truncation ...... 115
vi 3.3 The Main Distribution ...... 126 3.4 Regularized Periods of Eisenstein series ...... 128 3.5 Periods of Truncated Eisenstein Series ...... 136 3.6 Limit Formula for J T (f) ...... 137 3.7 An application ...... 146
Bibliography ...... 154
vii CHAPTER 1
INTRODUCTION
The relative trace formula is a tool to study period integrals of automorphic forms and their relation to the principle of functoriality. For a cusp form φ on the group
G(A) of ad`elepoints on a connected reductive group G over a number field F , these integrals are of the form Z φ(h)dh. S(F )\S(A) Here S is the group of fixed points of an involution of G over F . A representation
π of G(A) is called S-cyclic, or just cyclic if S is understood, if there is a φ in its representation space whose period integral is nonzero. It is expected that the cyclic representations are precisely those that are the functorial transfer from another group
G0. For more on the relative trace formula see [Lap1]. The present work consists of the study of to cases of the relative trace formula and their applications. Namely, the case first case studies the pair G = GSp(2) and S = SO(4), the second deals with
G = GL(n) and S = GL(n − 1).
Chapter 2 concerns the determination of cusp forms on an ad´elegroup G(A) whose period − namely integral − over a closed subspace (“cycle”) arising from a subgroup
H(A), is nonzero. Such forms contribute to the cohomology of the symmetric space
G/H, and play a role in lifting automorphic forms to G(A) from another group
1 G0(A). Many advances in these studies so far have been made by means of the theory of the Weil representation [We1]; see Waldspurger [Wa1], Howe and Piatetski-Shapiro
[HPS], [PS1], Kudla-Rallis [KR], Oda [O]. This technique has the advantage − in addition to early maturity − of constructing cusp forms on G(A) directly from such forms on G0(A). Miraculously, the cusp forms on G(A) so obtained happen to have nonzero H(A)-periods. Our approach is based on a more naive and direct method, focusing more on the representation and its properties rather than on its particular realization. Thus we integrate both the spectral and the geometric expressions for the kernel Kf (x, y) of the convolution operator on the space of cusp form on G(A), over the cycle associated with
H(A). If both variables x and y are integrated over the cycle, one obtains a bi-period summation formula, involving the periods of the automorphic forms over the cycles
(in Jacquet [J1], and later in [FH], this is named a “relative trace” formula, although there are no traces in that formula). The case where G(A) is G0(A) × G0(A) and
H(A) is G0(A) embedded diagonally, coincides with that of the usual trace formula on G0(A); this case is also referred to as the “group case”. If the second variable is integrated over a unipotent radical of a parabolic subgroup against an additive character, the Fourier summation formula − involving Fourier co- efficients of the automorphic forms − is obtained (see Jacquet [J2], where the formula is again named “relative trace” formula, and [F5], [F6] and [F4]). Only the cusp forms on G(A) with nonzero H(A)-periods survive on the spectral side. The geometric side is compared with the geometric side of an analogous summation formula on G0(A), for matching test functions on G(A) and G0(A). The resulting identity of spectral
2 sides can be used to establish lifting from G0(A) to G(A). In summary, both the bi-period and the Fourier summation formulae are special instances of the “relative trace” formula.
The derivation of the Fourier summation formula, and the characterization of the relevant orbital integrals, lead to deep chapters in global, and local, harmonic analy- sis, especially of symmetric spaces; cf., e.g., [OM], [BS]. The analytic problems thus raised might even be considered to be of greater importance than the motivating final applications in representation theory. Conversely, these applications justify some of the work which has been done on symmetric spaces. One expects to derive identi- ties of (bi-period or) Fourier-Period distributions intrinsically related to the (local) representation in question. These distributions are analogous to Harish-Chandra’s characters, which play a key role in studies of automorphic forms by means of the
Selberg trace formula. To fully harvest the (bi-period, or) Fourier-Period summation formulae, one would need an analogue of the orthogonality relations of characters, due to Harish-Chandra and Kazhdan [K], for these local distributions. The summation formula has been slow to evolve possibly since its application is based on panoply of techniques, substantially different from each other. Yet it could be a source of inspiration in various branches of contemporary harmonic analysis.
In chapter 2 we study the case of automorphic forms on G = GSp(2) and the cycle
H = SO(4, E/F ) associated to a quadratic separable extension E of the global base
field F . More precisely, G is the algebraic group of g ∈ GL(4) with gJ tg = λJ, λ =
0 w 0 1 × λ(g) ∈ GL(1),J = ( −w 0 ) , w = ( 1 0 ); A is the ring of ad´eles of F . We fix θ ∈ F which is not a square in F , put θ = ( 0 1 ) and Θ = θ1 0 , and let H be the centralizer 1 θ 0 0 θ1
3 of Θ in G. Denote by Z(F ),Z(A) the center of G(F ) and G(A) respectively. Also
IX x y consider the unipotent radical N = {n = ( 0 I ); X = ( z x )} of the Siegel parabolic subgroup P of type (2,2) of G, a complex valued nontrivial character ψ of the additive
0 1 group A/F , and the character ψθ(n) = ψ(trTX) = ψ(z − θy),T = ( −θ 0 ). Our main global achievement in this work is to advance the theory of the Fourier summation formula, namely develop such a formula by expanding geometrically and spectrally the integral of the kernel Kf (n, h) of the standard convolution operator r(f) (for a test function f) on the space of automorphic forms. In fact we multiply
Kf (n, h) by ψθ(n), and integrate over n ∈ N(F )\N(A) and h ∈ Z(A)H(F )\H(A). The Fourier summation formula is recorded in Proposition 18. On the spectral side we use the mixed truncation following [JLR]. On the geometric side we obtain a new type of orbital integrals of the form R R f (nγh)ψ (n)dndh. Nv Hv v θ Our main local achievement is in characterizing the functions of γ thus obtained, by studying their asymptotic behavior; see Proposition 8. This study involves integra- tion over a certain quadric in the affine 5-space. We are led to Fourier analysis with respect to quadratic forms, involving Weil’s factor γψ. Underlying our computations is the stationary phase method, where we use the Morse Lemma. We discover that the asymptotic behavior of these Fourier orbital integrals is compatible with that of analogous Fourier orbital integrals obtained in the analysis of the Fourier summa-
∗ 0 1 ∗ tion formula for N\PGL(2)/A, A = ( 0 1 ) ,N = ( 0 1 ); see Proposition 12 (due to Jacquet [J2]). We relate these Fourier orbital integrals on PGSp(2) and PGL(2),
4 proving the existence of matching, in Corollary 13 and compare the summation for- mula for PGSp(2) (Proposition 18) with that of PGL(2) (Proposition 16, [F5], [J2]) in Proposition 35.
In Proposition 17 we record the statement that naturally related spherical functions on PGSp(2) and PGL(2) have matching Fourier orbital integrals. The case of the unit elements in the Hecke algebras is proven in Proposition 14. The general case was proposed as a conjecture in an early draft of this work, and it has later been proved in Zinoviev’s OSU thesis [Z], using the case of the unit elements. It would be interesting to find an alternative proof of this “fundamental lemma”, possibly based on a “symmetric space” analogue of the regular functions technique of [F3], which might reduce the spherical case to that of smooth test functions, or to that of the unit element in the Hecke algebra, which are analyzed here.
As an application of our summation formula and study of orbital integrals we recover a result of Piatetski-Shapiro [PS1], which − together with the work of Weissauer [Wr1]
− in fact motivated our study. Let ρv be an admissible representation of PGL(2,Fv), and ζ a complex number. Write I(ρv, ζ) for the Gv = PGSp(2,Fv)-module on the space of φ : Gv → ρv which satisfy
3 A ∗ −1 ζ+ 2 φ 0 λw tA−1w g = |λ det A| ρv(A)(φ(g))
× 1 for A ∈ PGL(2,Fv), g ∈ Gv and λ ∈ Fv . Write J(ρv, 2 ) for the Langlands’ quotient 1 (see [BW]) of I(ρv, 2 ) (it is unramified if so is ρv, and nontempered if ρv is unitariz- able). Proposition 36 asserts that if ρ is a cuspidal representation of PGL(2, A) with
1 × × L( 2 , ρ⊗χθ) 6= 0 (χθ is the quadratic character of A /F associated with the quadratic
5 √ 1 extension E = F ( θ) of F ) and L( 2 , ρ) = 0, then there exists a cuspidal represen- 1 tation π of PGSp(2, A) with πv ' J(ρv, 2 ) for almost all v. This π is H(A)-cyclic, R namely the period ΠH (Φ) = Φ(h)dh over the cycle Z( )H(F )\H( ) is Z(A)H(F )\H(A) A A R nonzero for some Φ ∈ π, and is θ-generic, namely Wψ (Φ) = Φ(n)ψθ(n)dn θ N(F )\N(A) is nonzero for some (possibly other) Φ ∈ π. More precise local results could be obtained from our global theory had we had orthogonality relations for our Fourier-
Period distributions, analogous to those of Kazhdan [K] in the case of characters.
Conversely, Proposition 37 asserts (but only in the function field case) that given an H(A)-cyclic θ-generic discrete series representation π of PGSp(2, A), either there 1 1 exists a cuspidal PGL(2, A)-module ρ with L( 2 , ρ ⊗ χθ) 6= 0 and πv ' J(ρv, 2 ) for 1 almost all v, or πv ' J(χθ,v ◦det, 2 ) for almost all v (here χθ ◦det is a one-dimensional, residual, discrete series representation of PGL(2, A) defined by the quadratic char- acter χθ). A brief discussion of the description of packets of such representations of
PGSp(2, A) is given in the beginning of Section 2.12, see [F9].
In Section 2.13 we explain why there are no cusp forms on G(A) with periods by
ε 0 1 0 the split cycle SO(4) = H0 = ZG (( 0 −ε )) , ε = ( 0 −1 ). For this reason we consider only SO(4, E/F )-periods. In Section 2.14 we employ another form of the Fourier summation formula to study invariance of Fourier coefficients of cusp forms under the action of a certain stabilizer subgroup, recovering part of [PS2].
Under the isomorphism of PGSp(2) with SO(3, 2), the image of H0 is the split group SO(2, 2), and that of H is SO(3, 1; E/F ), the special orthogonal group associated with the sum of the hyperbolic form xy and the norm form z2 − θt2. Our techniques apply also with G = PGSp(1,D) ' SO(4, 1), where D is a quaternion algebra, and H
6 is again SO(3, 1; E/F ), if the field E embeds in D. It would be interesting to study this situation, and its relation to the present work. Further, here we consider cyclic cusp forms with nonzero Fourier coefficients with respect to a generic character of the unipotent radical of the Siegel parabolic subgroup. There are no generic (with respect to a maximal unipotent subgroup) cyclic cusp forms on GSp(2). It would be interesting to consider also cyclic cusp forms generic with respect to the Heisenberg unipotent subgroup. Naturally it would be interesting to consider cusp forms on
SO(n) cyclic with respect to SO(n − 1), for a suitable choice of inner forms of these groups. The present work is a first step in this direction.
In Chapter 3 we consider a variation of the relative trace formula initiated in [F5], in the case where G = GL(n) and S = GL(n − 1) × GL(1). In this case there are no cuspidal cyclic representations, and we are led to introducing periods of induced representations by means of regularization of analogous integrals. The definition of the regularized integrals is given in section 5, where their main properties are described. We derive an analogue of Arthur’s spectral expansion, which combined with the geometric side developed in [F5] constitutes our trace formula.
h 0 Embed S in G via (h, a) 7→ ( 0 a ) for h ∈ GL(n − 1) and a ∈ GL(1). Denote by N 1 p z the group of matrices of the form n = 0 I tq , where z is a scalar, I the identity 0 0 1 (n − 2) × (n − 2) matrix and p and q are row vectors. Here tq denotes the transpose of q. Let ψe be a complex valued nontrivial additive character of A trivial on F .
Denote by ψ the character of N(A) trivial on N(F ) defined by ψ(n) = ψe(p1 + qn−2) for n ∈ N(A).
7 ∞ 1 For a test function f ∈ Cc (G(A) ) let
X −1 Kf (x, y) = f(x γy) γ∈G(F ) be the kernel of the right regular representation of G(A)1 on L2(G(F )\G(A)1). Our main goal is the computation of the fine spectral expansion of the distribution
Z Z f 7→ J(f) = Kf (n, h)ψ(n)dndh. 1 N(F )\N(A) S(F )\S(A)∩G(A) This is recorded in Theorem 59. To derive our formula we use a modification of
Arthur’s truncation, as in [JLR]. We truncate the kernel Kf (n, h) with respect to the second variable, noting that for a sufficiently positive truncation parameter the truncated kernel is equal to the original kernel. As in [JLR] we compute, in sections 5 and 6, the period integrals of the truncated Eisenstein series. Using Arthur’s (G, M)- families we compute in section 7 the fine expansion of the distribution J(f).
In more detail, the constant term φP of an automorphic form φ has a decomposition
l X P φP (namk) = αi(HP (a))ψi (amk). i=1
P Here αi are polynomials and ψi are level P automorphic forms satisfying
P hλi+ρP ,HP (a)i P ψi (ag) = e ψi (g),
∗ for some λi ∈ aP . R ∗ The regularized integral 1 φ(h)dh is defined to be S(F )\S(A)∩G(A) ZZ Z # X X T,P P hλi,Xi (Λ (ψi ))(mk)dkdm αi(X)e τb(X − T )dX. P ⊂G i
8 The #-integral is defined through analytic continuation in [JLR]. Thus in section 5 we compute the regularized integrals of cuspidal Eisenstein series. Namely
Z ∗ E(h, φ, λ)dh = J(ω, φ, λ) 1 S(F )\S(A)∩G(A) where ! X Z Z J(ω, φ, λ) = ehλ+ρP ,HP (ηh)i φ(mηh)dm dh. 1 η Sη(A)\S(A) Mη(F )\Mη(A)∩G(A)
In section 6 we compute the (convergent) integrals of truncated Eisenstein series in terms of the regularized integrals of their constant terms. In section 7 we use the combinatorics of Arthur’s (G, M)-families to compute the fine spectral expansion of the distribution J(f). As noted above, the equality of the spectral expression of J(f) and the geometric expression of J(f), obtained in [F5], constitutes our trace formula.
For applications, in section 8 we change our setting from that of S(A) ∩ G(A)1 inside
G(A)1 (where G = GL(n) and S = GL(n − 1) × GL(1)), which is used by Arthur, to that of S = GL(n − 1) in G = PGL(n), which is more suitable for our applications.
Thus using this rephrased absolutely convergent expression and the results of [F5] and [F8] we describe the set of automorphic representations of PGL(n, A) which are S-cyclic. More precisely given a cuspidal representation π0 of PGL(2, A), let
0 π = I(n−2,2)(1n−2 × π ) be the induced representation of PGL(n, A). Then π is S- cyclic and every S-cyclic automorphic representation of PGL(n, A) is of this form.
9 CHAPTER 2
CUSP FORMS ON GSP(2) WITH SO(4)-PERIODS
2.1 Definitions and Notations
Let F be a number field and let G be the group GSp(2), the algebraic group defined over F by
t 0 w 0 w G = g ∈ GL(4) | g ( −w 0 ) g = λ(g)( −w 0 ) , λ(g) ∈ GL(1) ,
t 0 1 × ×2 where g denotes the transpose of g and w is ( 1 0 ). Fix θ ∈ F − F and put Θ = θ1 0 , where θ = ( 0 1 ) . Let H = Z (Θ) be the centralizer of Θ in G. 0 θ1 1 θ 0 G Let B be the upper triangular Borel subgroup. It has a Levi decomposition B =
MBNB. Here MB is the diagonal subgroup and NB the unipotent upper triangular subgroup. With this choice of a minimal parabolic subgroup the standard parabolic subgroups are the Borel subgroup B, the Siegel parabolic subgroup P1 = M1N1 and
IX x y the Heisenberg parabolic subgroup P2 = M2N2. Here N1 = {n1 = ( 0 I ); X = ( z x )} , A 0 λ M1 = 0 λwtA−1w where A ∈ GL(2), and M2 = diag(a, A, a ); detA = λ , N2 = n u(x) yu( z ) o y ; u(x) = ( 1 x ) . 0 u(−x) 0 1 Fix a nontrivial additive character ψ0 of A/F . Let ε be diag(1, −1). Define the 0 character ψ of N1(A)/N1(F ) by ψ(n) = ψ tr(εθ1X) .
10 Recall the following definitions (for a general connected reductive group G). Fix a maximal compact subgroup K of G(A) satisfying (see [MW], I.1.4)
1. G(A) = B(A)K;
2. K∩M(A) is a maximal compact group of M(A) for all Levi subgroups M ⊂ G;
3. P (A) ∩ K = (MP (A) ∩ K)(NP (A) ∩ K).
Let P be a standard parabolic subgroup of G. Let MP be its unique Levi subgroup containing MB and NP its unipotent radical. Let TP be the maximal split torus in the center of MP . Let X∗(TP ) = Hom(Gm,TP ) be the group of co-characters of TP .
∗ Let X (TP ) = Hom(TP , Gm) be the group of characters. Set
eaP = X∗(TP ) ⊗ R and d(P ) = dim(eaP ).
For any two parabolic subgroups Q ⊂ P , there are canonical injection eaP ,→ eaQ and surjection eaQ eaP , hence a canonical decomposition
P eaQ = aQ ⊕ eaP .
Since eaG is a summand of eaP for any P , set aP = eaP /eaG . Then we have the decomposition
P aQ = aQ ⊕ aP .
∗ Write aP = HomR(aP , R) for the dual space (over R) of aP . Let g be the Lie algebra of G and Ad : G → Aut(g) the adjoint representation of G on its Lie algebra g. Then Ad(TB) is a group of diagonalizable commuting
11 (T ) endomorphisms g → g. Hence they are simultaneously diagonalizable. Set gα =
∗ (T ) {X ∈ g|Ad(t)(X) = α(t)X} for α ∈ X (TB). If the space gα is not zero, α is called a weight. Hence we have
(T ) (T ) g = g0 ⊕ ⊕α∈Φgα .
Only finitely many nonzero α appear. These α are called roots of G with respect to
TB. The finite set of roots is denoted by Φ.
∗ There is a natural pairing h, i : X (TB) × X∗(TB) → Z, defined as follows: For
∗ α ∈ X (TB) and β ∈ X∗(TB) the composition α ◦ β : GL(1) → GL(1) gives a character of GL(1). Since X∗(GL(1)) ' Z there is an n ∈ Z such that α ◦ β(t) = tn. Define hα, βi = n.
∨ For each root α ∈ Φ the coroot α ∈ X∗(TB) is the unique homomorphism GL(1) →
∨ TB satisfying hα, α i = 2. A choice of a minimal parabolic subgroup determines an ordering of the set of roots
Φ. A root α is called simple if α > 0 and it cannot be written as α = β1 + β2 with βi both positive.
∗ Let ∆ and ∆b be the subsets of simple roots and simple weights in aG, respectively.
∨ ∨ We write ∆ for the basis of aG dual to ∆,b and ∆b for the basis of aG dual to ∆.
Thus, ∆∨ is the set of coroots and ∆b ∨ is the set of coweights.
For every P , let ∆P ⊂ aG be the set of nontrivial restrictions of elements of ∆ to
∨ ∨ aP . For each α ∈ ∆P , let α be the projection of β to aP , where β is the root in
∨ ∨ ∆ whose restriction to aP is α. Set ∆P = {α : α ∈ ∆P }. Then we may also define
∨ their dual bases. Namely, we denote the dual basis of ∆P by ∆b P and the dual basis
∨ of ∆P by ∆b P .
12 P If Q ⊂ P , we write ∆Q to denote the subset of α ∈ ∆Q appearing in the action of
TQ on the unipotent radical of Q ∩ MP . Then aP is the subspace of aQ annihilated
P P ∨ ∨ P ∨ P P by ∆Q. Let (∆Q) = {α : α ∈ ∆Q}. We then define (∆b )Q and ∆b Q to be the basis
P P ∨ dual to ∆Q and (∆Q) , respectively.
P P Given two parabolic subgroups Q ⊂ P denote by χQ and χbQ respectively the charac- teristic functions of the sets
P {H ∈ aB : hα, Hi > 0 for all α ∈ ∆Q} and
P {H ∈ aB : h$, Hi > 0 for all $ ∈ ∆b Q}.
For each parabolic subgroup P there is the Harish-Chandra map
HP : G(A) → aP characterized by:
hχ,HP (m)i ∗ 1. |χ|(m) = e for all m ∈ M(A) and χ ∈ X (MP ),
2. HP (nmk) = HP (m), n ∈ N(A), m ∈ M(A), k ∈ K.
1 For a group G, denote by G(A) the kernel of the Harish-Chandra map HP : G(A) → aP .
For a parabolic subgroup with Levi decomposition P = MN, denote by ρP the unique
∗ 2hρP ,H(m)i element in aP satisfying e = | AdN (m)|. Here AdN (m) is the adjoint action of m on the Lie algebra of N.
13 × r Let F∞ denote F ⊗Q R. There is an isomorphism (F∞) ' TP (F∞) (see [MW], I.1.11 and I.1.13 also [JLR], III.4). Let AP for P 6= G and AG be the intersections of the
× r 1 image of (R+) in TP (F∞) with G(A) and Z(A), respectively. The Harish-Chandra map HP induces an isomorphism of AP onto aP and we have the decomposition
1 X M(A) = AG × AP × M(A) . For X ∈ aP , write e for the unique element in AP such
X ∗ λ hλ,HP (p)i that HP (e ) = X, and for λ ∈ aP , write e for the character p 7→ e of P (A). Then for a suitable normalization of the Haar measure, we have
Z Z Z Z Z f(x)dx = f(namk)e−2hρP ,HP (a)idndadmdk. 1 1 G(A) N(A) AP M(A) K
Let Ω = NG(MB)/MB be the Weyl group of MB in G, where NG(MB) denotes the normalizer of MB in G. For a Levi subgroup M, let ΩM = NM (MB)/MB be the Weyl group of M. For two parabolic subgroups P and Q with Levi factors MP and MQ respectively, let Ω(P,Q) be the set of elements ω ∈ Ω of minimal length in their class
−1 ωΩMP , such that ωMP ω = MQ. The minimal length condition is equivalent to the
P Q condition ω∆G = ∆G.
Let K be a maximal compact subgroup of G(A). Let z be the center of the universal Q enveloping algebra of the complexified Lie algebra of G∞ = v G(Fv). Here the product is over the archimedean places v of F . Then a function φ(g) on G(A) is called K-finite if the span of the functions g 7→ φ(gk)(∀k ∈ K) is finite dimensional.
It is called z-finite if there is an ideal I in z of finite codimension such that I · φ = 0.
The function φ(g) is called smooth if the following condition is satisfied. For g = g∞gf , with g∞ ∈ G∞ and gf ∈ G(Af ), there exist a neighborhood V∞ of g∞ in G∞ and
14 ∞ a neighborhood Vf of gf in G(Af ), and a C -function φ∞ : V∞ → C, such that
φ(g∞gf ) = φ∞(g∞) for all g∞ ∈ V∞ and gf ∈ Vf . In the function field case φ is called smooth if it is locally constant.
Fix an embedding i : G,→ SL2n. For g ∈ G(A), write i(g) = (gkl)k,l=1,...,2n. Set Y ||g|| = sup{|gkl|v; k, l = 1,..., 2n} v where the product is over all places v of F . Then a function φ is called of moderate growth if there are C,N ∈ R such that for all g ∈ G(A) we have |φ(g)| ≤ C||g||N . The definition of moderate growth does not depend on the choice of the embedding i. As in [MW], I.2.17 we make:
Definition 1. A function
φ : G(F )\G(A) → C is called an automorphic form if
1. φ is smooth and of moderate growth,
2. φ is right K-finite,
3. φ is Z-finite.
The space of automorphic forms is denoted by A(G). For a parabolic subgroup
P = NM the space of automorphic forms of level P , denoted by AP (G), is the space of smooth right K-finite functions
(P ) φ : N(A)M(F )\G(A) → C
15 such that for every k ∈ K the function m 7→ φ(P )(mk) is an automorphic form of
(P ) M(A). For φ ∈ AP (G) and a parabolic subgroup Q ⊂ P with unipotent radical
NQ, the constant term is defined by: Z (P ) (P ) φQ (g) = φ (ng)dn. NQ(F )\NQ(A)
For a cuspidal representation σ of M(A) denote by AP (G)σ the subspace of AP (G) consisting of functions φ which are AP -invariant and for every k ∈ K the function m 7→ φ(P )(mk) belongs to the space of σ.
We define the truncation operator by
Definition 2. Let φ : G(F )\G(A) → C be a smooth function. For h ∈ H(A) define
T X d(P )−d(G) X (Λ φ)(h) = (−1) φP (δh)χbP (HP (δh) − T ). P ⊂G δ∈P (F )∩H(F )\H(F ) Here the first sum ranges over all standard parabolic subgroups P of G.
Similarly for a smooth function φ : P (F )\G(A) → C, for h ∈ H(A) define
T,P X d(R)−d(P ) X P (Λ φ)(h) = (−1) φR(δh)χbR(HR(δh) − T ). R⊂P δ∈R(F )∩H(F )\P (F )∩H(F ) Here the first sum ranges over all standard parabolic subgroups R of G contained in
P .
Proposition 1. For all h ∈ H(A) we have
X X T,P φ(h) = (Λ φ)(δh)χP (HP (δh) − T ). P ⊂G δ∈P (F )∩H(F )\H(F ) Proof. This proof is exactly the same as that of Lemma 1.5 in [A3]; see also [JLR],
III.5.
16 Proposition 2. For T sufficiently regular and fixed h ∈ H(A) the function m 7→
(ΛT,P φ)(mh) is rapidly decreasing on (M(F ) ∩ H(F ))\(M(A) ∩ H(A)1).
Proof. Same as Proposition 8 in [JLR]. The proof of [A3], page 92, needs a small modification for the “relative case” as in [JLR].
Definition of R #(following [JLR], pp. 178-185). Let V be a real vector space of finite dimension n. Let V ∗ be the space of complex valued linear forms on V and
V ∗ the space of real valued linear forms. An exponential polynomial function on V , R expoly for short, is a function f = f({λi,Pi}) of the form
X hλi,xi f(x) = e Pi(x), 1≤i≤r ∗ where λi are different elements of V and Pi(x) are nonzero polynomials. By a cone in V we mean a closed subset of the form
C = {x ∈ V : hµi, xi ≥ 0}, where {µ } is a basis of V ∗. Let {e } be a basis of V dual to {µ }. i R j i Consider the integral Z −hλ,xi IC (λ; f) := f(x)e dx. C
For the expoly f = f({λi,Pi}) the integral converges absolutely for λ in the open set
∗ U = {λ ∈ V : Rehλi − λ, eji < 0 for all 1 ≤ i ≤ r, 1 ≤ j ≤ n}.
Fix an index k. Let Ck be the intersection of C and the hyperplane Vk = {x ∈ V : hµk, xi = 0}. In the integral we can write x = hµk, xiek + y with y ∈ Vk and then m X X 1 Z ∂ I (λ; f) = ehλ−λi,yi P (y)dy. C hλ − λ , e im+1 ∂y i 1≤i≤r m≥0 i k Ck k
17 This formula gives the analytic continuation of IC (λ; f) to the tube domain
∗ Uk = {λ ∈ V : Rehλi − λ, eki < 0 for all 1 ≤ i ≤ r}.
The analytic continuation is a meromorphic function with hyperplane singularities, the singular hyperplanes being:
∗ Hk,i = {λ ∈ V : hλ, eki = hλi, eki for all 1 ≤ i ≤ r}.
∗ Therefore the function IC (λ; f) has analytic continuation to V by Hartogs’ lemma (see [JLR], pp. 180). Lemma 3 of [JLR] asserts:
Lemma 3. The function IC (λ; f) is holomorphic at λ = 0 if and only if hλi, eki 6= 0 for every pair (i, k).
∗ Denote the characteristic function of C by χC . Let T ∈ V . For λ ∈ V such that
Rehλi − λ, eji < 0 for all i and j set Z −hλ,xi fb(λ; C,T ) = f(x)χC (x − T )e dx V
Z X −hλ−λi,xi = Pi(x)χC (x − T )e dx 1≤i≤r V
X hλ−λi,T i hλi,xi = e IC (λ; Pi(x + T )e ). 1≤i≤r
The integrals are absolutely convergent and fb(λ; C,T ) extends to a meromorphic function on V ∗.
18 Definition 3. The function f(x)χC (x−T ) is called #integrable if fb(λ; C,T ) is regular at λ = 0. In this case, we define the #integral
Z # f(x)χC (x − T )dx (2.1.1) V to be the value fb(0; C,T ).
From [MW], I.3.2, any automorphic form φ ∈ AP (G) has the decomposition
l (P ) X hλi+ρP ,HP (a)i φ (namk) = αi(HP (a))ψi(mk)e . (2.1.2) i=1
1 Here n ∈ N(A), a ∈ AP , m ∈ M(F )\M(A) and k ∈ K. The αi(X) are polynomials,
∗ ψi ∈ AP (G) and λi ∈ aP . The decomposition is not unique. The set {λi} is uniquely determined by φ.
Definition 4. For an automorphic form φ ∈ A(G) define
Z # T,P Λ (φP ) (h)χP (HP (h) − T )dh 1 b P (F )∩H(F )\H(A) to be
Z # Z Z T,P X −hρP ,2Xi Λ (φP ) (e mk)dmdk e · χP (2X − T )dX, 1 b aP K M(F )\M(A) where the integral R # is as defined in (2.1.1). Write ψK(m) for R ψ(mk)dk. By aP K (2.1.2) this integral is equal to
l Z Z # X T,P K hλi,2Xi Λ (ψi ) (m)dm αi(X) · e · χP (2X − T )dX. 1 b i=1 M(F )\M(A) aP
19 R ∗ Definition 5. For an automorphic form φ ∈ A(G) define 1 φ(h)dh, also H(F )\H(A) denoted by ΠH (φ) and called the period of φ over H, to be
Z # X T,P (Λ (φP ))(h)χP (HP (h) − T )dh. 1 b P ⊂G P (F )∩H(F )\H(A)
(P ) For a level P automorphic form φ ∈ AP (G) with decomposition (2.1.2) define
Z ∗ (P ) φ (h)χP (HP (h) − T )dh 1 b P (F )∩H(F )\H(A) to be
l Z # X K hλi,2Xi ΠM∩H (ψi (m)) αi(X)e χbP (2X − T )dX. i=1 aP
R T Proposition 4. For φ ∈ A(G) we have that 1 (Λ φ)(h)dh is equal to H(F )\H(A) Z ∗ X d(P )−d(G) (−1) φP (h)χP (HP (h) − T )dh. 1 b P ⊂G P (F )∩H(F )\H(A) Proof. Same as Theorem 10 in [JLR].
2.2 The group
As in section 2.1, let G be the group GSp(2), the algebraic group defined over F by G = GSp(2) = g ∈ GL(4) | tgJg = λ(g)J, λ(g) ∈ GL(1) ,
0 w 0 1 where J = ( −w 0 ) and w = ( 1 0 ). It is the group of symplectic similitudes of a 4-dimensional space over F , a local or global field of characteristic other than two. Here tg denotes the transpose of g. The form J has the advantage that the upper triangular subgroup B of G is a minimal parabolic. The maximal parabolics
20 IX x y which contain B are the Siegel parabolic P = MN,N = {( 0 I ); X = ( z x )}, M =
A 0 A 0 1 0 0 λwtA−1w = ( 0 λεAε ) ), where ε = ( 0 −1 ); and the Heisenberg parabolic Q = a 0 0 1 x y z n o 0 1 0 y MQNQ,MQ = 0 A 0 ; det A = λ , with unipotent radical NQ = 0 0 λ/a 0 0 1 −x 0 0 0 1 1 0 0 z 0 1 0 0 which is an Heisenberg group with center ZQ = 0 0 1 0 . The centralizer H = 0 0 0 1 ε 0 ZG(Θ) of Θ = {( 0 −ε )} in G is
a 0 0 b 0 α β 0 a b α β 0 γ δ 0 ; det ( c d ) = det γ δ 6= 0 . c 0 0 d
Given θ ∈ F × put θ = ( 0 1 ) and Θ = θ1 0 . The centralizer H = Z (Θ) of 1 θ 0 0 θ1 G Θ in G consists of the matrices h = ( a b ); a = ( a1 a2 ) , ··· , d = d1 d2 , such c d θa2 a1 θd2 d1 √ √ a b that ( c d ) , a = a1 + a2 θ, ··· , d = d1 + d2 θ, has determinant λ(h) in GL(1). If θ ∈ F − F 2 is not a square then H(F ) is isomorphic via the map just defined to √ {g ∈ GL(2,E); det g ∈ GL(1,F )}, where E = F ( θ) is the quadratic extension of F √ generated by the square root θ of θ. There is a natural injection G/H → X(θ), g 7→
1 t −1 x = λ gΘJ g = gΘg J, where
X(θ) = {x ∈ G; λ(x) = θ, (xJ)2 = θ} = {x ∈ G; λ(x) = θ, tx = −x}.
1 t In particular, G/H → X(1), g 7→ x = λ gΘJ g, is an injection. Denote by A the diagonal subgroup, and by NB the unipotent upper triangular subgroup (thus B =
ANB). Let W = Norm(A)/A be the Weyl group of A in G, where Norm(A) is the normalizer of A in G. Identify W with a set of representatives s in G with ts = s−1 if s is of order two in W . Double cosets decompositions as below are well known (see, e.g., [Sp]).
21 t Proposition 5. (a) Each x in X(θ) has the form x = nsa n with n ∈ NB, a ∈ A, s ∈ W with (s2 = 1 in W and) sa = −as−1. (b) The group G is the disjoint union
PH ∪ P γ1H = BH ∪ Bγ1H ∪ Bγ2H ∪ Bγ3H, where
1 0 2 II 1 0 0 1 −1 w 0 γ 0 γ1 = 1 ,I = ( 0 1 ) , γ = ( 1 1 ) , γ2 = ( 0 w ) , γ3 = 0 1 εγ0ε . − 2 II 2
(c) If θ is not a square in F , then the group G is the disjoint union BH ∪ Bγ0H, and
0 1 G is the disjoint union PH ∪ P γ0H, γ0 = diag(1, ( −1 0 ) , 1). (d) G acts on X(θ) by √ 1 t −1 −1 g : x 7→ λ(g) gx g = gxJ g J. The variety X(θ) consists of the points ± θJ, which are fixed by G, and the orbit
0 a b1 b −a 0 −b b2 ; b b + b2 − ad = θ of ΘJ. −b1 b 0 d 1 2 −b −b2 −d 0
The stabilizer of ΘJ is ZG(Θ) = H.
t Proof. (a) Each x in G can be expressed in the form x = n1sa n2 (a ∈ A, s ∈
t 2 W, ni ∈ NB). For x in X(θ) we have sa = −a · s (hence s = 1 in W ). We may
+ − + + assume that n2 = 1, and write n1 = n1 n1 . Here n1 lies in the group Ns which is generated by the root subgroups Nα of NB associated to the positive roots α such
− − that sα is positive; n1 lies in the group Ns which is generated by the Nα with
+ − t t −t + + α > 0 and sα < 0. Then n1 n1 sa = x = − x = sa n1 n1 , and so n1 = 1 and
− t t −1 −1 n1 ∈ Ns . The relation n1sa = sa n1 can be written as n1 = a s n1sa, or on
−1 −1t −1 −1 −1t −1 −1 applying transpose and inverse, as n1 = as n1 sa . Put σ(n) = as n sa .
− − This is an automorphism of Ns of order two. Since Ns is a unipotent group, the first
22 1 − − cohomology set H (hσi,Ns ) of the group hσi generated by σ, with coefficients in Ns ,
1 − is trivial. This H is the quotient of the set {n ∈ Ns ; σ(n)n = 1} by the equivalence
−1 − relation n ≡ n0nσ(n0) . In particular, our n1 ∈ Ns satisfies n1σ(n1) = 1, hence it
− −1 is in the equivalence class of 1, namely there is n ∈ Ns with n1 = nσ(n) . Hence
−1 −1t −1 t x = n1sa = nσ(n) sa = nas nsa sa = nsa n, as required. (b) The Weyl group W of G consists of 8 elements, represented by the reflections 1, (14)(23), (12)(34), (13)(24), (14), (23), (2431), (3421). The last two are not of order 2, while for s = 1 there are no a ∈ A with sas = −a. The transposition
0 1 (23) is represented in G by s = diag(1, ( −1 0 ) , 1), but there is no a ∈ A with sas = −a. An analogous statement holds for (14).
Concerning the remaining 3 Weyl group elements, we have the following. Choose