Rankin Triple Products and Quantum Chaos
Thomas C. Watson
September 2003
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals 2
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals Contents
1 Introduction 5
2 Automorphic Forms 7 2.1 OrdersandUnitsofQuaternionAlgebras ...... 7 2.2 Automorphic Forms on Quaternion Groups ...... 8 2.3 HeckeOperators ...... 11 2.3.1 DoubleCosets ...... 11 2.3.2 Non-ArchimedeanOperators ...... 13 2.3.3 ArchimedeanOperators ...... 16 2.4 AutomorphicRepresentations ...... 17 2.5 WhittakerModels ...... 21
3 Theta Lifting 27 3.1 BasicSetup ...... 27 3.1.1 Groups ...... 27 3.1.2 Measures ...... 28 3.1.3 Dual-Pairs...... 29 3.2 Jacquet–LanglandsCorrespondence...... 30 3.2.1 Shimizu’sThetaLift ...... 30 3.2.2 Adjoint of Shimizu’s Lift ...... 35 3.3 LocalData ...... 36
4 L-functions 39 4.1 LocalLanglandsCorrespondence ...... 39 4.1.1 AdjointParameters ...... 41 4.1.2 TripleProductParameters ...... 44 4.2 ZetaIntegrals...... 45 4.2.1 Rankin–Selberg...... 45 4.2.2 Garrett/Rallis–Piateski-Shapiro ...... 48
5 Conclusion 59 5.1 TripleProductIdentities...... 59 5.2 ApplicationstoQuantumChaos ...... 60 5.2.1 Proofs...... 63
3
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals 4 CONTENTS
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals Chapter 1
Introduction
In this paper, we demonstrate the chaotic nature of some archetypical quantum dynamical systems, using machinery from analytic number theory. Consider the quantized geodesic flow on a finite-volume hyperbolic surface Γ\H, with Γ ⊂ SL2R consisting of the norm-1 units of an Eichler order in an indefinite quaternion algebra B over Q. Such Γ generalize the congruence sub- groups of SL2Z and are co-compact whenever B is ramified. For Γ = SL2Z, we prove that high-energy bound eigenstates obey the Random Wave conjecture of Berry / Hejhal for third moments. In fact we show that the third moment of − 1 +ǫ a wave’s amplitude distribution decays like E 12 . In the more general case of maximal orders, we reduce an optimal quantitative version of the Quantum Unique Ergodicity conjecture of Rudnick–Sarnak to the Lindel¨of Hypothesis, itself a consequence of the Riemann Hypothesis, for particular families of auto- morphic L-functions. Furthermore, our analysis shows that any lowering of the exponent in the Phragmen–Lindel¨of convexity bound implies QUE. In the mo- ment problem as well, a decisive role is played by ‘convexity-breaking.’ That is to say, the maximum non-trivial exponents precisely agree when translated be- tween physical and arithmetical formulations for both of these problems. (Those readers primarily interested in the physical side of this correspondence are ad- vised to skip ahead to the final chapter.) We accomplish this translation by proving identities expressing triple-cor- relation integrals of eigenforms in terms of central values of the corresponding Rankin triple-product L-functions. Very general forms of such identities were proved by Harris–Kudla, and certain (more precise) explicit classical versions of these were given by Gross–Kudla and B¨ocherer–Schulze-Pillot, for definite quaternion algebras. In using the Harris–Kudla method to prove our own clas- sical identities, we have to solve two main problems. The first problem is to explicitly compute the adjoint of Shimizu’s theta lift, which realizes the Jacquet–Langlands correspondence by transferring automor- × 2 phic forms from GL2 to GO(B), the latter being nearly the same as (B ) . As is well known, theta liftings from metaplectic to orthogonal groups are gen- erally more difficult to characterize than lifts in the opposite direction, which
5
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals 6 CHAPTER 1. INTRODUCTION can be evaluated directly in terms of Whittaker functions. Since B× and hence GO(B) have ‘multiplicity-one’—as Jacquet–Langlands proved with the trace- formula—we are able to determine the adjoint of Shimizu’s lift by duality, from explicit knowledge of Shimizu’s lift itself. It is, however, necessary to gener- alize Shimizu’s original calculations, since he only considered averages of lifts over isotypic bases of forms, which allowed him to employ Godement’s theory of spherical functions. In order to deal with individual ramified forms, we re- place this argument by explicit calculations involving Hecke operators. Thus we determine the Shimizu lifts of oldforms and newforms of square-free level, with any (possibly imprimitive) nebentypus. As a byproduct of these calculations, we obtain explicit formulas for all rele- vant GL2 Whittaker functions. These play an important role in our second main problem: evaluation of Garrett / Rallis–Piateski-Shapiro local zeta integrals in terms of the standard functorial triple-product L-factors. Our contribution lies in the calculation of archimedean zeta integrals with various types of ramifica- tion. In all of these ramified cases, the canonical choices of local data for theta lifting are apparently useful only for evaluating central values of the associ- ated zeta integrals. This was observed by Gross–Kudla in the non-archimedean case, and it appears to be a significant unexplained feature of the Harris–Kudla method. Finally, we re-prove (by elementary brute-force computation) the im- portant result of Rallis–Piateski-Shapiro on unramified non-archimedean zeta integrals, in anticipation of future generalizations to the many ramified non- archimedean cases. Acknowledgements. This work is the product of my doctoral research at Princeton University, directed by Peter Sarnak. I am very grateful to him for explaining this fascinating problem to me, and for his generous help and encouragement, which cannot be overstated. I also thank Henryk Iwaniec and Goro Shimura for teaching me analytic and algebraic number theory, as well as Michael Boshernitzan and Yakov Sinai for teaching me ergodic theory. I gratefully acknowledge receiving funding through an NSF Graduate Fellowship and a Charlotte E. Proctor Honorific Fellowship. Finally, I thank my friends and family for their love and support.
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals Chapter 2
Automorphic Forms
2.1 Orders and Units of Quaternion Algebras
Let B be a quaternion algebra over Q. Then Bv = B ⊗Q Qv is a quaternion algebra over Qv, for each place v of Q. B is called ramified at v if Bv is a division algebra. We will always assume that B is unramified at ∞, i.e. B is indefinite. Then the reduced discriminant dB of B over Q equals the product of the finite even number of finite places p where B is ramified. The 2 × 2 matrix algebra M = M2 over Q is totally unramified, and hence dM = 1. We denote by Dv the unique division quaternion algebra over Qv, while Mv is the unique split (i.e. non-division) quaternion algebra. We will sometimes use the coordinate α α functions α on α = 11 12 ∈ M , and also the notation ij α α v 21 22
[Xij ] = {α ∈ Mv ; αij ∈ Xij ⊂ Qv}.
ι Write the canonical anti-involution ι of B or Bv as α 7→ α , noting its compatibility with field extensions, and define the reduced trace and norm by ι ι τ(α)= α + α and ν(α)= αα . On M and Mv, ι is given explicitly as ι a b d −b = , c d −c a so τ and ν correspond to the usual trace and determinant. Dp has the valuation ordp ◦ ν, and its valuation ring Rp is also the unique maximal order of Dp. Choose a uniformizer ̟p ; all bilateral Rp ideals in Dp n n are principal and of the form ̟p Rp = Rp̟p , n ∈ Z. Every maximal order of Mp is conjugate to M2(Zp). We will be interested −n n −1 n Zp p Zp p p in those of the form n = M2(Zp) for p Zp Zp 1 1 n ∈ Z, although the set of all maximal orders is larger, having the structure of a (p + 1)-homogeneous tree for a natural notion of adjacency. It is a theorem of Hijikata that every Eichler order, an intersection of two maximal orders, is
7
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals 8 CHAPTER 2. AUTOMORPHIC FORMS
Z Z conjugate to p p for some n ≥ 0. All Eichler orders containing the pnZ Z p p Z p−n˙ Z previous one are of the form p p for 0 ≤ n˙ ≤ n¨ ≤ n. pn¨ Z Z p p Orders O of B exist and have the property that Op := O⊗Z Zp is a maximal ′ ′ order in Bp for almost all p. Furthermore, for any other order O of B, Op = Op for almost all p. Conversely, given a choice over all p of local orders (O˜p) which ′ satisfies O˜p = Op for a.a. p, then O := {α ∈ B ; αp ∈ O˜p ∀p} is an order in B ′ ˜ with p-adic completions Op = Op. The adele ring BA is defined as the restricted direct product of all Bv relative × to Ofin := p Op, for any order O of B, and the idele group BA as the restricted direct product of all B× relative to O× . B is contained diagonally in B , and Q v fin A we will write B = BQ ⊂ BA to emphasize this inculsion. By the previous paragraph, orders of B correspond 1-1 with compact open subgroups of Bfin. Maximal orders in B are characterized as orders O s.t. every Op is maximal. Choose a maximal order for each B, denoted O(dB ). Throughout the rest of the paper, for p ∤ dB we will identify Bp with Mp in such a way that Op(dB)= M2(Zp). Furthermore, in the case dB = 1 we identify B with M: O(1) = M2(Z). + Now for N,˙ N¨ ∈ Q relatively prime to dB and s.t. N = N˙ N¨ ∈ Z, we define the order O(dB , N~ ) of B by
Rp p | dB,
Op(dB , N~ ) = Z N˙ Z p p p ∤ d . ¨ B " NZp Zp #
−1 Then O(dB , N,˙ N˙ ) is a maximal order, and
−1 −1 O(dB , N~ ) = O(dB , N,˙ N˙ ) ∩ O(dB , N¨ , N¨) is an Eichler order. It has reduced discriminant dB N and is called of level N. Using strong approximation and Hijikata’s result, it is easy to show that any Eichler order of level N in B is conjugate to O(dB , (1,N)). Note that the only ′ ′ ′ Eichler orders containing O(dB , N~ ) are the O(dB , N~ ) s.t. N˙ | N˙ and N¨ | N¨. Throughout the rest of the paper, for any X possessing a character ν, we will use the notations
(a) X∗ = {β ∈ X∗ ; ν(β)= a}, (A) X∗ = {β ∈ X∗ ; ν(β) ∈ A},
± × ± (R ) [a] (aZp ) X∞ = X∞ , Xp = Xp .
2.2 Automorphic Forms on Quaternion Groups
~ × In this section, fix O = O(dB , N). Strong approximation for Ofin is proved in × × + × × × [26]: BA = BQ B∞Ofin since B is indefinite and ν(Ofin)= Zfin. Because of this,
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals 2.2. AUTOMORPHIC FORMS ON QUATERNION GROUPS 9
(1) × + × and since O = BQ ∩ B∞Ofin, there are homeomorphisms
× × × (1) + BQ \BA /Ofin ≃ O \B∞, × + × × (1) + BQ R \BA /Ofin ≃ O \PGL2 (R). χ Given a primitive Dirichlet character χ mod N , factor χ = p χp and extend its domain to Z× by means of the Chinese remainder theorem, fin Q χ × χ × (Z/N Z) ≃ p|N χ (Zp/N Zp) × χ = Qp|N χ Zp /(1 + N Zp). × Q× + × Using strong approximation, A = Q R Zfin , we define the grossen-character × χ χ˜ byχ ˜(qrz)= χ(z). Note thatχ ˜∞(−1) = χ(−1) andχ ˜p(pZp )= χ(p) if p ∤ N . Since O has level N, the map rp : Op → Zp/NZp defined by rp(β) = [β11] for p ∤ dB , is a ring homomorphism, and hence so is r : Ofin → Z/NZ, r(β) = χ × × (rp(βp))p|N . This permits us, for N | N, to defineχ ˜ on Ofin and χ on O by composition with r. (Note χ(β)= χ˜(β) for β ∈ O×.) Sinceχ ˜ is well-defined on × × × × × Qfin ∩ Ofin = Zfin, it extends to a character of QfinOfin. × As usual, B∞ = GL2(R) acts on CrR by linear fractional transformations, az + b a b z 7→ β|z = , β = , cz + d c d 2 2 2 2 + preserving the metric ds = (dx + dy )/y . Furthermore, B∞ acts on weight-k + k Maass forms (k ∈ Z) on H = R + ı R by ψ 7→ ψ|β,
k cz+d −k ψ|β(z) = |cz+d| ψ(β|z). Consider the space Ak(dB , N,χ~ ) of smooth weight-k Maass forms ψ on H which have moderate growth and satisfy the automorphy condition k (1) ~ ψ|γ∞ = χ(γ) ψ, γ ∈ O (dB , N). Since −1 ∈ O(1), such ψ must vanish identically unless χ(−1) = (−1)k, so we assume this from now on. Let Ck(dB , N,χ~ ) ⊂ Ak(dB, N,χ~ ) denote the cuspidal subspace, equipped with the Petersson inner-product
dx dy hψ1, ψ2i := ψ1(z) ψ2(z) y2 . (1) ZO \H 2 × × In the adelic setting, we define L (BQ \BA , χ˜) as usual to consist of auto- morphic functions Ψ with central characterχ ˜,
× × Ψ(γzβ) =χ ˜(z) Ψ(β), γ ∈ BQ , z ∈ A , and finite norm under the inner-product
1 × hΨ1, Ψ2i := 2 Ψ1(β) Ψ2(β) d β. P B×\P B× Z Q A
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals 10 CHAPTER 2. AUTOMORPHIC FORMS
The Tamagawa measure d×β is defined in §3.1.2. Note these two inner-products × are normalized differently. We denote the right regular action of α ∈ BA by
2 × × α|Ψ(β) = Ψ(βα), Ψ ∈ L (BQ \BA , χ˜),
2 × × and define L0(BQ \BA , χ˜) as the closed invariant subspace of cuspidal Ψ. Now let × ~ Op (dB , N) v = p, Kv(dB , N~ ) = (SO(2, R) v = ∞,
+ × and extendχ ˜∞ (dependent on k) to R K∞ from R by setting
cos θ sin θ χ˜ (rκ ) = eikθ, κ = . ∞ θ θ − sin θ cos θ ˜ ~ 2 × × We define Ck(dB , N, χ˜) to consist of all smooth Ψ ∈ L0(BQ \BA , χ˜) satisfying
κv|Ψ =χ ˜v(κv)Ψ, κv ∈ Kv(dB, N~ ).
× × + × (Note this is consistent with the inclusion of the center A ⊂ BQ R Ofin.) Then ∼ we have an isomorphism Ck(dB , N,χ~ ) −→ C˜k(dB , N,~ χ˜), given by
+ k Ψ(γβ κfin) =χ ˜(κfin) ψ| +(ı). ∞ β∞
′ χ ′ If O(dB , N~ ) ⊃ O(dB , N~ ) and N | N , there is an inclusion
′ .( ~Ck(dB , N~ ,χ) ֒→ Ck(dB , N,χ
Any form in the image of such a map with pN ′ | N is called p-old, as are linear combinations of such forms, and any form orthogonal to all p-old forms is called p-new. Likewise at the archimedean place, we have raising and lowering operators,
± ~ ~ Rk : Ck(dB, N,χ) −→ Ck±2(dB , N,χ), + z−z ∂ k − z−z ∂ k Rk = 2 ∂z + 2 , Rk = 2 ∂z + 2 .
± ′ A weight-k form ψ is called ∞-old if ψ = Rk∓2 ψ for ±k ≥ 0, and ∞-new if it is ± ∓ orthogonal to all ∞-old forms. Since Rk∓2 and Rk are adjoints, ψ is ∞-new iff ∓ |k| − 2 Rk ψ = 0 for ±k ≥ 0, and this is equivalent to y ψ(z) being a holomorphic / anti-holomorphic function of z. All forms with k =0 are ∞-old. For the rest of this paper we assume that N is square-free, and factoring it ♭ ♯ χ ♭ into N = N N N , we define Sk(dB , N,χ,N~ ) ⊂ Ck(dB , N,χ~ ) as the subspace of forms which are p-old for p | N ♭, p-new for p | N ♯, and ∞-new if |k| > 0. We will also assume that |k|= 6 1.
Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals 2.3. HECKE OPERATORS 11
2.3 Hecke Operators
χ In this section, fix O = O(dB , N~ ) and χ primitive with conductor N | N, and assume that N = N ♭N ♯N χ is square-free. For p | N ♭N ♯, set
˙ ˙ N¨ ¨ N˙ ¨ Op = Op(dB , (N, p )), Op = Op(dB, ( p , N)).
2.3.1 Double Cosets
× × × For α ∈ Bp , define the double coset Tp(α) = Op α Op . If p | dB , every Tp(α) n n × × n n is equal to some Tp(̟p ) = ̟p Rp = Rp ̟p for n ∈ Z, and Tp(̟ ) ⊂ Op iff × 2 n ≥ 0. Now for p ∤ dB and ~a ∈ (Qp ) , define
a˙ a˙ T (~a) = T , R (~a) = T . p p a¨ p p a¨ N˙ Since all a ∈ Q× and α = normalize O×, p N~ N¨ p × × ~ × × Tp(a) = aOp = Op a, Rp(N) = αN~ Op = Op αN~ .
~n If p ∤ dB N, every Tp(α) is equal to Tp(p ) for somen ˙ ≥ n¨ (set n =n ˙ − n¨), ~n and Tp(p ) ⊂ Op iffn ¨ ≥ 0. The parametersn, ˙ n¨ uniquely solve
n¨ n˙ +¨n × α ∈ p (Op r pOp), ν(α) ∈ p Zp .
Furthermore, these double cosets are partitioned by left cosets: (For right cosets, apply ι ; for lower triangular representatives, conjugate by Rp(N~ ).)
n pm+¨n Nxp˙ n¨ T (p~n) = O×. p pn−m+¨n p m=0 x mod pm [ p∤x if[ 0 We prove both well-known assertions of the previous paragraph at the same ~n time. Note that Tp(p ) contains the above union of left cosets, and they are disjoint since −1 ′ pm+¨n Nxp˙ n¨ pm +¨n Nx˙ ′pn¨ ′ pn−m+¨n pn−m +¨n m′−m ˙ ′ −m −m′ p N(x p − xp ) × = ′ ∈/ O pm−m p unless m = m′, x ≡ x′ mod pm. Now it suffices to show that the union of these left cosets over alln ˙ ≥ n¨ ≥ 0 has additive volume (see §3.1.2) equal to that of Feb 19 2009 7:55:20 PST Vers. 1 - Sub. to Annals 12 CHAPTER 2. AUTOMORPHIC FORMS × × × × 2 Op ∩ Bp , since for any α ∈ Op ∩ Bp , vol(αOp )/vol(Op )= |α|Bp = |ν(α)|p > 0. Finally, (pZ ) Z× N˙ Z× p pZ Np˙ Z O r O× = p p ∪ p p ∪ . . . p p N¨Z× Z× N¨Z× Z× p p p p pZ Np˙ Z pZ Np˙ Z ∪ p p ∪ . . . ∪ p p , so Np¨ Z Z× Np¨ Z pZ p p p p 3 2 1 × −1 p−1 −2 p−1 −3 p−1 −4 vol(Op ) = 1 − p p +4p p +4p p + p = 1 − p−1 1 − p−2 , and