Moments of automorphic L-functions at special points
Dissertation
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University
By
Alexander Lu Beckwith, B.A., M.S. Graduate Program in Mathematics
The Ohio State University 2020
Dissertation committee
Dr. Wenzhi Luo, Advisor Dr. James Cogdell Dr. Roman Holowinsky ⃝c Copyright by Alexander Lu Beckwith 2020 Abstract
We study the behavior of families of L-functions at exhibiting conductor-dropping behavior. We derive asymptotic expansions of the short interval first and second moments of GL(2) × GL(2) L-functions at special points with power-saving error terms. As a consequence, we obtain an essentially optimal lower bound on the number of cusp forms for the Hecke congruence surfaces Γ0(p) \ H of prime level are destroyed under quasiconformal deformation of these surfaces. Additionally, we show that a large number of cusp forms for the Hecke congruence surfaces Γ0(p) \ H of prime level are simultaneously destroyed in two directions of the associated Teichmüller space. We also establish upper bounds for the second moment of GL(2) × GL(3) L-functions and the sixth moment of GL(2) L-functions at special points as the spectral parameter varies in a short interval. An auxiliary twisted spectral large sieve inequality for short intervals is derived for the latter two results.
ii Dedicated to my grandparents
iii Acknowledgments
During my time as a student I have been very fortunate to have had the guidance and support of a number of wise and thoughtful people. There is a very long list of individuals whom I count myself lucky to have met and to whom I wish to extend my thanks in the following—I regret that I have surely neglected some from what appears below. First and foremost, to my advisor, Wenzhi Luo, for his thoughtful guidance, constant and enduring support, for his many kind words of encouragement, and for his boundless patience: Thank you, a thousand times, thank you. To Roman Holowinsky: for all his work helping graduate students navigate the nonacademic post- academic world, from which I have personally benefited so much. To Jim Cogdell: for organizing the automorphic forms seminar that I and many other students have benefited from greatly over the years. To Ghaith Hiary: for serving on my candidacy committee, and for teaching a topics course on computa- tional number theory that I found to be a formative experience. To my undergraduate professors, Judy Holdener, Bob Milnikel, and Marie Snipes: for encouraging me to study math a decade ago now. To my mentor and friend Jessen Book: for his wisdom and guidance when I needed a push in the right direction many years ago. And most of all, to my family, for all their love and support they have given me, no matter how much I stumbled.
Thank you.
iv Vita
1992 ...... Born in Midland, Michigan
2014 ...... B.A. in Mathematics, Concentration in Humane Studies, Kenyon College Gambier, Ohio
2017 ...... M.S. in Mathematics, The Ohio State University Columbus, Ohio
2014-2015 ...... University Fellow The Ohio State University Columbus, Ohio
2015-Present ...... Graduate Teaching Associate, Graduate Research Associate The Ohio State University Columbus, Ohio Fields of Study
Major field: Mathematics
Specialization: Analytic number theory, automorphic forms
v Contents
Abstract ...... ii Acknowledgments ...... iv Vita ...... v Notation and conventions ...... ix
1 Introduction 1
I Preliminaries 7
2 Preliminaries on automorphic forms 8 2.1 Automorphic forms for GL(2, R) ...... 9 2.2 Automorphic forms for GL(3, R) ...... 33
3 Preliminaries on automorphic L-functions 42 3.1 Automorphic L-functions ...... 42 3.2 On conductor-dropping behavior of L-functions in the spectral aspect ...... 48
4 Preliminaries on the spectral theory of hyperbolic Riemann surfaces and Teichmüller theory 51 4.1 Riemann surfaces ...... 51 4.2 Some remarks on Teichmüller theory and hyperbolic 2-surfaces ...... 52 4.3 The Phillips-Sarnak deformation theory of discrete groups ...... 55
II Upper bounds for moments of L-functions at special points 58
5 A short interval large sieve inequality with spectral twists 59 5.1 Background and survey of existing literature ...... 60 5.2 The short interval twisted spectral large sieve inequality ...... 67 5.3 Proof of Theorem 5.2.1 ...... 73 5.4 Proof of Lemmas 5.2.4 and 5.2.5 ...... 75
6 The second moment of GL(2) × GL(3) L-functions at special points over a shortened spectral interval 83 6.1 Background and survey of existing literature ...... 84 6.2 Preliminaries on GL(2) × GL(3) Rankin-Selberg L-functions ...... 86 6.3 The shortened spectral interval second moment of GL(2) × GL(3) L-functions at special points 88 6.4 The shortened spectral interval sixth moment of GL(2) L-functions at the special point ... 93
vi Contents III Asymptotic expansions of moments of L-functions at special points 97
7 The nonvanishing of GL(2) × GL(2) L-functions at special points over shortened spectral intervals 100 7.1 Properties of the Rankin-Selberg L-function ...... 102 7.2 An asymptotic expansion of the first moment of GL(2) × GL(2) L-functions at special points over shortened spectral intervals ...... 103
8 The simultaneous nonvanishing of GL(2) × GL(2) L-functions at special points 117 8.1 Proof of Theorem 8.0.1: an asymptotic expansion of the second moment of GL(2) × GL(2) L-functions at special points over shortened spectral intervals ...... 119
IV Future directions 138
9 Future directions 139 9.1 Direct analogues ...... 140 9.2 An asymptotic expansion of the fourth moment of GL(2) L-functions at special points .... 140 9.3 The first moment of GL(2) × GL(3) L-functions at special points ...... 141 9.4 The prime geodesic theorem ...... 142 9.5 Special point subconvexity problems ...... 144 9.6 Higher-order Fermi golden rules ...... 147 9.7 The newform Weyl law ...... 148
References 148
V Appendices 157
A Compendium of special functions and their properties 158 A.1 Bessel functions ...... 158
B Asymptotic behavoir of oscillatory integral 164
C Miscellaneous bounds for character sums 172 C.1 Compendium of properties of Kloosterman sums and Ramanujan sums and special character sums ...... 172
vii List of Figures
Chapter 2
2.1 Fundamental domain F0,1 for the modular group SL(2, Z) ...... 13 2.2 Some examples of fundamental domains for the Hecke congruence subgroups Γ0(N) ..... 14 2.3 Some examples of quotient surfaces for the Hecke congruence subgroups Γ0(N) ...... 15 2.4 Plot of cuspidal and residual spectrum of ∆ and poles of Eisenstein series for SL(2, Z) .... 22
viii List of Tables
Chapter 3
3.1 Some effects of conductor-dropping for L-functions in the spectral aspect ...... 49
Chapter 6
6.1 Status of upper bounds for families of L-functions at special points ...... 85
ix Notation and conventions
We standardize the common notation and conventions used throughout this document.
• N: the set of natural numbers 1, 2, 3,...
• H: the upper half of the complex plane, H = {x + iy : x, y ∈ R, y > 0}
• SL(2, R), SL(2, Z): the special linear group over R, Z, respectively
• GL(n, R): the general linear group of degree n over R
• Γ0(N): the Hecke congruence subgroup of level N
• Γa: the stability group of a cusp a of Γ
• νf (n): the normalized Fourier coefficients at ∞ of an automorphic form over GL(2, R)
• λf (n): the Hecke eigenvalues of an automorphic form f
• λf : the Laplace eigenvalue of an automorphic form f
• ΦΓ(s), φΓ(s): the scattering matrix for Γ and the determinant of ΦΓ(s)
• S(n, m; c): the Kloosterman sums for SL(2, Z), X∗ an + an S(n, m; c) = S∞∞(n, m; c) = e . c a(c)
Sab(m, n; c) denotes the Kloosterman sums of a discrete subgroup Γ ⊆ SL(2, R) attached to cusps a and b. As we will deal primarily with Kloosterman sums for the full modular group, we will not use this notation much.
• δ(condition): the delta symbol
• δ: the reflection operator, unless it appears in an exponent, in which case it is a positive constant.
• h·, ·i: the Petersson inner product
• k · k: the Petersson norm, unless looking at a sequence, in which case k · k denotes the (possibly truncated) ℓ2-norm
x List of Tables
• φe: the Mellin transform of a function φ (satisfying the appropriate conditions) defined by Z ∞ dx φe(s) = φ(x)xs 0 x
• φb: the Fourier transforms of a function φ (satisfying the appropriate conditions) defined by Z φb(ξ) = φ(x)e(−ξx)dx R
• τ2, τ3: the binary and ternary divisor functions defined by X X X τ2(n) = 1 and τ3(n) = 1 = τ2(n1).
n1n2=n n1n2n3=n n1n2=n
Note that these are distinct from the divisor functions τit given by X d it τ (n) = . it k dk=n
When the order of the divisor function is a natural number (i.e., τk with k ∈ N), we mean the k-ary th divisor function. We reserve σk for the k -power-of-divisors function.
− t−T 2 • `a˚u¯sfi¯sfi˚i`a‹nffl(t): the Gaussian function centered at T of width M, `a˚u¯sfi¯sfi˚i`a‹nffl(t) = e ( M ) GT,M GT,M • e(x): the additive character e(x) = e2πix
• hT,M (t): usually the (even) twisted Gaussian centered at T of width M
it −it `a˚u¯sfi¯sfi˚i`a‹nffl m `a˚u¯sfi¯sfi˚i`a‹nffl m hT,M (t, m, n) = (t) + (−t) GT,M n GT,M n
• ρ, ψ denote the weight function and phase function, respectively, in an oscillatory integral, except when possibly ψ denotes an SL(3, Z) Maass cusp form. This will hopefully be clear from context.
• Q, Q1, and Q2 denote holomorphic modular forms of some weight and level; we will always take these to be newforms and to have trivial nebentypus. P • ∗ a(c) : denotes a sum over congruence classes a (mod c) such that (a, c) = 1. When (a, c) = 1, the multiplicative inverse of a modulo c is denoted by a.
• T`eˇi`c‚hffl(R), T`eˇi`c‚hffl(S): the Teichmüller space of a Riemann surface R, or of the underlying surface S that marks a Riemann surface R.
• When not otherwise mentioned, sj will always refer to the special point, also known as the conductor-
dropping point or spectral point, for a Hecke-Maass cusp form uj for Γ ⊆ SL(2, R) a Fuchsian 1 group of the first kind. In this case sj = 2 +itj; for other sorts of automorphic forms, the corresponding special point may be defined differently.
xi List of Tables
Remark 0.0.1. In general, we observe a notational “reset” between chapters. This is done in order to avoid having too many symbols, especially when they may serve the same purpose for corresponding objects in each chapter. Throughout any ε appearing in an exponent will usually denote a small number that may actually change from line to line in a series of equations. When δ appears in an exponent, it is an absolute constant.
Remark 0.0.2. Σi, where i = 0, 1, 2,... , is shorthand for a spectral sum that appears as a result of applying an approximate functional equation to the L-functions in question. “Mathcal” script is used to denote objects that arise from application of the Kuznetsov trace formula. These will tend to be indexed according to whether they come from the diagonal term, the terms involving Kloosterman sums, etc.
Remark 0.0.3. A number of special functions appear throughout this document, primarily Bessel functions, but also others. A brief overview of the relevant theory of these appears in Appendix A, as well as a number of useful formulas and expansions. In particular, the various integral transforms appearing in the Kuznetsov trace formula are gathered in Appendix A; we will refer to these frequently.
In an effort to keep things as self-contained as possible, and to put the results presented here in greater context, throughout we will also include full statements of results attributed to others that we reference frequently. These theorems and propositions appear in blue-colored boxes. The original results in this document appear in red-colored boxes. Lemmas and other propositions appear in gray-colored boxes.
xii Chapter 1
Introduction
Let S be a compact smooth manifold of real dimension n—an “n-drum”—and let ∆S denote its associated
Laplace-Beltrami operator. The operator ∆S is a differential operator that acts on the space of square- integrable functions, L2(S), and it has the unique property that it is invariant under isometric diffeomor- phisms of S, which makes it an important invariant of the manifold S. It is formally self-adjoint, nonnegative, and it comes with a distinguished sequence of numbers—its spectrum or eigenvalues {λj}j≥1—that in the real world setting correspond to the resonant frequencies or tones that occur when one strikes S. Mark Kac famously once asked, “Can one hear the shape of a drum?” Put another way, what information about the geometry of S can be obtained from the λj, its resonant frequencies or spectrum? For example,
Weyl’s law for a general compact surface S expresses the volume of S in terms of the distribution of the λj: one has
p ω · (S) #{ λ ≤ T : λ is an eigenvalue of ∆ } = n ”vˆo˝l T n + O T n−1 , j j S (2π)n
n where ωn is the volume of the unit ball in R ; so one can hear the “volume” of the drum. In the time since Kac’s original question, a great deal of work has been done in studying the relationship between geometry and spectral theory. For particular classes of surfaces, one can often say a great deal more, and one can explore various refinements of Kac’s question. The latter is what we will do in this dissertation.
Let Γ ⊆ SL(2, R) be a Fuchsian group of the first kind that is not cocompact. Then the quotient surface Γ \ H is a noncompact hyperbolic Riemann surface of finite volume with finitely many punctures or cusps
{a}. In this setting, the study of the spectrum of the Laplacian ∆Γ\H is the study of classical GL(2, R) automorphic forms, and with certain weakenings of the invariance conditions, classical modular forms.
The noncompactness of Γ \ H is reflected in the spectral decomposition of ∆Γ\H: for these surfaces the spectrum of ∆Γ\H consists of both a discrete part and a continuous part. The continuous spectrum covers 1 ∞ the interval [ 4 , ) uniformly with multiplicity equal to the number of cusps, while the discrete eigenvalues
1 1.0. Introduction
1 2 λj can be written as λj = 4 + tj , and it is conjectured that all but finitely many of them are embedded in [ 1 , ∞). Accordingly, L2(Γ \ H) decomposes into two distinguished orthogonal subspaces: the span of the 4 { } 1 Maass cusp forms uj —the discrete spectrum—and the span of the Eisenstein series Ea z, 2 + it —the continous spectrum and residual spectrum1. The Eisenstein series have a simple definition in terms of the group Γ, but the Maass cusp forms are elusive and there are only extremely rare circumstances in which their explicit construction is known. In many circumstances, it is unclear whether any exist at all. One might wonder if a version of Weyl’s law still holds for the noncompact surface Γ \ H. Measurement
of the spectrum of ∆Γ\H is less straightforward when there are both discrete and continuous parts. When studying the distribution of eigenvalues of Γ \ H, it is difficult to separate the two contributions and one is forced to consider the pieces of the spectrum jointly. Denote the counting function for the discrete spectrum up to spectral height T by
NΓ(T ) = #{tj ≤ T : λj is an eigenvalue of ∆Γ\H}.
Analogously, the continuous spectrum is measured by the integral Z T −φ′ 1 + it 1 2 MΓ(T ) = 1 dt, 4π −T φ 2 + it
which yields the winding number of the determinant φΓ of the scattering matrix in the functional equation for the Eisenstein series. This counts the poles of the Eisenstein series, also known as scattering resonances
of ∆Γ\H. Weyl’s law in this setting for the counting functions considered simultaneously takes the form
(Γ \ H) N (T ) + M (T ) = ”vˆo˝l T 2 + O (T log T ) , Γ Γ 4π
as T → ∞. It is interesting to explore whether the discrete spectrum or continuous spectrum provides a larger contribution to the righthand side. There are special circumstances in which one can isolate the counting functions to address this question. In the case where Γ is a congruence subgroup, the determinant of the scattering matrix factors in terms of classical Dirichlet L-functions attached to Dirichlet characters. Using this fact, Selberg showed that the spectrum is essentially cuspidal in this case; the continuous spectrum counting function satisfies
MΓ(T ) Γ T log T,
and thus NΓ(T ) provides the main contribution to Weyl’s law; the implied constant in the above depends on the level of Γ. It follows that there are infinitely many cusp forms for congruence subgroups. This led Selberg to conjecture that cusp forms are the dominant part of the spectrum for generic finite volume hyperbolic
surfaces, for which such a clever approach to controlling the contribution from MΓ(T ) is no longer available. That the cusp forms apparently dominate for congruence subgroups is surprising and interesting, as there are currently only very limited ways of explicitly constructing any cusp forms.
1The residual spectrum actually comprises part of the discrete spectrum, see Chapter 2.
2 1.0. Introduction
As it turns out, the situation for congruence subgroups is rather special and also rather misleading; the dominance of the discrete spectrum appears to be related to the arithmeticity of the subgroup in question, and may be false for generic discrete, finite-volume subgroups of SL(2, R). In order to study the spectrum
of ∆Γ\H for generic discrete subgroups Γ ⊆ SL(2, R), Phillips-Sarnak restricted themselves to considering
those Γε, ε > 0, arising from quasiconformal deformations of congruence surfaces Γ \ H with Γ0 = Γ. Such \ H \ H \ H surfaces Γε lie in T`eˇi`c‚hffl(Γ ), the Teichmüller space of Γ . They showed that the presence of cusp forms on Γε \ H is implied by whether certain special values of a distinguished family of Rankin-Selberg L-functions are zero or not. Drawing on the work of Lax, Phillips, and Colin de Verdière, they consider perturbations of the Laplacian on the deformed surface and relate the destruction of the Maass cusp form to the nonvanishing of the inner product of a Maass cusp form and an Eisenstein series at the same spectral height. Namely, the condition is that if 1 L u ,E ·, + it =6 0, Q j 2 j
where LQ is a certain operator defined in terms of a weight 4 holomorphic modular form Q for Γ, then the \ H cusp form uj is destroyed under quasiconformal deformations Γε beginning in the direction generated by h · 1 i 1 Q. Amazingly, the inner product LQuj,E , 2 + itj can be expressed in terms of the value at sj = 2 + itj of the Rankin-Selberg L-function L (s, uj ⊗ Q), and thus the techniques used to study L-functions can be used to address the question of the destruction of Maass cusp forms. It is difficult to determine whether a particular L-function is nonzero on the critical line. Rather than focusing on an individual L-function, Luo studied the behavior of these special values on average. He showed
that in fact a positive proportion of the special values L (sj, uj ⊗ Q) are nonzero. Namely, he showed that
2 #{tj ≤ T : L(sj, uj ⊗ Q) =6 0} T ,
as T → ∞. His method proceeds by reducing the question of nonvanishing of the L-functions under study to obtaining an asymptotic expansion of the first moment of those L-functions and an upper bound for their second moment. Under the assumption that the multiplicities of the discrete spectrum of congruence surfaces are essentially bounded, it follows that a positive proportion of the discrete spectrum is destroyed under quasiconformal deformation of Γ\H, and therefore that the continuous spectrum dominates for Γε \H.
In this thesis we study various refinements of Luo’s nonvanishing results for the special values L (sj, uj ⊗ Q).
In short spectral intervals, many cusp forms for the surfaces Γ0(p) \ H are destroyed, where Γ0(p) denotes the Hecke congruence subgroup of prime level p ≥ 5.
Theorem 1.0.1. Let {uj} be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z) and let Q be a 1 +ε weight-2k cuspidal holomorphic Hecke newform for Γ0(p), where p ≥ 5 is prime. Then for M T 2 , as
3 1.0. Introduction
T → ∞ one has 1−ε #{T − M ≤ tj ≤ T + M : L(sj, uj ⊗ Q) =6 0} T M.
The lower bound T 1−εM is essentially the optimal bound predicted by Phillips-Sarnak and the Weyl law. We also study the question of whether a cusp form can be destroyed in only one basis type of quasicon-
formal deformation of the surfaces Γ0(p) \ H. Over the long spectral interval, many cusp forms are destroyed in at least two basis types of deformations for p ≥ 11.
Theorem 1.0.2. Let {uj} be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z) and let Q1 =6 Q2 be two weight-2k cuspidal holomorphic Hecke newforms for Γ0(p), where p ≥ 11 is prime. Then as T → ∞, one has 3 −ε #{tj ≤ T : L(sj, uj ⊗ Q1) =6 0 and L(sj, uj ⊗ Q2) =6 0} T 2 .
3 −ε 2−ε Unfortunately, the lower bound of T 2 in Theorem 1.0.2 is far from the optimal lower bound of T that is predicted by the Phillips-Sarnak theory. In Chapter 8 we address how this may possibly be improved by an forthcoming result of Chandee-Li on the second moment of GL(2) × GL(4) L-functions at special points. The behavior of the families of L-functions in question is actually quite interesting on its own, as at the 1 special value sj = 2 + itj the conductor of the L-function drops and the L-function behaves as though it has a lower degree. We will study this conductor-dropping behavior for additional classes of L-functions for which there is no immediate application in mind. The technical results in this dissertation are comprised of roughly of two general types: the first is a series of upper bounds for moments of families of L-functions exhibiting conductor-dropping behavior. The second type is a series of asymptotic expansions of the same sorts of moments of L-functions. For the nonvanishing results, we are interested in combining the two types to establish a nonvanishing result for the L-functions in question. A general theme in both settings is that often one can replace a certain hypothesis of square-root cancellation with a weaker Voronoi-type summation, which apply to a wider class of arithmetic functions. The upper bound type results are the following, and their proofs are found in Chapter 6. The first is an upper bound for the short interval second moment of GL(2) × GL(3) L-functions at the special point 1 sj = 2 + itj.
Theorem 1.0.3. Let {uj} be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z) and let ψ be a 21 +ε GL(3, R) automorphic form. Then for T 22 M T , as T → ∞ one has the upper bound X 2 1+ε |L(sj, uj ⊗ ψ)| T M.
T −M≤tj ≤T +M
We will also obtain the following upper bound for the sixth moment of GL(2) L-functions at the special point that is consistent with the Lindelöf hypothesis.
4 1.0. Introduction
Theorem 1.0.4. Let {uj} be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z). Then for 21 +ε T 22 M T , as T → ∞ one has X 6 1+ε |L(sj, uj)| T M.
T −M≤tj ≤T +M
In the course of proving Theorem 1.0.3 and Theorem 1.0.4, we will require a twisted large sieve inequality for short spectral intervals. The following is proved in Chapter 5.
Theorem 1.0.5. Let {uj} be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z) with normalized
Fourier coefficients νj(n). Let M,N, and T be parameters with 1 M T and N ≥ 1. Then for any N sequence A = (an) ∈ R , one has
2 ! 7 X X 4 3 1 N itj ε 2 anνj(n)n TM + N + M 2 N 2 + T (NT ) kAk . M T −M≤tj ≤T +M n≤N
The asymptotic expansions used to obtain the nonvanishing results are the following. Their proofs are contained in Chapter 7 and Chapter 8, respectively.
Theorem 1.0.6. Let {uj} be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z) and let Q be a 1 +ε weight 2k cuspidal holomorphic Hecke newform for Γ0(p), where p ≥ 5 is prime. Then for M T 2 , as T → ∞ one has the asymptotic expansion
( ) X t −T 2 − j 2 3 M | |2 ⊗ 2 +ε e νj(1) L(sj, uj Q) = 3 TM + O T . π 2 tj
Theorem 1.0.7. Let {uj}j be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z) and let Q1 =6
Q2 be two weight 2k cuspidal holomorphic Hecke newforms for Γ0(p), where p ≥ 11 is prime. Then for 5 +ε T 6 M T , as T → ∞ one has the asymptotic expansion
( ) X t −T 2 − j 11 M | |2 ⊗ ⊗ 6 +ε e νj(1) L(sj, uj Q1)L(sj, uj Q2) = c0,Q1,Q2 TM + O T , tj
for a certain constant c0,Q1,Q2 defined in terms of Q1 and Q2. Furthermore, if instead one has Q1 = Q2 = Q
5 1.0. Introduction
and p ≥ 5, then
( ) X t −T 2 − j 11 M 2 2 +ε e |νj(1)| |L(sj, uj ⊗ Q)| = c0,QTM log(T ) + c1,QTM + O T 6 ,
tj
for certain constants c0,Q and c1,Q that are defined in terms of Q.
Some remarks on organization
Part I provides an extended summary of the background material on automorphic forms for GL(2, R) and GL(3, R), automorphic L-functions, and Teichmüller theory and Phillips-Sarnak deformation theory of dis- crete groups. This part is quite lengthy and the familiar reader may wish to skip directly to Parts II and III, where the aforementioned results described in the introduction appear. At the head of the relevant chapters in these sections, we provide a more focused review of the relevant literature. Part IV contains some extended remarks on possible directions for future research. Integral transforms and properties of special functions that appear throughout this document can be found in Appendix A. Auxiliary tools used for evaluating integrals that appear in Parts II and III are contained in Appendix B.
6 Part I
Preliminaries
In Chapter 2 we review the classical spectral theory of automorphic forms on GL(2, R) and GL(3, R), and collect the general properties and summation formulas (Petersson, versions of Kuznetsov, Voronoi) that will be used in later chapters. We do not include the Selberg trace formula. We only discuss automorphic forms for trivial nebentypus. In Chapter 3, we collect some very basic theory of L-functions—for the sake of ease of reading, we have opted to include the more specific descriptions of various types of L-function in the relevant chapters, though the tools and formulas developed there may apply to a wider class. We also survey the literature on the behavior of L-functions exhibiting conductor-dropping in the spectral aspect. In Section 3.2, we provide a brief overview of the conductor-dropping phenomenon in the spectral aspect. In Chapter 4 we provide an outline of Phillips-Sarnak’s deformation theory of discrete groups and a brief review of the relevant material from Teichmüller theory.
7 Chapter 2
Preliminaries on automorphic forms
Contents 2.1 Automorphic forms for GL(2, R) ...... 9 2.1.1 The classical setting ...... 9 2.1.2 Fuchsian groups of the first kind ...... 11
2.1.3 The modular group and the Hecke congruence subgroups Γ0(N) ...... 12 2.1.4 The spectral theory of ∆ on Γ \ H ...... 16 2.1.4.1 The space of incomplete Eisenstein series ...... 17 2.1.4.2 The space of cuspidal automorphic functions ...... 19 2.1.4.2.1 The space of newforms and oldforms ...... 19 2.1.4.3 The spectral theorem ...... 20 2.1.5 The Hecke operators ...... 21 2.1.6 The Weyl law ...... 21
2.1.6.1 The Weyl law for Γ0(N) \ H ...... 22 2.1.7 Holomorphic modular forms ...... 23
2.1.7.1 Holomorphic modular forms in terms of the spectral theory of ∆k .... 25 2.1.7.2 The Petersson trace formula ...... 26 2.1.8 The Kuznetsov trace formula ...... 27 2.1.9 Bounds for Fourier coefficients and twists of Fourier coefficients ...... 29 2.1.9.1 The GL(2) Voronoi summation formula ...... 30 2.1.9.2 The Hecke bound ...... 32 2.1.10 An aside: the homogenous space setting ...... 33 2.2 Automorphic forms for GL(3, R) ...... 33 2.2.1 The Maass cusp forms for SL(3, Z) ...... 35 2.2.2 The Hecke operators and Hecke-Maass cusp forms for SL(3, Z) ...... 36 2.2.3 Bounds for Fourier coefficients and twists of Fourier coefficients ...... 37 2.2.3.1 The GL(3) Voronoi summation formula ...... 37 2.2.3.2 The Miller bound for additive twists of Fourier coefficients ...... 38 2.2.4 The minimal parabolic Eisenstein series for SL(3, Z) and Voronoi summation for the ternary divisor function ...... 39
8 2.1. Automorphic forms for GL(2, R)
2.1 Automorphic forms for GL(2, R)
In an effort to make this document as comprehensive as we can, in this section we provide an summary exposition of the relevant theory of GL(2, R) automorphic forms and geometry of the surfaces mentioned in the introduction. Our presentation largely follows Iwaniec’s book [Iwa02], and also draws on [Kat92], [Gol06], [Hej76], and [Mul10], to which we refer the reader for details. If the reader is already familiar with this background material we suggest they skip this section altogether.
2.1.1 The classical setting
There are many possible starting points for describing automorphic forms on GL(2, R). We begin with the classical setting, which is most natural for us in describing the Riemann surfaces mentioned in the introduction. Classically, one uses the upper half plane model of the hyperbolic plane,
H = {z = x + iy : x, y ∈ R, y > 0} .
The upper half plane is a Riemannian manifold, with line element
dx2 + dy2 ds2 = . y2
This equips H with a distance function, via the usual length formula Z Z 1 p dt ρ(z, w) = inf ds = inf x′(t)2 + y′(t)2 , p p p 0 y(t)
where the infimum is taken over smooth paths p = (x, y) : [0, 1] → H starting at z and ending at w. Explicitly, the distance function (the Poincaré distance function) is given by |z − w| + |z − w| ρ(z, w) = log , for z, q ∈ H, |z − w| − |z − w|
with geodesics given by half-circles intersecting the real line Rb. The special linear group consisting of 2 × 2 matrices with determinant 1, ( ! ) a b SL(2, R) = γ = : det(γ) = ad − bc = 1 , c d ! a b acts transitively on H via linear fractional transformations: for a matrix γ = ∈ SL(2, R), γ acts on c d H via az + b γz = ∈ H. cz + d This action is isometric in the sense that the metric derived from the line element ds2 is SL(2, R)-invariant. Note that both γ and −γ determine the same action on H, so one can instead consider the action of
9 2.1. Automorphic forms for GL(2, R)
PSL(2, R) on H. We will regularly abuse notation in this way. The whole group of isometries on H is generated by PSL(2, R) and the reflection operator δ : z 7→ −z. The associated area (or volume) element is the SL(2, R)-invariant Haar measure dxdy dµ(z) = , y2 also known as the Poincaré measure. ! a b The nontrivial motions given by γ = ∈ SL(2, R) on H take one of three types. Up to conjugacy c d class [γ] of γ, these are:
1. parabolic: [γ] fixes ∞: action by translation ⇔ |a + d| = 2,
2. hyperbolic: [γ] fixes 0 and ∞: action by dilation ⇔ |a + d| > 2,
3. elliptic: [γ] fixes i: action by rotation ⇔ |a + d| < 2.
We refer to the points fixed by parabolic and elliptic motions as parabolic or elliptic fixed points, respectively. Later we will outline how the geometry of quotient surfaces of H is described by the above types of motions, and how the geometry of the quotient surface is related to the spectrum of the Laplace-Beltrami on the surface. As a Riemannian manifold, H comes with an associated Laplace-Beltrami operator (or the Lapla-
cian) ∆H = div◦grad. In rectangular coordinates, the Laplacian on H (we will often suppress the underlying surface and simply denote the Laplacian by ∆) is given by 2 2 2 ∂ ∂ 2 ∂ ∂ ∆H = ∆ = y + = − (z − z) , ∂x2 ∂y2 ∂z ∂z
where ∂ 1 ∂ ∂ ∂ 1 ∂ ∂ = − i and = + i ∂z 2 ∂x ∂y ∂z 2 ∂x ∂y are the Wirtinger derivatives. It was mentioned in the introduction that the Laplace-Beltrami operator is an important invariant of a Riemannian manifold—in general, the Laplace-Beltrami operator associated with a Riemannian manifold determines the diffeomorphic isometries of the manifold: a diffeomorphism of the manifold is an isometry if and only if it leaves the Laplace-Beltrami operator invariant. Indeed, it is straightforward to check that ∆ is invariant under the action of SL(2, R) on H and δ. It is interesting to consider the eigenfunctions of ∆ on H; that is, those f : H → C satisfying
(∆ + λ)f = 0
with eigenvalue λ ∈ C and the appropriate differentiation conditions. The simplest eigenfunctions of ∆ on the full space H with nonzero eigenvalue1 are the simple power functions
1 1 I (z) = ys + y1−s and I (z) = ys − y1−s , 1,s 2 2,s 2s − 1
1The eigenfunctions of ∆ with eigenvalue λ = 0 are numerous: they are the harmonic functions on H.
10 2.1. Automorphic forms for GL(2, R)
− 6 1 both having eigenvalue λs = λs(∆) = s(1 s), when s = 2 . Generally, one makes the problem more inter- esting by restricting to subspaces of functions satisfying additional conditions. For example, by periodizing, one can show that the Whittaker functions
1 Ws(z) = 2y 2 K − 1 (2πy)e(x) s 2 ! 1 1 are eigenfunctions of ∆ that are invariant under the shifts z = z+1, and that the family of Whittaker 1 functions {W 1 (rz)}r,t∈R forms a complete eigenpacket on H. 2 +it
2.1.2 Fuchsian groups of the first kind
The action of the various kinds of subgroups of SL(2, R) are of interest. The subgroups that we are interested in are the ones yielding finite-volume Riemann surfaces. These are the Fuchsian groups of the first kind.
Definition 2.1.1. A subgroup Γ ⊆ SL(2, R) is said to be a Fuchsian group (or discrete) if Γ/{I} ⊆ PSL(2, R) acts discontinuously on H.
The action of a Fuchsian group Γ is completely determined by its action on a fundamental domain for Γ; a fundamental domain F ⊆ H for Γ is a domain in H such that ΓF ∩ F = ∅ and for all z ∈ H, Γz ∩ F =6 ∅.
Remark 2.1.1. Often we abuse notation and denote a fundamental domain for Γ by Γ \ H, although this is not well-defined since there are many fundamental domains for a given Fuchsian group; however, in the contexts where we use Γ \ H in this way the behavior of interest will be independent of our choice of a corresponding subset of H.
Definition 2.1.2. A Fuchsian group Γ is a Fuchsian group of the first kind if it is finitely generated
and has a fundamental domain FΓ = Γ \ H with finite volume, which is given by Z \ H ”vˆo˝l(Γ ) = dµ(z). FΓ
We say that Γ is co-compact if it contains no parabolic elements; in this case the fundamental domain is a compact polygon in H with an even number of sides.
With the appropriate choice of a complex atlas on the quotient space Γ \ H, one obtains a Riemann surface of constant negative curvature. A finite-volume group Γ may have finitely many cusps or punctures—these are points on the boundary of the fundamental domain lying on Rb, so they are the fixed point of a parabolic element of Γ. For any cusp
a of Γ, there is a scaling matrix σa ∈ SL(2, R) such that σa∞ = a and !
−1 1 1 σ γaσa = , a 1
where γa is the generator of the stability group Γa = {γ ∈ Γ: γa = a} = hγai of a.
11 2.1. Automorphic forms for GL(2, R)
The boundary of the fundamental domain of Γ may also contain elliptic fixed points. These correspond to branch points in the covering H → Γ \ H.
Definition 2.1.3. Let Γ be a Fuchsian group of the first kind such that Γ\H has genus g, inequivalent elliptic
fixed points ν1, . . . , νℓ of orders m1, . . . , mℓ, and h inequivalent cusps. Then we say that the signature of Γ
(or Γ \ H) is (g; m1, . . . , mℓ; h).
The signature of Γ \ H is an important invariant of Γ. For example, the volume of Γ \ H is completely determined by the signature of Γ \ H; one has the Gauss-Bonnet formula Xℓ 1 (Γ \ H) = 2π 2g − 2 + 1 − + h . ”vˆo˝l m j=1 j
In Chapter 4 we will see that the signature of Γ \ H characterizes the Riemann surface up to quasiconformal equivalence.
Remark 2.1.2. From this point forward, all Fuchsian groups of the first kind that we deal with will be assumed to be noncocompact unless otherwise specified, and will be denoted by Γ. Furthermore we will assume that ∞ is a cusp.
When Γ \ H has cusps, a great deal of information about the spectrum of ∆Γ\H is captured by the
Kloosterman sums associated with Γ \ H. Suppose that a and b are cusps of Γ, and c ∈ Cab, where ( ! ) ∗ ∗ −1 Cab = c > 0 : ∈ σ Γσb . c ∗ a
Then the Kloosterman sum of modulus c attached to a and b is defined by X md na Sab(n, m; c) = e + , (2.1) c c ∗ a −1 ∈Γ∞\σa Γσb/Γ∞ c d for m, n ∈ Z.
2.1.3 The modular group and the Hecke congruence subgroups Γ0(N)
Without imposing any restrictions on the eigenfunctions of ∆, the spectral theory on H is not particularly interesting. The natural restrictions to impose are invariance under the action of subgroups Γ ⊆ SL(2, R), which amounts to studying the spectrum of ∆ on the quotient surface Γ \ H. We shall deal with two distinguished types of Fuchsian groups of the first kind: the Hecke congruence subgroups of level N (which we will assume later to be prime), ( ! ) a b Γ0(N) = γ = ∈ SL(2, Z): c ≡ 0 (mod N) , c d
12 2.1. Automorphic forms for GL(2, R)
and those subgroups that arise from quasiconformal deformations of Γ0(N) \ H, described in Chapter 4. The
full modular group, SL(2, Z) = Γ0(1), has fundamental domain 1 F = z = x + iy : |x| < , |z| > 1 0,1 2
π Z \ H with volume 3 . The quotient surface SL(2, ) is a noncompact, finite-volume Riemann√ surface with one ∞ 1 i 3 cusp at , one elliptic fixed point i of order two, and one elliptic fixed point 2 + 2 of order three.
SL(2, Z) \ H
i √ i 1 i 3 2 + 2
ρ
(b) The Riemann surface Γ0(1) \ H
1 1 − 2 2
(a) Fundamental domain for Γ0(1) \ H
Figure 2.1: Fundamental domain F0,1 for the modular group SL(2, Z) and the resulting Riemann surface.
For higher level, the surfaces Γ0(N) \ H are noncompact, finite-volume Riemann surfaces whose volume is given by \ H \ H · ”vˆo˝l (Γ0(N) ) = ”vˆo˝l (Γ0(1) ) [Γ0(1) : Γ0(N)] ,
with Y Y 1 µ (N) := [Γ (1) : Γ (N)] = pα + pα−1 = N 1 + . 0 0 0 p pα∥N p|N
The fundamental domain of Γ0(N) is tessellated by µ0(N) copies of the fundamental domain of Γ0(1). Below are several examples.
13 2.1. Automorphic forms for GL(2, R)
i Γ0(11) \ H i Γ0(14) \ H i Γ0(37) \ H
√ 21+i 3 74 74 √ 53+i 3 74 74
6 + i 37 37
31+ i 37 37
1 4 1 1 2 1 12 1 1 2 53 1 15 1 1 2 1 3 2 3 55 1 3 2 3 47 3 2 3 74 6 4 3 5 2 5 3 4 6
(a) Γ0(11) \ H (b) Γ0(14) \ H (c) Γ0(37) \ H
Figure 2.2: Some examples of fundamental domains for Hecke congruence subgroups Γ0(N), truncated to finite height inside the standard polygon; the pairings of sides of the hyperbolic polygon are indicated by color. The Riemann surface corresponding to Γ0(11) \ H is a torus with two inequivalent cusps at 0 and ∞, \ H 1 1 ∞ \ H while Γ0(14) is a torus with four inequivalent cusps at 0, 7 , 2 and . As a Riemann surface, Γ0(37) ∞ has genus 2 and two inequivalent cusps at 0 and , as well as two inequivalent elliptic√ fixed points√ of order 6 i 31 i 21 i 3 53 i 3 2 at 37 + 37 and 37 + 37 , and inequivalent elliptic fixed points of order 3 at 74 + 74 and 74 + 74 .
The number of inequivalent cusps of Γ0(N) \ H is X c0(N) = φ ((n, m)) . nm=N
In the case where N is square-free, then c0(N) = τ2(N). The number of inequivalent elliptic fixed points of order 2 is 0 if 4|N, Y ν2(N) = −1 1 + otherwise, p p|N and the number of inequivalent elliptic fixed points of order 3 is 0 if 9|N, Y ν3(N) = −3 1 + otherwise, p p|N · where · is the Jacobi symbol and φ is the Euler totient function. One shows that Γ0(N) \ H has genus given by µ (N) ν (N) ν (N) c (N) (Γ (N) \ H) = 1 + 0 − 2 − 3 − 0 . `g´e›n˚u¯s 0 12 4 3 2
14 2.1. Automorphic forms for GL(2, R)
∞ ←− −→0
(a) Γ0(11) \ H
1 2 −→
←−∞ −→0
1 7 ←−
(b) Γ0(14) \ H √ i 11 µ = 31 + i µ = 26 + 2,0 37 37 3,0 37 37 ∞ ←− −→0
√ 6 i µ = + 10 3i 3 2,1 37 37 µ = + 3,1 37 37
(c) Γ0(37) \ H
Figure 2.3: Some examples of quotient surfaces for Hecke congruence subgroups Γ0(N).
15 2.1. Automorphic forms for GL(2, R)
With the additional symmetries imposed by requiring a function f : H → C to be invariant under the action
of Γ0(N), the spectral theory of ∆ on Γ0(N) \ H becomes much more interesting.
2.1.4 The spectral theory of ∆ on Γ \ H
Consider the automorphic functions defined on Γ \ H having moderate growth in cusps. These functions f : H → C are automorphic with respect to Γ; namely, those satisfying
f(γz) = f(z) for all γ ∈ Γ;
if f is additionally an eigenfunction of ∆, then we say that f is an automorphic form for Γ. An automorphic
form f with Laplace eigenvalue λf = s(1 − s) for some s ∈ C has Fourier expansions at each cusp a of Γ given by X b f(σaz) = Fa(y) + fa(n)Ws(nz), n=0̸
provided again that f does not grow too quickly in the cusps. We say that f is cuspidal if fa(y) ≡ 0 for all cusps a. Here the coefficients are denoted by Z 1 b fa(n) = f (σax) e (−nx) dx. 0 ≥ < 1 = = ∈ 1 Since ∆ is nonnegative and symmetric, we have λf 0, so that (s) = 2 and (s) = 0 or (s) 0, 2 . The space of square-integrable functions L2(Γ\H) is an inner product space with respect to the Petersson inner product, defined by Z hf, gi = f(z)g(z)dµ(z) Γ\H for f, g ∈ L2(Γ \ H). We denote the corresponding Petersson norm by kfk2 = hf, fi. The Laplacian acts on the space {f ∈ L2(Γ \ H): f and ∆f are smooth and ∆f ∈ L2(Γ \ H)}.
This is a dense subspace of L2(Γ \ H), from which ∆ has a self-adjoint extension to L2(Γ \ H) that is nonnegative. Then L2(Γ \ H) has the orthogonal decomposition
L2(Γ \ H) = C(Γ \ H) ⊕ E(Γ \ H), where C(Γ \ H) denotes the space of cuspidal automorphic functions, and E(Γ \ H) denotes the space of incomplete Eisenstein series; we will describe how these spaces decompose with respect to ∆ concretely in the two sections that follow.
16 2.1. Automorphic forms for GL(2, R)
2.1.4.1 The space of incomplete Eisenstein series
The incomplete Eisenstein series attached to cusps are defined by X | = −1 ∈ ∞ ∞ Ea(z ψ) = ψ (σa γz) , for ψ C0 (0, ).
γ∈Γa\Γ
Taking ψ(z) = zs yields a concrete class of automorphic forms: the Eisenstein series attached to the cusps {a} of Γ. These are defined by X = −1 s < Ea(z, s) = σa γz , for (s) > 1; (2.2)
γ∈Γa\Γ
However, the Eisenstein series are not in L2(Γ \ H), and the above series only converges absolutely for <(s) > 1. It turns out that the Eisenstein series span the space of incomplete Eisenstein series—we make this explicit shortly. The Eisenstein series have the Fourier expansions at cusps given by X s 1−s Ea(σbz, s) = δ(a = b)y + φab(s)y + φab(n, s)Ws(nz), n=0̸
where the functions defining the zeroth coefficients are given by the Dirichlet series √ Γ s − 1 X S (0, 0; c) φ (s) = π 2 ab ab Γ(s) c2s c≥1
and π2|n|s−1 X S (n, 0; c) φ (n, s) = ab . ab Γ(s) c2s c≥1
In the above the Sab(n, m; c) are the Kloosterman sums defined in (2.1).
Example 2.1.3. For the modular group SL(2, Z), there is only the classical Kloosterman sum S∞∞(n, m; c) = S(n, m; c), and we have S(0, 0; c) = φ(c), where φ is the Euler totient function. The Fourier coefficients of the single Eisenstein series E∞(z, s) = E(z, s) have pleasant closed forms, owing to the special formulas √ − 1 X √ − 1 − Γ s 2 φ(c) Γ s 2 ζ(2s 1) φ∞∞(s) = φ Z (s) = π = π SL(2, ) Γ(s) c2s Γ(s) ζ(2s) c≥1
and 2 s−1 X 2 s−1 π |n| S(n, 0; c) π |n| σ1−2s(n) φ∞∞(n, s) = = . Γ(s) c2s Γ(s) ζ(2s) c≥1 In this case the Fourier expansion takes the form √ X s 1−s E(z, s) = y + φSL(2,Z)(s)y + 4 y τ − 1 (n)K − 1 (2πny) cos(2πnx). s 2 s 2 n≥1
17 2.1. Automorphic forms for GL(2, R)
Since the Eisenstein series are only defined by the series (2.2) for <(s) > 1, one uses their functional equation to extend them meromorphically to C. Define the column vector E(z, s) of the Eisenstein series attached to the inequivalent cusps a1,..., ar of Γ by E (z, s) a1 . E(z, s) = . .
Ear (z, s) and define the scattering matrix of Γ by φ (s) ··· φ (s) a1a1 a1ar . . . ΦΓ(s) = . .. . . ··· φar a1 (s) φar ar (s)
Then E(z, s) satisfies the functional equation
E(z, s) = ΦΓ(s)E(z, 1 − s), (2.3) and this provides the analytic continuation of the Eisenstein series to <(s) ≤ 1. The Eisenstein series Ea(z, s) < ≥ 1 1 has at most finitely many poles in (s) 2 ; these are all simple, lie in ( 2 , 1], and they are also poles for 2 the diagonal entries φaa(s). The residues of these poles lie in L (Γ \ H) and are orthogonal to cusp forms, but form part of the discrete spectrum of ∆; we refer to these as the residual spectrum. When we have a particular cusp a in mind, we denote the Maass forms in the residual spectrum by uaj(z). However, there can be linear dependences between the uaj(z) for inequivalent cusps of Γ, so without reference to any cusps we denote a general basis element for the span of the residues by uj(z). We also normalize the residues to
have kuajk = 1. ≤ < 1 One also expects there to be many poles of the Eisenstein series in the segment in 0 (s) < 2 ; these are referred to as resonances.
Example 2.1.4. In the case of Hecke congruence subgroups Γ0(N), when N is odd and square-free the \ H ≡ 1 ≡ 1 entries of ΦΓ0(N)(s) are given as follows. For cusps a and b of Γ0(N) , say a v and b v′ with vw = v′w′ = N, we have Y Y ′ ′ 2s −1 s 1−s φab(s) = φSL(2,Z)(s) · φ ((w, w ) · (v, v )) p − 1 p − p . p|N p|(w,w′)·(v,v′)
< ≥ 1 These are all holomorphic in (s) 2 , except for a simple pole at s = 1, which correspond to the constant functions.
The space E(Γ \ H) of incomplete Eisenstein series Ea(z|ψ) is spanned by the Eisenstein series in the sense that each incomplete Eisenstein series has the expansion Z X 1 X 1 1 Ea(z|ψ) = hEa(·|ψ), uajiuaj(z) + Ea(·|ψ),Eb ·, + it Eb z, + it dt. 4π R 2 2 1 ≤ b 2 18 2.1. Automorphic forms for GL(2, R) 1 For the Eisenstein series Eab z, 2 + it , the normalized coefficients are defined to be 1 4π|n| 2 1 η (n, t) = φ n, + it . ab cosh(πt) ab 2 2.1.4.2 The space of cuspidal automorphic functions The cuspidal automorphic forms are called cusp forms or Maass cusp forms. They are those automorphic forms whose zeroth eigenvalues are zero at all cusps. Definition 2.1.4. A cusp form or Maass cusp form for Γ is a function u ∈ L2(Γ \ H) satisfying the following: − 1 2 1. ∆u = λuu, where λu = su(1 su) = 4 + tu is the Laplace eigenvalue of u, 2. For each cusp a of Γ, Z 1 fa(y) = f (σaz) dx ≡ 0. 0 By the abstract spectral theorem, C(Γ \ H) is spanned by the Maass cusp forms, and we denote an orthonormal basis for it by {uj}. The eigenspaces are all finite dimensional, and for any f ∈ C(Γ \ H) such that ∆f is smooth and k∆fk < ∞, one has X f(z) = hf, ujiuj(z), j with the eigenvalues being counted with multiplicity. Given an orthonormal basis {uj} for C(Γ \ H), we denote the corresponding Laplace eigenvalue of each uj by 1 1 λ = + t2 = s (1 − s ), with s = + it . j 4 j j j j 2 j The Fourier expansion at a cusp a is given by X uj(σaz) = ρj,a(n)Wsj (nz), n=0̸ and we define the normalized Fourier coefficients to be 1 4π|n| 2 νj,a(n) = ρj,a(n) cosh(πtj) The distinguished normalized Fourier coefficients in the cusp ∞ are denoted by νj(n). 2.1.4.2.1 The space of newforms and oldforms For the Hecke congruence subgroups, C(Γ \ H) decomposes further depending on whether the first Fourier coefficient is zero or not. For M|N and M < N, one has Γ0(N) ⊆ Γ0(M), and if v(z) is a Maass cusp form for 19 2.1. Automorphic forms for GL(2, R) | Γ0(M), then for each D such that DM N, vD(z) = v(Dz) is a cusp form for Γ0(N). Furthermore, νvD (n) = 0 for n 6≡ 0 (mod D). The span of such cusp forms is the space of oldforms, denoted by C`o˝l´dffl(Γ0(N) \ H). The orthogonal complement C`o˝l´dffl(Γ0(N) \ H) comprises the newforms of level N, and we denote the space of these by C”n`e›w(Γ \ H). A newform of level M|N with M < N residing in C`o˝l´dffl(Γ0(N) \ H) of the form u(z) = vD(z) for some DM|N is called an oldform. 2.1.4.3 The spectral theorem Combining the expansions in the previous two sections yields the spectral theorem, or the full spectral decomposition of L2(Γ \ H) with respect to ∆. Theorem 2.1.1. [The Spectral Theorem, [Iwa02]] Let Γ ⊆ SL(2, R) be a Fuchsian group of the first kind. The cuspidal space C(Γ \ H) is spanned by countably many eigenfunctions of ∆, the Maass cusp forms. The eigenspaces of ∆ on C(Γ \ H) each have finite dimension. The space E(Γ \ H) of incomplete Eisenstein series has the orthogonal decomposition M E(Γ \ H) = R(Γ \ H) ⊕ Ea(Γ \ H), a R \ H ∈ 1 where (Γ ) is the span of the Eisenstein series with residual eigenvalues having sj ( 2 , 1], M R \ H R \ H (Γ ) = sj (Γ ), 1 ≤ 2 R \ H where sj (Γ ) is the space spanned by the residue of all Eisenstein series regardless of cusps at sj. The multiplicity of each residual eigenvalue is bounded above by the number of cusps of Γ. For each cusp, the E \ H 1 ∞ ∈ 2 \ H spectrum of ∆ in a(Γ ) covers [ 4 , ) uniformly with multiplicity 1. Thus any f L (Γ ) has the expansion Z X X 1 ∞ 1 1 f(z) = hf, ujiuj(z) + f, Ea ·, + it Ea z, + it dt. 4π −∞ 2 2 j a Remark 2.1.5. In the case that Γ = SL2(Z), there is only one residual eigenvalue at s = 1, which corre- ≡ 3 sponds to the eigenvalue λ0 = 0. The corresponding residue of the Eisenstein series is u0(z) π , so we have R(SL(2, Z) \ H) = C. In general, there is always a residual eigenvalue at s = 1 for which the corresponding ≡ \ H −1 residue of the Eisenstein series is u0(z) (”vˆo˝l (Γ )) , which spans the constant functions. 20 2.1. Automorphic forms for GL(2, R) 2.1.5 The Hecke operators 2 The Hecke congruence subgroups Γ0(N) come with a special collection of operators, the Hecke operators 2 2 Tn : L (Γ0(N) \ H) → L (Γ0(N) \ H), that commute with ∆. These are defined by 1 X 1 X X az + d (T f)(z) = √ f(γz) = √ f , n n n d γ∈SL(2,Z)\Γn ad=n b(d) where Γn = {γ ∈ GL(2, Z) : det γ = n}. The Tn with (n, N) = 1 form a family of self-adjoint operators that 2 2 commute with the action of ∆ on L (Γ0(N) \ H), so we can choose a basis for L (Γ0(N) \ H) of simultaneous eigenfunctions for ∆, δ, and {Tn}(n,N)=1. These are the Hecke-Maass forms, and for cusp forms we denote the Hecke eigenvalues of a basis element uj by λj(n). For the Fourier coefficients in the cusp a = ∞, one has νj(n) = νj(1)λj(n) for (n, N) = 1, and the Hecke eigenvalues satisfy the multiplicativity relation X mn ν (m)λ (n) = ν , for (n, N) = 1. (2.4) j j j d2 d|(m,n) The Hecke operators preserve C`o˝l´dffl(Γ0(N) \ H) and C”n`e›w(Γ0(N) \ H), and we can choose a basis for these consisting of simultaneous eigenfunctions. When uj is a newform of level N, then νj(1) =6 0, and in this case one further has X mn λ (m)λ (n) = λ , for (n, N) = 1, j j j d2 d|(m,n) and λj(m)λj(p) = λj(mp), for p|N prime. 1 ∞ For Eisenstein series E∞ z, 2 + it , the eigenvalues at the cusp are τit(n), and these also satisfy the above multiplicativity relation. If δuj = uj, we say that uj is even, and if δuj = −uj we say it is odd. The Fourier coefficients of even Hecke-Maass cusp forms satisfy νj(−n) = νj(n) for all n, while the odd cusp forms satisfy νj(−n) = −νj(n). 2.1.6 The Weyl law As we described in the introduction, the distribution of the discrete eigenvalues of ∆ is of great interest and is related to the geometry of the surface in question. Recall that the counting function for the continuous spectrum is defined by Z T −φ′ 1 + it 1 Γ 2 MΓ(T ) = 1 dt, 4π −T φΓ 2 + it 2Hecke operators are also defined in more general settings; see [Ven90]. 21 2.1. Automorphic forms for GL(2, R) where φΓ is determinant of the scattering matrix for Γ. By the appropriate choice of a test function in the Selberg trace formula, one obtains the Weyl law for noncompact Γ \ H: (Γ \ H) h T N (T ) + M (T ) = ”vˆo˝l T 2 − Γ T log T + c T + O , Γ Γ 4π π Γ log T \ H hΓ − where hΓ is the number of inequivalent cusps of Γ and cΓ = π (1 log 2); see [Sel91]. 50i 40i 30i - cuspidal spectrum 20i < ≤ 1 - poles of φΓ0(1) in (s) 2 - residual spectrum 10i 1 1 1 4 2 Figure 2.4: Plot of small eigenvalues of ∆, the residual spectrum, and poles of Eisenstein series for SL(2, Z). The cusp forms dominate. 2.1.6.1 The Weyl law for Γ0(N) \ H In the case where Γ = Γ0(N) is a Hecke congruence subgroup, a great deal more can be said about the continuous spectrum. In the case where N = p is prime, the scattering matrix is given by ! 1 − 1 √ − − − − s − 1 s √ − − Γ s 2 ζ (2 2s) 2s 1 p 1 p p Γ s 2 ζ (2 2s) ΦΓ (p)(s) = π p − 1 = π Np(s). 0 Γ(s) ζ (2s) ps − p1−s p − 1 Γ(s) ζ (2s) 22 2.1. Automorphic forms for GL(2, R) More generally, for N = p1p2 ··· pr square-free, the scattering matrix for Γ0(N) is given by √ Γ s − 1 ζ (2 − 2s) O Φ (s) = π 2 N (s); Γ0(N) Γ(s) ζ (2s) p p|N the computation can be found in [Hej76]. Explicitly, then the determinant of the scattering matrix φΓ0(N)(s) has the factorization − − ⊮ 2 − 2r 1 Λ (2 2s, ) 1−2s Λ (2 2s, χN ) φΓ0(N)(s) = det ΦΓ0(N)(s) = N , Λ (2s, ⊮) Λ (2s, χN ) where χN is the principal Dirichlet modulo N, and Λ(s, ⊮) and Λ(s, χN ) are the completed L-functions for ζ(s) and L(s, χN ), respectively. In these cases, MΓ0(N)(T ) essentially counts the nontrivial zeros of those L-functions in the critical strip, so that by known standard techniques one has MΓ0(N)(T ) T log T, so that one can see that the discrete spectrum dominates. Namely, one has (Γ (N) \ H) T N (T ) = ”vˆo˝l 0 T 2 − c T log T − c T + O . Γ0(N) 4π 1,Γ0(N) 2,Γ0(N) log T 2.1.7 Holomorphic modular forms There is an extensive theory of holomorphic modular forms as well. We shall only describe a small portion of it; for a basic introduction, see Iwaniec-Kowalski’s book [IK04]. Definition 2.1.5. Let k ≥ 0 be an even integer. A modular form of weight k for Γ is a holomorphic function on H that respects the action of Γ as follows: ! a b Q(γz) = (cz + d)kQ(z) for all γ = ∈ Γ. c d We also require Q to be holomorphic at all cusps of Γ; that is, in local coordinates near each cusp a of Γ \ H, f is complex-differentiable. Remark 2.1.6.! The j-factor cz + d in the modularity condition is often denoted by j(γ, z) = cz + d, for a b γ = ∈ Γ. The j-factor satisfies j(γ1γ2, z) = j(γ1, γ2z) · j(γ2, z). c d Remark 2.1.7. The more convenient definition of holomorphicity at a cusp a of Γ \ H is phrased in terms −k of the Fourier series for Q. Note that by the modularity condition, the function Qa(z) = Q(σaz)j(σa, z) 23 2.1. Automorphic forms for GL(2, R) ! 1 1 is invariant under the action of , since 1 ! ! ! !−k 1 1 1 1 Qa(z + 1) = Q σa z · j σa , z 1 1 −k = Q(γaσaz) · j(γaσa, z) = Qa(z). Thus Q has Fourier expansion at a given by X X c k−1 Qa(z) = Qa(n)e(nz) = νQ,a(n)n 2 e(nz), (2.5) n∈Z n∈Z c and Q is holomorphic at a if Qa(n) = 0 for n < 0. We refer to the νQ,a(n) as the normalized Fourier coefficients of Q at a. If νQ,a(0) = 0 for all cusps, then we say that Q is cuspidal or a cusp form; note that in this case Q has exponential decay as z → a. If not, the growth is moderate. The Fourier expansion (2.5) coincides with the usual definition of holomorphicity in terms of the Riemann surface structure of Γ \ H in the following sense: taking local coordinates q = e(z) of Γ \ H near ∞, the condition is that the Laurent series expansion in q of Q at a has negative index coefficients all zero; hence, Q is holomorphic at a. Remark 2.1.8. The Fourier expansion at ∞ is special and we denote the corresponding Fourier coefficients d by Q∞(n) = ρQ(n) and νQ,∞(n) = νQ(n). By the work of Deligne ([Del74]) these are known to satisfy the Ramanujan conjecture: one has νQ(n) τ(n) for all n. The spaces Mk(Γ \ H) and Sk(Γ \ H) of weight-k holomorphic modular forms for Γ that are noncuspidal and cuspidal, respectively, are both finite dimensional. The space of weight-k cusp forms is an inner product space, with the Petersson inner product defined by Z k hQ1,Q2ik = y Q1(z)Q2(z)dµ(z), for Q1,Q2 ∈ Sk(Γ \ H). Γ\H By interpreting the (even) weight-k modular forms as k-fold differential forms on Γ \ H, using the Riemann- Roch theorem one finds that for Γ with signature (g; m1, . . . , mℓ; h), still assuming h > 0, the dimension of Mk(Γ \ H) is given by Xℓ k 1 k (k − 1)(g − 1) + 1 − + − 1 h if k ≥ 4, 2 mj 2 M \ H j=1 `d˚i‹mffl ( k(Γ )) = g + h − 1 if k = 2, 1 if k = 0, 24 2.1. Automorphic forms for GL(2, R) and the dimension of Sk(Γ \ H) is given by M \ H − ≥ `d˚i‹mffl ( k(Γ )) h if k 4, (Sk(Γ \ H)) = g if k = 2, `d˚i‹mffl 0 if k = 0. For the Hecke congruence subgroups specifically, the dimension of the space of cusp forms is given by k k k (S (Γ (N) \ H)) = (k − 1) · ( (Γ (N) \ H) − 1) + − 1 · c (N) + ν (N) · + ν (N) · , `d˚i‹mffl k 0 `g´e›n˚u¯s 0 2 0 2 4 3 3 \ H − for k > 2, and `d˚i‹mffl (S2(Γ0(N) )) = `g´e›n˚u¯s(Γ0(N))+c0(N) 1, where c0(N), ν2(N), and ν3(N) are as defined earlier. Remark 2.1.9. Note that modular forms are not actually well-defined on the surface Γ \ H, so we are abusing notation when we write Mk(Γ \ H) and Sk(Γ \ H); the standard notation is Mk(Γ) and Sk(Γ). By the uniformization theorem, however, it makes sense to write Sk(Γ\H). The interpretation of Q ∈ Sk(Γ\H) in terms of k-fold differential forms on Γ \ H is natural: there is a bijection between Sk(Γ \ H) and sections of the line bundle ω⊗k for the universal elliptic curve π : E → Γ \ H over Γ \ H. This connection allows one to fruitfully carry over homological tools from algebraic geometry. For congruence subgroups, the Hecke operators also act on Sk(Γ0(N) \ H), so we can choose a basis for it consisting of simultaneous eigenfunctions. We will occasionally denote such a basis for Sk(Γ0(N) \ H) by Bk(Γ0(N) \ H). We denote the normalized Hecke eigenvalues by λQ(n), with k−1 TnQ = λQ(n)n 2 f, for (n, N) = 1 so that these satisfy the Hecke relations X mn λ (m)λ (n) = λ , for (n, N) = 1. (2.6) Q Q Q d2 d|(m,n) One finds that νQ(n) = λQ(n)νQ(1) for (n, N) = 1. As before, one needs to restrict attention to the space of newforms in order for the Fourier coefficients to be multiplicative; in this case one has λQ(n)νQ(1) = νQ(n) for all n, and so in addition to satisfying (2.6), these satisfy λQ(m)λQ(p) = λ(mp), for p|N. Throughout we will take the basis newforms to be normalized so that λQ(1) = 1. 2.1.7.1 Holomorphic modular forms in terms of the spectral theory of ∆k In the previous discussion of the spectral theory of Γ \ H, one could also have viewed the action of SL(2, R) on H as an action on the space of smooth functions C∞(H) instead. Similarly, one can view the weight-k holomorphic modular forms for Γ in terms of a modified action of SL(2, R) on C∞(H): one defines the 25 2.1. Automorphic forms for GL(2, R) ∞ weight-k action SL(2, R) ⟳ C (H) sending γ : Q 7→ Q|kγ by − cz + d k (Q| γ)(z) = Q(γz). (2.7) k |cz + d| The weight-k automorphic functions are those f ∈ C∞(H) that are invariant under (2.7): ∞ ∞ C (Γ \ H, k) := {f ∈ C (H): f|kγ = Q} . The weight-k Laplacian is defined by ∂2 ∂2 ∂ ∆ = y2 + − iky , k ∂x2 ∂y2 ∂x ∞ and the weight-k Maass forms are those Q ∈ C (Γ \ H, k) that are eigenfunctions of ∆k and that have moderate growth in cusps. As in the weight-0 case, one has the spectral decomposition L2(Γ \ H, k) = C(Γ \ H, k) ⊕ E(Γ \ H, k), where the space of cuspidal weight-k automorphic functions and the space of incomplete Poincaré series are defined analogously. The spectral resolution of ∆k in this case decomposes C(Γ \ H, k) in terms of a basis of k ∈ \ H 2 weight-k Maass cusp forms. Furthermore, if Q Sk(Γ ) is a cusp form, then y Q(z) is a weight-k Maass k k − cusp form for Γ with weight-k Laplace eigenvalue λQ = 2 2 1 . The automorphic functions of various weights for Γ form a graded ring: one has M M 2 + 2 + ∼ 2 L (Γ \ GL(2, R) ) = L (Γ \ GL(2, R) , k) = L (Γ \ H, k), k k and the mass-raising and mass-lowering operators ∂ ∂ k ∂ k R = iy + y + = (z − z) + k ∂x ∂y 2 ∂z 2 and ∂ ∂ k ∂ k L = −iy + y − = −(z − z) − k ∂x ∂y 2 ∂z 2 2 2 2 2 such that Rk : L (Γ \ H, k) → L (Γ \ H, k + 2) and Lk : L (Γ \ H, k) → L (Γ \ H, k − 2). 2.1.7.2 The Petersson trace formula The Petersson trace formula relates the Fourier coefficients of weight-k holomorphic modular forms to the Kloosterman sums attached to Γ. The definitions of the various integral transforms in this section can be found in Appendix A. 26 2.1. Automorphic forms for GL(2, R) Theorem 2.1.2. Let k, m, n ∈ N and let k be even. Let a and b be two cusps of Γ \ H, and let Bk(Γ \ H) denote an orthonormal basis of weight-k holomorphic modular forms for Γ. Then √ − X X (k 2)! k Sab(m, n; c) 4π mn ν (m)ν (n) = δ(a = b)δ(m = n) + 2πi J − . (4π)k−1 f,a f,b c k 1 c f∈Bk(Γ\H) c∈Cab Summing over all weights k and applying the Sears-Titchmarsh inversion formulas (see Appendix A) one arrives at the following general form of the Petersson trace formula. This comprises an important piece of the Kuznetsov trace formula, and corresponds to the contribution holomorphic discrete series representations of Γ in the representation theoretic perspective; see [CPS90]. Theorem 2.1.3. Let m, n ∈ N. Let a and b be two cusps of Γ \ H. Let φ be a smooth function on [0, ∞) satisfying • |φ(x)| x as x → 0, • φ(r)(x) x−3 as x → ∞ for 0 ≤ r ≤ 3. Then √ X k X X ∞ i Γ(k)N (k − 1) S (m, n; c) 4π mn δ(a = b)δ(m = n)φ + φ ν (m)ν (n) = ab φS , (4π)k−1 f,a f,b c c k≥2 f∈Bk(Γ\H) c∈Cab 2|k where Z ∞ Z ∞ ∞ 1 2 T φ = 2 J0(x)φ(x)dx = 2 φ(t)t tanh(πt)dt. 2π 0 π 0 2.1.8 The Kuznetsov trace formula Similar to the Petersson trace formula, the Kuznetsov trace formula captures the deep connection between automorphic forms for Γ ⊆ SL(2, R) and the Kloosterman sums Sab(m, n; c). In one direction, it yields an expansion of spectral features of Γ in terms of algebro-geometric features of Γ. There are many generalizations of Kuznetsov’s original formula; in this section we document the versions we will make use of later. The various integral transforms that appear in the statements below can be found in Appendix A. |= | ≤ 1 Theorem 2.1.4. Let h be an even function that is regular in the strip (t) 2 , where it satisfies the bound 1 |h(t)| (1 + |t|)2+δ 27 2.1. Automorphic forms for GL(2, R) for some δ > 0. Let a and b be cusps of Γ \ H. Then for any m, n ∈ Z with mn =6 0, we have Z X X 1 ∞ h(tj)νj,a(m)νj,b(n) + 2 h(t)ηac(m, t)ηbc(n, t)dt 4π −∞ j c p ! Z X δ(a = b)δ(n = m) Sab(m, n; c) 4π |mn| ± = 2 t tanh(πt)h(t)dt + h , (2.8) π R c c c∈Cab where = sgn(mn) and the integral transforms are defined to be Z 2i tdt h+(x) = J2it (x) h(t) , π R cosh(πt) and Z 4 − h (x) = 2 K2it (x) h(t)t sinh(πt)dt. π R Remark 2.1.10. In the specific case of the full modular group Γ = SL(2, Z), there is only one cusp at ∞ and the normalized Fourier coefficients of the Eisenstein series are 1 1 − 1 1 1 | | 2 2 +it| | 2 +it 2 2 +it 4π n π n σ−2it(n) 4π π τit(n) η∞∞(n, t) = η(n, t) = 1 = 1 . cosh(πt) Γ 2 + it ζ(1 + 2it) cosh(πt) Γ 2 + it ζ(1 + 2it) In the contribution from the Eisenstein series, we then have 4πτ (m)τ (n) η (m, t)η (n, t) = it it , ac bc |ζ(1 + 2it)|2 π 1 1 − since cosh(πt) = Γ 2 + it Γ 2 it . Thus the Kuznetsov trace formula takes the form Z X 2 ∞ τ (n)τ (m) h(t )ν (m)ν (n) + h(t) it it dt j j j | |2 π 0 ζ(1 + 2it) j Z √ δ(n = m) X S(m, n; c) 4π mn ± = 2 t tanh(πt)h(t)dt + h . π R c c c≥1 There is a perhaps more illuminating form of Kuznetsov that fully displays the relationship between Kloosterman sums and the representation theory of GL(2, R). This form is better regarded as the full spectral decomposition of the Kloosterman zeta function X S (m, n; c) Z (m, n) = ab , for <(s) > 1. ab,s cs c∈Cab 28 2.1. Automorphic forms for GL(2, R) Theorem 2.1.5. Let φ be a smooth function on [0, ∞) satisfying • |φ(x)| x as x → 0, • φ(r)(x) x−3 as x → ∞ for 0 ≤ r ≤ 3. Then for m, n ∈ N, we have √ X X X Z ∞ Sab(m, n; c) 4π mn T 1 T φ = νj,a(m)νj,b(n) φ(tj) + 2 φ(t)ηac(m, t)ηbc(n, t)dt c c 4π −∞ c∈Cab j c X πikΓ(k)N (k − 1) X + φ ν (m)ν (n). (4π)k−1 f,a f,b k≥2 f∈Bk(Γ\H) 2|k Without the contribution from the holomorphic modular forms, one has √ X S (m, n; c) 4π mn δ(a = b)δ(m = n)φ∞+ ab φ c c c∈Cab X X Z ∞ T 1 T = νj,a(m)νj,b(n) φ(tj) + 2 φ(t)ηac(m, t)ηbc(n, t)dt, 4π −∞ j c and if mn < 0, then one has (by the Kontorovich-Lebedev inversion formula—see Appendix A) √ X X X Z ∞ Sab(m, n; c) 4π mn L 1 L φ = νj,a(m)νj,b(n) φ(tj) + 2 φ(t)ηac(m, t)ηbc(n, t)dt. c c 4π −∞ c∈Cab j c Remark 2.1.11. For the full modular group, Theorem 2.1.5 takes the form √ Z X S(m, n; c) 4π mn X 2 ∞ τ (m)τ (n) φ = ν (m)ν (n)T (t ) + T (t) it it dt c c j j φ j π φ |ζ(1 + 2it)|2 c≥1 j 0 X πikΓ(k)N (k − 1) X + φ ν (m)ν (n). (4π)k−1 f f k≥2 f∈Bk(Γ0(1)\H) 2|k 2.1.9 Bounds for Fourier coefficients and twists of Fourier coefficients For holomorphic modular forms, the individual Fourier coefficients satisfy λQ(n) τ(n), known as the Deligne bound for the Fourier coefficients ([Del74]). In general, the Ramanujan-Petersson conjecture ε predicts that the same bound holds for the Fourier coefficients of Maass cusp forms; namely, that |λj(n)| n 7 +ε for any ε > 0. The current best bound towards the Ramanujan-Petersson conjecture is |λj(n)| n 64 , due to Kim-Sarnak [Kim03]. On the other hand, by the Rankin-Selberg method, one shows that on average the Fourier coefficients 29 2.1. Automorphic forms for GL(2, R) satisfy the Ramanujan conjecture: for N 1, one has X 2 1+ε |λj(n)| N , n≤N so one can see that very few of the λj(n) can possibly deviate from their expected bound. 2.1.9.1 The GL(2) Voronoi summation formula The additive twists of the Fourier coefficients of GL(2) automorphic forms exhibit a special type of symmetry owing to the invariance of the underlying automorphic object under the action of Γ. These are the Voronoi summation formulas. In the modern formulation, one can regard these as a consequence of the functional equation for the twisted L-functions attached to the given automorphic form. The presence of gamma factors in the functional equation results in the appearance of integral transforms on the dual side of the formula defined in terms of Bessel functions. Associated to each GL(2) automorphic form are two distinguished Bessel functions. There is a pleasant representation theoretic description of these that the reader may find in Cogdell-Piatetskii-Shapiro’s book [CPS90] in terms of the Whittaker and Kirillov models for GL(2, R). For Q a holomorphic modular form of even weight k, we have + k JQ (ξ) = 2πi Jk−1(ξ) − + and JQ (ξ) = 0. Furthermore, JQ has a representation as an integral of Marnes-Barnes type, with s+ k−1 s+ k+1 Z −s 2 2 1 ξ Γ 2 Γ 2 J +(ξ) = ds Q 1−s+ k−1 1−s+ k+1 2πi (σ) 2 2 2 Γ 2 Γ 2 for σ > 0. For Maass cusp forms uj, the Bessel functions are − + π J (ξ) = J (ξ) − J− (ξ) uj 2itj 2itj sinh(πtj) and J − (ξ) = 4ε cosh(πt )K (ξ). uj j j 2itj For both of these, there is the integral representation of Mellin-Barnes type − − Z −s s+itj s itj 1+s+itj 1+s itj 1 ξ Γ 2 Γ 2 Γ 2 Γ 2 J ± (ξ) = ds, uj 1−s+it 1−s−it 2−s+it 2−s−it 2πi (σ) 2 j j j j Γ 2 Γ 2 Γ 2 Γ 2 for σ > 0. Theorem 2.1.6 (Kowalski-Michel-VanderKam, [KMV02], Theorem A.4). Let N ∈ N and let f be either a holomorphic modular form of even weight k and level N, or a newform Maass cusp form of level N and 30 2.1. Automorphic forms for GL(2, R) 1 2 ∈ N Laplace eigenvalue 4 + t ; in either case let f have trivial nebentypus. Let a, c with (a, c) = 1 and set N ∈ ∞ ∞ N1 = (c,N) . Let φ C0 ((0, )) be a smooth weight function. Then there is a newform fN1 of level N1 and weight k such that X an η (N ) X X naN n λ (n)e φ(n) = f√ 1 λ (n)e ∓ 1 Φ± , (2.9) f fN1 f 2 c c N c c N1 n≥1 1 ± n≥1 ± where the integral transforms Φf are defined by Z ∞ ± ± √ Φf (x) = Jf (4π xy) φ(y)dy. (2.10) 0 The Voronoi summation formula is named for Georgii Voronoi, who derived a similarly shaped summation formula for the binary divisor function, X τ2(n) = 1, n1n2=n in his work on the divisor problem and Gauss circle problem. In the modern formulation, the binary divisor function appears in the Fourier coefficients of the degenerate Eisenstein series E ∂ 2(z) = [E(z, s)]s= 1 ∂s 2 √ √ X = y log y + 4 y τ2(n)K0(2πny) cos(2πnx). n≥1 Explicitly, the τ2-Voronoi summation formula is the following. ∈ ∞ ∞ Theorem 2.1.7 (Kowalski-Michel-VanderKam, [KMV02], Theorem A.4). Let φ C0 ((0, )) be a smooth weight function, let a, c ∈ N with (a, c) = 1, and let 1 γ = γ0 = lim ζ(s) − s→1 s − 1 denote the Euler-Mascheroni constant. Then we have X an X X an n τ (n)e φ(n) = ∆(c; φ) + τ (n)e ∓ Φ± , (2.11) 2 2 E2 2 c c c N1 n≥1 ± n≥1 where the integral transforms Φ± are defined by E2 Z ∞ √ Φ± (ξ) = J ± (4π xy) φ(y)dy, E2 E2 0 31 2.1. Automorphic forms for GL(2, R) − with Bessel kernels given by JE2,+(ξ) = 2πY0(ξ) and JE2,+(ξ) = 4K0(ξ), and where we have denoted Z √ ∞ y ∆(c; φ) = 2 log + γ φ(y)dy. 0 c Remark 2.1.12. Both of the above GL(2) Voronoi-type summations has an interpretation in terms of the functional equation of the Estermann zeta function or an analogue of the Estermann zeta function. One considers the L-functions twisted by additive characters X an λf (n)e c L(s, f ⊗ ξ a ) = , for <(s) > 1, c ns n≥1 an where ξ a (n) = e for fixed a, c ∈ N with (a, c) = 1. For example, when f = Q is a weight-k holomorphic c c Hecke eigencuspform for SL(2, Z), then the completed L-function ! ! − −s k 1 k+1 π s + 2 s + 2 Λ(s, Q ⊗ ξ a ) = Γ Γ L(s, Q ⊗ ξ a ) c c 2 2 c k satisfies the functional equation Λ(s, Q ⊗ ξ a ) = i Λ(1 − s, Q ⊗ ξ −a ). The situation is similar but slightly c c more complicated for the other cases. 2.1.9.2 The Hecke bound The Fourier coefficients of GL(2, R) automorphic forms satisfy a square-root cancellation when twisted by an additive character. This is refered to as the Hecke bound, and it is essentially the strongest type of bound one expects for Fourier coefficients. Proposition 2.1.8 (Godber, [God13], Theorem 1.2). Let f be an automorphic form for a discrete subgroup Γ ⊆ SL(2, R). Let α ∈ R and N 1. Then X 1 +ε 1 λf (n)e (αn) N 2 Q(f, 0) 4 . n≤N The implied constant depends only on ε. One method of proving Proposition 2.1.8 is to use GL(2) Voronoi summation, and this is the approach taken when proving the analogous bound for additive twists of GL(3) Fourier coefficients (the Miller bound, 1 which only yields 4 -cancellation)—see Section 2.2.3.2. We will also require the following form of the Hecke bound for twisted double sums of Fourier coefficients It actually holds for the double sum of any two additively twisted arithmetic functions that exhibit square-root cancellation. Later we refer to this bound as the double Hecke bound. 32 2.2. Automorphic forms for GL(3, R) Proposition 2.1.9 (Deshouillers-Iwaniec, [DI86], Lemma 11.7). Let f be a cuspidal automorphic form for a discrete subgroup Γ ⊆ SL(2, R). Let α, β ∈ R and N 1, and suppose that 0 < λ0 < λ. Then X X 1 1 2 +ε λf (n1)λf (n2)e (αn1 + βn2) (N1N2) Q(f, 0) 4 . n1≤N1 n2≤N2 n1 λ0< <λ n2 The implied constant depends only on ε. 2.1.10 An aside: the homogenous space setting It is natural to seek to generalize of the classical setting for automorphic forms to higher dimensions. One way of doing this is to consider the invariant functions under actions of subgroups of GL(n, R); this is cast in the language of homogenous spaces and group actions. The generalization of automorphic forms to GL(n, R) will seem more natural if we rephrase the discussion in Section 2.1.1 in terms of GL(2, R). One can identify the upper half place H with ∼ 2 × H = h = GL(2, R)/hO(2, R) · R i. By the Iwasawa decomposition of GL(2, R), we see that every z ∈ h2 can be written as ! ! ! y x 1 x y z = = , 1 1 1 ! y x so the above isomorphism identifies z = x + iy ∈ H with the element of h2 that has Iwasawa form . 1 This makes GL(2, R) ⟳ h2 a homogenous space, and the (left) GL(2, R)-invariant measure is still dµ(z) = dxdy R 2 y2 . The center of the universal enveloping algebra of gl(2, ), D , is generated by the Laplacian ∆; that is, D2 = C(∆). Working with SL(2, Z), the minimal parabolic subgroup of SL(2, Z) consists of upper triangular matrices ( ! ) 1 m Γ∞ = : m ∈ Z . 1 We shall only be working with GL(2, R) and GL(3, R), so we won’t present the general theory of automorphic forms on GL(n, R); for that we refer the reader to Goldfeld’s book [Gol06]. 2.2 Automorphic forms for GL(3, R) In this section we record some of the basic theory of automorphic forms for the group GL(3, R). We only provide enough to reach the results in Chapter 5, where we study moments of GL(2) × GL(3) L-functions. 33 2.2. Automorphic forms for GL(3, R) Absent is any spectral theory in the guise of the spectral theorem and the GL(3) Kuznetsov trace formula. We only describe the minimal parabolic Eisenstein series, which we require for a GL(3)-type Voronoi summation for the ternary divisor function τ3—see Section 2.2.4. The material presented in this section follows Goldfeld’s book [Gol06]. Analogous to the description of H in Section 2.1.10 in terms of homogenous spaces, the generalized 3-dimensional upper half plane is defined by h3 = GL(3, R)/hO(3, R) · R×i. Owing to the Iwasawa decomposition, each z ∈ h3 can be written in Iwasawa form as 1 x x y y 2 3 1 2 z = 1 x1 y1 , 1 1 3 with x1, x2, x3 ∈ R and y1, y2 > 0. Then GL(3, R) ⟳ h , and the (left) GL(3, R)-invariant Haar measure has the explicit formula dx1dx2dx3dy1dy2 dµ(z) = 3 . (y1y2) While in the two-dimensional setting D2 is generated by a single operator, namely the hyperbolic Laplacian 2 ∂2 ∂2 R 3 ∆ = y ∂x2 + ∂y2 , the center of the universal enveloping algebra of gl(3, ), denoted by D , is generated by two invariant differential operators, 2 2 2 2 2 2 2 2 ∂ 2 ∂ − ∂ 2 2 2 ∂ 2 ∂ 2 ∂ 2 ∂ ∆1 = y1 2 + y2 2 y1y2 + y1 x2 + y2 2 + y1 2 + y2 2 + 2y1x1 ∂y1 ∂y2 ∂y1∂y2 ∂x3 ∂x2 ∂x1 ∂x2∂x3 and 3 3 3 3 − 2 ∂ 2 ∂ − 3 2 ∂ 2 ∂ ∆2 = y1y2 2 + y1y2 2 y1y2 2 + y1y2 2 ∂y1∂y2 ∂y1∂y2 ∂x3∂y1 ∂x1∂y1 3 3 3 3 − 2 ∂ 2 − 2 2 ∂ − 2 ∂ 2 2 ∂ 2y1y2x1 + y2 x1 y1y2 2 y1y2 2 + 2y1y2 ∂x2∂x3∂y2 ∂x3∂y2 ∂x2∂y2 ∂x2∂x1∂x3 3 2 2 2 2 2 2 2 2 ∂ 2 ∂ − 2 ∂ 2 ∂ 2 2 2 ∂ 2 ∂ − 2 ∂ + 2y1y2x1 2 + y1 2 y2 2 + 2y1x1 + x1 + y2 y1 2 + y1 2 y2 2 . ∂x1∂x3 ∂y1 ∂y2 ∂x2∂x3 ∂x3 ∂x2 ∂x1 The simplest eigenfunctions of D3 are the power functions defined by ν1+2ν2 2ν1+ν2 Iν (z) = I(ν1,ν2)(z) = y1 y2 , 2 for a tuple ν = (ν1, ν2) ∈ C . We have 2 2 ∆1Iν (z) = (ν1 + 2ν2) − (ν1 + 2ν2)(2ν1 + ν1) + (2ν1 + ν1) Iν (z) = λν (∆1)Iν (z) 34 2.2. Automorphic forms for GL(3, R) and ∆2Iν (z) = λν (∆2)Iν (z), 3 writing λν (∆2) for the corresponding polynomial in ν1, ν2. We say that Iν (z) is an eigenfunction of D of type ν = (ν1, ν2). 2.2.1 The Maass cusp forms for SL(3, Z) Z ∈ The Maass cusp forms for SL(3, ) of type ν = (ν1, ν2) are those (non-zero) smooth functions ψ 2 3 L SL(3, Z) \ h that are eigenfunctions of ∆1 and ∆2 with eigenvalues λν (∆1) and λν (∆2), respectively, satisfying a certain decay condition. Z Definition 2.2.1. AMaass cusp form for SL(3, ) of type ν = (ν1, ν2) is a non-zero smooth function 2 3 ψ ∈ L SL(3, Z) \ h such that ∆1ψ = λν (∆1)ψ and ∆2ψ = λν (∆2), and which satisfies Z ψ(uz)du = 0 (SL(3,Z)∩U)\U for all subgroups U of SL(2, R) consisting of upper triangular matrices of the form I ∗ r1 Ir2 Ir3 with r1 + r2 + r3 = 3. A Maass cusp form ψ of type ν has the Fourier-Whittaker expansion | | ! X X X m1m2 Aψ(m1, m2) γ ψ(z) = WJacquet m1 z, ν, φ1, m2 , |m m | |m2| ∈ Z \ Z ≥ ̸ 1 2 1 γ U2( ) SL(2, ) m1 1 m2=0 1 (2.12) × where Un(G) denotes the n n upper triangular matrices with 1 on the diagonal and entries in G, φε1,ε2 : U3(R) → C denotes the character of U3(R) defined by 1 u u 1 3 ∈ R φε1,ε2 1 u2 = e (ε1u1 + ε2u2) , with ε1, ε2 , 1 and WJacquet (z, ν, φε1,ε2 ) is the Jacquet-Whittaker function Z Z Z · · WJacquet (z, ν, φε1,ε2 ) Iν (w3 u z)φε1,ε2 (u)du1du2du3, R R R 35 2.2. Automorphic forms for GL(3, R) −1 with w3 = 1 . The Fourier coefficients Aψ(m1, m2) are normalized such that A(m1, m2) = 1 O(|m1m2|). In what follows the Langlands parameters, or spectral parameters, for a Maass cusp form ψ of type ν will be denoted by α = −ν1 − 2ν2 + 1, β = −ν1 + ν2 and γ = 2ν1 + ν2 − 1. ∗ ∗ Associated to ψ is a dual Maass cusp form ψ of type (ν2, ν1). Furthermore, the Fourier coefficients of ψ are given by Aψ∗ (m1, m2) = Aψ(m2, m1), and the Langlands parameters are ∗ ∗ ∗ α = ν2 + 2ν1 − 1, β = −ν2 + ν1, and γ = 1 − 2ν2 − ν1. 2.2.2 The Hecke operators and Hecke-Maass cusp forms for SL(3, Z) Z Z There is an analogous theory of Hecke operators for SL(3, ) as in the SL(2, ) setting. The Hecke operators 2 3 2 3 Tn : L SL(3, Z) \ h → L SL(3, Z) \ h are defined by a b1 c1 1 X T f(z) = f b c z , n n 2 abc=n 0≤c1,c2 for n ∈ N. For diagonal matrices δ δ 1 2 δ = δ1 , 1 2 3 2 3 where δ1, δ2 = 1, we have the involutions Tδ : L SL(3, Z) \ h → L SL(3, Z) \ h given by Tδψ(z) = ψ(δzδ). These all commute with ∆1 and ∆2, and in fact, for Maass cusp forms, one has Tδψ = ψ for each Tδ; that is, all Maass cusp forms for SL(3, Z) are even. The Hecke operators {Tn} comprise a family of commuting normal operators, and they also commute { } with all the Tδ and ∆1 and ∆2. It follows from the spectral theorem we may choose a basis ψj for 2 3 L SL(3, Z) \ h consisting of simultaneous eigenfunctions of all the Tn, Tδ, ∆1, and ∆2; accordingly, the ψj are referred to as Hecke-Maass cusp forms. ∈ N We normalize these such that Aψj (1, 1) = 1, and we have Tnψj = Aψj (n, 1) for all n . Furthermore, 36 2.2. Automorphic forms for GL(3, R) ′ ′ ∈ Z the Fourier coefficients satisfy the following Hecke relations: for m1, m2, m1, m2, n , ′ ′ ′ ′ Aψj (m1m1, m2m2) = Aψj (m1, m2)Aψj (m1, m2), ′ ′ provided (m1m1, m2m2) = 1, X d0m1 d1m2 Aψj (n, 1)Aψj (m1, m2) = Aψj , . d1 d2 d0d1d2=n d1|m1 d2|m2 2.2.3 Bounds for Fourier coefficients and twists of Fourier coefficients The Fourier coefficients of an SL(3, Z) Maass cusp form ψ are predicted to satisfy the Ramanujan-Petersson ε conjecture as well; namely, one expects that |Aψ(m, n)| (mn) . The current best bound towards this is 5 +ε |Aψ(m, n)| (mn) 14 , due to Luo-Rudnick-Sarnak ([LRS99])3. Meanwhile, on average, one has the Rankin-Selberg bound X 2 |Aψ(m, n)| ψ N. m2n≤N By the Cauchy-Schwarz, it follows that X X 2 |Aψ(m, n)| ψ N|m|. m≤N n≤N 2.2.3.1 The GL(3) Voronoi summation formula As in the GL(2) setting, additive twists of the Fourier coefficients of ψ an SL(3, Z) automorphic form satisfy a Voronoi summation formula owing to the invariance of ψ under the action of SL(3, Z). The proof amounts to the functional equation for a twisted L-function, so that gamma factors appear on the dual side. These give rise to an integral transform similar to the classical Hankel-type transforms defined in terms of Bessel functions in the GL(2) setting. As described in [Qi15], an analogously to the GL(2) situation, associated to each ψ of type (ν1, ν2) outlined above are two GL(3) Bessel functions. For ξ > 0 they are defined by the Mellin-Barnes integrals Z s−α+δ s−β+δ s−γ+δ X δ −s Γ Γ Γ ± ( i) ξ 2 2 2 Jψ (ξ) = ds, (2.13) 4πi C 2 1−s+α+δ 1−s+β+δ 1−s+γ+δ δ=0,1 Γ 2 Γ 2 Γ 2 where C is a curved contour avoiding the poles of the gamma functions described in [Qi15]. Note that the Langlands parameters appearing in the denominator of (2.13) are the Langlands parameters of ψ∗, so the 3 2 7 +ε For symmetric square lifts ψ = sym uj , where uj is a Maass cusp forms for SL(2, Z), the bound is improved to (mn) 32 by the result of Kim-Sarnak ([Kim03]). 37 2.2. Automorphic forms for GL(3, R) ratio of gamma functions is the ratio of gamma factors of the associated L-function L(s, ψ). In [Qi15], Qi shows that the GL(3) Bessel functions have the asymptotic expansion − ± e (3ξ) KX1 B J ±(ξ) = ψ,k + E±(ξ) + O ξ−K , (2.14) ψ ξ ξk ψ K,ψ k=0 ∈ Z for K ≥0 and ξ 1 and where √ −3 3πξ (j) e E± (ξ) . ψ ψ,j ξ The Fourier coefficients of ψ satisfy the following GL(3) Voronoi summation formula. Theorem 2.2.1 (Li, [Li11]). Let ψ be an SL(3, Z) Maass cusp form of type with Fourier coefficients denoted ∈ ∞ ∞ ∈ Z by Aψ(m, n). Let φ Cc ((0, )). Let a, c, m with (a, c) = 1. Then we have X X X X 2 an 1 n1 cm ± n2n1 ∗ Aψ(m, n)e φ(n) = Aψ (n1, n2)S am, n2; Φψ∗ 3 , (2.15) c c ± cm n1 c m n≥1 n1|cm n2≥1 where the integral transform is defined by Z ∞ ± ∓ 1 3 Φψ∗ (x) = φ(y)Jψ∗ 2π (xy) dy. (2.16) 0 2.2.3.2 The Miller bound for additive twists of Fourier coefficients While the Fourier coefficients of GL(2, R) automorphic forms satisfy the square-root cancellation of the Hecke bound, the best known bound for additive twists of Fourier coefficients of SL(3, Z) automorphic forms 1 is weaker; one only has a 4 -saving bound. The first nontrivial bound of this sort for GL(3) was achieved by Miller ([MS06]), who used GL(3) Voronoi summation. The following version is due to Xiannan Li ([Li]). Proposition 2.2.2 (Li, [Li]). Let ψ be a (tempered) cusp form for SL(3, Z) of type (ν1, ν2). Let α ∈ R and N 1. Then X 3 +ε D Aψ(1, n)e (αn) N 4 Q(ψ, 0) , n≤N 1 5 where the exponent can be taken to be D = 4 assuming the Ramanujan conjecture and D = 12 uncondition- ally. The implied constant depends only on ε. 38 2.2. Automorphic forms for GL(3, R) 2.2.4 The minimal parabolic Eisenstein series for SL(3, Z) and Voronoi summa- tion for the ternary divisor function In this section we provide a statement of a GL(3)-type Voronoi summation formula for the ternary divisor function τ3 that was obtained in [Li14]. This is best viewed as a consequence of the symmetries of a degenerate minimal parabolic Eisenstein series for SL(3, Z). The minimal parabolic subgroup Γ∞ of SL(3, Z) consists of the upper triangular matrices 1 m2 m3 Γ∞ = 1 m : m1, m2, m3 ∈ Z . 1 1 < < 2 ∈ 3 For (ν1), (ν2) > 3 and z h , the minimal parabolic Eisenstein series E(z, ν1, ν2) is constructed by the same method as are the GL(2) Eisenstein series; namely, by periodizing the power functions Iν (z). Explicitly, they are defined by X 2ν1+ν2 ν1+2ν2 E(z, ν1, ν2) = (y1(γz)) (y2(γz)) . γ∈Γ∞\SL(3,Z) One then has the completed Eisenstein series 2 2 Eb(z, ν , ν ) = ν − ν − ζ(3ν )ζ(3ν )E(z, ν , ν ), 1 2 1 3 2 3 1 2 1 2 which extends to an entire function of ν1, ν2 ∈ C, and which has a Fourier expansion where the Fourier coefficients for the principal part4 are defined by X X 1−3ν1 2−3ν1−3ν2 σ1−3ν1,2−3ν1−3ν2 (n, m) = d1 d2 . | | m d1 m d2 d d >0 1 1 d2>0 (d2,n)=1 A relationship similar to (2.11) involving the degenerate GL(2, R) Eisenstein series and the binary divisor function τ2(n) holds for the ternary divisor function X τ3(n) = 1, n1n2n3=n n1,n2,n3∈N which forms the Fourier coefficients of a certain derivative of the minimal parabolic Eisenstein series for SL(3, Z); namely, the degenerate Eisenstein series is ∂3 E3(z) = [E(z, ν1, ν2)] 1 . 2 ν1=ν2= 3 ∂ν1∂ν2 Furthermore, one can obtain a Voronoi-type summation formula for the ternary divisor function τ3(n) from 4There are several other complicated terms that arise with “Fourier coefficients” defined in terms of the binary divisor function. We will not reproduce them here: see [Li14]. 39 2.2. Automorphic forms for GL(3, R) the functional equation for additive twists of E(z, ν1, ν2); see [Li14] for a full exposition. We will need this GL(3) Voronoi-type summation formula in Chapter 6, where we derive an upper bound for the sixth moment of L(sj, uj), which is essentially L(sj, uj ⊗ E3). Specifically, one has the following. ∈ ∞ ∞ Theorem 2.2.3 (Li, [Li14]). Let φ(x) Cc (0, ) be a smooth compactly supported weight function, and let h, ℓ, c ∈ N with hℓ ≡ 1 (mod c). Denote X X σ0,0(m, n) = 1 | | n d1 n d2 d d >0 1 1 d2>0 (d2,m)=1 and let 1 γ = γ0 = lim ζ(s) − s→1 s − 1 be the Euler-Mascheroni constant and − d − 1 γ1 = ζ(s) − ds s 1 s=1 the first Stieltjes constant5. Let Φ± denote the GL(3) Bessel function defined in (2.13) with Langlands E3 parameters α = β = γ = 0. Then τ3(n) satisfies the Voronoi-type summation X mℓ τ (m)e φ(m) 3 c m≥1 c X X X 1 X X n c mn2 = σ , m S ηm, h; Φ± 3 0,0 E3 3 2 2π ± nm n n1n2 n c η= 1 n|c m≥1 n1|n n2| n1 + ∆0(h, c; φ) + ∆1(h, c; φ) + ∆2(h, c; φ), (2.17) where we have denoted 1 X nτ (n) n ∆ (h, c; φ) = 2 S 0, h; T (n, c)ψe(1), (2.18) 0 2 c2 c 3,0 n|c with 2 9 2 2 T3,0(n, c) = (log n) − 5(log c)(log n) + (log c) + 3γ + 7γ log n − 9γ log c − 3γ1 2 1 X 3 + log(d) (log(nc) − 5γ) − (log d)2 , τ2(n) 2 d|n 1 X nτ (n) n ∆ (h, c; φ) = 2 S 0, h; T (n, c)ψe′(1), (2.19) 1 2 c2 c 3,1 n|c 40 2.2. Automorphic forms for GL(3, R) with 5 1 X T3,1(n, c) = log n − 3 log c + 3γ − log d, 3 3τ2(n) d|n and 1 X nτ (n) n ∆ (h, c; φ) = 2 S 0, h; ψe′′(1). (2.20) 2 4 c2 c n|c Remark 2.2.1. Before continuing, we remark that there is a much more general Voronoi-type summation formula for the n-ary divisor functions. The general shape of the formula is similar, but the proof is via the functional equation for an analogue of the Estermann zeta function, rather than the automorphy of an explicit GL(2, R) automorphic form. See [Ivi97] for a full treatment. 5 b The Stieltjes constants arise naturally here because of the presence of the zeta function in the definition of the E(z, ν1, ν2), since they appear in the Laurent series expansion of ζ(s) at s = 1. 41 Chapter 3 Preliminaries on automorphic L-functions Contents 3.1 Automorphic L-functions ...... 42 3.1.1 The approximate functional equation ...... 44 3.1.2 The convexity bound ...... 47 3.2 On conductor-dropping behavior of L-functions in the spectral aspect .... 48 The main tools we will use to study automorphic forms are their associated L-functions. In general, all L-functions we study have the same basic properties: a Dirichlet series definition in a half-plane <(s) > 1, a functional equation reflecting s → 1 − s giving the L-function a meromorphic continuation to C, an Euler product, etc. There is a general formal framework of L-functions, an outline of which we provide here, but we will leave specific properties of the L-functions we study in each chapter to the preliminaries section of each. We largely follow the presentation in Iwaniec-Kowalski’s book [IK04]. Since later we study the behavior of L-functions exhibiting conductor-dropping, we will describe this phenomenon and the literature surrounding it in greater detail in Section 3.2. 3.1 Automorphic L-functions Definition 3.1.1. We say that a function L(s, f), where f is some sort of automorphic object and s ∈ C (possibly not at s = 0 or s = 1), is an L-function if it satisfies the following. N 1. There is a sequence of Dirichlet series coefficients (λf (n)) ∈ C such that for <(s) > 1, one has an absolutely convergent Dirichlet series X λ (n) L(s, f) = f . ns n≥1 42 3.1. Automorphic L-functions 2. Furthermore, we assume that for <(s) > 1 the Dirichlet series has an Euler product of degree d, − − Y α (p) 1 α (p) 1 L(s, f) = 1 − 1 ··· 1 − d ps ps p where the p are prime (possibly excluding some finite subset) and local parameters αi(p) ∈ C and |αi(p) < p for all 1 ≤ i ≤ d. We say that d is the degree of L(s, f), or L(s, f) is an L-function of degree d. That L(s, f) has an Euler product means the Dirichlet series coefficients are multiplicative. From the adelic point-of-view, it makes sense to occasionally denote the local L-functions by − − α (p) 1 α (p) 1 L (s, f) = 1 − 1 ··· 1 − d . p ps ps 3. There are gamma factors associated to L(s, f), Yd − ds s + κj γ(s, f) = π 2 Γ , 2 j=1 where κj ∈ C with <(κj) > −1 are the local parameters of L(s, f) at ∞, or the archimedean param- eters of L(s, f). For this reason, sometimes the adelic notation L∞(s, f) = γ(s, f) is used. The κj are either real or come in complex conjugate pairs. 4. The (arithmetic) conductor Q(f) ∈ Z is such that αi(p) =6 0 for p 6 |Q(f) and 1 ≤ i ≤ d, and we say such a prime p is unramified. 5. The completed L-function s Λ(s, f) = Q(f) 2 γ(s, f)L(s, f) satisfies a functional equation of the form Λ(s, f) = ε(f)Λ(1 − s, f ∗), which extends L(s, f) to a meromorphic function on C with at most poles at s = 0 and s = 1. In the above, |ε(f)| = 1 is the root number of L(s, f). The dual object f ∗ is another automorphic object whose L-function L(s, f ∗) has the same degree, conductor, and gamma factors as L(s, f), and has ∗ Dirichlet series coefficients λf ∗ (n) = λf (n). In the case where L(s, f) = L(s, f ), we say that the underlying object is self-dual. Typically we are interested of the behavior of L-functions in a particular family. In this case it is useful to describe the behavior of the L-functions in question in terms of their analytic conductor, defined to be Yd Q(s, f) = Q(f) (|s + κj| + 3) = Q(f)Q∞(s, f). j=1 43 3.1. Automorphic L-functions This is essentially just the bound that comes from applying Stirling’s approximation, √ it σ− 1 − πt t 1 Γ(σ + it) = 2πt 2 e 2 1 + O for t > 0, e σ t to the gamma factors. If one prefers, one can simply define the analytic conductor to be γ(s, f). Example 3.1.1. An example of a family of L-functions we will use throughout this section are the L- functions associated to a basis {uj} of Hecke-Maass cusp forms for SL(2, Z). We define the L-function L(s, uj) by the Dirichlet series X λ (n) L(s, u ) = j , for <(s) > 1, j ns n≥1 and these have the degree two Euler product Y −1 −1 Y −1 − αj(p) − βj(p) − λj(p) 1 L(s, uj) = 1 s 1 s = 1 s + 2s . p p p p p p prime prime The gamma factors are − Γ s itj Γ s+itj if u is even, −s 2 2 j γ(s, uj) = π s+1−itj s+1+itj Γ 2 Γ 2 if uj is odd, with the archimedean parameters being itj if uj is even and 1 itj if uj is odd. The completed L-function Λ(s, uj) = γ(s, uj)L(s, uj) is entire and satisfies the functional equation Λ(s, uj) = εjΛ(1 − s, uj), where εj = 1 if uj is even and εj = −1 if uj is odd; in both cases uj is self-dual. The conductor is Q(uj) = 1. 3.1.1 The approximate functional equation Let L(s, f) be an L-function. The standard approximate functional equation allows one to express L(s, f) as the sum of two weighted Dirichlet series for s in the critical strip 0 < <(s) < 1, where the L(s, f) lacks a Dirichlet series representation. Theorem 3.1.1 (Iwaniec-Kowalski, [IK04], Theorem 5.3). Let L(s, f) be an L-function. Let G(u) be any function which is holomorphic and bounded in the strip −4 < <(u) < 4, even, and normalized by G(0) = 1. Let X > 0. Then for s in the strip 0 ≤ σ ≤ 1 we have ! ! X X λf (n) n λf (n) nX L(s, f) = V p + ε(s, f) V − p + R (3.1) ns s n1−s 1 s f n≥1 X Q(f) n≥1 Q(f) 44 3.1. Automorphic L-functions where Vs(y) is a smooth function defined by Z 1 −u γ(s + u, f) du Vs(y) = y G(u) 2πi (3) γ(s, f) u and − 1 −s γ(1 s, f) ε(s, f) = ε(f)Q(f) 2 . γ(s, f) The last term is Rf = 0 if Λ(s, f) is entire, otherwise Λ(s + u, f) G(u) u Rf = Res + Res X . u=1−s u=−s Q(f)s/2γ(s, f) u The general mantra is that the approximate functional equation splits L(s, f) into two sums of length 1 Q(s, f) 2 . One can see that this is the case by shifting the contours defining Vs(y) very far to the right, where they satisfy !−A y Vs(y) 1 + p Q∞(s, f) and !α! y Vs(y) = 1 + O p , Q∞(s, f) if <(s + κj) ≥ 3α > 0 and where A > 0 is arbitrary. The weight functions appearing in the sums are defined in terms of the gamma factors of L(s, f). Often we will be interested in obtaining an asymptotic expansion of a moment of L-functions in some family, and to extract the main term of the asymptotic expansion we will need to remove the dependence of the weight function on the parameters in question. To do so, we use an unbalanced approximate functional equation that contains all the parameters of interest in only one of the weight functions. This provides us with the additional flexibility needed to achieve power-saving error terms in the asymptotic expansions. The derivation is similar for the other L-functions we will study, but since the proofs are all similar, we will only provide one for the example below. 1 Example 3.1.2 (Unbalanced approximate functional equation for L 2 + itj, uj ). Let uj be an even Hecke- 1 2 1 Maass cusp form with Laplace eigenvalue λj = 4 + tj . Then at the special point sj = 2 + itj (where the analytic conductor drops), for any N > 0, L(sj, uj) can be expanded as X λ (n) n X λ (n) j 2itj j L(sj, uj) = H + π Pt (πnN) , (3.2) nsj N nsj j n≥1 n≥1 45 3.1. Automorphic L-functions where Z s+ 1 Γ 2 1 2 −s ds H(ξ) = 1 ξ , (3.3) 2πi (σ) Γ 4 s and 1 1 − Z s+ 2 2 +s 2itj 1 Γ 2 Γ 2 ds P (ξ) = ξ−s , (3.4) tj 1 1 −s+2it 2πi (σ) Γ 2 j s 4 Γ 2 1 with σ > 2 . In this form, the weight function H(ξ) will not participate if we apply the Kuznetsov trace formula. Furthermore, we are free to choose N however we wish in order to compensate for whatever crude estimates to the integral transforms that appear after applying Kuznetsov to the dual sum, where the weight function does participate. To prove (3.2), as typical, we evaluate the following integral in two different ways: Z s+ 1 Γ 2 1 2 ds I(σ) = 1 L(s + sj, uj) . 2πi (σ) Γ 4 s 1 For σ > 2 , we can expand the L-function as a Dirichlet series, yielding Z s+ 1 X Γ 2 λj(n) 1 2 ds I(σ) = L(s + sj, uj) , nsj 2πi Γ 1 s n≥1 (σ) 4 whence we define H(ξ) as in (3.3). Meanwhile, by the functional equation we have 1 − 1 1 − − s+ 2 s+ 2 2 s 2itj Γ 2 Γ 2 Γ 2 L(s + s , u ) = πs+2itj L(s − s, u ). 1 j j 1 1 +s+2it j j Γ Γ 2 j 4 4 Γ 2 Since L(s, uj) is entire, shifting the contour to <(s) = −σ we cross a simple pole at s = 0, so that Z s+ 1 Γ 2 1 2 ds I(σ) = L(sj, uj) + 1 L(s + sj, uj) 2πi − Γ s ( σ) 4 − 1 1 − − Z s+ 2 2 s 2itj 1 Γ 2 Γ 2 ds s+2itj − = L(sj, uj) + π 1 L(sj s, uj) 1 +s+2itj 2πi (−σ) Γ 4 Γ 2 s 2 1 1 Z s+ +s−2itj Γ 2 Γ 2 1 − 2 2 ds = L(s , u ) − π2itj π s L(s + s, u ) , j j 1 1 −s+2it j j 2πi (σ) Γ 2 j s 4 Γ 2 46 3.1. Automorphic L-functions after changing variables s 7→ −s. Expanding the L-function appearing in the final integral, we see that 1 1 Z s+ +s−2itj Γ 2 Γ 2 1 − 2 2 ds L(s , u ) = I(σ) + π2itj π s L(s + s, u ) j j 1 1 −s+2it j j 2πi (σ) Γ 2 j s 4 Γ 2 X λ (n) n X λ (n) j 2itj j = H + π Pt (πnN) , nsj N nsj j n≥1 n≥1 having defined Ptj (ξ) as in (3.4). 3.1.2 The convexity bound Apart from questions about the location of the nontrivial zeros of an L-function, we are often interested < 1 its magnitude along the critical line (s) = 2 . It is often the case that other problems in analytic number theory can be reduced to a statement about the growth of an L-function at a point on the critical line as a parameter grows. The trivial estimate, or convexity bound (or convex bound), for L(s, f) at a point < 1 with (s) = 2 follows from the Phragmen-Lindelöf convexity principle and standard bounds for L-functions in vertical strips. For any ε > 0 and −ε ≤ <(s) ≤ 1 + ε, one has 1−σ +ε L(s, f) Q(s, f) 2 , and in particular, along the critical line one has 1 +ε 1 1 4 L + it, f Q + it, f . (3.5) 2 2 In fact, the ε appearing in the exponent can be removed along the critical line by the method of Heath-Brown, see [HB08]. The bound (3.5) is quite weak and on its own it is rarely sufficient for applications. The Lindelöf hypothesis predicts that the best bound that can be achieved is essentially logarithmic, with the conjectured 1 1 ε L 2 + it, f Q 2 + it, f . Any bound better than (3.5) is a subconvexity bound or subconvex bound. Example 3.1.3 (The convexity bound and subconvexity bound for L(s, uj)). In the case where f = uj, the convexity bound is 1 1 +ε L + it, u ((|t + t | + 1)(|t − t | + 1)) 4 . 2 j j j 1 1 +ε For fixed values of t, as t → ∞, one therefore has L + it, u t 2 . Meanwhile, for fixed t , as t → ∞ j 2 j t j j 1 1 2 +ε the bound is L 2 + it, uj tj t . Jutila-Motohashi ([JM05]) achieved a uniform subconvexity bound (of “Weyl bound” strength), with 1 1 +ε L + it, u (|t − t| + 1) 3 . 2 j j Notably, this bound is actually weaker than the convexity bound when the spectral parameter and height on the critical line are correlated, where the analytic conductor drops: one of the gamma factors is fixed at 47 3.2. On conductor-dropping behavior of L-functions in the spectral aspect 1 the special point sj = 2 + itj as tj varies, so that the convexity bound is instead 1 1 +ε L + it , u t 4 . 2 j j j At the special point, the L-function behaves like it is associated to a GL(1) object, rather than a GL(2) object. We discuss this phenomenon more generally in the next section. 3.2 On conductor-dropping behavior of L-functions in the spectral aspect As mentioned in Section 3.1, it is possible to vary the parameters of an L-function in such a way that some of its gamma factors are fixed or essentially fixed, causing an abnormal drop in the magnitude of its analytic conductor. Consequently, as the parameter of interest varies, the L-function artificially behaves like it has a lower degree. This conductor-dropping behavior can be observed in many aspects, but here we will focus on the spectral aspect. The simplest setting where this conductor-dropping behavior can be observed is with L-functions defined Z 1 2 for Eisenstein series: for example, attached to an SL(2, ) Eisenstein series Et with eigenvalue 4 + t is the degree 2 L-function X τ (n) L(s, E ) = it , for <(s) > 1, t ns n≥1 which factors as the product of two GL(1) L-functions, with L(s, Et) = ζ(s − it)ζ(s + it). (3.6) 1 1 1 1 Thus, at the point s = 2 + it, we have L 2 + it, Et = ζ 2 ζ 2 + 2it and the L-function literally behaves as a GL(1) L-function as the spectral parameter t varies. This phenomenon is observed in general as the degree increases; for example, one has L(s, Et ⊗ Q) = L(s − it, Q)L(s + it, Q) for Q an automorphic form for GL(2, R). When considering the features of the L-function in question at the special conductor-dropping point (often referred to as the special point or conductor-dropping point or spectral point), the lengths of sums, convexity bounds, etc are changed accordingly. Below is a table summarizing the sorts of changes that may occur for various families of L-functions. 48 3.2. On conductor-dropping behavior of L-functions in the spectral aspect Convex Special pt. Sp. pt. L-function Behavior bound behavior convex 1 +ε 1 +ε L (s, uj) GL(2) tj 2 GL(1) tj 4 1 1 ⊗ × 2 +ε 4 +ε L (s, uj χ) GL(2) GL(1) tj GL(1) tj 2 1 +ε 1 +ε L s, sym uj GL(3) tj 2 GL(2) tj 2 2 1 +ε 1 +ε L s, sym uj ⊗ χ GL(3) × GL(1) tj 2 GL(2) tj 2 1+ε 1 +ε L (s, uj ⊗ Q) GL(2) × GL(2) tj GL(2) tj 2 3 3 ⊗ × 2 +ε 4 +ε L (s, uj ψ) GL(2) GL(3) tj GL(3) tj 1 2 ⊗ × 2 +ε × 1+ε L s, sym uj f GL(3) GL(2) tj GL(2) GL(2) tj 2 3 +ε 3 +ε L s, sym uj ⊗ ψ GL(3) × GL(3) tj 2 GL(2) × GL(3) tj 2 Table 3.1: A table summarizing changes to L-functions when conductor-dropping occurs in the spectral aspect for various GL(2)-associated L-functions. Here uj is a Hecke-Maass cusp form for SL(2, Z), χ is a primitive Dirichlet character of modulus N, Q is a holomorphic Hecke eigencuspform for Γ0(N), and ψ is a Maass cusp form for SL(3, Z). It is interesting that conductor-dropping has actually been used to great effect, but only ever in situations where there is a convenient factorization involving Eisenstein series like (3.6). For example, Luo-Sarnak ([LS95]) used the factorization to reduce quantum unique ergodicity (QUE) for Eisenstein series on the modular surface SL(2, Z) \ H to subconvexity bounds of any strength for the L-functions L(s, uj). Similarly, such a factorization was used to prove the analogous equidistribution results for the Picard modular surface Z \ H3 < 1 Z SL(2, [i]) by Koyama ([Koy00]), and for the residues in (s) < 2 of Eisenstein series for SL(2, ) by Petridis-Raulf-Risager ([PRR13]). Similarly, Young [You16] was able to prove QUE for Eisenstein series restricted to the geodesic in SL(2, Z) \ H connecting 0 to i∞ by effectively reducing estimates on shifted convolution sums of divisor functions to subconvexity bounds for GL(2) L-functions of any strength. We describe similar problems of interest in the Part IV. Unfortunately, thus far the subconvexity problem in the presence of conductor-dropping has proven to be very difficult and has resisted dedicated attacks focusing on this aspect. There is only one instance where subconvexity at the special point is known; the monumental work of Michel-Venkatesh ([MV10]) achieves a subconvexity bound for GL(2, R) L-functions that is uniform in all parameters of the analytic conductor, and implicitly yields a special point subconvexity bound. Subsequently the exponent-of-saving was made explicit in [WA18] by Wu. It is unclear what technique—since they make heavy use of measure rigidity results—are strictly needed if at all to achieve special point subconvexity when one ignores the other parameters, or whether the exponent-of-saving can be significantly improved in this situation. There is another type of conductor-dropping that can occur in the spectral aspect: when considering the Rankin-Selberg convolution L-function L(s, f ⊗ g) for two (tempered spherical) automorphic forms f and g for GL(n), the conductor drops if the spectral parameters of f and g are within a fixed range of each other. For this situation, Blomer ([Blo12a]) has established a strong second moment upper bound that is consistent 49 3.2. On conductor-dropping behavior of L-functions in the spectral aspect with the Lindelöf hypothesis. Namely, one has Z !1+ε X 2 1 ∗ L , f ⊗ g dspecµ . 2 ∥µf −ν∥≤1 ∥µf −µg ∥≤1 Interestingly, he only requires the use of “soft methods”: he makes no use of the trace formula, approximate functional equation, or Voronoi summation. 50 Chapter 4 Preliminaries on the spectral theory of hyperbolic Riemann surfaces and Teichmüller theory Contents 4.1 Riemann surfaces ...... 51 4.2 Some remarks on Teichmüller theory and hyperbolic 2-surfaces ...... 52 4.2.1 The Teichmüller space of a surface ...... 53 4.2.2 Holomorphic quadratic differentials ...... 54 4.2.3 Teichmüller’s theorem ...... 54 4.3 The Phillips-Sarnak deformation theory of discrete groups ...... 55 In this chapter we provide the necessary background on the Phillips-Sarnak deformation theory of discrete groups. As a practical matter, this requires us to make a brief digression to Teichmüller theory. However, our focus is on the spectral theory of the Riemann surfaces Γ \ H, so we will not cover any of the rich geometric \ H theory of T`eˇi`c‚hffl(Γ )—there is simply too much to cover. For the sake of providing the appropriate amount of context to the unfamiliar reader for the results described in the introduction, we will at least provide the highlights. The material found in this section primarily follows the book of Gardiner-Lakic ([GL00]). The main result we need is Teichmüller’s description of the quasiconformal mapping between two Riemann surfaces in M(S) in terms of holomorphic quadratic differentials, which explains the connection between the annihilation of cusp forms and the special values L(sj, uj ⊗ Q). 4.1 Riemann surfaces Let S be a differentiable manifold of two real dimensions. There are two ways to think define a C-structure (marking) on S. One could consider complex atlases {φα} defined on S: these are collections of charts 51 4.2. Some remarks on Teichmüller theory and hyperbolic 2-surfaces between open subsets of C and S with transition maps that are holomorphic. The atlas endows S with the structure of a Riemann surface. The other perspective is to consider maps to S from surfaces already endowed with C-structure. A C- structure on S is given by a diffeomorphism φ : R → S where R is a Riemann surface; in this case one refers to the structure as (φ, R), and one says that R is marked by S. One may then wish to study the various C-structures that can be assigned to S. Two complex structures ◦ −1 → (φ1,R1) and (φ2,R2) on S are conformally equivalent if φ2 φ1 : R1 R2 is isotopic to a conformal diffeomorphism. On an infinitesimal level, a map f being conformal means that it preserves circles: circles in the tangent space at z ∈ R1 are mapped to circles in the tangent space at f(z) ∈ R2. The celebrated uniformization theorem of Koebe classifies all Riemann surfaces up to conformal equiva- lence as belonging to one of five broad classes of surfaces of constant curvature: 1. Cˆ: the Riemann sphere: positive curvature 2. C: the complex plane: zero curvature 3. S1 × (0, 1): a complex cylinder: zero curvature 4. T2: a complex torus: zero curvature 5. Γ \ H: a finite volume hyperbolic 2-surface; Γ is a Fuchsian group: negative curvature Thus the C-structures that can be assigned to a surface S must come from the above classes. As a conse- quence, the C-structure on the surface is induced by a Riemannian metric with fundamental form ds2 = ρ2|dz|2 = ρ2 dx2 + dy2 , with ρ > 0. In order to study the possible C-structures, one considers maps between two structures R1 and R2 that are quasiconformal: the map must send infinitesimal circles in R1 to infinitesimal ellipses in R2 of bounded eccentricity. The relationship between the various complex structures on S is the subject of Teichmüller theory, and it turns out that the space of C-structures on S is endowed with a metric defined between in terms of quasiconformal mappings. We are interested in the final of the above classes, the surfaces of constant negative curvature. 4.2 Some remarks on Teichmüller theory and hyperbolic 2-surfaces As we have said, a quasiconformal map f sends infinitesimal circles to infinitesimal ellipses of bounded eccentricity. By the Cauchy-Riemann equations, f must essentially (to a first order approximation) satisfy the Beltrami equation fz = µfz for some constant µ. If µ = 0, then f is holomorphic. When one allows µ = µ(z) to be nonconstant, µ must be essentially bounded; that is, µ ∈ L∞(R). Analytically, quasiconformality is defined on charts as follows. 52 4.2. Some remarks on Teichmüller theory and hyperbolic 2-surfaces Definition 4.2.1. Let U, V ⊆ C. A homeomorphism f : U → V is quasiconformal if there is a constant k < 1 such that 1. f has locally integrable (distributional) derivatives fz and fz on U, 2. |fz| ≤ k|fz| almost everywhere. 1+k In this case, for the minimal such k we denote the dilation constant by K(f) = 1−k . The Beltrami fz (z) 1+|µ(z)| coefficient of f is given by µ(z) = , and dilation of f at z ∈ U is Kz(f) = . fz (z) 1−|µ(z)| As is suggested by the above, the quasiconformal homeomorphisms of a Riemann surface R are in bijection with the “Beltrami differentials” µ(z) on R, up to conformal equivalence. That is, the conformal classes of quasiconformal deformations are determined by a Beltrami differential. dz Definition 4.2.2 ([GL00]). A Beltrami differential µ(z) dz on a Riemann surface R is an assignment to α each chart zα on Uα of an L∞ C-valued function µ defined on zα(Uα) such that dzβ α β dzα µ (zα) = µ (zβ) . dzβ dzα 4.2.1 The Teichmüller space of a surface One defines quasiconformal deformations of a surface R up to conformal maps factoring through a homotopy through quasiconformal maps. Definition 4.2.3 ([GL00]). Let f1 : R → R1 and f2 : R → R2 be quasiconformal. Then f1 and f2 are Teichmüller equivalent if there is a conformal map g : R1 → R2 and a homotopy H : R × [0, 1] → R through quasiconformal self-maps gt : R → R such that 1. g0 = idR, −1 2. g1 = (f2) ◦ g ◦ f1, and 3. gt(p) = p for 0 ≤ t ≤ 1 and p ∈ ∂R. C The Teichmüller space, T`eˇi`c‚hffl(S), of a surface S admitting -structures consists of all (φ, R) marked by S up to Teichmüller equivalence. When we have a Riemann surface R in mind, we may also denote the Teichmüller surface by T`eˇi`c‚hffl(R). The space T`eˇi`c‚hffl(S) is a metric space where the metric is defined in the following way. Definition 4.2.4. Let (φ1,R1) and (φ2,R2) be two Riemann surfaces marked by S. Then the Teichmüller distance between R1 and R2 is defined by 1 d (R ,R ) = inf {log K(f)} . T`eˇi`c‚hffl 1 2 2 f:R1→R2 quasiconformal That is, d (R ,R ) is the infimum of the dilation of quasiconformal maps from R to R . T`eˇi`c‚hffl 1 2 1 2 The way that d is defined actually only yields a pseudometric on (S); by identifying Riemann T`eˇi`c‚hffl T`eˇi`c‚hffl surfaces marking S up to conformal equivalence it becomes a true metric. 53 4.2. Some remarks on Teichmüller theory and hyperbolic 2-surfaces 4.2.2 Holomorphic quadratic differentials The Beltrami coefficients form a Banach space dual to T`eˇi`c‚hffl(R). Definition 4.2.5. Let R be a Riemann surface with an atlas of charts {Aα}α. Locally, a holomorphic 2 quadratic differential φ(dz) is a collection of holomorphic maps {φα} such that for local coordinates z1 and z2, one has 2 dz2 φ1(z1) = φ2(z2) . (4.1) dz1 The integrable holomorphic quadratic differentials form a Banach space, which we denote by Q(R). This is a closed subspace Q(R) ⊆ L1(R) of the Banach space of integrable quadratic differentials on R. ∗ ∼ It is a standard fact that the dual space is L1(R) = L∞(R), where L∞(R) consists of essentially bounded (−1, 1) forms µ on R. This induces the isomorphism ∗ ∼ Q(R) = L∞(R)/N, where N consists of Beltrami differentials µ ∈ L∞(R) such that (φ, µ) = 0 for all φ ∈ Q(R). Furthermore, ∗ ∼ when Q(R) is finite-dimensional, one can show that Q(R) = T`eˇi`c‚hffl(R), so Q(R) can be regarded as the tangent space to T`eˇi`c‚hffl(R) at R. 4.2.3 Teichmüller’s theorem Teichmüller’s theorem provides the existence of a quasiconformal map f : R → R realizing d (R ,R ); 1 2 T`eˇi`c‚hffl 1 2 furthermore, f is uniquely determined by a (integrable) holomorphic quadratic differential of unit norm in Q(R1). Theorem 4.2.1. Let R1 and R2 be Riemann surfaces with finitely many cusps and such that Q(R1) and Q(R2) are finite-dimensional. Suppose that f : R1 → R2 is a quasiconformal map that is not conformal. Then in the Teichmüller equivalence class of f there is a map f0 : R1 → R2 such that 1 log K(f0) = d (R1,R2), 2 T`eˇi`c‚hffl |φ(z)| 1+k the Beltrami coefficient of f0 is given by µ(z) = k φ(z) , where 0 < k < 1, K(f0) = 1−k , and where 2 φ(z)(dz) is a holomorphic quadratic differential on R1 of unit norm. Put another way, the fundamental form of the Riemannian metric on the deformed surface defining its C-structure can be represented in terms of the original surface via ds2 = ρ(z)|dz + µ(z)dz|2 where ρ > 0. In the case where R = Γ \ H, where Γ is a Fuchsian group of the first kind, this allows us to describe the 54 4.3. The Phillips-Sarnak deformation theory of discrete groups condition (4.1) in terms of holomorphic modular forms of weight 4. Let φ = {φα}α ∈ Q(Γ \ H). Then the local holomorphic maps φα lift to a holomorphic function ψ on H, and by (4.1), ψ must satisfy the following: 1. The automorphy condition az + b − ψ(z) = ψ(γ(z)) (γ′(z))2 = ψ (cz + d) 4 cz + d ! a b for any γ = ∈ Γ c d R | | ∞ 2. φ is integrable: Γ\H ψ(z) dµ(z) < . That is, ψ must be a holomorphic modular cusp form of weight 4 for Γ! In the case where Γ is arithmetic, we have a basis for S4(Γ \ H) consisting of Hecke eigencuspforms. By Teichmüller’s theorem, any path in \ H \ H T`eˇi`c‚hffl(Γ ) starting at Γ can be described in terms of Hecke eigencuspforms for the initial surface. In terms of Beltrami differentials, the automorphy condition can be phrased as ! γ′(z) cz + d 2 µ(z) = µ(γ(z)) = µ(γ(z)) . γ′(z) cz + d In the next section we describe how this fact can be exploited to get a handle on the relative sizes of NΓ and MΓ for the deformed surface. The characterization of the quasiconformal deformations in terms of holomorphic quadratic differentials allows us to recover classical results in a simple way. The dimension of the space of holomorphic quadratic differentials is constant as one moves through the Teichmüller space of a given hyperbolic Riemann surface Γ \ H. From the description of the Beltrami differentials, one sees that \ H \ H − `d˚i‹mffl(Q(Γ )) = `d˚i‹mffl(S4(Γ )) = 3g 3 + k + ℓ, where g is the genus, k is the number of cusps, and ℓ is the number of elliptic fixed points. That 3g −3+k +ℓ parameters are necessary to describe all C-structures on a surface of genus g ≥ 2 has been known since Riemann. More can be said: it turns out that the signature of a finite-volume hyperbolic surface is an invariant up to quasiconformal deformation. Theorem 4.2.2. Let R1 = Γ1 \ H and R2 = Γ2 \ H be two finite-volume hyperbolic Riemann surfaces. If → ∈ Γ1 and Γ2 have the same signature, then there is a quasiconformal map f : R1 R2, and R2 T`eˇi`c‚hffl(R1). ∈ Conversely, if R2 T`eˇi`c‚hffl(R1), then R1 and R2 have the same signature. 4.3 The Phillips-Sarnak deformation theory of discrete groups In order to meaningfully study the distribution of the eigenvalues of the Laplacian ∆ on a noncompact surface Γ \ H, one has to consider the contributions from the discrete spectrum and continuous spectrum 55 4.3. The Phillips-Sarnak deformation theory of discrete groups simultaneously; again, Weyl’s law for these surfaces is \ H 2 NΓ(T ) + MΓ(T ) = ”vˆo˝l(Γ )T + O (T log T ) . The work of Colin de Verdière ([CdV82], [CdV83]) showed that for perturbations in a compact subset of 1 of congruence surfaces, all but finitely many discrete eigenvalues are destroyed. In order to study the spectrum of ∆ on generic discrete subgroups Γ ⊆ SL(2, R), Phillips-Sarnak ([PS85b], [PS85a]) restricted \ H ≤ ≤ \ H themselves to considering paths Γε , where 0 ε 1, in T`eˇi`c‚hffl(Γ ) beginning at Γ0 = Γ; by the \ H uniformization theorem, a deformed surface along such a path in T`eˇi`c‚hffl(Γ ) corresponds to a subgroup Γε. The behavior of a cusp form under quasiconformal deformations of the underlying surface is real analytic in ε. Phillips-Sarnak considered deformations of the metric given by 1 ds2 = |dz + εµ(z)dz|2, ε y2 where µ(z) is a Beltrami differential.2 As we have seen in the previous section, the Beltrami differential can be taken to be µ(z) = y2Q(z), where Q is a weight k holomorphic modular form for Γ. The corresponding Laplacian for the surface essentially3 has the “perturbation series” ∆(ε) = ∆ + ε∆(1) + ε2∆(2) + ··· , where the ∆(i) are certain operators defined on a Hilbert space. In particular, the first of these has the explicit form ! ∂ 2 ∆(1) = L = −8< Q y2 . Q ∂z Through a series of approximations away from cusps, they then reduce the question of whether a cusp form uj remains in the discrete spectrum under deformation to the vanishing of the inner product Z (1) · 1 (1) 1 dxdy ∆ uj,Ea , + itj = ∆ uj (z)Ea z, + itj 2 , 2 Γ\H 2 y 1 \ H where Ea z, 2 + itj is an Eisenstein series attached to one of the finitely many cusps a of Γ . Theorem 4.3.1 (Phillips-Sarnak, [PS85b]). Let Γ ⊆ SL(2, R) be a Fuchsian group of the first kind that is not cocompact. Suppose that u is a Maass cusp form for Γ with Laplace eigenvalue λ = 1 + t2, and let j j 4 j 1 \H Ea z, 2 + itj be an Eisenstein series attached to a cusp a of Γ . Let Q be a weight-4 cuspidal holomorphic modular form for Γ. Then a sufficient condition for uj to be destroyed under quasiconformal deformation of 1These are non-curvature-preserving deformations, and the resulting surface is not a Riemann surface; the spectral problem is of course still interesting on manifolds without complex structure. 2 2 | |2 Actually, the deformations of the metric are given by dsε = ρε(z) dz + εµ(z)dz , where ρε > 0, but the dependence of ρε ∂ ≡ on ε is such that ∂ε [ρε(z)]ε=0 0, so the metrics described above suffice. 3One actually needs to first make some approximation arguments, since ∆ does not have compact resolvent so perturbation theory does not apply. 56 4.3. The Phillips-Sarnak deformation theory of discrete groups \ H \ H Γ in the direction generated by Q in T`eˇi`c‚hffl(Γ ) is 1 1 L u ,E ·, + it = ∆(1)u ,E ·, + it =6 0. Q j a 2 j j a 2 j The connection to the special values of Rankin-Selberg L-functions is as follows. Starting with the inner product hLQuj,Ea (·, s)i with <(s) > 1, one unfolds the integral and integrates by parts to obtain an expression in terms of the Rankin-Selberg L-function L(s, uj ⊗ Q). Evaluating this at s = sj then yields ! ! 1 s(s + 1) π−2s s − it + k−1 s + it + k−1 L u ,E ·, + it = Γ j 2 Γ j 2 Q j a 2 j 4 Γ (2 + s) 2 2 ! ! s − it + k+1 s + it + k+1 1 × Γ j 2 Γ j 2 L + it , u ⊗ Q . 2 2 2 j j The proportion of these special values that are nonzero was studied by Deshouillers-Iwaniec ([DI86]) and Luo ([Luo93], [Luo01]), who showed that in fact, for congruence subgroups of prime level, a positive proportion of 2 the special values are nonzero. That is, one has MΓ(T ) T for those discrete subgroups of SL(2, R) arising as quasiconformal deformations of Γ0(p) \ H. In Part III we provide a longer overview of these nonvanishing results and the techniques used in their proofs. It is conjectured that, except for the case of the once-punctured torus, for “generic” Γ, there are only finitely many cusp forms, while for arithmetic subgroups cusp forms are known to exist in abundance. ⊗ \ H The connection between the special value of L(s, uj Q) at sj and the actual deformation in T`eˇi`c‚hffl(Γ ) was developed in [PS92a]. The value L(sj, uj ⊗ Q) characterizes the movement of the special point sj off < 1 < 1 the line (s) = 2 into the resonances—poles of the scattering matrix—of the Laplacian in (s) < 2 for the deformed surface. In reference to classical perturbation theory in quantum physics, this is referred to as Fermi’s golden rule. Theorem 4.3.2 (Phillips-Sarnak, [PS92a], [PS92b]). Let sj(ε) be the special point associated to a Maass cusp form uj for Γ under perturbation Γε \ H and let Q ∈ S4(Γ \ H). Then d2 [<(s (ε))] = −c |L(s , u ⊗ Q)|2 , dε2 j ε=0 j j j where cj > 0 is a constant. Additional work on the distribution of eigenvalues under deformations of finite-volume noncompact dis- crete groups is studied in [BV98], [Wol91], [Wol92], [Wol94], [Jud95], [JP97], [Phi97], and [PR13]. An interesting computational investigation for certain discrete groups is carried out in [FL05]. 57 Part II Upper bounds for moments of L-functions at special points In this part, we obtain upper bounds for the moments of L-functions at special points. For each of these results, we will need a large sieve type inequality that incorporates additional twisting from the special point. In Chapter 5 we derive a twisted large sieve type inequality over shortened spectral intervals. In Chapter 6 we use this to obtain upper bounds for the short interval second moment of the L-functions L(sj, uj ⊗ ψ), where ψ is a fixed SL(3, Z) Maass cusp form. In Section 6.4 we modify the proof of the result for GL(2) × GL(3) L-functions to obtain a short interval sixth moment upper bound on L(sj, uj). 58 Chapter 5 A short interval large sieve inequality with spectral twists Contents 5.1 Background and survey of existing literature ...... 60 5.1.0.1 The untwisted spectral large sieve inequality ...... 60 5.1.0.2 The twisted spectral large sieve inequality ...... 61 5.1.0.3 The short interval twisted large sieve inequality of Luo ...... 64 5.1.0.4 Some extended remarks on integral transforms appearing in Kuznetsov . 65 5.2 The short interval twisted spectral large sieve inequality ...... 67 5.2.1 Proof of Theorem 5.2.2 ...... 69 5.2.1.1 Application of Kuznetsov trace formula ...... 69 5.2.1.2 Extraction of the problematic main term ...... 71 5.3 Proof of Theorem 5.2.1 ...... 73 5.4 Proof of Lemmas 5.2.4 and 5.2.5 ...... 75 5.4.1 Proof of Lemma 5.2.4: the k = 0 terms ...... 75 d∗”m`a˚i‹nffl 5.4.1.1 Extraction of the main term: evaluation of Kℓ M,k=0(A; X) ...... 75 5.4.1.2 Estimation of the error term ...... 76 5.4.2 Proof of Lemma 5.2.5: the k ≠ 0 terms ...... 78 5.4.2.1 Proof of Lemma 5.4.1: properties of the weight function ...... 80 Let {uj} be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z) with the notation from Chapter 2. Let A = (an) ∈ R be a sequence and 1 ≤ M T . In this section we will prove the following short interval mean value estimate for twisted sums of Fourier coefficients of the uj. Theorem 5.0.1. Let M,N, and T be parameters with 1 M T . Then for any sequence A = (an) ∈ R, 59 5.1. Background and survey of existing literature we have 2 ! 7 X X 4 3 1 N itj ε 2 anνj(n)n TM + N + M 2 N 2 + T (NT ) kAk . M T −M≤tj ≤T +M n≤N The proof of Theorem 5.0.1 begins in Section 5.2.1, and is a minor adaptation of the work carried out by Luo in [Luo95] using a slightly different weight function, combined with the refinement of Young in [You13]. Apart from one major difference involving the integral transforms of the weight function, outlined in Section 5.1.0.4, the proof essentially amounts to just checking that the technical features of certain integral transforms of auxiliary weight functions developed in [You13] still hold in this setting. The familiar reader may wish to skip those sections entirely. We include the details, however, for completeness. 5.1 Background and survey of existing literature The large sieve type inequalities we discuss here all focus on the spectral aspect, although there is an extensive body of work that is dedicated to proving large sieve type inequalities in other various aspects (weight, level, etc.) and for other objects in harmonic analysis. In this section we review the existing literature on spectral large sieve inequalities with additional spectral twists, and we highlight the subtle differences between this problem and the standard large sieve inequality for GL(2, R) automorphic forms. The general large sieve framework is as follows. The goal in the large sieve problem is to efficiently bound 2 the L -mean value of (Fourier) coefficients λf (n) for f in some family F as |F| grows; that is, to obtain bounds of the form 2 X X 2 anλf (n) (F,N) · kAk , (5.1) B f∈F n≤N A ∈ CN F F where = (an) and B( ,N) is a function depending on the size of and N. By duality, the best bound one can possibly hope to achieve would have F |F| B( ,N) = + N. 5.1.0.1 The untwisted spectral large sieve inequality In the spectral aspect, one takes the family F = F(T ) to consist of Maass cusp forms uj restricted to having eigenparameter tj in some range. For the long spectral interval, the range is tj ≤ T for large T , and consequently, by Weyl’s law the size of F(T ) is T 2. For this family, Iwaniec [Iwa80] established the optimal 60 5.1. Background and survey of existing literature (untwisted) large sieve inequality 2 X X 2 ε 2 anνj(n) T + N (TN) kAk . (5.2) tj ≤T n≤N In fact, the family F(T ) can actually be taken to also include the contribution from the continuous spectrum, and one has the following. Theorem 5.1.1 (Iwaniec, [Iwa80]). Let {uj} be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z). Let A = (an) be a sequence of complex numbers, and let N ≥ 2. Then as T → ∞, 2 Z 2 X X 4 T 1 X a ν (n) + a τ (n) dt T 2 + N (TN)εkAk2. (5.3) n j 2 | |2 n it π −T ζ(1 + 2it) tj ≤T n≤N n≤N Jutila-Motohashi ([Jut00], [JM05]) later extended (5.2) to shortened spectral intervals, with 2 Z 2 X X 4 T +M 1 X a λ (n) + a τ (n) dt (TM + N)(TN)εkAk2. (5.4) n j 2 | |2 n it π T −M ζ(1 + 2it) T −M≤tj ≤T +M n≤N n≤N Again, this (untwisted) spectral large sieve inequality is optimal because Weyl’s law shows that the size of the family is (Z Z ) −(T +M) T +M φ′ 1 + it { − ≤ ≤ } 2 # uj : T M tj T + M + + 1 dt TM. −(T −M) T −M φ 2 + it 5.1.0.2 The twisted spectral large sieve inequality itj If one includes additional spectral twists by replacing νj(n) with νj(n)n on the lefthand side of (5.2), complicating new features arise. In [DI86], Deshouillers-Iwaniec implicitly established a nonoptimal upper bound of the following form. Theorem 5.1.2 (Deshouillers-Iwaniec, [DI86]). Let {uj} be an orthonormal basis of Hecke-Maass cusp forms for SL(2, Z). Let A = (an) be a sequence of complex numbers, and let N ≥ 2. Then for T 1, 2 Z 2 X X 4 T 1 X a ν (n)nitj + a τ (n) dt T 2 + N 2+ε kAk2 (5.5) n j 2 | |2 n t π −T ζ(1 + 2it) tj ≤T n≤N n≤N for A ∈ R. 61 5.1. Background and survey of existing literature The issue arises from the additional twisting factors in the continuous spectrum, whence for that contri- bution they are only able to make the bound Z 2 T X 4 1 it kAk2 2 2 anτt(n)n dt NT , π − |ζ(1 + 2it)| T n≤N and upon application of Kuznetsov trace formula, the resulting sum of Kloosterman sums is O N 2kAk2 . With the additional assumption of square-root cancellation for additive twists of the sequence A,(5.5) is 3 1 2 ε 2 2 N kAk2 improved to O T + N T + T (NT ) . The original application of Deshouillers-Iwaniec was to establish a lower bound on the nonvanishing of the special values L(sj, uj ⊗ Q), with a view towards Phillips-Sarnak’s deformation theory of discrete groups. We provide further discussion of this result in Part III. For arbitrary A, in [Luo95], Luo was able to partially circumvent the issue of having ruinously large contributions from the continuous spectrum and Kloosterman sums by recognizing and extracting the large contribution from the continuous spectrum in (5.5) and exactly cancelling it with a corresponding contribu- tion from the sum of Kloosterman sums on the geometric side of the Kuznetsov trace formula. He established the following. Theorem 5.1.3 (Luo, [Luo95]). For any sequence A = (an)n∈N ∈ R, we have 2 ! ! X X 3 3 1 N 2 itj 2 ε 2 anνj(n)n T + T 2 N 2 + (NT ) kAk . (5.6) T tj ≤T n≤N In fact, the term extracted by Luo’s method can be shown to be bounded below by TNkAk2 for certain sequences, so that an optimal bound of the same shape as the untwisted large sieve inequality (5.2) is simply not true when the additional spectral twists are included. It is easier to work with smoothed versions of the spectral sums above; denote the smoothed versions of the discrete spectrum average by 2 X X −tj /T itj ST (A) = e anνj(n)n , tj n≤N and of the continuous spectrum average by Z 2 ∞ −t/T X 2 e it TT (A) = anτit(n)n dt. π |ζ(1 + 2it)|2 0 n≤N Theorem 5.1.3 is implied by the following asymptotic expansion of the smoothed sums that is implicit in 62 5.1. Background and survey of existing literature [Luo95]. Theorem 5.1.4 (Luo, [Luo95]). For any sequence A = (an)n∈N ∈ R and 1 ≤ X ≤ T , the smoothed averages satisfy 2T X X 1 S A T A S ˜bˆa`dffl A T ( ) + T ( ) = aman 2 S(0, m; r)S(0, n; r) + T ( ; X) (5.7) π ≤ ≥ r m,n N r 1 ! ! 3 NT N 2 + O T 2 + + (NT )εkAk2 , (5.8) X T ≥ ≤ 1 where, for a smooth weight function η satisfying η(x) = 1 for x 1 and η(x) = 0 for x 2 , 1 X X X sn + νm S ˜bˆa`dffl(A; X) = 2 cos a a |m − n| e T 2T m n r m≠ n r