Sums of Fourier Coefficients of Modular Forms and the Gauss Circle Problem
Alexander Weston Walker
B.A. in Mathematics, Boston College, Chestnut Hill, MA, 2012 M.Sc. in Mathematics, Brown University, Providence, RI, 2015
A Dissertation Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Mathematics at Brown University
Recommended for Acceptance by the Department of Mathematics Advisor: Professor Jeffrey Hoffstein
May 2018 c Copyright by Alexander Weston Walker, 2018. All rights reserved. This dissertation by Alexander Walker is accepted in its present form by the Department of Mathematics as satisfying the dissertation requirement for the degree of Doctor of Philosophy.
Date Jeffrey Hoffstein, Advisor
Recommended to the Graduate Council
Date Maria Nastasescu, Reader
Date Michael Rosen, Reader
Approved by the Graduate Council
Date Andrew Campbell, Dean of the Graduate School Vitae
Alexander Weston Walker was born in Concord, New Hampshire on January 17, 1990 to Marilyn Walker (n´eeMcNeil) and Kenneth Walker. He grew up in Londonderry, New Hampshire and graduated from Londonderry High School in 2008. He received his B.A. in Mathematics from Boston College in 2012 and was awarded the Paul J. Sally Distinguished Alumnus Prize upon graduation.
Alexander began his graduate studies in the fall of 2012 at Brown University, where he received his M.Sc. in Mathematics in 2015. During his time at Brown, he taught undergraduate courses in calculus, linear algebra, and multi-variable calculus. In addition, he taught summer classes in number theory through the Summer@Brown program and led a readings course in analytic number theory through the math department’s Directed Reading Program.
When he’s not writing his thesis, Alexander is an active participant in the math department’s informal game theory seminar.
iv Abstract of “Sums of Fourier Coefficients of Modular Forms and the Gauss Circle Problem,” by Alexander Walker, Ph.D., Brown University, May 2018.
The Gauss circle problem is a classic problem in number theory that concerns estimates for the number of lattice points contained in a circle of large radius. This question originates with Gauss, who proved that the number of lattice points can be approximated by the area of the enclosing circle. Well-supported conjectures suggest that the error of this approximation is surprisingly small. In this thesis, we investigate the Gauss circle problem and several variants by means of Dirichlet series.
We begin by studying the partial sums of Fourier coefficients of GL(2) cusp forms. These partial sums are conjectured to behave much like the error term in the Gauss circle problem, but are simpler in many analytic regards. We introduce Dirichlet series whose coefficients are the squares of the partial sums and prove that these series have meromorphic continuation to the entire complex plane. Much of this material has been introduced elsewhere, but this simplified analogy of the Gauss circle problem serves as an important foundation for later chapters.
We then turn our attention to the Gauss circle problem itself. Specifically, we address the generalized Gauss circle problem, which concerns estimates for the number of lattice points in k-dimensional spheres. Techniques developed in the previous (cusp form) case are modified and applied to understand the meromorphic behavior of the Dirichlet series associated to the second moment of the error term in the k-dimensional Gauss circle problem. Integral transforms are then applied to prove sharp and smooth second moment results for the lattice point discrepancies. Our results are particularly interesting in dimension three, where we develop the first
v power-savings error for the second moment of the lattice point discrepancy.
The conjectural bounds in the Gauss circle problem and its generalization to higher dimensions are of a fundamentally different form. We recognize this phase change as a property of the weight of the underlying modular form and use this to motivate a new variant of the Gauss circle problem concerning twisted divisor sums that we call the Eisenstein series analogy.
Finally, we consider a variant of the Gauss circle problem that concerns the size of iterated partial sums of coefficients of modular forms. We apply the theory of iterated partial sums to recover information about non-iterated partial sums and show how questions regarding iterated partial sums may be approached using Dirichlet series.
vi This thesis is dedicated to all those who gave me opportunities. With special mention to my wife, for also taking them with me.
vii Acknowledgements
This thesis would not have been possible without the support of my mathematical family. I’d like to thank Jeff for encouraging the collaboration that makes this family metaphor work, and the automorphic N, for making room for one more newform.1
In particular, I’d like to thank my brothers and collaborators Tom Hulse, Chan Ieong Kuan, and David Lowry-Duda. Between our five joint papers and at least that many productive weekly meetings, you have left your fingerprints on every page.
For everything he’s done for me outside of these collaborations, I’d like to give a second round of thanks to David the office-mate, David the roommate, and David the friend. You have turned me into a better mathematician and a better person.
Thank you to Doreen, for all the work you did to help me secure my current fellowship. Thank you to Audrey, for talking to the graduate school so that I never seem to have to. And thank you to Jeff, Maria, Mike and Min, for serving on my committees.
Lastly, I’d like to thank my family and my wife for their endless love and support. You might not appreciate point-counting in conic sections but you do appreciate me.
1I’d also like to thank my tenth cousin nine times removed, Johann Carl Friedrich Gauss. viii Contents
1 Introduction1 1.1 The Gauss Circle Problem...... 1 1.2 The Generalized Gauss Circle Problem...... 3 1.3 A Connection to Modular Forms and the Cusp Form Analogy....6 1.4 Outline and Summary of Major Results...... 8
2 Background 12 2.1 Modular Forms...... 12 2.2 Eisenstein Series and Poincar´eSeries...... 15 2.3 Some L-functions attached to Modular Forms...... 16
3 The Cusp Form Analogy 18 3.1 An Elementary Decomposition...... 20 3.2 Analytic Properties of Wf (s)...... 23 3.2.1 Spectral Expansion of the Shifted Convolution...... 23 3.3 Meromorphic Continuation of the Shifted Convolution...... 28 3.3.1 Continuation of the Discrete Spectral Part of Zf (s, w)..... 28 3.3.2 Continuation of the Continuous Spectral Part of Zf (s, w)... 29 3.4 Polar Analysis of D(s, Sf × Sf )...... 34 3.4.1 Polar Analysis of Zf (s, w)...... 34 3.4.2 Polar Analysis of Wf (s)...... 36 3.4.3 Polar Analysis of D(s, Sf × Sf )...... 37 3.5 Arithmetic Applications...... 38
4 The Generalized Gauss Circle Problem 44 4.1 Algebraic Decompositions...... 47 4.2 Spectral Expansion of Zk(s, w)...... 50 4.2.1 Spectral Expansion of the Shifted Convolution Zk(s, w).... 51 k 2 k 4.2.2 Modifying |θ (z)| Im(z) 2 to be Square Integrable...... 52 4.2.3 Applying the Spectral Expansion of the Poincar´eSeries.... 56 4.3 Meromorphic Continuation of Zk(s, w)...... 59 4.3.1 Meromorphic Continuation of the Non-Spectral Part..... 59 4.3.2 Meromorphic Continuation of the Discrete Spectral Part... 62 4.3.3 Meromorphic Continuation of the Continuous Spectral Part. 63 4.4 Analytic Behavior of Wk(s)...... 67 ix 4.4.1 The Diagonal Part...... 67 4.4.2 The Discrete Part...... 70 4.4.3 The Continuous Part...... 70 4.4.4 The Non-Spectral Part...... 71 4.4.5 Polar Analysis of Wk(s)...... 72 4.5 Polar Analysis of D(s, Sk × Sk) and D(s, Pk × Pk)...... 74 4.5.1 Polar Analysis of D(s, Sk × Sk)...... 74 4.5.2 Polar Analysis of D(s, Pk × Pk)...... 76 4.5.3 Cancellation in the Poles of D(s, Pk × Pk)...... 78 4.6 Modifications in the Planar Case...... 79 4.6.1 Spectral Expansion of Z2(s, w)...... 80 4.6.2 Meromorphic Continuation of Z2(s, w)...... 82 4.6.3 Analytic Behavior of W2(s)...... 83 4.6.4 Polar Analysis of D(s, S2 × S2) and D(s, P2 × P2)...... 87
5 Arithmetic Results for the Generalized Gauss Circle Problem 92 5.1 The Discrete Laplace Transform...... 95 5.2 The Laplace Transform...... 96 5.3 The Discrete Second Moment...... 101 5.4 The Second Moment...... 107
6 A Brief Note on Sums of Coefficients of Modular Forms 113 6.1 Sums of Coefficients of Eisenstein Series...... 114 6.2 Weight One Eisenstein Series...... 115 6.3 Conjectures for the Eisenstein Series Analogy...... 117
7 Iterated Partial Sums 120 7.1 C´esaroand Riesz Means...... 120 7.2 Relation to the Gauss Circle Problem and the Cusp Form Analogy.. 122 7.3 Applications to (Non-Iterated) Partial Sums...... 124 7.4 Dirichlet Series Attached to Riesz Means...... 126 7.4.1 An Elementary Decomposition...... 127 α α 7.4.2 General Analytic Behavior of D(s, Rf × Rf )...... 128 1 1 7.4.3 Polar Analysis of D(s, Rf × Rf )...... 128
A Explicit Bounds for D(s, P3 × P3) 133 A.1 Explicit Bounds for W3(s)...... 133 A.1.1 The Diagonal Part...... 134 A.1.2 The Non-Spectral Part...... 134 A.1.3 The Discrete Part...... 135 A.1.4 The Continuous Part...... 136 A.2 Explicit Bounds for D(s, P3 × P3)...... 138
Index 140
Bibliography 142 x Chapter 1
Introduction
This thesis applies new techniques to investigate a number of variants on the classic Gauss circle problem. In this introductory chapter, we summarize the history and progress towards Gauss’ problem. We describe two variants of the circle problem and the common theme that ties them together: the partial sums of Fourier coefficients of modular forms. We conclude with a summary of the main results of this thesis.
1.1 The Gauss Circle Problem
The story of the Gauss circle problem begins with an innocent question about point- counting, first posed by Gauss in 1798:
How many integer lattice points (x, y) lie inside of the circle of radius √ R? In other words, how many integer solutions does x2 + y2 ≤ R have?
In the same work, Gauss gives a geometric proof that the number of solutions above is well-approximated by the area of the bounding circle, πR. To describe
Gauss’ bound precisely, we let S2(R) denote the number of lattice points in the circle √ of radius R, and define the lattice point discrepancy P2(R) by
S2(R) = πR + P2(R).
1 By relating the magnitude of the error term P2(R) to the area of a narrow annulus,
1/2 Gauss proved that P2(R) = O(R ). In some respects, this bound is unsurprising; indeed, by writing X S2(R) = r2(n), n≤R
in which r2(n) is the number of representations of n as a sum of two integer squares √ (ie. the number of lattice points on the circle of radius n), one may recognize Gauss’ error bound as a form of square-root cancellation, which is common in arithmetic problems.
Gauss’ bound would not be improved for over a century, until Sierpi´nski[Sie06]
1/3 proved that P2(R) = O(R ) by applying some recent ideas of Voronoi (1903). Fol- lowing Sierpi´nski’sresult mathematicians were naturally led to the following question, which might be called the actual “problem” underlying the Gauss circle problem:
α+ Question. What is the smallest exponent α for which P2(R) = O(R ) for all > 0?
It is widely conjectured that the optimal exponent in the Gauss circle problem
1 1 is α = 4 , despite the fact that any progress past Sierpi´nski’s α ≤ 3 bound has been hard-fought. Incremental improvements over the last century by the likes of Littlewood [LW24], van der Corput [vdCN28], Iwaniec–Mozzochi [IM88], and Huxley
131 has managed to show only that α ≤ 416 ≈ 0.3149 [Hux03]. A recent preprint due to 517 Bourgain and Watt [BW17] claims the marginal improvement α ≤ 1648 ≈ 0.31371.
On the other hand, a classic result due to Hardy [Har15] has already established
1 that the conjectured lower bound α ≥ 4 must hold. A second proof of Hardy’s lower bound was given seven years later as a consequence of the following mean-square theorem of Cram´er.
2 Theorem ([Cra22]). There exists a constant C > 0 such that
Z R 2 3/2 5/4 |P2(t)| dt = CR + O(R ). 0
1 Thus α = 4 represents not only a lower bound but the true average order of the exponent in the error bound. This is compelling evidence in support of the conjecture
1 α = 4 , which is often called the Gauss Circle Problem in the strict sense.
1.2 The Generalized Gauss Circle Problem
Of the circle problem’s many variants, none is more honest than the extension from point-counting in circles to point-counting in spheres of higher dimension.
Extending notation from the previous section, we let Sk(R) denote the number of √ lattice points in the k-dimensional sphere of radius R. A direct generalization of
k/2 Gauss’ proof shows that Sk(R) ∼ vkR as R → ∞, in which vk is the volume of the k-dimensional ball of radius one. As in dimension two, we quantify this approximation by defining an error term Pk(R), such that
k Sk(R) = vkR 2 + Pk(R).
Written this way, the generalized Gauss circle problem asks the following:
αk+ Question. What is the least exponent αk for which Pk(R) = O(R ) for all > 0?
Gauss’ geometric bound readily extends to general dimension k > 1 to show that
αk ≤ (k − 1)/2. Known improvements verify that additional cancellation occurs in the error term, and partial results (Ω±-results and mean-square estimates, e.g.) point
3 towards the conjecture
k 2 − 1, k > 2 αk = (1.1) 1 4 , k = 2.
Interestingly, this suggests that the behavior of the two-dimensional Gauss circle problem is fundamentally different than the corresponding case in higher dimensions.
It does not appear that mathematicians gave serious consideration to the general- ized Gauss circle problem until after Sierpi´nski’sseminal improvement in the planar
3 case. The first result in higher dimensions is due to Landau [Lan15] and gives α3 ≤ 4
by extending Sierpi´nski’smethod to dimension three. Since then, bounds on P3(R)
2 have been the subject of many author’s works, and include α3 ≤ 3 (due to [Vin63]), 29 21 α3 ≤ 44 (due to [CI95]), and the current estimate α3 ≤ 32 ≈ 0.6563 (due to [HB99]).
The history of the mean-square estimate for the lattice point discrepancy in di- mension three is far easier to state – it begins with a result due to Jarnik [Jar40] giving
Z R 2 2 2p P3(t) dt = CR log R + O(R log R), (1.2) 0
and ends with a recent result due to Lau [Lau99] which reduces the error to O(R2). Weak results such as these underline a sentiment of the authors of the excellent survey article [IKKN06], who write
“As most experts agree, in our familiar three-dimensional Euclidean space the analogue of the Gaussian problem is the most difficult and enigmatic.”
In sharp contrast, the generalized Gauss circle problem in dimensions four and above is almost entirely understood. In dimension four, the conjectured error bound
4 for P4(R) follows from the closed-form
X m S (R) = 1 + 8 σ(m) − 32 σ = v R2 + O(R log R), 4 4 4 m≤R
in which σ(m) is the sum-of-divisors function for integer m and is 0 otherwise by slight abuse of notation. In dimension k ≥ 5, uniform application of the k = 4 bound and the recursive convolution formula
√ R X 2 Sk(R) = Sk−1(R) + 2 Sk−1(R − m ) m=1
k/2−1 suffice to give Pk(R) = O(R log R) by induction. Of course, this implies that the generalized Gauss circle problem is known (and true!) in dimensions four and above. (More is known than has been written. For example, it is known that the log-powers are not needed in dimension k ≥ 5.)
Our understanding of the generalized Gauss circle problem in dimensions k ≥ 4 highlights how much harder the problem becomes in dimensions two and three. Yet this is not to say that all is known about these higher-dimensional cases. As one example, we present the mean-square result for dimension four, which satisfies
Z R 2 3 5 P4(t) dt = CR + O(R 2 log R) 0
following a result of Jarnik [Jar40]. This error term is almost surely too large and could potentially be reduced to O(R2+) for any > 0 by introducing a secondary main term of size R5/2 that has yet to be isolated.
5 More information on the generalized Gauss circle problem and its relation to the Dirichlet divisor problem and point-counting problems in generic convex bodies can be found in the survey article [IKKN06].
1.3 A Connection to Modular Forms and the Cusp
Form Analogy
Analytic number theory is guided by the principle that arithmetic functions can be understood by the meromorphic properties of their generating functions. For this reason, many attacks on the generalized Gauss circle problem begin with the identity
X Sk(n) = rk(m), m≤n
in which rk(n) is the number of representations of n as a sum of k integer squares.
This idea is particularly fruitful because the generating function of the rk(n) satisfies
k X 2πinz X 2πinz k rk(n)e = r1(n)e = θ(z) , n∈Z n∈Z in which θ(z) is the classic theta function, first introduced by Jacobi (1829). In modern terminology, we describe θ(z) as a modular form of weight 1/2 and level 4. Casting θ(z) in these terms suggests that the generalized Gauss circle problem is but one of a myriad of problems regarding the error terms in sums of Fourier coefficients of modular forms. In particular, it may be that progress towards the Gauss circle problem could be motivated by progress on “simpler” analogues of θ(z).
Shortly after Sierpi´nskiinitiated the modern crusade on the Gauss circle problem, Ramanujan (1916) began investigating what would later become one of the first cusp
6 forms, the modular discriminant. Defined as the infinite product
Y X ∆(z) := e2πiz (1 − e2πinz)24 = τ(n)e2πinz, n≥1 n≥1 the modular discriminant satisfies a family of functional equations related to the action of the modular group on the upper half plane. Ramanujan made several claims about the coefficients τ(n) in 1916, including the claim that τ(n) n11/2+. This conjecture became known as the Ramanujan Conjecture (and later, as a part of the Ramanujan–Petersson Conjecture) until it was settled in 1974 following Deligne’s resolution of the Weil Conjectures.
Inspired by the Gauss circle problem analogy, one might hope that the partial sums of τ(n) experience greater-than-squareroot-cancellation, or even that
X 11 + 1 + S∆(X) := τ(n) X 2 4 . (1.3) n≤X for each > 0. And indeed, this bound was realized on average in mean-square by the work of Chandrasekharan–Narasimhan (1962-64), who proved that
Z X 2 11+ 3 12+ S∆(t) dt = CX 2 + O(X ) (1.4) 0 for all > 0.
Work by Hecke and many others has contextualized the properties of ∆(z) into the general theory of cusp forms, and the aforementioned results of Chandrasekharan– Narasimhan prove that extraordinary cancellation occurs in the partial sums of Fourier coefficients of cusp forms, at least on average.
7 However, like the original Gauss circle problem, this cusp form analogy remains open. State-of-the-art is essentially due to Jutila and Hafner–Ivi´c,who establish that
1 summation increases the exponent of the error by at most 3 over the size of a cusp form’s individual coefficients. Thus progress in the cusp form analogy matches that of the Gauss circle problem, circa 1906! The lack of further improvements could be due to greater abstraction, a lack of study, or both.
1.4 Outline and Summary of Major Results
Chapter2: Background Material
Chapter2 introduces many of the automorphic forms that we require in this the- sis. We present the basic properties of these forms and their associated L-functions, directing the reader to in-depth references where appropriate.
Chapter3: The Cusp Form Analogy
Chapter3 discusses a variant of the Gauss circle problem dating back to the work of Jutila [Jut87] and Hafner–Ivi´c[HI89] which we call the cusp form analogy. To attack this problem, we introduce the Dirichlet series
2 X |Sf (n)| D(s, S × S ) := , f f ns+k−1 n≥1
in which Sf (n) denotes the sum of the first n Fourier coefficients of the cusp form f(z).
We prove that the Dirichlet series D(s, Sf × Sf ) has a meromorphic continuation to the entire complex plane. In Theorem 3.4.2, we take this analysis a step further to
classify rightmost poles and residues of D(s, Sf × Sf ). This information is used for
8 arithmetic applications in §3.5. For example, we prove in Theorem 3.5.4 that
2 3 X Sf (n) −n/X L( 2 , f × f) 3 3 1 + e = Γ( )X 2 + O X 2 nk−1 4π2ζ(3) 2 n≥1 for all > 0, where k is the weight of f(z).
The material in this chapter closely follows the paper [HKLW17c]. However, as the main purpose of this chapter is to introduce important ideas for later use in our attack on the Gauss circle problem, the scope of our results will be artificially limited.
Chapters4–5: The Generalized Gauss Circle Problem
Chapters4–5 present a novel attack on the generalized Gauss circle problem through the study of the Dirichlet series
2 2 X Sk(n) X Pk(n) D(s, S × S ) = ,D(s, P × P ) = (1.5) k k ns+k k k ns+k−2 n≥1 n≥1
and their meromorphic properties. Here, Sk(n) denotes the number of lattice points √ in the k-ball of radius n and Pk(n) denotes the error term in the approximation
k/2 Sk(n) = vkn + Pk(n), where vk is the volume of the unit k-sphere.
Our treatment of the two Dirichlet series in (1.5) is inspired by the investigation into D(s, Sf × Sf ) conducted in Chapter3. We prove that the series in (1.5) have meromorphic continuations and classify their rightmost poles, at first for k ≥ 3 in §4.5 and then for k = 2 in §4.6.4.
Arithmetic applications are delayed until Chapter5, where we prove a variety of second moment results for the lattice point discrepancy Pk(n). One of these results
9 2 is Theorem 5.2.5, which proves new bounds for the Laplace transform of Pk(n) . In dimension k = 2, this estimate takes the form
∞ Z 3 1 2 −t/X 0 3 2 2 + P2(t) e dt = C2Γ( 2 )X − X + O X , 0 which improves a result of Ivi´c[Ivi01].
We conclude the chapter with Theorem 5.4.6, which presents asymptotics for the
(sharp and continuous) second moment of Pk(n). This result is novel in dimension k = 3, where it improves results of Jarnik [Jar40] and [Lau99] and produces, for the first time, a power-savings error.
In AppendixA, we prove Lemma A.2.2, a technical lemma which further improves our result in the three-dimensional case. When this lemma is taken into account, our result becomes
Z X 0 0 2 2 C3 2 C3 C3 π 2 2− 2 + P3(t) dt = X log X + − − X + O X 11 . 0 2 2 4 3
Chapters4 and5 unify and further develop material which also appears in the recently submitted papers [HKLW17a] and [HKLW17b].
Chapter6: Sums of Coefficients of Modular Forms
Chapter6 discusses a generalization of the cusp form analogy to sums of coefficients of non-cuspidal modular forms. We show that theory is typically trivial, except in the case of modular forms of extremely low weight, and use this to explain the phase change in the generalized Gauss circle problem between dimensions two and three.
10 Based on our results we present a new variant of the Gauss circle problem for sums of Fourier coefficients of weight one Eisenstein series. This Eisenstein series analogy is stated precisely and illustrated with some large-scale computation.
Chapter7: Iterated Partial Sums
Chapter7 discusses a variant of the Gauss circle problem which concerns iterated partial sums of Fourier coefficients of modular forms. These iterated sums are mostly understood following work of Hardy [Har17] and Jutila [Jut87], which we adapt to suit our particular needs.
In §7.3, we demonstrate new ways in which these higher partial sums can be used to control the error terms in the Gauss circle problem and the cusp form analogy. We apply this theory in Corollary 7.3.2 to give a new proof of the result
k−1 + 1 − Sf (X) = Ω(X 2 4 ), which is conjecturally sharp (up to logarithmic factors, etc.).
We conclude Chapter7 by showing that the general techniques of Chapter3 can be applied to study the Dirichlet series associated to the squares of the Riesz means of the Fourier coefficients of cusp forms.
11 Chapter 2
Background
In this chapter, we give a quick review of the automorphic forms we will be using frequently, including modular forms, Eisenstein series, and Poincar´eseries. We then introduce two types of L-functions associated to modular forms.
2.1 Modular Forms
Before defining modular forms, let us recall that the group GL(2) acts on the upper half-plane H by means of linear fractional transformations, ie.
az + b a b γ · z := , γ = ∈ GL(2). cz + d c d
A modular form f(z) of integer weight k (and level 1) is a holomorphic function f : H → C which satisfies
k f(γz) = (cz + d) f(z) for all γ ∈ SL(2, Z) (2.1) and which is holomorphic at infinity. For many applications, it is useful to define modular forms which satisfy (2.1) when γ is restricted to a finite index subgroup Γ ⊂
12 SL(2, Z). One family of finite index subgroups consists of the principal congruence subgroups of level n,
1 0 Γ(n) := γ ∈ SL(2, ): γ ≡ mod n , Z 0 1
and subgroups Γ that contain Γ(n) for some n are known as congruence subgroups.
The congruence subgroup that we use most frequently in this thesis is Γ0(4), in which
a b Γ (n) := ∈ SL(2, ): c ≡ 0 mod n . 0 c d Z
We restrict our attention to modular forms defined on congruence subgroups in this thesis. In this case, holomorphicity at infinity implies that f(z) has a Fourier expansion at infinity, which we may write in the form
X f(z) = a(n)e(nz), e(z) := e2πiz. n≥0
If f(z) is chosen to be an eigenfunction for each of the Hecke operators, then the coefficients of a(n) are multiplicative (though not completely multiplicative). For this reason as well as for numerous connections to point-counting on curves over finite fields, these coefficients are of great arithmetic interest. Following Deligne’s resolution of the Weil Conjectures, we have a(n) d(n)n(k−1)/2 [Del74].
The hyperbolic manifold defined by the quotient Γ\H has a cusp at infinity and in general has many inequivalent cusps. A cusp form is a modular form on Γ that
vanishes at each cusp of Γ\H.
To each cusp a of Γ\H, let Γa denote the stabilizer of the cusp a under the group
action of Γ. A scaling matrix σa ∈ PSL2(R) is any matrix which satisfies σa∞ = a and
13 ∼ a b induces the isomorphism Γa = Γ∞ via conjugation. For each γ = ( c d ) ∈ PSL2(R), we define the slash operator of weight k, denoted |γ, as the operator which maps
−k f : H → C to the function f|γ(z) := (cz + d) f(γz). Thus cusp forms are exactly those modular forms for which f|σa (z) vanishes as z → ∞ for each cusp a.
The theory of modular forms has been generalized to include modular forms with character, which are holomorphic functions f : H → C satisfying the “twisted” functional equation
f(γz) = χ(d)(cz + d)kf(z) for all γ ∈ Γ, where χ is a Dirichlet character and Γ is a congruence subgroup. A classic example in this theory is the function
2 2 X 2πin2z X 2πinz θ (z) = e = r2(n)e , n∈Z n≥0