Dokumen - Pub - Problems in The Theory of Modular Forms 978 981 10 2651 5 9811026513 978 93 80250 72 4

HBA Lecture Notes in Mathematics
IMSc Lecture Notes in Mathematics
M. Ram Murty
Michael Dewar
Hester Graves
Problems in
the Theory of
Modular
Forms
Series Editor
Sanoli Gun, Institute of Mathematical Sciences, Chennai, Tamil Nadu, India
Editorial Board
R. Balasubramanian, Institute of Mathematical Sciences, Chennai
Abhay G. Bhatt, Indian Statistical Institute, New Delhi
Yuri F. Bilu, Université Bordeaux I, France
Partha Sarathi Chakraborty, Institute of Mathematical Sciences, Chennai
Carlo Gasbarri, University of Strasbourg, Germany
Anirban Mukhopadhyay, Institute of Mathematical Sciences, Chennai
V. Kumar Murty, University of Toronto, Toronto
D.S. Nagaraj, Institute of Mathematical Sciences, Chennai
Olivier Ramaré, Centre National de la Recherche Scientifique, France
Purusottam Rath, Chennai Mathematical Institute, Chennai
Parameswaran Sankaran, Institute of Mathematical Sciences, Chennai
Kannan Soundararajan, Stanford University, Stanford
V.S. Sunder, Institute of Mathematical Sciences, Chennai
About the Series
The IMSc Lecture Notes in Mathematics series is a subseries of the HBA Lecture
Notes in Mathematics series. This subseries publishes high-quality lecture notes
of the Institute of Mathematical Sciences, Chennai, India. Undergraduate and
graduate students of mathematics, research scholars, and teachers would find this
book series useful. The volumes are carefully written as teaching aids and highlight
characteristic features of the theory. The books in this series are co-published with
Hindustan Book Agency, New Delhi, India.
More information about this series at http://www.springer.com/series/15465

M. Ram Murty Michael Dewar
•
Hester Graves
Problems in the Theory

of Modular Forms
123
M. Ram Murty Hester Graves
Department of Mathematics and Statistics University of Michigan
Queen’s University Ann Arbor, MI, USA
Kingston, ON, Canada
Michael Dewar
Queen’s University
Kingston, ON, Canada
This work is a co-publication with Hindustan Book Agency, New Delhi, licensed for sale in
all countries in electronic form only. Sold and distributed in print across the world by
Hindustan Book Agency, P-19 Green Park Extension, New Delhi 110016, India. ISBN:
978-93-80250-72-4 © Hindustan Book Agency 2016.
ISSN 2509-8071 (electronic)

ISSN 2509-8098 (electronic)
ISBN 978-981-10-2651-5 (eBook)
DOI 10.1007/978-981-10-2651-5
Library of Congress Control Number: 2016952892
© Springer Science+Business Media Singapore 2016 and Hindustan Book Agency 2016
This work is subject to copyright. All rights in this online edition are reserved by the Publishers, whether
the whole or part of the material is concerned, specifically the rights of reuse of illustrations, recitation,
broadcasting, and transmission or information storage and retrieval, electronic adaptation, computer
software, or by similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publishers, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publishers nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.
Printed on acid-free paper
This Springer imprint is published by Springer Nature

The registered company is Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #22-06/08 Gateway East, Singapore 189721, Singapore
A great discovery solves a great problem but
there is a grain of discovery in the solution of
any problem.
- George Pólya, How to solve it.
Contents
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Part I Problems
1 Jacobi’s q-series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 The q-exponential function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Jacobi’s Triple Product Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Jacobi’s two-square theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Jacobi’s four square theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Supplementary problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2 The Modular Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.1 The full modular group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Subgroups of the modular group . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 The Hecke subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Groups acting on topological spaces . . . . . . . . . . . . . . . . . . . . . . . . 19
3 The Upper Half-Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.1 The group GL+ n (R) and its subgroups . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Fundamental domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 The extended upper half-plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 The Poincaré metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4 Modular Forms of Level One . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.1 Fourier expansions and q-series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 Eisenstein series for SL2 (Z) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3 The valence formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.4 The dimension formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
vii
viii Contents
4.5 The j-function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5 The Ramanujan τ -function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.1 The Eisenstein series E2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2 Hecke operators of level one . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.3 Modular forms and Dirichlet series . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.4 Ramanujan congruences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6 Modular Forms of Higher Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

6.1 Modular forms for congruence subgroups . . . . . . . . . . . . . . . . . . 73
6.2 Eisenstein series for Γ(N ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.3 Fourier expansions of higher level Eisenstein series . . . . . . . . . . 76
6.4 The valence and dimension formulas . . . . . . . . . . . . . . . . . . . . . . . 79
6.5 The Jacobi four-square theorem revisited . . . . . . . . . . . . . . . . . . . 81
7 The Petersson Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

7.1 The Hilbert space of cusp forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.2 Commutativity of the Hecke operators . . . . . . . . . . . . . . . . . . . . . 87
7.3 Hecke operators as Hermitian operators . . . . . . . . . . . . . . . . . . . . 90
7.4 Basis of eigenforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
8 Hecke Operators of Higher Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

8.1 Hecke operators for congruence subgroups . . . . . . . . . . . . . . . . . 95
8.2 Nebentypus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
8.3 Oldforms and newforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
9 Dirichlet Series and Modular Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

9.1 General Dirichlet series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
9.2 The Poisson summation formula . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
9.3 L-functions attached to modular forms . . . . . . . . . . . . . . . . . . . . . 123
9.4 Twists of L-series attached to modular forms . . . . . . . . . . . . . . . . 129
10 Special Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

10.1 Elliptic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
10.2 The modular invariant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
10.3 Theta series and lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
10.4 Special values of zeta and L-functions . . . . . . . . . . . . . . . . . . . . . . 147
Contents ix
Part II Solutions
1 Jacobi’s q-series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

1.1 The q-exponential function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
1.2 Jacobi’s Triple Product Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
1.3 Jacobi’s two-square theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
1.4 Jacobi’s four square theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
2 The Modular Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

2.1 The full modular group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
2.2 Subgroups of the modular group . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
2.3 The Hecke subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
2.4 Groups acting on topological spaces . . . . . . . . . . . . . . . . . . . . . . . . 168
3 The Upper Half-Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

3.1 The group GL+ n (R) and its subgroups . . . . . . . . . . . . . . . . . . . . . . 175
3.2 Fundamental domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
3.3 The extended upper-half plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
3.4 The Poincaré metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
4 Modular Forms of Level One . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

4.1 Fourier expansions and q-series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
4.2 Eisenstein series for SL2 (Z) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
4.3 The valence formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
4.4 The dimension formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
4.5 The j-function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5 The Ramanujan τ -function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

5.1 The Eisenstein series E2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.2 Hecke operators of level one . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
5.3 Modular forms and Dirichlet series . . . . . . . . . . . . . . . . . . . . . . . . . 205
5.4 Ramanujan congruences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
6 Modular Forms of Higher Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

6.1 Modular forms for congruence subgroups . . . . . . . . . . . . . . . . . . 217
6.2 Eisenstein series for Γ(N ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
6.3 Fourier expansions of higher level Eisenstein series . . . . . . . . . . 220
6.4 The valence and dimension formulas . . . . . . . . . . . . . . . . . . . . . . . 224
6.5 The Jacobi four-square theorem revisited . . . . . . . . . . . . . . . . . . . 227
x Contents
7 The Petersson Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

7.1 The Hilbert space of cusp forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
7.2 Commutativity of the Hecke operators . . . . . . . . . . . . . . . . . . . . . 234
7.3 Hecke operators as Hermitian operators . . . . . . . . . . . . . . . . . . . . 235
7.4 Basis of eigenforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
8 Hecke Operators of Higher Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239

8.1 Hecke operators for congruence subgroups . . . . . . . . . . . . . . . . . 239
8.2 Nebentypus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
8.3 Oldforms and newforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
9 Dirichlet Series and Modular Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

9.1 General Dirichlet series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
9.2 The Poisson summation formula . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
9.3 L-functions attached to modular forms . . . . . . . . . . . . . . . . . . . . . 260
9.4 Twists of L-series attached to modular forms . . . . . . . . . . . . . . . . 263
10 Special Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269

10.1 Elliptic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
10.2 The modular invariant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
10.3 Theta series and lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
10.4 Special values of zeta and L-functions . . . . . . . . . . . . . . . . . . . . . . 279
A short guide for further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Acknowledgements
This book is based on an upper level undergraduate course given at Queen’s

University in the fall semester of 2011. The class consisted of primarily un-
dergraduates, several graduate students, and a few post-doctoral fellows.
Since the topic of modular forms has become central not only in number
theory, but also other parts of mathematics (the most spectacular being the-
oretical physics), I felt an urgency to deliver this beautiful theory to the sin-
cere student as quickly and painlessly as possible. As the topic of modular
forms can be approached from various directions, some balance had to be
achieved where only the most minimal background is assumed. The great
nineteenth century philosopher Vivekananda has said that “it is practice
first, and knowledge afterwards.” Somehow, knowledge is gained through
doing, by applying our knowledge, though it may be incomplete and our
understanding imperfect. From this perspective, I have always felt that the
best way to learn anything is via a “hands on” approach where the student
learns the nuances of the theory through practical problem solving.
My two earlier books “Problems in Analytic Number Theory” and “Prob-
lems in Algebraic Number Theory” published by Springer were written
with this underlying philosophy. Judging from the feedback I have received
from students and researchers around the world, this teaching philosophy
is well-founded and I hope the same is true with this book.
I would like to thank my two post-doctoral fellows, Michael Dewar and
Hester Graves, in assisting me in this endeavour. They not only transferred
my hand-written notes into LATEX, but added substantial sections to amplify
and elucidate the material. I also thank V. Kumar Murty, Sanoli Gun, Pu-
rusottam Rath, Alia Hamieh, and Kannappan Sampath for their valuable
comments on earlier versions of this book. I especially thank Kannappan for
his help in the production of LATEX diagrams included in this book and the
referees for their helpful comments.
Ram Murty
Kingston, Ontario
xi
Preface
The earliest murmurs of a theory of modular forms can be traced back to the
work of Jacobi in 1829 when he wrote his famous treatise Fundamenta Nova
Theoriae Functionum Ellipticarum dealing with q-series and elliptic functions.
In some parenthetic sense, this was further developed by Riemann, Hur-
witz, Dedekind, Eisenstein, and Kronecker. However, it is in the work of
Ramanujan, in his celebrated paper [29] of 1916 in which he introduced the
τ -function, where we find the seeds of a comprehensive theory. There, Ra-
manujan studied the infinite product (in the variable q) given by
∞
Y 24
q (1 − q n ) (0.1)
n=1
which he expands as an infinite series and writes

∞
X
τ (n)q n .
n=1
Ramanujan computes by hand many of these coefficients and makes the

following empirical observations about the integers τ (n):
(i) if (m, n) = 1 then τ (mn) = τ (m)τ (n),
(ii) if p is prime and α ≥ 0 then τ (pα+2 ) = τ (p)τ (pα+1 ) − p11 τ (pα ),
(iii) if p is prime then |τ (p)| ≤ 2p11/2 .
Items (i) and (ii) were proved by Mordell in 1917 [23], a year after Ra-
manujan’s paper appeared. But conjecture (iii) defied attempts by many em-
inent mathematicians until 1974, when Deligne [8] resolved it as a corollary
of his solution to the Weil conjectures. In fact, only after Ramanujan’s con-
jecture was reformulated in the context of algebraic geometry and its con-
nection to the Weil conjectures made explicit, did mathematicians realize its
central place in number theory, and mathematics in general.
xiii
xiv Preface
At first, Ramanujan’s conjectures are amazing. Why should coefficents of

the power series defined by the infinite product in (0.1) show such structure?
Why should they be multiplicative?
These questions were first addressed in a serious and fundamental way
by Erich Hecke, who in the 1930’s wrote a sequence of papers enunciating
what we now call the Hecke theory of modular forms. For the most part,
his theory was quite satisfactory in the sense that it explained conceptually
why (i) and (ii) should be true. However, when it came to (iii), the theory
suggested only a general conjecture but could not “explain” why it should
be true. Such an explanation had to await further developments in algebraic
geometry, largely due to Weil and Grothendieck.
Afterwards, in the late 1940’s, Hecke’s theory was further developed on
the one hand by Maass, who noticed the existence of a “real-analytic” coun-
terpart, and by Rankin and Selberg, who developed a theory of L-series and
obtained significant results towards conjecture (iii).
In the 1950’s and 1960’s, Harish-Chandra and, subsequently, Langlands
reformulated the notion of a modular form in the larger framework of rep-
resentation theory of Lie groups. This opened up a new universe linking
number theory and representation theory, and subsequently led to the de-
velopment of the Langlands program in the theory of automorphic repre-
sentations. For a readable account of this connection, we refer the reader to
Kumar Murty’s article “Ramanujan and Harish-Chandra” [26].
In the 1960’s and 1970’s, Hecke’s theory was extended to “higher levels,”
notably by Atkin and Lehner. At the same time, its connection to the the-
ory of elliptic curves, and more generally abelian varieties, forged the link
to arithmetic, algebraic geometry, and more precisely to the theory of Ga-
lois representations. Perhaps the most exciting event in this context was the
insight of Hellegouarch and Frey relating Fermat’s Last Theorem and mod-
ular forms. This led Serre to formulate precise conjectures that paved the
way for a method of attack on Fermat’s Last Theorem. The turning point
was when Ribet showed that if every elliptic curve “arose from a modular
form” (Taniyama’s conjecture) then Fermat’s Last Theorem follows.
Andrew Wiles recalls that when he heard this result, he set himself the
task of proving Taniyama’s conjecture. Wiles completed his proof in 1995.
His solution of Fermat’s Last Theorem required the full force of number
theory, algebraic geometry, and representation theory.
Only a special case of Serre’s conjecture was needed for Fermat’s Last
Theorem, but other cases had important consequences. A landmark theorem
is the two-dimensional reciprocity law for Galois representations. In 2008,
Khare and Wintenberger showed that every two-dimensional “odd” Galois
representation “arises from a modular form.”
This theorem can be viewed as a generalization of Wiles’ work and at
the same time, as the methods are different, it offers yet another (although
equally difficult) resolution to Fermat’s Last Theorem, which is only one of
many conjectures to which the theory of modular forms has been applied.
Preface xv
A conjecture or a problem in mathematics acts like a muse that inspires

further developments, new concepts, and a rich tapestry of fundamental
ideas. The theory of modular forms is an essential part of mathematics and
it is becoming increasingly clear that it will play a central role in the devel-
opment of twenty-first century mathematics. This is our main motivation
for the writing of this book. It is to acquaint the graduate student in a pain-
less manner to the essential ideas of the theory. At the same time, as theory
is sterile without practice, we have tried to invite and engage the student in
this topic through the problem-solving approach. Along with my other two
books, “Problems in Algebraic Number Theory” and ‘Problems in Analytic
Number Theory, ” this book should serve as a practical guide for the seri-
ous student to teach herself or himself the rudiments of number theory and
to embark in the exciting pursuit of research work in this area. At the end,
I have listed some references for further study to assist the student in this
lofty endeavour.
Kingston, Ontario Ram Murty

August 2014
About the Authors
M. Ram Murty is Professor at the Department of Mathematics and Statistics at

Queen’s University, Canada, where he is a Queen’s Research Chair in Mathematics.
He is also Professor of Philosophy at Queen’s. He was elected a fellow of the Royal
Society of Canada in 1990, the Indian National Science Academy in 2008, and won
the Coxeter–James Prize, Jeffery–Williams Prize, the E.W.R. Steacie Fellowship,
and the Killam Fellowship. His research areas include number theory, modular
forms, elliptic curves, and sieve theory. An author of over five books with Springer,
his book Non-vanishing of L-functions and Applications, coauthored by his brother
V. Kumar Murty, won the 1996 Balaguer Prize and was published by Birkhauser.
In addition, Ram is adjunct professor at McGill University; TIFR; IMSc; CMI; IIT
Bombay; IISER, West Bengal; Vivekananda University; and Harish Chandra
Research Institute, Uttar Pradesh.
Michael Dewar is a post-doctorate fellow at the Queen’s University, Canada. He
did his Ph.D. on “Ramanujan congruences in modular forms” from the University
of Illinois at Urbana-Champaign, United States. His research interest lies in mod-
ular forms, Jacobi forms, and harmonic weak Maass forms.
Hester Graves is Professor at the Department of Mathematics, University of
Michigan in Ann Arbor, United States.
xvii
Part I
Problems
Chapter 1
Jacobi’s q-series
1.1 The q-exponential function
It is not clear exactly how to define a q-series. Some experts humorously

suggest that it is any power series in q. To some extent this may be true.
However, one can say that part of the theory is connected with modular
forms and another part with combinatorics. Euler, Jacobi, Ramanujan and
many others have made expert use of q-series to derive remarkable num-
ber theoretic results ranging from the study of the partition function, the
number of representations of natural numbers as sums of squares to the de-
velopment of exotic continued fractions as is evidenced by the recent book
[7] on the subject. During the last fifty years, more connections have arisen,
most notably with Lie theory and representation theory as well as theoretical
physics. Since the subject is gaining prominence and significance, pointing
to a parallel world of mathematics in which q-analogues of classical theo-
ries exist, and since the prerequisites for the theory are minimal, we found
it fitting to introduce the reader to the world of modular forms through q-
series. Already, one sees several classical results such as Jacobi’s triple prod-
uct identity and his celebrated formula for the number of ways of writing a
natural number as a sum of four squares as immediate consequences. More-
over, as the problem of representing a natural number as a sum of k squares
(with k even) is intimately tied to the theory of modular forms of integral
weight, it seems fitting to begin with a study of how q-series can be used to
study this problem. The number of ways of writing a number as a sum of k
squares with k odd is related to the study of modular forms of half-integral
weight, which is beyond the scope of this book.
To ease the reader into this fascinating world, we begin our study with
the function
∞
X xn
Eq (x) = 1 + ,
n=1
(q n − 1)(q n−1 − 1) · · · (q − 1)
© Springer Science+Business Media Singapore 2016 and Hindustan Book Agency 2016 3
M.R. Murty et al., Problems in the Theory of Modular Forms, HBA Lecture Notes
in Mathematics, DOI 10.1007/978-981-10-2651-5_1
4 1 Jacobi’s q-series
and view it as the q-analog of the classical exponential function. We work

with Eq (x) as a formal Laurent series. If we define the empty product as 1,
then the series can be written as
∞
X xn
Eq (x) = ,
n=0
(q n − 1)(q n−1 − 1) · · · (q − 1)
It may not be clear to the reader why this should be considered as an

analog of the exponential function. The underlying “philosophy” of mov-
ing from the world of natural numbers to the q-world seems to be partially
based on the observation that
qn − 1
lim = n,
q→1 q − 1
so that one views q n −1 (or more precisely (q n −1)/(q −1)) as the q-analog of
the natural number n. Once this is understood, many of the functions in the
q-world become meaningful and exhibit remarkable structural properties.
Exercise 1.1.1. Show that

x x x
Eq (x) − Eq = Eq .
q q q
Exercise 1.1.2. Prove that if |q| > 1 then

∞
Y x
Eq (x) = 1+ .
n=1
qn
Exercise 1.1.3. Show that if |q| < 1 then

∞ ∞ n
Y
n
X q ( 2 ) xn
(1 + q x) = .
n=0 n=0
(1 − q) · · · (1 − q n )
Exercise 1.1.4. Prove that if |q| > 1 then
1
(1 + x)E q1 (x) = .
Eq (x)
Deduce that if |q| < 1 then

∞ ∞
Y X xn
(1 + q n x)−1 = .
n=0 n=0
(q n − 1) · · · (q − 1)
1.2 Jacobi’s Triple Product Identity 5
1.2 Jacobi’s Triple Product Identity
We now give a simple proof of the celebrated identity of Jacobi:

Theorem 1.2.1 (Jacobi Triple Product). If |q| < 1 and x 6= 0 then
∞ ∞
q 2n+1

Y X 2
2n+2 2n+1
(1 − q )(1 + q x) 1 + = q n xn . (1.1)
n=0
x n=−∞
Remark. According to Askey, this was also contained in some unpublished

work of Gauss (see [3]). In our proof of (1.1), we follow Andrews (see [1]).
Proof. By Exercise 1.1.3 we have, upon replacing q by q 2 and x by xq,

∞ ∞ 2
Y
2n+1
X q n xn
(1 + xq )=
n=0 n=0
(1 − q 2 ) · · · (1 − q 2n )
∞ Q∞ 2n+2j+2
n2 n j=0 (1 − q )
X
= q x Q∞ 2j+2
n=0 j=0 (1 − q )
∞ ∞ ∞
Y X 2 Y
= (1 − q 2j+2 )−1 q n xn (1 − q 2n+2j+2 ),
j=0 n=−∞ j=0
since for negative n, the product inside the sum is zero. Again by Exercise
1.1.3, replacing q by q 2 and replacing x by −q 2n+2 , we can write the product
inside the summation as
∞ 2
X (−1)m q m +m+2nm
.
m=0
(1 − q 2 ) · · · (1 − q 2m )
The sum under consideration becomes

∞ ∞ 2
X X (−1)m q m +m+2nm
n2 n
q x .
n=−∞ m=0
(1 − q 2 ) · · · (1 − q 2m )
Interchanging the sums, we get

∞ ∞
X (−1)m ( xq )m X 2
2 2m
q (m+n) xn+m .
m=0
(1 − q ) · · · (1 − q ) n=−∞
The innermost sum is

∞
X 2
q n xn
n=−∞
and can be factored out. The remaining sum is

∞ −1
Y q 2j+1
1+
j=0
x
q
by Exercise 1.1.4 (where we have set q 2 for q and x for x). Putting everything
together gives the desired result. t
u
Exercise 1.2.2 (Euler’s pentagonal number theorem). Show that if |q| < 1
then
∞ ∞
Y X k(3k−1)
(1 − q n ) = (−1)k q 2 .
n=1 k=−∞
Theorem 1.2.3 (Jacobi’s formula). If |q| < 1 then

∞ ∞
Y X k(k+1)
(1 − q n )3 = (−1)k (2k + 1)q 2 .
n=1 k=0
Proof. From the triple product identity, with x replaced by −x, we see that
∞ ∞
Y X 2
(1 − q 2n )(1 − xq 2n−1 )(1 − x−1 q 2n−1 ) = (−x)k q k .
n=1 k=−∞
The left hand side has a factor (1−xq) coming from n = 1, and consequently
vanishes when x = 1q . The same is true of the right hand side. Writing
∞
Y ∞
Y
(1 − xq 2n−1 ) = (1 − xq) (1 − xq 2n+1 ),
n=1 n=1
we obtain
∞ ∞
Y 1 X 2
(1 − q 2n )(1 − xq 2n+1 )(1 − x−1 q 2n−1 ) = (−x)k q k .
n=1
1 − xq
k=−∞
1
Putting x = q in the left hand side gives
∞
Y
(1 − q 2n )3 .
n=1
1
For the right hand side, we use l’Hopital’s rule to take the limit as x → q to
get
∞ ∞
1 X 2 X 2
− (−1)k kq 1−k+k = − (−1)k kq k −k .
q
k=−∞ k=−∞
We observe that the function f (k) = k 2 − k has the property that f (k) =
f (−(k − 1)). Thus, pairing up k and −(k − 1) in the sum, we get that it is
1.3 Jacobi’s two-square theorem 7
∞ ∞
X 2 X 2
−k −k
− {(−1)k k + (−1)k−1 (1 − k)}q k =− (−1)k (2k − 1)q k .
k=1 k=1
We can rewrite this sum as

∞
X ∞
X
(−1)k−1 (2(k − 1) + 1)q k(k−1) = (−1)k (2k + 1)q k(k+1) .
k=1 k=0
1
Changing q by q 2 gives the desired result. t
u
Exercise 1.2.4. Prove that
∞ ∞
X 2 X k(k+1)
(4n + 1)q 2n +n
= (−1)k (2k + 1)q 2 .
n=−∞ k=0
1.3 Jacobi’s two-square theorem
In this section, our goal is to obtain formulas for the number of ways a nat-
ural number m can be written as a sum of two squares. We recognize this as
the computation of the m-th coefficient in the power series expansion
∞
!2
n2
X
q .
n=−∞
We prove:
Theorem 1.3.1.
∞
!2 ∞
q 4n+1 q 4n+3

n2
X X
q =1+4 − .
n=−∞ n=0
1 − q 4n+1 1 − q 4n+3
An immediate corollary is deduced by expanding the right hand side as a

power series in q. Let rk (n) be the number of ways of writing n as a sum of
k squares.
Corollary 1.3.2. For n ≥ 1, we have r2 (n) = 4(d1 (n) − d3 (n)) where di (n) is the
number of divisors of n congruent to i (mod 4).
The following proof is due to M.D. Hirschhorn [15].
Proof of Theorem 1.3.1. In the Jacobi triple product identity, we put a4 q for x
and q 2 for q to get
∞ ∞
4n 2n2 +n
X Y
a q = (1 + a4 q 4n−1 )(1 + a−4 q 4n−3 )(1 − q 4n ).
n=−∞ n=1
Multiply both sides by a and differentiate with respect to a, to obtain

∞ ∞
X 2 Y
(4n + 1)a4n q 2n +n
= (1 + a4 q 4n−1 )(1 + a−4 q 4n−3 )(1 − q 4n )
n=−∞ n=1
∞
( )
d Y
4 4n−1 −4 4n−3 4n
+a (1 + a q )(1 + a q )(1 − q ) .
da n=1
To differentiate the product, it is useful to observe that if

∞
Y
P (a) = fn (a),
n=1
then by taking the logarithmic derivative of both sides, we get

∞
P 0 (a) X fn0 (a)
= .
P (a) f (a)
n=1 n
In our case, we get

∞ ∞
X 2 Y
(4n + 1)a4n q 2n +n
= (1 + a4 q 4n−1 )(1 + a−4 q 4n−3 )(1 − q 4n )
n=−∞ n=1
∞
( )
X 4a3 q 4n−1 4a−5 q 4n−3
× 1+a − .
n=1
1+a q4 4n−1 1 + a−4 q 4n−3
We now put a = 1 and get

∞ ∞
!
2n2 +n
X Y
4n−1 4n−3 4n
(4n + 1)q = (1 + q )(1 + q )(1 − q )
n=−∞ n=1
( ∞
) (1.2)
q 4n+1 q 4n+3
X
× 1−4 −
n=0
1 + q 4n+1 1 + q 4n+3
after changing the index of summation to start from n = 0.

By Theorem 1.2.3 and Exercise 1.2.4, the sum on the left hand side is
∞
Y
(1 − q n )3 .
n=1
This can be rewritten as

∞
Y
(1 − q 2n−1 )3 (1 − q 2n )3 .
n=1
1.4 Jacobi’s four square theorem 9
The product on the right hand side of (1.2) can be written as

∞ ∞
1 − q 4n−2 1 − q 2n
Y Y
(1 − q 4n ) = ,
n=1
1 − q 2n−1 n=1
1 − q 2n−1
so we now deduce that

∞ ∞
q 4n+1 q 4n+3
Y X
2n−1 4 2n 2
(1 − q ) (1 − q ) = 1 − 4 − .
n=1 n=0
1 + q 4n+1 1 + q 4n+3
But the left hand side is

∞
!2
n n2
X
(−1) q
n=−∞
from the triple product formula. Changing q to −q gives
∞
!2 ∞
q 4n+1 q 4n+3

n2
X X
q =1+4 − ,
n=−∞ n=0
1 − q 4n+1 1 − q 4n+3
which is the desired result. t

u
Exercise 1.3.3. Let x4 (n) be defined as follows:


0
 if n is even
x4 (n) = +1 if n ≡ 1 (mod 4)

−1 if n ≡ 3 (mod 4)

Prove that x4 (n) is completely multiplicative. That is, show that x4 (mn) =
x4 (m)x4 (n).

X
r2 (n) = 4 x4 (d).
d|n
1.4 Jacobi’s four square theorem
Using results obtained for r2 (n), we will derive an explicit formula for r4 (n),
the number of ways of writing n as a sum of four squares. This formula is
due to Jacobi who derived it using the theory of elliptic functions. Here, we
will follow a method due to Ramanujan that is completely elementary and
based on the following exercise.

1 θ 1 1
(a) cot sin nθ = + cos θ + cos 2θ + · · · + cos(n − 1)θ + cos nθ,
2 2 2 2
(b) 2(sin mθ)(sin nθ) = cos(m − n)θ − cos(m + n)θ.
qr
Exercise 1.4.2. For |q| < 1, we let ur = 1−q r so that
qr
= ur (1 + ur ).
(1 − q r )2
Prove that
∞
X ∞
X
um (1 + um ) = nun .
m=1 n=1
Exercise 1.4.3. With notation as in the previous exercise, show that

∞
X ∞
X
(−1)m−1 u2m (1 + u2m ) = (2n − 1)u4n−2 .
m=1 n=1
We are now ready to prove the following theorem due to Ramanujan.
Theorem 1.4.4. Let θ be real and not a multiple of π. Set

∞
1 θ X
L = L(q, θ) = cot + un sin nθ,
4 2 n=1
2 X ∞
1 θ
T1 = T1 (q, θ) = cot + un (1 + un ) cos nθ,
4 2 n=1
∞
1X
T2 = T2 (q, θ) = nun (1 − cos nθ),
2 n=1
qn
where un = 1−q n . Then L2 = T1 + T2 .
Before we begin the proof of this theorem, we indicate how it implies

Jacobi’s theorem on representing a natural number as a sum of four squares.
Corollary 1.4.5.
( ∞
)2 ∞
q 4n+1 q 4n+3 nq n
X X
1+4 − =1+8 .
n=0
1 − q 4n+1 1 − q 4n+3 n=1
1 − qn
n6≡0 (mod 4)
π
Proof. Put θ = 2 in Theorem 1.4.4. Then
∞
1 X qn π
L= + sin n .
4 n=1 1 − q n 2
If n is even, sin n π2 = 0 so the right hand side is

∞
1 X q 2n+1 π
+ 2n+1
sin(2n + 1) .
4 n=0 1 − q 2
Now, if n is even, sin(2n + 1) π2 = 1 and if n is odd, sin(2n + 1) π2 = −1. In

other words,
∞
q 4n+1 q 4n+3
X
4L = 1 + 4 − .
n=0
1 − q 4n+1 1 − q 4n+3
n
Since cos n π2 = 0 if n is odd and (−1) 2 if n is even, computing T1 and T2 in
Theorem 1.4.4 gives
∞
X
16T1 = 1 − 16 (−1)m−1 u2m (1 + u2m ),
m=1
∞
X ∞
X
16T2 = 8 (2m − 1)u2m−1 + 32 (2m − 1)u4m−2 .
m=1 m=1
Putting everything together and using Exercise 1.4.3 gives

∞
X
16L2 = 16(T1 + T2 ) = 1 + 8 nun ,
n=1
n6≡0 (mod 4)

u
Corollary 1.4.6. Let r4 (n) be the number of ways of writing n as a sum of four
squares. Then
X
r4 (n) = 8 d.
d|n
d6≡0 (mod 4)
In particular, r4 (n) > 0 for all n ≥ 1.

Proof. We need only invoke Theorem 1.3.1 and Corollary 1.4.5 to deduce that
∞ ∞
!4 ∞
X
n
X
n2
X nq n
r4 (n)q = q =1+8 .
n=0 n=−∞ n=1
1 − qn
n6≡0 (mod 4)
We write the right hand side as

 
∞
X ∞
X ∞
X X
q nm = 1 + 8 qn 
 
1+8 n  ,
d
n=1 m=1 n=1 d|n
n6≡0 (mod 4) d6≡0 (mod 4)
from which we deduce the desired formula for r4 (n). t

u
Proof of Theorem 1.4.4. Squaring L, we obtain

( ∞
)2
1 θ X
L2 = cot + un sin nθ
4 2 n=1
2 ∞ ∞ X ∞
1 θ 1 θ X X
= cot + cot un sin nθ + um un sin mθ sin nθ
4 2 2 2 n=1 m=1 n=1
2
1 θ
= cot + S1 + S2 ,
4 2
where ∞
X
1 θ
S1 = cot un sin nθ
2 2 n=1
and
∞ X
X ∞
S2 = um un sin mθ sin nθ.
m=1 n=1
By Exercise 1.4.1,

1 θ 1 1
cot sin nθ = + cos θ + cos 2θ + · · · + cos(n − 1)θ + cos nθ
2 2 2 2
and
2 (sin mθ) (sin nθ) = cos(m − n)θ − cos(m + n)θ,
so that
∞
X 1 1
S1 = un + cos θ + cos 2θ + · · · + cos(n − 1)θ + cos nθ
n=1
2 2
and
∞ ∞
1 XX
S2 = um un {cos(m − n)θ − cos(m + n)θ} .
2 m=1 n=1
In other words,
2 ∞
2 1 θ X
L = cot + c0 + ck cos kθ
4 2
k=1
for certain constants ck , after re-arranging S1 + S2 as a cosine series. Now

∞ ∞
1X 1X
c0 = un + u2n = nun ,
2 n=1 2 n=1
by Exercise 1.4.2. For k ≥ 1, we have

∞
1 X 1 X 1 X 1 X
c k = uk + un + um un + um un − um un ,
2 2 2 2
n=k+1 m−n=k n−m=k m+n=k
where m, n ≥ 1 in all of the summations. We may rewrite this as

∞ ∞ k−1
1 X X 1X
c k = uk + uk+j + uj uj+k − uj uk−j .
2 j=1 j=1
2 j=1
It is easily checked that
uj uk−j = uk (1 + uj + uk−j )
and
uk+j + uj uk+j = uk (uj − uk+j )
so that
 
1 ∞ k−1
X 1X 
c k = uk + (uj − uk+j ) − (1 + uj + uk−j ) .
2 2 j=1 
j=1
The first sum telescopes and we obtain

1 1
c k = uk + (u1 + u2 + · · · + uk ) − (k − 1) − (u1 + u2 + · · · + uk−1 )
2 2
1
= uk (1 + uk − k).
2
Hence
2 ∞ ∞
1 θ 1X X 1
L2 = cot + nun + uk (1 + uk − k) cos kθ
4 2 2 n=1 2
k=1
2 ∞ ∞
1 θ X 1X
= cot + uk (1 + uk ) cos kθ + kuk (1 − cos kθ),
4 2 2
k=1 k=1

u
The methods utilised above to determine formulas for r2 (n) and r4 (n)
can be extended for other values r2k (n) for certain values of k. We refer the
reader to consult [11] for more examples. In [11], the student will also find a
discussion for finding formulas for rk (n) when k is odd and this is intimately
connected with the theory of modular forms of half-integral weight which
is beyond the scope of this book.
1.5 Supplementary problems
Exercise 1.5.1. Define τ (n) by

∞
Y ∞
X
q (1 − q n )24 = τ (n)q n .
n=1 n=1
Show that τ (n) is odd if and only if n = (2m + 1)2 for some m.
Exercise 1.5.2. Let rk (n) be the number of ways of writing n as a sum of k squares.
Show that
Xn
rk (n) = ri (a)rk−i (n − a)
a=0
for any i satisfying 1 ≤ i ≤ k.
Exercise 1.5.3. Define the q-logarithm function Lq (x) as

∞
X xn
Lq (x) = .
n=1
qn−1
Suppose |x| < |q| and |q| > 1. Show that

x x
Lq (x) − Lq = .
q q−x
Deduce that
∞
X x
Lq (x) = n−x
.
n=1
q
Exercise 1.5.4. Show that if |x| < |q| and |q| > 1 then
xEq0 (−x)
Lq (x) = ,
Eq (−x)
where the derivative is with respect to x.

Chapter 2
The Modular Group
2.1 The full modular group
Exercise 2.1.1. Let R be a commutative ring with identity. Show that the set

ab a, b, c, d ∈ R
SL2 (R) = :
cd ad − bc = 1
forms a group under matrix multiplication.
The (full) modular group SL2 (Z) plays a pivotal role in the theory of mod-
ular forms. One also considers PSL2 (Z) = SL2 (Z) /{±I}. The relationship
of SL2 (Z) to SL2 (R) is similar to the relationship of Z to R in the sense that
Z is a discrete subgroup of R and SL2 (Z) is a discrete subgroup of SL2 (R).
We will show below that SL2 (Z) is generated by the elements

0 −1 11
S= and T = .
1 0 01
Theorem 2.1.2. The matrices S and T generate SL2 (Z).
Proof. Observe that

1n
Tn =
01
so that
ab a + nc b + nd
Tn = . (2.1)
cd c d
Also S 2 = −I and
ab −c −d
S = .
cd a b
16 2 The Modular Group

ab
Now let g = be any element of SL2 (Z). If c = 0, then ad = 1
cd
implies that a = d = ±1. In this case
0
1b 0
g=± = ±T b .
0 1
0 0
where b0 = ±b. Since S 2 = −I, either g = T b or S 2 T b . If c 6= 0, we proceed
as follows. Without loss of generality we may suppose |a| ≥ |c|, for oth-
erwise we can apply S to arrange this. By the Division Algorithm, we can
write a = cq + r with 0 ≤ r < |c|. Then T −q g has upper left entry r = a − qc
which is smaller than |c|. Applying S switches the rows (with a sign change)
and so we can iterate the process if r 6= 0. After a finite number of steps, we
are reduced to the case c = 0 and we are done. t
u
Exercise 2.1.3. Show that S has order 4, ST has order 6, and T has infinite order.
Exercise 2.1.4. Show that SL2 (Z) is generated by two elements of finite order,
namely S and ST of order 4 and 6 respectively.
Exercise 2.1.5. Show that any homomorphism
φ : SL2 (Z) → C×
has image contained in the finite subgroup of C× consisting of 12th roots of unity.
Exercise 2.1.6. Show that SL2 (Z) is generated by

11 10
T = and U = .
01 11
Exercise 2.1.7. Suppose that (c, d, N ) = 1. Show that there are elements c0 =
c + tN and d0 = d + sN for some integers s, t such that (c0 , d0 ) = 1.
2.2 Subgroups of the modular group
For each natural number N , the principal congruence subgroup of level N ,

denoted Γ(N ), is the group

ab ab 10
∈ SL2 (Z) : ≡ (mod N ) .
cd cd 01
In particular, Γ(1) = SL2 (Z) .
Exercise 2.2.1. Show that the natural map

2.2 Subgroups of the modular group 17
SL2 (Z) → SL2 (Z/N Z)

g 7→ g (mod N )
is a surjective homomorphism.
Exercise 2.2.2. Prove that Γ(N ) is a normal subgroup of SL2 (Z) of finite index.
Exercise 2.2.3. For any commutative ring R with identity, let GL2 (R) be the set
of 2 × 2 invertible matrices with entries in R. Show that GL2 (R) is a group.
Exercise 2.2.4. Let p be a prime. Show that the order of GL2 (Z/pZ) is given by
(p2 − 1)(p2 − p).
Exercise 2.2.5. Let p be a prime. Show that SL2 (Z/pZ) is a normal subgroup of
GL2 (Z/pZ) of index p − 1. Deduce that SL2 (Z/pZ) has order p(p2 − 1).
Exercise 2.2.6. Show that SL2 (Z/p2 Z) has size p4 (p2 − 1).
Exercise 2.2.7. Apply an induction argument to show that
SL2 (Z/pn Z)
has size p3n−2 (p2 − 1).

Y 1

3
| SL2 (Z/N Z)| = N 1− 2 .
p
p|N
Deduce that
Y 1

[SL2 (Z) : Γ(N )] = N 3 1− .
p2
p|N
A subgroup Γ ⊂ SL2 (Z) is called a congruence subgroup if Γ(N ) ⊂ Γ for

some N . Since Γ(N ) is of finite index in SL2 (Z), it follows that any congru-
ence subgroup is also of finite index in SL2 (Z) . One can ask if the converse
is true. Is every subgroup of finite index in SL2 (Z) a congruence subgroup?
The answer is no and the first counter example was given by Fricke and Pick
in 1887, independently. However, in the theory of modular forms, congru-
ence subgroups will play a dominant role. If Γ is a congruence subgroup,
the smallest N such that Γ(N ) ⊂ Γ is called the level of Γ.
It is clear from the previous discussion that more generally, we can con-
sider SLn (Z) and SLn (Z/N Z). One can prove that the natural map
φN : SLn (Z) → SLn (Z/N Z)
is again surjective and the index of ker φN in SLn (Z) can be given explicitly.
Again a subgroup Γ ⊂ SLn (Z) is called a congruence subgroup if ker φN ⊂ Γ
for some N . The group Γ is again of finite index. It is a celebrated theorem

of Bass, Lazard, and Serre that for n ≥ 3, every subgroup of finite index is
a congruence subgroup. For n ≥ 3, Bass, Milnor, and Serre extended this
result in 1967 as follows. Let K be an algebraic number field with the ring of
integers OK . All subgroups of finite index in SLn (OK )(n ≥ 3) are congru-
ence subgroups if and only if K has at least one real embedding.
Exercise 2.2.9. Show that SL2 (Z/N Z) is generated by

11 10
T = and U = ,
01 11
both of order N .
Exercise 2.2.10. Show that Γ(2) is generated by −I, T 2 , and U 2 , where

12 10
T2 = and U 2 = .
01 21
Exercise 2.2.11. In SL2 (Z), show that the group Γ generated by S and T 2 is a
congruence subgroup of index three.
2.3 The Hecke subgroups
The Hecke subgroups are denoted Γ0 (N ) and defined by

ab
Γ0 (N ) = ∈ SL2 (Z) : c ≡ 0 (mod N ) .
cd
It is easy to see that this is a group. Clearly Γ(N ) ⊂ Γ0 (N ) and so these are
congruence subgroups.
Exercise 2.3.1. Show that the map Γ0 (N ) → (Z/N Z)× given by

ab
7→ d (mod N )
cd
The kernel of the map in the previous exercise is denoted Γ1 (N ) and is

given by

ab c ≡ 0 (mod N )
Γ1 (N ) = ∈ SL2 (Z) : .
cd d ≡ 1 (mod N )
Thus we have
2.4 Groups acting on topological spaces 19
Γ(N ) ⊂ Γ1 (N ) ⊂ Γ0 (N ) ⊂ SL2 (Z) .
Exercise 2.3.2. Show that the map Γ1 (N ) → Z/N Z given by

ab
7→ b (mod N )
cd
is a surjective homomorphism with kernel Γ(N ).

1
Exercise 2.3.3. Show that Γ1 (N ) has index N 2
Q
p|N 1− p2 in SL2 (Z) .

1
Q
Exercise 2.3.4. Show that Γ0 (N ) has index N p|N 1+ p in SL2 (Z) .
2.4 Groups acting on topological spaces
Let G be a group and X a topological space. We say that there is a left action
of G on X if, for each element g ∈ G, we have a continuous map x 7→ gx ∈ X
for x ∈ X satisfying the conditions:
(a) 1x = x ∀x ∈ X
(b) (gh)x = g(hx) ∀g, h ∈ G, x ∈ X.
From these axioms, we see that x 7→ g −1 x is the inverse map of x 7→ gx.
Thus each element g ∈ G gives rise to a topological automorphism of X.
It may be clearer to define an action as a pairing G × X → X, where we
use the notation g · x to denote the image of (g, x) under this mapping. We
suppress the ‘dot’ for aesthetic reasons.
For each element x ∈ X, we define the stabilizer subgroup Gx as
Gx := {g ∈ G : gx = x}.
Exercise 2.4.1. If G acts on X and g ∈ G, x ∈ X, show that
Ggx = gGx g −1 .
An element x ∈ X is called a fixed point of g ∈ G if gx = x. This is

equivalent to saying g ∈ Gx .
For each element x ∈ X, we define the G-orbit of x to be
Gx := {gx : g ∈ G} .
The set of all G-orbits in X will be denoted G \ X. We say two elements

x, y ∈ X are G-equivalent if there is a g ∈ G such that gx = y. In particular,
y ∈ Gx. This is clearly an equivalence relation on X and we can decompose
X as a union of disjoint orbits:
a
X= Gx. (2.2)
If there is only one orbit, we say G acts transitively on X. This is equivalent

to saying that any two elements of X are G-equivalent. That is, given any
two elements x, y ∈ X, there is a g ∈ G such that gx = y.
Exercise 2.4.2. If G acts transitively on X, show that all stabilizers are conjugate.
Exercise 2.4.3. (Orbit-Stabilizer Formula) Suppose that G acts on X.

(a) If x ∈ X and g, h ∈ G, show that gx = hx if and only if g and h lie in the same
left coset of Gx .
(b) If G and X are both finite, show that |Gx| = [G : Gx ]. Deduce that
X
|X| = [G : Gxi ],
i
where the summation is over a complete set of representatives of the orbits of G.
We denote by π the surjective map X → G \ X given by
x 7→ π(x) = Gx.
In other words, we associate to each x its G-orbit. We put the strongest topol-
ogy on G \ X such that π is continuous. More precisely, a subset U of G \ X
is open if and only if π −1 (U ) is open in X. The topological space G \ X with
this topology is called the quotient space of X by G. If U is any open set of X,
we have
[
π −1 (π(U )) = gU
g∈G
is again a union of open sets. Thus π(U ) is open so that π is an open contin-
uous map of X onto G \ X.
Now suppose that G has the topological structure of a Hausdorff space
and the map G × G → G given by (g, h) 7→ gh−1 is a continuous map, we
say G is a topological group.
We say that a topological group G acts on a topological space X if, in
addition to conditions (a) and (b) defining group actions, we impose that the
map (g, x) 7→ gx is a continuous map with respect to the product topology
on G × X.
We will not prove but simply state the following important theorem. (The
reader can find a proof in Chapter 1 of [37].)
Theorem 2.4.4. Let G be a topological group that acts transitively on a topological

space X. If G is a locally compact group with a countable basis and X is a locally
compact Hausdorff space, then for each x ∈ X, the space of right cosets G/Gx is
homeomorphic to X, with the correspondence given by gGx 7→ gx.
Remark. If G acts transitively on X, this theorem allows us to view X as a

coset space G/Gx for any x ∈ X. Functions on X can be viewed as functions
on G which are constant on Gx . This point of view has advantages in the
higher dimensional theory of modular forms.
We also define a right action of a group G on a set X in which

(a) x1 = x ∀x ∈ X
(b) x(gh) = (xg)h ∀g, h ∈ G, x ∈ X.
The only difference between left and right actions is in the order in which
group elements are applied: when gh acts from the left, (gh)x = g(hx) with
h being applied first, but when gh acts from the right, x(gh) = (xg)h with g
being applied first.
Suppose H and K are subgroups
` of a group G. The left action of H parti-
tions G into right cosets: G = i Hgi . Now K acts from the right on the set
X = H\G of cosets via the rule
(Hgi )k = H(gi k).
In particular, an element k ∈ K simply permutes the right cosets.

More generally, suppose that X is a subset of G which is invariant under
the left action of H and the right action of K. That is,
∀h ∈ H, ∀k ∈ K, hX = X = Xk.
`
Now as before, X is a disjoint union of its H orbits: X = i Hxi . Moreover,
K acts on the set of orbits via
(Hxi )k = H(xi k),
where we use the fact that xi k ∈ X by the K-invariance of X. The upshot is

that k ∈ K simply permutes the orbits.
Example 2.4.5. For any g ∈ G where G is a group with subgroups H and K,

the set
X = {hgk : h ∈ H, k ∈ K}
is invariant under the left action of H and the right action of K. This set is
called a double coset and is denoted HgK. Moreover, there exist gi ∈ G such
that a
HgK = Hgi .
i
Exercise 2.4.6. Let G be a group with subgroups H and K. Suppose that X is a

subset of G which is invariant under the left action of H and the right action of K.
Show that X is a disjoint union of double cosets HgK.
Exercise 2.4.7. Suppose a group H acts on a set Y and |H\Y | < ∞. If X ⊆ Y

is invariant under the action of H and if H1 is a finite-index subgroup of H, then
prove that |H1 \X| < ∞.
Exercise 2.5.1. If Γ is a congruence subgroup of level N and γ ∈ SL2 (Z), show

that γΓγ −1 is again a congruence subgroup of level N .
Γ(M ) ∩ Γ(N ) = Γ(lcm(M, N )).
Γ(M )Γ(N ) = Γ(gcd(M, N )).
Exercise 2.5.4. Define

1 10
Γ (N ) = g ∈ SL2 (Z) : g ≡ (mod N ) .
?1
Show that Γ1 (N ) is a congruence subgroup of level N .
Exercise 2.5.5. If SL2 (Z) has a non-congruence subgroup, show that it has in-
finitely many. (Hint: consider Γ ∩ Γ (p) for a non-congruence subgroup Γ
and rational primes p.)
Exercise 2.5.6. For each g ∈ SL2 (Z) and a subgroup Γ of finite index in SL2 (Z),
we define ng (Γ ) to be the least positive integer such that

11
T ng (Γ ) ∈ gΓ g −1 , T = as usual.
01
We say Γ is of class N if the least common multiple of the set of numbers {ng (Γ ) :
g ∈ SL2 (Z)} equals N . Show that Γ (N ) is of class N . Also show that the normal
closure of T N , namely
D(N ) := hgT N g −1 : g ∈ SL2 (Z)i
is of class N . (This exercise extends the notion of congruence level defined for
congruence subgroups to all subgroups of finite index, and is due to Fricke
and Wohlfahrt.)
Exercise 2.5.7. (a) Using the Chinese remainder theorem (or otherwise), show that
for given x, y, the following congruence
2.5 Supplementary problems 23
x(αy + β) ≡ γ mod m
has a solution in integers α, β, γ such that (α, β, m) = 1.

(b) Using (a), show that given g ∈ Γ (N ), there exists
0 0
a b
g0 = ∈ Γ (N )
c 0 d0
such that b0 ≡ c0 ≡ 0 mod mN and g 0 ∈ D(N )gD(N ) with D(N ) defined

in Exercise 2.5.6. (Hint: consider conjugates of T by the transpose of T
that is in D(N ).)
Exercise 2.5.8. With notation as in Exercise 2.5.6, show that D(N )Γ (M N ) =

Γ (N ) for all positive integers M, N .
Exercise 2.5.9. Let Γ be a subgroup in SL2 (Z) of class N . Show that Γ is a con-
gruence subgroup if and only if it is a congruence subgroup of level N . Deduce that
if Γ is a subgroup of SL2 (Z) of class N and the index [SL2 (Z) : Γ ] does not divide
[SL2 (Z) : Γ (N )], then Γ is not a congruence subgroup.
Exercise 2.5.10. Let F be a free group on n generators, say x1 , ..., xn . For a word
w ∈ F , define ì (w) to be the sum of all the exponents of xi occuring in w. For a
subset I of {1, ..., n}, and a positive integer `, define
Γ` (I) = {w ∈ F : ì (w) ≡ 0 mod ` ∀i ∈ I}.
Show that Γ` (I) is a normal subgroup of F of index `|I| .
Exercise 2.5.11. Recall that Γ (2) is generated by −I, T 2 , U 2 (see Exercise 2.2.10)
and that the subgroup generated by x1 = T 2 and x2 = U 2 is free of index 2 in
Γ (2).
(a) With notation as in the previous exercise, compute the class of the group
Γ` ({1}) and Γ` ({1, 2}).
(b) Show that Γ` ({1, 2}) is not a congruence subgroup if ` is not a power of 2.
Chapter 3
The Upper Half-Plane
3.1 The group GL+

n (R) and its subgroups
Let H denote the upper half-plane,
H = {z ∈ C : Im(z) > 0},
viewed as an open subset of C with the usual topology. We define an action

of the group
+ ab a, b, c, d ∈ R
GL2 (R) = :
cd ad − bc > 0
on H via the formula
ab az + b
z= . (3.1)
cd cz + d
Exercise 3.1.1. Show that (3.1) defines an action of GL+
2 (R) on H.
The transformations
az + b
z 7→
cz + d
are called fractional linear transformations or Möbius transformations. Note that
the scalar matrices act trivially on H.
Exercise 3.1.2. Show that the group PSL2 (R) = SL2 (R)/ ± I acts faithfully on
H. That is, gz = z for all z ∈ H implies g = 1 ∈ PSL2 (R).
Exercise 3.1.3. Show that the action of PSL2 (R) on H is transitive, i.e. there is
only one orbit and for all z, w ∈ H, there exists some g ∈ PSL2 (R) such that
gz = w.
Exercise 3.1.4. Show that the map
z−i
z 7→
z+i
26 3 The Upper Half-Plane
is an analytic isomorphism of the upper half-plane H and the unit disc D = {w ∈

C : |w| < 1}.
Theorem 3.1.5. The group of analytic automorphisms of H is isomorphic to PSL2 (R).
Proof. Let φ : H → H be an analytic automorphism of H. Write φ(i) = a + bi

with a, b ∈ R. Then the map
φ(z) − a
ψ(z) =
b
maps i to i. But now, let h : H → D be given by

z−i 1 −i
h(z) = = z.
z+i 1 i
By Exercise 3.1.4, this is an analytic isomorphism of H onto the unit disc

D = {z ∈ C : |z| < 1}. Put g(z) = (h◦ψ ◦h−1 )(z). Then g : D → D is analytic
with g(0) = 0. By the Schwarz Lemma of complex analysis we deduce that
|g(z)| ≤ |z|. The Schwarz Lemma applied to the inverse map of g, namely
h ◦ ψ −1 ◦ h−1 , shows that |z| = |g −1 (g(z))| ≤ |g(z)|. Hence |g(z)| = |z| and
by the Schwarz Lemma g(z) = eiθ z for some θ ∈ R. In particular,
iθ
e 0
g(z) = z.
0 1
From the solution to Exercise 3.1.4,

i(1 + z) i i
h−1 (z) = = z,
1−z −1 1
and so
ψ(z) = h−1 ◦ g ◦ h (z)

iθ
i i e 0 1 −i
= z
−1 1 0 1 1 i
iθ
ie + i eiθ − 1

= z.
−eiθ + 1 ieiθ + i
Matrices representing fractional linear transformations may be rescaled with-

out changing the corresponding function. Multiply every entry by e−iθ/2 /2i
to obtain
cos(θ/2) sin(θ/2)
ψ(z) = z.
− sin(θ/2) cos(θ/2)
Thus ψ ∈ PSL2 (R). On the other hand, φ(z) can be written as
3.2 Fundamental domains 27
√ a
! √ a
!
θ θ

b √ b √ cos
φ(z) = 1
b ψ(z) = 1
b 2
θ
sin 2
θ z,
0 √
b
0 √
b
− sin 2 cos 2
and hence φ ∈ PSL2 (R). t

u
If G is a topological group, a subgroup Γ of G is called a discrete subgroup
if any sequence of elements γn ∈ Γ with γn → 1 is eventually constant.
Alternately, every neighborhood of the identity contains only finitely many
elements of Γ.
Exercise 3.1.6. Show that SL2 (Z) is a discrete subgroup of SL2 (R) ⊂ R4 , where
R4 is endowed with the Euclidean topology.
3.2 Fundamental domains
If Γ is a subgroup of SL2 (Z) and F ⊂ H is a closed set with connected inte-

rior, we say that F is a fundamental domain for Γ (or Γ\H) if
a) any z ∈ H is Γ-equivalent to a point in F ;
b) no two interior points of F are Γ-equivalent;
c) the boundary of F is a finite union of smooth curves.
Example 3.2.1. Let Γ∞ be the subgroup of SL2 (Z) given by

1n
Γ∞ = :n∈Z .
01
A fundamental domain for Γ∞ \H is given by

1 1
F = z ∈ H : − ≤ Re(z) ≤ .
2 2
Indeed, given any z ∈ H, we can apply a transformation γ of Γ∞ to ensure

γz ∈ F . If two interior points z, z 0 of F were equivalent, then z 0 = z + n
for some integer n. But then Re(z 0 ) = Re(z) + n forces n = 0. Finally, the
boundary of F in H consists of the two vertical lines Re(z) = −1/2 and
Re(z) = 1/2, both of which are smooth curves.
We will now construct what is called the standard fundamental domain for
the action of SL2 (Z) on H.
Theorem 3.2.2. Let

1
F = z ∈ H : | Re(z)| ≤ , |z| ≥ 1 .
2
Then F is a fundamental domain for the action of SL2 (Z) on H.

ρ
ρ2 i
−1 − 12 0 1 1
2
Fig. 3.1 Fundamental Domain for SL2 (Z)
To prove this, we will first need the following lemma.

Lemma 3.2.3. Fix z ∈ H. The set of (c, d) ∈ Z × Z such that (c, d) 6= (0, 0) and
|cz + d| ≤ 1 is finite and non-empty.
Proof. We can write z as x + iy. We know that |cz + d| ≤ 1 if and only if
(cx + d)2 + c2 y 2 ≤ 1, which implies that c2 y 2 ≤ 1. Equivalently, |c| ≤ 1/y, so
there are only finitely many possibilities for c.
On the other hand, since |cx + d| ≤ 1, we have −1 ≤ cx + d ≤ 1, implying
that −1 − cx ≤ d ≤ 1 − cx. Thus we conclude there are only finitely many
possibilities for d.
Finally, it is clear that (c, d) = (0, 1) satisfies the requirements. t
u
Proof of Theorem 3.2.2. The boundary of F is clearly a union of smooth curves
and so condition (c) in the definition of a fundamental domain is satisfied.
We need only check (a) and (b). To show (a), we fix z ∈ H and observe that
for any given
ab
γ= ∈ SL2 (Z),
cd
we have
Im(z)
Im(γz) = . (3.2)
|cz + d|2
Since c, d ∈ Z and are not both zero, the function |cz + d|2 attains a posi-
tive minimum as γ ranges over elements of SL2 (Z). In fact, by Lemma 3.2.3
above, there are only finitely many c, d ∈ Z such that |cz + d| ≤ 1. Thus we
may choose γ ∈ SL2 (Z) so that |cz + d| is minimal and thus Im (γz) is max-
imal. By translating z by an integer n (which is equivalent to T n z), we may
3.2 Fundamental domains 29
suppose that | Re(z)| ≤ 1/2. We claim |γz| ≥ 1 so that γz ∈ F . Indeed, if we

suppose otherwise, writing w = γz, we have |w| < 1 so that Sw = −1/w is
such that Im(Sw) = Im(w)
|w|2 > Im(w). This means Im(Sγz) > Im(γz), which
is a contradiction to our choice of γ. This proves that every element of H is
SL2 (Z)-equivalent to some point in F .
It only remains to prove (b), i.e. to show that no two interior points of F
are SL2 (Z)-equivalent. Indeed, suppose
that z, w ∈ F are interior points and
that γz = w for some γ = ac db ∈ SL2 (Z). Without loss of generality, we
may suppose that Im(w) ≥ Im(z). This implies √ |cz + d| ≤ 1. In particular,
| Im(cz + d)|√= |c| Im(z) ≤ 1. Since Im(z) ≥ 3/2 for all z ∈ F , this means
that |c| ≤ 2/ 3. Thus c = 0 or c = ±1.
If c = 0, then ad = 1. Multiplying γ by −I if necessary, we may suppose
a = d = 1. Thus γ is a power of T and it is clear that if z ∈ F , then γz ∈ /
F unless z lies on the boundary. If c = ±1, we may multiply γ by −I if
necessary and assume c = 1. Then, if z = x + iy, we have
3
1 ≥ |cz + d|2 = (x + d)2 + y 2 > (x + d)2 +
4
so that |x+d| < 1/2. Since |x| < 1/2, this means d = 0. But now 1 ≥ |cz+d| =
|z| which implies that z is not an interior point of F . t
u
Exercise 3.2.4. Suppose that z, w ∈ F are distinct points which are SL2 (Z)-
equivalent. Show that Im(z) = Im(w) and Re(z) = − Re(w).
Exercise 3.2.5. If F is the standard fundamental domain for SL2 (Z) and γ ∈
SL2 (Z), show that γF is again a fundamental domain for SL2 (Z).
One can prove that any discrete subgroup Γ of SL2 (R) has a fundamental
domain (see Theorem 1.6.2 of [22]). The next theorem allows us to construct
fundamental domains in certain cases.
Theorem 3.2.6. Let F be a fundamental domain for SL2 (Z) and let Γ be a congru-
ence subgroup
Sr with −I ∈ Γ. If g1 , . . . , gr are coset representatives of Γ such that
D = i=1 gi−1 F has a connected interior, then D is a fundamental domain for Γ.
Proof. It is clear that the boundary of each gi−1 F consists of a finite number
of smooth curves so that condition (c) in the definition of a fundamental
domain is satisfied. We need to verify (a) and (b). Now any z ∈ H is SL2 (Z)-
equivalent to some point w ∈ F . Thus w = τ z with τ ∈ SL2 (Z). We can
write τ = gi γ for some i and γ ∈ Γ. Hence γz ∈ gi−1 F so that γz ∈ D and (a)
is satisfied.
Now suppose two interior points of D are Γ-equivalent. In other words,
there is some γ ∈ Γ, and some interior point z ∈ D such that γz ∈ D is also
an interior point. The action of γ is a homeomorphism and so there exists an
open set U ⊂ D containing z such that γz ∈ γU ⊂ D. By the construction of
D, there exist indices 1 ≤ i, j ≤ r such that gi z, gj γz ∈ F . Let V = gi U ∩ F ◦ ,
where F ◦ denotes the interior of F . The set V is open and non-empty. Take
any x ∈ V. Then
γgi−1 x ∈ γgi−1 V ⊆ γU ⊂ D.

Let k be such that γgi−1 x ∈ gk−1 F . Now x and gk γgi−1 x are SL2 (Z)-

equivalent interior points of F . Hence they are equal. The map w 7→ gk γgi−1 w
is an automorphism of H. Thus, there is an open set W ⊂ F ◦ containing x
such that gk γgi−1 W ⊂ F ◦ . But then every element of W is SL2 (Z)-equivalent
to some element of F ◦ . This can only happen if gk γgi−1 acts trivially on W.
As any two holomorphic functions agreeing on an open set must agree ev-
erywhere, we deduce that gk γgi−1 = ±I by Exercise 3.1.2. Now
gi Γ = gk γ{±I}Γ = gk Γ.
As the gi are distinct coset representatives, we conclude that gi = gk and so

γ = ±I. In particular, z = γz. This proves (b). t
u
Exercise 3.2.7. Show that I, S, T −1 S are coset representatives for Γ0 (2) in SL2 (Z).
Deduce that a fundamental domain for Γ0 (2) is given by F ∪ SF ∪ ST F , where F
is the standard fundamental domain for SL2 (Z).
3.3 The extended upper half-plane
We define the extended upper half-plane H? as
H? = H ∪ Q ∪ {i∞}.
That is, H? is obtained by adjoining all the rational numbers and i∞, called
the cusps. We topologize H? in the following way. For z ∈ H, we take the
usual fundamental system of neighborhoods. For a cusp s 6= i∞, we take all
sets of the form
{s} ∪ C ◦ ,
where C ◦ is the interior of a circle in H tangent to the real axis at s. If s = i∞,
we take all sets of the form
{i∞} ∪ {z ∈ H : Im(z) > c},
for all c > 0, as a fundamental system of open neighborhoods of i∞. This

makes H? into a Hausdorff space.
We define an action of SL2 (Z) on H? as follows. For z ∈ H, the action is
defined via the usual formula

ab az + b
z= .
cd cz + d
3.3 The extended upper half-plane 31
For the cusps Q ∪ {i∞}, we identify i∞ with 10 and define

ab r ar + bs
= .
cd s cr + ds
The reader may verify directly that this is indeed an action.

Exercise 3.3.1. Show that every cusp is SL2 (Z)-equivalent to i∞.
For any subgroup Γ of SL2 (Z) of finite index, we can define a topology
on the orbit space Γ\H? . Indeed, we have the natural map
π : H? −→ Γ\H? ,
sending each point z ∈ H? to its orbit under Γ. We impose the strongest

topology to ensure π is continuous. In other words, the open sets of Γ\H?
are
{U ⊆ Γ\H? : π −1 (U ) is open in H? }.
This makes Γ\H? into a Hausdorff space (see Theorem 1.28 of [37], p. 12). If
Γ = SL2 (Z), then Γ\H? has the complex analytic structure of the Riemann
sphere. More generally, if Γ is a congruence subgroup, Γ\H? is a compact
Riemann surface.
Exercise 3.3.2. If Γ is of finite index in SL2 (Z), show that the number of Γ-
inequivalent cusps is ≤ [SL2 (Z) : Γ].
Exercise 3.3.3. Show that {0, i∞} is a complete set of inequivalent cusps of Γ0 (2).
Exercise 3.3.4. With the action of SL2 (Z) on H? , show that the stabilizer Γi∞ of
i∞ is given by
1n
± :n∈Z .
01
Exercise 3.3.5. Let p be prime. Show that the matrices

k 0 −1
ST = , 0 ≤ k ≤ p − 1,
1 k
along with ( 10 01 ), give a complete set of right coset representatives for Γ0 (p) in
SL2 (Z).
Exercise 3.3.6. Let p be prime. Show that the set of matrices

k 1
, 0 ≤ k ≤ p − 1,
−1 0
along with ( 10 01 ), is a complete set of left coset representatives for Γ0 (p).

Exercise 3.3.7. Let p be prime. Show that a complete set of inequivalent cusps for
Γ0 (p) is given by {0, i∞}.
3.4 The Poincaré metric
Since the topology on H is the subspace topology inherited from C, and as

C is a metric space with the usual metric, H is also a metric space. However,
this metric cannot be lifted to Γ\H with Γ being a subgroup of PSL2 (R) since
it is not PSL2 (R) invariant. Poincaré defined a metric on H which is SL2 (R)-
invariant and lifts to Γ\H. The Poincaré metric on H is given by the distance
function (also called the hyperbolic metric)
|z − w|2

−1
d(z, w) = cosh 1+ ,
2 Im(z) Im(w)
where z, w ∈ H and cosh−1 is the inverse function of the hyperbolic cosine.

Notice that this gives rise to the same topology on H as before. The fact
that d(z, w) is a metric on H is easily seen from simple differential geometric
considerations which we shall not discuss here.
Exercise 3.4.1. Prove that d(z, w) is SL2 (R)-invariant. That is, show that d(gz, gw) =
d(z, w) for all g ∈ SL2 (R).
Exercise 3.4.2. Show that for x ≥ 1,

p
cosh−1 x = log(x + x2 − 1).
Exercise 3.4.3. Show that for y > 0,
d(i, iy) = | log y|.
One can also define an SL2 (R)-invariant measure on H which lifts to Γ\H?
for any subgroup Γ of SL2 (R). This is given by
dxdy
dµ := .
y2
Theorem 3.4.4. The measure dµ is SL2 (R)-invariant on H.
Proof. We need to check that
dµ(gz) = dµ(z) for all g ∈ SL2 (R).

Indeed, let z = x + iy and g = ac db ∈ SL2 (R) so that
(ad + bc)x + bd + ac(x2 + y 2 ) + iy

gz =
c2 (x2 + y 2 ) + 2cdx + d2
= u(x, y) + iv(x, y) (say).
By the change of variable formula in Rn (see p. 67 of [42]), we have

dudv = (detJ)dxdy,
where J is the Jacobian matrix

!
∂u ∂u
∂x ∂y
∂v ∂v .
∂x ∂y
A tedious calculation shows that det J = (1/y)2 , from which the result fol-
lows. t
u
Exercise 3.4.5. Let F be the standard fundamental domain for SL2 (Z). Show that
Z Z
dxdy π
2
= .
F y 3
Exercise 3.5.1. Show that a set of coset representatives for Γ∞ in SL2 (Z) is given
by
??
: (c, d) = 1, c, d ∈ Z ,
cd
where the asterisks indicate that we can take any a, b such that ad − bc = 1.
Exercise 3.5.2. Let F , Γ, g1 , . . . , gr and D be as in Theorem 3.2.6. Show that

Z Z
dxdy π
2
= [SL2 (Z) : Γ] .
D y 3
Exercise 3.5.3. Let G = SL2 (R) act on H. Show that K, the stabilizer of i, consists
of matrices
cos θ sin θ
for θ ∈ R.
− sin θ cos θ
Then show that the upper half-plane is homeomorphic to G/K.
Chapter 4
Modular Forms of Level One
4.1 Fourier expansions and q-series
We are interested in holomorphic functions f (z) : H → C for which

az + b k ab
f = (cz + d) f (z), ∀ ∈ SL2 (Z). (4.1)
cz + d cd
We will use the topology on the extended upper half-plane H? to define

what it means for f (z) to be holomorphic at the cusp i∞.
Exercise 4.1.1. Show that the punctured fundamental neighborhood
Uc = {z ∈ H : Im(z) > c}
of i∞ is mapped via z 7→ e2πiz into the punctured open disc centered at zero, of
radius e−2πc .
The function q(z) = e2πiz from Exercise 4.1.1 can be extended (continu-
ously) by sending i∞ to 0. Hence, the fundamental neighborhoods of i∞ are
mapped to the fundamental neigborhoods of the origin.
Since T = ( 10 11 ) ∈ SL2 (Z), condition (4.1) implies that f (z + 1) = f (z) for
all z ∈ H. Hence, there is a well-defined induced map, which we also call f ,
from the unit disc to C:
q 7→ f (z),
where z ∈ H is any point with q = e2πiz . If f (z) : H → C is holomorphic,
then f (q) will be holomorphic on the punctured unit disc and hence will
have a Laurent expansion centered at q = 0,
∞
X
f (q) = an q n ,
n=−∞
which is called a q-expansion. We call
36 4 Modular Forms of Level One
∞
X
f (z) = an e2πizn
n=−∞
the Fourier series at i∞. We define the order of f at i∞ to be the index of the
first non-zero coefficient in the q-series:
vi∞ (f ) := inf {n ∈ Z : an 6= 0} .
If −∞ < vi∞ (f ) < 0, then f is said to be meromorphic at i∞ with a pole

of order |vi∞ (f )|. If vi∞ (f ) ≥ 0 then f is said to be holomorphic at i∞. If
vi∞ (f ) > 0 then we say that f vanishes at i∞ and has a zero of order vi∞ (f ).
A (holomorphic) modular form of weight k ∈ Z for the full modular group
SL2 (Z) is a holomorphic function f (z) : H → C which satisfies (4.1) and
which is holomorphic at i∞. A modular form on SL2 (Z) is called a cusp form
if it vanishes at the cusp i∞, i.e. a0 = 0 in the Fourier expansion at i∞. In this
chapter, we will write f (i∞) = a0 . Unfortunately, the value at i∞ will not
be well-defined for the modular forms of higher level which we will meet in
Chapter 6.
Exercise 4.1.2. Let γ = ac db ∈ GL+

2 (R) and z ∈ H. Define j(γ, z) = cz + d.
Show that
j(γ1 γ2 , z) = j(γ1 , γ2 z)j(γ2 , z).
For γ ∈ GL+2 (R), it is convenient to introduce the “slash” notation: for

any holomorphic function f on H,
(f |γ)(z) := (det γ)k/2 j(γ, z)−k f (γz). (4.2)
Then the modularity condition (4.1) is equivalent to
f |γ = f, ∀γ ∈ SL2 (Z).
Exercise 4.1.3. Show that f |(γ1 γ2 ) = (f |γ1 )|γ2 for all γ1 , γ2 ∈ GL+
2 (R).
Exercise 4.1.4. Prove that if f : H → C is holomorphic, satisfies

(i) f (z +1) = f (z)
(ii) f − z1 = z k f (z)
for all z ∈ H, and is holomorphic at i∞, then f is a modular form of weight k for
SL2 (Z).
Let Mk (SL2 (Z)) denote the set of all modular forms of weight k on
SL2 (Z). Let Sk (SL2 (Z)) ⊆ Mk (SL2 (Z)) denote the set of cusp forms of
weight k on SL2 (Z). The use of the letter S for the space of cusp forms comes
from the German word “Spitzenform” which means cusp form.
Exercise 4.1.5. Let k1 , k2 ∈ Z. Show that:

4.2 Eisenstein series for SL2 (Z) 37
(a) Both Mk1 (SL2 (Z)) and Sk1 (SL2 (Z)) are C-vector spaces.
(b) If f ∈ Mk1 (SL2 (Z)) and g ∈ Mk2 (SL2 (Z)), then f g ∈ Mk1 +k2 (SL2 (Z)).
(c) If f ∈ Mk1 (SL2 (Z)) and g ∈ Sk2 (SL2 (Z)), then f g ∈ Sk1 +k2 (SL2 (Z)).
Later, we will see that Mk (SL2 (Z)) and Sk (SL2 (Z)) are finite-dimensional
vector spaces. Consequently, methods of linear algebra can be applied to
study them and to construct “nice” bases for Mk (SL2 (Z)) and Sk (SL2 (Z)).
The previous exercise also allows us to speak about a graded C-algebras of
holomorphic modular forms and holomorphic cusp forms.
Exercise 4.1.6. Show that Mk (SL2 (Z)) = {0} for k odd.
4.2 Eisenstein series for SL2 (Z)
We begin with a very simple observation. Suppose we are given a binary

quadratic form
Q(m, n) = Am2 + 2Bmn + Cn2 , A, B, C ∈ R,
which is positive definite. That is, A > 0, and AC − B 2 > 0. We may write
Q(m, n) in matrix form:

AB m
Q(m, n) = (m, n) .
BC n
A B ) is a real symmetric matrix, we can diagonalize it.
Since the matrix ( B C
Thus, we can find an orthogonal matrix P such that

AB λ1 0
P Pt = ,
BC 0 λ2
and if we put
u m
=P ,
v n
then we can write Q(m, n) = λ1 u2 + λ2 v 2 . Since Q is positive definite, the
eigenvalues λ1 , λ2 are real and positive. We see that
Q(m, n) ≥ λ(u2 + v 2 ),
where
λ = min(λ1 , λ2 ) > 0.
Moreover, P is an orthogonal transformation and thus preserves length.
Consequently, u2 + v 2 = m2 + n2 . This proves that
Q(m, n) ≥ λ(m2 + n2 )
for some λ > 0. If z is a complex number, then |mz + n|2 is a positive definite
quadratic form in m, n.
These elementary observations from linear algebra allow us to construct
an important family of modular forms. For k > 2, define
X0 1
Gk (z) = , z ∈ H,
m,n
(mz + n)k
where the summation is over all (m, n) ∈ Z2 , with (m, n) 6= (0, 0). The dash
on the summation indicates this latter condition.
Exercise 4.2.1. Show that the series for Gk (z) converges absolutely for all z ∈ H
and for k > 2.
By the Weierstrass M-test, we see that Gk (z) defines a holomorphic func-

tion for z ∈ H. Its behavior at i∞ is easily seen to come from the terms when
m = 0. Thus X 1
Gk (i∞) = .
nk
n∈Z
n6=0
Observe that Gk (z) is identically zero if k is odd as is easily seen by pairing

up the summands corresponding to (m, n) and (−m, −n). However, for k
even, Gk (z) is a non-trivial function and
Gk (i∞) = 2ζ(k),
where ζ(s) denotes the Riemann zeta function. In addition, Gk (z) is a mod-
ular form of weight k since
Gk (z + 1) = Gk (z),

1
Gk − = z k Gk (z),
z
as is easily verified. Indeed,

X0 1
Gk (z + 1) = ,
m,n
(mz + m + n)k
and as (m, n) ranges over elements of Z2 \ {(0, 0)}, so does (m, m + n). Sim-
ilarly,
1 X0 1
Gk − = zk k
= z k Gk (z)
z m,n
(nz − m)
since (n, −m) ranges over elements of Z2 \ {(0, 0)} as (m, n) ranges over
Z2 \{(0, 0)}. By Exercise 4.1.4, we see that Gk (z) is a modular form of weight
k for SL2 (Z).
To study the behavior of Gk (z) near i∞, we now derive its q-expansion.
Recall
∞
z2
Y
sin πz = πz 1− 2 .
n=1
n
Taking the logarithmic derivative gives

∞
1 X 1 1
π cot πz = + + , (4.3)
z n=1 z+n z−n
which is easily verified to be an absolutely convergent series. On the other

hand, for any z ∈ H we have
cos πz eiπz + e−iπz

π cot πz = π = πi iπz
sin πz e − e−iπz
∞ (4.4)
q+1 2πi X
= πi = πi − = πi − 2πi qn ,
q−1 1−q n=0
where q = e2πiz . By successive differentiation of (4.3), we deduce:

Theorem 4.2.2 (Lipschitz formula). For k ≥ 2 and for q = e2πiz where z ∈ H,
we have
∞
X 1 (−2πi)k X k−1 n
= n q .
(z + n)k (k − 1)! n=1
n∈Z
This allows us to prove:

Theorem 4.2.3. For every even k ≥ 4,
∞
2(−2πi)k X
Gk (z) = 2ζ(k) + σk−1 (n)q n ,
(k − 1)! n=1
where q = e2πiz and σs (n) = ds .

P
d|n
Proof. We have, by separating m = 0 from the sum defining Gk (z),

∞ X
X 1
Gk (z) = 2ζ(k) + 2 .
m=1 n∈Z
(mz + n)k
We can expand the inner sum using the Lipschitz formula (Theorem 4.2.2):
∞ ∞
(−2πi)k X X k−1 mn
Gk (z) = 2ζ(k) + 2 n q .
(k − 1)! m=1 n=1
Collecting terms for which mn = r, we see the coefficient of q r is σk−1 (r).

t
u
Exercise 4.2.4. Define the Bernoulli numbers Bk by the power series
X Bk xk ∞
x
x
= ,
e −1 k!
k=0
so that B0 = 1, B1 = − 12 , B2 = 16 , B3 = 0, B4 = − 30
1
, . . . . Show that
(2πi)2k B2k
2ζ(2k) = − , k ≥ 1.
(2k)!
Exercise 4.2.5. Prove that the Bernoulli numbers satisfy the recursion formula:
m−1
X
m+1
(m + 1)Bm = − Bk .
k
k=0
Exercise 4.2.6. Define the n-th Bernoulli polynomial by

n
X n
Bn (x) = Bk xn−k .
k
k=0
Prove that
∞
text X xn
t
= Bn (x) .
e − 1 n=0 n!
Using Exercise 4.2.4, we obtain:
Corollary 4.2.7. For every even k ≥ 4,
∞
2k X
Ek (z) = 1 − σk−1 (n)q n
Bk n=1
is a modular form of weight k for the full modular group SL2 (Z) satisfying
Ek (i∞) = 1.
Proof. Since Gk (z) = 2ζ(k)Ek (z), we can use Exercise 4.2.4 and Theorem
4.2.3 and the result is immediate. t
u
Here are some examples:
∞
X ∞
X
E4 (z) = 1 + 240 σ3 (n)q n , E10 (z) = 1 − 264 σ9 (n)q n ,
n=1 n=1
∞ ∞
X 65520 X
E6 (z) = 1 − 504 σ5 (n)q n , E12 (z) = 1 + σ11 (n)q n ,
n=1
691 n=1
X∞ ∞
X
E8 (z) = 1 + 480 σ7 (n)q n , E14 (z) = 1 − 24 σ13 (n)q n .
n=1 n=1
4.3 The valence formula 41
Exercise 4.2.8. Prove that Mk (SL2 (Z)) = CEk ⊕ Sk (SL2 (Z)) for all k ≥ 4.
Deduce that dim Mk (SL2 (Z)) = 1 + dim Sk (SL2 (Z)).
It is easily verified that
E43 − E62
∆ :=
1728
is a non-trivial cusp form of weight 12 for the full modular group. The
Fourier coefficients of ∆ define the celebrated Ramanujan τ -function:
∞
X
∆(z) = τ (n)q n = q − 24q 2 + 252q 3 + · · · .
n=1
Later, we will see that

∞
Y
∆(z) = q (1 − q n )24 .
n=1
4.3 The valence formula
Given a meromorphic function f : H → C, not identically zero, and z0 ∈ H,

there is a unique integer n such that f (z)/(z − z0 )n is holomorphic and non-
zero at z0 . We call n the order of f at z0 and denote it by vz0 (f ). If f is
holomorphic, then vz0 (f ) is the order of the zero of f at z0 .
If f is a modular form of weight k for the full modular group, the identity

az + b
f (z) = (cz + d)−k f
cz + d
shows that vz0 (f ) = vgz0 (f ) for all g ∈ SL2 (Z). In other words, vz0 (f ) de-
pends only on the orbit of z0 under SL2 (Z) so that we need only study vz (f )
for z in a fundamental domain of SL2 (Z). Recall that we have defined vi∞ (f )
to be the order of the zero at q = 0 in the q-expansion
∞
X
f (z) = an q n .
n=0
That is, vi∞ (f ) is the smallest value of n such that an 6= 0.

Let us note that f ∈ Mk (SL2 (Z)) has only a finite number of zeros in
the standard fundamental domain F since f being holomorphic on H im-
plies that the zeros are isolated. Moreover, for some c > 0, the region
{z ∈ F : Im(z) > c} ∪ {i∞}, being a fundamental neighborhood of i∞,
contains no zero of f except possibly i∞. All remaining zeros are contained
in the compact region {z ∈ F : Im(z) ≤ c} and this number is finite (see, for
example, Theorem 10.18 on page 208 of [33]).
Theorem 4.3.1 (The valence formula). Let f be a modular form, not identically
zero, of weight k for the full modular group SL2 (Z). Then
1 1 X0 k
vi∞ (f ) + vi (f ) + vρ2 (f ) + vz (f ) = ,
2 3 2
12
z6=i,ρ
z∈F
1
where the primed summation excludes points with Re z = 2 and points with both
√ πi
|z| = 1 and Re z > 0. Here i = −1 and ρ = e 3 .
We use primed summation in the valence formula to account for the fact
that half of the boundary of F is SL2 (Z)-equivalent to the other half.
Proof. We first give the proof in the case that f has no zeros on the bound-
ary of the standard fundamental domain F in H. Then, we indicate how to
modify the proof so as to treat the case when some zeros lie on the boundary.
iT
ρ
ρ2 i
−1 − 12 0 1 1
2
Fig. 4.1 Contour without zeros on ∂F
We choose T large enough so that all zeros z0 of f have Im(z0 ) ≤ T , apart

from possibly z0 = i∞. We integrate f 0 /f around the contour C indicated in
Figure 4.1 . By Cauchy’s argument principle,
Z 0
1 f (z) X
dz = vz (f )
2πi C f (z)
z∈F
The contour has several segments. Since f (z + 1) = f (z), the two vertical
segments cancel each other. The horizontal segment maps (under z 7→ q =
e2πiz ) into the circle of radius e−2πT about q = 0 oriented clockwise and thus
contributes −vi∞ (f ) to the integral. The map z 7→ −1/z maps the arc from
ρ2 to i to the arc from ρ to i. Since

1
f − = z k f (z),
z
we find upon logarithmic differentiation (actually, logarithmic differential-

ization!) that
f 0 − z1

k f 0 (z)

1
d − = + dz.
f − z1

z z f (z)
Hence the integral along the circular arc from ρ2 to ρ is
i ρ Z i Z ρ 0
f 0 (z) f 0 (z) k f 0 (z)
Z Z
1 1 f (z)
dz + dz = + dz + dz
2πi ρ2 f (z) i f (z) 2πi ρ z f (z) i f (z)
Z i
k dz
= .
2πi ρ z
But this is easily parametrized by z = eiθ :

π
i
d(eiθ )
Z Z
k dz k 2 k π π k
= = − = .
2πi ρ z 2πi π
3
eiθ 2π 2 3 12
Hence
X k
vz (f ) = −vi∞ (f ) + .
12
z∈F
This gives the result in the case that f has no zeros on the boundary of F . We
now indicate how to modify this proof to allow for zeros on the boundary.
We modify C to create C 0 as follows. If there is a zero z1 with Re z1 = −1/2
and z1 6= ρ2 then we deform the contour as indicated in Figure 4.2. The arc
around T z1 is the transform by T of the arc around z1 . If there is a zero z2
with |z2 | = 1 and − 12 < Re z2 < 0, then we deform the contour as indicated
in Figure 4.3. The arc around Sz2 is the image under S of the arc around z2 .
We make as many such deformations as are necessary. Finally, we deform
the contour near the points ρ2 , i, and ρ so that it moves along circular arcs
of small enough radius > 0. See Figure 4.4. The argument principle now
gives
f 0 (z)
Z
1 X0
dz = vz (f ).
2πi C 0 f (z) 2
z6=i,ρ
z∈F
The modified vertical contours still cancel each other since their orientations
are opposite. Moreover, the arcs along |z| = 1 including the deformations
around points z2 and Sz2 , but excluding the arcs around ρ2 , i, and ρ, will
contribute k/12 once we take the limit → 0. To determine the contributions
z1 T z1
−1 − 12 0 1 1
2
Fig. 4.2 Contour with zeros on vertical edge of F
of the arcs centered at ρ2 , i, and ρ, we take the limit → 0 and proceed as

follows.
i Sz2
z2
ρ2 ρ
−1 − 12 0 1 1
2
Fig. 4.3 Contour with zeros on lower edge of F
We consider i first. If the arc were the full circle (oriented clockwise)
around i, the contribution would be −vi (f ). Since it is only a half-circle,
the arc is easily parametrized with the angle ranging from −π to 0 so that
the final contribution is − 12 vi (f ). The same is true of the arcs centered at ρ
i
ρ2 ρ
−1 − 12 0 1 1
2
Fig. 4.4 Contour with zeros at i and ρ
and ρ2 . The arc around ρ is parametrized with the angle ranging from π/2 to
2π/3 (see Figure 4.5). The integral around the arc centered at ρ contributes
−vρ (f )/6. The same is true of the contribution from the arc centered at ρ2 .
But ρ and ρ2 are SL2 (Z)-equivalent so the final contribution is −vρ (f )/3. t
u
π
3
π
3
−1 0 1
Fig. 4.5 Tangential angle at ρ
Exercise 4.3.2. If k < 0, show that Mk (SL2 (Z)) = {0}.

Exercise 4.3.3. Show that M0 (SL2 (Z)) = C.
Exercise 4.3.4. Show that M2 (SL2 (Z)) = {0}.
Exercise 4.3.5. Show that E4 (z) has a simple zero at z = ρ.
Exercise 4.3.6. Show that E6 (z) has a simple zero at z = i.
Exercise 4.3.7. Show that Sk (SL2 (Z)) = {0} for k ≤ 10.
Exercise 4.3.8. Show that Mk (SL2 (Z)) = CEk for k = 4, 6, 8, and 10.
Exercise 4.3.9. Show that E42 = E8 . Deduce that

n−1
X
σ7 (n) = σ3 (n) + 120 σ3 (m)σ3 (n − m).
m=1
Exercise 4.3.10. Show that E4 E6 = E10 . Deduce that

n−1
X
11σ9 (n) = 21σ5 (n) − 10σ3 (n) + 5040 σ3 (n)σ5 (n − m).
m=1
4.4 The dimension formula
Ramanujan’s cusp form

E43 − E62
∆=
1728
plays a special role in the theory of modular forms. Observe that the valence
formula shows that ∆(z) 6= 0 for all z ∈ H. It only has a simple zero at i∞.
This observation allows us to prove:
Theorem 4.4.1. The following statements hold:
(i) Mk (SL2 (Z)) = {0} for all k < 0 and k = 2;
(ii) For k = 0, 4, 6, 8, and 10, Mk (SL2 (Z)) is one-dimensional and spanned by

1, E4 , E6 , E8 , and E10 , respectively;
(iii) Multiplication by ∆ defines an isomorphism of Mk−12 (SL2 (Z)) onto Sk (SL2 (Z))
for all k ∈ Z.
Proof. We have already noted that (i) and (ii) are immediate by the va-
lence formula (see Exercises 4.3.2, 4.3.3, 4.3.4, and 4.3.8). Clearly, if g ∈
Mk−12 (SL2 (Z)), then g∆ ∈ Sk (SL2 (Z)). If f ∈ Sk (SL2 (Z)), then g = f /∆
is a modular form of weight k − 12 since ∆ does not vanish on the upper
half-plane, implying g is analytic. This proves (iii). t
u
4.4 The dimension formula 47
Corollary 4.4.2 (The dimension formula). For k ≥ 0,

(
k
b 12 c if k ≡ 2 (mod 12)
dim Mk (SL2 (Z)) = k
b 12 c + 1 if k 6≡ 2 (mod 12)
Proof. The formula is true for k ≤ 10 by (i) and (ii) of Theorem 4.4.1. By
Exercise 4.2.8 and by part (iii) of Theorem 4.4.1, for k ≥ 12 we have
dim Mk (SL2 (Z)) = 1 + dim Sk (SL2 (Z))

= 1 + dim Mk−12 (SL2 (Z)).
This proves the result by induction. t

u
Exercise 4.4.3. For k ≥ 4, show that

(
k
b 12 c − 1 if k ≡ 2 (mod 12)
dim(Sk (SL2 (Z)) = k
b 12 c if k 6≡ 2 (mod 12)
Corollary 4.4.4. The modular forms E4α E6β with 4α + 6β = k, and α, β integers
≥ 0, form a basis for Mk (SL2 (Z)).
Proof. We first show that any modular form f of weight k for SL2 (Z) can be
written as a linear combination of such elements. We proceed by induction
on k. This is clear for k ≤ 10 by Exercises 4.3.8-4.3.10. For general k, choose
positive integers α, β such that 4α + 6β = k. This is clearly possible since k is
even and we can find positive integers α, β such that 2α+3β = k/2. (Indeed,
if k/2 is even, we write 2α = k/2. If k/2 is odd, we write 2(α − 1) + 3 = k/2.)
Then g = E4α E6β has weight k and is non-zero at i∞. So there exists λ ∈ C
such that f −λg vanishes at i∞. Thus we can write f = λg +∆h for some h ∈
Mk−12 (SL2 (Z)) by (iii) of Theorem 4.4.1. Applying the induction hypothesis
to h shows that the monomials E4α E6β span Mk (SL2 (Z)). To show that these
elements are linearly independent, we suppose that
X
cαβ E4α E6β = 0.
4α+6β=k
Since E4 (ρ) = 0 and E6 (ρ) 6= 0, we see that cαβ with α = 0 must vanish.
The same argument shows that cαβ with β = 0 also vanishes, using the fact
E6 (i) = 0 and E4 (i) 6= 0. Thus, we have α ≥ 1, β ≥ 1 in the summation.
Hence X
cαβ E4α−1 E6β−1 E4 E6 = 0.
4(α−1)+6(β−1)=k−10
Dividing through by E4 E6 and repeating the process completes the proof by

induction. t
u
E43 − E62
Exercise 4.4.5. Prove that ∆ = has integer Fourier coefficients.
1728
4.5 The j-function
The j-function is defined by j = E43 /∆. Clearly

az + b ab
j = j(z) ∀ ∈ SL2 (Z).
cz + d cd
It is not holomorphic at i∞ since ∆ vanishes there to first order. Conse-

quently, it has a simple pole there. Since ∆ has integer coefficients (Exer-
cise 4.4.5) with lead coefficient 1,
∆ = q − 24q 2 + . . . ,
we see that the Fourier coefficients of the j-function’s expansion at i∞ are

also integers. We have
1
j(z) = + 744 + 196884q + 21493760q 2 + . . . (4.5)
q
If we write c(n) for the n-th Fourier coefficient of the j-function, then Peters-
son [28] showed in 1932 that
√
e4π n
c(n) ∼ √ ,
2n3/4
as n → ∞. Thus the integer coefficients c(n) grow at an exponential rate.
These coefficients are related to the representation theory of a sporadic sim-
ple group known as the Monster group that played a decisive role in the
classification of all finite simple groups. The j-function has many more
remarkable properties that we cannot adumbrate here. It is ubiquitous in
mathematics and one is compelled to study its beauty and character irre-
spective of any subdiscipline. We will content ourselves here with the fol-
lowing.
Theorem 4.5.1. Let F denote the fundamental domain from Theorem 3.2.2 and let
F 0 denote the set obtained by removing points with Re z = 12 and points with both
|z| = 1 and Re z > 0. The j-function z 7→ j(z) defines a bijection from F 0 onto C.
Proof. We must show that given λ ∈ C, there is a unique z ∈ F 0 such that

j(z) = λ. In other words, we have to show that E43 − λ∆ has a unique zero
in F 0 . But this is clear from the valence formula. Indeed, f = E43 − λ∆ has
weight 12 and the only decomposition of
4.5 The j-function 49
b c
a+ + =1
2 3
with a, b, c, integers ≥ 0 arises from
(a, b, c) = (1, 0, 0), (0, 2, 0), or (0, 0, 3).
This means f has a unique zero in F 0 . t

u
4.5.2. Show that j(ρ) = 0 and j(i) = 1728.

Exercise
Moreover, show that
vρ j(z) = 3, that vi j(z) − 1728 = 2, and that vz0 j(z) − j(z0 ) = 1 for all
z0 ∈ H which are not SL2 (Z)-equivalent to either ρ or i.
A modular function of level one is a meromorphic function f : H → C that
satisfies
az + b ab
f = f (z) ∀ ∈ SL2 (Z)
cz + d cd
and is meromorphic at i∞. In particular, it admits a q-expansion with only
finitely many negative powers of q.
Exercise 4.5.3. Show that any modular function f has a finite number of zeros and
poles in F .
It turns out that every modular function of level one is a rational function
in j as is proved in the following theorem.
Theorem 4.5.4. Let f be a meromorphic function on H. The following are equiva-
lent:
i) f is a modular function;
ii) f is a quotient of two modular forms of the same weight;
iii) f is a rational function of j.
Proof. The implications (iii) ⇒ (ii) ⇒ (i) are immediate since j = E43 /∆. It
therefore suffices to prove (i) ⇒ (iii). Let f be a modular function. Suppose
that it has a pole at z0 of order r. Then
(j(z) − j(z0 ))r f (z)
is again a modular function without a pole at z0 . In this way, we can multiply

f by a suitable polynomial in j and assume f is holomorphic on H. Since
∆ has a simple zero at i∞, we can multiply f by a suitable power of ∆
and assume that ∆n f is holomorphic at i∞, for some positive integer n. But
now, ∆n f is holomorphic and is a modular form of weight 12n. By Corollary
4.4.4, it can be written as a linear combination of monomials E4α E6β . We are
therefore reduced to showing E4α E6β /∆n is a polynomial in j. But 4α + 6β =
12n, so that 3|α and 2|β. Thus, we can write this as (E43 )α/3 (E62 )β/2 /∆n =
(E43 /∆)α/3 (E62 /∆)β/2 . Since E43 − E62 = 1728∆ we see that E62 /∆ = j − 1728.
As j = E43 /∆, we are done. t
u
The reader may easily check that the valence formula also holds for mod-
ular functions, the proof being essentially identical. This permits an alter-
native proof of (i) ⇒ (iii) that treats zeros and poles on the same foot-
ing. Let f be a modular function, not identically zero. Assume first that
vi (f ) = vρ (f ) = 0. Let z1 , z2 , . . . , zr and w1 , w2 , . . . , ws denote, respectively,
the zeros and poles in H of f (z), listed according to multiplicity. Set
Qr
(j(z) − j(zi ))
g(z) = Qsi=1 .
i=1 (j(z) − j(wi ))
Then f /g is a modular function with no zeros or poles in H. By the valence

formula, vi∞ (f /g) = 0. Hence, f /g is a non-vanishing, holomorphic modu-
lar form of weight 0. By Exercise 4.3.3, f = cg for some c ∈ C, as desired. If
vi (f ) 6= 0, then an analysis using the valence formula shows that 2|vi (f ). By
Exercise 4.5.2, j(z) − j(i) has a double zero at z = i. Hence, i should only
appear in the list of zeros and poles |vi (f )|/2 many times. Similarly, ρ should
appear |vρ (f )|/3 times.
Exercise 4.6.1. Show that the Bernoulli numbers Bk vanish for odd k ≥ 3.
Exercise 4.6.2. Show that for k even and positive, (−1)k/2 Bk < 0.
Bk+1 (n + 1) − Bk+1
1k + 2k + · · · + nk = ,
k+1
for every natural number k.
Exercise 4.6.4. Prove that ζ(k) tends to 1 as k tends to infinity and use this to
show that √
|B2k | ∼ 4π e(k/πe)2k+1/2
as k tends to infinity.
Exercise 4.6.5. Show that if the Fourier coefficients of f ∈ Mk (SL2 (Z)) are inte-
gers, then f can be written as a polynomial in E4 , E6 , and ∆ with integer coeffi-
cients.
P∞
Exercise 4.6.6. Suppose f = n=0 an q n ∈ Mk (SL2 (Z)). Show that for any x ∈
R we have limy→∞ f (x+iy) = a0 . Show that for any x ∈ Q, if f is not a cusp form,
then limy→0 f (x + iy) = ∞, and if f is a cusp form, then limy→0 f (x + iy) = 0.
Deduce that the real line is a natural boundary for any modular form.
Exercise 4.6.7. Prove that E42 E6 = E14 .
Exercise 4.6.8. Let rk = dim Mk (SL2 (Z)). Put ` = k − 12rk + 12 and E0 = 1.

Show that
E` E14−` = E14
for ` = 0, 4, 6, 8, 10 and 14.
Exercise 4.6.9. Let f and g be modular forms of weight k and ` respectively, for the
full modular group. Show that
kf g 0 − `f 0 g
is a modular form of weight k + ` + 2 for the full modular group.
Exercise 4.6.10. Let j(z) = E43 /∆. Prove that
dj
= −2πiE14 /∆.
dz
Exercise 4.6.11. Let rk = dim Mk (SL2 (Z)) and ` = k − 12rk + 12. For any
modular form f ∈ Mk (SL2 (Z)), show that
f E`−1 ∈ M12rk −12 (SL2 (Z)).
Exercise 4.6.12. With notation as in the previous exercise, show that
f E`−1 ∆1−rk
is a modular function and hence a polynomial in j.
Exercise 4.6.13. Let f be a modular form of weight k for the full modular group
whose Fourier expansion at i∞ is given by
∞
X
f (z) = an e2πinz , an ∈ C.
n=0
Recall that by Corollary 4.4.4, there are unique cab ’s such that
X
f= cab E4a E6b , cab ∈ C.
4a+6b=k
Let F be an algebraic extension of Q. Suppose that
#{n : an ∈ F } ≥ dim Mk (SL2 (Z)).
Then, show that cab belong to F , whence all the Fourier coefficients belong to F .
Exercise 4.6.14. If f is a modular form whose Fourier coefficients at i∞ are alge-

braic numbers, then show that the smallest exension field of the rationals containing
all the Fourier coefficients of f is a finite extension of Q.
Exercise 4.6.15. Suppose that f is a modular form of weight k whose Fourier ex-
pansion at i∞ is given by
∞
X
f (z) = an e2πinz , an ∈ Q.
n=0
Show that, for every σ ∈ Gal(Q/Q), the series

∞
X
f σ (z) := σ(an )e2πinz ,
n=0
defines a modular form of weight k.

Chapter 5
The Ramanujan τ -function
5.1 The Eisenstein series E2
In his fundamental paper of 1916, Ramanujan introduced the τ -function as

being the coefficients in the power series expansion of the infinite product
∞
Y ∞
X
q (1 − q n )24 = τ (n)q n . (5.1)
n=1 n=1
He made three conjectures regarding τ (n):

(i) τ (mn) = τ (m)τ (n) for (m, n) = 1
(ii) if p is prime, then τ (pα+1 ) = τ (p)τ (pα ) − p11 τ (pα−1 ) for α ≥ 1;
(iii) if p is prime, then |τ (p)| ≤ 2p11/2 .
Our goal in this chapter is to prove the first two of these conjectures in
the setting of the theory of modular forms and develop what is now called
Hecke theory of level one.
Ramanujan discovered an intimate connection between the function τ (n)
and r24 (n) which counts the number of ways of expressing n as a sum of 24
squares. More precisely, let σv∗ (n) be σv (n) if n is odd and σve (n) − σvo (n) if n
is even with σve (n), σvo (n) being the sumof the v-th powers of the even and
odd divisors of n respectively. Ramanujan proved that
16 ∗
r24 (n) = σ (n) + e24 (n),
691 11
where
128
e24 (n) = (−1)n−1 259τ (n) − 512τ (n/2) ,
691
with the last term appearing only if n is even. We will not prove this formula
in this book but refer the reader to Chapters 9 and 10 of [13] and [11] for
further details.
54 5 The Ramanujan τ -function
Our first goal is to prove that the left hand side of (5.1) is a cusp form of
weight 12 for the full modular group for q = e2πiz and z ∈ H. We follow a
method due to Hurwitz and study the Eisenstein series E2 (defined below)
to realize this goal.
The double series
∞ ∞
X X 0 1
G2 (z) := , z∈H (5.2)
m=−∞ n=−∞
(mz + n)2
can be shown to converge by using a technique due to Hurwitz. The primed

summation again means that we omit (m, n) = (0, 0). We introduce
1 1
am,n (z) = −
mz + n − 1 mz + n
for (m, n) 6= (0, 0), (0, 1). Let us observe that
1 −1
− am,n (z) = ,
(mz + n)2 (mz + n)2 (mz + n − 1)
and that the double series

XX 1
(mz + n)2 (mz + n − 1)
(m,n)6=(0,0),(0,1)
converges absolutely. Thus,

∞
X X 1
G2 (z) = 2ζ(2) +
(mz + n)2
m6=0 n=−∞
∞ X X∞
X X 1
= 2ζ(2) + − a m,n (z) + am,n (z),
n=−∞
(mz + n)2 n=−∞
m6=0 m6=0
where now the first double sum in the last equality is absolutely convergent
and the inner sum in the second double sum telescopes to zero. Therefore
∞
X X 1
G2 (z) = 2ζ(2) + − am,n (z) .
(mz + n)2
m6=0 n=−∞
Because the double sum is now absolutely convergent, we may interchange

the order of summation to get
∞ X
X 1
G2 (z) = 2ζ(2) + − a m,n (z) . (5.3)
n=−∞ m6=0
(mz + n)2
From (5.2), we have

5.1 The Eisenstein series E2 55
∞
X X 1
G2 (z) = 2ζ(2) +
n=−∞
(mz + n)2
m6=0
so that
∞
z2

1 X X
G2 − = 2ζ(2) +
z n=−∞
(nz − m)2
m6=0
XX z2
= 2ζ(2)(1 + z 2 ) +
(nz − m)2
m6=0 n6=0
∞
X X z2
= 2ζ(2)z 2 + .
m=−∞ n6=0
(nz − m)2
Re-labelling our index of summation, we obtain

∞
z2

1 2
X X
G2 − = 2ζ(2)z + .
z n=−∞ m6=0
(mz + n)2
From (5.3), we get

∞
1 X X
G2 (z) = 2ζ(2) + G2 − − 2ζ(2)z 2 z −2 − am,n (z).
z n=−∞ m6=0
We now proceed to evaluate the last sum:

∞ ∞ X
X X X 1 1
am,n (z) = − .
n=−∞ m6=0 n=−∞ m6=0
mz + n − 1 mz + n
Recall the cotangent expansion (4.3):

∞
1 X 1 1
π cot (πz) − = + .
z m=1 z + m z − m
Using this, we see that the sum in question is
∞ πn z
1 X π(n − 1) z
π cot − − π cot +
z z n−1 z n
n6=−∞
n6=0,1
X 1 1
X 1 1

+ − + − .
mz − 1 mz mz mz + 1
m6=0 m6=0
The last two sums combine to give

X 1 1

1 X

1 1

− =− +
mz − 1 mz + 1 z (1/z) + m (1/z) − m
m6=0 m6=0
2 π
=− π cot −z
z z
−2π π
= cot + 2.
z z
We now examine
∞ πn z
1 X π(n − 1) z
π cot − − π cot +
z z n−1 z n
n6=−∞
n6=0,1
N
1 X π(n − 1) πn z z
= lim π cot − π cot + − .
z N →∞ z z n n−1
n6=−N
n6=0,1
Keeping in mind that cotangent is an odd function, we easily see that this
telescopes to

1 π π(N + 1) πN z z
lim 2π cot − π cot − π cot + + − 2z
z N →∞ z z z N N +1

2π π 2π πN
= cot −2− lim cot .
z z z N →∞ z
Now, as z ∈ H, by (4.4) we have
e2πi(N/z) + 1

πN 1
lim cot = i lim = i + 2i lim 2πi(N/z) = i.
N →∞ z N →∞ e2πi(N/z) − 1 N →∞ e −1
Putting everything together, we obtain that

∞
X X 2πi
am,n (z) = − .
n=−∞ m6=0
z
This proves:
Theorem 5.1.1. For all z ∈ H,

1
G2 − = z 2 G2 (z) − 2πiz.
z
Thus G2 is not a modular form, but rather an example of a “quasi”-

modular form. We can normalize G2 and consider
∞
1 X X 1
E2 (z) = 1 + .
2ζ(2) n=−∞
(mz + n)2
m6=0
Exercise 5.1.2. Show that E2 (z) has the Fourier expansion at i∞ given by
∞
X
E2 (z) = 1 − 24 σ1 (n)q n ,
n=1
where q = e2πiz and σ1 (n) =

P
d.
d|n
Exercise 5.1.3. Show that for all z ∈ H,

1 6
E2 − = z 2 E2 (z) + z.
z πi
E43 − E62
Theorem 5.1.4 (Jacobi). The cusp form ∆ = ∈ S12 (SL2 (Z)) is Ra-
1728
manujan’s infinite product
∞
Y
∆=q (1 − q n )24 .
n=1
Proof. Take q = e2πiz and let

∞
Y
F (z) = q (1 − q n )24 .
n=1
Clearly F (z + 1) = F (z) for all z ∈ H. If we can show that

1
F − = z 12 F (z),
z
then by Exercises 4.1.4 and 4.4.3, we conclude that F = c∆ for some c ∈ C.

Comparing Fourier coefficients, we easily see that c = 1. Now, logarithmic
differentiation with respect to z gives
∞
F 0 (z) X nq n
= 2πi − 48πi ,
F (z) n=1
1 − qn
Noting that
∞
qn X
= q mn ,
1 − qn m=1
we see from Exercise 5.1.2 that

F 0 (z)
= 2πiE2 (z).
F (z)
Replacing z by −1/z and using Exercise 5.1.3, we obtain
F 0 (−1/z) F 0 (z)

2 6z
= 2πi z E2 (z) + = z2 + 12z.
F (−1/z) πi F (z)
Writing this equation as
F 0 (−1/z) 1 F 0 (z) 12
· 2 = + ,
F (−1/z) z F (z) z
we immediately see that F (−1/z) and z 12 F (z) have the same logarithmic
derivative. Thus F (−1/z) = λz 12 F (z) for some constant λ. Putting z = i,
we have λ = 1, as F (z) 6= 0 for z ∈ H. This completes the proof. t
u
Exercise 5.1.5. For z ∈ H, we define the Dedekind η-function by

∞
Y
2πiz/24
η(z) = e (1 − e2πinz ).
n=1
Show that r
1 z
η − = η(z),
z i
where the square root is the branch with non-negative real part.

∆0 (z)
= 2πiE2 (z).
∆(z)

Exercise 5.1.7. Prove that for all ac db ∈ SL2 (Z) and all z ∈ H, we have

az + b 6c(cz + d)
E2 = (cz + d)2 E2 (z) + .
cz + d πi
Exercise 5.1.8. The Serre derivative of a modular form f ∈ Mk (SL2 (Z)) is de-
fined as
1 0 k
θk f := f − E2 f.
2πi 12
Prove that θk : Mk (SL2 (Z)) → Mk+2 (SL2 (Z)).
Theorem 5.1.9 (Ramanujan). Let

∞
X
P = 1 − 24 σ1 (n)q n ,
n=1
X∞
Q = 1 + 240 σ3 (n)q n ,
n=1
X∞
R = 1 − 504 σ5 (n)q n .
n=1
d
For the differential operator D = q dq , we have
1
DP = (P 2 − Q),
12
1
DQ = (P Q − R),
3
1
DR = (P R − Q2 ).
2
Remark. The reader will note that if we replace q by e2πiz , then P , Q, and
R become E2 , E4 , and E6 respectively. The differential equations stated in
the theorem were first proved by Ramanujan in his 1916 paper using the
theory of elliptic functions. The differential equations have now become
central in the theory of modular forms and have even found application
in Nesterenko’s work in transcendental number theory. Our proof here will
involve only the basic property of the Serre derivative stated in Exercise
5.1.8.
Proof. Noting that if q = e2πiz , then
d 1 d
D=q = ,
dq 2πi dz
we see that the Serre derivative can be written as

k
θk f = Df − E2 f.
12
By Exercise 5.1.8, we have θ4 E4 ∈ M6 (SL2 (Z)) and θ6 E6 ∈ M8 (SL2 (Z)).
Since these vector spaces are 1-dimensional and spanned by E6 and E8 re-
spectively, we see that θ4 E4 = λ1 E6 and θ6 E6 = λ2 E8 for appropriate con-
stants λ1 , λ2 ∈ C. A quick check of the q-expansion of both sides shows
λ1 = −1/3 and λ2 = −1/2. Since E8 = E42 , this proves that DQ = 13 (P Q−R),
and DR = 12 (P R − Q2 ). To show that DP = 12 1
(P 2 − Q), we need only note
that
1 0 1
E (z) − E22 (z) ∈ M4 (SL2 (Z)),
2πi 2 12
by a straightforward computation using Exercise 5.1.7. Since M4 (SL2 (Z)) is
also one-dimensional and spanned by E4 , we quickly find on comparing
coefficients that
1 2 1
DP − P = − Q. t
u
12 12
5.2 Hecke operators of level one
Let Xm be the set of 2 × 2 matrices with integer entries and determinant m,

with m a natural number. The group SL2 (Z) acts on Xm via left multipli-
cation. We first show that the number of orbits is finite and then find orbit
representatives. Using this result, we will construct special linear operators
Tm : Mk (SL2 (Z)) → Mk (SL2 (Z)) called Hecke operators and proceed to
study their eigenfunctions and eigenvalues. This study will lead to a proof
of the first two conjectures of Ramanujan concerning the τ -function.
Theorem 5.2.1. The number of orbits for the action of SL2 (Z) on Xm is σ1 (m),
and a set of orbit representatives can be taken as

ab
: ad = m, 0 ≤ b ≤ d − 1 .
0d
That is,
a a ab
Xm = SL2 (Z) .
0d
ad=m b (mod d)
d>0
Proof. First we show that every matrix in Xm appears in some orbit as given
in the theorem. Indeed, let

αβ
∈ Xm .
γ δ
We want to find matrices

rs ab
∈ SL2 (Z) and ∈ Xm
tu 0d
such that
αβ rs ab ra rb + sd
= = .
γ δ tu 0d ta tb + ud
This suggests we take a = (α, γ), and r = α/a, t = γ/a. As (r, t) = 1, we can
find integers s, u such that
ru − ts = 1.
Then −1
ab rs αβ u −s αβ
= =
0d tu γ δ −t r γ δ
5.2 Hecke operators of level one 61
so that
b = βu − δs
d = −βt + δr.
Note that ad = m simply by taking determinants. To ensure 0 ≤ b ≤ d − 1,

we note that
1 −k ab a b − kd
=
0 1 0d 0 d

ab
may also be taken as a representative of the orbit SL2 (Z) so we can
0d
ensure 0 ≤ b ≤ d − 1.
Finally, we need to show these representatives are SL2 (Z)-inequivalent.
Indeed, if g = ( rt us ) ∈ SL2 (Z) is such that
0 0
ab a b
g = ,
0d 0 d0
then we get
a 0 b0

ar br + sd
= .
at bt + ud 0 d0
Thus, t = 0 and ru = det g = 1 so that r = ±1. As a, a0 > 0, we must have
r = 1 so that a = a0 . But then d = d0 since ad = a0 d0 = m. Finally, b + sd = b0
leads to s = 0 since 0 ≤ b ≤ d − 1 and 0 ≤ b0 ≤ d − 1. This completes the
proof. t
u
This theorem allows us to define the m-th Hecke operator Tm as follows.

For f ∈ Mk (SL2 (Z)), let

k/2−1
X X ab
Tm (f ) := m f .
0d
ad=m b (mod d)
d>0
Theorem 5.2.2. Let m be a positive integer. If f ∈ Mk (SL2 (Z)), then Tm (f ) ∈

Mk (SL2 (Z)). If f ∈ Sk (SL2 (Z)), then Tm (f ) ∈ Sk (SL2 (Z)). In other words, Tm
is a linear transformation of the vector spaces Mk (SL2 (Z)) and Sk (SL2 (Z)).
Proof. First we need to show that Tm (f )|γ = Tm (f ) for all γ ∈ SL2 (Z).
Fix γ ∈ SL2 (Z) and an orbit representative a0 db . Since SL2 (Z) Xm =
Xm SL2 (Z) (simply by taking determinants), we see that
0 0
ab 0 a b
γ=γ
0d 0 d0
for some γ 0 ∈ SL2 (Z) and a0 d0 = m, 0 ≤ b0 ≤ d0 − 1. But f |γ 0 = f and so

a0 b 0
a 0 b0

ab 0
f γ =f γ =f .
0d 0 d0 0 d0
0 0
Moreover, the new representatives a0 db 0 obtained this way are again dis-
tinct as we range over ad = m, 0 ≤ b ≤ d − 1. Indeed, suppose also that
0 0
a 1 b1 a b
γ = γ 00
0 d1 0 d0
for another a1 d1 = m and 0 ≤ b1 ≤ d1 − 1. Now

a 1 b1 00 0 −1 ab
γ = γ (γ ) γ.
0 d1 0d
After cancelling the γ from both sides, we conclude (as in the proof of The-
orem 5.2.1) that this implies that γ 00 (γ 0 )−1 is the identity and that a0 db =

a 1 b1

0 d1 .
The fact that Tm (f ) is holomorphic at i∞ when f ∈ Mk (SL2 (Z)) and that
Tm (f ) vanishes at i∞ when f ∈ Sk (SL2 (Z)) will follow from the explicit
q-expansion which we develop in Theorem 5.2.4 below. t
u
d−1
(
X
2πinb/d d if d | n
e =
b=0
0 otherwise
Theorem 5.2.4. If f ∈ Mk (SL2 (Z)) has q-expansion

∞
X
f= λ(n)q n
n=0
at i∞, then the q-expansion at i∞ of Tm (f ) is

 
X∞ X mn
Tm (f ) =  dk−1 λ 2
 qn .
n=0
d
d|n,m
Proof. Writing
∞
X
f (z) = λ(n)e2πinz ,
n=0
and recalling the definition (4.2) of the slash operator for matrices in GL+
2 (R),
we see that
d−1 ∞
1 X m k X X
Tm (f )(z) = λ(n)e2πin(az+b)/d .
m d n=0
ad=m b=0
d>0
Interchanging the two inner sums and using Exercise 5.2.3, we obtain
X m k−1 X∞
Tm (f )(z) = λ(n)e2πinaz/d .
d n=0
ad=m
d>0 d|n
Writing n = dr, we can write the inner sum as

∞
X
λ(dr)e2πiarz .
r=0
Collecting terms with ar = s, we get

 
X∞ X m k−1
Tm (f )(z) = λ(dr) e2πisz .
 
d

s=0 ad=m
ar=s
The coefficient of e2πisz can be re-written as

X ms
ak−1 λ ,
a2
a|m,s
as claimed. t
u
Exercise 5.2.5. Prove that Tm (∆) = τ (m)∆. Deduce that τ (mn) = τ (m)τ (n)
for (m, n) = 1.
Exercise 5.2.6. Prove that for any m, n ≥ 1,

X mn
τ (m)τ (n) = d11 τ .
d2
d|m,n
Exercise 5.2.7. Let p be prime and a ≥ 1. Show that
τ (pa+1 ) = τ (p)τ (pa ) − p11 τ (pa−1 ).
Thus, the theory of Hecke operators resolves the first two of Ramanujan’s
conjectures.
Exercise 5.2.8. Let f be E4 ∆, E6 ∆, E8 ∆, E10 ∆, or E14 ∆. Show that the Fourier

coefficients of f are multiplicative.
Exercise 5.2.9. Let p be a prime number. Show that

∞
X 1
τ (pn )xn = .
n=0
1 − τ (p)x + p11 x2
Exercise 5.2.10. Prove that for n > 1,

n−1
X
(1 − n)τ (n) = 24 σ1 (j)τ (n − j).
j=1
Exercise 5.2.11. If n is even, prove that 8|τ (n).
Exercise 5.2.12. If p is prime, show that τ (p) is even.
5.3 Modular forms and Dirichlet series
Ramanujan’s third conjecture that |τ (p)| ≤ 2p11/2 was resolved in 1974 by P.

Deligne as a consequence of his solution to the celebrated Weil conjectures.
In this section, we will establish a much weaker bound (due to Hecke) and
initiate a study of Dirichlet series attached to modular forms. In his 1916
paper, Ramanujan suggested the study of the associated Dirichlet series
∞
X τ (n)
L∆ (s) :=
n=1
ns
as a means to understand the nature of τ (n). Indeed, Ramanujan noted that

his conjectures (i) and (ii) imply that we can write
∞ −1
X τ (n) Y τ (p) 1
= 1 − s + 2s−11 . (5.4)
n=1
ns p
p p
We will see below that L∆ (s) converges absolutely for Re(s) > 7 and ex-
tends to an entire function satisfying the functional equation
(2π)−s Γ(s)L∆ (s) = (2π)−(12−s) Γ(12 − s)L∆ (12 − s), (5.5)
where Z ∞
Γ(s) = e−t ts−1 dt
0
is Euler’s Γ-function.
In his lectures on the work of Ramanujan, Hardy [13, p. 185] writes “The
functional equation (5.5) must have been familiar to Ramanujan, but I can-
not find an explicit statement of it either in his collected papers or in the
notebooks.” Apparently, equation (5.5) was proved by Wilton in 1929, but it
was Erich Hecke who recognized (5.5) and (5.4) as a special case of a general
theory of modular forms, Dirichlet series, and Euler products. This theory
we now call Hecke theory. In this section, we discuss some aspects of this
theory pertaining to level one.
5.3 Modular forms and Dirichlet series 65
Exercise 5.3.1. Let p be prime. Prove that Ramanujan’s conjecture
|τ (p)| ≤ 2p11/2
is equivalent to the assertion that the roots of the quadratic polynomial 1 − τ (p)x +
p11 x2 are non-real.
Exercise 5.3.2. Assuming Ramanujan’s conjecture that |τ (p)| ≤ 2p11/2 , prove

that |τ (n)| ≤ n11/2 d(n) for every natural number n. Here d(n) = σ0 (n) is the
number of divisors of n.
Exercise 5.3.3. Let f ∈ Mk (SL2 (Z)). Show that y k/2 |f (z)| is invariant under
SL2 (Z), where y = Im(z).
Theorem 5.3.4 (Hecke). If f ∈ Sk (SL2 (Z)) has Fourier expansion

∞
X
f (z) = af (n)q n
n=1
k/2

at i∞, then af (n) = O n .
Proof. By the previous exercise, the continuous function φ(z) = y k/2 |f (z)| is
invariant under the modular group. Hence we consider z ∈ F , the standard
fundamental domain. As Im z → ∞, q → 0 and so since f is a cusp form,
|f (z)| = O(q) = O e−2πy .

(5.6)
Thus φ(z) = O y k/2 e−2πy and so φ(z) → 0 as y → ∞. This implies that φ is

bounded in F and hence in H. In other words, there is a constant c such that
y k/2 |f (z)| ≤ c, ∀z ∈ H. (5.7)
Now fix y and vary x between 0 and 1. Then q = e2πi(x+iy) traverses counter-
clockwise the circle Cy of radius e−2πy , centered at zero. By Cauchy’s residue
theorem,
Z Z 1
1
af (n) = f (z)q −n−1 dq = f (x + iy)q −n dx.
2πi Cy 0
Using (5.7) we obtain

|af (n)| ≤ cy −k/2 e−2πny

for any y > 0. Choosing y = 1/n gives af (n) = O nk/2 . t
u

Corollary 5.3.5. We have τ (n) = O n6 .
P∞
Corollary 5.3.6. Suppose f = n=1 af (n)q n ∈ Sk (SL2 (Z)). The Dirichlet se-
ries
∞
X af (n)
Lf (s) =
n=1
ns
converges absolutely for Re(s) > k/2 + 1.
Proof. This is immediate from Hecke’s estimate for af (n) given by Theo-
rem 5.3.4. t
u
We now proceed to derive the analytic continuation and functional equa-

tion for Lf (s), following Hecke.
Theorem 5.3.7. Let f ∈ Sk (SL2 (Z)). The function Lf (s) extends to an entire
function and satisfies the functional equation
(2π)−s Γ(s)Lf (s) = ik (2π)−(k−s) Γ(k − s)Lf (k − s).
Proof. The estimate (5.6) shows that the integral (for fixed y0 > 0)
Z ∞
f (iy)y s−1 dy
y0
converges absolutely for all complex values of s. Since f is of exponential

decay on the imaginary axis, we may differentiate under the integral sign
and deduce that the integral defines an entire function of s. Moreover, by
modularity f (iy) = i−k y −k f (i/y) = O y −k e−2π/y as y → 0. Hence the

limit as y0 → 0 of the integral exists and defines an entire function. Thus we

have the entire function:
Z ∞ ∞
X Z ∞
f (iy)y s−1 dy = af (n) e−2πny y s−1 dy = (2π)−s Γ(s)Lf (s),
0 n=1 0
after a change of variable (t = 2πny) in the integral. We split the integral on

the left hand side as
Z 1 Z ∞
f (iy)y s−1 dy + f (iy)y s−1 dy.
0 1
We change variables (y 7→ 1/y) in the first integral and the integral becomes
Z ∞
i
f y −s−1 dy.
1 y
Since i/y = −1/iy, and f is a modular form of weight k,

1
f − = (iy)k f (iy),
iy
so that
5.4 Ramanujan congruences 67
Z ∞ Z ∞
i −s−1 k
f y dy = i f (iy)y k−s−1 dy.
1 y 1
Thus Z ∞
−s
dy
(2π) Γ(s)Lf (s) = f (iy) y s + ik y k−s ,
1 y
from which the functional equation is immediate. t
u
Exercise 5.3.8. Suppose that f ∈ Mk (SL2 (Z)) with Fourier expansion at i∞

given by
∞
X
f (z) = af (n)q n ,
n=0
P∞ a (n)
and that af (0) 6= 0. Show that Lf (s) = n=1 fns has an analytic continuation
to the entire complex plane, except for a simple pole at s = 0 and s = k and satisfies
the functional equation
5.4 Ramanujan congruences
There are many interesting arithmetical properties of the τ -function, too nu-
merous to expound here. We will content ourselves with a sampling of such
results and invite the reader to consult the article [45] by H.P.F. Swinnerton-
Dyer. Here is an example of an elementary congruence for the τ -function.
Exercise 5.4.1. Prove that τ (n) ≡ nσ1 (n) (mod 5).
Some of these congruences are intertwined with yet another interesting

arithmetical function, the partition function. In fact, in many of the proofs of
Ramanujan’s congruences (to be stated below), Theorem 5.1.9 plays a fun-
damental role. We illustrate this through the following discussion.
Let p(n) be defined by the series
∞
X ∞
Y
p(n)q n = (1 − q n )−1 .
n=0 n=1
The function p(n) is the partition function and counts the number of ways
of writing n = λ1 + λ2 + · · · + λr with λ1 ≥ λ2 ≥ · · · ≥ λr ≥ 0, λi ∈ Z.
The partition function appears in many branches of mathematics ranging
from number theory, combinatorics, group theory, representation theory, to
mathematical physics. Ramanujan discovered that
p(5n + 4) ≡ 0 (mod 5)
p(7n + 5) ≡ 0 (mod 7)
p(11n + 6) ≡ 0 (mod 11),
and these were proved by him using the theory of elliptic functions. These
are often called Ramanujan congruences. We show below how the theory of
modular forms leads to “quick” proofs of these congruences. We follow the
treatment of Berndt [4].
Theorem 5.4.2. For all n ≥ 0, we have p(5n + 4) ≡ 0 (mod 5).
Proof. We have already seen in Exercise 5.4.1 that Q ≡ 1 (mod 5) and R ≡ P

(mod 5). Thus
1728∆ = Q3 − R2 ≡ Q − P 2 (mod 5),
and from Theorem 5.1.9,
3DP ≡ Q − P 2 (mod 5).
Now by Theorem 5.1.4,

∞ ∞
Y
n 24
Y (1 − q n )25
1728∆ = 1728q (1 − q ) ≡ 3q (mod 5)
n=1 n=1
1 − qn
and (1 − q n )25 ≡ 1 − q 25n (mod 5). Hence

∞
Y ∞
Y
3q (1 − q n )−1 (1 − q 25n ) ≡ 3DP (mod 5).
n=1 n=1
Therefore,
∞ ∞
!
X Y
q p(n)q n
≡ DP (1 − q 25n )−1 (mod 5).
n=0 n=1
We comparePthe coefficient of q 5m of both sides of the congruence and note

∞
that DP ≡ n=1 nσ(n)q n (mod 5) so that the 5m-th coefficient of the right
hand side is also divisible by 5. This yields p(5m − 1) ≡ 0 (mod 5) as re-
quired. t
u
Exercise 5.4.3. Prove that Q2 ≡ P (mod 7) and R ≡ 1 (mod 7).

dP 2 dQ
(Q3 − R2 )2 ≡ (QP − 1)2 ≡ −q − 3q (mod 7).
dq dq

∞ ∞
!
X Y
3 2 2 2 n
(Q − R ) ≡ q p(n)q (1 − q 49n ) (mod 7).
n=0 n=1
Deduce that p(7n + 5) ≡ 0 (mod 7).
We close this section by mentioning Lehmer’s conjecture that τ (n) 6= 0.

Recently, work of Brunier, Ono, and Rhoades [6] relates the question to ir-
rationality of Fourier coefficients of harmonic weak Maass forms.
762048
Exercise 5.5.1. Prove that E62 = E12 − 691 ∆.
65520
P∞
Exercise 5.5.2. Show that E12 = 1 + 691 n=1 σ11 (n)q n .
Exercise 5.5.3. Prove that τ (n) ≡ σ11 (n) (mod 691).
Exercise 5.5.4. Prove that E14 ≡ E2 (mod 13).
Exercise 5.5.5. Prove that τ (n) ≡ nσ3 (n) (mod 7).

n−1
X 65 σ5 (n) 3τ (n)
σ5 (j)σ5 (n − j) = σ11 (n) + − .
j=1
(252)(691) 252 691
Deduce that
n−1
X 65
σ11 (n) + O n6 .

σ5 (j)σ5 (n − j) =
j=1
(252)(691)
Exercise 5.5.7. If k1 > k2 ≥ 4 are even, prove that for any > 0,
n−1
X
σk1 −1 (j)σk2 −1 (n − j) = c(k1 , k2 )σk1 +k2 −1 (n) + O nk1 −1+ ,

j=1
where c(k1 , k2 ) ∈ Q is a constant depending on k1 and k2 .
Exercise 5.5.8. Let fk be a sequence of complex numbers satisfying the recurrence

with f0 = 1, and
fk+1 = f fk − gfk−1 , k≥1
for some constants f and g. Write
1 − f x + gx2 = (1 − αx)(1 − βx).
Show that
αk+1 − β k+1
fk = .
α−β
If α = β, the right hand side is viewed as αk + αk−1 β + · · · + αβ k−1 + β k =
(k + 1)αk .
Exercise 5.5.9. Let g be a completely multiplicative function. An arithmetical func-

tion f is called g-multiplicative if the equation
X mn
f (m)f (n) = g(d)f (5.8)
d2
d|m,n
holds for all natural numbers m and n. Show that f is g-multiplicative if and only
if f is multiplicative and for all primes p and all natural numbers k ≥ 1, we have
f pk+1 = f (p)f (pk ) − g(p)f (pk−1 ).

(5.9)
Exercise 5.5.10. Let gk (n) = nk . Show that the Ramanujan τ -function is g11 -
multiplicative.
Exercise 5.5.11. Let gs (n) = ns . Show that σs (n) := d|n ds is gs -multiplicative.

P
Exercise 5.5.12. Let F and G be two complex-valued functions defined on N × N.

Suppose that X
F (m, n) = G(m/d, n/d).
d|n
Show that X
G(m, n) = µ(d)F (m/d, n/d)
d|m,n
and conversely. (Here µ denotes the Möbius function.)
Exercise 5.5.13. Show that the Ramanujan τ -function satisfies

X
τ (mn) = d11 µ(d)τ (m/d)τ (n/d)
d|m,n
for any two natural numbers m, n.
Exercise 5.5.14. Let A be the n×n matrix whose (i, j) entry is τ (ij). Show that the
determinant of A is given by n!11 µ(1)µ(2) · · · µ(n). In particular, the determinant
is zero if n ≥ 4.
Exercise 5.5.15. Show that τ (n) is odd if and only if n is an odd square.
Exercise 5.5.16. Let r = dim Mk (SL2 (Z)). Put
Fk := E12r−k+2 ∆−r .
Show that Fk has q-expansion
Ckr q −r + · · · + Ck1 q −1 + Ck0 + · · ·
with Ckr = 1 and Ckj ∈ Z for all j.
Exercise 5.5.17. Let j denote the j-function. Show that for any integer m ≥ 0,
dj
jm
dz
has a q-expansion without a constant term.
Exercise 5.5.18. Let f be a modular form of weight k for the full modular group
with q-expansion
f (z) = a0 + a1 q + a2 q 2 + · · ·
With Ckj defined as in Exercise 5.5.16, show that
Ck0 a0 + Ck1 a1 + · · · + Ckr ar = 0.
Exercise 5.5.19. Suppose that k ≡ 2 (mod 4). With notation as in the previous
exercise, show that Ck0 6= 0.
Exercise 5.5.20. Suppose that k ≡ 0 (mod 4). With notation as in the previous
exercise, show that Ck0 6= 0.
Chapter 6
Modular Forms of Higher Level
6.1 Modular forms for congruence subgroups
Let Γ be a congruence subgroup of SL2 (Z) of level N . Recall that H? is the

extended upper half-plane:
H? = H ∪ Q ∪ {i∞}.
We are interested in holomorphic functions f (z) : H → C for which

az + b ab
f = (cz + d)k f (z), ∀ ∈ Γ. (6.1)
cz + d cd
The reader may wish to stop and review Section 4.1 before proceeding.
The punctured fundamental neighborhood
Uc = {z ∈ H : Im(z) > c}
of i∞ is mapped via z 7→ e2πiz/N onto the punctured open disc centered at

zero, of radius e−2πc/N . We may write q 1/N = e2πiz/N for a typical point in
this punctured disc.
Let h be the least positive integer such that ( 10 h1 ) ∈ Γ. Since Γ is of level N ,
we have h ≤ N . (In fact, a straightforward gcd argument shows that h|N .)
For any function satisfying (6.1) we have f (z + h) = f (z). Hence we have a
well-defined map, which we also call f , from the unit disc to C:
q 1/h 7→ f (z)
where z ∈ H is any point such that q 1/h = e2πiz/h . If f (z) : H → C is

holomorphic, then f (q 1/h ) will be holomorphic on the punctured unit disc
and hence will have a Laurent expansion centered at q 1/h = 0,
74 6 Modular Forms of Higher Level
∞
X
f (q 1/h ) = an q n/h .
n=−∞
We call
∞
X
f (z) = an e2πizn/h
n=−∞
a Fourier series at i∞.

More generally, for any γ = ( rt us ) ∈ SL2 (Z), the function

−k rz + s
(f |γ) (z) = (tz + u) f
tz + u
will again be holomorphic on H and will now be modular under γ −1 Γγ.

That is
ab ab
(f |γ) | = f |γ, ∀ ∈ γ −1 Γγ.
cd cd
Since Γ(N ) is normal in SL2 (Z) by Exercise 2.2.2, the group γ −1 Γγ will also
be a level N congruence subgroup. Thus we let h be the least positive integer
such that ( 10 h1 ) ∈ γ −1 Γγ so that f |γ will be h-periodic:
(f |γ) (z + h) = (f |γ) (z).
As above, we consider the holomorphic function on the unit disc given by
q 1/h 7→ (f |γ) (z)
where z ∈ H is any point such that q 1/h = e2πiz/h . Now (f |γ) will have a
Laurent expansion
X∞
(f |γ)(q 1/h ) = bn q n/h
n=−∞
1/h
centered at q = 0. We say that
∞
X
(f |γ)(z) = bn e2πizn/h (6.2)
n=−∞
is a Fourier expansion of f at γ(i∞).

It is tempting to define the order of f at γ(i∞) to be the index of the
least non-zero coefficient in (6.2). However, there is a technical complica-
−1 0

tion
which
can occur when 0 −1 ∈
6 Γ. It is possible that there is some
−1 h0 −1 0 1 h 0
∈ γ Γγ with h ≥ 1 but 0 1 6∈ γ −1 Γγ. In this case, h = 2h0 .
0 −1
The theory of Riemann surfaces (which we sidestep because the reader
may not yet be familiar with Riemann surfaces) suggests that the Laurent
6.2 Eisenstein series for Γ(N ) 75
0
series (6.2) be considered as an expansion in terms of q 1/h rather than
0 1/2

q 1/h = q 1/h . This leads to the possibility of half-integer orders at these
so-called irregular cusps. Therefore, we define the width of the cusp γ (i∞) to
be the least positive integer h0 such that at least one of
1 h0

∈ γ −1 Γγ
0 1
or
−1 h0

∈ γ −1 Γγ
0 −1
holds and write the expansion of f |γ as
∞
X 0
(f |γ)(z) = cn e2πizn/h . (6.3)
n=−∞
Now we define the order of f at γ(i∞) to be the index of the least non-zero
coefficient in (6.3):

1
vγ(i∞) (f ) := inf n ∈ Z : cn 6= 0 .
2
Exercise 6.1.1. Suppose f : H → C is holomorphic and satisfies (6.1) for a con-

gruence subgroup Γ of level N . Suppose γ, τ ∈ SL2 (Z) and that γ(i∞) and τ (i∞)
are Γ-equivalent. Prove that vγ(i∞) (f ) = vτ (i∞) (f ). That is, the order of f at a
Γ-equivalence class of cusps is well-defined.
If −∞ < vγ(i∞) (f ) < 0, then f is said to be meromorphic at γ(i∞) with a

pole of order |vγ(i∞) (f )|. If vγ(i∞) (f ) ≥ 0 then f is said to be holomorphic at
γ(i∞). If vγ(i∞) (f ) > 0 then we say that f vanishes at γ(i∞) and has a zero
of order vγ(i∞) (f ).
A modular form of weight k for Γ is a holomorphic function f : H → C
satisfying (6.1) which is holomorphic at every cusp. We denote by Mk (Γ)
the C-vector space of modular forms of weight k for Γ. If f ∈ Mk (Γ) is such
that f vanishes at all the cusps, we say f is a cusp form of weight k for Γ and
denote by Sk (Γ) the C-vector space of such forms.
6.2 Eisenstein series for Γ(N )
Exercise 6.2.1. Let k ≥ 3 and N a natural number. Let g ∈ (Z/N Z)2 . Define the
Eisenstein series
X 0 1
Gk,g (z) = , z∈H
(mz + n)k
(m,n)≡g (mod N )
where the summation is over all integers m, n satisfying the congruence condition
(m, n) ≡ g (mod N ) and the dash on the summation means we exclude (m, n) =
(0, 0) in the case g = (0, 0). For any γ ∈ SL2 (Z), show that Gk,g |γ = Gk,gγ .
Conclude that Gk,g ∈ Mk (Γ(N )).
Exercise 6.2.2. Show that Gk,g = (−1)k Gk,−g . Deduce that if k is odd and 2g ≡
(0, 0) (mod N ), then Gk,g = 0.
Exercise 6.2.3. Let Γ be a congruence subgroup of SL2 (Z) of level N . If γ1 , . . . , γr

are distinct right coset representatives of Γ(N ) in Γ, show that
r
X
f := Gk,gγj ∈ Mk (Γ).
j=1
Exercise 6.2.4. Let Γ1 , Γ2 be congruence subgroups of SL2 (Z) such that Γ1 ⊂ Γ2 .

If f ∈ Mk (Γ1 ) and {γ1 , . . . , γr } is a complete set of right coset representatives of
Γ1 in Γ2 , show that
r
X
(a) f |γj ∈ Mk (Γ2 ),
j=1
Yr
(b) (f |γj ) ∈ Mkr (Γ2 ).
j=1
11 10
6.2.5. Show that Γ0 (4) is generated by T = ( 0 1 ), by ( 4 1 ), and by
Exercise
−1 0
0 −1 .
Exercise 2.3.3 shows that [SL2 (Z) : Γ0 (4)] = 6 and the reader may easily
verify that

10 10 0 −1 0 −1 0 −1 0 −1
, , , , , (6.4)
01 21 1 1 1 2 1 3 1 4
is a set of right coset representatives.
Exercise 6.2.6. Show that there are exactly three inequivalent cusps for the group
Γ0 (4). Find the widths of each cusp.
6.3 Fourier expansions of higher level Eisenstein series
As in the previous section, let g = (a1 , a2 ) ∈ (Z/N Z)2 and consider Gk,g (z).
We derive the q-expansion of Gk,g (z). Let us write
6.3 Fourier expansions of higher level Eisenstein series 77
∞
X
Gk,g (z) = bk,g (0) + bk,g (n)e2πinz/N . (6.5)
n=1
Theorem 6.3.1. The Fourier coefficients bk,g (n) in (6.5) are given as follows:
(
0 if a1 6≡ 0 (mod N ),
bk,g (0) = P −k
m≡a2 (mod N ) m if a1 ≡ 0 (mod N ),
and
(−2πi)k X
bk,g (n) = dk−1 (sgn d)e2πia2 d/N
N k (k − 1)!
d|n
n/d≡a1 (mod N )
where the summation is over all divisors of n (positive and negative) and
(
+1 if d > 0,
sgn d =
−1 if d < 0.
Proof. The computation of bk,g (0) was done at the end of the first paragraph
of the solution to Exercise 6.2.1. We separate this term out and write
k
X X 1
Gk,g (z) = bk,g (0) + .
mz + n
m≡a1 (mod N ) n≡a2 (mod N )
m6=0
Writing n = N j + a2 , we can write the inner sum as
X X mz + a2 −k
−k −k
(mz + N j + a2 ) =N +j .
N
j∈Z j∈Z
As in the level one case, we use Lipschitz’s formula (Theorem 4.2.2) which
states that for k ≥ 2 and z ∈ H, we have
∞
X 1 (−2πi)k X k−1 2πinz
= n e .
(z + n)k (k − 1)! n=1
n∈Z
Since (mz + a2 )/N ∈ H if and only if m > 0, we first consider the cases with
m > 0 for which we have
−k ∞
X mz + a2 (−2πi)k X k−1 2πida2 /N 2πidmz/N
+j = d e e .
N (k − 1)!
j∈Z d=1
If m < 0, we can re-write m = −m0 with m0 > 0 and get

X m0 z − a2 −k ∞
(−2πi)k X k−1 −2πida2 /N 2πidm0 z/N
(−1)k −j = (−1)k d e e .
N (k − 1)!
j∈Z d=1
We can combine both Fourier expansions (by grouping together dm = n and

dm0 = n) to get
(−2πi)k X
bk,g (n) = dk−1 (sgn d) e2πia2 d/N
N k (k − 1)!
d|n
n/d≡a1 (mod N )
allowing for both positive and negative integers d that divide n in the sum-
mation. t
u
Exercise 6.3.2. For g = (a1 , 0) ∈ (Z/N Z)2 and n 6= 0, show that
(−2πi)k X
bk,g (n) = dk−1 (sgn d).
N k (k − 1)!
d|n
n/d≡a1 (mod N )
Exercise 6.3.3. If g = (0, a2 ) ∈ (Z/N Z)2 , show that bk,g (n) = 0 if N - n, and
(−2πi)k X k−1 da2

bk,g (N n) = d (ζ + (−1)ζ −da2 ),
N k (k − 1)!
d|n
where ζ = e2πi/N . Deduce that Gk,g (z), with g = (0, a2 ), has a q-expansion of the
form
X∞
an q n , q = e2πiz .
n=0
Exercise 6.3.4. Prove that if g = (0, a2 ) ∈ (Z/N Z)2 , then Gk,g ∈ Mk (Γ1 (N ))
for k ≥ 3.
Exercise 6.3.5. Let

E2,N (z) := E2 (z) − N E2 (N z).
Show that E2,N ∈ M2 (Γ0 (N )).
Exercise 6.3.6. If f ∈ Mk (SL2 (Z)), show that
fN (z) := f (z) − N f (N z) ∈ Mk (Γ0 (N )).
In particular, deduce that for k ≥ 3,
Ek,N (z) := Ek (z) − N Ek (N z) ∈ Mk (Γ0 (N )).

6.4 The valence and dimension formulas 79
6.4 The valence and dimension formulas
It is possible to derive the analogs of the valence and dimension formulas for
any congruence subgroup Γ of SL2 (Z). The general method used to derive
these formulas is via the Riemann-Roch theorem, which is usually a topic in
Riemann surface theory and algebraic geometry. Thus, we content ourselves
by merely stating the results and referring the interested reader to [9] for
further details. Here are the analogues of Theorem 4.3.1 and Corollary 4.4.2,
for an arbitrary congruence subgroup Γ of SL2 (Z).
We first need to define elliptic points. A point z ∈ H is called an elliptic
point for Γ if {±I} Γz is strictly larger than {±I}, where Γz = {γ ∈ Γ : γz =
z} is the stabilizer subgroup of z. In other words, z is an elliptic point for Γ
if and only if {±I}Γz 6= {±I}. We define the order of an elliptic point to be
|{±I}Γz /{±I}|.
Exercise 6.4.1. Show that the only elliptic points for SL2 (Z) lying in the standard
fundamental domain F are i, ρ = eπi/3 , and ρ2 .
Exercise 6.4.2. Find the stabilizer subgroups SL2 (Z)i and SL2 (Z)ρ in SL2 (Z) and
show that SL2 (Z)i / {±I} and SL2 (Z)ρ / {±I} have orders two and three, respec-
tively.
Exercise 6.4.3. If Γ has finite index in SL2 (Z), show that there are only finitely
many Γ-inequivalent elliptic points for Γ.
Theorem 6.4.4 (The valence formula). Let Γ be a congruence subgroup of

SL2 (Z) and 0 6= f ∈ Mk (Γ). Then
X vz (f )
k 2 23
= + + ∞ + 2g − 2 ,
?
|Γz | 2 2 3
z∈Γ\H
where the sum is over Γ-equivalence classes of z ∈ H∗ , vz (f ) denotes the order of

f at z, Γz is the stabilizer of z in {±I}Γ/{±I}, g is the genus of Γ\H? , 2 is the
number of elliptic points of order 2, 3 is the number of elliptic points of order 3,
and ∞ is the number of Γ-inequivalent cusps.
Exercise 6.4.5. Verify that Theorem 6.4.4 reduces to Theorem 4.3.1 in the case Γ =
SL2 (Z).
Theorem 6.4.6. (The dimension formula) Let Γ be a congruence subgroup of SL2 (Z)
and k an even non-negative integer. Then
(k − 1)(g − 1) + [ k4 ]2 + [ k3 ]3 + k2 ∞ if k ≥ 2

dim Mk (Γ) =
1 if k = 0
and

 dim Mk (Γ) − ∞ if k ≥ 4
dim Sk (Γ) = g if k = 2 ,
0 if k = 0

where g, 2 , 3 , ∞ have the same meaning as in Theorem 6.4.4.
For most applications, it is not essential to have results as precise as The-

orems 6.4.4 and 6.4.6. Often, crude bounds suffice. To keep this monograph
self-contained, we therefore take an alternate route using an idea of Sturm
that allows us to reduce to the level one case.
Theorem 6.4.7 (Sturm’s bound). Let Γ be a congruence subgroup and f ∈

Mk (Γ). Let r1 , . . . , rt be the Γ-inequivalent cusps of Γ. If
t
X k[SL2 (Z) : {±I} Γ]
vri (f ) > ,
i=1
12
then f = 0.
Proof (Sketch). If {±I} Γ = SL2 (Z), the result is clear from the valence for-
mula (Theorem 4.3.1). Let M = [SL2 (Z) : {±I} Γ] and 1 = γ1 , . . . , γM be a
complete set of coset representatives for {±I} Γ in SL2 (Z). Then, the func-
tion
M
Y
F (z) = (f |γj )(z)
j=1
is clearly an element of MkM (SL2 (Z)) since for γ ∈ SL2 (Z), the products γj γ,
1 ≤ j ≤ M , are again a complete set of coset representatives so that F |γ = F
for all γ ∈ SL2 (Z).
Consider one of the Γ-inequivalent
n cusps rj . Ifothis cusp has width h, then
(1) (h)
there are h coset representatives γrj , . . . , γrj ⊆ {γ1 , . . . , γM } such that
(s)
rj = γrj (i∞).Moreover, each of the corresponding Fourier expansions will
have the same order:
(f |γr(s)
j
)(z) = bs q vrj (f )/h + · · · .
Hence the product of these will have order vrj (f ). Thus,

t
X kM
v∞ (F ) = vri (f ) > .
i=1
12
By the level one case, we deduce that F = 0 and hence some f |γj = 0. But
then f = (f |γj )|γj−1 = 0|γj−1 = 0. t
u
6.5 The Jacobi four-square theorem revisited 81
∞
X ∞
X
Exercise 6.4.8. Let f = an q n and g = bn q n ∈ Mk (Γ) for some con-
n=0 n=0
gruence subgroup Γ of SL2 (Z). Suppose that an = bn for all n ≤ k[SL2 (Z) :
{±}Γ]/12. Show that f = g.
Exercise 6.4.9. For any congruence subgroup Γ of SL2 (Z), prove that
k
dim Mk (Γ) ≤ [SL2 (Z) : {±}Γ] + 1.
12
dim M4 (Γ0 (4)) = 3.
(Hint: consider E4 , E4,2 , and E4,4 from Exercise 6.3.6.)
M2 (Γ0 (2)) = CE2,2 ,
where E2,2 (z) = E2 (z) − 2E2 (2z).
Exercise 6.4.12. Show that dim M2 (Γ0 (4)) = 2 and that {E2,2 , E2,4 } is a basis
for M2 (Γ0 (4)).
6.5 The Jacobi four-square theorem revisited
In Chapter 1, Corollary 1.4.6, we derived a formula for r4 (n), the number

of ways of writing n as a sum of four squares. In this section, we will use
the theory of modular forms to derive the same formula. The method will
also apply to derive formulas for rk (n), the number of ways of writing n as
a sum of k squares, for other values of k.
Exercise 6.5.1. Let η(z) denote the Dedekind η-function. Prove that
∞
X 2 η 5 (2z)
θ(z) := qn = ,
n=−∞
η 2 (z)η 2 (4z)
where q = e2πiz .
Theorem 6.5.2. Let

∞
X 2
θ(z) = e2πin z .
n=−∞
Then
√

z
θ = 4z + 1 θ(z),
4z + 1
where the branch of the square root is the principal branch (whose image is contained
in the right half-plane).
Proof. We use Exercises 6.5.1 and 5.1.5. From the former we have
η 5 (2z)
θ(z) =
η 2 (z)η 2 (4z)
and from the latter we have

r
1 z
η − = η(z).
z i
πi
It is immediate from the definition that η(z + 1) = e 12 η(z). Now,

4z 1 πi 1
η =η − + 1 = e 12 η −
4z + 1 4z + 1 4z + 1
r r
πi 4z + 1 πi 4z + 1
= e 12 η(4z + 1) = e 6 η(4z).
i i
Also,
−1 ! s
2z 1 4z + 1 1
η =η − − −2 = i η − −2
4z + 1 2z 2z 2z
s
−πi 4z + 1 1
=e 6 i η −
2z 2z
s r
−πi 4z + 1 2z
=e 6 i η(2z).
2z i
Finally,
−1 ! s
z 1 4z + 1 1
η =η − − −4 = i η − −4
4z + 1 z z z
s
−πi 4z + 1 1
=e 3 i η −
z z
s r
−πi 4z + 1 z
=e 3 i η(z).
z i
Putting everything together gives

√

z
θ = 4z + 1 θ(z). t
u
4z + 1
Theorem 6.5.3 (Jacobi’s four square theorem). We have θ4 ∈ M2 (Γ0 (4)) and
X
r4 (n) = 8 d.
d|n
4-d

z
Proof. By Theorem 6.5.2, θ4 4z+1 = (4z + 1)2 θ4 (z). Since θ4 (z + 1) = θ4 (z)
and since Γ0 (4) is generated by ( 10 11 ) and ( 14 01 ) and −I by Exercise 6.2.5,
we deduce θ4 ∈ M2 (Γ0 (4)). By Exercise 6.4.12, E2,2 and E2,4 comprise a
basis for M2 (Γ0 (4)). Thus θ4 = aE2,2 + bE2,4 . A quick calculation comparing
coefficients of q-expansions of both sides of this equation leads to
1
θ4 = − E2,4 ,
3
which leads to the formula

r4 (n) = 8 σ(n) − 4σ(n/4) ,
where σ(n/4) = 0 if 4 - n. If 4|n, we write n = 4n1 and 4σ(n1 ) is the sum of

the divisors d of n with 4|d. This observation leads to the desired formula.
t
u
Exercise 6.6.1. Show that η 12 (2z) ∈ S6 (Γ0 (4)).
Exercise 6.6.2. Show that S6 (Γ0 (4)) = η 12 (2z)C.
Exercise 6.6.3. Let k ≥ 6 be even. Show that Sk (Γ0 (4)) = η 12 (2z)Mk−6 (Γ0 (4)).
k
Exercise 6.6.4. For even k ≥ 0, prove that dim Mk (Γ0 (4)) = 2 + 1.
Exercise 6.6.5. Let r12 (n)P be the number of ways of writing n as a sum of 12
∞
squares and let η 12 (2z) = n=1 an q n . Show that
n
r12 (n) = 8σ5 (n) − 512σ5 + 16an .
4
Exercise 6.6.6. Let f (z) ∈ Mk (Γ1 (N )) and suppose r ∈ Z is positive. Show that
f (rz) ∈ Mk (Γ1 (rN )). Moreover, if f (z) is a cusp form, then so is f (rz).
Chapter 7
The Petersson Inner Product
7.1 The Hilbert space of cusp forms
Let Γ be a congruence subgroup of SL2 (Z) and F any fundamental domain

for Γ. Let Γ = Γ/ {±I}. For f, g ∈ Sk (Γ), we define the Petersson inner
product as ZZ
1 dxdy
(f, g) = y k f (z)g(z) 2 ,
SL2 (Z) : Γ F y
where z = x+iy. Since dxdy k

y 2 and y f (z)g(z) are both Γ-invariant (by (3.2) and
Theorem 3.4.4), the integral is well-defined and independent of the choice
of fundamental domain. Moreover, it is easily seen to converge since both f
and g are cusp forms and |f (z)| = |g(z)| = O e−2πy as y → ∞.
Exercise 7.1.1. Let f , g ∈ Mk (Γ) and let F be any fundamental domain for the
congruence subgroup Γ. Show that the integral
ZZ
dxdy
y k f (z)g(z) 2 < ∞
F y
as long as one of f, g ∈ Sk (Γ).

(i) (f, g) is linear in f ,
(ii) (f, g) is conjugate symmetric. That is, (f, g) = (g, f ),
(iii) (f, f ) > 0 for f 6= 0.
Therefore, the Petersson inner product defines a Hermitian inner product on Sk (Γ).
Theorem 7.1.3. Suppose that Γ1 ⊂ Γ2 are congruence subgroups. Let f, g ∈

Sk (Γ2 ). Then (f, g) is independent of whether f, g are considered as elements of
Sk (Γ1 ) or Sk (Γ2 ).
86 7 The Petersson Inner Product
Proof. Let F2 be a fundamental domainSfor Γ2 , and take coset representatives

αj such that Γ2 = j αj Γ1 and F1 = j αj−1 (F2 ) is a fundamental domain
S
for Γ1 . Now
ZZ
1 dxdy
y k f (z)g(z) 2
SL2 (Z) : Γ1 F1 y
1 X ZZ dxdy
= y k f (z)g(z) 2
SL2 (Z) : Γ2 · Γ2 : Γ1 j −1
αj (F2 ) y
Z Z
1 X dxdy
= (Im αj−1 z)k f (αj−1 z)g(αj−1 z) 2 .
SL2 (Z) : Γ2 · Γ2 : Γ1 j F2 y

There are Γ2 : Γ1 summands and, since f, g ∈ Sk (Γ2 ), we find that they
are all equal to ZZ
dxdy
y k f (z)g(z) 2 ,
F2 y
so that the original integral is
ZZ
1 dxdy
y k f (z)g(z) ,
y2

SL2 (Z) : Γ2 F2
as desired. t
u
Exercise 7.1.4. Suppose α ∈ GL+ has integer entries. Let D = det α. If
2 (Q)
Γ(N ) ⊆ Γ for some N , show that Γ(N D) ⊆ α−1 Γα and Γ(N D) ⊆ αΓα−1 .
Exercise 7.1.5. Let α ∈ GL+
2 (Q) have integer entries. Show that there exists γ ∈
SL2 (Z) such that
ab
γ −1 α = ,
0d
where a and d are positive integers and b ∈ Z.
Exercise 7.1.6. Let f ∈ Mk (Γ) with Γ a congruence subgroup of SL2 (Z). Let α ∈
GL+2 (Q). Show that f |α is holomorphic at every cusp s ∈ Q ∪ {i∞}.
Exercise 7.1.7. Let f ∈ Sk (Γ) with Γ a congruence subgroup of SL2 (Z). Let α ∈
GL+2 (Q). Show that f |α vanishes at every cusp s ∈ Q ∪ {i∞}.
Theorem 7.1.8. Let Γ be a congruence subgroup of SL2 (Z). Let α ∈ GL+

2 (Q) have
integer entries and put Γ0 = α−1 Γα ∩ Γ. Then Γ0 is a congruence subgroup and
the map
f 7→ f |α
takes Mk (Γ) to Mk (Γ0 ) and Sk (Γ) to Sk (Γ0 ).
Proof. If D = det α, then by Exercise 7.1.4, Γ(N D) ⊆ α−1 Γα, where N is the
level of Γ. Thus Γ(N D) ⊂ Γ ∩ α−1 Γα ⊆ SL2 (Z) so that Γ0 is a congruence
subgroup of SL2 (Z).
7.2 Commutativity of the Hecke operators 87
If τ ∈ Γ0 , we can write τ = α−1 γα for some γ ∈ Γ so that
(f |α)|τ = f |(ατ ) = f |(γα) = (f |γ)|α = f |α
since f |γ for all γ ∈ Γ. By Exercise 7.1.6, f |α is holomorphic at the cusps for

Γ0 . The analogous result for Sk (Γ0 ) follows similarly from Exercise 7.1.7. t u
Exercise 7.1.9. Let Γ0 be a subgroup of GL+ 0

2 (R) and F a fundamental domain for
0 + −1 0
the action of Γ on H. If α ∈ GL2 (R), show that α F is a fundamental domain
for the action of α−1 Γ0 α on H.
Theorem 7.1.10. Let f, g ∈ Sk (Γ) with Γ a congruence subgroup of SL2 (Z). Let
α ∈ GL+
2 (Q). Then
(f, g) = (f |α, g|α),
where the latter inner product is with respect to Γ0 = α−1 Γα ∩ Γ.
Proof. By Theorem 7.1.8, f |α, g|α ∈ Sk (Γ0 ) where Γ0 = α−1 Γα ∩ Γ. Let Γ00 =
Γ ∩ αΓα−1 . Then α−1 Γ00 α = Γ0 , and so if F 00 is a fundamental domain for Γ00 ,
then α−1 F 00 is a fundamental domain for Γ0 , by Exercise 7.1.9. Thus,
ZZ
1 dxdy
(f |α, g|α) = 0
(f |α)(z)(g|α)(z)y k 2
SL2 (Z) : Γ −1
α F 00 y
ZZ
1 dxdy
= 0
f (z)g(z)y k 2
SL2 (Z) : Γ F 00 y
= (f, g)
by the invariance of the inner product and the fact that

h i
SL2 (Z) : Γ0 = SL2 (Z) : α−1 Γ00 α = SL2 (Z) : Γ00 . t
u
Exercise 7.1.11. Let α ∈ GL+ 2 (Q) have integer entries and determinant D. Let
α0 = Dα−1 . Show that for f, g ∈ Sk (Γ), with Γ a congruence subgroup of SL2 (Z),
we have
(f |α, g) = (f, g|α0 ).
7.2 Commutativity of the Hecke operators
Let Tn denote the nth Hecke operator for SL2 (Z), as developed in Section 5.2.
We prove that Tn Tm = Tm Tn so that the Tn ’s form a commuting family of
linear transformations on Mk (SL2 (Z)) and Sk (SL2 (Z)). We will use this fact
along with the multiplicative properties of the Hecke operators in the next
section to prove that the Tn ’s are Hermitian with respect to the Pettersson
inner product.
Theorem 7.2.1. Let (m, n) = 1. Then Tn Tm = Tmn = Tm Tn .

Proof. Let f ∈ Mk (SL2 (Z)) have q-expansion
∞
X
f (z) = λ(r)q r .
r=0
∞
X
By Theorem 5.2.4, we have Tn (f ) = b(r)q r , where
r=0
X nr
b(r) = dk−1 λ .
d2
d|(n,r)
∞
X
Writing Tm Tn (f ) = c(r)q r , we have
r=0
X mr X X mnr
c(r) = ek−1 b = ek−1 dk−1 λ .
e2 d2 e2
e|(m,r) e|(m,r) d|( n, mr )
e2
As (m, n) = 1, the inner sum over divisors d of (n, mr/e2 ) is really a sum
over divisors d of (n, r/e) as (n, m/e) = 1. Therefore,
X X mnr
c(r) = ek−1 dk−1 λ 2 2
d e
e|(m,r) d|(n, re )
X X mnr
= (ed)k−1 λ 2 2 .
d e
e|(m,r) d|(n,r)
As d runs over divisors of (n, r) and e runs over divisors of (m, r), de runs
over divisors of (mn, r). Thus
X mnr
c(r) = tk−1 λ
t2
t|(mn,r)
so that Tm Tn = Tmn when (m, n) = 1. t

u
Theorem 7.2.2. Let p be a prime number. For all s ≥ 1,
Tps Tp = Tps+1 + pk−1 Tps−1 .
Proof. Since the Hecke operators are defined in terms of the slash operator,
it is notationally convenient to introduce
j
−1 s−j
s pX
X p b
Tf
p s = ,
0 pj
j=0 b=0
7.2 Commutativity of the Hecke operators 89
where we view the right hand side as a “formal sum” of matrices. Then
k
ps(1− 2 ) Tps (f ) = f |Tf
ps .
Thus,
p−1
p0 X 1c
T
fp = + .
01 0p
c=0
Observe that for all a, b, d ∈ Z and all f ∈ Mk (SL2 (Z)), by the definition
of the slash operator (4.2) we have

ab a k/2 a b
f = f z+ , (7.1)
0d d d d

a0
f = f (z) . (7.2)
0a
Since f ∈ Mk (SL2 (Z)) is periodic, the right hand side of (7.1) depends only
on b (mod d), rather than b itself. In particular, we actually view Tfps as an
equivalence class of formal sums of matrices in which matrices are equiva-
lent if their corresponding slash operators are equal.
Although notationally Tps is a left action, Tf
ps acts from the right. Hence
we consider
j j
s pX −1 p−1 X
s pX −1
ps−j b
X s−j
p0 X 1c p b
T
fp Tf
ps = + . (7.3)
01 0 pj 0p 0 pj
j=0 b=0 c=0 j=0 b=0
Now
ps−j b ps+1−j bp

p0
=
01 0 pj 0 pj
and
ps−j b ps−j b + pj c

1c
=
0p 0 pj 0 pj+1
so that the first sum of (7.3) can be rewritten as
j
−1 s+1−j
s pX
ps+1 0

X p bp
+ , (7.4)
0 1 0 pj
j=1 b=0
while the second sum of (7.3) is

j
−1 s−j
p−1 pX
s X
b + pj c

X p
.
0 pj+1
j=0 c=0 b=0
In this last sum, the two inner summations can be combined to give
pj+1
X−1 s−j b0

p
.
0 pj+1
b0 =0
This combines with the first term

in equation
(7.4) to give
Tps+1 . Inequation
]
ps+1−j bp ps−j b
(7.4), the effect of the matrix 0 pj
is the same as 0 pj−1
by (7.2).
j
By (7.1), as the b’s range over residue classes (mod p ), we effectively get p
copies of each matrix
s−j 0
p b
, with 0 ≤ b0 ≤ pj−1 − 1,
0 pj−1
so that these combine to give pT]

ps−1 . Thus,
T
fp Tf
ps = Tps+1 + pTps−1 .
] ]
s( k −1)
Since Tf
ps = Tps /p 2 , we immediately deduce the theorem. t
u
Exercise 7.2.3. Let p be prime. Show that for all r, s ≥ 1, we have
Tp r Tp s = Tp s Tp r .
It may be easier to prove Exercise 7.2.4 before Exercise 7.2.3. In fact, after
Exercise 7.2.4, Exercise 7.2.3 becomes a trivial statement about commuting
polynomials.
Exercise 7.2.4. Let p be prime. Show that there is a polynomial Ps (x) ∈ Z[x] such
that Tps = Ps (Tp ).
Exercise 7.2.5. Prove that for all m, n ≥ 1, the Hecke operators commute:
Tn T m = Tm T n .
7.3 Hecke operators as Hermitian operators
We prove that Hecke operators are Hermitian.

Theorem 7.3.1. For any f , g ∈ Sk (SL2 (Z)) and any n ≥ 1,
(Tn (f ), g) = (f, Tn (g)).
Proof. By Theorem 7.2.1 and Exercise 7.2.4 we see that the algebra of Hecke
operators are generated by the Tp ’s, with p prime. Thus, to prove the theo-
rem, it suffices to show (Tp (f ), g) = (f, Tp (g)) for all primes p. By the defini-
tion of Tp ,
7.3 Hecke operators as Hermitian operators 91
p−1
( )
k
2 −1
p0 X 1b
Tp (f ) = p f + f
01 0p
b=0
and so by linearity
p−1
( )
k
2 −1
p0 X 1b
(Tp (f ), g) = p f ,g + f ,g
01 0p
b=0
By Exercise 7.1.11, we have

p0 10
f , g = f, g
01 0p
and
1b p −b
f ,g = f, g .
0p 0 1
1 ≤ b ≤ p − 1, we observe that α := 01 −1

For each b ∈ SL2 (Z) and β :=
0 1

−1 −b ∈ SL2 (Z). Now by Theorem 7.1.10, we have

p −b p −b
f, g = f |β, g β
0 1 0 1
Since f |β = f and g|α = g, we have

p −b p −b 1b
f |β, g β = f, g α β = f, g .
0 1 0 1 0p
Putting all of these together shows that (Tp (f ), g) = (f, Tp (g)). t

u
An immediate consequence of this theorem is:

Corollary 7.3.2. The eigenvalues of Tn are real algebraic numbers.
Proof. If Tn (f ) = λn f , then by the theorem (Tn (f ), f ) = (f, Tn (f )) so that

(λn f, f ) = (f, λn f ) which implies λn (f, f ) = λn (f, f ) so that λn = λn .
Hence the eigenvalues are real.
To see that they are algebraic numbers, we recall that by Corollary 4.4.4
any element of Sk (SL2 (Z)) is a polynomial in E4 and E6 . Therefore, Sk (SL2 (Z))
has a basis g1 , . . . , gr with the gi having rational Fourier coefficients in their
q-expansions. Since Tn (gi ) has again rational Fourier coefficients (see Theo-
rem 5.2.4), we can represent Tn by a rational matrix and so its characteristic
polynomial has rational coefficients. Therefore, the eigenvalues of Tn are al-
gebraic numbers. t
u
P∞
Exercise 7.3.3. Suppose f = n=1 an q n ∈ Sk (SL2 (Z)) is an eigenfunction for
all Tm . Prove that a1 6= 0 and that the eigenvalue of Tm is am /a1 .
By the previous exercise, we can speak of a “normalized” eigenfunction,

where we have a1 = 1. The Fourier coefficients of a normalized eigenfunc-
tion are the eigenvalues of the Tn ’s.
Exercise 7.3.4. Let f , g ∈ Sk (SL2 (Z)) be normalized eigenfunctions for all the
Hecke operators. Show that f = g or (f, g) = 0.
From this exercise, we can deduce that if f1 , . . . , fr is a maximal set of

distinct normalized eigenfunctions, then they are linearly independent. To
see this, suppose that j is the smallest index such that we can write fj+1 =
Pj
i=1 αi fi . Then
j j
!
X X
0 6= (fj+1 , fj+1 ) = αi fi , fj+1 = αi (fi , fj+1 ) = 0.
i=1 i=1
Hence, r ≤ dim Sk (SL2 (Z)). Applying a basic theorem of linear algebra, we

show in the next section that Sk (SL2 (Z)) always has a basis of simultaneous
eigenfunctions of all Hecke operators. Normalized eigenfunctions are also
called Hecke eigenforms.
7.4 Basis of eigenforms
We begin with a basic theorem from linear algebra.

Theorem 7.4.1. Let R be a commutative ring of Hermitian operators on a finite di-
mensional Hilbert space V . Then V has an orthogonal basis f1 , f2 , . . . , fr of eigen-
vectors of R.
Proof. We induct on the dimension of V . Since V is finite dimensional, the

space of linear operators on V is also finite dimensional, and so we can find
a finite set of operators, say {S1 , S2 , . . . , Sm }, that span R.
Assume dim V 6= 0. We first show by induction on m that V contains one
eigenvector f1 of S1 , S2 , . . . , Sm . If m = 1, the result is clear. Let λ1 be an
eigenvalue of S1 and let
V1 = {f ∈ V : S1 f = λ1 f }
be the corresponding eigenspace. Since Sj S1 = S1 Sj , we see that Sj V1 ⊆ V1 ,

for j = 2, . . . m. Therefore, V1 contains an eigenvector f1 of S2 . . . . , Sm by the
induction hypothesis. Then
V = Cf1 ⊕ (Cf1 )⊥ ,
where
(Cf1 )⊥ = {g ∈ V : (f1 , g) = 0}.
Clearly, the Sj ’s preserve this space since for g ∈ (Cf1 )⊥ , we have (f1 , Sj g) =
(Sj f1 , g) = 0 since f1 is an eigenfunction for the Sj ’s. By the induction hy-
pothesis on the dimension of V , we see that (Cf1 )⊥ has an orthogonal basis
f2 , . . . , fr . This completes the proof. t
u
Theorem 7.4.2. The space Sk (SL2 (Z)) has an orthogonal basis of Hecke eigen-
forms.
Proof. We apply Theorem 7.4.1 to the commutative ring of Hecke operators
acting on the finite dimensional vector space Sk (SL2 (Z)) to deduce the re-
sult. t
u
Exercise 7.4.3. Show that Ek is an eigenfunction for all the Hecke operators acting
on Mk (SL2 (Z)).
Exercise 7.4.4. Show that (Ek , f ) = 0 for any f ∈ Sk (SL2 (Z)).
Exercise 7.4.5. Show that ∆k is not a Hecke eigenform for any k ≥ 2.
Exercise 7.5.1. Let 12 | k and d = dim Sk (SL2 (Z)). For each 1 ≤ j ≤ d, define
∞
2(d−j)
X
j
fj := ∆ E6 = a(j) n
n q .
n=1
(j) (j)
Verify that an = 0 for n < j and aj = 1. Conclude that the fj form a basis for
Sk (SL2 (Z)).
Exercise 7.5.2. Let k ≥ 4 be even and d = dim Sk (SL2 (Z)). Choose non-negative
integers a, b such that 12 6= 4a + 6b ≤ 14 and 4a + 6b ≡ k (mod 12). For each
1 ≤ j ≤ d, define
∞
2(d−j)+b a
X
fj := ∆j E6 E4 = a(j) n
n q .
n=1
(j) (j)
Verify that an = 0 for n < j and aj= 1. Conclude that the fj form a basis for
Sk (SL2 (Z)). (This basis is called the Miller basis in the literature.)
Exercise 7.5.3. Let k, d, a, and b be as in Exercise 7.5.2. Verify that there exists a
basis g0 , g1 , . . . , gd for Mk (SL2 (Z)) such that
∞
X
gi = q i + b(i)
n q
n
n=d+1
(i)
with bn ∈ Z for all n and all i. Conclude that if
∞
X
f= cn q n ∈ Mk (SL2 (Z))
n=0
has c0 , c1 , . . . , cd ∈ Z, then cn ∈ Z for all n.
Exercise 7.5.4. Let d = dim Sk (SL2 (Z)) and fix a positive integer m. Suppose
∞
X
n=0
is such that for i = 0, 1, . . . , d, we have that ci ∈ Z and that ci ≡ 0 (mod m).

Prove that ci ≡ 0 (mod m) for all i ≥ 0.
Exercise 7.5.5. Let f be a Hecke eigenform for the full modular group. Show that
the Fourier coefficients of f are algebraic integers.
Chapter 8
Hecke Operators of Higher Level
8.1 Hecke operators for congruence subgroups
Let Γ1 and Γ2 be subgroups of GL+ +

2 (Q) and let α ∈ GL2 (Q). Recall the
double coset
Γ1 αΓ2 = {γ1 αγ2 : γ1 ∈ Γ1 , γ2 ∈ Γ2 }
from Example 2.4.5. The group Γ1 acts on the double coset from the left and
Γ2 acts from the right. Moreover, we can decompose the double coset as a
disjoint union of Γ1 -orbits:
a
Γ1 αΓ2 = Γ1 αγj
j
with suitable γj in Γ2 .
Exercise 8.1.1. Let p be prime and Γ = SL2 (Z). Let Xp be the set of matrices with
integer entries of determinant p. Show that

p0
Xp = Γ Γ.
01
Fix N ∈ N. Let S + be an additive subgroup of Z, i.e. S + = M Z for some

integer M . Let S × be a multiplicative subgroup of (Z/N Z)× . We shall also
use S × to indicate the pre-image of S × under the natural map Z → S × . Let
Xn = Xn (N, S × , S + )

ab × +
:= : a, b, c, d ∈ Z, ad − bc = n, N |c, a ∈ S , b ∈ S .
cd
Here are some examples:

Example 8.1.2. Take N = 1, S + = Z, and S × = {1}. Then Xn (1, S × , S + ) =
Xn from Section 5.2.
96 8 Hecke Operators of Higher Level
Example 8.1.3. The standard Hecke congruence subgroups can all be writ-
ten in this form:
Γ1 (N ) = X1 (N, 1, Z)
Γ(N ) = X1 (N, 1, N Z)
Γ0 (N ) = X1 (N, (Z/N Z)× , Z).
Exercise 8.1.4. For fixed N , S + , and S × , show that
Xm Xn ⊆ Xmn .

X1 (N, S × , S + )
is a congruence subgroup of SL2 (Z).
In particular, Exercise 8.1.5 shows that X1 is a group. Exercise 8.1.4 shows

that the set Xn is invariant under the left and the right action of X1 since
X1 Xn ⊆ Xn and Xn X1 ⊆ Xn . We apply Exercise 2.4.7 with H = SL2 (Z),
H1 = X1 , Y = Xn and X = Xn along with Theorem 5.2.1 to deduce that
|X1 \Xn | < ∞. Thus there is a finite set of orbit representatives αi ∈ Xn such
that a
Xn = X1 α i .
i
Moreover, multiplying the orbits on the right by any fixed γ ∈ X1 simply

permutes the orbits: a a
X1 α i γ = X1 α i . (8.1)
i i
We define the nth Hecke operator on Mk (X1 ) as the map which sends f ∈
Mk (X1 ) to the finite sum
k
X
Tn (f ) := n 2 −1 f |αi . (8.2)
i
Theorem 8.1.6. Let Γ = X1 (N, S × , S + ) as above. For any k, n ≥ 1, the Hecke

operator in (8.2) is a well-defined linear map from Mk (Γ) to Mk (Γ) and from Sk (Γ)
to Sk (Γ).
Proof. Suppose f ∈ Mk (Γ) and γ ∈ Γ. Then for any αi , since f |γαi = f |αi ,
we see that the map does not depend on the choice of orbit representatives
and is thus well-defined. Moreover,
8.1 Hecke operators for congruence subgroups 97
!
k
X
Tn (f ) γ = n 2 −1 f |αi γ
i
k
X
= n 2 −1 f |αi γ
i
= Tn (f ),
where the final step follows from (8.1) which shows that γ simply permutes
the orbits. By Exercise 7.1.6, each f |αi is holomorphic at every cusp Q ∪
{i∞} and hence the finite sum Tn (f ) is also holomorphic at every cusp. We
conclude that Tn (f ) ∈ Mk (Γ). If in addition f ∈ Sk (Γ), then by Exercise 7.1.7
we conclude that Tn (f ) ∈ Sk (Γ). The linearity of the map follows from the
linearity of the slash operator. t
u
We write down the Hecke operators explicitly in the case Γ = Γ1 (N ) =

X1 (N, 1, Z) in the following theorem.
Theorem 8.1.7. For each a ∈ (Z/N Z)× , fix σa ∈ SL2 (Z) such that
−1
a 0
σa ≡ (mod N ),
0 a
where a−1 denotes the multiplicative inverse of a (mod N ). Then Xn = Xn (N, 1, Z)

has the decomposition
d−1
a a ab
Xn = Γ1 (N )σa ,
0d
ad=n b=0
a>0
(a,N )=1
where the first disjoint union is taken over all positive integers a dividing n that are
coprime to N .
Proof. The terms on the right hand side are clearly contained in Xn . We first
show the union is a disjoint union. Suppose we had

a 1 b1 a 2 b2
γ 1 σa1 = γ 2 σa2 , γ1 , γ2 ∈ Γ1 (N ).
0 d1 0 d2
Then SL2 (Z) contains the matrix

−1
a 1 b1 a 2 b2 a1 /a2 ?
=
0 d1 0 d2 0 d1 /d2
implying that a1 = a2 and d1 = d2 since ai > 0. Thus,

a 1 b1 1j a 2 b2
=
0 d1 01 0 d2
for some j. Since b1 and b2 are both ≤ d − 1, we must have j = 0, which

means b1 = b2 .
To complete the proof, we need to
show that any element of Xn is con-
tained in the right hand side. Let αγ βδ ∈ Xn . Choose coprime integers g, h
such that gα + hγ = 0. (This can be done as the reader can directly verify.)
Let x, y be integers such that xh − gy = 1. Then set

xy
A= ∈ SL2 (Z).
gh
Now
αβ ??
A =
γ δ 0?
and has determinant n. Replacing A by ± 10 1j A if necessary, we may sup-

pose that
αβ ab a > 0,
A = ,
γ δ 0d 0 ≤ b ≤ d − 1.

Since αγ βδ ∈ Xn , considered modulo N this equality gives

1? ab
A ≡ (mod N )
0n 0d
which implies that A−1 has the form

−1
a ?
(mod N ),
0 a
which can be written as

−1
1? a 0
(mod N ).
01 0 a
Therefore we can find a matrix B ∈ Γ1 (N ) such that
A−1 = Bσa .
This completes the proof. t

u
This gives, for (n, N ) = 1, an explicit description of Tn which is

d−1
X X ab
Tn (f ) = nk/2−1 f σa .
0d
ad=n b=0
a>0
8.2 Nebentypus 99
8.2 Nebentypus
A Dirichlet character modulo N is a homomorphism χ : (Z/N Z)× → C× . This

implies χ(1) = 1.

Exercise 8.2.1. Let χ be a Dirichlet character modulo N . For each γ = ac db ∈
Γ0 (N ), define ψ(γ) = χ(d). Show that ψ(γ1 γ2 ) = ψ(γ1 )ψ(γ2 ).
Exercise 8.2.2. Show that the set of Dirichlet characters modulo N forms a group
of order φ(N ) under multiplication, where φ denotes the Euler function.

(
X φ(N ) if d ≡ 1 (mod N )
χ(d) =
χ
0 otherwise
where the sum is over all Dirichlet characters modulo N .
Exercise 8.2.4. If χ 6= 1 is a Dirichlet character modulo N , show that

X
χ(a) = 0.
a∈(Z/N Z)×
For any Dirichlet character χ modulo N , we define the following vector

subspace of Mk (Γ1 (N )):

Mk (Γ0 (N ), χ) := f ∈ Mk (Γ1 (N )) : f |γ = χ(d)f, ∀γ = ac db ∈ Γ0 (N ) .
In particular, if χ is the trivial character, then Mk (Γ0 (N ), χ) = Mk (Γ0 (N )).
Theorem 8.2.5. We have the decomposition

M
Mk (Γ1 (N )) = Mk (Γ0 (N ), χ),
χ
where the direct sum is over all Dirichlet characters modulo N .
Proof. We first show that any f ∈ Mk (Γ1 (N )) can be written as a sum of

functions fχ ∈ Mk (Γ0 (N ), χ). Let
1 X
fχ = χ(d)f |γd ,
φ(N )
d∈(Z/N Z)×
where γd ∈ Γ0 (N ) such that the lower right entry of γd is congruent to d

(mod N ). This is well-defined since f ∈ Mk (Γ1 (N )).
We now check that we
αβ
indeed have fχ ∈ Mk (Γ0 (N ), χ). For any γ = γ δ ∈ Γ0 (N ), we have
1 X 1 X
fχ |γ = χ(d)f |γd γ = χ(d)f |γdδ
φ(N ) φ(N )
d∈(Z/N Z)× d∈(Z/N Z)×
1 X
= χ(dδ)χ(δ)f |γdδ .
φ(N )
d∈(Z/N Z)×
Now as d runs through coprime residue classes modulo N , so does dδ. Thus,
we can change variables in the sum and deduce that
fχ |γ = χ(δ)fχ .
Moreover, fχ is holomorphic at all cusps by Exercise 7.1.6. In other words,

f ∈ Mk (Γ0 (N ), χ). Now
!
X X 1 X
fχ = χ(d) f |γd ,
χ ×
φ(N ) χ
d∈(Z/N Z)
and the inner sum is zero by Exercise 8.2.3 unless d = 1, in which case it is 1
so that X
f= fχ .
χ
To show the sum is a direct sum, we consider g ∈ Mk (Γ0 (N ), ψ). For any
character χ, we compute
  (
1 X g if χ = ψ
gχ =  ψχ(d) g =
φ(N ) 0 otherwise,
d∈(Z/N Z)×
by Exercise 8.2.4. In particular, if

M
g ∈ Mk (Γ0 (N ), ψ) ∩ Mk (Γ0 (N ), χ),
χ6=ψ
then  
X
g = gψ =  gχ  = 0. t
u
χ6=ψ
ψ
In the previous section, we defined Hecke operators for Γ1 (N ). Theorem

8.1.7 allows us to give the general formula for any n as
d−1
k
X X ab
Tn (f ) = n 2 −1 f |σa . (8.3)
0d
ad=n b=0
a>0
(a,N )=1
If f ∈ Mk (Γ0 (N ), χ), we have f |σa = χ(a)f so that

8.2 Nebentypus 101
d−1
k
2 −1
X X ab
Tn (f ) = n χ(a) f .
0d
ad=n b=0
a>0
We drop the condition (a, N ) = 1 since χ(a) = 0 if (a, N ) 6= 1.

P∞
Exercise 8.2.6. Let f ∈ Mk (Γ0 (N ), χ). If f (z) = m=0 λ(m)q m , then prove that
 
X∞ X mn
Tn (f ) =  χ(a)ak−1 λ 2
 qm .
m=0
a
a|m,n
If a modular form is an eigenvector of a Hecke operator, then we call it an

eigenfunction (or eigenform).
P∞
Exercise 8.2.7. If f = m=0 λ(m)q m ∈ Mk (Γ0 (N ), χ) is an eigenfunction for
all Tn ’s, show that λ(1) 6= 0 and λ(mn)λ(1) = λ(m)λ(n) for (m, n) = 1.
P∞
In light of Exercise 8.2.7, we say that an eigenvector f = m=0 λ(m)q m
is normalized if λ(1) = 1.
Exercise 8.2.8. For p|N , and f ∈ Mk (Γ0 (N ), χ), show that
p−1
k
2 −1
X 1b
Tp (f ) = p f .
0p
b=0
P∞
Exercise
P∞ 8.2.9. Let f = n=0 an q n ∈ Mk (Γ0 (N ), χ) and then write Tp (f ) =
n
n=0 b n q . Show that
bn = apn + χ(p)pk−1 an/p ,
where we interpret an/p = 0 when p - n.

P∞
Exercise 8.2.10. Let f = m=0 λ(m)q m ∈ Mk (Γ0 (N ), χ) be a normalized eigen-
function for all the Tn ’s. Show that for p prime,
λ(pa+1 ) = λ(p)λ(pa ) − χ(p)pk−1 λ(pa−1 )
for a ≥ 1.
As before, we deduce that the Tn ’s form a commutative algebra of oper-
ators. We now state (without proof) the action of these Hecke operators on
the Petersson inner product.
Theorem 8.2.11. Let f ∈ Mk (Γ1 (N )), and g ∈ Sk (Γ1 (N )). For each a ∈
(Z/N Z)× , fix σa ∈ SL2 (Z) such that
−1
a 0
σa ≡ (mod N ),
0 a
where a−1 denotes the multiplicative inverse of a (mod N ). If (n, N ) = 1,

then (Tn (f ), g) = (f |σn , Tn (g)). In particular, if f ∈ Mk (Γ0 (N ), χ), then
(Tn (f ), g) = χ(n)(f, Tn (g)).
This allows us to prove:

Corollary 8.2.12. Let (n, N ) = 1 and let χ be a Dirichlet character modulo N . Let
cn be a square root of χ(n). Then cn Tn is a Hermitian operator on Sk (Γ0 (N ), χ).
That is, (cn Tn (f ), g) = (f, cn Tn (g)).
Proof. This is clear from the theorem. Indeed, since cn is a root of unity, we
have (cn Tn (f ), g) = cn (Tn (f ), g) = cn χ(n)(f, Tn (g)) = cn cn 2 (f, Tn (g)) =
cn (f, Tn (g)) = (f, cn Tn (g)). t
u
Exercise 8.2.13. Show that Sk (Γ0 (N ), χ) has a basis of eigenforms for all the Tn
with (n, N ) = 1.
8.3 Oldforms and newforms
Exercise 8.2.13 only gives us a basis of eigenforms for all the Tn ’s under the
restriction that (n, N ) = 1. One would like a basis for all Tp ’s, including p|N .
Since all the Tp ’s commute, we know that the Tp ’s preserve each eigenspace.
If each eigenspace were one-dimensional, then we would deduce that the
basis is also a basis of eigenforms for Tp ’s with p|N . However, this need not
be the case. The problem arises from eigenforms arising from lower levels,
as the following examples show.
Example 8.3.1. Let f ∈ Sk (Γ0 (N ), χ) and r|N . If f is an eigenform for all

Tn with (n, N ) = 1, show that f |( 0r 10 ) is also an eigenform for all Tn with
(n, N ) = 1 with the same eigenvalues.
Solution. We have

r0 k
X X r0 ab
Tn f = n 2 −1 χ(a) f .
01 01 0d
ad=n b (mod d)
a>0
Now
r0 ab a br r0
=
01 0d 0 d 01
so that the sum becomes

k
2 −1
X X a br r0
n χ(a) f .
0 d 01
ad=n b (mod d)
a>0
As b runs through residue classes modulo d, so does br since (r, n) = 1.

Hence
r0 r0 r0
Tn f = Tn (f ) = λn f ,
01 01 01
where Tn (f ) = λn f . t
u
Exercise 8.3.2. Show that ∆(z) and ∆(2z) are both eigenforms for the Tn ’s (n odd)
acting on S12 (Γ0 (2)), with the same eigenvalues.
Exercise 8.3.3. Show that ∆(z) and ∆(2z) are linearly independent in S12 (Γ0 (2)).
In 1970, Atkin and Lehner realized how to resolve this difficulty by fo-
cusing on forms “genuinely” of level N . Indeed, let d1 d2 = N . Suppose f ∈
Mk (Γ1 (d1 )). Then f ∈ Mk (Γ1 (N )). If g(z) := f (d2 z), then g ∈ Mk (Γ1 (N ))
by Exercise 6.6.6. The subspace of Sk (Γ1 (N )) spanned by forms obtained in
these two ways from f ∈ Sk (Γ1 (d)), as we range over proper divisors d|N
is called the space of oldforms, denoted Skold (Γ1 (N )). One can show that the
Tn ’s preserve this space. The orthogonal complement of this space with re-
spect to the Petersson inner product is called the space of newforms, denoted
Sknew (Γ1 (N )). It turns out that this space has the multiplicity one property:
the intersection of the eigenspaces for all the Tn ’s with (n, N ) = 1 is one-
dimensional and so is also an eigenspace for the Tp ’s with p|N . This fact al-
lows us to construct a complete theory of L-series and functional equations
analogous to the level one case.
0 −1 −1

Exercise 8.4.1. Let wN = N 0 . Prove that wN Γ0 (N )wN ⊂ Γ0 (N ).
The action of wN on modular forms is called the Atkin-Lehner involution.
Exercise 8.4.2. Show that if f ∈ Mk (Γ0 (N )), then f |wN ∈ Mk (Γ0 (N )).
Exercise 8.4.3. Show that if f ∈ Sk (Γ0 (N )), then f |wN ∈ Sk (Γ0 (N )).
Exercise 8.4.4. Show that if f ∈ Mk (Γ0 (N )), then (f |wN ) |wN = f . Thus wN is,
in fact, an involution.
Chapter 9
Dirichlet Series and Modular Forms
9.1 General Dirichlet series
The general Dirichlet series is a discrete analog of the more familiar Laplace
transform of a function in the theory of complex variables. It is of the form
∞
X
an e−λn s (9.1)
n=1
where s ∈ C, λn ∈ R with λn monotonically increasing to infinity, and the

an ’s are complex numbers. Such series give us a convenient way to package
the sequence of an ’s for further study. For example, if λn = n, then equation
(9.1) becomes a power series
X∞
an xn
n=1
with x = e−s . The reader is probably familiar with the theory of power
series from a first course in calculus. Power series have a wide variety of
applications, most notably to combinatorial problems.
Ordinary Dirichlet series (which we will now consider) arise with λn =
log n. Then (9.1) becomes
∞
X an
(9.2)
n=1
ns
and such series are useful in analytic number theory, especially in the con-
text of multiplicative number theory. Indeed, series of the form (9.2) were in-
troduced in 1839 by P.G.L. Dirichlet to prove his celebrated theorem on the
infinitude of primes in a given arithmetic progression. However, Dirichlet
studied such series only for a real variable s. It was Riemann who initiated
in 1860 an ardent study of such series for complex variables of s. In his fa-
mous memoir, he focused on the special case of (9.2) with an = 1 for all n
106 9 Dirichlet Series and Modular Forms
and introduced (what we now call) the Riemann zeta function ζ(s), which
for Re(s) > 1 is given by the Dirichlet series
∞
X 1
s
. (9.3)
n=1
n
Exercise 9.1.1. Show that the series (9.3) converges absolutely for Re(s) > 1.
Exercise 9.1.2. Given a sequence of complex numbers an , we suppose that for any
> 0, there is a constant C() such that |an | ≤ C()n for all natural numbers n.
Show that
∞
X an
n=1
ns
converges absolutely for Re(s) > 1.
Riemann showed that ζ(s) can be analytically continued to the entire

complex plane, apart from a simple pole at s = 1 and that it satisfies a func-
tional equation relating ζ(s) with ζ(1−s). More importantly, Riemann noted
that (9.3) can be rewritten as an infinite product over prime numbers:
∞ Y −1
X 1 1
ζ(s) = = 1 − (9.4)
n=1
ns p
ps
for Re(s) > 1. It is this connection that relates the distribution of prime
numbers with the zeros of the Riemann zeta function.
Exercise 9.1.3. Prove the identity (9.4) for Re(s) > 1.
Exercise 9.1.4. Let f (n) be a multiplicative function. That is, f (mn) = f (m)f (n)
for (m, n) = 1. Assume that for any > 0, there is a constant C() such that
|f (n)| ≤ C()n . Show that for Re(s) > 1, we have
∞
f (p) f (p2 )

X f (n) Y
= 1 + + + · · · ,
n=1
ns p
ps p2s
where the product is over the prime numbers.
In our study of Dirichlet series, it is useful to begin with several results

from classical analysis.
Proposition 9.1.5 (Abel’s Lemma). Let m ∈ Z and suppose an and bn are se-
quences of complex numbers. Let
X
A(r) = an .
m≤n≤r
9.1 General Dirichlet series 107
Then
X X
an bn = A(r)br − A(m − 1)bm − A(n)(bn+1 − bn ).
m≤n≤r m≤n<r
Proof. We have
X X
a n bn = A(n) − A(n − 1) bn
m≤n≤r m≤n≤r
X X
= A(n)bn − A(n)bn+1 ,
m≤n≤r m−1≤n≤r−1
after a change of the index of summation. The two sums can be combined to
give X
A(r)br − A(m − 1)bm − A(n) (bn+1 − bn ) . t
u
m≤n<r
Exercise 9.1.6. Suppose that {an }∞ n=1 is a sequence of complex numbersP and f (u)
is a continuously differentiable function on the interval [1, x]. Let A(t) = n≤t an .
Prove that Z x
X
an f (n) = A(x)f (x) − A(t)f 0 (t)dt.
n≤x 1
√
Exercise 9.1.7. Let α > β > 0. Write z = x + iy, with x, y ∈ R, i = −1, and
x > 0. Show that
|z| −αx
e−αz − e−βz ≤ − e−βx .

e
x
Here is the main result regarding the region of convergence for a general
Dirichlet series.
Theorem 9.1.8. Let s0 ∈ C and suppose {an }∞

n=1 is a sequence of complex num-
bers. If the series
X∞
an e−λn s
n=1
converges for s = s0 , then it converges uniformly in any region of the form Re(s −
s0 ) ≥ 0, Arg(s − s0 ) ≤ α with α < π/2.
Proof. Without loss of any generality, we may assume s0 = 0 and that λn > 0
(since by our monotonicity condition for the λn ’s, only a finite number can
be negative and these terms converge for all values of s). We now apply
Abel’s lemma (Proposition 9.1.5). Since
∞
X
an
n=1
converges, we have that for any > 0, there is an N () such that for
X
A(r) = an ,
n≤r
we have
|A(r1 ) − A(r2 )| < for r1 , r2 > N ().
Applying Abel’s lemma with bn = e−λn s , we get
X X
an e−λn s = A(r)e−λr s −A(m−1)e−λm s − A(n) e−λn+1 s − e−λn s .

m≤n≤r m≤n<r
Writing s = σ + it, with σ, t ∈ R and applying Exercise 9.1.7, we get for

r, m > N () + 1,
 
X X
an e−λn s ≤ e−λr σ + e−λm σ + e−λn+1 s − e−λn s 
m≤n≤r m≤n<r
 
|s| X
≤ e−λr σ + eλm σ + e−λn σ − e −λn+1 σ

.
σ
m≤n<r
The last sum telescopes to give
e−λm σ − e−λr σ
so that

X
−λn s |s|
an e ≤ 2 1 +
σ
m≤n≤r
since e−λn σ ≤ 1 for σ > 0. Thus, in every domain where
|s|
≤ M,
σ
we have for m, r ≥ N () that
X
an e−λn s ≤ 2(M + 1).
m≤n≤r
In other words, in such a region, the series converges uniformly. But the re-
gion defined by the condition |s|/σ ≤ M is precisely the region cos arg(s) ≥
1/M . Since M can be arbitrarily large, this means arg(s) ≤ α with α < π/2.
t
u
Exercise 9.1.9. Suppose that
X
A(x) = an = O(xδ )
n≤x
for some δ > 0. Show that

∞ Z ∞
X an A(t)dt
s
=s
n=1
n 1 ts+1
and that the integral converges absolutely for Re(s) > δ.
The following exercise gives the analytic continuation of ζ(s) to the region
Re(s) > 0.
∞
{t}
Z
s
ζ(s) = −s dt,
s−1 1 ts+1
where {t} = t − btc and the integral converges absolutely for Re(s) > 0. Deduce
that lims→1+ (s − 1)ζ(s) = 1.
P∞ −λn s
Given a general Dirichlet series f (s) = n=1 an e , Theorem 9.1.8
shows that if it converges at s = s0 , then it converges for any s with
Re(s) > Re(s0 ), and the function thus defined is holomorphic in this region.
Thus, the set of values of s for which the Dirichlet series f (s) converges
contains a maximal open half-plane, called the half-plane of convergence. It is
convenient to view the empty set ∅ and the entire complex plane C as open
half-planes. If the half-plane of convergence is given by Re(s) > σ0 , we call
σ0 the abscissa of convergence. The half-plane of convergence of the Dirichlet
series
X∞
|an |e−λn s
n=1
is called the half-plane of absolute convergence of f . If we denote by σ1 its ab-

scissa of convergence, it is clear that
σ 0 ≤ σ1 .
Exercise 9.1.11. Show that the series

1 1 1 1
1− + s − s + s − ···
2s 3 4 5
has abscissa of convergence σ0 = 0 and abscissa of absolute convergence σ1 = 1.
Since the Dirichlet series

∞
X
f (s) = an e−λn s
n=1
converges for any s with Re(s) > σ0 , we can apply Theorem 9.1.8 to deduce
that if f (s) converges at s0 , then f (s) → f (s0 ) when s → s0 in the region
Re(s) ≥ Re(s0 ) and arg(s − s0 ) ≤ α with α < π2 . This follows from the
uniform convergence and the fact that e−λn s tends to e−λn s0 as s → s0 .

∞
X
n=1
can be identically zero only if all its coefficients an are zero.
A fundamental theorem in the theory of general Dirichlet series is the

following due to Landau.
Theorem 9.1.13 (Landau). Let

∞
X
n=1
be a Dirichlet series with non-negative coefficients. Then f (s) has a singularity at

its abscissa of convergence.
Proof. Let σ0 be its abscissa of convergence. Without loss of any generality,

we may suppose σ0 = 0 (since we can replace an by an e−λn σ0 ). If f (s) is
analytic at s = 0, we can find > 0 so that f (s) is analytic in the disc |s−1| ≤
1 + , centered at 1. In particular, its Laurent series converges there:
∞
X f (k) (1)
f (s) = (s − 1)k .
k!
k=0
Now
∞
X
f (k) (1) = (−1)k an λkn e−λn ,
n=1
since the Dirichlet series is uniformly convergent in its half-plane of absolute

convergence. Thus, our Laurent series becomes
∞ ∞
!
X 1 X k −λn
f (s) = a n λn e (1 − s)k ,
k! n=1
k=0
valid for |s − 1| ≤ 1 + . In particular, this is valid for s = −. Thus

∞ ∞
X 1 X
f (−) = an λkn e−λn (1 + )k
k! n=1
k=0
is convergent. This being a series of non-negative terms, we can re-arrange

the sums to get
∞ ∞
!
k k
X X λ (1 + )
f (−) = an e−λn n
.
n=1
k!
k=0
The inner sum is eλn (1+) so that we deduce

∞
X
a n e λn
n=1
converges. Thus, our Dirichlet series converges for s = − and hence for
Re(s) > −, by our earlier result. Hence the abscissa of convergence cannot
be zero.
Exercise 9.1.14. Prove the identity

∞ n+1
− β n+1
n+1
− δ n+1

X α γ
Tn
n=0
α−β γ−δ
1 − αβγδT 2
= .
(1 − αγT )(1 − αδT )(1 − βγT )(1 − βδT )
Exercise 9.1.15. A function a : N → C is called an arithmetical function. Given

two arithmetical functions a and b, we define the Dirichlet convolution a ∗ b, to
be the arithmetical function given by
X n
(a ∗ b)(n) = a(d)b
d
d|n
where the sum is over positive divisors d of n. Show that

∞ ∞
! ∞ !
X (a ∗ b)(n) X a(n) X b(n)
= .
n=1
ns n=1
ns n=1
ns
Exercise 9.1.16. An arithmetical function a : N → C is said to be completely

multiplicative if a(mn) = a(m)a(n) for all m, n. Show that
∞ −1
X a(n) Y a(p)
= 1− s .
n=1
ns p
p
Exercise 9.1.17. Given any arithmetical function a : N → C, we will write

∞
X a(n)
L(s, a) := .
n=1
ns
If a, b, c, d are completely multiplicative functions, show that
L(s, ac)L(s, ad)L(s, bc)L(s, bd)

L(s, (a ∗ b)(c ∗ d)) = .
L(2s, abcd)
Here ac denotes the arithmetical function given by (ac)(n) := a(n)c(n).
Exercise 9.1.18. Let X

σu (n) = du .
d|n
Show that for any complex numbers a, b,

∞
X σa (n)σb (n) ζ(s)ζ(s − a)ζ(s − b)ζ(s − a − b)
s
= .
n=1
n ζ(2s − a − b)
Exercise 9.1.19. Let t0 ∈ R. With a = it0 , b = −it0 in the previous exercise, show
that
∞
X (σit0 (n))2 ζ(s)2 ζ(s − it0 )ζ(s + it0 )
= .
n=1
ns ζ(2s)
Using Landau’s theorem (Theorem 9.1.13), show that ζ(1 + it0 ) 6= 0.
9.2 The Poisson summation formula
In section 5.3, we introduced the Dirichlet series associated to a modular

form of weight k on the full modular group. There we extended this Dirich-
let series to the entire complex plane and derived a functional equation re-
lating its value at s to k − s. This result is prototypical of a larger theme. One
can say that this theme begins with Riemann’s celebrated memoir of 1860
in which he introduced his zeta function and derived its analytic continu-
ation and functional equation using the rudimentary ideas from the theory
of Fourier transforms, most notably the Poisson summation formula. These
ideas are highlighted in the following theorem.
Let us suppose we are given two sequences of complex numbers {an }∞ n=0
and {bn }∞n=0 . We suppose that these sequences have polynomial growth, i.e.
an = O(nc ) and bn = O(nc ) for some c > 0. To each of these sequences, we
associate for t ≥ 0,
∞
X
f (t) = an e−πnt
n=0
X∞
g(t) = bn e−πnt .
n=0
9.2 The Poisson summation formula 113
We also associate the following Dirichlet series:

∞
X an
Lf (s) = ,
n=1
ns
∞
X bn
Lg (s) = s
.
n=1
n
Theorem 9.2.1. With f and g as above, assume that for t > 0,

1
f = wtk g(t)
t
for some complex number w and k ∈ R. Then Lf (s) and Lg (s) extend analytically
to the entire complex plane apart from possible simple poles at s = 0 and k and
satisfy the functional equation
π −s Γ(s)Lf (s) = wπ −(k−s) Γ(k − s)Lg (k − s).
The proof will show that if a0 = b0 = 0, there are no poles.
Proof. By the assumption of polynomial growth on the coefficients an , we

see that
∞
Z ∞ Z ∞ X !
s dt −πnt dt
(f (t) − a0 )t = an e ts
0 t 0 n=1
t
is an absolutely convergent integral. Moreover, Lf (s) converges absolutely

for Re(s) sufficiently large and so in this region the integral can be re-written
as
X∞ Z ∞
an e−πnt ts−1 dt = π −s Γ(s)Lf (s)
n=1 0
after an easy change of variable. Now

Z ∞ Z 1 Z ∞
s−1 s−1
(f (t) − a0 )t dt = (f (t) − a0 )t dt + (f (t) − a0 )ts−1 dt.
0 0 1
The first integral is

Z 1
a0
− + f (t)ts−1 dt.
s 0
We change t to 1/t in this integral expression to get

Z ∞
a0 1 −s−1
− + f t dt.
s 1 t
Since f (1/t) = wtk g(t), this becomes

Z ∞ Z ∞
a0 dt a0 dt wb0
− +w g(t)tk−s =− +w (g(t) − b0 )tk−s − ,
s 1 t s 1 t k−s
for Re(s) sufficiently large. All of this can be combined to give

Z ∞ Z ∞
−s a0 wb0 s dt dt
π Γ(s)Lf (s) = − − + (f (t) − a0 )t + (g(t) − b0 )tk−s .
s k−s 1 t 1 t
The right hand side is analytic for all s ∈ C apart from a simple pole at s = 0
with residue −a0 and s = k with residue −wb0 .
Changing f to g in the above identity shows that, upon using g(1/t) =
(1/w)tk f (t),
Z ∞ Z ∞
b0 (1/w)a0 dt dt
π −s Γ(s)Lg (s) = − − + (g(t)−b0 )ts + (f (t)−a0 )tk−s ,
s k−s 1 t 1 t
from which the functional equation is immediate. t

u
A wealth of examples of f, g satisfying the conditions of Theorem 9.2.1

arise from the theory of modular forms as was evident from Theorem 5.3.7
and Exercise 5.3.8. More examples can be derived by applying the Poisson
summation formula, which we now discuss.
Let F : R → C be a smooth function (that is, infinitely differentiable). We
define the Schwartz space S as the set of smooth functions F such that for
every pair of integers m, n ≥ 0,
sup xm F (n) (x) < ∞.

x∈R
It is clear that S is a vector space over the field of complex numbers.
Exercise 9.2.2. If F ∈ S, show that

Z ∞
tm F (n) (t)dt < ∞
−∞
for all pairs of integers m, n ≥ 0.
For F ∈ S, we define the Fourier transform F̂ by setting

Z ∞
F̂ (x) = F (t)e−2πitx dt.
−∞
Exercise 9.2.3. If F ∈ S, show that F̂ ∈ S.
The Fourier inversion formula shows that for F ∈ S, we have

Z ∞
F (t) = F̂ (x)e2πitx dx.
−∞
ˆ
In other words, F̂ (x) = F (−x).
Exercise 9.2.4. Show that Z ∞
2
e−πx dx = 1.
−∞
2 2
Exercise 9.2.5. For F (x) = e−πx , show that F̂ (u) = e−πu . In other words, F is
its own Fourier transform.
Exercise 9.2.6. Let F ∈ S and define G(t) = F (At + B) for real constants A, B,
with A 6= 0. Show that AĜ(t) = e2πiBt/A F̂ (t/A).
Theorem 9.2.7 (Poisson summation formula). Let F ∈ S. Then
X X
F (n) = F̂ (n).
n∈Z n∈Z
Proof. Let X
G(v) = F (n + v).
n∈Z
It is easily checked that G is a continuous function of v with period one. We

can compute the Fourier coefficients of G:
Z 1 Z 1 X !
−2πimv
cm : = G(v)e dv = F (n + v) e−2πimv dv
0 0 n∈Z
XZ n+1 Z ∞
= F (x)e−2πimx dx = F (x)e−2πimx dx = F̂ (m).
n∈Z n −∞
Since F̂ ∈ S, X
|F̂ (m)| < ∞
m∈Z
so X
G(v) = F̂ (m)e2πimv
m∈Z
is the Fourier series representing G. Setting v = 0 gives the Poisson summa-

tion formula. t
u
2
Exercise 9.2.8. Show that if F (x) = e−(Ax+B) π
with A, B ∈ R and A 6= 0, then
2
AF̂ (x) = e2πiBx/A e−π(x+B) .
Exercise 9.2.9. For any a ∈ R, show that

X 2 X 2
e−π(n+a) /x = x1/2 e−πn x+2πina .
n∈Z n∈Z
Exercise 9.2.10. Show that for x > 0,

X 2 X 2
e−πn x = x−1/2 e−πn /x .
n∈Z n∈Z
Following Riemann, we now derive the analytic continuation and func-

tional equation for the function ζ(s). Indeed, let
X 2
ϑ(z) = eπin z ,
n∈Z
so that f (t) = ϑ(it) satisfies

−1
f = t1/2 f (t)
t
by virtue of Exercise 9.2.10. We can then apply Theorem 9.2.1 with k = 1/2
to deduce that ζ(2s) extends analytically to the entire complex plane apart
from a possible simple pole at s = 0 and s = 1/2 and satisfies the functional
equation
1 1
π −s Γ(s)ζ(2s) = π −( 2 −s) Γ − s ζ(1 − 2s).
2
Changing s to s/2, we deduce the following.
Theorem 9.2.11 (Riemann, 1860). The function
Z ∞
dt
(ϑ(it) − 1) ts = 2π −s Γ(s)ζ(2s)
0 t
defined originally for Re(s) > 1/2, extends analytically to the entire complex plane
via the formula
Z ∞
−s 1 1 1 dt
2π Γ(s)ζ(2s) = − − 1 + (ϑ(it) − 1) (t2 + t 2 −s ) .
2 −s
s 1 t
Changing s to s/2 and dividing by 2 gives

s Z ∞
1 1 1 1−s
dt
π −s/2 Γ ζ(s) = − + (ϑ(it) − 1) ts/2 + t 2
2 s−1 s 2 1 t
and the right hand side is invariant under the map s → 1 − s. Moreover, the
expression is analytic for all s ∈ C except at s = 0 and s = 1 where it has a simple
pole.
Exercise 9.2.12. Show that (s − 1)ζ(s) extends to an entire function.
Exercise 9.2.13. Show that ζ(s) has simple zeros at s = −2n, for each positive
integer n.
Exercise 9.2.14. Show that ζ(0) = −1/2.
Exercise 9.2.15. Let f ∈ S. For x > 0 show that

X
f (nx)
n∈Z
converges absolutely and satisfies

X X n
x f (nx) = fˆ .
x
n∈Z n∈Z
Exercise 9.2.16. Let f ∈ S be an even function. That is, f (x) = f (−x). For any
x > 0, show that
∞ n ∞
X X 1 ˆ
f =x fˆ(nx) + xf (0) − f (0) .
n=1
x n=1
2
R∞
Exercise 9.2.17. Let f ∈ S. Set ζ(s, f ) = 0
f (x)xs dx
x . Show that the integral
converges for all s with Re(s) > 1.
Exercise 9.2.18. Let f ∈ S be an even function. With ζ(s, f ) defined as in the

previous exercise, show that ζ(s, f ) extends for all s ∈ C, apart from possible simple
poles at s = 0, 1 and satisfies the functional equation
ζ ? (s, f ) = ζ ? (1 − s, fˆ),
where ζ ? (s, f ) = ζ(s)ζ(s, f ).
It is clear from the previous discussion that the essential tool in deriv-
ing the analytic continuation and functional equation for the Riemann zeta
function is the transformation law of Exercise 9.2.10. The function
X 2
Θ(z) = e2πin z
n∈Z
is a classical theta-function. It is a modular form of weight 1/2 in the follow-

ing sense. Let d be an odd integer and c any integer. If d is a positive odd
prime number, the familiar Legendre symbol dc is defined as 0 if d|c, +1 if

c is a square (mod d) and −1 otherwise.Q This is extended by multiplicativity

for positive odd d. That is, if d = j pj with pj odd primes (not necessarily
distinct) then
c Y c
= .
d j
pj

0
For d = ±1, we adopt the convention ±1 = 1. For d negative, we define

c
c  |d| if c > 0
=
d − c if c < 0
|d|
c

Exercise 9.2.19. Show that the symbol d is completely multiplicative in each of
the variables.
√ To describe the modular transformation law of the Θ function, we define

z for z ∈ C to be the branch whose
argument is in the interval (−π/2, π/2].
Finally, for any z ∈ H and ac db , we set

1 if d ≡ 1 (mod 4)
εd =
i if d ≡ 3 (mod 4),
and c √
j(γ, z) := ε−1
d cz + d.
d
Then
Θ(γz) = j(γ, z)Θ(z) ∀γ ∈ Γ0 (4).
We will not prove this fact here, but only present it to indicate to the student
that our study has a natural entry into the theory of modular forms of half-
integral weight.
However, Riemann’s method of deriving the analytic continuation and
functional equation for ζ(s) extends to similar results for the Dirichlet L-
functions L(s, χ) defined as follows. Let χ : (Z/N Z)? → C be a Dirichlet
character. Then
∞
X χ(n)
L(s, χ) = .
n=1
ns
Riemann’s method can be adapted to derive the analytic continuation and

functional equation of L(s, χ). The discussion splits into two parts, accord-
ing as χ is even (that is, χ(−1) = 1) or χ is odd (that is, χ(−1) = −1).
We begin with a preliminary discussion of Dirichlet characters. Let χ0
denote the trivial character modulo N . Thus χ0 (n) = 1 if (n, N ) = 1 and
0 otherwise. If d|N and ψ is a Dirichlet character modulo d, then χ0 ψ is
a Dirichlet character modulo N . A Dirichlet character modulo N obtained
in this way from a proper divisor d of N (that is, d|N , d < N ) is called
imprimitive. Otherwise we call the character primitive. Primitive characters
modulo N are periodic with period N and no smaller period. By definition,
every imprimitive character χ of modulus N is induced from a character
whose modulus d is a proper divisor of N . The conductor of a character χ
is the least modulus d such that χ is induced from a (necessarily primitive)
character of modulus d.
Exercise 9.2.20. Let d|N and let ψ be a character modulo d. Prove that
Y ψ(p)

L(s, χ0 ψ) = L(s, ψ) 1− .
ps
p|N
The previous exercise shows that to derive the analytic continuation and
functional equation for L(s, χ), we need only focus our attention on the
primitive characters χ (mod N ). The following exercises reduce the prob-
lem to a simple application of the Poisson summation formula. To this end,
we introduce for every character χ (mod N ), the Gauss sum τ (χ), defined as
N
X m
τ (χ) = χ(m)e , where e(t) = e2πit .
m=1
N
This sum plays a fundamental role in number theory.
Exercise 9.2.21. If (n, N ) = 1, show that

N
X mn
χ(n)τ (χ) = χ(m)e ,
m=1
N
where χ denotes the complex conjugate of χ.
Theorem 9.2.22. If χ is a primitive non-trivial character modulo N , show that

N
X mn
χ(n)τ (χ) = χ(m)e .
m=1
N
That is, the formula in the previous exercise holds without the condition (n, N ) = 1.
Proof. By virtue of Exercise 9.2.21, we need to show the identity also holds
for (n, N ) > 1. In this case, the left hand side is zero and so it suffices to
show the right hand size is zero. Write n/N = n1 /N1 with (n, N1 ) = 1. Thus
N1 |N and N1 < N . If N1 = 1, then n is a multiple of N so that e(mn/N ) = 1
and the right hand side is zero since
N
X
χ(m) = 0
m=1
as χ is a non-trivial character. So we may suppose 1 < N1 < N . We have to

prove that
N
X mn1
χ(m)e = 0.
m=1
N1
Write N = N1 N2 and put m = aN1 + b with 0 ≤ a < N2 , 1 ≤ b ≤ N1 . The

sum in question can be rewritten as
X
X bn1
e χ(aN1 + b)
N1
1≤b≤N1 0≤a<N2
and it suffices to prove that the inner sum is zero. Let us write
X
S(b) = χ(aN1 + b).
0≤a<N2
Let us note that

N2
X X
S(b + N1 ) = χ(aN1 + b) = χ(b) + χ(aN1 + b)
a=1 1≤a<N2
= S(b).
Thus S(b) is periodic with period N1 .

We see that for any integer c satisfying (c, N ) = 1 and c ≡ 1 (mod N1 ),
X
χ(c)S(b) = χ(caN1 + cb)
0≤a<N2
and as a ranges over 0 ≤ a < N2 , we see that ca ranges over the same set
(mod N2 ). Thus,
χ(c)S(b) = S(bc).
Since c ≡ 1 (mod N1 ), we write c = qN1 + 1 and so S(bc) = S(bqN1 + b) =
S(b) as S(b) is periodic with period N1 . As χ is a primitive character modulo
N , it is not periodic to any modulus N1 with N1 a proper factor of N . Thus,
there are integers c1 , c2 such that (c1 , N ) = (c2 , N ) = 1, c1 ≡ c2 (mod N1 )
and χ(c1 ) 6= χ(c2 ). From the preceding, upon setting c ≡ c1 c−1 2 (mod N ), we
have that c ≡ 1 (mod N1 ) so that χ(c)S(b) = S(b). But χ(c) = χ(c1 )χ(c2 ) 6= 1
and we conclude S(b) = 0 as desired. t
u
Exercise 9.2.23. If χ is a primitive character modulo N , then |τ (χ)|2 = N .
The previous
√ exercise shows that if χ is a primitive character modulo N
then |τ (χ)| = N .
Now suppose χ is an even character (that is, χ(−1) = 1). We have
Z ∞
dx
Γ(s) = e−x xs
0 x
so that after a simple change of variables:

s Z ∞
−s/2 s/2 −s 2 dx
π N Γ n = e−n πx/N xs/2 .
2 0 x
We multiply this equation by χ(n) and sum over n ≥ 1 to get

∞
!
s Z ∞
−s/2
X 2 dx
π N s/2
Γ L(s, χ) = χ(n)e−n πx/N
xs/2
2 0 n=1
x
for Re(s) > 1. Since χ(−1) = 1 and χ(0) = 0, we can rewrite the sum in the
integrand as
∞
X 2
θ(x, χ) = χ(n)e−n πx/N .
n=−∞
Theorem 9.2.24. If χ is a primitive character modulo N is even, then

1/2
N 1
τ (χ)θ(x, χ) = θ ,χ .
χ x
Proof. The left hand side is

∞
X 2
τ (χ)χ(n)e−n πx/N
.
n=−∞
By Theorem 9.2.22,
N
X mn
χ(n)τ (χ) = χ(m)e
m=1
N
so that
N ∞
X X 2
τ (χ)θ(x, χ) = χ(m) e−n πx/N +2πimn/N
.
m=1 n=−∞
By Exercise 9.2.9, the inner sum is equal to

∞
1/2 X
N 2
e−(n+m/N ) πN/x
x n=−∞
so that
∞
N
1/2 X !
N X
−(nN +m)2 π/N x
τ (χ)θ(x, χ) = χ(m) e
x m=1 n=−∞
1/2 X
N 2
= χ(r)e−πr /N x
x
r∈Z
1/2
N 1
= θ ,χ . t
u
x x
We are now ready to derive the analytic continuation and functional

equation for L(s, χ) with χ even.
Theorem 9.2.25. Let χ be a primitive character modulo N . Suppose χ is even. Then

∞ ∞
dx N 1/2
s Z Z
1−s dx
2π −s/2 N s/2 Γ L(s, χ) = θ(x, χ)xs/2 + θ(x, χ)x 2 .
2 1 x τ (χ) 1 x
Both integrals converge for all s ∈ C and thus we obtain analytic continuation of
L(s, χ) to the entire complex plane. Moreover,

−s/2 s/2
s
−(1−s)/2 (1−s)/2 1−s
π N Γ L(s, χ) = wχ π N Γ L(1 − s, χ)
2 2
where wχ = N 1/2 /τ (χ).
Proof. Since
s Z ∞
−s/2 s/2 dx
2π N Γ L(s, χ) = θ(x, χ)xs/2
2 x
Z0 ∞ Z ∞
dx 1 dx
= θ(x, χ)xs/2 + θ , χ x−s/2
1 x 1 x x
after a simple change of variables. By Theorem 9.2.24,

x 1/2
1
θ , χ = τ (χ)θ(x, χ) .
x N
Therefore
∞ ∞
N 1/2
s Z Z
dx 1−s dx
2π −s/2 N s/2 Γ L(s, χ) = θ(x, χ)xs/2 + θ(x, χ)x 2 .
2 1 x 1 τ (χ) x
t
u

∞ ∞
3/2 X
X
−n2 πx/N +2πimn/N N m −πN (n+ m )2 /x
ne =i n+ e N .
n=−∞
x n=−∞
N
Exercise 9.2.27. Let χ be an odd Dirichlet character modulo N . Let

∞
X 2
ϑ1 (x, χ) = nχ(n)e−πn x/N
.
n=−∞
Show that
Z ∞
s+1 s+1 s+1 s+1 dx
2π −( 2 )Γ N 2 L(s, χ) = ϑ1 (x, χ)x 2 .
2 0 x
9.3 L-functions attached to modular forms 123
Exercise 9.2.28. Prove that if χ is an odd primitive Dirichlet character modulo N ,

then
1 3/2 τ (χ)
ϑ1 ,χ = x ϑ1 (x, χ)
x iN 1/2
Exercise 9.2.29. Let χ be an odd primitive Dirichlet character modulo N . Let

s+1
ξ(s, χ) := π −s/2 N s/2 Γ L(s, χ).
2
Then ξ(s, χ) is entire and satisfies the functional equation
ξ(s, χ) = wχ ξ(1 − s, χ),
where wχ = τ (χ)/iN 1/2 .
9.3 L-functions attached to modular forms
In Chapter 6, we defined the notion of the Fourier expansion of a modular

form at each cusp. Here we will study the nature of these expansions in
the context of the Dirichlet series attached to them. More precisely, let f ∈
Sk (Γ0 (N ), χ) be a cusp form and suppose that
∞
X
f (z) = an e2πinz
n=1
is its Fourier expansion at the cusp i∞. Let

0 −1
w=
N 0
and put g = f |w so that,P by a simple extension of Exercise 8.4.3, g ∈

∞ 2πinz
SK (Γ0 (N ), χ). Let g(z) = n=1 bn e be its Fourier expansion at i∞.
Thus, this is the Fourier expansion of f at the cusp 0. Since

k/2 −k 1
g(z) = N (N z) f −
Nz
we see that by putting z = it, t > 0, we have

i
g(it) = i−k N −k/2 t−k f .
Nt
We can apply the method of Theorem 9.2.1 to deduce:

Theorem 9.3.1. Let f and g be as above and let

∞
X an
Lf (s) = ,
n=1
ns
∞
X bn
Lg (s) = s
.
n=1
n
Then Lf (s) and Lg (s) extend to entire functions and satisfy the functional equation
Λf (s) = ik Λg (k − s),
where
√ !s
N
Λf (s) = Γ(s)Lf (s)
2π
√ !s
N
Λg (s) = Γ(s)Lg (s).
2π
Proof. We have Z ∞
(2π)−s Γ(s)Lg (s) = g(it)ts−1 dt.
0
Since g(it) = i−k N −k/2 t−k f (i/N t), the integral is

Z ∞
i
i−k N −k/2 f t−k+s−1 dt.
0 N
We put v = 1/N t and change variables to get

Z ∞ Z ∞
i−k N −k/2 f (iv)N k−s v k−s−1 dv = i−k N k/2−s f (iv)v k−s−1 dv
0 0
= i−k N k/2−s (2π)−(k−s) Γ(k − s)Lf (k − s)
from which the theorem follows. t

u
The assumption that f and g are cusp forms is not essential in the previ-
ous theorem, as shown in the following exercise.

0 −1
Exercise 9.3.2. Let f ∈ Mk (Γ0 (N ), χ) and g = f |w with w = . Sup-
N 0
posing that
∞
X
f (z) = an e2πinz
n=0
X∞
g(z) = bn e2πinz
n=0
are the Fourier expansions of f and g at i∞, prove that

√ !s
N
Λf (s) := Γ(s)Lf (s)
2π
√ !s
N
Λg (s) := Γ(s)Lg (s),
2π
P∞ P∞
where Lf (s) = n=1 anns and Lg (s) = n=1 nbns extend analytically to the entire
complex plane with poles at s = 0 and s = k. Moreover,
a0 b0 i k
Λf (s) + +
s k−s
and
b0 a0 ik
Λg (s) + +
s k−s
are entire and satisfy the functional equation
Λf (s) = ik Λg (k − s).
Exercise 9.3.3. With f , g as in the previous exercise, show that Lf (0) = −a0 , Lg (0) =
−b0 ,
k
2πi
lim (s − k)Lf (s) = b0 √ /Γ(k)
s→k N
and k
2πi
lim (s − k)Lg (s) = a0 √ /Γ(k).
s→k N
Theorem 9.3.1 and Exercise 9.3.2 have a converse. The reader will note
that the integral expressions for
a0 b0 i k
Λf (s) + +
s k−s
and
b0 a0 ik
Λg (s) + +
s k−s
are bounded in any vertical strip of the form σ1 ≤ Re(s) ≤ σ2 .
Using some fundamental results from complex analysis, we proceed in

the converse direction. The first result concerns the Γ function, defined for
Re(s) > 0 by the integral,
Z ∞
Γ(s) = e−t ts−1 dt.
0
A simple integration by parts shows that Γ(s + 1) = sΓ(s) and this allows
us to extend Γ(s) analytically to the entire complex plane with only simple
poles at s = 0, −1, −2, . . .
Stirling’s formula for the asymptotic behavior of Γ(s) as |s| → ∞ implies
that if σ is fixed, and |t| → ∞, then
√ π
|Γ(σ + it)| ∼ 2πe− 2 |t| |t|σ−1/2
(see, for example, Exercise 6.3.15 of [24]). This asymptotic formula allows
us to deduce that if Λf (s) and Λg (s) are bounded in every vertical strip,
then interval σ1 ≤ Re(s) ≤ σ2 , the series Lf (s) and Lg (s) are
for any fixed
1
O |t| 2 −σ eπ|t|/2 as |t| → ∞. A second result from complex analysis, namely
the Phragmén-Lindelöf theorem, allows us to sharpen this estimate. We state
this without proof but refer the reader to Theorem 8.1.13 of [24].
Theorem 9.3.4 (Phragmén-Lindelöf). Suppose that f (s) is analytic in the ver-
tical strip a ≤ Re(s) ≤ b, and that for some α ≥ 1,
α
|f (s)| = O e|t|
as |t| → ∞. If f (s) is bounded on the two vertical lines Re(s) = a and Re(s) = b,
then f (s) is bounded in the entire vertical strip.
We apply this theorem in our context via the following exercise.
Exercise 9.3.5. Suppose that f (s) is analytic in the vertical strip defined by a ≤
Re(s) ≤ b and that for some α ≥ 1,
α
f (s) = O e|t|

as |t| → ∞. If for some A> 0, f (s) = O |t|A on Re(s) = a and Re(s) = b, then
show that f (s) = O |t|A for all s in the vertical strip a ≤ Re(s) ≤ b.
In our context, both Lf (s) and Lg (s) are bounded in some fixed half-
plane. Moreover s(s − k)Lf (s) and s(s − k)Lg (s) are entire. By the functional
equation relating Λf (s) to Λg (k − s) and Stirling’s formula, we see that for
some A > 0,
s(s − k)Lf (s) = O |t|A ,

s(s − k)Lg (s) = O |t|A ,

in any vertical strip.
Exercise 9.3.6. Show that for any σ > 0,

Z σ+i∞
1
e−x = Γ(s)x−s ds.
2πi σ−i∞
Exercise 9.3.7. With f as in Exercise 9.3.2, show that for σ > 0 we have
Z σ+i∞
iy 1
f √ − a0 = Λf (s)y −s ds.
N 2πi σ−i∞
Using Exercise 9.3.7 and moving the line of integration to Re(s) = −U

(where U is positive and not an integer), we see by Cauchy’s Theorem that
Z −U +i∞
iy 1
f √ − a0 = Λf (s)y −s ds − a0 − ik bo y −k .
N 2πi −U −i∞
By the functional equation
Λf (s) = ik Λg (k − s),
we have
Z −U +i∞
iy 1
f √ = ik Λg (k − s)y −s ds − ik b0 y −k .
N 2πi −U −i∞
We replace k − s by s in the integral and obtain

Z k+U +i∞
iy 1
f √ = ik Λg (s)y s−k ds − ik b0 y −k .
N 2πi k+U −i∞
By Exercise 9.3.7, the integral is easily seen to be

iy
k
i g √ − b0 y −k .
N
We deduce
iy iy
f √ = ik y −k g √ .
N N
√
Changing y to 1/ N t, we get

i
f = ik N k/2 tk g(it).
Nt
In other words, for z = it,


1
f − = N k/2 z k g(z),
Nz
which now holds for all z ∈ H by the principle of analytic continuation. Thus
f |wN = g. This proves the converse of Theorem 9.3.1 and Exercise 9.3.2.
Our discussion can be summarized by the following theorem, due essen-
tially to E. Hecke.
Theorem 9.3.8 (Hecke). Suppose f and g are given by the Fourier series
∞
X
f (z) = an e2πinz
n=0
X∞
g(z) = bn e2πinz
n=0
P∞ an
with an , bn bounded by O(nα ) for some α ≥ 0. Put Lf (s) = n=1 ns , Lg (s) =
P ∞ bn
n=1 ns and
√ !s
N
2π
√ !s
N
Λg (s) = Γ(s)Lg (s)
2π
with N a positive number (not necessarily an integer). Set

0 −1
w=
N 0
and √ −k 1
f |w = Nz f −
Nz
where k is a fixed positive integer. Then the following are equivalent:
(a) g = f |w
(b) Λf (s) and Λg (s) extend to the entire complex plane and both
a0 b0 i k
Λf (s) + +
s k−s
and
b0 a0 i−k
Λg (s) + +
s k−s
are entire and bounded in every vertical strip and satisfy the functional equation
Λf (s) = ik Λg (k − s).
9.4 Twists of L-series attached to modular forms 129
9.4 Twists of L-series attached to modular forms
Given Lf as in the previous section, and χ a primitive Dirichlet character

modulo q, then it is natural to consider
∞
X an χ(n)
Lf (s, χ) = .
n=1
ns
Such a Dirichlet series is called a twist of Lf (s).

Exercise 9.4.1. Verify the matrix identity
1 −d2 u/r a + uc/r b − bcdu/r − cd2 u2 /r2

1 u/r ab
=
0 1 cd 0 1 c d − cd2 u/r
where ad − bc = 1.
We use this to prove:
Theorem 9.4.2. Let f ∈ Mk (Γ0 (q), χ) where χ is a Dirichlet character of conduc-
tor q ? with q ? |q. Let ψ be a primitive Dirichlet character modulo r. If
∞
X
f (z) = an e2πinz ,
n=0
then the “twisted series”

∞
X
fψ (z) = an ψ(n)e2πinz
n=0
is an element of Mk (Γ0 (N ), χψ 2 ), where N is the least common multiple of q, q ? r,

and r2 . Moreover, if f is a cusp form, then so is fψ .
Proof. Let τ (ψ) be the Gauss sum
X u
τ (ψ) = ψ(u)e .
r
u (mod r)
Since ψ is a primitive Dirichlet character modulo r, we have |τ (ψ)|2 = r by

Exercise 9.2.23. Also,
X un
τ (ψ)ψ(n) = ψ(u)e
r
u (mod r)
by Theorem 9.2.22. Thus we can write fψ in terms of f as

X 1 u/r
τ (ψ)fψ = ψ(u)f . (9.5)
0 1
u (mod r)
By Exercise 9.4.1,

1 u/r ab
α := =
0 1 cd 0 1 c d − cd2 u/r

so that if γ = ac db ∈ Γ0 (N ), then α ∈ Γ0 (N ) also, because N is the least
common multiple of q, q ? r, and r2 and N |c (hence r2 |c). Also

a?
α≡ (mod q ? ).
0d
Thus,
1 d2 u/r
X
τ (ψ)fψ | γ = ψ(u)χ(α)f
0 1
u (mod r)
since −1 −1

1 −d2 u/r

1 u/r ab
α = =γ
0 1 0 1 cd
and f |α = f . As (d, r) = 1, we see that d2 u (mod r) forms a complete set
of residue classes modulo r as u ranges over all residue classes modulo r.
Thus,

2
X 1 u/r
τ (ψ)fψ |γ = χ(d)ψ(d) ψ(u)f
0 1
u (mod r)
2
= χ(d)ψ (d)τ (ψ)fψ
by (9.5). Since τ (ψ) 6= 0, we deduce fψ ∈ Mk (Γ0 (N ), χψ 2 ) as claimed. More-

over, if f is a cusp form, then from (9.5), we see that fψ is also a cusp form.
t
u
Exercise 9.4.3. Let (q, r) = 1 and (u, r) = 1. Let d and w be integers such that
dr − quw = 1. Verify the matrix identity with N = qr2 ,

1 u/r 0 −1 0 −1 r −w 1 w/r
=r .
0 1 N 0 q 0 −qu d 0 1
Exercise 9.4.4. Let

0 −1 0 −1
wN = , wq = ,
N 0 q 0
and f ∈ Sk (Γ0 (q), χ) where χ is a Dirichlet character modulo q. Let ψ be a prim-

itive Dirichlet character modulo r with (q, r) = 1. With fψ as in Theorem 9.4.2,
N = qr2 , g = f |wq , show that
fψ |wN = w(ψ)gψ
where
w(ψ) = χ(r)ψ(q)τ (ψ)2 /r.
Given f ∈ Sk (Γ0 (q), χ) and ψ a primitive Dirichlet character modulo r,
we associate the “twisted” L-series:
∞
X an ψ(n)
Lf (s, ψ) =
n=1
ns
where the an ’s are the Fourier coefficients of f at i∞. We put

√ !s
N
Λf (s, ψ) = Γ(s)Lf (s, ψ).
2π
By Theorem 9.3.8 and Exercise 9.4.4, we deduce:

Theorem 9.4.5. Let f ∈ Sk (Γ0 (q), χ), ψ a primitive Dirichlet character modulo
r with (q, r) = 1 and N = qr2 . Then Λf (s, ψ) is an entire function, bounded in
vertical strips and satisfying the functional equation
Λf (s, ψ) = ik w(ψ)Λg (k − s, ψ)
where
w(ψ) = χ(r)ψ(q)τ (ψ)2 /r
and g = f |wq , with wq = 0q −1

0 .
There is a family of results that aim at deriving a sort of converse to

Theorem 9.4.5. This program of research goes under the name of “converse
theory” in the theory of modular forms (and more generally the theory of
automorphic forms). Theorem 9.3.8 shows that for two functions f, g with
Fourier expansions
∞
X
f (z) = an e2πinz
n=0
X∞
g(z) = bn e2πinz ,
n=0
with an , bn = O(nα ) for some α ≥ 0 and L-series

∞
X an
Lf (s) =
n=1
ns
∞
X bn
Lg (s) =
n=1
ns
related by the functional equation

√ !s √ !k−s
N k N
Γ(s)Lf (s) = i Γ(k − s)Lg (k − s),
2π 2π
we have g = f N0 −1

0 . In the
special case N = 1, the functional equation
implies that g(z) = f 01 −1 0 . In particular if f = g, we see that the func-
0 −1
tional equation implies f 1 0 = f . Since f |( 10 11 ) = f, we deduce that
f |γ = f for all γ in the group generated by ( 10 11 ) and 01 −10 which is the full
modular group SL2 (Z) by Theorem 2.1.2. If N > 1, one functional equation
is not sufficient. However, in a celebrated paper A. Weil [46] showed the fol-
lowing holds, which we state without proof. An elegant and simple proof of
this theorem in the English language has been found by Razar [32] and we
invite the student to study it.
Theorem 9.4.6 (Weil). Let k be a positive even integer and χ a character modulo
q. Suppose f and g are given by the Fourier series
∞
X
f (z) = an e2πinz
n=0
X∞
g(z) = bn e2πinz ,
n=0
with an , bn = O(nα ) for a fixed α ≥ 0. Let

√ s
q
2π
√ s
q
Λg (s) = Γ(s)Lg (s).
2π
Suppose that Λf (s) and Λg (s) have meromorphic continuations for all s ∈ C and
a0 b0 i k
Λf (s) + + ,
s k−s
b0 a 0 ik
Λg (s) + +
s k−s
are entire and bounded in vertical strips and satisfy Λf (s) = ik Λg (k − s). Further,
let R be a set of prime numbers coprime with q which meets every coprime residue
class modulo c for any c ≥ 1. For each r ∈ R and every primitive character ψ
modulo r, suppose that Λf (s, ψ) and Λg (s, ψ) are entire and bounded in vertical
strips and satisfy the functional equation given as in Theorem 9.4.5 with w(ψ)
as specified there. Then f ∈ Mk (Γ0 (q), χ), g ∈ Mk (Γ0 (q), χ) and g = f |wq .
Moreover, if Lf (s) and Lg (s) converge absolutely on some line Re(s) = σ with
0 < σ < k, then f and g are cusp forms.
Exercise 9.5.1. If τ (χ) denotes the Gauss sum, prove that
τ (χ) = χ(−1)τ (χ).
Exercise 9.5.2. For Re(c) > 0, let F (x) = e−c|x| . Show that
2c
F̂ (u) = .
c2 + 4π 2 u2
Exercise 9.5.3. Show that for Re(c) > 0,
∞
ec + 1 X 2c
= .
ec − 1 n=−∞ c2 + 4π 2 n2
Exercise 9.5.4. From the previous exercise, deduce Euler’s formula that
∞
X 1 π2
2
= .
n=1
n 6
Exercise 9.5.5. By differentiating the identity in Exercise 9.5.3, deduce that

∞
X 1
2k
∈ π 2k Q.
n=1
n
Exercise 9.5.6. For Re(c) > 0, and 0 ≤ v < 1, show that

X e2πinv 1

2 cosh cv

−cv
= e + .
c2 + 4π 2 n2 2c ec − 1
n∈Z
Exercise 9.5.7. Prove that for 0 ≤ v < 1,

X e2πinv
1

2
= 2 v − v + π2 .
n2 6
n6=0
Chapter 10
Special Topics
10.1 Elliptic functions
Let ω1 and ω2 be two non-zero complex numbers which are linearly inde-
pendent over R. They generate a lattice L which is the additive group con-
sisting of elements of the form
mω1 + nω2 , m, n ∈ Z.
We denote by Cb the extended complex plane C ∪ {∞}. A meromorphic func-

tion f : C → C is said to be an elliptic function with respect to L if
b
f (z + ω) = f (z) ∀z ∈ C, ∀ω ∈ L.
Sometimes, such functions are also called doubly periodic in the literature.
We can identify the space of such functions with meromorphic functions
on the complex torus C/L. The values of any such function are completely
determined by its values on the fundamental parallelogram which we take to
be the (closed) region in the complex plane enclosed by the parallelogram P
with vertices 0, ω1 , ω2 , and ω1 + ω2 :
P = {rω1 + sω2 : 0 ≤ r, s ≤ 1} .
It will be convenient to let P denote the interior of P along with the two
sides incident to the origin, including the origin itself:
P = {rω1 + sω2 : 0 ≤ r, s < 1} .
Sometimes we take any translate of P as a fundamental parallelogram (see

Exercise 10.1.4 below). It is easily verified that the set of all elliptic functions
for a lattice L, denoted E(L), forms a field.
Exercise 10.1.1. If f is a holomorphic elliptic function, then f is constant.
136 10 Special Topics
Exercise 10.1.2. If two elliptic functions (for a fixed lattice L) have the same poles
with the same principal parts, show that they differ by a constant.
Exercise 10.1.3. If f is an elliptic function with no poles on the boundary of the
fundamental parallelogram P , show that the sum of the residues of f in P is zero.
Exercise 10.1.4. Let f be an elliptic function with respect to a lattice L. Show that
there is an α ∈ C such that f is holomorphic on the boundary of the translate α+P ,
where P is the fundamental parallelogram of L.
Exercise 10.1.5. Show that there is no non-constant elliptic function having ex-
actly one simple pole in the fundamental parallelogram P .
Exercise 10.1.6. Let f be an elliptic function. Show that the number of zeros of f
in the fundamental parallelogram P is equal to the number of poles of f (counted
with multiplicity).
The order of an elliptic function f is defined as the number of poles of f
(counted with multiplicity) in the fundamental parallelogram. By the previ-
ous exercise, the order of f is also equal to the number of zeros of f (counted
with multiplicity) in P .
Exercise 10.1.7. If f is an elliptic function of order m, then for any c ∈ C, show
that the number of solutions of f (z) = c with z ∈ P is exactly m (counted with
multiplicities).
The simplest non-constant elliptic function with respect to the lattice L is
the celebrated Weierstrass ℘ function defined by
1 X 1 1

℘(z) := 2 + − 2 .
z (z − ω)2 ω
06=ω∈L
P
One might naively hope for a simpler elliptic function like ω∈L 1/(z − ω),
but this does not converge. Moreover, Exercise 10.1.5 implies that 2 is the
smallest possible order for a non-constant elliptic function. We now address
the question of convergence of the series defining ℘(z).
Exercise 10.1.8. For any s > 2, show that the series
X 1
ωs
ω∈L
ω6=0
is absolutely convergent.
Exercise 10.1.9. Show that the sum
X 1 1

− 2
(z − ω)2 ω
06=ω∈L
is absolutely convergent for any z ∈ C \ L.

10.1 Elliptic functions 137
By the solution to Exercise 10.1.9, the infinite sum defining ℘(z) converges
uniformly on compact subsets of C\L. Thus the sum defines a meromorphic
function on C. Now X 1
℘0 (z) = −2 3
ω∈L
(z − ω)
and it is clear that ℘0 (z) is a doubly periodic function, with periods ω1 and
ω2 . Hence for any ω ∈ L,
℘0 (z + ω) = ℘0 (z) ∀z ∈ C \ L.
Hence
℘ (z + ω) − ℘(z)
is a constant function of z. But for z = −ω/2, this term is ℘(ω/2) − ℘(−ω/2)
and since ℘(z) is an even function (an easy consequence of Exercise 10.1.9),
this term is zero. Hence we have proved:
Theorem 10.1.10. The Weierstrass ℘-function is an elliptic function of the lattice
L. Moreover, it is an even function of z. The derivative ℘0 (z) is an odd elliptic
function. Moreover, ℘(z) is holomorphic for all z ∈ P with z 6= 0. Its only poles are
at elements of L and these are all double order poles.
We now show that the field of elliptic functions on L is generated by ℘(z)
and ℘0 (z).
Theorem 10.1.11. Any even elliptic function on L is a rational function of ℘(z).
Any elliptic function f on L can be written as

f (z) = g ℘(z) + ℘0 (z)h ℘(z)
where g and h are rational functions.

Proof. It suffices to prove the first assertion because any function f (z) can be
written as f (z) = fe (z) + fo (z) with
f (z) + f (−z)
fe (z) = ,
2
which is even, and
f (z) − f (−z)
fo (z) = ,
2
which is odd. Now
0 fo (z)
fo (z) = ℘ (z)
℘0 (z)
and f0 /℘0 is an even elliptic function.
Now suppose f is even. Then ordz0 (f ) = ord−z0 (f ) for any z0 ∈ C. We
now show that if 2z0 ∈ L, then ordz0 (f ) is even. To see this we differentiate
the equation f (z) = f (−z) to get
i
f (i) (z) = (−1) f (i) (−z),
so that if 2z0 ∈ L, then f (i) (z0 ) = f (i) (z0 − 2z0 ) = f (i) (−z0 ) and this implies
f (i) (z0 ) = 0 for all odd i. In other words, if 2z0 ∈ L, then ordz0 (f ) is even.
Now let Pe = P/ ± 1 be “half” of the fundamental parallelogram. Con-
sider the function Y nv
g(z) = ℘(z) − ℘(v)
v∈Pe
v6=0
where (
ordv (f ) if 2v ∈
6 L
nv =
ordv (f )/2 if 2v ∈ L.
For 2v 6∈ L, the term ℘(z) − ℘(v) has zeros at the distinct points z = ±v, and
a pole of order 2 at z = 0. Hence the zeros are simple and so
nv
ordv ℘(z) − ℘(v) = ordv (f ).
If v ∈ 2L, then ℘(z) − ℘(v) has a double zero at the point z = v = −v. In this
way, we see that f (z) and g(z) have exactly the same zeros and poles (with
the same orders) except possibly at z = 0. But by Exercise 10.1.6, they have
the same order at z = 0 also. Therefore f (z)/g(z) is a holomorphic elliptic
function which, by Exercise 10.1.1, is a constant. t
u
Thus, the field E(L) of elliptic functions associated to a lattice L is gener-

ated by ℘ and ℘0 .
Exercise 10.1.12. Let X

Gk = ω −k
ω∈L
ω6=0
Prove that
∞
X
℘(z) = z −2 + (k + 1)Gk+2 z k
k=2
k even
is the Laurent expansion of ℘(z) at z = 0.
Observe that if k is odd, then Gk = 0 because the ω and −ω terms cancel.
Exercise 10.1.13. With notation as in the previous exercise, show that for z ∈ C\L,
℘0 (z)2 = 4℘(z)3 − 60G4 ℘(z) − 140G6 .

The previous exercise shows that the points ℘(z), ℘0 (z) for z 6∈ L lie on
the curve
y 2 = 4x3 − g2 x − g3 (10.1)
where
g2 = 60G4 , g3 = 140G6
Curves of the form (10.1) are called elliptic curves. The cubic polynomial on
the right hand side has discriminant given by
∆ = g23 − 27g32
Exercise 10.1.14. Let L be a lattice and g2 , g3 defined as above. Show that all of the
complex solutions of the equation
y 2 = 4x3 − g2 x − g3

are given by ℘(z), ℘0 (z) where ℘(z) is the Weierstrass ℘-function attached to L
and z ranges over all complex numbers in C \ L.
The Weierstrass ℘-function satisfies a remarkable addition formula which
we record in the following theorem.
Theorem 10.1.15. Suppose z, w, z + w ∈ C \ L. If z 6= w, we have
2
℘0 (z) − ℘0 (w)

1
℘ (z + w) = −℘(z) − ℘(w) + .
4 ℘(z) − ℘(w)
When z = w, we have
2
℘00 (z)

1
℘ (2z) = −2℘(z) + .
4 ℘0 (z)
Proof. Let (x1 , y1 ) = (℘(z), ℘0 (z)) and (x2 , y2 ) = (℘(w), ℘0 (w)) be the corre-
sponding points on the elliptic curve y 2 = 4x3 − g2 x − g3 . Let y = ax + b be
the line through these two points. Then
℘0 (z) = a℘(z) + b
℘0 (w) = a℘(w) + b.
Now the elliptic function
℘0 (u) − a℘(u) − b
has poles in the fundamental parallelogram P only at u = 0 and this pole has
order 3. By Exercise 10.1.6, it has three zeroes in P (counting multiplicities).
Working modulo L, we already have two of these zeros at u = z and u = w.
Let u = t be the third zero. Then
℘0 (t) = a℘(t) + b.
Now, for any elliptic function f it is a fact that

X
w ordw f ∈ L
w∈P
(see Exercise 10.5.1). Thus

z + w + t ∈ L.
Also,
℘0 (t)2 = 4℘(t)3 − g2 ℘(t) − g3 .
Thus, the cubic equation
(ax + b)2 = 4x3 − g2 x − g3
has roots x = ℘(z), ℘(w), and ℘(t). Since the sum of the roots of the cubic is
a2 /4, we get
a2
℘(z) + ℘(w) + ℘(t) = .
4
Since a is the slope of the line joining (x1 , y1 ) and (x2 , y2 ), we get
2
℘0 (z) − ℘0 (w)

1
℘(z) + ℘(w) + ℘(t) = .
4 ℘(z) − ℘(w)
Since t = −z − w (mod L) and ℘(−t) = ℘(t), we deduce the first assertion of

the theorem. The second assertion is obtained by taking the limit as z → w.
t
u
Exercise 10.1.16. Let L be the lattice spanned

by ω1, ω2 and let ℘(z)be the associ-
ated Weierstrass function. Show that ℘ ω21 , ℘ ω22 , and ℘ ω1 +ω 2
2
are the roots
of the cubic 4x3 − g2 x − g3 with g2 and g3 defined by (10.1).
Exercise 10.1.17. Suppose that ω1 , ω2 are linearly independent over R and that L
is the lattice spanned by ω1 , ω2 . Let ℘(z) be the associated ℘-function. Show that
℘ ω21 , ℘ ω22 , and ℘ ω1 +ω

2
2
are all distinct.
10.2 The modular invariant
For a complex number z ∈ H, let Lz be the lattice spanned by 1 and z. We

denote the corresponding g2 , g3 associated to Lz by g2 (z) and g3 (z) respec-
tively. Thus
X
g2 (z) = 60 (mz + n)−4
(m,n)6=(0,0)
X
g3 (z) = 140 (mz + n)−6 .
(m,n)6=(0,0)
10.2 The modular invariant 141
Consistent with our notation introduced in Chapter 4, these equations are

g2 (z) = 60G4 (z) and g3 (z) = 140G6 (z). These are modular forms of weight
4 and 6, respectively, for the full modular group SL2 (Z). We define the dis-
criminant function
∆0 (z) = g2 (z)3 − 27g3 (z)2 .
Theorem 10.2.1. For z ∈ H, we have ∆0 (z) 6= 0.
Proof. This statement is equivalent to showing that the roots of the associ-
ated cubic 4x3 − g2 (z)x − g3 (z) are all distinct. But this follows from Exer-
cise 10.1.17. t
u
It is clear that ∆0 (z) is a modular form of weight 12 for the full modular
group. It is related to the ∆(z) introduced in Chapter 4. Indeed, it is easily
verified that
12
∆0 (z) = (2π) ∆(z)
using our explicit knowledge of ζ(2k) given in Exercise 4.2.4. As before, we
let
E 3 (z)
j(z) = 4
∆(z)
and it is immediately checked that in our notation above,
g2 (z)3
j(z) = 1728 (10.2)
g2 (z)3 − 27g3 (z)2
which by the previous theorem is well-defined for z ∈ H. The j-function is

also called the modular invariant for several reasons. The first is that

az + b
j = j(z) ∀z ∈ H.
cz + d
However, now that we have related the j-function to elliptic curves, we will
elaborate on this relationship.

Exercise 10.2.2. If ω1 , ω2 are linearly independent over R and ac db ∈ SL2 (Z),
show that the lattice generated by aω1 + bω2 and cω1 +dω2 is the same as the lattice
generated by ω1 and ω2 .
0 0
L and ω1 , ω2 also generate the same
Exercise 10.2.3. If ω1 , ω2 generate the lattice
lattice L, show that there is a matrix ac db ∈ SL2 (Z) such that ω10 = aω1 + bω2
and ω20 = cω1 + dω2 .
Exercise 10.2.4. Let L be the lattice spanned by ω1 , ω2 with associated functions

g2 , g3 . If λ is a non-zero scalar, and L0 is the lattice spanned by λω1 , λω2 , show that
the corresponding elliptic curve for L0 is
y 2 = 4x3 − λ−4 g2 x − λ−6 g3 .

Given any elliptic curve E
y 2 = 4x3 − g2 x − g3 ,
we define
1728g23
j(E) = .
g23 − 27g32
One can show that j(E) is isomorphism invariant. In fact E1 ∼
= E2 over C if
and only if j(E1 ) = j(E2 ).
In Theorem 4.5.1 we proved that the map z 7→ j(z) is a bijection from
the standard fundamental domain F onto the complex numbers C. This fact
allows us to prove the following uniformization theorem.
Theorem 10.2.5. Given any two A, B ∈ C with A3 − 27B 2 6= 0, there exists a
lattice L such that its associated Weierstrass function satisfies
℘0 (z)2 = 4℘(z)3 − A℘(z) − B.
Proof. We consider two cases. The first case is when A = 0. The second case
is when A 6= 0.
In the first case (A, B) = (0, B) with B 6= 0. We choose z0 ∈ H so that
g2 (z0 ) = 0. This can be done since g2 (z) is, up to a scalar multiple, E4 (z)
and by Exercise 4.3.5 z = ρ works. Since ∆0 (z0 ) 6= 0, we have g3 (z0 ) 6= 0.
Choosing λ so that λ−6 g3 (z0 ) = B, the lattice L = [λ, λz0 ] has the required
property, by Exercise 10.2.4.
Now consider the case when A 6= 0. Let a = B 2 /A3 . Since A3 − 27B 2 6= 0,
we have a 6= 1/27. By Theorem 4.5.1, we can choose z0 ∈ H so that j(z0 ) =
1728/(1 − 27a). If B = 0, then we choose λ so that λ−4 g2 (z0 ) = A and we see
that L = [λ, λz0 ] works. If B 6= 0, we choose λ so that
Ag3 (z0 )
λ2 =
Bg2 (z0 )
and consider the lattice L = [λ, λz0 ]. From (10.2) we deduce that a =
g3 (z0 )2 /g2 (z0 )3 . Thus for the lattice L, the associated values of g2 and g3
are A and B respectively. Indeed,
g2 (z0 ) B 2 g2 (z0 )3
g2 = 4
= 2 =A
λ A g3 (z0 )2
and
g3 (z0 ) B 3 g2 (z0 )3
g3 = = =B
λ6 A3 g3 (z0 )2
as desired. t
u
The previous theorem shows that all elliptic curves over C have associ-
ated with them a lattice L and a Weierstrass ℘-function such that (℘(z), ℘0 (z))
10.3 Theta series and lattices 143
parametrizes all the complex points of the curve. By Exercise 10.2.2, we

see that for any given elliptic curve E, any two lattices that give rise to E
are SL2 (Z)-equivalent. Since ω1 , ω2 are linearly independent over R, we
can ensure that our lattice satisfies Im (ω2 /ω1 ) > 0. If z = ω2 /ω1 with
Im (ω2 /ω1 ) > 0, then the j-invariant of E is given by the modular invari-
ant j(z). Combined with the remark after Exercise 10.2.4 and Theorem 4.5.1,
we see that to each point z in the standard fundamental domain F, we can
associate an elliptic curve Ez and this is a one-to-one correspondence. Thus,
points of F parametrize all elliptic curves. (This is an example of a moduli
space.)
Using our earlier notation, we see that
E43 (z)
j(z) = .
∆(z)
2E4 (z)E60 (z) − 3E6 (z)E40 (z)
is a cusp form of weight 12 for the full modular group.
Exercise 10.2.7. Prove that E42 E6 = E14 . Deduce that

X
m/(mρ + n)13 = 0,
(m,n)6=(0,0)
where ρ is a primitive cube root of unity.

E14 (z)
j 0 (z) = −2πi .
∆(z)
10.3 Theta series and lattices
In this section, we describe a general method of constructing modular forms

arising from theta series associated to lattices.
Let V be a vector space over R of dimension n equipped with an inner
product. That is, we have a map
(·, ·) : V × V → R
which is bilinear, symmetric and positive definite.

Recall the Kronecker delta function
(
1 if i = j,
δij =
0 if i 6= j.
Let µ be an invariant measure on V . That is, for any measurable subset U

of V , we have µ (v + U ) = µ (U ) for all v ∈ V (translation invariance) and
µ (λU ) = λn µ (U ) for any λ ∈ R. Also, if e1 , . . . , en is a basis of V such that
(ei , ej ) = δij then the set
 
X n 
λj ej : 0 ≤ λj ≤ 1, 1 ≤ j ≤ n
 
j=1
has measure 1.
A lattice L of V is a subset of V which can be written as
L = ω1 Z + · · · + ωn Z
where {ω1 , . . . , ωn } is a basis of V . The volume of L is defined as

 
X n 
µ (V /L) := µ  λj ωj : 0 ≤ λj ≤ 1, 1 ≤ j ≤ n 
 
j=1
Recall that the dual vector space V ? of V is the vector space of linear function-
als on V . Since V is an inner product space these functionals are of the form
(·, v0 ) for some v0 ∈ V (Riesz representation theorem). We define the dual
lattice L? of L as the collection of elements v ? ∈ V ? such that v ? (ω) ∈ Z for
all ω ∈ L.
Exercise 10.3.1. If L = ω1 Z + · · · ωn Z and if ω1? , . . . , ωn? ∈ L? is the dual basis,

that is ωi? (ωj ) = δij , show that
L? = ω1? Z + · · · + ωn? Z
To each lattice L we associate a theta series

X
θL (z) = eπiz(ω,ω)
ω∈L
for z ∈ H. We first address the question of convergence.

Since each ω ∈ L is of the form
n
X
ω= xi ωi , xi ∈ Z,
i=1
we see that
n X
X n
(ω, ω) = (ωi , ωj )xi xj
i=1 j=1
is a positive definite quadratic form. With respect to the basis ω1 , . . . , ωn ,

this quadratic form is represented by a positive definite real symmetric ma-
trix. As such, it can be diagonalized by an orthogonal transformation (by a
fundamental theorem in linear algebra). Moreover, the eigenvalues are real
and positive (since the matrix is positive definite). In other words, there is
λmin > 0 such that
(ω, ω) ≥ λmin x21 + · · · + x2n .

Thus if z = u + iv with v > 0, we have

X 2 2
|θL (z)| ≤ e−πvλmin (x1 +···+xn )
x1 ,...,xn ∈Z
!n
−πλmin vx2
X
≤ e < ∞.
x∈Z
This proves:
Theorem 10.3.2. The theta series θL (z) is a holomorphic function for z ∈ H.
Proof. From the proof immediately above, we easily see that the series repre-
senting θL (z) converges uniformly and absolutely on every compact subset
of H (since for any compact set in H, v is bounded from below.) Thus θL (z)
is holomorphic there. t
u

The matrix A = (ωi , ωj ) is called the Gram matrix associated to the
lattice L.
√
Exercise 10.3.3. Prove that µ(V /L) = det A.
Exercise 10.3.4. With A as above, show that the matrix A−1 is the Gram matrix of
the dual lattice L? .
To derive the modular transformation law for θL (z), we need the follow-
ing version of Poisson’s summation formula for lattices.
Let V be as above a vector space over R of dimension n. We view V as
being isomorphic to Rn with the usual Euclidean topology. A bounded func-
tion f : V → R is called rapidly decreasing if it can be partially differentiated
arbitrarily often and each of these partial derivatives are bounded.
The Fourier transform of f is a map fˆ : V ? → R defined by
Z
fˆ(x) = f (u)e−2πi(u,x) dµ(x)
V
where (u, x) denotes x(u). Once can show that fˆ is again a rapidly decreas-
ing function on V ? .
Theorem 10.3.5. For f rapidly decreasing,
X 1 X
f (ω) = fˆ(ω ? ).
µ(V /L) ? ?
ω∈L ω ∈L
Proof. We use the basis of L to establish a topological isomorphism with Rn

so that L is then mapped to Zn . This is a lattice of volume 1. Identifying µ
with the product measure dx1 · · · dxn , we reduce the identity to be proved
to X X
f (x) = fˆ(x)
x∈Zn x∈Zn
which is the classical Poisson summation formula essentially proved earlier

in Chapter 9. t
u
This allows us to prove the following fundamental theorem.
Theorem 10.3.6. Let L be a lattice as above. Then

1 z n/2
θL ? − = µ (V /L) θL (z)
z i
for all z ∈ H. (Here, the square root is chosen so that it is positive for any positive
real number.)
Proof. By complex analysis, it suffices to prove the identity for purely imag-
inary z. In other words, it suffices to show that
X ? ? X
e−π(ω ,ω )/y = y n/2 µ (V /L) e−πy(ω,ω) .
ω ? ∈L? ω∈L
Now let f : V → R be given by f (x) = e−π(x,x) . Then f is rapidly de-

creasing and fˆ = f once we identify V with V ? using the pairing pro-
2
vided by the inner product. Since the Fourier transform of e−π|x| is itself,
we now proceed as follows. Let Ω = y 1/2 L. Then Ω ? = y −1/2 L? so that
µ (V /Ω) = y n/2 µ(V /L). The Poisson summation formula applied to Ω and
f now yields
X X
y n/2 µ (V /L) e−π(x,x) = e−π(x,x)
x∈y 1/2 L x∈y −1/2 L?
and a simple change of variables in the summation now gives the desired
result. t
u
A lattice L is said to be self-dual if L? = L. It is said to be even if ||ω||2 ∈ 2Z
for all ω ∈ L.
10.4 Special values of zeta and L-functions 147
Exercise 10.3.7. If L is self-dual and even, show that 8 | n.
Exercise 10.3.8. If L is self-dual show that the Gram matrix of L and L? have
determinant 1.
Exercise 10.3.9. Show that the theta function of a self-dual, even lattice is a modu-
lar form of weight n/2 for the full modular group SL2 (Z).
10.4 Special values of zeta and L-functions
The logarithmic derivative of the Γ -function is called the digamma function.

Higher derivatives of the digamma function give rise to the polygamma func-
tions. More precisely, the digamma function is given by
∞
1 X 1 1
ψ(z) = −γ − − − ,
z n=1 n+z n
so that the polygamma functions ψk (z) are given by

∞
X 1
ψk (z) = (−1)k−1 k! .
n=0
(n + z)k+1

(−1)k k!
ψk (z + 1) = ψk (z) + .
z k+1
Exercise 10.4.2. For every integer k ≥ 0, prove that
dk k!
− k
(π cot πz) = ψk (z) + (−1)k+1 ψk (−z) + (−1)k k+1 .
dz z
dk
− (π cot πz) = ψk (z) + (−1)k+1 ψk (1 − z).
dz k
Let N be a natural number and let χ : (Z/N Z)∗ → C be a Dirichlet char-
acter and define the L-series L(s, χ) by
∞
X χ(n)
L(s, χ) := .
n=1
ns

N
X
(k − 1)!L(k, χ) = (−N )−k χ(a)ψk−1 (a/N ).
a=1
Exercise 10.4.5. If χ is an even character (that is, χ(−1) = 1) and k is even, show
that
N
X dk
−2(k − 1)!L(k, χ) = (−N )−k χ(a) k (π cot πz) .
a=1
dz z=a/N
Deduce that L(k, χ) is an algebraic multiple of π k .
Exercise 10.4.6. If χ is an odd character (that is, χ(−1) = −1) and k is odd, show
that
N
X dk
−2(k − 1)!L(k, χ) = (−N )−k χ(a) k (π cot πz) .
a=1
dz z=a/N
The gist of the preceding exercises is that the value of L(k, χ) when k
and χ have the same parity, is a non-zero algebraic multiple of π k . This fact
was first proved by Hecke [14] and he noted that this implies an interesting
result for real quadratic fields. Namely, if F is a real quadratic field, then
ζF (2m) is an algebraic multiple of π 4m . This motivated him to ask if such
a result holds generally for any totally real field F . That is, if F is totally
real of degree r over the rationals, then is it true that ζF (2m) is an algebraic
multiple of π 2rm ? This was anwered in the affirmative by Siegel and Klingen
and is now known as the Siegel-Klingen theorem.
The proof of the Siegel-Klingen theorem makes use of the theory of classi-
cal modular forms and Hilbert modular forms. Since a detailed explanation
of the proof is beyond the scope of this book, we content ourselves with
a brief outline which should enable the interested student to make a more
detailed study.
We summarise the contents of the appendix in [39]. For the relevant back-
ground in algebraic number theory, we refer the student to [25]. For a de-
tailed study of Hilbert modular forms, we suggest the book by Garrett [10].
Let F be a totally real number field of degree r and discriminant D. Let
x 7→ x(i) be an indexing of real embeddings of F and let OF be the ring of
integers of OF . The group SL2 (OF ) is called the Hilbert modular group and
it acts on Hr via the map:
a(1) z1 + b(1) a(r) zr + b(r)

ab
g := ∈ SL2 (OF ), g·(z1 , .., zr ) = , ..., (r) .
cd (1)
c z1 + d (1) c zr + d(r)
For a, b ∈ F and z ∈ Hr , it is convenient to use the notation

r
Y
N (az + b) := (a(i) zi + b(i) ).
i=1
Let k be a natural number (which is even if K has a unit of norm −1). For an
ideal a, we define the analogue of the Eisenstein series as
X
Fk (a, z) = N (a)k N (λz + µ)−k ,
(λ,µ)6=(0,0)
where the sum runs over a complete system of pairs of numbers in the ideal
a different from (0, 0) and not differing from one another by a factor which
is a unit (that is, not associated). One can show that the series converges
absolutely for k > 2 and for k = 2, one can apply a limit process (Hecke’s
trick). The function Fk (a, z) is an example of a Hilbert modular form in the
sense that
Fk (a, gz) = N (cz + d)k Fk (a, z) ∀ g ∈ SL2 (OF ).
As OF is a lattice and its dual is the inverse different d−1 of OF (see [25]), one
has a Fourier expansion of Fk (a, z) of the form
r
(2πi)k
X
ζ(a, k) + D1/2−k σk−1 (a, ν)e2πiT r(zν) ,
(k − 1)!
ν∈d−1 ,ν0
where ν runs over totally positive numbers in d−1 and

X
ζ(a, k) = N (a)k N (µ)−k ,
a|(µ)
X
σk−1 (a, ν) = sign(N (α)k )N ((α)ad)k−1 .
d−1 |(α)a|ν
The summation is over principal ideals (µ), (α) under the conditions given.
If we set all the variables z1 , ..., zr to z, then Fk (a, z) becomes a classical
modular form of weight rk. By Exercises 5.5.18, 5.5.19, and 5.5.20, we deduce
that ζ(a, k) is a an algebraic multiple of π kr . Since the Dedekind zeta function
ζF (k) can be written as a rational linear combination of the values ζ(a, k), the
Siegel-Klingen theorem follows from this.
Exercise 10.5.1. Let f be an elliptic function with period lattice L. Show that
X
ordz (f )z ∈ L
z∈C/L
Exercise 10.5.2. With ℘ denoting the Weierstrass ℘-function as in Section 10.1,

prove that
℘00 (z) = 6℘(z)2 − g2 /2.
Exercise 10.5.3. With Gk as in Exercise 10.1.12, prove that for m ≥ 4,

m−2
X
(2m + 1)(m − 3)(2m − 1)G2m = 3 (2j − 1)(2m − 2j − 1)G2j G2m−2j .
j=1
Exercise 10.5.4. Show that for k ≥ 2, each G2k can be expressed as a polynomial
in G4 and G6 with positive rational coefficients.
Exercise 10.5.5. Let ϑ(z) be given by the series

∞
X 2
eπin z .
n=−∞
Show that
∞
Y
ϑ(z) = (1 − e2πinz )(1 + e(2n−1)πiz )2 .
n=1
Deduce that ϑ(z) does not vanish in the upper half-plane.
Exercise 10.5.6. With η being the Dedekind η-function and ϑ as in the previous
exercise, show that
ϑ(z) = η 2 ((z + 1)/2)/η(z + 1).
Exercise 10.5.7. Show that if j is the modular invariant, then j 0 (z) satisfies

az + b ab
j0 = (cz + d)2 j 0 (z) ∀ ∈ SL2 (Z)
cz + d cd
but that j 0 (z) is not a modular form of weight 2 for the full modular group.
Part II
Solutions
Chapter 1
Jacobi’s q-series
1.1 The q-exponential function
1.1.1. Show that

x x x
Eq (x) − Eq = Eq .
q q q
Observe that
∞
q n xn − xn
X
x
Eq (x) − Eq =
q n=1
q n (q n − 1)(q n−1 − 1) · · · (q − 1)
∞
( xq )n

X x x
= n−1
= E q . t
u
n=1
(q − 1) · · · (q − 1) q q
1.1.2. Prove that if |q| > 1 then

∞
Y x
Eq (x) = 1+ .
n=1
qn
From Exercise 1.1.1, we see that

x x
Eq (x) = 1 + Eq .
q q
Iterating this relation and using that Eq ( qxn ) → 1 as n → ∞ gives the result.
t
u
1.1.3. Show that if |q| < 1 then

∞ ∞ n
Y
n
X q ( 2 ) xn
(1 + q x) = .
n=0 n=0
(1 − q) · · · (1 − q n )
1
In Exercise 1.1.2, we replace q by q to obtain
∞
Y
E q1 (x) = (1 + q n x).
n=1
By the definition of Eq (x), we find

∞ ∞ n(n+1)
!
Y
n q 2 xn
X
(1 + q x) = (1 + x) ,
n=0 n=0
(1 − q) · · · (1 − q n )
n(n+1)
on noting that q 1+2+···+n = q 2 . Thus
∞ ∞ n(n+1)
Y
n
X q 2 xn (1 + x)
(1 + q x) =
n=0 n=0
(1 − q) · · · (1 − q n )
∞
X xn n(n+1) n(n−1)
n

=1+ q 2 + q 2 (1 − q )
n=1
(1 − q) · · · (1 − q n )
∞ n
X q ( 2 ) xn
=1+ ,
n=1
(1 − q) · · · (1 − q n )
as required. t
u
1.1.4. Prove that if |q| > 1 then
1
(1 + x)E q1 (x) = .
Eq (x)
Deduce that if |q| < 1 then

∞ ∞
Y X xn
(1 + q n x)−1 = .
n=0 n=0
(q n − 1) · · · (q − 1)
Using the same method as in Exercise 1.1.1, we find that

∞
xn
X
x 1
E q1 (x) − E q1 = 1− n
q n=1
1
− 1 · · · 1q − 1 q
qn
∞
X xn
= −x
1
n=0 qn − 1 · · · 1q − 1
= −xE q1 (x).
Therefore
1 x
E (x) =
1 E1 .
q (1 + x) q q
1.2 Jacobi’s Triple Product Identity 155
Iterating this gives the first result since as n → ∞, we have x/q n → 0, and
hence E1/q ( qxn ) → 1. In particular
∞ −1
Y x
E q1 (x) = 1+ n . (1.1)
n=0
q
By Exercise 1.1.2, we have that
(1 + x) E q1 (x) = Eq (x)−1 .
Replace q with 1/q in (1.1) for the final required deduction. t

u
1.2 Jacobi’s Triple Product Identity
1.2.2 (Euler’s pentagonal number theorem). Show that if |q| < 1 then
∞ ∞
Y X k(3k−1)
n
(1 − q ) = (−1)k q 2 .
n=1 k=−∞
In the Jacobi triple product identity

∞ ∞
q 2n−1

Y X 2
(1 − q 2n )(1 + q 2n−1 x) 1 + = xk q k ,
n=1
x
k=−∞
3 1
we replace q by q 2 and x by −q − 2 to get
∞ ∞
Y X (3k−1)k
(1 − q 3n )(1 − q 3n−2 )(1 − q 3n−1 ) = (−1)k q 2 .
n=1 k=−∞
Q∞
But the left hand side is simply n=1 (1 − q n ). t
u
1.2.4. Prove that

∞ ∞
k(k+1)
2n2 +n
X X
(4n + 1)q = (−1)k (2k + 1)q 2 .
n=−∞ k=0
The left hand side can be written as

∞ ∞
X 2 X 2
−n
1+ (4n + 1)q 2n +n
+ (1 − 4n)q 2n .
n=1 n=1
We compare coefficients with the right hand side. When k = 0, the sum-
2
mand is 1. When k = 2n, the summand is (4n + 1)q 2n +n . When k = 2n − 1,
2
the summand is −(4n − 1)q 2n −n . t
u
1.3 Jacobi’s two-square theorem
1.3.3. Let x4 (n) be defined as follows:


0
 if n is even
x4 (n) = +1 if n ≡ 1 (mod 4)

−1 if n ≡ 3 (mod 4)

Prove that x4 (n) is completely multiplicative. That is, show that x4 (mn) =
x4 (m)x4 (n).
If m or n is even, this is clear. If m and n are both ≡ 1 (mod 4) or both

≡ 3 (mod 4), this is also clear since their product is ≡ 1 (mod 4). If m ≡ 1
(mod 4) and n ≡ 3 (mod 4) then their product is ≡ 3 (mod 4) and the result
is clear. t
u
1.3.4. Show that X

r2 (n) = 4 x4 (d).
d|n
By Corollary 1.3.2, we must show that the sum is equal to d1 (n) − d3 (n),
where di (n) is the number of divisors of n congruent to i (mod 4). By Exer-
cise 1.3.3, the sum is precisely equal to the desired quantity. t
u
1.4 Jacobi’s four square theorem
1.4.1. Show that

1 θ 1 1
(a) cot sin nθ = + cos θ + cos 2θ + · · · + cos(n − 1)θ + cos nθ,
2 2 2 2
(b) 2(sin mθ)(sin nθ) = cos(m − n)θ − cos(m + n)θ.
Part (b) follows easily from two applications of the usual formula
cos(x + y) = cos x cos y − sin x sin y.
Part (a) follows by an induction argument. Indeed, the result is clear if

n = 1 since
θ

1 θ 1 cos θ θ θ
cot sin θ = 2
θ
2 cos sin = cos2
2 2 2 sin 2
2 2 2
and as
θ θ θ
cos θ = cos2 − sin2 = 2 cos2 − 1,
2 2 2
we see that
1 θ 1 + cos θ
cot sin θ = .
2 2 2
Now suppose the result has been proved for n. Observe that
1 θ 1 1
cot sin nθ + cos nθ + cos(n + 1)θ
2 2 2 2
cos θ2 sin nθ + cos nθ sin θ2 1
= θ
+ cos(n + 1)θ
2 sin 2 2
1

sin n + 2 θ 1
= + cos(n + 1)θ.
2 sin θ2 2
Writing (n + 12 ) = (n + 1) − 12 , we see that this is
sin(n + 1)θ cos θ2 − sin θ2 cos(n + 1)θ 1

θ
+ cos(n + 1)θ
2 sin 2
2
1 θ
= cot sin(n + 1)θ,
2 2
as desired. t
u
r
q
1.4.2. For |q| < 1, we let ur = 1−q r so that
qr
= ur (1 + ur ).
(1 − q r )2
Prove that
∞
X ∞
X
um (1 + um ) = nun .
m=1 n=1
Since
∞
1 X
= qn ,
1 − q n=0
by differentiating both sides with respect to q and then multiplying both

sides by q, we obtain
∞
q X
= nq n .
(1 − q)2 n=1
Hence,
∞ ∞ X ∞ ∞ ∞ ∞
X qm X
mn
X X
mn
X nq n
m 2
= nq = n q = ,
m=1
(1 − q ) m=1 n=1 n=1 m=1 n=1
1 − qn
as required. t
u
1.4.3. With notation as in the previous exercise, show that

∞
X ∞
X
(−1)m−1 u2m (1 + u2m ) = (2n − 1)u4n−2 .
m=1 n=1
As in the previous exercise, we have

∞ ∞ ∞
X q 2m X X
(−1)m−1 = (−1) m−1
rq 2mr
m=1
(1 − q 2m )2 m=1 r=1
∞
X ∞
X
= r (−1)m−1 q 2mr
r=1 m=1
∞
X rq 2r
=
r=1
1 + q 2r
∞
rq 2r 2rq 4r
X
= −
r=1
1 − q 2r 1 − q 4r
∞ ∞
X rq 2r X rq 2r
= −
r=1
1 − q 2r r=1
1 − q 2r
r even
∞
X (2n − 1)q 4n−2
= ,
n=1
1 − q 4n−2
as required. t
u
1.5.1. Define τ (n) by

∞
Y ∞
X
q (1 − q n )24 = τ (n)q n .
n=1 n=1
Show that τ (n) is odd if and only if n = (2m + 1)2 for some m.
We use the fact that for any prime p and for 1 ≤ j ≤ pα − 1, the binomial
coefficient α
p
≡ 0 (mod p).
j
Thus (1 − q n )8 ≡ 1 − q 8n (mod 2) and hence
∞
Y ∞
Y
q (1 − q n )24 ≡ q (1 − q 8n )3 (mod 2).
n=1 n=1
From Jacobi’s formula (Theorem 1.2.3), we deduce

∞ ∞
X X 2
n
τ (n)q ≡ q (2k + 1)q 4k +4k
(mod 2)
n=1 k=0
∞
X 2
≡ (2k + 1)q (2k+1) (mod 2),
k=0
from which the result follows immediately. t

u
1.5.2. Let rk (n) be the number of ways of writing n as a sum of k squares. Show
that
n
X
rk (n) = ri (a)rk−i (n − a)
a=0
for any i satisfying 1 ≤ i ≤ k.
This is clear from the identity
∞
!k ∞
!i ∞
!k−i
X 2 X 2 X 2
qn = qn qn
k=−∞ k=−∞ k=−∞
on comparing the coefficients of both sides of the equation. t

u
1.5.3. Define the q-logarithm function Lq (x) as

∞
X xn
Lq (x) = .
n=1
qn −1
Suppose |x| < |q| and |q| > 1. Show that

x x
Lq (x) − Lq = .
q q−x
Deduce that
∞
X x
Lq (x) = n−x
.
n=1
q
We have n
X ∞ xn − x
x q
Lq (x) − Lq = n−1
q n=1
q
∞ n
X x
=
n=1
q
x
= .
q−x
Since |q| > 1, as n → ∞ we have x/q n → 0 and thus Lq (x/q n ) → 0. There-
fore, iteration leads to
x
x q x
Lq (x) = + + Lq
q − x q − xq q2
= ···
∞
X x
= n−x
. t
u
n=1
q
1.5.4. Show that if |x| < |q| and |q| > 1 then
xEq0 (−x)
Lq (x) = ,
Eq (−x)
where the derivative is with respect to x.
From the product

∞
Y x
Eq (x) = 1+ n ,
n=1
q
we take the logarithmic derivative and use Exercise 1.5.3 to deduce the de-
sired result. t
u
Chapter 2
The Modular Group
2.1 The full modular group
2.1.1. Let R be a commutative ring with identity. Show that the set

ab a, b, c, d ∈ R
SL2 (R) = :
cd ad − bc = 1
forms a group under matrix multiplication.

It is easily checked that SL2 (R) is closed under multiplication and has an
identity element. One can verify directly that

ab d −b 10
=
cd −c a 01
so that inverses also exist. t

u
2.1.3. Show that S has order 4, ST has order 6, and T has infinite order.
This is verified directly since S 2 = −I and

0 −1
ST =
1 1

1n
so that (ST )3 = −I. Finally, T n = for any n ∈ Z so T has infinite
01
order. t
u
2.1.4. Show that SL2 (Z) is generated by two elements of finite order, namely S and
ST of order 4 and 6 respectively.
Since S has order 4 and ST has order 6, we see from Theorem 2.1.2 that
SL2 (Z) = hS, T i = hS, ST i. t

u
2.1.5. Show that any homomorphism
φ : SL2 (Z) → C×
has image contained in the finite subgroup of C× consisiting of 12th roots of unity.
By Exercise 2.1.4, SL2 (Z) is generated by S of order four and ST of order
six. The function φ must send these elements to a fourth root of unity and
a sixth root of unity, respectively.
√
Thus the image is contained in the group
1+ −3
generated by i and 2 , which is the group of 12th roots of unity. t
u
2.1.6. Show that SL2 (Z) is generated by

11 10
T = and U = .
01 11
We need only observe that

1 −1 10 1 −1
T −1 U T −1 = =S
0 1 11 0 1
to deduce the result. t

u
2.1.7. Suppose that (c, d, N ) = 1. Show that there are elements c0 = c + tN and
d0 = d + sN for some integers s, t such that (c0 , d0 ) = 1.
Write c = c1 c2 with c2 = (c, N ). Thus (c1 , N ) = 1, and so by the Extended

Euclidean Algorithm we can find u, v ∈ Z such that 1 = c1 u + N v. Hence
1 ≡ N v (mod c1 ). Set m = v(1 − d) so that
d + mN ≡ 1 (mod c1 ). (2.1)
Take c0 = c and d0 = d + mN. We claim (c0 , d0 ) = 1. Indeed, if p is a prime

such that p|c0 and p|d0 then p|c and p|d + mN. Since c = c1 c2 , we have p|c1
or p|c2 . If p|c1 , we get a contradiction to (2.1). Thus p|c2 which implies that
p|N . From p|d + mN , we deduce p|d and so p|(c, d, N ), a contradiction. t
u
2.2 Subgroups of the modular group
2.2.1. Show that the natural map
SL2 (Z) → SL2 (Z/N Z)

g 7→ g (mod N )
The map is obviously a homomorphism. Let

ab
∈ SL2 (Z/N Z).
cd
We must lift this to an element of SL2 (Z). By Exercise 2.1.7, since (c, d, N ) =
1, we can find c0 , d0 such that
c0 ≡ c (mod N )
d0 ≡ d (mod N )
and (c0 , d0 ) = 1. We need to find k, l ∈ Z so that

a + kN b + lN
∈ SL2 (Z).
c0 d0
We know that ad0 −bc0 = 1+jN for some j so the problem reduces to solving
jN + kd0 N − lc0 N = 0,
or equivalently,
j = lc0 − kd0 .
Since (c0 , d0 ) = 1, the extended Euclidean Algorithm produces k, l ∈ Z satis-
fying this requirement. t
u
2.2.2. Prove that Γ(N ) is a normal subgroup of SL2 (Z) of finite index.
By Exercise 2.2.1 we have Γ(N ) is the kernel of the natural map
SL2 (Z) → SL2 (Z/N Z)
and consequently normal. The image of this map is a finite set and the result
follows from the first isomorphism theorem. t
u
2.2.3. For any commutative ring R with identity, let GL2 (R) be the set of 2 × 2
invertible matrices with entries in R. Show that GL2 (R) is a group.
A matrix A is invertible if and only if det(A) ∈ R× and the result is im-

mediate from the equality det(AB) = det(A) det(B). t
u
2.2.4. Let p be a prime. Show that the order of GL2 (Z/pZ) is given by (p2 − 1)(p2 − p).
Each element of GL2 (Z/pZ) gives rise to a basis of (Z/pZ)2 . Conversely,

given a basis, we can construct an element of GL2 (Z/pZ) by taking the basis
elements as columns of the matrix. Thus, to count the size of GL2 (Z/pZ), we
need only count the number of bases of (Z/pZ)2 . For the first column, we
can choose any non-zero vector, so we have p2 − 1 choices. For the second
column, we can choose any vector which is not a scalar multiple of the first
column. Thus, there are p2 − p choices. Hence, the group GL2 (Z/pZ) has
(p2 − 1)(p2 − p) elements. t
u
2.2.5. Let p be a prime. Show that SL2 (Z/pZ) is a normal subgroup of GL2 (Z/pZ)
of index p − 1. Deduce that SL2 (Z/pZ) has order p(p2 − 1).
The map
GL2 (Z/pZ) → (Z/pZ)×

A 7→ det A
is a homomorphism whose kernel is SL2 (Z/pZ). Consequently, SL2 (Z/pZ)

is a normal subgroup. Moreover, this homomorphism is surjective since

a0
7→ a ∈ (Z/pZ)× ,
01
so the image has size (p − 1). Thus SL2 (Z/pZ) has index p − 1 in GL2 (Z/pZ).
The cardinality is now deduced from Exercise 2.2.4. t
u
2.2.6. Show that SL2 (Z/p2 Z) has size p4 (p2 − 1).
The natural map given by reduction modulo p
SL2 (Z/p2 Z) → SL2 (Z/pZ) (2.2)
can be seen to be surjective by first lifting a matrix in SL2 (Z/pZ) to SL2 (Z)
by Exercise 2.2.1 and then reducing modulo p2 . The kernel of the map in
(2.2) consists of matrices of the form

1 + t1 p t 2 p
t3 p 1 + t4 p
with 0 ≤ ti ≤ p − 1 and determinant 1 modulo p2 . This last condition is

equivalent to t1 + t4 ≡ 0 (mod p). Hence, the kernel contains p3 elements.
From Exercise 2.2.5, we see that SL2 (Z/p2 Z) has p4 (p2 − 1) elements. t
u
2.2.7. Apply an induction argument to show that SL2 (Z/pn Z) has size p3n−2 (p2 − 1).
The result is true for n = 1 and n = 2 by the previous exercises. Now

consider the natural map
SL2 (Z/pn+1 Z) → SL2 (Z/pn Z),
which is seen to be surjective by first lifting a matrix in SL2 (Z/pn Z) to SL2 (Z)
by Exercise 2.2.1 and then reducing modulo pn+1 . The kernel consists of
matrices
1 + t 1 pn t 2 p n

t 3 pn 1 + t 4 pn
with 0 ≤ ti ≤ p − 1 and determinant 1 modulo pn+1 . This last condition is

equivalent to t1 + t4 ≡ 0 (mod p). So again, there are only p3 possibilities for
the ti s. Hence
SL2 (Z/pn+1 Z) = p3 |SL2 (Z/pn Z)|
and the result follows by induction. t
u
2.2.8. Show that

Y 1

3
| SL2 (Z/N Z)| = N 1− 2 .
p
p|N
Deduce that
Y 1

3
[SL2 (Z) : Γ(N )] = N 1− 2 .
p
p|N
Suppose N = pn1 1 · · · pnt t where the pi are unique primes. By the Chinese
Remainder Theorem,
SL2 (Z/N Z) = SL2 (Z/pn1 1 Z) × · · · × SL2 (Z/pnt t Z).

Exercise 2.2.7 now gives | SL2 (Z/N Z)| = N 3 p|N 1 − p12 . As in the solu-
Q
tion to Exercise 2.2.2, since Γ(N ) is the kernel of the natural map SL2 (Z) →
SL2 (Z/N Z), we deduce that [SL2 (Z) : Γ(N )] = | SL2 (Z/N Z)|. t
u
2.2.9. Show that SL2 (Z/N Z) is generated by

11 10
T = and U = ,
01 11
both of order N .
It is clear that T and U have order N since

1n 10
Tn = and U n = .
01 n1
The natural map

SL2 (Z) → SL2 (Z/N Z)
is surjective and since SL2 (Z) is generated by T and U (by Exercise 2.1.6),
we are done. t
u
2.2.10. Show that Γ(2) is generated by −I, T 2 , and U 2 , where

2 12 2 10
T = and U = .
01 21
Clearly −I, T 2 , and U 2 are in Γ(2) so that the group generated by these
three elements is contained in Γ(2). We adapt the proof of Theorem 2.1.2 and
use the following modified division algorithm: given a and b, with b 6= 0,
there are q and r such that a = bq + r with |r| ≤ |b|/2. Observe that r can be
negative. Now choose any

ab
g= ∈ Γ(2).
cd

1m
If c = 0, then as before, g = ± for some m ∈ Z. Since g ∈ Γ(2), m is
0 1
even and hence of the form m = 2k. We can thus write g = ±T 2k ∈ h−I, T 2 i.
Now suppose c 6= 0. If |a| > |c|, with c 6= 0, we write a = (2c)q + r with
|r| ≤ |c| and note that

1 −2q ab r b − 2qd
T −2q g = = .
0 1 cd c d
As c is even, we must have r and c of opposite parity and so |r| < |c|. If
|a| < |c|, then as a is odd, it is not zero and we can write c = (2a)q + r with
|r| ≤ |a|. Thus

1 0 ab a b
U −2q g = = .
−2q 1 cd r d − 2qb
As before, we must have |r| < |a|, as a and r are of opposite parity. Using
T 2 and U 2 in this way, we can rearrange the lower left entry to be zero and
thus we are done. t
u
2.2.11. In SL2 (Z), show that the group Γ generated by S and T 2 is a congruence
subgroup of index three.
We first show that Γ(2) ⊂ Γ and hence that Γ is a congruence subgroup.

By Exercise 2.2.10, Γ(2) is generated by −I, T 2 , and U 2 , and so it suffices to
show that −I and U 2 lie in Γ. Since −I = S 2 and U 2 = ST −2 S −1 , this is
clear. To compute the index, we observe that
SL2 (Z) /Γ(2) ∼

= SL2 (Z/2Z)
and use the correspondence theorem and work in SL2 (Z/2Z). The image of
Γ/Γ(2) in SL2 (Z/2Z) is the group generated by S (mod 2), which has order
2. Thus [Γ : Γ(2)] = 2. As Γ(2) has index 6 in SL2 (Z), it follows that Γ has
index 3 in SL2 (Z). t
u
2.3 The Hecke subgroups 167
2.3 The Hecke subgroups
2.3.1. Show that the map Γ0 (N ) → (Z/N Z)× given by

ab
7→ d (mod N )
cd
The fact that the map is a homomorphism is easily checked. The surjec-
tivity follows from the fact (N, d) = 1 implies there are integers a, b such that
ad − bN = 1. Consequently, the matrix

a b
∈ Γ0 (N )
N d
and has image d under this map. t

u
2.3.2. Show that the map Γ1 (N ) → Z/N Z given by

ab
7→ b (mod N )
cd
is a surjective homomorphism with kernel Γ(N ).
It is easily checked that this is a homomorphism. The matrix

11
T =
01

b 1b
lies in Γ1 (N ). We see that T = ∈ Γ1 (N ) for any integer b, so the
01
surjectivity is immediate. The fact that Γ(N ) is the kernel is immediate. t u

2.3.3. Show that Γ1 (N ) has index N 2 p|N 1 − p12 in SL2 (Z).
Q
By the previous exercise,
Γ1 (N )/Γ(N ) ∼
= Z/N Z
so that from
[SL2 (Z) : Γ1 (N )][Γ1 (N ) : Γ(N )] = [SL2 (Z) : Γ(N )]
we deduce
[SL2 (Z) : Γ(N )]
[SL2 (Z) : Γ1 (N )] = .
N
The result now follows from Exercise 2.2.8. t
u

1
Q
2.3.4. Show that Γ0 (N ) has index N p|N 1+ p in SL2 (Z).
From Exercise 2.3.1, we have
Γ0 (N )/Γ1 (N ) ∼
= (Z/N Z)×
and so
[SL2 (Z) : Γ1 (N )]
[SL2 (Z) : Γ0 (N )] = ,
φ(N )
where φ denotes Euler’s function. Since by Exercise 2.3.3 the numerator is
Y 1

N2 1− 2
p
p|N
and
Y 1

φ(N ) = N 1− ,
p
p|N
the result now follows. t

u
2.4 Groups acting on topological spaces
2.4.1. If G acts on X and g ∈ G, x ∈ X, show that
Ggx = gGx g −1 .
For any h ∈ G, we have
h ∈ Ggx ⇐⇒ h(gx) = gx ⇐⇒ (g −1 hg)x = x

⇐⇒ g −1 hg ∈ Gx ⇐⇒ h ∈ gGx g −1 . t
u
2.4.2. If G acts transitively on X, show that all stabilizers are conjugate.

For any x, y ∈ X, by transitivity there exists g ∈ G with y = gx. Hence by
Exercise 2.4.1, we have Gy = Ggx = gGx g −1 . t
u
2.4.3. (Orbit-Stabilizer Formula) Suppose that G acts on X.
(a) If x ∈ X and g, h ∈ G, show that gx = hx if and only if g and h lie in the same
left coset of Gx .
(b) If G and X are both finite, show that |Gx| = [G : Gx ]. Deduce that
X
|X| = [G : Gxi ],
i
where the summation is over a complete set of representatives of the orbits of G.

(a) We have
gx = hx ⇐⇒ (h−1 g)x = x ⇐⇒ h−1 g ∈ Gx ⇐⇒ g ∈ hGx ,
so that g and h lie in the same left coset of Gx .

(b) For a fixed x, the listing
gx : g ∈ G
contains |G| elements. From (a), we see that any element that appears in
the list is repeated |Gx | times. Thus |Gx| = [G : Gx ]. Since X is a disjoint
union of its orbits, we deduce that
X
|X| = [G : Gxi ]
i
where the sum is over a complete set of representatives of the orbits of

G. t
u
2.4.6. Let G be a group with subgroups H and K. Suppose that X is a subset of G

which is invariant under the left action of H and the right action of K. Show that
X is a disjoint union of double cosets HgK.
Clearly, for any x ∈ X we have HxK ⊆ X and in fact

[
X= HxK.
x∈X
But double cosets HxK and HyK are either disjoint or equal. To see this,
suppose they share a common element h1 xk1 = h2 yk2 . Then x = h−1 −1
1 h2 yk2 k1 .
For any h3 xk3 ∈ HxK, we have
h3 xk3 = h3 h−1 −1

1 h2 y k2 k1 k3 ∈ HyK
and we deduce that HxK ⊆ HyK. The reverse inclusion is analogous. t

u
2.4.7. Suppose a group H acts on a set Y and |H\Y | < ∞. If X ⊆ Y is invariant

under the action of H and if H1 is a finite-index subgroup of H, then prove that
|H1 \X| < ∞.
Write n = |H\Y | and m = [H : H1 ]. There exist y1 , . . . , yn ∈ Y and

h1 , . . . hm ∈ H such that
an
Y = Hyi
i=1
and
m
a
H= H 1 hi .
j=1
Every x ∈ X ⊆ Y must be of the form x = hyi for some h ∈ H and some

1 ≤ i ≤ n. Now Hx = Hyi . Thus H\X is a subset of H\Y , and hence it must
be finite. There is in index set I ⊆ {1, 2, . . . , n} such that
a m
aa
X= Hyi = H 1 hi y i
i∈I i∈I j=1
and so |H1 \X| < ∞. t

u
2.5.1. If Γ is a congruence subgroup of level N and γ ∈ SL2 (Z), show that γΓγ −1
is again a congruence subgroup of level N .
By Exercise 2.2.2, we know that Γ(N ) is a normal subgroup of SL2 (Z).

Hence Γ(N ) = γΓ(N )γ −1 ⊆ γΓγ −1 .
2.5.2. Prove that

Γ(M ) ∩ Γ(N ) = Γ(lcm(M, N )).
We know that
g ∈ Γ(M ) ∩ Γ(N ) ⇐⇒ g ∈ Γ(M ) and g ∈ Γ(N )

⇐⇒ g ≡ I (mod M ) and g ≡ I (mod N )
⇐⇒ g ≡ I (mod lcm(M, N ))
⇐⇒ g ∈ Γ(lcm(M, N ))
where the penultimate equivalence follows from the Chinese Remainder

Theorem and the definition of an integer congruence. t
u
2.5.3. Prove that

Γ(M )Γ(N ) = Γ(gcd(M, N )).
If g ∈ Γ(M )Γ(N ), then g = ab with a ∈ Γ(M ) and b ∈ Γ(N ). In particular,

a ≡ I (mod M ) and b ≡ I (mod N ) and thus g = ab ≡ I (mod gcd(M, N )).
Consequently,
Γ(M )Γ(N ) ⊂ Γ(gcd(M, N )).
Thus, it suffices to show that [Γ(1) : Γ(M )Γ(N )] = [Γ(1) : Γ(gcd(M, N ))].
By the Second Isomorphism Theorem in group theory, we have
Γ(M )Γ(N )/Γ(N ) ∼

= Γ(M )/Γ(M ) ∩ Γ(N )
and hence [Γ(M )Γ(N ) : Γ(N )] = [Γ(M ) : Γ(M ) ∩ Γ(N )]. Thus, by the Tower
Rule for the indices of groups, we deduce that
[Γ(1) : Γ(N )]
[Γ(1) : Γ(M )Γ(N )] =
[Γ(M )Γ(N ) : Γ(N )]
[Γ(1) : Γ(N )]
=
[Γ(M ) : Γ(M ) ∩ Γ(N )]
[Γ(1) : Γ(N )][Γ(1) : Γ(M )]
= .
[Γ(1) : Γ(M ) ∩ Γ(N )]
By Exercise 2.5.2, Γ(M ) ∩ Γ(N ) = Γ(lcm(M, N )). Thus, by Exercise 2.2.8,

3
Y 1
[Γ(1) : Γ(M )Γ(N )] = gcd(M, N ) 1− 2
p
p| gcd(M,N )
= [Γ(1) : Γ(gcd(M, N ))],
as desired. t
u
2.5.4. Define

10
Γ1 (N ) = g ∈ SL2 (Z) : g ≡ (mod N ) .
?1
Show that Γ1 (N ) is a congruence subgroup of level N .
This is clear. t
u
2.5.5. If SL2 (Z) has a non-congruence subgroup, show that it has infinitely many.
(Hint: consider Γ ∩ Γ (p) for a non-congruence subgroup Γ and rational
primes p.)
It is clear that Γ ∩ Γ (p) is a non-congruence subgroup if Γ is a non-

congruence subgroup since Γ ⊇ Γ ∩ Γ (p). Thus, we only need to show
that as p varies, we produce infinitely many distinct groups. If Γ ∩ Γ (p) = H
(say), for infinitely many rational primes p, then the index
[SL2 (Z) : Γ (p)] = p(p2 − 1)
divides [SL2 (Z) : H] for infinitely many primes p, which is a contradiction.

t
u
2.5.6. For each g ∈ SL2 (Z) and a subgroup Γ of finite index in SL2 (Z), we define
ng (Γ ) to be the least positive integer such that

ng (Γ ) −1 11
T ∈ gΓ g , T = as usual.
01
We say Γ is of class N if the least common multiple of the set of numbers {ng (Γ ) :
g ∈ SL2 (Z)} equals N . Show that Γ (N ) is of class N . Also show that the normal
closure of T N , namely
D(N ) := hgT N g −1 : g ∈ SL2 (Z)i
is of class N . (This exercise extends the notion of congruence level defined for
congruence subgroups to all subgroups of finite index, and is due to Fricke
and Wohlfahrt.)
Clearly, nI (Γ (N )) = N and as Γ (N ) is a normal subgroup of SL2 (Z), we
see that Γ (N ) is of class N . Moreover, as D(N ) ⊆ Γ (N ), and T N ∈ D(N ), it
follows that D(N ) is of class N . t
u
2.5.7.
(a) Using the Chinese remainder theorem (or otherwise), show that for given x, y,
the following congruence
x(αy + β) ≡ γ mod m
has a solution in integers α, β, γ such that (α, β, m) = 1.

(b) Using (a), show that given g ∈ Γ (N ), there exists
0 0
0 a b
g = ∈ Γ (N )
c 0 d0
such that b0 ≡ c0 ≡ 0 mod mN and g 0 ∈ D(N )gD(N ) with D(N ) defined

in Exercise 2.5.6. (Hint: consider conjugates of T by the transpose of T
that is in D(N ).)
(a) If p is a prime dividing m, we may take yp = 0 if p|α and choose yp so
that αyp + β ≡ 1 mod p when p is coprime to α. In either case, αyp + β
and p are relatively prime and so xp can be chosen to satisfy the given
congruence mod pe where pe ||m. The proof is now completed using the
Chinese remainder theorem.
(b) Let
ab
g=
cd
be given and let L be the transpose of T . Consider the elements of the
form
−y nx y a 1 b1
g1 = L T L g = (say)
c 1 d1
where b1 = b + nx(by + d). We choose x and y so that
b1 = b + nx(by + d) ≡ 0 mod mN
and as b is divisible by N , this is equivalent to solving b/N +x(by+d) ≡ 0

mod m which by (a) has a solution. Now consider the elements of the
form
nz a 2 b2
g2 = g1 L = (say)
c 2 d2
where c2 = c1 + d1 N z. We observe that we may find z such that
c2 = c1 + dN z ≡ 0 mod mN
which is equivalent to c1 /N + d1 z ≡ 0 mod mN because a1 d1 = 1 +

b1 c1 ≡ 1 mod mN or in other words, a1 d1 ≡ 1 mod m. Thus, g2 is the
required modification.
t
u
2.5.8. With notation as in Exercise 2.5.6, show that D(N )Γ (M N ) = Γ (N ) for all
positive integers M, N .
Clearly, it suffices to prove the inclusion ⊇ for M > 1. In view (b) of

Exercise 2.5.7, it suffices to show that an element of the form

ab
g= ∈ Γ (N )
cd
with b ≡ c ≡ 0 mod M N belongs to ∆(N )Γ (M N ). Indeed, consider the

matrix
0 a ad − 1
g = .
−ad + 1 −d(ad − 2)
The g ≡ g 0 mod M N so that there exists g 00 ∈ Γ (M N ) such that g = g 0 g 00 . A
simple calculation shows that (with L the transpose of T )
g 0 = L−d+1 T −1 L1−a T d
whence g 0 ∈ D(N ). This completes the proof. t

u
2.5.9. Let Γ be a subgroup in SL2 (Z) of class N . Show that Γ is a congruence

subgroup if and only if it is a congruence subgroup of level N . Deduce that if Γ
is a subgroup of SL2 (Z) of class N and the index [SL2 (Z) : Γ ] does not divide
[SL2 (Z) : Γ (N )], then Γ is not a congruence subgroup.
One direction is a tautology. For the other direction, it suffices to observe

that Γ (M ) ⊇ Γ (`M ) for any integer `. Thus, if Γ is a congruence subgroup,
then Γ ⊇ Γ (N ) for some N . Also, Γ ⊇ ∆(N ) since Γ is of class N . Therefore,
we have
Γ ⊇ Γ (M N )∆(N ) = Γ (N ).
Now, N is the least such is clear from the definition of the level. The conclu-
sion is now evident. t
u
2.5.10. Let F be a free group on n generators, say x1 , ..., xn . For a word w ∈ F ,

define ì (w) to be the sum of all the exponents of xi occuring in w. For a subset I of
{1, ..., n}, and a positive integer `, define
Γ` (I) = {w ∈ F : ì (w) ≡ 0 mod ` ∀i ∈ I}.

Show that Γ` (I) is a normal subgroup of F of index `|I| .
Normality is evident since Γ` (I) is the kernel of the surjective homomor-

phism F → (Z/`Z)|I| sending w to (ì (w))i∈I . It is easily seen that
Y
{ xei i : 0 ≤ ei < `}
i∈I
is a set of (left) coset representatives of Γ` (I) in F , from which the result is

evident. t
u
2.5.11. Recall that Γ (2) is generated by −I, T 2 , U 2 (see Exercise 2.2.10) and that
the subgroup F generated by x1 = T 2 and x2 = U 2 is free of index 2 in Γ (2).
(a) With notation as in the previous exercise, compute the class of the group
Γ` ({1}) and Γ` ({1, 2}).
(b) Show that Γ` ({1, 2}) is not a congruence subgroup if ` is not a power of 2.
(a) Let H denote any of the two groups Γ` ({1}) or Γ` ({1, 2}). Then T k ∈ H
if and only if T k ≡ I mod 2 and `1 (T k ) ≡ 0 mod `. This happens if and
only if k ≡ 0 mod 2 and k ≡ 0 mod `. That is, if and only if 2`|k if ` is
odd and or k|` if ` is even. Thus, H is of class 2` if ` is odd and ` if ` is
even.
(b) We have shown the following facts:
[SL2 (Z) : Γ` ({1, 2})] = [SL2 (Z) : Γ (2)][Γ (2) : F ][F : Γ` ({1, 2})] = 12`2 .
Since Γp ({1, 2}) ⊇ Γ` ({1, 2}) for any prime p dividing `, it follows that
Γp ({1, 2}) is a congruence subgroup if Γ` ({1, 2}) is a congruence sub-
group. Also, [SL2 (Z) : Γ (2p)] = 6p(p2 − 1) if p > 2 and 48 if p = 2. Since
` is not a power of 2, let p be an odd prime that divides `. If Γ` ({1, 2}) is
a congruence subgroup, then so is Γp ({1, 2}) by our remark. However,
since Γp ({1, 2}) is of level 2p, it follows that if Γp ({1, 2}) is a congruence
subgroup, then Γp ({1, 2}) ⊇ Γ (2p) so that 12p2 |6p(p2 − 1), a contradic-
tion. Thus, Γ` ({1, 2}) is not a congruence subgroup.
. t
u
Chapter 3
The Upper Half-Plane
3.1 The group GL+

n (R) and its subgroups
3.1.1. Show that

ab az + b
z=
cd cz + d
defines an action of GL+
2 (R) on H.
Let g = ac db ∈ GL+

2 (R). Then it is easily checked that
(ad − bc) Im(z)

Im(gz) = > 0.
|cz + d|2
In addition, the identity I acts trivially (as well as −I). The other axiom is
easily verified directly. t
u
3.1.2. Show that the group PSL2 (R) = SL2 (R)/ ± I acts faithfully on H. That is,
gz = z for all z ∈ H implies g = 1 ∈ PSL2 (R).

Let g = ac db and suppose that for all z ∈ H, we have
az + b
= z.
cz + d
Then cz 2 + (d − a)z − b = 0 for all z ∈ H which implies c = b = 0 and d = a
since a quadratic polynomial has at most two roots. Since the determinant
a2 = 1, we have a = ±1 so that g = ±I. t
u
3.1.3. Show that the action of PSL2 (R) on H is transitive, i.e. there is only one orbit
and for all z, w ∈ H, there exists some g ∈ PSL2 (R) such that gz = w.
It suffices to show that for any z ∈ H, there is a g ∈ PSL2 (R) such that
gz = i. Then given w ∈ H, there is also an element g1 ∈ PSL2 (R) such that
g1 (w) = i. Thus gz = g1 w and (g1−1 g)z = w.
Now, as z = x + iy ∈ H, we have y > 0 and it is immediate that

√ √ √
1x y 0 y x/ y
√ i= √ i = x + iy. t
u
01 0 1/ y 0 1/ y
3.1.4. Show that the map

z−i
z 7→
z+i
is an analytic isomorphism of the upper half-plane H and the unit disc D = {w ∈
C : |w| < 1}.
Clearly the map is analytic since −i ∈ / H. Moreover, |z − i| < |z + i|

for z ∈ H so that the image lies in D. Injectivity is easily checked. To prove
surjectivity, we must show that for any w ∈ D, there is a z such that z−i
z+i = w.
Solving for z, we obtain
i(1 + w)
z= .
1−w
We must ensure Im(z) > 0. But this is clear since
1 − |w|2

i(1 + w)(1 − w)
Im(z) = Im = > 0. t
u
|1 − w|2 |1 − w|2
3.1.6. Show that SL2 (Z) is a discrete subgroup of SL2 (R) ⊂ R4 , where R4 is en-
dowed with the Euclidean topology.
In R4 , convergence is component-wise convergence. Consequently, if

γn → I where γn ∈ SL2 (Z), then for n sufficiently large, γn = I. t
u
3.2 Fundamental domains
3.2.4. Suppose that z, w ∈ F are distinct points which are SL2 (Z)-equivalent.
Show that Im(z) = Im(w) and Re(z) = − Re(w).
Without loss of generality,

we may suppose Im(z) ≥ Im(w)√so that if
γz = w and γ = ac db ∈ SL2 (Z), then |cz + d|2 ≤ 1. Thus |c| 3/2 ≤ 1
implies c ∈ {0, ±1}.
If c = 0, then
±1 b
γ= .
0 ±1
But then Re(w) = Re(z) ± b so that |b| = 1. Hence Re(w) = ±1/2 so that
Re(z) = − Re(w). Also, Im(w) = Im(z ± b) = Im(z).
If |c| = 1, then |cz+d|2 ≤ 1 becomes 2
√ |z±d| ≤ 1. Thus, assuming z = x+iy,
2 2
we have (x ± d) + y ≤ 1. As y ≥ 3/2, this means |x ± d| ≤ 1/2, forcing
3.3 The extended upper-half plane 177
|d| ≤ 1. If |d| = 1, we must have z = ρ or ρ2 (see Figure 3.1 on page 28). If

d = 0, our condition becomes |z| = 1. Now
Im(z)
Im(w) = = Im(z)
|cz + d|2
and
a ∓1 1
w= z = ±a − .
±1 0 z
As Sz = −1/z maps the arc of the unit circle with π/3 ≤ arg z ≤ π/2 to the
arc with π/2 ≤ arg z ≤ 2π/3, we see that either w = −1/z or that {z, w} =
{ρ, ρ2 }. Hence Re(w) = − Re(z), as required. t
u
3.2.5. If F is the standard fundamental domain for SL2 (Z) and γ ∈ SL2 (Z), show
that γF is again a fundamental domain for SL2 (Z).
The map z 7→ γz is an analytic automorphism of H. Consequently, the

topology of F is preserved under the mapping. Thus γF is again a closed set
with connected interior and its boundary is again a finite union of smooth
curves. Since every point z ∈ H is SL2 (Z)-equivalent to some w ∈ F , it is
SL2 (Z)-equivalent to γw ∈ γF . If two interior points z, w of γF are SL2 (Z)-
equivalent, then γ −1 z and γ −1 w would be interior points of F , which are
SL2 (Z)-equivalent, a contradiction.
3.2.7. Show that I, S, T −1 S are coset representatives for Γ0 (2) in SL2 (Z). Deduce
that a fundamental domain for Γ0 (2) is given by F ∪ SF ∪ ST F , where F is the
standard fundamental domain for SL2 (Z).
The group Γ0 (2) has index three in SL2 (Z) and it is easily verified that
I, S, T −1 S are coset representatives for Γ0 (2). To see that F ∪SF ∪ST F has a
connected interior, we need only sketch the region explicitly as in Figure 3.1
on page 178. t
u
3.3 The extended upper-half plane
3.3.1. Show that every cusp is SL2 (Z)-equivalent to i∞.

Given a rational number s 6= i∞, we need to find a matrix ac db ∈ SL2 (Z)
such that cs + d = 0. Writing s = −d/c with (c, d) = 1, we can find by
the Euclidean algorithm, integers a, b, such that ad − bc = 1. This gives the
desired matrix. t
u
3.3.2. If Γ is of finite index in SL2 (Z), show that the number of Γ-inequivalent
cusps is ≤ [SL2 (Z) : Γ].
i
ρ
ρ2
SF
ST F
−1 − 12 0 1 1
2
Fig. 3.1 Fundamental Domain for Γ0 (2)
Let s ∈ Q ∪ {i∞} be a cusp. By Exercise 3.3.1, there exists τ ∈ SL2 (Z) such
that s = τ (i∞). If g1 , . . . , gr are right coset representatives of Γ in SL2 (Z),
then s = τ (i∞) = γgj (i∞) for some γ ∈ Γ. Hence the set of Γ-inequivalent
cusps is contained in the set
{g1 (i∞), . . . , gr (i∞)}
whose size is at most [SL2 (Z) : Γ]. t

u
3.3.3. Show that {0, i∞} is a complete set of inequivalent cusps of Γ0 (2).

For any g = ac db ∈ SL2 (Z), since g ∈ Γ0 (2) if and only if c ≡ 0 (mod 2),
we see that i∞ is equivalent to a rational number a/c if and only if c is even
and a is odd. Similarly, 0 is equivalent to all b/d with d odd. If 0 and i∞ are
equivalent, there is an element g ∈ Γ0 (2) such that g(i∞) = 0. This means
a = 0 and the matrix g would then be singular modulo 2, a contradiction.
tu
3.3.4. With the action of SL2 (Z) on H? , show that the stabilizer Γi∞ of i∞ is given
by
1n
± :n∈Z .
01
It is clear that if
a(1) + b(0) a 1
= =
c(1) + d(0) c 0
then we must have c = 0 so that ad = 1 and hence a = d = 1 or a = d = −1.
t
u
3.3 The extended upper-half plane 179
3.3.5. Let p be prime. Show that the matrices

k 0 −1
ST = , 0 ≤ k ≤ p − 1,
1 k
along with ( 10 01 ), give a complete set of right coset representatives for Γ0 (p) in
SL2 (Z).
It is easily checked that these matrices are distinct coset representatives.

Indeed, 01 −1

k ∈
6 Γ 0 (p) and if

0 −1 0 −1
Γ0 (p) = Γ0 (p) ,
1 k 1 k0
then −1
k0 1

0 −1 0 −1 0 −1
= ∈ Γ0 (p)
1 k 1 k0 1 k −1 0
which means k ≡ k0 (mod p).
Now given ac db ∈ SL2 (Z), if ac db 6∈ Γ0 (p) then we want to show that
there exists some k such that

ab 0 −1
∈ Γ0 (p) .
cd 1 k
This is equivalent to ensuring

−1
ab 0 −1 ab k 1 ak − b a
= = ∈ Γ0 (p),
cd 1 k cd −1 0 ck − d c
so we need only find k so that ck ≡ d (mod p). This is always possible when
p - c. t
u
3.3.6. Let p be prime. Show that the set of matrices

k 1
, 0 ≤ k ≤ p − 1,
−1 0
along with ( 10 01 ), is a complete set of left coset representatives for Γ0 (p).
If g1 , . . . , gr are right coset representatives of a subgroup H in G, then

since
Hgi 6= Hgj ⇐⇒ gi gj−1 6∈ H ⇐⇒ gi−1 H 6= gj−1 H,
we see that g1−1 , . . . , gr−1 are left coset representatives and the result is im-
mediate from Exercise 3.3.5. t
u
3.3.7. Let p be prime. Show that a complete set of inequivalent cusps for Γ0 (p) is
given by {0, i∞}.
For 0 ≤ k ≤ p−1, we have that 01 −1

k (i∞) = 0. Hence, by the solution to
Exercise 3.3.2 and by an application of Exercise 3.3.5, the Γ0 (p)-inequivalent
cusps are contained in the set containing i∞ and 0. We need only show that
0 and i∞ are Γ0 (p)-inequivalent. Indeed, if there is a matrix,

ab
∈ Γ0 (p)
cd
such that
ab
(i∞) = 0,
cd
6 0 and i∞ if c = 0. This means that our
then the left hand side is a/c if c =
matrix is of the form
0b
,
cd
which has determinant −bc and is consequently divisible by p, a contradic-
tion. t
u
3.4 The Poincaré metric
3.4.1. Prove that d(z, w) is SL2 (R)-invariant. That is, show that d(gz, gw) =
d(z, w) for all g ∈ SL2 (R).
It suffices to show that
|z − w|2
Im(z) Im(w)

is SL2 (R)-invariant. Indeed, let g = ac db ∈ SL2 (R). Then by the solution to
Exercise 3.1.1,
Im(z) Im(w)
Im(gz) Im(gw) =
|cz + d|2 |cw + d|2
and
az + b aw + b z−w
− = ,
cz + d cw + d (cz + d)(cw + d)
since ad − bc = 1. The result is now immediate. t
u
3.4.2. Show that for x ≥ 1,
p
cosh−1 x = log(x + x2 − 1).
We want y such that

ey + e−y
= x.
2
This is a quadratic equation in ey which is easily solved and we get
3.4 The Poincaré metric 181
p
ey = x + x2 − 1
since the inverse hyperbolic cosine function is non-negative (and hence we

need ey ≥ 1). t
u
3.4.3. Show that for y > 0,

d(i, iy) = | log y|.
We have
(y − 1)2

−1
d(i, iy) = cosh 1+
2y

−1 1 1
= cosh y+
2 y
s 2 !
1 1 y 1 1
= log y+ + + 2−
2 y 4 4y 2
 s 
2
1 1 y 1
= log  y+ + − .
2 y 2 2y
Thus
1 1 1 1
d(i, iy) = log y+ + y− .
2 y 2 y
If y > y1 , this is log y. If y < y1 , this is − log y. In either case, it can be written
as | log y|. t
u
3.4.5. Let F be the standard fundamental domain for SL2 (Z). Show that
Z Z
dxdy π
2
= .
F y 3
We compute easily Z Z
dxdy
= 1.
F :y≥1 y2
The integral
1
Z Z Z Z 1
dxdy 2 dy
=2 √ dx
F :y≤1 y2 0 1−x2 y2
Z 1
2 1
=2 −1 + √ dx
0 1 − x2
1
= −1 + 2 sin−1
2
π
= −1 + ,
3
from which the result follows. t

u
3.5.1. Show that a set of coset representatives for Γ∞ in SL2 (Z) is given by

??
: (c, d) = 1, c, d ∈ Z ,
cd
where the asterisks indicate that we can take any a, b such that ad − bc = 1.
The coset

ab a + nc b + nd
Γ∞ = :n∈Z .
cd c d
Thus, for fixed c, d with (c, d) = 1, the coset parametrizes all integer solu-
tions of ad − bc = 1, by the Euclidean algorithm. t
u
3.5.2. Let F , Γ, g1 , . . . , gr and D be as in Theorem 3.2.6. Show that

Z Z
dxdy π
2
= [SL2 (Z) : Γ] .
D y 3
Using the notation of Theorem 3.2.6, we have coset representatives g1 , . . . , gr

of Γ in SL2 (Z) such that
[r
D= gi−1 F
i=1
is a fundamental domain for Γ. Thus

Z Z r Z Z
dxdy X dxdy
2
= .
D y i=1 gi−1 F y2
Since the measure dxdy/y 2 is SL2 (Z)-invariant, each integral is equal to π/3,
the area of F , by Exercise 3.4.5. t
u
3.5.3. Let G = SL2 (R) act on H. Show that K, the stabilizer of i, consists of matri-
ces
cos θ sin θ
for θ ∈ R.
− sin θ cos θ
Then show that the upper half-plane is homeomorphic to G/K.

If the matrix ac db ∈ SL2 (R) fixes i, then we must have
ai + b = i(ci + d) = −c + di.
In other words, b = −c and a = d. Since ad − bc = 1, this means a2 +

b2 = 1. We can write a = cos θ and b = sin θ to deduce the result on K.
Since SL2 (R) acts transitively on H, we deduce that the upper half-plane is
homeomorphic to G/K by Theorem 2.4.4.
Chapter 4
Modular Forms of Level One
4.1 Fourier expansions and q-series
4.1.1. Show that the punctured fundamental neighborhood
Uc = {z ∈ H : Im(z) > c}
of i∞ is mapped via z 7→ e2πiz into the punctured open disc centered at zero, of
radius e−2πc .
If z = x + iy, then e2πiz = e−2πy e2πix . For fixed y and x ∈ R, the map
z 7→ e2πiz takes the line Im(z) = y in H to the circle centered at zero of
radius e−2πy . t
u
4.1.2. Let γ = ac db ∈ GL+

2 (R) and z ∈ H. Define j(γ, z) = cz + d. Show that
j(γ1 γ2 , z) = j(γ1 , γ2 z)j(γ2 , z).
This is just a brute force calculation. Let

a 1 b1 a 2 b2
γ1 = , γ2 = ,
c 1 d1 c 2 d2
so that
j(γ1 γ2 , z) = (c1 a2 + d1 c2 )z + (c1 b2 + d1 d2 ).
Now j(γ2 , z) = c2 z + d2 and

a 2 z + b2
j(γ1 , γ2 z) = c1 + d1
c 2 z + d2
(c1 a2 + d1 c2 )z + (c1 b2 + d1 d2 )
=
c 2 z + d2
from which the result follows immediately. t
u
4.1.3. Show that f |(γ1 γ2 ) = (f |γ1 )|γ2 for all γ1 , γ2 ∈ GL+

2 (R).
Since (f |γ1 )(z) = (det γ1 )k/2 j(γ1 , z)−k f (γ1 z), we have

(f |γ1 ) γ2 (z) = (det γ2 )k/2 j(γ2 , z)−k (f |γ1 )(γ2 z)

= (det γ2 )k/2 j(γ2 , z)−k (det γ1 )k/2 j(γ1 , γ2 z)−k f (γ1 (γ2 z))
= f | (γ1 γ2 ) ,
by the multiplicativity of determinants and by Exercise 4.1.2. t

u
4.1.4. Prove that if f : H → C is holomorphic, satisfies

(i) f (z +1) = f (z)
(ii) f − z1 = z k f (z)
for all z ∈ H, and is holomorphic at i∞, then f is a modular form of weight k for
SL2 (Z).
Using the slash notation, we have f |T = f and f |S = f by (i) and (ii). By

Exercise 4.1.3, we see that f |γ = f for all γ in the group generated by S, T ,
which is SL2 (Z). t
u
4.1.5. Let k1 , k2 ∈ Z. Show that:

(a) Both Mk1 (SL2 (Z)) and Sk1 (SL2 (Z)) are C-vector spaces.
(b) If f ∈ Mk1 (SL2 (Z)) and g ∈ Mk2 (SL2 (Z)), then f g ∈ Mk1 +k2 (SL2 (Z)).
(c) If f ∈ Mk1 (SL2 (Z)) and g ∈ Sk2 (SL2 (Z)), then f g ∈ Sk1 +k2 (SL2 (Z)).

Suppose f , g ∈ Mk1 (SL2 (Z)), α, β ∈ C, and γ = ac db ∈ SL2 (Z). Then
(αf + βg) (γz) = αf (γz) + βg (γz)

= α(cz + d)k1 f (z) + β(cz + d)k1 g(z)
= (cz + d)k1 (αf + βg) (z) .
an q nand g = bn q n . We clearly have

P P
Consider the q-expansions f =
αf + βg = (αan + βbn )q n so that vi∞ αf + βg ≥ min {vi∞ (f ), vi∞ (g)}.
P
We conclude that αf + βg ∈ Mk1 (SL2 (Z)). It is now clear that Mk1 (SL2 (Z))
and Sk1 (SL2 (Z)) are vector spaces.
Now for f ∈ Mk1 (SL2 (Z)) and g ∈ Mk2 (SL2 (Z)), if γ = ac db ∈ SL2 (Z),
then
(f g)(γz) = f (γz) · g(γz)
= (cz + d)k1 f (z) · (cz + d)k2 g(z)
= (cz + d)k1 +k2 (f g)(z).
Moreover, vi∞ (f g) = vi∞ (f ) + vi∞ (g). This proves (b) and (c).
4.1.6. Show that Mk (SL2 (Z)) = {0} for k odd.

For γ = −I, the modular condition becomes f (z) = (−1)k f (z) from
which we deduce f (z) ≡ 0 since k is odd. t
u
4.2 Eisenstein series for SL2 (Z)
4.2.1. Show that the series for Gk (z) converges absolutely for all z ∈ H and for
k > 2.
We note that for z = x + iy, x, y ∈ R, we have
|mz + n|2 = (mx + n)2 + y 2 m2

= x2 + y 2 m2 + 2xmn + n2

which is a binary quadratic form corresponding to the matrix

2
x + y2 x

.
x 1
For z ∈ H, we have y 6= 0 and so this is positive definite. By our earlier

remark, there is a constant λ > 0 such that
|mz + n|2 ≥ λ(m2 + n2 ).
We are reduced to studying the convergence of

∞
X0 1 X r2 (t)
= ,
m,n
(m2 + n2 )k/2 t=1
tk/2
where r2 (t) is the number of ways of writing t as a sum of two squares of

integers. Since r2 (t) = O(t ) (see Exercise 1.3.4) for any > 0, the series
converges for k > 2. t
u
4.2.4. Define the Bernoulli numbers Bk by the power series
∞
X Bk xk
x
x
= ,
e −1 k!
k=0
so that B0 = 1, B1 = − 12 , B2 = 16 , B3 = 0, B4 = − 30
1
, . . . . Show that
(2πi)2k B2k
2ζ(2k) = − , k ≥ 1.
(2k)!
From (4.4) on page 39,

2πiz
πz cot πz = πiz + .
e2πiz − 1
But the right hand side is
∞
X (2πiz)k
πiz + Bk . (4.1)
k!
k=0
On the other hand, from (4.3) on page 39, we have

∞ ∞ X∞ 2k
X z2 X z
πz cot πz = 1 + 2 2 2
= 1 − 2 ,
n=1
z −n n=1
n
k=1
upon writing
∞
z2 (z/n)2 X z 2k
= = .
n2 − z 2 1 − (z/n)2 n
k=1
Thus
∞
X
πz cot πz = 1 − 2 ζ(2k)z 2k .
k=1
Combining this with (4.1) gives

∞ ∞
X (2πiz)k X
πiz + Bk =1−2 ζ(2k)z 2k (4.2)
k!
k=0 k=1
and the desired result follows by comparing coefficients. t

u
4.2.5. Prove that the Bernoulli numbers satisfy the recursion formula:
m−1
X
m+1
(m + 1)Bm = − Bk .
k
k=0
This is easily deduced by comparing coefficients of the identity

∞
X Bk xk
x = (ex − 1) ,
k!
k=0
upon writing ex − 1 as a power series in x. t

u
4.2.6. Define the n-th Bernoulli polynomial by

n
X n
Bn (x) = Bk xn−k .
k
k=0
Prove that
∞
text X xn
t
= Bn (x) .
e − 1 n=0 n!
This is immediate from comparing coefficients of the multiplication of

power series for ext and t/(et − 1). t
u
4.2.8. Prove that Mk (SL2 (Z)) = CEk ⊕ Sk (SL2 (Z)) for all k ≥ 4. Deduce that
dim Mk (SL2 (Z)) = 1 + dim Sk (SL2 (Z)).
Suppose f ∈ Mk (SL2 (Z)). Let λ = f (i∞) denote the constant term in the
Fourier expansion. Clearly

f = λEk + f − λEk
where f −λEk ∈ Sk (SL2 (Z)) since it vanishes at the cusp i∞. The conclusion
follows. t
u
4.3 The valence formula
4.3.2. If k < 0, show that Mk (SL2 (Z)) = {0}.
By the valence formula, we deduce that such an f is identically zero. t

u
4.3.3. Show that M0 (SL2 (Z)) = C.
If k = 0, the left hand side of the valence formula is zero. However, for
any f (z) ∈ M0 (SL2 (Z)), the form f (z) − f (i) ∈ M0 (SL2 (Z)) clearly has a
zero at z = i and hence f (z) must be constant. t
u
4.3.4. Show that M2 (SL2 (Z)) = {0}.
The right hand side of the valence formula is 1/6. This is impossible un-
less f ≡ 0. t
u
4.3.5. Show that E4 (z) has a simple zero at z = ρ.
Applying the valence formula to f = E4 , of weight 4, we see the right

hand side is 1/3. This can only arise if E4 has a simple zero at z = ρ. t
u
4.3.6. Show that E6 (z) has a simple zero at z = i.
Applying the valence formula to f = E6 , of weight 6, we find the right

hand side is 1/2 and this can arise only if E6 has a simple zero at z = i. t
u
4.3.7. Show that Sk (SL2 (Z)) = {0} for k ≤ 10.

If the weight of a modular form f is ≤ 10, then the right hand side of the
valence formula is ≤ 5/6 < 1. If there were a non-trivial cusp form of weight
≤ 10, there would be a contribution of at least 1 from the zero at i∞. t
u
4.3.8. Show that Mk (SL2 (Z)) = CEk for k = 4, 6, 8, and 10.
This follows immediately from Exercises 4.2.8 and 4.3.7. t

u
4.3.9. Show that E42 = E8 . Deduce that

n−1
X
σ7 (n) = σ3 (n) + 120 σ3 (m)σ3 (n − m).
m=1
By the previous exercise, M8 (SL2 (Z)) is one-dimensional and spanned

by E8 . Since E42 has weight 8, it must be a scalar multiple of E8 . The scalar
must be 1 since E4 (i∞) = 1. The identity for σ7 (n) is derived by comparing
coefficients in the q-series expansions of both sides of the equation E8 = E42 .
t
u
4.3.10. Show that E4 E6 = E10 . Deduce that

n−1
X
11σ9 (n) = 21σ5 (n) − 10σ3 (n) + 5040 σ3 (n)σ5 (n − m).
m=1
As in Exercise 4.3.9, E4 E6 = E10 holds since M10 (SL2 (Z)) is one-dimensional.

The identity for σ9 (n) arises from comparing coefficients in the q-expansions
of both sides of the equation E10 = E4 E6 . t
u
4.4 The dimension formula
4.4.3. For k ≥ 4, show that

(
k
b 12 c − 1 if k ≡ 2 (mod 12)
dim(Sk (SL2 (Z)) = k
b 12 c if k 6≡ 2 (mod 12)
The result is immediate by Exercise 4.2.8 and Corollary 4.4.2. t

u
E43 − E62
4.4.5. Prove that ∆ = has integer Fourier coefficients.
1728
We can write
E4 = 1 + 240A
E6 = 1 − 504B
where A, B are q-series with integer coefficients. Now
4.5 The j-function 191
E43 − E62 = (1 + 240A)3 − (1 − 504B)2

= 3(240A) + 3(240A)2 + (240A)3 + 2(504)B − (504B)2 .
Since 1728 = 123 , 240 = 12 × 20, and 504 = 12 × 42, we see that
3(240)2 ≡ (240)3 ≡ (504)2 ≡ 0 (mod 1728).
Thus E43 − E62 ≡ 144(5A + 7B) (mod 1728). But

∞
X
5A + 7B = (5σ3 (n) + 7σ5 (n))q n
n=1
and X
5σ3 (n) + 7σ5 (n) = (5d3 + 7d5 ).
d|n
Now d3 (5 + 7d2 ) ≡ d3 (d2 − 1) ≡ 0 (mod 3) and d3 (5 + 7d2 ) ≡ d3 (1 − d2 ) ≡ 0

(mod 4) so that all the coefficients of 5A + 7B are divisible by 12 and hence
all the coefficients of E43 − E62 are divisible by 1728. t
u
4.5 The j-function

4.5.2. Show that j(ρ) = 0 and j(i) = 1728. Moreover, show that vρ j(z) = 3,

that vi j(z) − 1728 = 2, and that vz0 j(z) − j(z0 ) = 1 for all z0 ∈ H which
are not SL2 (Z)-equivalent to either ρ or i.
By Exercises 4.3.5 and 4.3.6, we know that E4 (ρ) = 0 and E6 (i) = 0, and
that these are simple zeros. By the valence formula, there are no other zeros
of E4 and E6 in the fundamental domain F . Now
E4 (ρ)3
j(ρ) = =0
∆(ρ)
and
1728E4 (i)3
j(i) = = 1728.
E4 (i)3 − E6 (i)2
Moreover, it is clear that vρ (j) = 3vρ (E4 ) − vρ (∆) = 3, and since
1728E4 (z)3 1728E6 (z)2

j(z) − 1728 = − 1728 = ,
E4 (z)3 − E6 (z)2 E4 (z)3 − E6 (z)2

we also have vi (j − 1728) = 2vi (E6 ) − vi E43 − E62 = 2.
Suppose z0 is SL2 (Z)-inequivalent to ρ and to i. Then
E4 (z)3 − j(z0 )∆(z)

j(z) − j(z0 ) =
∆(z)

vanishes at z = z0 . Since ∆(z0 ) 6= 0, we have 1 ≤ vz0 j(z) − j(z0 ) =

vz0 E4 (z)3 − j(z0 )∆(z) . By the valence formula, this must be 1. t
u
4.5.3. Show that any modular function f has a finite number of zeros and poles in
F.
The set {z ∈ H : Im(z) > c} being a fundamental neighborhood of i∞,
and as the zeros and poles of any meromorphic function are isolated, we can
choose c sufficiently large such that i∞ is the only zero or pole of f in this
neighborhood. Now as the region
F ∩ {z ∈ H : Im(z) ≤ c},
is compact, it can only contain a finite number of zeros and poles. t

u
4.6.1. Show that the Bernoulli numbers Bk vanish for odd k ≥ 3.

From the equation defining the Bernoulli numbers, we have
∞
x x X Bk xk
x
−1+ = .
e −1 2 k!
k=2
Writing the left hand side as f (x), we see that
xex x x(ex − 1) + x x
f (−x) = x
− 1 − = x
− 1 − = f (x)
e −1 2 e −1 2
is even. Hence Bk = 0 for odd k ≥ 3.
Alternatively, we may conclude that the Bk vanish for odd k ≥ 3 by com-
paring coefficients on both sides of equation (4.2) in the solution to Exer-
cise 4.2.4. t
u
4.6.2. Show that for k even and positive, (−1)k/2 Bk < 0.
This follows from (4.2) in the solution to Exercise 4.2.4 since ζ(2k) > 0. t
u
4.6.3. Prove that
Bk+1 (n + 1) − Bk+1
1k + 2 k + · · · + n k = ,
k+1
for every natural number k.
Consider the series  

∞ n
X X tk
 jk
j=0
k!
k=0
which is easily evaluated upon interchanging the summations. We get

n
X e(n+1)t − 1
ejt =
j=0
et − 1
and the desired result now follows upon applying Exercise 4.2.6. t
u
4.6.4. Prove that ζ(k) tends to 1 as k tends to infinity and use this to show that
√
|B2k | ∼ 4π e(k/πe)2k+1/2
as k tends to infinity.
By the integral test, it is easily seen that ζ(k) = 1 + O(1/2k−1 ) so that ζ(k)
tends to 1 as k tends to infinity. The asymptotic for the Bernoulli numbers
follows upon using this together with Stirling’s formula:
√
(k)! ∼ (k/e)k 2πk.
t
u
4.6.5. Show that if the Fourier coefficients of f ∈ Mk (SL2 (Z)) are integers, then f
can be written as a polynomial in E4 , E6 , and ∆ with integer coefficients.
We proceed by induction on the weight. Since Mk (SL2 (Z)) is one dimen-

sional for k ≤ 10 and spanned by Ek , the result is clear. If f is a cusp form,
then f /∆ is a modular form of weight k − 12 with integral coefficients since
1/∆ has a q-expansion with Z-coefficients. If f is not a cusp form, write
k = 4a + 6b and consider g = f − λE4a E6b , where f (i∞) = λ. Note that
λ ∈ Z since λ is the constant term in the Fourier expansion of f at i∞. Since
E4 (i∞) = E6 (i∞) = 1. we see that g is a cusp form and we can apply induc-
tion to g/∆ and ultimately deduce the desired result. t
u
P∞
4.6.6. Suppose f = n=0 an q n ∈ Mk (SL2 (Z)). Show that for any x ∈ R we have
limy→∞ f (x + iy) = a0 . Show that for any x ∈ Q, if f is not a cusp form, then
limy→0 f (x+iy) = ∞, and if f is a cusp form, then limy→0 f (x+iy) = 0. Deduce
that the real line is a natural boundary for any modular form.
Write z = x + iy ∈ H. Then
∞
X
lim |f (x + iy) − a0 | = lim an e2πi(x+iy)n
y→∞ y→∞
n=1
∞
X
≤ lim |an | e−2πyn
y→∞
n=1
= 0.
a b

If x ∈ Q, then there exists γ = c d ∈ SL2 (Z) such that γx = i∞. In
particular, cx + d = 0 and so

ax + b + aiy ax + b + aiy a ax + b i
γ(x + iy) = = = − .
cx + d + ciy ciy c c y
Since γ(x + iy) ∈ H, we must have (ax + b)/c < 0. Thus we set t :=
exp (2π(ax + b)/c) < 1. Writing m = vi∞ (f ) now gives
∞ ∞
f (γ(x + iy)) X an 2πi( ac − ax+b
cy i)(n−m) =
X an (n−m)/y
− 1 ≤ e t
am e2πiγ(x+iy)m n=m+1
a m n=m+1
a m
which goes to 0 as y goes to 0. Hence limy→0 amfe(γ(x+iy))

2πiγ(x+iy)m = 1.
By modularity,
−k
f (z) = (cz + d) f (γz)
and hence
−k
lim f (x + iy) = lim (cx + iyc + d) f (γ(x + iy))
y→0 y→0
−k
= lim (iyc) f (γ(x + iy))
y→0
am e2πiγ(x+iy)m
= lim .
y→0 (iyc)k
If f is not a cusp form, then m = 0 and we have

a0
lim f (x + iy) = lim = ∞.
y→0 y→0 (iyc)k
If f is a cusp form, then
am tm/y
lim f (x + iy) = k
lim k = 0
y→0 |c| y→0 y
by l’Hopital’s rule. In either case, since Q is dense in R = ∂H and since mero-

morphic functions have isolated zeros and poles, there can be no meromor-
phic continuation of f to a larger domain. t
u
4.6.7. Prove that E42 E6 = E14 .

The dimension formula shows that M14 (SL2 (Z)) is one-dimensional and
so is spanned by E14 . Since E42 E6 also has weight 14, we have E42 E6 = cE14
for some constant c. A quick verification using Bernoulli numbers and
Corollary 4.2.5 gives the result. t
u
4.6.8. Let rk = dim Mk (SL2 (Z)). Put ` = k − 12rk + 12 and E0 = 1. Show that
E` E14−` = E14
for ` = 0, 4, 6, 8, 10 and 14.

The result is clear for ` = 0. For ` = 4, we must show that E4 E10 = E14
which again is checked as in the previous exercise. Alternately, by exercise
4.3.10, E4 E6 = E10 so that what we need to show is E42 E6 = E14 which is
the previous exercise. For ` = 6, we need to show E6 E8 = E14 . But again
by exercise 4.3.9, we have E42 = E8 and the question reduces to the previous
exercise. Finally, for ` = 8, 10, we are back to the above cases. This completes
the proof. t
u
4.6.9. Let f and g be modular forms of weight k and ` respectively, for the full
modular group. Show that
kf g 0 − `f 0 g
is a modular form of weight k + ` + 2 for the full modular group.
It is easily checked that

az + b
f0 = (cz + d)k+1 (kcf (z) + (cz + d)f 0 (z))
cz + d
so that the desired result is easily deduced by direct verification. t

u
4.6.10. Let j(z) = E43 /∆. Prove that
dj
= −2πiE14 /∆.
dz
We have
dj/dz = 3E42 E40 ∆−1 − ∆−2 E43 ∆0
so that
E43 − E62

d
2
∆ dj/dz = 3E42 E40 ∆ − E43
dz 1728
which equals
E42 E6
(2E4 E60 − 3E6 E40 ) .
1728
By the previous exercise, the expression in the brackets is a modular form of
weight 12. A quick verification shows that it is a cusp form and so must be a
suitable multiple of ∆. Comparing coefficients in the q-expansion immedi-
ately gives us the result. t
u
4.6.11. Let rk = dim Mk (SL2 (Z)) and ` = k − 12rk + 12. For any modular form
f ∈ Mk (SL2 (Z)), show that
f E`−1 ∈ M12rk −12 (SL2 (Z)).
Since f ∈ Mk (SL2 (Z)), it is a linear combination of
E4a E6b
such that 4a + 6b = k. From the previous exercise, E`−1 = E14−` /E14 , so that
f E`−1 is a linear combination of terms like
E4a E6b E14−` /E14 . (4.3)
Now E14 = E42 E6 and E14−` is one of E0 , E4 , E6 , E8 , E10 or E14 . Recalling

that E82 = E42 and E10 = E4 E6 , we see that E14−` /E14 is one of 1/E42 E6 ,
1/E4 E6 , 1/E42 , 1/E6 , or 1. If a ≥ 2, b ≥ 1, then the term in 4.3 is holomorphic.
Thus, we need only consider the five cases
(a, b) ∈ {(0, 1), (1, 0), (1, 1), (2, 0), (2, 1)}.
But in all these cases, rk = 1 so that ` = k ≤ 14 and the result is evident. t

u
4.6.12. With notation as in the previous exercise, show that
f E`−1 ∆1−rk
is a modular function and hence a polynomial in j.
By the previous exercise f E`−1 is a modular form of weight 12rk − 12 so

that f E`−1 ∆1−rk has weight zero and hence is a modular function. By Theo-
rem 4.5.4, it is a rational function of j. However, since ∆ does not vanish, it
must be a polynomial in j. t
u
4.6.13. Let f be a modular form of weight k for the full modular group whose Fourier
expansion at i∞ is given by
∞
X
f (z) = an e2πinz , an ∈ C.
n=0
Recall that by Corollary 4.4.4, there are unique cab ’s such that
X
f= cab E4a E6b , cab ∈ C.
4a+6b=k
Let F be an algebraic extension of Q. Suppose that
#{n : an ∈ F } ≥ dim Mk (SL2 (Z)).

Then, show that cab belong to F , whence all the Fourier coefficients belong to F .
We begin by noting that the Fourier coefficients of E4 and E6 are ra-

tional (in fact, integral). Inserting the q-expansions of f and E4a E6b in our
expression, and comparing the coefficients of q n on both sides for exactly
dim Mk (SL2 (Z)) many values of n such that an ∈ F , we get a system of
linear equations satisfied by the cab say, A(cab ) = v (with obvious notation),
for some matrix A with rational entries and vector v with entries from F .
Now the uniqueness (and the existence) of complex numbers cab imply the
invertibility of A. Since A has rational entries, so does A−1 . But then, the
cab ’s belong to F . t
u
4.6.14. If f is a modular form whose Fourier coefficients at i∞ are algebraic num-

bers, then show that the smallest exension field of the rationals containing all the
Fourier coefficients of f is a finite extension of Q.
Let F be the smallest field over Q containing the first dk Fourier coeffi-
cients of f , where dk is the dimension of Mk (SL2 (Z)) and k is the weight of
f . Then, by the previous exercise, all the Fourier coefficients are in F , which
is a finite extension of Q since it was obtained by adjoining finitely many
algebraic numbers. t
u
4.6.15. Suppose that f is a modular form of weight k whose Fourier expansion at

i∞ is given by
∞
X
f (z) = an e2πinz , an ∈ Q.
n=0
Show that, for every σ ∈ Gal(Q/Q), the series

∞
X
f σ (z) := σ(an )e2πinz ,
n=0
defines a modular form of weight k.
Using Corollary 4.4.4 and the earlier exercises, there are unique algebraic
numbers cab such that X
f= cab E4a E6b .
4a+6b=k
Since σ fixes the rationals pointwise, it follows that

X
fσ = σ(cab )E4a E6b
4a+6b=k
because E4 and E6 have rational Fourier coefficients. Thus, f σ is a modular

form of weight k. t
u
Chapter 5
The Ramanujan τ -function
5.1 The Eisenstein series E2
5.1.2. Show that E2 (z) has the Fourier expansion at i∞ given by

∞
X
E2 (z) = 1 − 24 σ1 (n)q n ,
n=1
where q = e2πiz and σ1 (n) =

P
d.
d|n
We first write
∞ ∞
1 X X 1
E2 (z) = 1 + .
ζ(z) m=1 n=−∞ (mz + n)2
Using the Lipschitz formula (Theorem 4.2.2) with k = 2 for the inner infinite
series, we get
∞ ∞
(−2πi)2 X X md
E2 (z) = 1 + dq .
ζ(2) m=1
d=1
As the double series is absolutely convergent, we may group together terms

with md = n to get
∞
4π 2 X
E2 (z) = 1 − σ1 (n)q n .
ζ(2) n=1
Finally, using the fact ζ(2) = π 2 /6 gives the result. t

u
5.1.3. Show that for all z ∈ H,

1 6
E2 − = z 2 E2 (z) + z.
z πi
Since G2 = 2ζ(2)E2 , this is immediate from Theorem 5.1.1. t

u
5.1.5. For z ∈ H, we define the Dedekind η-function by
∞
Y
η(z) = e2πiz/24 (1 − e2πinz ).
n=1
Show that r
1 z
η − = η(z),
z i
where the square root is the branch with non-negative real part.
The product defining η(z) clearly converges to a non-zero value for any
z ∈ H and thus defines a holomorphic function on H. Taking the logarithmic
derivative, we find
∞
!
η 0 (z) πi X nq n
= 1 − 24 ,
η(z) 12 n=1
1 − qn
where q = e2πiz . As in the proof of Theorem 5.1.4, the right hand side is
easily seen to be πiE2 (z)/12. Therefore
η 0 (−1/z)

πi 1 πi 2 6z
= E2 − = z E2 (z) + .
η(−1/z) 12 z 12 πi
In other words
η 0 (−1/z) 1 πiE2 (z) 1
· = + .
η(−1/z) z 2 12 2z
so that
η 0 (−1/z) 1 η 0 (z) 1
· 2 = + .
η(−1/z) z η(z) 2z
√
But the right hand side
√ is clearly the logarithmic derivative of zη(z). There-
fore, η(−1/z) = λ zη(z) for some constant√λ. Putting z = i shows that
√
λ = 1/ i. Finally, since z ∈ H, it is clear that √zi = zi .
p
t
u
5.1.6. Show that
∆0 (z)
= 2πiE2 (z).
∆(z)
Since
∞
Y
∆(z) = e2πiz (1 − e2πinz )24 ,
n=1
we take the logarithmic derivative of both sides and get

∞
!
∆0 (z) X ne2πinz
= 2πi 1 − 24 .
∆(z) n=1
1 − e2πinz
Putting q = e2πiz and expanding

∞
X nq n
n=1
1 − qn
as in the proof of Theorem 5.1.4 gives us the result. t

u

5.1.7. Prove that for all ac db ∈ SL2 (Z) and all z ∈ H, we have

az + b 6c(cz + d)
E2 = (cz + d)2 E2 (z) + .
cz + d πi
Recall that for any modular form f in Mk (SL2 (Z)), we have

az + b k ab
f = (cz + d) f (z) ∀ ∈ SL2 (Z) .
cz + d cd
Differentiating this with respect to z gives

az + b d az + b
f0 = (cz + d)k f 0 (z) + k(cz + d)k−1 cf (z).
cz + d dz cz + d
Now
d az + b
= (cz + d)−2
dz cz + d
so that
az + b
f0 = (cz + d)k+2 f 0 (z) + k(cz + d)k+1 cf (z).
cz + d
We apply this to ∆ ∈ M12 (SL2 (Z)) to get

az + b
∆0 = (cz + d)14 ∆0 (z) + 12c(cz + d)13 ∆(z).
cz + d
Dividing through by

az + b
∆ = (cz + d)12 ∆(z),
cz + d
we obtain using Exercise 5.1.6,

az + b
2πiE2 = (cz + d)2 2πiE2 (z) + 12c(cz + d),
cz + d

u
5.1.8. The Serre derivative of a modular form f ∈ Mk (SL2 (Z)) is defined as

1 0 k
θk f := f − E2 f.
2πi 12
Prove that θk : Mk (SL2 (Z)) → Mk+2 (SL2 (Z)).
First, we must check that

az + b k+2 ab
(θk f ) = (cz + d) (θk f )(z) ∀ ∈ SL2 (Z) .
cz + d cd

Since f ∈ Mk (SL2 (Z)), for any ac db ∈ SL2 (Z), we have

az + b
f = (cz + d)k f (z).
cz + d
Now from Exercise 5.1.7 and its solution,

az + b 6c(cz + d)
E2 = (cz + d)2 E2 (z) + ,
cz + d πi
and
0 az + b
f = (cz + d)k+2 f 0 (z) + kc(cz + d)k+1 f (z).
cz + d
Putting these together, we find

az + b 1 0 az + b k az + b az + b
(θk f ) = f − E2 f
cz + d 2πi cz + d 12 cz + d cz + d

1 k
= (cz + d)k+2 f 0 (z) − E2 (z)f (z)
2πi 12
= (cz + d)k+2 (θk f )(z)
as required.
It remains to check that θk f is holomorphic at the cusp i∞. In a neighbor-
hood of i∞,
X∞
f (z) = an e2πizn
n=0
and so
∞
1 0 X
f (z) = an ne2πizn .
2πi n=0
By Exercise 5.1.2, E2 (z) also has a Fourier expansion at i∞, and putting the
expansions together easily shows that θk f is holomorphic at i∞. t
u
5.2 Hecke operators of level one
5.2.3. Show that (

d−1
X d if d | n
e2πinb/d =
b=0
0 otherwise
If d|n, the result is clear. Otherwise the sum in question is a geometric

sum
d−1
X
rb , r = e2πin/d 6= 1,
b=0
which is easily summed to be (rd − 1)/(r − 1) = 0. t

u
5.2.5. Prove that Tm (∆) = τ (m)∆. Deduce that τ (mn) = τ (m)τ (n) for (m, n) =
1.
By Theorem 5.2.2, Tm (∆) is again a cusp form of weight 12. Since S12 (SL2 (Z))
is one-dimensional, and spanned by ∆, we must have Tm (∆) = λm ∆ for
some constant λm . On the other hand, by Theorem 5.2.4, the n-th coefficient
in the q-expansion of Tm (∆) is
X mn
d11 τ = τ (mn)
d2
d|m,n
whenever (m, n) = 1. But Tm (∆) = λm ∆ and the right hand side of the
equation shows that the n-th coefficient is λm τ (n). Thus τ (mn) = λm τ (n).
Putting n = 1 shows τ (m) = λm from which the result follows. t
u
5.2.6. Prove that for any m, n ≥ 1,

X mn
τ (m)τ (n) = d11 τ .
d2
d|m,n
Again, comparing the n-th coefficient of both sides of the equation Tm (∆) =
τ (m)∆, we get mn
X
τ (m)τ (n) = d11 τ ,
d2
d|m,n
by an application of Theorem 5.2.4. t

u
5.2.7. Let p be prime and a ≥ 1. Show that τ (pa+1 ) = τ (p)τ (pa ) − p11 τ (pa−1 ).
We apply Exercise 5.2.6 with m = p and n = pa to get
τ (p)τ (pa ) = τ (pa+1 ) + p11 τ (pa−1 ),
as required. t
u
5.2.8. Let f be E4 ∆, E6 ∆, E8 ∆, E10 ∆, or E14 ∆. Show that the Fourier coefficients

of f are multiplicative.
The weight of f is, respectively, 16, 18, 20, 22 or 26. As the space of cusp
forms is one-dimensional in each of these cases, it follows that Tm (f ) = λm f
for some constant λm . By Theorem 5.2.4, upon writing
∞
X
f (z) = af (n)q n ,
n=1
we have that the n-th Fourier coefficent of Tm (f ) is

X mn
dk−1 af ,
d2
d|n,m
where k = 16, 18, 20, 22, or 26, respectively. If (m, n) = 1, the sum becomes
af (mn). Hence af (mn) = λm af (n), whenever (m, n) = 1. Choosing n = 1,
we deduce af (m) = λm . The result now follows. t
u
5.2.9. Let p be a prime number. Show that
∞
X 1
τ (pn )xn = .
n=0
1 − τ (p)x + p11 x2
Equivalently, we must prove that

∞
11 2
X
1 − τ (p)x + p x τ (pn )xn = 1.
n=0
For n ≥ 2, the coefficient of xn in the product on the left hand side is τ (pn ) −
τ (p)τ (pn−1 ) + p11 τ (pn−2 ) which is zero by Exercise 5.2.7. For n = 1, the
coefficient of x is easily checked to be zero also, and as the coefficient of x0
is one, we are done. t
u
5.2.10. Prove that for n > 1,
n−1
X
(1 − n)τ (n) = 24 σ1 (j)τ (n − j).
j=1
By Exercise 5.1.6,
1 0
∆ (z) = ∆(z)E2 (z)
2πi
In other words,
∞ ∞ ∞
! !
X X X
nτ (n)q n = τ (n)q n 1 − 24 σ1 (n)q n .
n=1 n=1 n=1
5.3 Modular forms and Dirichlet series 205
The result now follows upon comparing coefficients of q n of both sides of

the equation. t
u
5.2.11. If n is even, prove that 8|τ (n).
This is evident from the previous exercise. t

u
5.2.12. If p is prime, show that τ (p) is even.
This follows from Exercise 1.5.1 where it was shown that τ (n) is odd if
and only if n is an odd square. t
u
5.3 Modular forms and Dirichlet series
5.3.1. Let p be prime. Prove that Ramanujan’s conjecture
|τ (p)| ≤ 2p11/2
is equivalent to the assertion that the roots of the quadratic polynomial 1 − τ (p)x +
p11 x2 are non-real.
The discriminant of the quadratic is τ (p)2 − 4p11 and this is < 0 if and
only if |τ (p)| < 2p11/2 . We have strict inequality here since τ (p) is an integer
and p11/2 is irrational. t
u
5.3.2. Assuming Ramanujan’s conjecture that |τ (p)| ≤ 2p11/2 , prove that |τ (n)| ≤
n11/2 d(n) for every natural number n. Here d(n) = σ0 (n) is the number of divisors
of n.
From Exercise 5.2.9, we have

∞
X
τ (pm )xm = (1 − τ (p)x + p11 x2 )−1 .
m=0
Writing 1 − τ (p)x + p11 x2 = (1 − αp x)(1 − βp x), we see from Exercise 5.3.1

that Ramanujan’s conjecture is equivalent to |αp | = |βp | = p11/2 . Thus, we
may write αp = p11/2 eiθp and βp = p11/2 e−iθp so that we have
∞
X 1 1 1
τ (pm )xm = − .
m=0
(αp − βp )x 1 − αp x 1 − βp x
Expanding both geometric series, we deduce that the mth coefficient is
αpm+1 − βpm+1 11m e

i(m+1)θp
− e−i(m+1)θp 11m sin(m + 1)θp
τ (pm ) = =p 2 =p 2 .
αp − βp e − e−iθp
iθ p sin θp
Now
sin(m + 1)θ ei(m+1)θ − e−i(m+1)θ

= = eimθ +ei(m−2)θ +· · ·+e−i(m−2)θ +e−imθ
sin θ eiθ − e−iθ
so that
sin(m + 1)θ
≤ (m + 1).
sin θ
Consequently, as τ (n) is multiplicative, we obtain
Y
|τ (n)| = τ (pα ) ≤ n11/2 d(n). t
u
pα ||n
5.3.3. Let f ∈ Mk (SL2 (Z)). Show that y k/2 |f (z)| is invariant under SL2 (Z),
where y = Im(z).
We have
az + b
f = (cz + d)k f (z)
cz + d
and
az + b Im(z)
Im =
cz + d |cz + d|2
from which the result follows. t
u
5.3.8. Suppose that f ∈ Mk (SL2 (Z)) with Fourier expansion at i∞ given by

∞
X
f (z) = af (n)q n ,
n=0
P∞ a (n)
and that af (0) 6= 0. Show that Lf (s) = n=1 fns has an analytic continuation
to the entire complex plane, except for a simple pole at s = 0 and s = k and satisfies
the functional equation
Put a0 = af (0). Let us note that f (iy) − a0 has exponential decay and we
can consider
Z ∞
(f (iy) − a0 )y s−1 dy = (2π)−s Γ(s)Lf (s)
0
as in the proof of Theorem 5.3.7. We split the integral into two parts:
Z 1 Z ∞
(f (iy) − a0 )y s−1 dy + (f (iy) − a0 )y s−1 dy.
0 1
5.4 Ramanujan congruences 207
In the first integral, we change variables y 7→ 1/y. Using the modular rela-
tion, we get
Z ∞
{ik y k (f (iy) − a0 ) + ik a0 y k − a0 }y −s−1 dy.
1
The integrals for Re(s) > k, the integrals

Z ∞ Z ∞
k−s dy dy
y and y −s
1 y 1 y
1
are easily evaluated as s−k and 1s , respectively. Therefore, for Re(s) > k,
∞
ik a0
Z
a0 dy
(2π)−s Γ(s)Lf (s) = − + (f (iy) − a0 )(y s + y k−s ik ) ,
s−k s 1 y
from which the analytic continuation and functional equation are now im-
mediate. t
u
5.4 Ramanujan congruences
5.4.1. Prove that τ (n) ≡ nσ1 (n) (mod 5).
Using the notation of Theorem 5.1.9, we have

∞
X
1728 τ (n)q n = Q3 − R2 ≡ Q − R2 (mod 5)
n=1
since Q ≡ 1 (mod 5). Now, R ≡ P (mod 5) as d5 ≡ d (mod 5) so we obtain

∞
X
3 τ (n)q n ≡ Q − P 2 (mod 5).
n=1
From Theorem 5.1.9, Q − P 2 = −12DP so that

∞
X
τ (n)q n ≡ DP (mod 5).
n=1
P∞
As DP = −24 n=1 nσ(n)q n , the result follows. t
u
5.4.3. Prove that Q2 ≡ P (mod 7) and R ≡ 1 (mod 7).
The second congruence is obvious from the q-expansion of R. For the first,
we note that Q2 = E8 and
∞
X n7 q n
E8 = 1 + 480 .
n=1
1 − qn
Since n7 ≡ n (mod 7) by Fermat’s little theorem, we conclude that
Q 2 = E8 ≡ P (mod 7). t
u
5.4.4. Prove that

dP 2 dQ
(Q3 − R2 )2 ≡ (QP − 1)2 ≡ −q − 3q (mod 7).
dq dq
(Q3 − R2 )2 ≡ (QP − 1)2 (mod 7).
Now
(QP − 1)2 ≡ Q2 P 2 − 2P Q + 1 (mod 7)

3
≡ P − 2P Q + R (mod 7)
by Exercise 5.4.3. Hence
(QP − 1)2 ≡ P (P 2 − Q) − P Q + R (mod 7).
By Theorem 5.1.9, P 2 − Q = 12DP and P Q − R = 3DQ. Thus
(Q3 − R2 )2 ≡ P (12DP ) − 3DQ (mod 7).

2
Now q dP dP
dq = 2P q dq so we may rewrite this as
dP 2 dQ
(Q3 − R2 )2 ≡ 6q − 3q (mod 7)
dq dq
dP 2 dQ
≡ −q − 3q (mod 7),
dq dq
as required. t
u
5.4.5. Prove that

∞ ∞
!
X Y
3 2 2 2 n
(Q − R ) ≡ q p(n)q (1 − q 49n ) (mod 7).
n=0 n=1
Deduce that p(7n + 5) ≡ 0 (mod 7).
By Theorem 5.1.4,
2 ∞ ∞
Q3 − R 2 (1 − q n )49
Y Y
= q2 (1 − q n )48 = q 2 .
1728 n=1 n=1
1 − qn
Hence we obtain
∞ ∞
!
X Y
3 2 2 2 n
(Q − R ) ≡ q p(n)q (1 − q 49n ) (mod 7),
n=0 n=1
upon reducing modulo 7. Combining this with Exercise 5.4.4, we get

∞ ∞
dP 2

2
X
n
Y
49n −1 dQ
q p(n)q ≡ (1 − q ) −q − 3q (mod 7). (5.1)
n=0 n=1
dq dq
Noting that for any q-series J with integer coefficients, the nth coefficient
of q dJ
dq is always divisible by n, we need only compare coefficients of q
7n
of
both sides of the congruence (5.1) to deduce the result. t
u
762048
5.5.1. Prove that E62 = E12 − 691 ∆.
Since E62 has weight 12, we can write it as αE12 +β∆ because M12 (SL2 (Z))
is spanned by E12 and ∆. Computing the coefficients of q 0 and q 1 leads to
the result. t
u
∞
5.5.2. Show that E12 = 1 + 65520 n
P
691 n=1 σ11 (n)q .
By Corollary 4.2.7, we need only compute the Bernoulli number B12 ,

which is easily done by directly computing it from the generating function
m−1
X
m Bk
or using the well-known recurrence B0 = 1, Bm = − k m−k+1 . t
u
k=0
5.5.3. Prove that τ (n) ≡ σ11 (n) (mod 691).
By Exercises 5.5.1 and 5.5.2, we find
65520σ11 (n) ≡ 762048τ (n) (mod 691),
from which the result immediately follows. t

u
5.5.4. Prove that E14 ≡ E2 (mod 13).
A quick computation of the Bernoulli number B14 shows that E14 ≡ 1 −

P ∞ n 13
24 n=1 σ13 (n)q (mod 13). Since d ≡ d (mod 13), the result follows. t
u
5.5.5. Prove that τ (n) ≡ nσ3 (n) (mod 7).
We have Q3 − R2 = 1728∆ ≡ −∆ (mod 7). Now Q2 ≡ P (mod 7) and

R ≡ 1 (mod 7) by Exercise 5.4.3. Thus QP − R ≡ −∆ (mod 7). By Theorem
5.1.9, 3DQ = QP − R. Thus −3DQ ≡ ∆ (mod 7). Now
∞
X
Q = 1 + 240 σ3 (n)q n ,
n=1
P∞
so that DQ = 240 n=1 nσ3 (n)q n , from which the result follows. t
u
5.5.6. Prove that

n−1
X 65 σ5 (n) 3τ (n)
σ5 (j)σ5 (n − j) = σ11 (n) + − .
j=1
(252)(691) 252 691
Deduce that
n−1
X 65
σ11 (n) + O n6 .

σ5 (j)σ5 (n − j) =
j=1
(252)(691)
This is immediate from Exercise 5.5.1 on comparing coefficients of q n of

both sides of the equation. The error term arises from using the estimate
τ (n) = O(n6 ) and σ5 (n) = n5+ . The error term can be sharpened using
Deligne’s result to O(n11/2+ ). t
u
5.5.7. If k1 > k2 ≥ 4 are even, prove that for any > 0,

n−1
X
σk1 −1 (j)σk2 −1 (n − j) = c(k1 , k2 )σk1 +k2 −1 (n) + O nk1 −1+ ,

j=1
where c(k1 , k2 ) ∈ Q is a constant depending on k1 and k2 .
If we consider the form Ek1 Ek2 of weight k1 + k2 , we may write it as

c1 Ek1 +k2 + c2 f , where f is a cusp form of weight k1 + k2 . The result fol-
lows upon comparing coefficients of q n in the q-expansions of both sides
and using the estimate
of Hecke given in Theorem 5.3.4 and the estimate

σk−1 = O nk−1+ . If k1 = k2 ≥ 4 is even, then the error term is O nk1 , as
in Exercise 5.5.6 above. t
u
5.5.8. Let fk be a sequence of complex numbers satisfying the recurrence with f0 =

1, and
fk+1 = f fk − gfk−1 , k≥1
for some constants f and g. Write
1 − f x + gx2 = (1 − αx)(1 − βx).

Show that
αk+1 − β k+1
fk = .
α−β
If α = β, the right hand side is viewed as αk + αk−1 β + · · · + αβ k−1 + β k =
(k + 1)αk .
The recurrence for fk implies that we have the equality of power series
∞
X −1
fk xk = 1 − f x − gx2 .
k=0
By partial fractions, we can write this as

1 1 1
− .
x(α − β) 1 − αx 1 − βx
Expanding the right hand side as a power series gives the desired result. t
u
5.5.9. Let g be a completely multiplicative function. An arithmetical function f is

called g-multiplicative if the equation
X mn
f (m)f (n) = g(d)f
d2
d|m,n
holds for all natural numbers m and n. Show that f is g-multiplicative if and only
if f is multiplicative and for all primes p and all natural numbers k ≥ 1, we have
f pk+1 = f (p)f (pk ) − g(p)f (pk−1 ).

If f is g-multiplicative, then setting m = p and n = pk in (5.8) from page

70 gives
f (p)f (pk ) = f (pk+1 ) + g(p)f (pk−1 )
which is (5.9) from page 70. For the converse direction, it suffices to observe
that we need only establish (5.8) for m, n powers of a fixed prime since f is
a multiplicative function. We write
1 − f (p)x + g(p)x2 = (1 − αp x)(1 − βp x)
so that by an application of the previous exercise, we obtain
αpk+1 − βpk+1
f pk = . (5.2)
αp − βp
Now since g is completely multiplicative, (5.8) on page 70 reduces to show-

ing
min(j,k)
X
f (pj )f (pk ) = g(p)i f (pj+k−2i ).
i=0
From (5.2), we have to show for j ≤ k,

j
!
αpj+1 − βpj+1 αpk+1 − βpk+1 X i αpj+k−2i+1 − βpj+k−2i+1
· = (αp βp )
αp − βp αp − βp i=0
αp − βp
since g(p) = αp βp . The right hand side of the above equation is easily
summed as
j j
αpk+1 X j−i i βpk+1 X i j−i
α β − α β
αp − βp i=0 p p αp − βp i=0 p p
from which the desired result is immediate. t

u
5.5.10. Let gk (n) = nk . Show that the Ramanujan τ -function is g11 -multiplicative.
This is immediate from the identity

X mn
τ (m)τ (n) = d11 τ
d2
d|m,n
established in Exercise 5.2.6.
5.5.11. Let gs (n) = ns . Show that σs (n) := ds is gs -multiplicative.

P
d|n
By Exercise 5.5.9, it suffices to show that for primes p,
σs pk+1 = σs (p)σs pk − ps σs pk−1

for k ≥ 1. But this is easily verified upon noting that
ps(α+1) − 1
σs (pα ) = 1 + ps + · · · + psα = .
ps − 1
5.5.12. Let F and G be two complex-valued functions defined on N × N. Suppose

that X
F (m, n) = G(m/d, n/d).
d|n
Show that X
G(m, n) = µ(d)F (m/d, n/d)
d|m,n
and conversely. (Here µ denotes the Möbius function.)
This is easily deduced using the Möbius inversion formula or by the

defining property of the Möbius function:
X
µ(d) = 0
d|n
unless n = 1, in which case it is equal to 1. Thus,

X X X
µ(d)F (m/d, n/d) = µ(d) G(m/de, n/de)
d|m,n d|m,n e|m/d,n/d
which is equal to (on setting r = de):

X X
G(m/r, n/r) µ(d) = G(m, n),
r|m,n de=r
as desired. The converse is similar.
5.5.13. Show that the Ramanujan τ -function satisfies

X
τ (mn) = d11 µ(d)τ (m/d)τ (n/d)
d|m,n
for any two natural numbers m, n.
Since
τ (m) τ (n) X
2 d11
= τ (mn/d ) ,
m11/2 n11/2 (mn)11/2
d|m,n
we can apply the previous exercise to the functions F (m, n) = τ (m)τ (n)/(mn)11/2
and G(m, n) = τ (mn)/(mn)11/2 to deduce the result. t
u
5.5.14. Let A be the n × n matrix whose (i, j) entry is τ (ij). Show that the determi-
nant of A is given by n!11 µ(1)µ(2) · · · µ(n). In particular, the determinant is zero
if n ≥ 4.
By Exercise 5.5.13, we can write our matrix A as a product LDLt , where

D is a diagonal matrix whose i-th diagonal entry is i11 µ(i) and L is the lower
triangular matrix whose (i, j) entry is given by τ (j/i) is i|j and zero other-
wise. The result is now immediate since the determinant of U is 1. t
u
5.5.15. Show that τ (n) is odd if and only if n is an odd square.
We have
∞
X ∞
Y
τ (n)q n = q (1 − q n )24
n=1 n=1
which modulo 2 is congruent to

∞
Y
q (1 − q 8n )3 .
n=1
By Jacobi’s identity (Theorem 1.2.3), this is congruent (modulo 2) to

∞ ∞
X 2 X 2
q q 4n +4n
= q (2n+1) .
n=0 n=0
t
u
5.5.16. Let r = dim Mk (SL2 (Z)). Put
Fk := E12r−k+2 ∆−r .
Show that Fk has q-expansion
Ckr q −r + · · · + Ck1 q −1 + Ck0 + · · ·
with Ckr = 1 and Ckj ∈ Z for all j.
Since
∞
Y
∆−1 = q −1 (1 + q n + q 2n + · · · )24 ,
n=1
the Fourier coefficients in the q-expansion of any integer power of ∆ are in-
tegers. Writing k = 12t + δ with 0 ≤ δ ≤ 10, we have by the dimension
formula that r = t + 1 or t according as δ 6≡ 2(mod 12) or δ ≡ 2(mod 12) re-
spectively. Thus, 12r − k + 2 = 14 − δ or 2 − δ according as δ 6≡ 2(mod 12) or
δ ≡ 2(mod 12). The Eisenstein series E12r−k+2 has accordingly integer coef-
ficients. Thus, the product E12r−k+2 ∆−r also has integer coefficients. More-
over, it is easily checked that the leading coefficient is 1. t
u
5.5.17. Let j denote the j-function. Show that for any integer m ≥ 0,
dj
jm
dz
has a q-expansion without a constant term.
For m = 0, this is clear. But since

1
j(z) = + ··· ,
q
and
dj 1 dj m+1
jm = ,
dz m + 1 dz
the right hand side has no constant term from which the result is immediate.
t
u
5.5.18. Let f be a modular form of weight k for the full modular group with q-
expansion
f (z) = a0 + a1 q + a2 q 2 + · · ·
With Ckj defined as in Exercise 5.5.16, show that
Ck0 a0 + Ck1 a1 + · · · + Ckr ar = 0.
By Exercise 5.5.16, the q-expansion of
Fk = E12r−k+2 ∆−r
has integer coefficients. On the other hand, by Exercise 4.6.12,
W := f E`−1 ∆1−r , ` = k − 12r + 12,
is a modular function (without poles except at infinity) and hence a polyno-

mial in j. Moreover,
dj
W
dz
has a q-expansion with zero constant term (by Exercise 5.5.17. By Exercise
4.6.10,
dj
= −2πiE14 /∆
dz
so that
1 dj
− E = f E`−1 ∆1−r E14 /∆.
2πi dz
From Exercise 4.6.8, E` E14−` = E14 so that
1 dj
− W = f E14−` ∆−r = f Fk
2πi dz
(using Exercise 5.5.16). Comparing the constant term of both sides gives us
the desired result. t
u
5.5.19. Suppose that k ≡ 2 (mod 4). With notation as in the previous exercise, show
that Ck0 6= 0.
Ck0 is the constant term of Fk = E12r−k+2 ∆−r with r = dim Mk (SL2 (Z)).
Since k ≡ 2(mod 4), we have k ≡ 2t(mod 12) with t = 1, 3 or 5. Then 12r =
k − 2, k + 6, k + 2 respectively, so that 12r − k + 2 = 0, 8, 4 respectively. Thus,
E12r−k+2 = E0 , E42 , E4
respectively. As all the Fourier coefficients of E4 are positive and the same is
true for ∆−r , we deduce that all Ck0 , Ck1 , ..., Ckr are positive and so Ck0 > 0.
t
u
5.5.20. Suppose that k ≡ 0 (mod 4). With notation as in the previous exercise, show
that Ck0 6= 0.
Since k ≡ 0 (mod 4), we have k ≡ 4t (mod 12), with t = 0, 1, 2. Then 12r =

k − 4t + 12 so that
Ek−12r+12 = E4t = E4t
since t = 0, 1 or 2. Moreover,
−1 dj
Fk = E12r−k+2 ∆−r = −E12r−k+2 ∆1−r E14 ,
2πidz
by Exercise 4.6.10. Again, by Exercise 4.6.8, E` E14−` = E14 with ` = k −
12r + 12 so that
−1 dj dj
Fk = −E14−` ∆1−r E14 = −E`−1 ∆1−r .
2πidz 2πidz
As ` = 4t, we can rewrite this as
dj
Fk = −E4−t ∆1−r .
2πidz
Since E43 /∆ = j, it is not difficult to see that
3 dj 1−t/3 3 d(E43−t ∆−r ) 3r + t − 3 3−t d∆−r

Fk = ∆1−r−t/3 = + E
t−3 2πidz t−3 2πidz (3 − t)r 4 2πidz
so that Ck0 is the constant term of the Fourier expansion of
3r + t − 3 3−t d∆−r
Vk := E .
(3 − t)r 4 2πidz
Since k > 2, we have 3r + t − 3 > 0 and the series E43−t begins with 1. More-
over, it has all strictly positive coefficients. Furthermore, the coefficients of
∆−r are all positive and when we differentiate and divide by 2πi, the coef-
ficients are all strictly negative, while the constant term is zero. Hence the
constant term of Vk is negative. That is, Ck0 < 0 and hence Ck0 6= 0. t
u
Chapter 6
Modular Forms of Higher Level
6.1 Modular forms for congruence subgroups
6.1.1. Suppose that f : H → C is holomorphic and satisfies (6.1) for a congruence

subgroup Γ of level N . Suppose γ, τ ∈ SL2 (Z) and that γ(i∞) and τ (i∞) are
Γ-equivalent. Prove that vγ(i∞) (f ) = vτ (i∞) (f ). That is, the order of f at a Γ-
equivalence class of cusps is well-defined.
By assumption, there exists δ ∈ Γ such that δτ (i∞) = γ(i∞). In particular,

γ −1 δτ stabilizes i∞. Hence, by Exercise 3.3.4 we must have

−1 1b
γ δτ = ±
01
for some b ∈ Z. Suppose that the expansions at γ(i∞) and τ (i∞) are
X
f |τ = bn q n/N
X
f |γ = cn q n/N .
Then since f |δ = f , we have
f |τ = f |δτ

1b
= f |γ| ±
01
X
k
= (±1) cn e2πi(z+b)n/N
X
= (±1)k cn e2πibn/N q n/N .
In particular, for all n we have bn = ±cn e2πibn/N . Thus, bn and cn are either
both zero or both non-zero. Hence the order of f is well-defined. t
u
6.2 Eisenstein series for Γ(N )
6.2.1. Let k ≥ 3 and N a natural number. Let g ∈ (Z/N Z)2 . Define the Eisenstein
series X 0 1
Gk,g (z) = , z∈H
(mz + n)k
(m,n)≡g (mod N )
where the summation is over all integers m, n satisfying the congruence condition
(m, n) ≡ g (mod N ) and the dash on the summation means we exclude (m, n) =
(0, 0) in the case g = (0, 0). For any γ ∈ SL2 (Z), show that Gk,g |γ = Gk,gγ .
Conclude that Gk,g ∈ Mk (Γ(N )).
The summands of Gk,g (z) are a proper subset of the summands of Gk (z).
We have already seen (Exercise 4.2.1) that series of this form converge ab-
solutely and uniformly on compact subsets of H for k > 2. Thus, Gk,g (z)
is holomorphic on the upper half-plane. Moreover, if g = (a1 , a2 ) with
a1 6≡ 0 (mod N ), then limIm z→∞ Gk,g (z) = 0. On the other hand, if a1 ≡ 0
(mod N ), then X
lim Gk,g (z) = n−k .
Im z→∞
n≡a2 (mod N )
In either case, Gk,g (z)

is holomorphic at i∞.
Now let γ = ac db ∈ SL2 (Z). Then a direct calculation shows that
k
az + b k
X 0 1
Gk,g = (cz + d)
cz + d (ma + nc)z + (mb + nd)
(m,n)≡g (mod N )
k
= (cz + d) Gk,gγ (z)
since
ab
(m n) = ma + nc mb + nd
cd
and if we write gγ = (b1 , b2 ) the map (m, n) 7→ (m, n)γ sets up a one-to-one
correspondence between pairs of integers (m, n) ≡ g (mod N ) and pairs
(m0 , n0 ) ≡ gγ (mod N ). This proves Gk,g |γ = Gk,gγ for all γ ∈ SL2 (Z). Since
Gk,g (z) is holomorphic at i∞, it follows that it is holomorphic at all cusps. Fi-
nally, if γ ≡ I (mod N ), then gγ ≡ g (mod N ) so that Gk,g (z) ∈ Mk (Γ(N )).
t
u
6.2.2. Show that Gk,g = (−1)k Gk,−g . Deduce that if k is odd and 2g ≡ (0, 0)
(mod N ), then Gk,g = 0.
Putting γ = −I in Exercise 6.2.1, we see that Gk,g (z) = (−1)k Gk,−g (z).
To say 2g ≡ (0, 0) (mod N ) means g ≡ −g (mod N ) so that if k is odd,
Gk,g (z) ≡ 0. t
u
6.2.3. Let Γ be a congruence subgroup of SL2 (Z) of level N . If γ1 , . . . , γr are distinct
right coset representatives of Γ(N ) in Γ, show that
6.2 Eisenstein series for Γ(N ) 219
r
X
f := Gk,gγj ∈ Mk (Γ).
j=1
First observe that f is independent of the choice of coset representatives:

If Γ(N )δ = Γ(N )δ 0 , then δ 0 δ −1 ∈ Γ(N ) and hence Gk,g |δ 0 δ −1 = Gk,g which
implies that Gk,gδ0 = Gk,g |δ 0 = Gk,g |δ = Gk,gδ .
By Exercise 6.2.1, we have for γ ∈ Γ,
r
X
f |γ = Gk,gγj γ .
j=1
As the γj run through a complete set of coset representatives of Γ(N ) in Γ,

so do the γj γ and so f |γ = f , for all γ ∈ Γ. t
u
6.2.4. Let Γ1 , Γ2 be congruence subgroups of SL2 (Z) such that Γ1 ⊂ Γ2 . If f ∈

Mk (Γ1 ) and {γ1 , . . . , γr } is a complete set of right coset representatives of Γ1 in
Γ2 , show that
r
X
(a) f |γj ∈ Mk (Γ2 ),
j=1
Yr
(b) (f |γj ) ∈ Mkr (Γ2 ).
j=1
The essential idea is the same as in Exercise 6.2.3. We have

 
Xr r
X
 f |γj  |γ = f |(γj γ)
j=1 j=1
and {γ1 γ, . . . , γr γ} is again a complete set of coset representatives of Γ1 in

Γ2 . Thus the function is invariant upon ’slashing’ and this proves (a).
Noting that (f1 f2 )|γ = (f1 |γ)(f2 |γ), we see that (b) follows similarly. t u
6.2.5. Show that Γ0 (4) is generated by T = ( 10 11 ), by ( 14 01 ), and by −1 0

0 −1 .
k
Let γ = ac db ∈ Γ0 (4). Then a is odd. If b = 0, then γ = ± ( 14 01 ) for some
k. If b 6= 0, then noting that γT k = ac b+ka

d+kc , and choosing k appropriately,
|a| k
we may suppose that |b| < 2 (as a is odd). Noting γ ( 14 01 ) = a+4kb b

c+4kd d , we
may assume |a| < 2|b|. Iteration eventually gives b = 0. t
u
6.2.6. Show that there are exactly three inequivalent cusps for the group Γ0 (4). Find
the widths of each cusp.
A complete set of inequivalent cusps is contained in {γ(i∞)} as γ runs

through the coset representatives of (6.4) on page 76. In particular, there are
three cusps: i∞, 1/2, and 0. To determine the width of γ(i∞), we look for
h ∈ Z such that
1h
γ(±I) γ −1 ∈ Γ0 (4).
01
Since −I ∈ Γ0 (4), we consider γ ( 10 h1 ) γ −1 for each of the coset representa-
tives γ. We first consider γ = ( 12 01 ). Now

1h −1 10 1h 1 0 1 − 2h h
γ γ = = .
01 21 01 −2 1 −4h 1 + 2h
It is clear that this matrix is in Γ0 (4) for all h ∈ Z. Hence, the cusp ( 12 01 ) i∞ =
1
2 has width 1.
Now consider γ = 01 −1

m for m = 1, 2, 3, or 4. We have

1h 0 −1 1h m 1 10
γ γ −1 = = ,
01 1 m 01 −1 0 h1
which is in Γ0 (4) if and only if 4 | h. Hence, the cusp 01 −1

m i∞ = 0 has
width 4. Finally, it is clear that the cusp ( 10 01 ) i∞ = i∞ has width 1.
6.3 Fourier expansions of higher level Eisenstein series
6.3.2. For g = (a1 , 0) ∈ (Z/N Z)2 and n 6= 0, show that
(−2πi)k X
bk,g (n) = dk−1 (sgn d).
N k (k − 1)!
d|n
n/d≡a1 (mod N )
This is immediate from Theorem 6.3.1 since a2 = 0. t

u
6.3.3. If g = (0, a2 ) ∈ (Z/N Z)2 , show that bk,g (n) = 0 if N - n, and
(−2πi)k X k−1 da2

bk,g (N n) = d (ζ + (−1)ζ −da2 ),
N k (k − 1)!
d|n
where ζ = e2πi/N . Deduce that Gk,g (z), with g = (0, a2 ), has a q-expansion of the
form
X∞
an q n , q = e2πiz .
n=0
Again from Theorem 6.3.1, we see that the condition n/d ≡ a1 ≡ 0

(mod N ) implies that summation is empty, and hence that bk,g (n) = 0,
whenever N - n.
If N | n, we see that the summation formula for bk,g (n) in Theorem 6.3.1
shows that if we write n = N n0 , the summation over divisors of N n0 can be
re-written as a sum over divisors of n0 which reduces to the stated formula.
Finally, since the Fourier expansion is only supported on multiples of N , we
deduce that Gk,g has a traditional q-expansion, as given. t
u
6.3.4. Prove that if g = (0, a2 ) ∈ (Z/N Z)2 , then Gk,g ∈ Mk (Γ1 (N )) for k ≥ 3.
By Exercise 6.2.1, we have
Gk,g |γ = Gk,gγ ∀γ ∈ Γ1 (N )
and we need only observe that (0, a2 )γ ≡ (0, a2 ) (mod N ), to deduce the
desired result. t
u
6.3.5. Let
E2,N (z) := E2 (z) − N E2 (N z).
Show that E2,N ∈ M2 (Γ0 (N )).
It is clear that E2,N (z) is a holomorphic function on H. Now let γ =
a b ∈ Γ0 (N ). Then
c d

az + b az + b aN z + bN
E2,N = E2 − N E2 .
cz + d cz + d cz + d
By Exercise 5.1.7 (the transformation law for E2 ), we have

az + b 6c(cz + d)
E2 = (cz + d)2 E2 (z) + .
cz + d πi
Also,
a(N z) + bN 6c(cz + d)
E2 c = (cz + d)2 E2 (N z) +
N (N z) + d N πi
so that

az + b
E2,N = (cz + d)2 (E2 (z) − N E2 (N z))
cz + d
= (cz + d)2 E2,N (z).
We must check that E2,N is holomorphic at every cusp. This will be a

more general argument that will obviate the calculation above. Suppose
now that γ = ac db ∈ SL2 (Z). We must determine a Fourier expansion of
E2,N (z) at ac db (i∞). Equivalently, we determine an expansion at i∞ of
E2,N (z)| ac db . The heart of this calculation deals with

ab −2 N 0 ab
N E2 (N z)| = (cz + d) N E2 z .
cd 0 1 cd
Recalling Theorem 5.2.1, since

N 0 ab Na Nb
= ∈ XN ,
0 1 cd c d
we know that there exist

rs x y
∈ SL2 (Z) and ∈ XN
tu 0w
such that
N 0 ab rs x y
= .
0 1 cd tu 0w
In particular, some computations as in the proof of Theorem 5.2.1 easily
show that we must have
x = gcd(c, N )
c = tx
N a = ux
xw = N
d = yt + uw.
Now

ab rs x y
N E2 (N z)| = (cz + d)−2 N E2 z
cd tu 0w

rσ + s
= (cz + d)−2 N E2
tσ + u
where
x y x y
σ := z= z+ .
0w w w
Observe that
w(tσ + u) = xtz + (yt + uw) = cz + d.
Moreover, by Exercise 5.1.7,

rσ + s 6t(tσ + u)
E2 = (tσ + u)2 E2 (σ) + .
tσ + u πi
Thus,

ab 6t(tσ + u)
N E2 (N z)| = (cz + d)−2 N (tσ + u)2 E2 (σ) +
cd πi
2

−2 (cz + d) 6t(cz + d)
= (cz + d) N E2 (σ) +
w2 wπi
N 6c
= 2 E2 (σ) + .
w πi(cz + d)
Therefore,

ab ab ab
E2,N (z)| = E2 (z)| − N E2 (N z)|
cd cd cd
x
= E2 (z) − E2 (σ).
w
Now
x ∞
y y
σ1 (n)e2πin( w z+ w )
X x
E2 (σ) = E2 z+ = 1 − 24
w w n=1
∞
X
= 1 − 24 σ1 (n)e2πiny/w q nx/w .
n=1

In particular,
we see that E2,N (z)| ac db is holomorphic at i∞. Also, observe
that if ac db ∈ Γ0 (N ), then
N |c and sox x = N and w = 1. In particular, this
shows that E2,N (z)| ac db = E2 (z) − w E2 (z + y) = E2,N (z) is modular. t
u
6.3.6. If f ∈ Mk (SL2 (Z)), show that
fN (z) := f (z) − N f (N z) ∈ Mk (Γ0 (N )).
In particular, deduce that for k ≥ 3,
Ek,N (z) := Ek (z) − N Ek (N z) ∈ Mk (Γ0 (N )).
It is clear that fN is holomorphic on H since f is. The modular transfor-

mation rule is immediate from

az + b az + b a(N z) + bN
fN =f − Nf = (cz + d)k fN (z)
cz + d cz + d ( Nc )(N z) + d

which holds for all ac db ∈ Γ0 (N ).
We must check that fN is holomorphic at every cusp. Suppose now that
γ = ac db ∈ SL2 (Z). As in the solution to Exercise 6.3.5, we use Theo-
rem 5.2.1 to deduce the existence of ( rt us ) ∈ SL2 (Z) and ( x0 wy ) ∈ XN such
that
N 0 ab rs x y
= .
0 1 cd tu 0w
Let σ := ( x0 wy ) z = w
x
z + wy . Now w(tσ + u) = xtz + yt + uw = cz + d as in
the solution to Exercise 6.3.5. Thus

ab ab ab
fN (z)| = f (z)| − N f (N z)|
cd cd cd

rs
= f (z) − N (cz + d)−k f σ
tu
= f (z) − N (cz + d)−k (tσ + u)k f (σ)
k
−k cz + d
= f (z) − N (cz + d) f (σ)
w
N x y
= f (z) − k f z+ .
w w w
P∞
If we assume that f (z) = n=0 an q n is the Fourier expansion of f at i∞,
then
∞
N x y N X
f z + = an e2πiny/w q xn/w
wk w w wk n=0

is clearly holomorphic at i∞ and hence so is fN (z)| ac db . t
u
6.4 The valence and dimension formulas
6.4.1. Show that the only elliptic points for SL2 (Z) lying in the standard funda-
mental domain F are i, ρ = eπi/3 , and ρ2 .

Suppose that z ∈ F and ±I 6= ac db ∈ SL2 (Z) are such that
az + b
z= . (6.1)
cz + d

We must have c 6= 0 since otherwise ac db = ± ( 10 1b ) is a translation which
will only fix z when b = 0. Now by (6.1) above, we have
cz 2 + (d − a)z − b = 0
which implies that

p
a−d± (d − a)2 + 4bc
z= .
2c
a b

Since det c d = 1, we have bc = ad − 1 and hence
p
a−d± (a + d)2 − 4
z= .
2c
6.4 The valence and dimension formulas 225
Now z ∈ H and so (a+d)2 −4 < 0 which can only happen when a+d =√0, ±1.
If a + d = 0 then a = −d and z = ac ± ci ∈ F . Thus |c|1
= Im z ≥ 23 and
hence |c| = 1. Now − 12 ≤ Re z ≤ 12 implies that a = 0. Hence z = i.
√ √ √
If a + d = ±1, then Im z = 2|c|3 . But z ∈ F implies that 2|c|3 ≥ 23 which
√
3
forces |c| = 1 and Im z = 2 . Thus z = ρ or ρ2 . t
u
6.4.2. Find the stabilizer subgroups SL2 (Z)i and SL2 (Z)ρ in SL2 (Z) and show that
SL2 (Z)i / {±I} and SL2 (Z)ρ / {±I} have orders two and three, respectively.

Suppose ±I 6= ac db ∈ SL2 (Z)i . From the solution to Exercise 6.4.1, we
know that 0 = a = −d and |c| = 1. Since ad − bc = 1, wededuce that b = −c
and hence that ac db = ± 01 −1 0 . Therefore SL2 (Z)i = ±I, ± 1 0
0 −1
and
so | SL2 (Z)i / ± I| = 2.
Now suppose that ±I 6= ac db ∈ SL2 (Z)ρ . From the solution to Exer-
cise 6.4.1, we know that |a + d| = 1 and |c| = 1. We also know that
√ p
1 3 a − d ± (a + d)2 − 4
− + i= .
2 2 2c
Hence Re ρ = −1/2 = (a − d)/2c. Thus c = d − a. It is easy to determine the
four possible combinations of a, c and d which satisfy these requirements.
In each case,
b can be determined from ad − bc = 1. We may conclude that
SL2 (Z)ρ = ±I, ± 01 −1 1 1

1 , ± −1 0 . In particular, | SL2 (Z)ρ / ± I| = 3. t
u
6.4.3. If Γ has finite index in SL2 (Z), show that there are only finitely many Γ-
inequivalent elliptic points for Γ.
Suppose that z ∈ H is an elliptic point for Γ. By the definition of a fun-

damental domain, there exists γ ∈ SL2 (Z) such that γz ∈ F , where F
is the standard fundamental domain for SL2 (Z). It is easily verified that
Γγz = γΓz γ −1 6= {±I}. Hence γz is an elliptic point for Γ and hence also
for SL2 (Z). By Exercise 6.4.1, we see that z is SL2 (Z)-equivalent to either i or
ρ (or to ρ2 = 10 −1

0 ρ and hence to ρ).
Let γ1 , . . . , γd be a complete set of coset representatives for Γ in SL2 (Z).
We deduce that z is Γ-equivalent to an element of the finite set
{γj (i), γj (ρ) : 1 ≤ j ≤ d}. t

u
6.4.5. Verify that Theorem 6.4.4 reduces to Theorem 4.3.1 in the case Γ = SL2 (Z).
By Exercises 6.4.1 and 6.4.2, the only SL2 (Z)-inequivalent elliptic points
are i and ρ and the appropriate stabilizers have orders 2 and 3 respectively.
This tallies with the left hand side of the formula in Theorem 4.3.1. Since
2 = 1, 3 = 1, g = 0, and ∞ = 1, this gives us k/12, which agrees with
Theorem 4.3.1. t
u
∞
X ∞
X
6.4.8. Let f = an q n and g = bn q n ∈ Mk (Γ) for some congruence sub-
n=0 n=0
group Γ of SL2 (Z). Suppose that an = bn for all n ≤ k[SL2 (Z) : {±}Γ]/12. Show
that f = g.
Apply Theorem 6.4.7 to f − g ∈ Mk (Γ). t

u
6.4.9. For any congruence subgroup Γ of SL2 (Z), prove that
k
dim Mk (Γ) ≤ [SL2 (Z) : {±}Γ] + 1.
12
Consider the linear map Mk (Γ) → Cr+1 , where r = k[SL2 (Z) : {±}Γ]/12,
X∞
given by f = an q n 7→ (a0 , a1 , . . . , ar ). Sturm’s bound (Theorem 6.4.7)
n=0
says that this map is injective. Hence dim Mk (Γ) ≤ r + 1. t
u
6.4.10. Prove that

dim M4 (Γ0 (4)) = 3.
(Hint: consider E4 , E4,2 , and E4,4 from Exercise 6.3.6.)
By Exercises 2.3.4 and 6.4.9, we have
1
dim M4 (Γ0 (4)) ≤ [SL2 (Z) : Γ0 (4)] + 1 = 3.
3
Now E4 , E4,2 , E4,4 ∈ M4 (Γ0 (4)) and a quick check of their q-expansions at
i∞ shows that
E4 = 1 + 240q + 2160q 2 + · · ·
E4,2 = 1 − 16q + 112q 2 + · · ·
E4,4 = 1 − 16q 2 + 112q 4 + · · ·
and these are easily verified to be linearly independent over C. Thus, the
forms are a basis for M4 (Γ0 (4)). t
u
6.4.11. Prove that

M2 (Γ0 (2)) = CE2,2 ,
where E2,2 (z) = E2 (z) − 2E2 (2z).
By Exercise 6.4.9,
dim M2 (Γ0 (2)) ≤ 3/2,
so that dim M2 (Γ0 (2)) ≤ 1. Since E2,2 ∈ M2 (Γ0 (2)) by Exercise 6.3.5, and as
E2,2 6= 0, we are done. t
u
6.4.12. Show that dim M2 (Γ0 (4)) = 2 and that {E2,2 , E2,4 } is a basis for M2 (Γ0 (4)).
By Exercise 6.4.9,
dim M2 (Γ0 (4)) ≤ 2.
Now E2,2 and E2,4 ∈ M2 (Γ0 (4)) by Exercise 6.3.5. It is easy to check that
E2,2 = −1 − 24q + · · ·
E2,4 = −3 − 24q + · · ·
Clearly, these elements are linearly independent over C. t

u
6.5 The Jacobi four-square theorem revisited
6.5.1. Let η(z) denote the Dedekind η-function. Prove that

∞
X 2 η 5 (2z)
θ(z) := qn = ,
n=−∞
η 2 (z)η 2 (4z)
where q = e2πiz .
Set x = 1 in Jacobi’s Triple Product Identity (Theorem 1.2.1) to get

∞ ∞
X 2 Y
qn = (1 − q 2n )(1 + q 2n−1 )2 ,
n=−∞ n=1
which can be rewritten as

∞ ∞
X 2 Y (1 − q 2n )(1 + q n )2
qn = .
n=−∞ n=1
(1 + q 2n )2
Now 1 + q n = (1 − q 2n )/(1 − q n ) and 1 + q 2n = (1 − q 4n )/(1 − q 2n ), so this

becomes
∞ ∞
X 2 Y (1 − q 2n )5 η 5 (2z)
qn = = . t
u
n=−∞ n=1
(1 − q n )2 (1 − q 4n )2 η 2 (z)η 2 (4z)
6.6.1. Show that η 12 (2z) ∈ S6 (Γ0 (4)).
By the proof of Theorem 6.5.2, we see that

s r
2z −πi 4z + 1 2z
η =e 6 i η(2z)
4z + 1 2z i
so that η 12 (2z) transforms with weight 6 using Exercise 6.2.5. Now Γ0 (4) has
three inequivalent cusps, 0, 1/2 and i∞. Clearly
∞
Y 12
η 12 (2z) = q 1 − q 2n = q + ···
n=1
has a zero of order 1 at the cusp i∞.

0 −1

To check the behaviour of η 12 (2z) at 0, we need to consider η 12 (2z) 1 0
at i∞. But

0 −1 1 1
η 12 (2z) = z −6 η 12 2 − = z −6 η 12 −
1 0 z (z/2)
r !12
(z/2) z
= z −6 η 12
i 2
1 z
= − 6 η 12
2 2
1 1/4
= − 6q + ··· .
2
In particular, η 12 (2z) has a zero of order 1 at the cusp 0. (Recall that the cusp
0 has width 4).
Finally, to check the behaviour at 1/2, we need to consider η 12 (2z) |( 12 01 )
at i∞. Now

12 10 −6 12 2z
η (2z) = (2z + 1) η
21 2z + 1

−6 12 1
= (2z + 1) η 1−
2z + 1

−6 12 1
= − (2z + 1) η −
2z + 1
= η 12 (2z + 1)
= −η 12 (2z)
= −q + · · · .
In particular, η 12 (2z) has a zero of order 1 at the cusp 1/2. We conclude that
η 12 (2z) ∈ S6 (Γ0 (4)). t
u
6.6.2. Show that S6 (Γ0 (4)) = η 12 (2z)C.

If f ∈ S6 (Γ0 (4)), we must have a zero of order at least 1 at each of i∞,

0, and 1/2. Then for some λ, f (z) − λη 12 (2z) has a zero of order 2 at i∞. By
Sturm’s bound (Theorem 6.4.7), we deduce f (z) = λη 12 (2z). t
u
6.6.3. Let k ≥ 6 be even. Show that Sk (Γ0 (4)) = η 12 (2z)Mk−6 (Γ0 (4)).
From the solution to Exercise 6.6.1, η 12 (2z) has a zero of order 1 at ev-
ery cusp. Moreover, η 12 (2z) does not vanish for z ∈ H. Hence, for any
f ∈ Sk (Γ0 (4)), we deduce that f /η 12 (2z) is holomorphic in H and at all
cusps of Γ0 (4). Hence f /η 12 (2z) ∈ Mk−6 (Γ0 (4)). t
u
k
6.6.4. For even k ≥ 0, prove that dim Mk (Γ0 (4)) = 2 + 1.
By Exercise 6.4.9, we have dim Mk (Γ0 (4)) ≤ k2 + 1. Hence, it suffices to

find enough linearly independent forms. We modify the basis from Exer-
cise 6.4.12 by setting
F := −E2,2
E2,4 − 3E2,2
G :=
48
so that F = 1 + 24q + · · · and G = q + · · · ∈ M2 (Γ0 (4)) are linearly indepen-
dent. Now
k
F 2 −i Gi = q i + · · ·
for i = 0, 1, . . . , k2 , are clearly linearly independent.
6.6.5. Let r12
P(n) be the number of ways of writing n as a sum of 12 squares and let
∞
η 12 (2z) = n=1 an q n . Show that
n
r12 (n) = 8σ5 (n) − 512σ5 + 16an .
4
By Exercise 6.6.4 applied to k = 6, we have dim M6 (Γ0 (4)) = 4. Moreover,
E6 (z), E6,2 (z), E6,4 (z), η 12 (2z)
comprise a basis. Now θ12 (z) ∈ M6 (Γ0 (4)) and a quick comparison of the
q-expansions immediately leads to the result. t
u
6.6.6. Let f (z) ∈ Mk (Γ1 (N )) and suppose r ∈ Z is positive. Show that f (rz) ∈
Mk (Γ1 (rN )). Moreover, if f (z) is a cusp form, then so is f (rz).
By Equation (4.2) on page 36, f (rz) = r−k/2 f (z)|( 0r 10 ). Now the cusp
conditions (holomorphy/vanishing) follow from Exercises 7.1.6 and 7.1.7. It
remains to show that r 0
a b
f (z)|( 0 1 ) is invariant under the modular transforma-
tion. For any c d ∈ Γ1 (rN ), we have
! !
r0 ab ra rb a rb r0
f (z) = f (z) = f (z) .
01 cd c d c/r d 01

a rb
Since rN |c, we see that c/r d ∈ Γ1 (N ). Hence
!
a rb r0 r0
f (z) = f (z) .
c/r d 01 01
We conclude that f (rz) ∈ Mk (Γ1 (rN )). t

u
Chapter 7
The Petersson Inner Product
7.1 The Hilbert space of cusp forms
7.1.1. Let f , g ∈ Mk (Γ) and let F be any fundamental domain for the congruence
subgroup Γ. Show that the integral
ZZ
dxdy
y k f (z)g(z) 2 < ∞
F y
as long as one of f, g ∈ Sk (Γ).
The integral is
X ZZ dxdy
y k f (z)g(z) ,
j
y2
α−1
j (D)
where D is the standard fundamental domain for SL2 (Z) and

[
SL2 (Z) = αj Γ
j
is a disjoint coset decomposition of Γ in SL2 (Z). Each of the integrals is

ZZ
k dxdy
Im(αi−1 z) f (αi−1 z)g(αi−1 z) 2 .
D y
Since one of the f, g ∈ Sk (Γ), we see that the integral is O y k e−2πy and

hence the integral converges. t

u
7.1.2. Prove that

(i) (f, g) is linear in f ,
(ii) (f, g) is conjugate symmetric. That is, (f, g) = (g, f ),
(iii) (f, f ) > 0 for f 6= 0.
Therefore, the Petersson inner product defines a Hermitian inner product on Sk (Γ).
This is immediate from the definitions. t

u
7.1.4. Suppose α ∈ GL+2 (Q) has integer entries. Let D = det α. If Γ(N ) ⊆ Γ for
some N , show that Γ(N D) ⊆ α−1 Γα and Γ(N D) ⊆ αΓα−1 .
Note that α is not necessarily in SL2 (Z) and so Exercise 2.2.2 is not appli-
cable. The inverse α−1 does not necessarily have integer entries.
Let γ ∈ Γ(N D). Write γ = I + N Dβ for some 2 × 2 matrix β with inte-
ger entries. We want to show γ ∈ α−1 Γα, which is equivalent to showing
αγα−1 ∈ Γ. Now
αγα−1 = α(I + N Dβ)α−1 = I + N αβ(Dα−1 )
and Dα−1 has integer entries. Therefore αγα−1 ∈ Γ(N ). A similar argument
shows α−1 γα ∈ Γ(N ). t
u
7.1.5. Let α ∈ GL+

2 (Q) have integer entries. Show that there exists γ ∈ SL2 (Z)
such that
−1 ab
γ α= ,
0d
where a and d are positive integers and b ∈ Z.
Let
r t
α=
su
and suppose that δ = (r, s). Put r1 = r/δ and s1 = s/δ. Since (r1 , s1 ) = 1, we
can find integers x, y such that r1 x + s1 y = 1. Now,

x y r t δ?
=
−s1 r1 su 0?
has the desired form since the determinants of both matrices on the left hand
side of the equation are positive. t
u
7.1.6. Let f ∈ Mk (Γ) with Γ a congruence subgroup of SL2 (Z). Let α ∈ GL+
2 (Q).
Show that f |α is holomorphic at every cusp s ∈ Q ∪ {i∞}.
A typical cusp s ∈ Q∪{i∞} is of the form s = β(i∞) for some β ∈ SL2 (Z).
We must show that for every β ∈ SL2 (Z), the function (f |α)|β = f |(αβ) is
holomorphic at i∞.
From (4.2) on page 36, we see that αβ can be multiplied by a positive
scalar without affecting f |αβ. Hence we may suppose that αβ has integer
entries. By Exercise 7.1.5, we can find γ ∈ SL2 (Z) such that

ab
αβ = γ ,
0d
7.1 The Hilbert space of cusp forms 233
with a, d positive integers and b ∈ Z. Now

ab ab
f |αβ = f |γ = (f |γ) .
0d 0d
Since f ∈ M (Γ) is holomorphic at every cusp, f |γ has a Fourier expansion

Pk ∞
(f |γ)(z) = n=0 an q n/h . Consequently,
∞
!
ab X
n/h ab
(f |γ) = an q
0d 0d
n=0
∞
a k/2 X
= an e2πin(az+b)/dh
d n=0
∞
a k/2 X
= an e2πib/dh q an/dh
d n=0
is holomorphic. t
u
7.1.7. Let f ∈ Sk (Γ) with Γ a congruence subgroup of SL2 (Z). Let α ∈ GL+
2 (Q).
Show that f |α vanishes at every cusp s ∈ Q ∪ {i∞}.
This solution is identical to the previous solution, except that all Fourier
expansions now start at n = 1 instead of n = 0. t
u
7.1.9. Let Γ0 be a subgroup of GL+ 0

2 (R) and F a fundamental domain for the action
0 + −1 0
of Γ on H. If α ∈ GL2 (R), show that α F is a fundamental domain for the
action of α−1 Γ0 α on H.
It suffices to show that every point in H is (α−1 Γ0 α)-equivalent to some

point in α−1 F 0 and that no two interior points of α−1 F are (α−1 Γ0 α)-
equivalent since the map z 7→ α−1 z is an analytic map. Indeed, if z ∈ H,
then αz ∈ H, and there is a γ ∈ Γ0 such that γ(αz) = w ∈ F 0 since F 0 is
a fundamental domain for Γ0 . Hence (α−1 γα)z = α−1 w, which means that
every point in H is (α−1 Γ0 α)-equivalent to a point in α−1 F 0 .
Now, if w1 , w2 ∈ α−1 F 0 are (α−1 Γ0 α)-equivalent, then there is a γ1 ∈ Γ0
such that (α−1 γ1 α)w1 = w2 . Since wi ∈ α−1 F 0 for i = 1, 2, we can write
wi = α−1 vi with vi ∈ F 0 . Thus, α−1 γ1 v1 = α−1 v2 which implies v1 is Γ0
equivalent to v2 . This means that v1 , v2 lie on the boundary of F 0 , and so w1
and w2 also lie on the boundary of α−1 F 0 . t
u
7.1.11. Let α ∈ GL+ 0

2 (Q) have integer entries and determinant D. Let α = Dα
−1
.
Show that for f, g ∈ Sk (Γ), with Γ a congruence subgroup of SL2 (Z), we have
(f |α, g) = (f, g|α0 ).
From (4.2) on page 36, we see that g|α0 = g|α−1 . Thus, we need only
replace g with g|α−1 in Theorem 7.1.10 to deduce the result. t
u
7.2 Commutativity of the Hecke operators
7.2.3. Let p be prime. Show that for all r, s ≥ 1, we have
Tp r Tp s = Tp s Tp r .
Without loss of generality, r ≤ s. We induct on s. For the base case s = 1,

there is nothing to show. Assume Tpr Tps = Tps Tpr whenever 1 ≤ r ≤ s ≤ t.
We consider Tpr Tpt+1 where 1 ≤ r ≤ t+1. If r = t+1, then there is nothing to
show. Otherwise, r ≤ t and by Theorem 7.2.2 and the inductive hypothesis,
we have
Tpr Tpt+1 = Tpr Tpt Tp − pk−1 Tpt = Tpt Tp − pk−1 Tpt Tpr = Tpt+1 Tpr . t

u
7.2.4. Let p be prime. Show that there is a polynomial Ps (x) ∈ Z[x] such that
Tps = Ps (Tp ).
For s = 0 and s = 1, we take P0 (x) = 1 and P1 (x) = x, respectively. We

proceed by induction on s. From Theorem 7.2.2, we have
Tps+1 = Tps Tp − pk−1 Tps−1

= Ps (Tp )Tp − pk−1 Ps−1 (Tp ),
which is a polynomial in Tp with integer coefficients by the induction hy-

pothesis. t
u
7.2.5. Prove that for all m, n ≥ 1, the Hecke operators commute:
Tn Tm = Tm Tn .
Take prime power decompositions n = ps11 · · · psaa and m = q1t1 · · · qbtb so

that by Theorem 7.2.1 we have
Tn Tm = Tps11 · · · Tpsaa Tqt1 · · · Tqtb .

1 b
When pi and qj are distinct primes, Tpsi i Tqtj = Tqtj Tpsi i by Theorem 7.2.1.
j j
When pi = qj , we have Tpsi i Tqtj = Tqtj Tpsi i by Exercise 7.2.3. Hence
j j
Tn Tm = Tqt1 · · · Tqtb Tps11 · · · Tpsaa = Tm Tn ,

1 b
as desired. t
u
7.4 Basis of eigenforms 235
7.3 Hecke operators as Hermitian operators

P∞ n
7.3.3. Suppose f = n=1 an q ∈ Sk (SL2 (Z)) is an eigenfunction for all Tm .
Prove that a1 6= 0 and that the eigenvalue of Tm is am /a1 .
Suppose Tm (f ) = λm f . By Theorem 5.2.4, the coefficient of q 1 in Tm (f )

is am . Hence am = λm a1 . If a1 = 0, then am = 0 for all m and f = 0,
contradicting the fact that f is an eigenvector. Thus a1 6= 0. t
u
7.3.4. Let f , g ∈ Sk (SL2 (Z)) be normalized eigenfunctions for all the Hecke oper-
ators. Show that f = g or (f, g) = 0.
P∞ P∞
Write f (z) = n=1 an q n and g(z) = n=1 bn q n . Since f and g are nor-
malized, the Fourier coefficients are the eigenvalues of the Tn ’s. Thus,
an (f, g) = (Tn (f ), g) = (f, Tn (g)) = (f, bn (g)) = bn (f, g).
The bn ’s are real by Corollary 7.3.2 so that if (f, g) 6= 0, then an = bn = bn for

all n. Hence f = g. t
u
7.4 Basis of eigenforms
7.4.3. Show that Ek is an eigenfunction for all the Hecke operators acting on
Mk (SL2 (Z)).
It suffices to show this for the Tp ’s. Moreover, by linearity P it suffices to

∞
show that f =P −B k
E
2k k is an eigenfunction of T p . We write f = n=0 a(n)q
n
∞
so that Tp f = n=0 b(n)q n where
X np
b(n) = dk−1 a 2 . (7.1)
d
d|(n,p)
Notice that
b(0) = pk−1 a(0) + a(0) = pk−1 + 1 a(0),

and so, by comparing the constant terms, we see that the only possible eigen-
value is pk−1 + 1. Thus, we need to check Tp (f ) = (pk−1 + 1)f , but this is
directly verified by comparing the Fourier coefficients of both sides and us-
ing the multiplicativity of a(n) = σk−1 (n) as follows.
If (n, p) = 1, then by (7.1)
b(n) = a(np) = a(p)a(n) = pk−1 + 1 a(n).

If (n, p) = p, then write n = pt m where (m, p) = 1. Now (7.1) gives

b(n) = pk−1 a(pt−1 m) + a(pt+1 m)

= pk−1 a(pt−1 ) + a(pt+1 ) a(m).
Pt+1
Since a(pt+1 ) = pr(k−1) = p(t+1)(k−1) + a(pt ), we have
r=0

b(n) = pk−1 a(pt−1 ) + p(k−1) pt(k−1) + a(pt ) a(m)

= pk−1 a(pt ) + a(pt ) a(m)
= pk−1 + 1 a(pt m)

= pk−1 + 1 a(n),

as desired. t
u
7.4.4. Show that (Ek , f ) = 0 for any f ∈ Sk (SL2 (Z)).
It suffices to prove this assertion when f is a Hecke eigenform since
Sk (SL2 (Z)) has a basis of eigenforms by Theorem 7.4.2. Observe that Theo-
rem 7.3.1 actually holds when one of the forms is a cusp form and the other
is non-vanishing at i∞. Thus we have
(pk−1 + 1)(Ek , f ) = (Tp (Ek ), f ) = (Ek , Tp (f )) = λp (Ek , f ),
where Tp (f ) = λp f . If (Ek , f ) 6= 0, we have pk−1 + 1 = λp for all primes p.

But by Theorem 5.3.4 , λp = O pk/2 and this is a contradiction. t
u
7.4.5. Show that ∆k is not a Hecke eigenform for any k ≥ 2.
The coefficient of q in the q-expansion of ∆k is zero if k ≥ 2. By Exercise
7.3.3, ∆k is not an eigenform. t
u
7.5.1. Let 12 | k and d = dim Sk (SL2 (Z)). For each 1 ≤ j ≤ d, define

∞
2(d−j)
X
fj := ∆j E6 = a(j) n
n q .
n=1
(j) (j)
Verify that an = 0 for n < j and aj = 1. Conclude that the fj form a basis for
Sk (SL2 (Z)).
(j)
From the q-expansion ∆j = q j + · · · , it is evident that an = 0 for n < j
(j)
and aj = 1. From this, we immediately see that the fj ’s are linearly inde-
pendent over C. Moreover, the weight of fj is 12d which equals k by Exer-
cise 4.4.3. t
u
7.5.2. Let k ≥ 4 be even and d = dim Sk (SL2 (Z)). Choose non-negative integers
a, b such that 12 6= 4a + 6b ≤ 14 and 4a + 6b ≡ k (mod 12). For each 1 ≤ j ≤ d,
define
∞
2(d−j)+b a
X
fj := ∆j E6 E4 = a(j) n
n q .
n=1
(j) (j)
Verify that an = 0 for n < j and = 1. Conclude that the fj form a basis for
aj
Sk (SL2 (Z)). (This basis is called the Miller basis in the literature.)
As in the previous exercise, the assertion about the vanishing of the

Fourier coefficients is immediate. This implies that the fj ’s are linearly in-
dependent. Moreover, the weight of fj is 12d + 4a + 6b which is k, by an
application of Exercise 4.4.3. t
u
7.5.3. Let k, d, a, and b be as in Exercise 7.5.2. Verify that there exists a basis g0 ,
g1 , . . . , gd for Mk (SL2 (Z)) such that
∞
X
gi = q i + b(i)
n q
n
n=d+1
(i)
with bn ∈ Z for all n and all i. Conclude that if
∞
X
n=0
has c0 , c1 , . . . , cd ∈ Z, then cn ∈ Z for all n.
It is straightforward to use linear algebra to explicitly produce g1 , g2 , . . . ,

gd from the basis of fi from Exercise 7.5.2 as follows. First of all, it is clear
that all the fj have integral coefficients and so we may set gd = fd . Then we
use gd to clear out the coefficient of q d in fd−1 by setting
(d−1)
gd−1 := fd−1 − ad gd

(d−1) d (d−1)
= q d−1 + ad q + O q d+1 − ad q d + O q d+1

= q d−1 + O q d+1 .

In general, we recursively set

j−1
(d−j)
X
gd−j = fd−j − ad−s gd−s .
s=0
In this way we produce g1 , g2 , . . . , gd . To create g0 , observe that E62d+b E4a =

1 + O(q) ∈ Mk (SL2 (Z)) has integral coefficients. Use the gi to clear out the
coefficients of q, q 2 , . . . , q d as before.
P Now since we have a basis for Mk (SL2 (Z)), any modular form f =
cn q n ∈ Mk (SL2 (Z)) can be expressed as a linear combination of g0 , . . . ,
gd . In fact, the weights are exactly the initial coefficients c0 , . . . , cd . If these
are all integers, then clearly f is an integer linear combination of forms with
integral coefficients. t
u
7.5.4. Let d = dim Sk (SL2 (Z)) and fix a positive integer m. Suppose
∞
X
n=0
is such that for i = 0, 1, . . . , d, we have that ci ∈ Z and that ci ≡ 0 (mod m).

Prove that ci ≡ 0 (mod m) for all i ≥ 0.
This is a straightforward generalization from Exercise 7.5.3. It is clear that
f = c0 g0 + c1 g1 + · · · + cd gd
where the gi are the basis vectors from Exercise 7.5.3. Since m | ci for i = 0,
1, . . . , d, we have that
c cd
0
f =m g0 + · · · gd ∈ mZ[[q]]
m m
as desired. t
u
7.5.5. Let f be a Hecke eigenform for the full modular group. Show that the Fourier
coefficients of f are algebraic integers.
We show that the eigenvalues of the nth Hecke operator Tn are alge-
braic integers. Indeed, let f1 , . . . , fd be the Miller basis constructed in Exer-
cise 7.5.2. The fj ’s have Z-coefficients and Tn (fj ) can be written as a Z-linear
combination of the fj ’s. Thus, the matrix representing Tn with respect to the
Miller basis has Z-coefficients. Consequently, the characteristic polynomial
of Tn is monic with coefficients from Z, so the eigenvalues of Tn are alge-
braic integers. t
u
Chapter 8
Hecke Operators of Higher Level
8.1 Hecke operators for congruence subgroups
8.1.1. Let p be prime and Γ = SL2 (Z). Let Xp be the set of matrices with integer
entries of determinant p. Show that

p0
Xp = Γ Γ.
01
Clearly Γ p0 10 Γ ⊆ Xp . To show the reverse inclusion, suppose σ ∈ Xp .

By Theorem 5.2.1, there exists γ1 ∈ Γ such that

ab
σ = γ1
0d
with ad = p and 0 ≤ b ≤ d − 1. If a = p, then d = 1, b = 0 and

p0 10 p0
σ = γ1 ∈Γ Γ.
01 01 01
Otherwise a = 1 and d = p. Observe that

1b 10 1b
=
0p 0p 01
and
10 0 −1 p0 0 1
= ,
0p 1 0 01 −1 0
so that
0 −1 p0 0 1 1b p0
σ = γ1 ∈Γ Γ. t
u
1 0 01 −1 0 01 01
8.1.4. For fixed N , S + , and S × , show that
Xm Xn ⊆ Xmn .
Clearly, the determinant of any element of Xm Xn is mn. Moreover, since

a 1 b1 a 2 b2 a 1 a 2 + b 1 c 2 a 1 b2 + b 1 d 2
= ,
c 1 d1 c 2 d2 c 1 a 2 + d 1 c 2 c 1 b2 + d 1 d 2
we see that if N |c1 and N |c2 , then N |(c1 a2 + d1 c2 ). Also, a1 , a2 ∈ S × implies

a1 a2 ∈ S × . Thus a1 a2 + b1 c2 ∈ S × because N |c2 . Finally, a1 b2 + b1 d2 ∈ S + .
t
u
8.1.5. Show that

X1 (N, S × , S + )
is a congruence subgroup of SL2 (Z).
By Exercise 8.1.4, we see that X1 is closed under matrix multiplication.

Moreover, X1 clearly contains the identity element and is a subgroup of
SL2 (Z). Finally, if we write S + = M Z, then X1 ⊇ Γ(N M ). t
u
8.2 Nebentypus

8.2.1. Let χ be a Dirichlet character modulo N . For each γ = ac db ∈ Γ0 (N ),
define ψ(γ) = χ(d). Show that ψ(γ1 γ2 ) = ψ(γ1 )ψ(γ2 ).
Suppose ac db , ( rt us ) ∈ Γ0 (N ). Then c ≡ 0 (mod N ) and so

ab ef ? ?
ψ =ψ = χ(cf + dh) = χ(dh) = χ(d)χ(h),
cd gh ? cf + dh
since χ is a homomorphism. t
u
8.2.2. Show that the set of Dirichlet characters modulo N forms a group under
multiplication of order φ(N ), where φ denotes the Euler function.
ByQthe Chinese remainder theorem (Z/N Z)× ∼ α ×

Q
= p (Z/p Z) where
α
N = p p is the unique factorization of N into distinct prime powers. Thus
any Dirichlet character factors as a product of Dirichlet characters modulo
pα for p|N . For p odd, (Z/pα Z)× is a cyclic group of order φ(pα ) (see for ex-
ample Exercise 2.2.4 of [24]). Thus, a Dirichlet character modulo pα is com-
pletely determined once it is defined on a generator and there are φ(pα ) such
Dirichlet characters for each odd prime p.
If p = 2 and α ≥ 3, then (Z/2α Z)× ∼ = Z/2Z × Z/2α−2 Z (see Exercise 2.2.6
of [24]) which has 2 · 2α−2
= φ(2 ) elements. Since (Z/2Z)× and (Z/4Z)× are
α
cyclic of orders 1 and 2 respectively, we see that in all cases there are φ(2α )
Dirichlet characters modulo 2α . tu
8.2 Nebentypus 241
8.2.3. Show that

(
X φ(N ) if d ≡ 1 (mod N )
χ(d) =
χ
0 otherwise
where the sum is over all Dirichlet characters modulo N .
If d = 1, the result is clear by Exercise 8.2.2. Therefore, suppose d 6≡ 1

(mod N ). Then there is a character ψ such that ψ(d) 6= 1. Thus
X X
T = χ(d) = (χψ)(d) = ψ(d)T
χ χ
because as χ ranges over all Dirichlet characters, so does χψ. Hence T = 0

since ψ(d) 6= 1. t
u
8.2.4. If χ 6= 1 is a Dirichlet character modulo N , show that

X
χ(a) = 0.
a∈(Z/N Z)×
Since χ is not trivial, there is a b 6= 1 such that χ(b) 6= 1 with (b, N ) = 1.

Then X X
S= χ(a) = χ(ab) = χ(b)S
a∈(Z/N Z)× a∈(Z/N Z)×
because as a ranges over coprime residue classes, so does ab. Therefore s = 0

since χ(b) 6= 1. t
u
P∞
8.2.6. Let f ∈ Mk (Γ0 (N ), χ). If f (z) = m=0 λ(m)q m , then prove that
 
X∞ X mn
Tn (f ) =  χ(a)ak−1 λ 2
 qm .
m=0
a
a|m,n
We proceed exactly as in the proof of Theorem 5.2.4 to derive this result.

t
u
P∞ m
8.2.7. If f = m=0 λ(m)q ∈ Mk (Γ0 (N ), χ) is an eigenfunction for all Tn ’s,
show that λ(1) 6= 0 and λ(mn)λ(1) = λ(m)λ(n) for (m, n) = 1.
Let Tn (f ) = µn f . Then by Exercise 8.2.6 when (m, n) = 1 the coefficient

of q m is λ(mn) = µn λ(m). Putting m = 1, we get λ(n) = µn λ(1). Plugging
this into the previous equation, we get λ(mn)λ(1) = λ(m)λ(n) as desired.
If λ(1) = 0, then all coefficients λ(n) = 0, contrary to the definition of an
eigenvector. t
u
8.2.8. For p|N , and f ∈ Mk (Γ0 (N ), χ), show that

p−1
k
2 −1
X 1b
Tp (f ) = p f .
0p
b=0
In the sum defining Tn , we see from Equation (8.3) on page 100 that this
is immediate. t
u
P∞ ∞
8.2.9. Let f = n=0 an q n ∈ Mk (Γ0 (N ), χ) and then write Tp (f ) = n=0 bn q n .
P
Show that
bn = apn + χ(p)pk−1 an/p ,
where we interpret an/p = 0 when p - n.
This is clear from Exercise 8.2.6. t
u
P∞
8.2.10. Let f = m=0 λ(m)q m ∈ Mk (Γ0 (N ), χ) be a normalized eigenfunction
for all the Tn ’s. Show that for p prime,
λ(pa+1 ) = λ(p)λ(pa ) − χ(p)pk−1 λ(pa−1 )
for a ≥ 1.
This again is immediate from Exercise 8.2.6 and the solution to Exer-
cise 8.2.7. We have Tn (f ) = λ(n)f . When n = pa , we compare coefficients of
q p of both sides. On the right hand side, we have λ(pa )λ(p). On the left hand
side, we have λ(pa+1 ) + χ(p)pk−1 λ(pa−1 ). t
u
8.2.13. Show that Sk (Γ0 (N ), χ) has a basis of eigenforms for all the Tn with
(n, N ) = 1.
This is immediate from Theorem 7.4.1, Corollary 8.2.12, and the fact that
the algebra of the Tn ’s is a commutative algebra. t
u
8.3 Oldforms and newforms
8.3.2. Show that ∆(z) and ∆(2z) are both eigenforms for the Tn ’s (n odd) acting
on S12 (Γ0 (2)), with the same eigenvalues.
This is clear from Example 8.3.1 applied to k = 12, χ = 1 and N = 2. t
u
8.3.3. Show that ∆(z) and ∆(2z) are linearly independent in S12 (Γ0 (2)).
We only need to look at the q-expansions:
∆(z) = q − 24q 2 + · · ·
∆(2z) = q 2 − 24q 4 + · · ·
which are clearly linearly independent. t

u
0 −1 −1

8.4.1. Let wN = N 0 . Prove that wN Γ0 (N )wN ⊂ Γ0 (N ).

d −c/N
−1
Let ac db ∈ Γ0 (N ). Then wN ac db wN = −bN a
∈ Γ0 (N ). t
u
8.4.2. Show that if f ∈ Mk (Γ0 (N )), then f |wN ∈ Mk (Γ0 (N )).

If γ = ac db ∈ Γ0 (N ), then by Exercise 8.4.1, we have

d −c/N
(f |wN )|γ = f |wN = f |wN .
−bN a
The form f |wN is holomorphic at the cusps by Exercise 7.1.6. t

u
8.4.3. Show that if f ∈ Sk (Γ0 (N )), then f |wN ∈ Sk (Γ0 (N )).
The solution is identical to that of Exercise 8.4.2 except that now we ap-
peal to Exercise 7.1.7 to deduce that f |wN vanishes at every cusp. t
u
8.4.4. Show that if f ∈ Mk (Γ0 (N )), then (f |wN ) |wN = f . Thus |wN is, in fact,
an involution.
= −N 0
2

We compute wN 0 −N . By Equation (4.2) on page 36, we see that
2
f |wN = f. t
u
Chapter 9
Dirichlet Series and Modular Forms
9.1 General Dirichlet series
9.1.1. Show that the series (9.3) on page 106 converges absolutely for Re(s) > 1.
Writing s = σ + it, with σ > 1, and t ∈ R, we have
∞ ∞
X 1 X 1
s
≤ <∞
n=1
n n=1
nσ
for σ > 1, as is easily shown by a simple application of the integral test. t

u
9.1.2. Given a sequence of complex numbers an , we suppose that for any > 0,
there is a constant C() such that |an | ≤ C()n for all natural numbers n. Show
that
∞
X an
n=1
ns
converges absolutely for Re(s) > 1.

Let s = σ + it, with t ∈ R and σ > 1. Choose > 0 so that < σ − 1. Then
∞ ∞
X an X 1
s
≤ C() σ−
< ∞,
n=1
n n=1
n
as in the previous exercise. t

u
9.1.3. Prove the identity (9.4) on page 106 for Re(s) > 1.
Since every natural number can be written as a product of powers of dis-
tinct primes, the result is immediate upon noting that
Y −1 Y
1 1 1
1− s = 1 + s + 2s + · · · ,
p
p p
p p
where the product is over distinct prime numbers p. A typical term in the
Dirichlet series obtained by expanding the product on the right hand side is
1
s.
(pa1 1 · · · pakk )
αk
Since every natural number n can be written uniquely as n = pα 1 · · · pk
1
with distinct prime numbers p1 , . . . , pk , we immediately deduce (9.4) of page

106. t
u
9.1.4. Let f (n) be a multiplicative function. That is, f (mn) = f (m)f (n) for
(m, n) = 1. Assume that for any > 0, there is a constant C() such that
|f (n)| ≤ C()n . Show that for Re(s) > 1, we have
∞
f (p) f (p2 )

X f (n) Y
= 1 + + + · · · ,
n=1
ns p
ps p2s
where the product is over the prime numbers.
By Exercise 9.1.2, the Dirichlet series converges absolutely and so does

each factor of the product. A typical term of the expanded product has the
form
f (pa1 1 ) · · · f (pakk )
s .
(pa1 1 · · · pakk )
αk
If n = pα
1 · · · pk is the unique factorization of the natural number n, then
1
we see that this is f (n)/ns since f is multiplicative. t

u
9.1.6. Suppose that {an }∞n=1 is a sequence of complex numbers and Pf (u) is a con-
tinuously differentiable function on the interval [1, x]. Let A(t) = n≤t an . Prove
that Z x
X
an f (n) = A(x)f (x) − A(t)f 0 (t)dt.
n≤x 1
First suppose that x is a natural number. We apply Abel’s lemma (Propo-

sition 9.1.5) with bn = f (n), m = 1, r = x to get
X X
an f (n) = A(x)f (x) − A(n) f (n + 1) − f (n) .
n≤x n≤x−1
Writing
Z n+1
f (n + 1) − f (n) = f 0 (t)dt,
n
we obtain
X X Z n+1
an f (n) = A(x)f (x) − A(n) f 0 (t)dt.
n≤x n≤x−1 n
Since A(t) is a step function, we see that the latter sum can be written as
X Z n+1 Z x
A(t)f 0 (t)dt = A(t)f 0 (t)dt
n≤x−1 n 1
which proves the result if x is a natural number. If x is not a natural number,

let us write bxc for the greatest integer ≤ x and note that
X X
an f (n) = an f (n).
n≤x n≤bxc
To complete the proof, we need to show that

Z x
A(x) (f (x) − f (bxc)) − A(t)f 0 (t)dt = 0,
bxc
which is evident since A(t) is a step function. t

u
√
9.1.7. Let α > β > 0. Write z = x + iy, with x, y ∈ R, i = −1, and x > 0.
Show that
|z| −αx
e−αz − e−βz ≤ − e−βx .

e
x
We have Z β
−αz −βz
e −e =z e−tz dt
α
so that
Z β
e−αz − e−βz ≤ |z| e−tz dt
α
Z β
= |z| e−tx dt
α
|z| −αx
− e−βx .

= e t
u
x
9.1.9. Suppose that X
A(x) = an = O(xδ )
n≤x
for some δ > 0. Show that

∞ Z ∞
X an A(t)dt
s
=s
n=1
n 1 ts+1
and that the integral converges absolutely for Re(s) > δ.
We apply Exercise 9.1.6 with f (n) = n−s to get

Z x
X
−s A(x) A(t)
an n = +s dt.
xs 1 ts+1
n≤x
For any s with Re(s) > δ, we have
A(x)
lim =0
x→∞ xs
since A(x) = O(xδ ). Thus
∞ Z ∞
X an A(t)dt
s
= s ,
n=1
n 1 ts+1
and the integral converges absolutely for Re(s) > δ. t

u
9.1.10. Show that ∞
{t}
Z
s
ζ(s) = −s dt,
s−1 1 ts+1
where {t} = t − btc and the integral converges absolutely for Re(s) > 0. Deduce
that lims→1+ (s − 1)ζ(s) = 1.
∞
btc
Z
ζ(s) = s dt.
1 ts+1
Since btc = t − {t}, we can rewrite the integral as

Z ∞ Z ∞
t − {t} s {t}
s s+1
dt = − s dt.
1 t s−1 1 ts+1
Since 0 ≤ {t} < 1, the integral converges absolutely for Re(s) > 0. Multiply-
ing by (s − 1) and taking limits as s → 1+ , we deduce
lim (s − 1)ζ(s) = 1. t
u
s→1+
9.1.11. Show that the series

1 1 1 1
1− + s − s + s − ···
2s 3 4 5
has abscissa of convergence σ0 = 0 and abscissa of absolute convergence σ1 = 1.
The fact that σ1 = 1 is clear by the divergence of the harmonic series. To
see that σ0 = 0, we apply Exercise 9.1.9 in which the corresponding sum
A(x) = O(1) so that the series converges for any s with Re(s) > 0. Clearly,
for s = 0, the series does not converge and so σ0 = 0. t
u
9.1.12. Show that

∞
X
n=1
can be identically zero only if all its coefficients an are zero.
Multiplying f (s) by eλ1 s , we get

∞
X
eλ1 s f (s) = a1 + an e−(λn −λ1 )s .
n=2
Now, assuming s is real, let s → +∞. By the uniform convergence, we see

that eλ1 s f (s) tends to a1 . Since f (s) is identically zero, a1 = 0. Proceeding
inductively, we deduce an = 0 for all n. t
u
9.1.14. Prove the identity

∞ n+1
− β n+1
n+1
− δ n+1

X α γ
Tn
n=0
α−β γ−δ
1 − αβγδT 2
= .
(1 − αγT )(1 − αδT )(1 − βγT )(1 − βδT )
The series on the left hand side is easily seen to be

−1 −1 αγ αδ βγ βδ
(α − β) (γ − δ) − − + ,
1 − αγT 1 − αδT 1 − βγT 1 − βδT
which simplifies to the expression on the right hand side upon choosing a
common denominator. This example is due to S. Ramanujan. t
u
9.1.15. A function a : N → C is called an arithmetical function. Given two

arithmetical functions a and b, we define the Dirichlet convolution a ∗ b, to be the
arithmetical function given by
X n
(a ∗ b)(n) = a(d)b
d
d|n
where the sum is over positive divisors d of n. Show that

∞ ∞
! ∞ !
X (a ∗ b)(n) X a(n) X b(n)
= .
n=1
ns n=1
ns n=1
ns
Expanding the product on the right hand side, we obtain

X a(d)b(e)
.
ds e s
d,e|n
Collecting terms with de = n gives the identity. The identity should be

viewed in the ring of formal Dirichlet series so that questions of conver-
gence need not be considered here. t
u
9.1.16. An arithmetical function a : N → C is said to be completely multiplica-
tive if a(mn) = a(m)a(n) for all m, n. Show that
∞ −1
X a(n) Y a(p)
= 1 − .
n=1
ns p
ps
This is evident from Exercise 9.1.4. Indeed,

∞
a(p) a(p2 )

X a(n) Y
= 1 + + + · · · .
n=1
ns p
ps p2s
Since a is completely multiplicative, a(pm ) = a(p)m so that each Euler factor

in the product is a geometric series that is easily summed to give the desired
result. Again, the identity should be viewed in the ring of formal Dirichlet
series. t
u
9.1.17. Given any arithmetical function a : N → C, we will write
∞
X a(n)
L(s, a) := .
n=1
ns
If a, b, c, d are completely multiplicative functions, show that
L(s, ac)L(s, ad)L(s, bc)L(s, bd)

L(s, (a ∗ b)(c ∗ d)) = .
L(2s, abcd)
Here ac denotes the arithmetical function given by (ac)(n) := a(n)c(n).

Since a, b, c, d are completely multiplicative, a ∗ b and c ∗ d are multiplica-
tive. By Exercise 9.1.4, the left hand side has an Euler product whose p-th
Euler factor is given by
∞
a(p)n+1 − b(p)n+1 c(p)n+1 − d(p)n+1
X
p−ns .
n=0
a(p) − b(p) c(p) − d(p)
By Exercise 9.1.14, with T = p−s , we see that this simplifies to
1 − a(p)b(p)c(p)d(p)T 2
(1 − a(p)c(p)T )(1 − a(p)d(p)T )(1 − b(p)c(p)T )(1 − b(p)d(p)T )
from which the result is now evident. t

u
9.1.18. Let
X
σu (n) = du .
d|n
Show that for any complex numbers a, b,

∞
X σa (n)σb (n) ζ(s)ζ(s − a)ζ(s − b)ζ(s − a − b)
s
= .
n=1
n ζ(2s − a − b)
The function σu is a Dirichlet convolution of two completely multiplica-

tive arithmetical functions so that the previous exercise is applicable. The
result is now immediate. We leave the details to the student. t
u
9.1.19. Let t0 ∈ R. With a = it0 , b = −it0 in the previous exercise, show that
∞
X (σit0 (n))2 ζ(s)2 ζ(s − it0 )ζ(s + it0 )
= .
n=1
ns ζ(2s)
Using Landau’s theorem (Theorem 9.1.13), show that ζ(1 + it0 ) 6= 0.

The identity is immediate from the previous exercise. If ζ(1 + it0 ) = 0,
then ζ(1 − it0 ) = 0, as can be deduced from the expression for ζ(s) given
in Exercise 9.1.10. Thus, the double pole at s = 1 of ζ(s)2 is canceled by the
double zero of ζ(s − it0 )ζ(s + it0 ) at s = 1. Therefore, the right hand side
of the identity is analytic for Re(s) > 12 . By Landau’s theorem, the Dirichlet
series has a singularity at its abscissa of convergence which must be ≤ 12 .
However, the right hand side has a zero at s = 1/2 (since the denominator
has a pole there). However, the left hand side is ≥ 1 (coming from n = 1) as
s → 1/2+ . This contradiction shows ζ(1 + it0 ) 6= 0. t
u
9.2 The Poisson summation formula
9.2.2. If F ∈ S, show that

Z ∞
tm F (n) (t)dt < ∞
−∞
for all pairs of integers m, n ≥ 0.

Since F ∈ S,
sup tm+2 F (n) (t) < ∞.
t∈R
Thus Z ∞ Z ∞
m (n) dt
t F (t)dt < ∞.
1 1 t2
Similarly,
Z 1
tm F (n) (t)dt < ∞.
−∞
Finally
Z 1
tm F (n) (t)dt < ∞
−1
and this completes the proof. t

u
9.2.3. If F ∈ S, show that F̂ ∈ S.

Since F ∈ S, the integral
Z ∞
F (t)e−2πitx dt
−∞
is easily shown to be absolutely convergent.

Differentiating n times gives
Z ∞
F̂ (n) (x) = (2πit)n F (t)e−2πitx dt.
−∞
Keeping in mind that F ∈ S, we integrate the right hand side by parts m

times to get
Z ∞ m
1 d
F̂ (n) (x) = m m
((2πit)n F (t)) e−2πitx dt.
(2πix) −∞ dt
Thus ∞
dm n
Z
m
x F̂ (n)
(x) ≤ (2π) n−m
(t F (t)) e−2πitx dt .
−∞ dtm
The integral is a sum of terms of the form
Z ∞
tr F (s) (t)e−2πitx dt.
−∞
By Exercise 9.2.2, these integrals are absolutely convergent. This completes

the proof. t
u
9.2.4. Show that Z ∞
2
e−πx dx = 1.
−∞
Let Z ∞
2
I= e−πx dx.
−∞
Then Z ∞ Z ∞
2
+y 2 )
I2 = e−π(x dxdy.
−∞ −∞
Switching to polar co-ordinates, we put x = r cos θ, y = r sin θ, 0 ≤ θ ≤ 2π,

r > 0. We find
Z ∞ Z 2π
2
I2 = e−πr rdθdr
Z0 ∞ 0
2
= e−πr (2πr)dr = 1.
0
Since I > 0, we conclude I = 1. t

u
2 2
9.2.5. For F (x) = e−πx , show that F̂ (u) = e−πu . In other words, F is its own
Fourier transform.
We must prove that

Z ∞
2 2
e−πx e−2πixu dx = e−πu .
−∞
This is equivalent to
Z ∞
2
J(u) := e−π(x+iu) dx = 1.
−∞
For u = 0, this follows from Exercise 9.2.4.

For the general case, we observe that
Z ∞ Z ∞
∂ 2 ∂ −π(x+iu)2
e−π(x+iu) dx = e dx
∂u −∞ −∞ ∂u
Z ∞
2
= 2πi (x + iue−π(x+iu) dx
−∞
Z ∞
∂ −π(x+iu)2
=i e dx
−∞ ∂x
2 ∞
h i
= ie−π(x+iu) = 0.
−∞
Thus J 0 (u) = 0 so that J(u) is constant. Since J(0) = 1, we deduce J(u) = 1,

as desired. t
u
9.2.6. Let F ∈ S and define G(t) = F (At + B) for real constants A, B, with
A 6= 0. Show that AĜ(t) = e2πiBt/A F̂ (t/A).
This is immediate from the definition of the Fourier transform and an

appropriate change of variables. t
u
2
9.2.8. Show that if F (x) = e−(Ax+B) π
with A, B ∈ R and A 6= 0, then
2
AF̂ (x) = e2πiBx/A e−π(x+B) .
This is immediate from Exercises 9.2.5 and 9.2.6. t

u
9.2.9. For any a ∈ R, show that

X 2 X 2
e−π(n+a) /x = x1/2 e−πn x+2πina .
n∈Z n∈Z
2
We apply the Poisson summation formula to the function F (t) = e−π(t+a) /x
.
By Exercise 9.2.8, the Fourier transform of F is
√ 2
xe2πita e−πxt .
Applying the Poisson summation formula gives the desired result. t

u
9.2.10. Show that for x > 0,

X 2 X 2
e−πn x
= x−1/2 e−πn /x
.
n∈Z n∈Z
Put a = 0 in Exercise 9.2.9. t

u
9.2.12. Show that (s − 1)ζ(s) extends to an entire function.
Theorem 9.2.11 shows that
s(s − 1)π −s/2 Γ(s/2)ζ(s)
is entire. Since sΓ(s) = Γ(s + 1), we see that sΓ(s/2) = 2Γ( 2s + 1) is analytic
at s = 0. Therefore ζ(s) is
analytic at s = 0. Since Theorem 9.2.11 gives an
expression for π −s/2 Γ 2s ζ(s) which is valid for all s and as Γ(s) is analytic
at s = 1, we see that ζ(s) has a simple pole at s = 1 which is also evident
from Exercise 9.1.10. t
u
9.2.13. Show that ζ(s) has simple zeros at s = −2n, for each positive integer n.
By Theorem 9.2.11, s

π −s/2 Γ
ζ(s)
2
is analytic for all s ∈ C, s 6= 0, 1. Since Γ(s) has simple poles at s =
0, −1, −2, . . . we deduce that ζ(s) vanishes for s = −2, −4, . . . . t
u
9.2.14. Show that ζ(0) = −1/2.
Since
Z ∞
−s/2 1 1 1 1−s dt
π Γ (s/2) ζ(s) = − + (ϑ(it) − 1)(ts/2 + t 2 ) ,
s−1 s 2 1 t
we multiply this expression by s and then set s = 0 to obtain

s
lim π −s/2 sΓ ζ(s) = −1.
s→0 2
But 2s Γ s s s

2 =Γ 2 + 1 so that lims→0 sΓ 2 = 2. Thus ζ(0) = −1/2. t
u
9.2.15. Let f ∈ S. For x > 0 show that
X
f (nx)
n∈Z
converges absolutely and satisfies

X X n
x f (nx) = fˆ .
x
n∈Z n∈Z
Since f ∈ S, xm f (n) (x) is bounded for all x ∈ R and each choice of inte-
gers m, n ≥ 0. In particular, x2Pf (x) is bounded from which we easily deduce
the absolute convergence of n∈Z f (nx). By Exercise 9.2.6 for the function
G(t) = f (xt), we have xĜ(t) = fˆ(t/x). Applying the Poisson summation
formula to G gives X X
G(n) = Ĝ(n).
n∈Z n∈Z
In other words, X X n

x f (nx) = fˆ ,
x
n∈Z n∈Z
as desired. t
u
9.2.16. Let f ∈ S be an even function. That is, f (x) = f (−x). For any x > 0,
show that
∞ n ∞
X X 1 ˆ
f =x fˆ(nx) + xf (0) − f (0) .
n=1
x n=1
2
By the previous exercise, we have

X X n
x f (nx) = fˆ .
x
n∈Z n∈Z
Since f is even, we have

∞
X X∞ n
xf (0) + 2x f (nx) = fˆ(0) + 2 fˆ .
n=1 n=1
x
Changing x to 1/x and multiplying by x/2, we obtain the desired result after
rearranging terms. t
u
R∞
9.2.17. Let f ∈ S. Set ζ(s, f ) = 0 f (x)xs dx
x . Show that the integral converges
for all s with Re(s) > 1.
Since f is bounded, the integral

Z 1
dx
f (x)xs
0 x
is convergent. Take an integer m > Re(s) > 1. Then as f ∈ S, |xm f (x)| is

bounded for all x ∈ R. Thus, the integral
Z ∞ Z ∞
dx
f (x)xs xRe(s)−m−1 dx < ∞. t
u
1 x 1
9.2.18. Let f ∈ S be an even function. With ζ(s, f ) defined as in the previous

exercise, show that ζ(s, f ) extends for all s ∈ C, apart from possible simple poles at
s = 0, 1 and satisfies the functional equation
ζ ? (s, f ) = ζ ? (1 − s, fˆ),
where ζ ? (s, f ) = ζ(s)ζ(s, f ).

In the integral definition of ζ(s, f ), we change x to nx:
Z ∞
s dx
ζ(s, f ) = n f (nx)xs .
0 x
By Exercise 9.2.16, the second sum is

∞ n ∞
X X 1 ˆ
f =x fˆ(nx) + xf (0) − f (0)
n=1
x n=1
2
so that
∞ ∞
! !
∞
fˆ(0)
Z X X dx 1 f (0)
ζ(s)ζ(s, f ) = s
f (nx)x + fˆ(nx)x1−s − + .
1 n=1 n=1
x 2 s 1−s
The integral converges absolutely for all s ∈ C and thus defines an entire
function. Moreover, the right hand side is invariant under the maps s →
1 − s, f → fˆ from which the result follows. t
u
c

9.2.19. Show that the symbol d is completely multiplicative in each of the vari-
ables.
This follows easily from the definition and the multiplicativity of the Leg-
endre symbol. t
u
9.2.20. Let d|N and let ψ be a character modulo d. Prove that
Y ψ(p)

L(s, χ0 ψ) = L(s, ψ) 1− s .
p
p|N
We have
Y −1
χ0 (p)ψ(p)
L(s, χ0 ψ) = 1−
p
ps
Y −1
ψ(p)
= 1− s
p
p-N
Y −1 Y
ψ(p) ψ(p)
= 1− s 1− s
p p
p-d p-d
p|N
Y ψ(p)

= L(s, ψ) 1−
ps
p|N
since if p|N and p|d then ψ(p) = 0. t

u
9.2.21. If (n, N ) = 1, show that

N
X mn
χ(n)τ (χ) = χ(m)e ,
m=1
N
where χ denotes the complex conjugate of χ.
We have
N
X m
χ(n)τ (χ) = χ(m)χ(n)e .
m=1
N
Since (n, N ) = 1, let n0 be such that nn0 ≡ 1 (mod N ) so that setting h ≡ mn0
(mod N ), we get
N
X nh
χ(n)τ (χ) = χ(h)e .
N
h=1
Changing h to m in the sum gives the desired expression. t

u
9.2.23. If χ is a primitive character modulo N , then |τ (χ)|2 = N .
By Theorem 9.2.22
N
X mn
χ(n)τ (χ) = χ(m)e
m=1
N
so that
N N
X X n(m1 − m2 )
|χ(n)|2 |τ (χ)|2 = χ(m1 )χ(m2 )e
m1 =1 m2 =1
N
and summing over n, 1 ≤ n ≤ N to deduce
φ(N )|τ (χ)|2 = N φ(N )
using the orthogonality relations (see Exercise 8.2.3). t

u
9.2.26. Prove that

∞ ∞
3/2 X
X 2 N m −πN (n+ m )2 /x
ne−n πx/N +2πimn/N
=i n+ e N .
n=−∞
x n=−∞
N
By Exercise 9.2.9,
∞ ∞
X 2 X 2
e−n πy+2πinα
= y −1/2 e−π(n+α) /y
.
n=−∞ n=−∞
We differentiate both sides with respect to α to get

∞ ∞
X 2 X 2
2πi ne−πn y+2πinα
= −2πy −3/2 (n + α)e−π(n+α) /y
.
n=−∞ n=−∞
Setting y = x/N , and α = m/N gives the result. t

u
9.2.27. Let χ be an odd Dirichlet character modulo N . Let

∞
X 2
ϑ1 (x, χ) = nχ(n)e−πn x/N
.
n=−∞
Show that
Z ∞
−( s+1
2 )
s+1 s+1 s+1 dx
2π Γ N 2 L(s, χ) = ϑ1 (x, χ)x 2 .
2 0 x
In the formula
s Z ∞
−s/2 s/2 −s 2
πx/N s/2 dx
π Γ N n = e−n x ,
2 0 x
we change s to s + 1 to get
Z ∞
− s+1 s+1 s+1 −s 2 s+1 dx
π 2 N 2 Γ n = ne−πn x/N x 2 .
2 0 x
Multiplying by χ(n) and summing over n ≥ 1 gives

Z ∞
− s+1 s+1 s+1 s+1 dx
π 2 N 2 Γ L(s, χ) = ϑ1 (x, χ)x 2 ,
2 0 x
because χ is odd and so

∞ ∞
X 2 X 2
2 nχ(n)e−πn x/N
= nχ(n)e−πn x/N
. t
u
n=1 n=−∞
9.2.28. Prove that if χ is an odd primitive Dirichlet character modulo N , then

1 τ (χ)
ϑ1 , χ = x3/2 ϑ1 (x, χ)
x iN 1/2
By Theorem 9.2.22, we have

N
X mn
τ (χ)χ(n) = χ(m)e .
m=1
N
Thus,
∞ N
!
X X mn 2
τ (χ)ϑ1 (x, χ) = n χ(m)e e−πn x/N
.
n=−∞ m=1
N
This is equal to
N ∞
X X 2
χ(m) ne−πn x/N +2πimn/N
.
m=1 n=−∞
By Exercise 9.2.26, the inner sum is

∞
3/2 X
N m −πN (n+m/N )2 /x
i n+ e ,
x n=−∞
N
so that
N
3/2 X ∞
N X m −πN (n+m/N )2 /x
τ (χ)ϑ1 (x, χ) = i χ(m) n+ e
x m=1 n=−∞
N
1/2 ∞
X
iN 2
= χ(n)ne−πn /N x
x3/2 n=−∞
1/2

iN 1
= ϑ1 ,χ .
x3/2 x
In other words, by changing χ to χ in the equation, we get

1 τ (χ)
ϑ1 , χ = x3/2 ϑ1 (x, χ),
x iN 1/2
as required. t
u
9.2.29. Let χ be an odd primitive Dirichlet character modulo N . Let

s+1
ξ(s, χ) := π −s/2 N s/2 Γ L(s, χ).
2
Then ξ(s, χ) is entire and satisfies the functional equation
ξ(s, χ) = wχ ξ(1 − s, χ),
where wχ = τ (χ)/iN 1/2 .
By Exercise 9.2.27,
Z ∞
s+1
) N s+1 s+1 dx
2π −( 2 2 L(s, χ) = θ1 (x, χ)x 2 .
0 x
This is equal to
Z ∞ Z ∞
s+1 dx 1 s+1 dx
θ1 (x, χ)x 2 + θ1 , χ x− 2
1 x 1 x x
after a suitable change of variables. By the previous exercise,

1
θ1 , χ = wχ x3/2 θ1 (x, χ)
x
so that our expression is

Z ∞ Z ∞
s+1 dx 2−s dx
θ1 (x, χ)x 2 + wχ θ1 (x, χ)x 2
1 x 1 x
Z ∞ Z ∞
s/2 dx 1−s dx
= θ1 (x, χ)x √ + wχ θ1 (x, χ)x 2 √ .
1 x 1 x
Both integrals are absolutely convergent for all s ∈ C. The functional equa-
tion is easily deduced by noting that changing s to 1 − s, changing χ to χ,
and multiplying by wχ , the expression remains invariant. t
u
9.3 L-functions attached to modular forms

0 −1
9.3.2. Let f ∈ Mk (Γ0 (N ), χ) and g = f |w with w = . Supposing that
N 0
∞
X
f (z) = an e2πinz
n=0
X∞
g(z) = bn e2πinz
n=0
are the Fourier expansions of f and g at i∞, prove that

√ !s
N
Λf (s) := Γ(s)Lf (s)
2π
√ !s
N
Λg (s) := Γ(s)Lg (s),
2π
P∞ P∞
where Lf (s) = n=1 anns and Lg (s) = n=1 nbns extend analytically to the entire
complex plane with poles at s = 0 and s = k. Moreover,
a0 b0 i k
Λf (s) + +
s k−s
and
b0 a0 ik
Λg (s) + +
s k−s
are entire and satisfy the functional equation
Λf (s) = ik Λg (k − s).
As in the proof of Theorem 9.3.1 and Theorem 9.2.1, we have the easily
verified equality:
Z ∞
b0 i k

a0 it
Λf (s) + + = f √ − a0 ts−1 dt
s k−s 1 N
Z ∞
it
+ ik g √ − b0 tk−s−1 dt.
1 N
Both integrals converge for all s ∈ C since the an ’s and bn ’s have polynomial
growth. Similarly,
Z ∞
a0 i−k

b0 it
Λg (s) + + = g √ − b0 ts−1 dt
s k−s 1 N
Z ∞
−k it
+i f √ − a0 tk−s−1 dt.
1 N
If we change s to k − s in the integral expression we deduce the desired

functional equation, since the polar part also satisfies the same functional
equation. t
u
9.3.3. With f , g as in the previous exercise, show that Lf (0) = −a0 , Lg (0) = −b0 ,
k
3πi
lim (s − k)Lf (s) = b0 √ /Γ(k)
s→k N
and k
2πi
lim (s − k)Lg (s) = a0 √ /Γ(k).
s→k N
Since Γ(s) has a simple pole at s = 0 with residue 1,
√ !s
N
2π
has a simple pole at s = 0 with residue Lf (0). By Exercise 9.3.2, the expres-
sion
a0 b0 i k
Λf (s) + +
s k−s
is entire. Therefore Lf (0) + a0 = 0. Similarly Lg (0) = −b0 . From the same
expression, Λf (s) has a simple pole at s = k with residue b0 ik . Thus
lim (s − k)Λf (s) = b0 ik

s→k
√ k
from which we deduce lims→k (s − k)Lf (s) = b0 2πi/ N Γ(k)−1 , as re-
quired. The result for Lg (s) is similarly derived. t
u
9.3.5. Suppose that f (s) is analytic in the vertical strip defined by a ≤ Re(s) ≤ b
and that for some α ≥ 1, α
f (s) = O e|t|

as |t| → ∞. If for some A> 0, f (s) = O |t|A on Re(s) = a and Re(s) = b, then
show that f (s) = O |t|A for all s in the vertical strip a ≤ Re(s) ≤ b.
We apply the Phragmén-Lindelöf theorem to the function g(s) = f (s)/(s−
u)A where u > b. Then g is anlytic in the strip and bounded on Re(s) = a
and Re(s) = b. By Theorem 9.3.4, g(s) is bounded throughout the vertical
strip which completes the proof. t
u
9.3.6. Show that for any σ > 0,
Z σ+i∞
1
e−x = Γ(s)x−s ds.
2πi σ−i∞
We consider the oriented rectangular contour CT given by vertices σ − iT ,

σ + iT , −U + iT , −U − iT , where U is positive and not an integer and
T > 0. Since the Γ-function has simple poles at s = 0, −1, −2, . . . and
Ress=−j Γ(s) = (−1)j /j!, we find by Cauchy’s Theorem that
1
Z X (−1)j
Γ(s)x−s ds = xj .
2πi CT j!
0≤j<U
By Stirling’s formula, the contribution of the horizontal integrals goes to 0

as T → ∞. Letting U → ∞, we obtain
σ+i∞ ∞
(−1)j xj
Z
1 X
Γ(s)x ds =j
= e−x . t
u
2πi σ−i∞ j=0
j!
9.3.7. With f as in Exercise 9.3.2, show that for σ > 0 we have

Z σ+i∞
iy 1
f √ − a0 = Λf (s)y −s ds.
N 2πi σ−i∞
The integral is
Z σ+i∞
√ !s
1 N
Γ(s)Lf (s)y −s ds
2πi σ−i∞ 2π
∞ −s !
Z σ+i∞
X 1 2πny
= an Γ(s) √ ds
n=1
2πi σ−i∞ N
X∞ √
= an e−2πny/ N
n=1

iy
=f √ − a0 . t
u
N
9.4 Twists of L-series attached to modular forms
9.4.1. Verify the matrix identity

1 u/r ab
=
0 1 cd 0 1 c d − cd2 u/r
where ad − bc = 1.
This is straightforward matrix multiplication. t

u
9.4.3. Let (q, r) = 1 and (u, r) = 1. Let d and w be integers such that dr−quw = 1.
Verify the matrix identity with N = qr2 ,

1 u/r 0 −1 0 −1 r −w 1 w/r
=r .
0 1 N 0 q 0 −qu d 0 1
−1
The left hand side is uN/r

N 0
. Using dr − quw = 1, we see that the right
hand side is
2
0 −1 r 0 0 −1 r 0 qru −1
r = =
q 0 −qu 1/r q 0 −qru 1 qr2 0
which is the left hand side since N = qr2 . t

u
9.4.4. Let
0 −1 0 −1
wN = , wq = ,
N 0 q 0
and f ∈ Sk (Γ0 (q), χ) where χ is a Dirichlet character modulo q. Let ψ be a prim-
itive Dirichlet character modulo r with (q, r) = 1. With fψ as in Theorem 9.4.2,
N = qr2 , g = f |wq , show that
fψ |wN = w(ψ)gψ
where
w(ψ) = χ(r)ψ(q)τ (ψ)2 /r.
By a simple modification of Exercise 8.4.3, we see that g ∈ Sk (Γ0 (q), χ).
Since (q, r) = 1, we have for any (u, r) = 1, integers d and w such that
dr − quw = 1. By Exercise 9.4.3, with N = qr2 , we have the identity

1 u/r 0 −1 0 −1 r −w 1 w/r
=r
0 1 N 0 q 0 −qu d 0 1
so that

1 u/r 1 w/r 1 w/r
f wN = χ(d)g = χ(r)g .
0 1 0 1 0 1
By (9.5) on page 129, since ψ is primitive,

X 1 u/r
τ (ψ)fψ |wN = ψ(u)f wN
0 1
u (mod r)

X 1 w/r
= ψ(u)χ(r)g
0 1
u (mod r)
where quw ≡ −1 (mod r). As u ranges over coprime residue classes modulo
r, so does w. Since quw ≡ −1 (mod r), we have ψ(quw) = ψ(−1) so that
ψ(u) = ψ(−q)ψ(w). Thus

X 1 w/r
τ (ψ)fψ |wN = ψ(−q)χ(r) ψ(w)g
0 1
w (mod r)
= ψ(−q)χ(r)τ (ψ)gψ .
It is easy to see that (see Exercise 9.5.1)
τ (ψ) = ψ(−1)τ (ψ).
Together with Exercise 9.2.23, we obtain
τ (ψ)fψ |wN = ψ(q)χ(r)τ (ψ)gψ
so that
rfψ |wN = ψ(q)χ(r)τ (ψ)2 gψ
u
9.5.1. If τ (χ) denotes the Gauss sum, prove that
τ (χ) = χ(−1)τ (χ).
We have
X a
τ (χ) = χ(a)e −
N
a (mod N )

X b
= χ(−b)e
N
b (mod N )
= χ(−1)τ (χ). t
u
9.5.2. For Re(c) > 0, let F (x) = e−c|x| . Show that

2c
F̂ (u) = .
c2 + 4π 2 u2
We have
Z ∞
F̂ (u) = e−c|x| e−2πiux dx
−∞
Z ∞ Z 0
= e−(c+2πiu)x dx + e(c−2πiu)x dx
0 −∞
Z ∞ Z ∞
= e−(c+2πiu)x dx + e−(c−2πiu)x dx.
0 0
Since Z ∞
1
e−vx dx = ,
0 v
for Re(v) > 0, we get
1 1 2c
F̂ (u) = + = 2 ,
c + 2πi c − 2πiu c + 4π 2 u2
as required. t
u
9.5.3. Show that for Re(c) > 0,

∞
ec + 1 X 2c
= .
ec − 1 n=−∞ c2 + 4π 2 n2
Using Exercise 9.5.2, we apply the Poisson summation formula to get

X X 2c
e−c|n| = .
c2 + 4π 2 n2
n∈Z n∈Z
The left hand side is essentially a geometric series and easily summed to be
∞
X 2e−c
1+2 e−cn = 1 +
n=1
1 − e−c
2 ec + 1
=1+ = c . t
u
ec −1 e −1
9.5.4. From the previous exercise, deduce Euler’s formula that
∞
X 1 π2
2
= .
n=1
n 6
From the previous exercise, we have

∞
ec + 1 1 1 X 1
· − = 2 .
ec − 1 2c c2 n=1
c 2 + 4π 2 n2
If we set c = 0 in the right hand side, we get

∞
1 X 1
.
2π 2 n=1 n2
On the left hand side, we have

c
e +1 1 1 1
lim · − 2 =
c→0 ec − 1 2c c 12
by an easy computation. For example, invert the power series expansion of

ec − 1 at c = 0 to get

1 1 1 c 2

= − + +O c
ec − 1 c 2 12
and simplify the expression in the limit above. t

u
9.5.5. By differentiating the identity in Exercise 9.5.3, deduce that

∞
X 1
2k
∈ π 2k Q.
n=1
n
By Exercise 9.5.3,
∞
2 1 1 X 1
1+ c − 2 =2 2 + 4π 2 n2
.
e −1 2c c n=1
c
The left hand side is easily seen to be a power series in c with rational coef-
ficients. Differentiating both sides k times and then setting c = 0 gives the
desired result. t
u
9.5.6. For Re(c) > 0, and 0 ≤ v < 1, show that

X e2πinv 1

2 cosh cv

−cv
= e + c .
c2 + 4π 2 n2 2c e −1
n∈Z
From the proof of Theorem 9.2.7, we can see that for any Schwartz func-
tion F , we have X X
F (n + v) = F̂ (n)e2πinv .
n∈Z n∈Z
Applying this generalized Poisson summation formula to F (x) = e−c|x| , we

have X X 2c
e−c|n+v| = e2πinv .
c2 + 4π 2 n2
n∈Z n∈Z
The sum on the left hand side is summed easily as

∞
X ∞
X
e−cv + e−c(n+v) + ec(−n+v)
n=1 n=1
since 0 ≤ v < 1. The two sums above are geometric sums and the result is
1
e−cv + ecv + e−cv

ec −1
from which the result is immediate. t
u
9.5.7. Prove that for 0 ≤ v < 1,

X e2πinv
1

2
=2 v −v+ π2 .
n2 6
n6=0
By Exercise 9.5.6,
1 X e2πinv 1

2 cosh cv

−cv
2
+ 2 2 2
= e + c .
c c + 4π n 2c e −1
n6=0
We need to compute
e−cv

1 cosh cv
lim − 2+ .
c→0 2c c e(ec − 1)
Using the fact that
c2 v 2 c4 v 4
cosh cv = 1 + + + ···
2! 4!
and that
1 1 c 1
= 1− + c2 + · · ·
ec − 1 c 2 12
we see that the limit in question is
−cv
1 2 c2 v 2

e 1 1 c
lim − 2 + 2 1− + c + + ···
c→0 2c c c 2 12 2
c2 v 2 v2

1 1 1
= lim 1 − cv + + ··· − + +
c→0 2c 2 2c 12 2

1 1
= v2 − v + ,
2 6
as required. t
u
Chapter 10
Special Topics
10.1 Elliptic functions
10.1.1. If f is a holomorphic elliptic function, then f is constant.
This is a consequence of Liouville’s theorem which says that any bounded

entire function is constant. Since f is holomorphic and completely deter-
mined by its values on the compact region P , we deduce that f is bounded
and hence constant. t
u
10.1.2. If two elliptic functions (for a fixed lattice L) have the same poles with the
same principal parts, show that they differ by a constant.
The difference of the two elliptic functions is again elliptic and it is clearly
holomorphic. By the previous exercise, this is constant. t
u
10.1.3. If f is an elliptic function with no poles on the boundary of the fundamental

parallelogram P , show that the sum of the residues of f in P is zero.
This is an application of Cauchy’s Residue Theorem. The sum in question

is Z
1
f (z)dz,
2πi ∂P
where ∂P is the oriented boundary of P , traversing from 0 to ω1 , then to
ω1 + ω2 , then to ω2 and then back to 0. But this integral is the sum of four
line integrals, each corresponding to a side of the parallelogram P . The line
integrals corresponding to opposite sides cancel because of the double peri-
odicity of f and because of the oriented direction of the sides of ∂P . t
u
10.1.4. Let f be an elliptic function with respect to a lattice L. Show that there is an
α ∈ C such that f is holomorphic on the boundary of the translate α + P , where P
is the fundamental parallelogram of L.
Any meromorphic function has only a finite number of poles in a bounded

subset of C. Thus if f has poles on the boundary of P , there are only finitely
many. Thus we can find α such that f has no poles on the boundary of α+P .
t
u
10.1.5. Show that there is no non-constant elliptic function having exactly one sim-
ple pole in the fundamental parallelogram P .
Suppose there is such a function f . By Exercise 10.1.4, we can find α such
that the translate α+P has no poles of f on the boundary. By Exercise 10.1.3,
the sum of the residues of f is zero. This means that f is holomorphic and
hence constant, a contradiction. t
u
10.1.6. Let f be an elliptic function. Show that the number of zeros of f in the
fundamental parallelogram P is equal to the number of poles of f (counted with
multiplicity).
If f is elliptic, then so is its derivative f 0 . Thus f 0 /f is elliptic. By choosing
an appropriate translate α+P of P we can ensure there are no zeros or poles
of f on the boundary of α + P . By the same reasoning as in the solution to
Exercise 10.1.3, we see that
f 0 (z)
Z
1
dz = 0.
2πi ∂(α+P ) f (z)
The result now follows immediately from Cauchy’s Argument Principle.

t
u
10.1.7. If f is an elliptic function of order m, then for any c ∈ C, show that the
number of solutions of f (z) = c with z ∈ P is exactly m (counted with multiplici-
ties).
We apply the previous exercise to the elliptic function f (z) − c. t
u
10.1.8. For any s > 2, show that the series
X 1
ωs
ω∈L
ω6=0
is absolutely convergent.
The sum is
X 1 X 1
s = s .
(mω1 + nω2 ) ω2s ω1
mω +n
(m,n)6=(0,0) (m,n)6=(0,0) 2
Since ω1 , ω2 are linearly independent over R, we have ω1 /ω2 6∈ R. Now we

appeal to the solution of Exercise 4.2.1 to deduce that there exists λ > 0 such
that
ω1
+ n ≥ λ m2 + n2 .

m
ω2
Thus,
X 1 1 X 1
s ≤ s/2
ω1 λ (m2 + n2 )
(m,n)6=(0,0) m ω2 +n (m,n)6=(0,0)
and the solution now proceeds as in Exercise 4.2.1. t

u
10.1.9. Show that the sum
X 1 1

− 2
(z − ω)2 ω
06=ω∈L
is absolutely convergent for any z ∈ C \ L.

Now
1 1 2ωz − z 2
− = .
(z − ω)2 ω2 ω 2 (z − ω)2
Let R = |z| and exclude the finite number of ω ∈ L with |ω| ≤ 2R. Then
|ω − z| ≥ |ω| − |z| = |ω| − R > R.
It is now clear that
1 1 R (2|w| + R) −3

2 − ≤ 2 = O |ω| ,
(z − ω) ω2 2
|ω | (|ω| − R)
so that our series converges absolutely by Exercise 10.1.8. t

u
10.1.12. Let X
Gk = ω −k
ω∈L
ω6=0
Prove that
∞
X
℘(z) = z −2 + (k + 1)Gk+2 z k
k=2
k even
is the Laurent expansion of ℘(z) at z = 0.

For |z| < |ω|, we have
h i
−2 −2
(z − ω) − ω −2 = ω −2 (1 − z/ω) − 1
∞ (10.1)
X zk
= (k + 1)
ω k+2
k=1
using the familiar identity

∞
1 X
2 = kxk−1
(1 − x) k=1
−1
obtained easily by differentiating the power series for (1 − x) . We insert
(10.1) into the series defining ℘(z). Reversing the order of summation (which
we can do by the absolute convergence of the series) yields
∞
X
℘(z) = z −2 + (k + 1)Gk+2 z k .
k=1
It is clear that Gk = 0 whenever k is odd since the ω and −ω terms in the

definition of Gk cancel. t
u
10.1.13. With notation as in the previous exercise, show that for z ∈ C \ L,
℘0 (z)2 = 4℘(z)3 − 60G4 ℘(z) − 140G6 .
From the previous exercise we have
℘(z) = z −2 + 3G4 z 2 + 5G6 z 4 + · · ·
and so
℘(z)3 = z −6 + 9G4 z −2 + 15G6 + · · · .
Differentiating we obtain
℘0 (z) = −2z −3 + 6G4 z + 20G6 z 3 + · · ·
and so
℘0 (z)2 = 4z −6 − 24G4 z −2 − 80G6 + · · · .
Hence the elliptic function ℘0 (z)2 −4℘(z)3 +60G4 ℘(z)+140G6 is holomorphic

at z = 0 and vanishes there. By periodicity, this function vanishes for all
z ∈ L. By Exercise 10.1.9, this function is holomorphic for all z ∈ C \ L.
Finally, by Exercise 10.1.1 this function is constant (and in fact zero since it
vanishes at z = 0). t
u
10.1.14. Let L be a lattice and g2 , g3 defined as above. Show that all of the complex
solutions of the equation
y 2 = 4x3 − g2 x − g3

are given by ℘(z), ℘0 (z) where ℘(z) is the Weierstrass ℘-function attached to L
and z ranges over all complex numbers in C \ L.
Let (x, y) ∈ C2 lie on the curve. We want to find a z ∈ C \ L such
that x = ℘(z) and y = ℘0 (z). If the equation ℘(z) = x has no solution,
.
then 1 ℘(z) − x is an elliptic function which is holomorphic. By Exer-
cise 10.1.1, it must be a constant. This is a contradiction since ℘(z) is not a
constant function. Therefore, there must be a z ∈ C \ L such that x = ℘(z).
By Exercise 10.1.13 we deduce that y = ±℘0 (z). As ℘(z) is an even func-
tion and ℘0 (z) is odd, we may adjust the sign of z appropriately to ensure
that (x, y) = ℘(z), ℘0 (z) . t
u
10.1.16. Let L be the lattice spanned

by ω1 , ω2 and let ℘(z) be the associated Weier-
strass function. Show that ℘ ω21 , ℘ ω22 , and ℘ ω1 +ω 2
2
are the roots of the cubic
4x3 − g2 x − g3 with g2 and g3 defined by (10.1).
Since ℘0 (z) is an odd function, ℘0 (ω1 /2) = −℘0 (−ω1 /2). But ℘0 (z) is el-
liptic and so ℘0 (−ω1 /2) = ℘0 (−ω1 /2 + ω1 ) = ℘0 (ω1 /2). Thus ℘0 (ω1 /2) = 0.
Similarly, ℘0 (ω2/2) = 0and ℘0 ((ω1 + ω2 )/2) = 0. By Exercise 10.1.13, we de-
duce that ℘ ω21 , ℘ ω22 , and ℘ ω1 +ω 2
2
are the roots of the associated cubic.
t
u
10.1.17. Suppose that ω1 , ω2 are linearly independent over R and that L is the
latticespanned by ω1 , ω2 . Let ℘(z) be the associated ℘-function. Show that ℘ ω21 ,

℘ ω22 , and ℘ ω1 +ω2

2
are all distinct.
Consider the function
f1 (z) = ℘(z) − ℘(ω1 /2).
Since (by the solution of the previous exercise) ℘0 (ω1 /2) = 0 , we see that
f1 (z) has a double order zero at z = ω1 /2. As ℘(z) has a double order pole at
z = 0 and as there are no other poles in the fundamental parallelogram, this
accounts for all of the zeros of f1 (z) by Exercise 10.1.6. If ℘(ω1 /2) = ℘(ω2 /2),
then z = ω2 /2 would be a zero of f1 (z) and hence ω1 ≡ ω2 (mod L), which
contradicts the linear independence of ω1 and ω2 over R. Thus, ℘(ω1 /2) and
℘(ω2 /2) are distinct.
A similar argument applies for the other 2-division points, namely ℘(ω2 /2)
and ℘((ω1 + ω2 )/2). t
u
10.2 The modular invariant

10.2.2. If ω1 , ω2 are linearly independent over R and ac db ∈ SL2 (Z), show that
the lattice generated by aω1 +bω2 and cω1 +dω2 is the same as the lattice generated
by ω1 and ω2 .
The lattice L generated by ω1 and ω2 consists of all linear combinations

mω1 +nω2 with m, n ∈ Z. The lattice L0 generated by aω1 +bω2 and cω1 +dω2
consists of all linear combinations
x(aω1 + bω2 ) + y(cω1 + dω2 ) = (xa + yc)ω1 + (xb + yd)ω2
with x, y ∈ Z. Clearly L0 ⊆ L. But given any pair of integers m, n we can

find integers x, y such that

ac x m
=
bd y n
since ad − bc = 1. Thus L ⊆ L0 and this completes the proof. t

u
10.2.3. If ω1 , ω2 generate the lattice L and ω10 , ω20 also generate the same lattice
L, show that there is a matrix ac db ∈ SL2 (Z) such that ω10 = aω1 + bω2 and
ω20 = cω1 + dω2 .
Since ω10 , ω20 generate the same lattice as ω1 , ω2 , there are integers a, b, c, d
such that
ω10 = aω1 + bω2

ω20 = cω1 + dω2 .
Similarly there are integers e, f , g, h such that
ω1 = eω10 + f ω20
ω2 = gω10 + hω20 .
Thus 0
ω1 ef ω1 ef ab ω1
= =
ω2 gh ω20 gh cd ω2
and because ω1 , ω2 are linearly independent,

ef ab 10
=
gh cd 01
from which we derive the result. t

u
10.2.4. Let L be the lattice spanned by ω1 , ω2 with associated functions g2 , g3 . If
λ is a non-zero scalar, and L0 is the lattice spanned by λω1 , λω2 , show that the
corresponding elliptic curve for L0 is
y 2 = 4x3 − λ−4 g2 x − λ−6 g3 .
Since g2 = 60G4 and g3 = 140G6 where

X 1
Gk = k
,
(m,n)6=(0,0)
(mω1 + nω2 )
by Exercise 10.1.12 we see that changing ω1 by λω1 and changing ω2 by λω2

alters the sum by a factor of λ−k . t
u
10.2.6. Show that

2E4 (z)E60 (z) − 3E6 (z)E40 (z)
is a cusp form of weight 12 for the full modular group.
Let f (z) = 2E4 (z)E60 (z) − 3E6 (z)E40 (z). Now

az + b
E4 = (cz + d)4 E4 (z)
cz + d
and by differentiating both sides we get

az + b
E40 (cz + d)−2 = 4(cz + d)3 cE4 (z) + (cz + d)4 E40 (z).
cz + d
Similarly,

az + b
E60 (cz + d)−2 = 6(cz + d)5 cE6 (z) + (cz + d)6 E60 (z).
cz + d
Thus,

az + b az + b
6c(cz + d)7 E6 (z) + (cz + d)8 E60 (z)

f = 2E4
cz + d cz + d

az + b
4c(cz + d)5 E4 (z) + (cz + d)6 E40 (z)

− 3E6
cz + d
= (cz + d)12 f (z).
To see that it is a cusp form, we check its behaviour at i∞ and see that from
the q-expansion that it vanishes, since E40 (z) and E60 (z) both vanish there.
t
u
10.2.7. Prove that E42 E6 = E14 . Deduce that

X
m/(mρ + n)13 = 0,
(m,n)6=(0,0)
where ρ is a primitive cube root of unity.
This is easily deduced from the dimension formula (Corollary 4.4.2) or

from Exercise 4.6.7. Indeed, dim M14 (SL2 (Z)) = 1 and comparing the con-
stant terms of E42 E6 and E14 gives the result. For the second part, we see
that E14 (z) has a double zero at z = ρ and so the derivative also vanishes at
z = ρ from which the second assertion follows. t
u
10.2.8. Show that

E14 (z)
j 0 (z) = −2πi .
∆(z)
Since j(z) = E43 (z)/∆(z), we see that
∆0 (z)E43 (z) − 3∆(z)E42 (z)E40 (z)

−j 0 (z) = .
∆(z)
But 1728∆(z) = E43 (z) − E62 (z) so that
1728∆0 (z) = 3E42 (z)E40 (z) − 2E6 (z)E60 (z).
Hence
−1728∆2 j 0 = E42 E6 (2E4 E60 − 3E6 E40 ) .
By Exercise 10.2.7, E42 E6 = E14 . By Exercise 10.2.6, the term in the bracket is
a cusp form of weight 12. By considering the q-expansion, this cusp form is
easily seen to be 2πi(1728)∆ from which the result follows. t
u
10.3 Theta series and lattices
10.3.1. If L = ω1 Z + · · · ωn Z and if ω1? , . . . , ωn? ∈ L? is the dual basis, that is

ωi? (ωj ) = δij , show that
L? = ω1? Z + · · · ωn? Z
First, consider an element of the form c1 ω1? + · · · + cn ωn? where ci ∈ Z. If

d1 ω1 + · · · dn ωn ∈ L with dj ∈ Z, then by linearity we clearly have
n X
X n
(c1 ω1? + · · · + cn ωn? )(d1 ω1 + · · · dn ωn ) = ci dj ωi? (ωj )
i=1 j=1
n X
X n
= ci dj δij
i=1 j=1
Xn
= ci di ∈ Z.
i=1
Thus L? ⊇ ω1? Z + · · · ωn? Z.

Now given any f ? ∈ L? , it suffices to show that we the equality of func-
tions
n
X
f? = f ? (ωj )ωj? .
j=1
since by the definition of the dual lattice, we know f ? (ωj ) ∈ Z for j =

1, . . . , n. It suffices to check this for each basis element ω1 , . . . , ωn . Indeed,
by linearity we have
 
Xn n
X n
X
 f ? (ωj )ωj?  (ωi ) = f ? (ωj )ωj? (ωi ) = f ? (ωj )δij = f ? (ωi ).
j=1 j=1 j=1
Pn
Thus, f ? = j=1 f ? (ωj )ωj? and so L? ⊆ ω1? Z + · · · ωn? Z. t
u
√
10.3.3. Prove that µ(V /L) = det A.
V /L is a fundamental domain for the lattice. By linear algebra (and what

we have said earlier), there is a basis e1 , . . . , en of V which is orthogonal
with respect to (·, ·). In other words, we can write
n
X
ωj = bij ej
j=1
t 2
so that if B = (bij ), the Gram matrix A is √ equal to B B. Thus (det B) =
det A and as det A > 0, we have det B = det A. At the same time, having
fixed our own basis e1 , . . . , en , the lattice spanned by ω1 , . . . , ωn is that lattice
spanned
nP by the columns of o B and the fundamental domain for this lattice
n
is j=1 λj ej : 0 ≤ λj ≤ 1 which has measure µ(V /L). Since this measure
is the volume of the parallelepiped spanned by the columns of B, we are
done. t
u
10.3.4. With A as above, show that the matrix A−1 is the Gram matrix of the dual
lattice L? .
In Exercise 10.3.1, we have already seen that ω1? , . . . , ωn? is a basis of L? .

The Gram matrix of L? is therefore ωi? , ωj? . At the same time, L is
spanned by e1 , . . . , en (using the notation of the previous exercise), and so if
we put e?j = A−1 ej , we see that
t
(e?j , ek ) = A−1 ej Aek
= etj A−1 Aek
= etj ek
= δjk
so that e?1 , . . . , e?n is the dual basis also. But

t
(e?j , e?k ) = A−1 ej A A−1 ek = ej A−1 ek

so that A−1 is the Gram matrix of L? . t

u
10.3.7. If L is self-dual and even, show that 8 | n.

Recall that 11
0 −1
the full modular group SL2 (Z) is generated by T = ( 0 1 ) and
S = 1 0 by Theorem 2.1.2. We first consider how θL (z) transforms under
T . Since L is even, we have
X X
θL (z + 1) = eπi(z+1)(ω,ω) = eπiz(ω,ω) = θL (z),
ω∈L ω∈L
for all z ∈ H. The transformation under S is the content of Theorem 10.3.6.

In particular, since Si = i, we have
θL (i) = µ(V /L)θL (i),
so that µ(V /L) = 1 as θL (i) 6= 0.

The fact that (ST )3 z = z inspires the following sequence of calculations.
Start with
n/2 n/2
Tz z+1
θL (ST z) = θL (T z) = θL (z).
i i
Replace z with ST z to obtain

n/2
ST z + 1
θL (ST )2 z =

θL (ST z)
i
n/2 n/2
ST z + 1 z+1
= θL (z).
i i
Replace z with ST z again to obtain

n/2 n/2
(ST )2 z + 1

ST z + 1
θL (ST )3 z =

θL (ST z)
i i
n/2 n/2 n/2
(ST )2 z + 1

ST z + 1 z+1
= θL (z).
i i i
Since (ST )3 z = z, we clearly have

n/2 n/2 n/2
(ST )2 z + 1

ST z + 1 z+1
1= .
i i i
Now set z = i so that ST z = −1/(i + 1) and (ST )2 z = i − 1. Now the

previous equation becomes in/2 = 1 which implies that 8 | n. t
u
10.3.8. If L is self-dual show that the Gram matrix of L and L? have determinant
1.
Since L = L? , µ(V /L) = µ(V /L? ) so that det A = det A−1 by Exer-
cise 10.3.4. Hence det A = ±1. Since A is positive definite, det A = 1. t
u
10.4 Special values of zeta and L-functions 279
10.3.9. Show that the theta function of a self-dual, even lattice is a modular form of
weight n/2 for the full modular group SL2 (Z).
From the solution to Exercise 10.3.7, we have that µ(V /L) = 1 and that
θL (z + 1) = θL (z). Now since 8 | n by Exercise 10.3.7, the result follows
immediately from Theorem 10.3.6,

1
θL − = z n/2 θL (z),
z
since S and T generate SL2 (Z).
10.4 Special values of zeta and L-functions
10.4.1. Prove that

(−1)k k!
ψk (z + 1) = ψk (z) + .
z k+1
This is immediate from the series expansion of ψk (z). t
u
10.4.2. For every integer k ≥ 0, prove that
dk k!
− (π cot πz) = ψk (z) + (−1)k+1 ψk (−z) + (−1)k k+1
dz k z
From the partial fraction expansion of the cotangent function (see (4.3) in
Section 4.2), we have
∞
1 X 1 1
π cot πz = + + .
z n=1 z−n z+n
The result is now easily deduced by taking successive derivatives. t

u
10.4.3. Show that

dk
− (π cot πz) = ψk (z) + (−1)k+1 ψk (1 − z).
dz k
This is clear from Exercises 10.4.2 and 10.4.1. t
u
10.4.4. Show that

N
X
−k
(k − 1)!L(k, χ) = (−N ) χ(a)ψk−1 (a/N ).
a=1
We may write
N
X X N
X ∞
X
L(k, χ) = χ(a) n−k = χ(a) (N j + a)−k
a=1 n≡a(mod N ) a=1 j=0
from which the desired result is apparent. t

u
10.4.5. If χ is an even character (that is, χ(−1) = 1) and k is even, show that
N
X dk
−2(k − 1)!L(k, χ) = (−N )−k χ(a) (π cot πz) .
a=1
dz k z=a/N
From the previous exercise, we can write

N
X
−k
2(k − 1)!L(k, χ) = (−N ) χ(a)ψk−1 (a/N ) + χ(N − a)ψk−1 (1 − a/N )
a=1
which is
N
X
= (−N )−k χ(a)(ψk−1 (a/N ) + ψk−1 (1 − a/N ).
a=1
By exercise 10.4.3, the result follows. The last assertion is immediate upon
noting that
dk
(π cot πz)
dz k z=a/N
is an algebraic multiple of π k . t
u
10.4.6. If χ is an odd character (that is, χ(−1) = −1) and k is odd, show that
N
X dk
−2(k − 1)!L(k, χ) = (−N )−k χ(a) (π cot πz) .
a=1
dz k z=a/N
As before we can write

N
X
2(k − 1)!L(k, χ) = (−N )−k χ(a)ψk−1 (a/N ) + χ(N − a)ψk−1 (1 − a/N )
a=1
which is now equal to (since χ is odd)

N
X
= (−N )−k χ(a)(ψk−1 (a/N ) − ψk−1 (1 − a/N ).
a=1
By exercise 10.4.3, the result follows. The last assertion is also immediate
upon noting (as before) that
dk
(π cot πz)
dz k z=a/N
is an algebraic multiple of π k . t
u
10.5.1. Let f be an elliptic function with period lattice L. Show that

X
ordz (f )z ∈ L
z∈C/L
We apply Cauchy’s residue formula to obtain that the above sum is
wf 0 (w)
Z
1
dw.
2πi ∂P f (w)
This is a sum of four line integrals corresponding to the oriented (counter-

clockwise) edges of the fundamental parallelogram.
Assuming first that f has no zeros or poles on the boundary of P , we
have four oriented integrals: from 0 to ω1 ; from ω1 to ω1 + ω2 ; from ω1 + ω2
to ω2 ; and from ω2 to 0.
In the second integral, we change variables from w to w − ω1 . In the third
integral, change from w to w − ω2 . Using the periodicity of f 0 /f , this simpli-
fies to
Z ω1 0 Z ω2 0
X ω2 f (w) ω1 f (w)
ordz (f )z = dw + dw.
2πi 0 f (w) 2πi 0 f (w)
z∈C/L
But for any meromorphic function g(z), the integral

b
g 0 (z)
Z
1
dz
2πi a g(z)
is the winding number around zero of the path
γ : [0, 1] → C
given by
γ(t) = g (1 − t)a + tb
and in particular if g(a) = g(b), the integral is an integer. Since f is elliptic,

we indeed have f (ω1 ) = f (0) = f (ω2 ). This completes the proof in the case
that f has no zero or pole on the boundary of P . If it does, we translate P
appropriately and proceed as above. t
u
10.5.2. With ℘ denoting the Weierstrass ℘-function as in Section 10.1, prove that
℘00 (z) = 6℘(z)2 − g2 /2.
This is easily seen upon differentiating the differential equation satisfied

by the Weierstrass ℘-function. t
u
10.5.3. With Gk as in Exercise 10.1.12, prove that for m ≥ 4,

m−2
X
(2m + 1)(m − 3)(2m − 1)G2m = 3 (2j − 1)(2m − 2j − 1)G2j G2m−2j .
j=1
The result is immediate upon using Exercise 10.5.2 and comparing coef-
ficients of the Laurent expansions about z = 0 for the differential equation
satisfied by ℘00 (z). t
u
10.5.4. Show that for k ≥ 2, each G2k can be expressed as a polynomial in G4 and
G6 with positive rational coefficients.
This is immediate from the preceding exercise and mathematical induc-

tion. t
u
10.5.5. Let ϑ(z) be given by the series

∞
X 2
eπin z .
n=−∞
Show that
∞
Y
ϑ(z) = (1 − e2πinz )(1 + e(2n−1)πiz )2 .
n=1
Deduce that ϑ(z) does not vanish in the upper half-plane.
By Jacobi’s triple product identity (Theorem 1.2.1), we have upon setting

q = eπiz and x = 1,
∞
Y
ϑ(z) = (1 − e2(n+1)πiz )(1 + e(2n+1)πiz )2 ,
n=0
from which the result is immediate upon changing variables appropriately.

The non-vanishing of the product shows that ϑ(z) does not vanish in the
upper half-plane. t
u
10.5.6. With η being the Dedekind η-function and ϑ as in the previous exercise,
show that
ϑ(z) = η 2 ((z + 1)/2)/η(z + 1).
By Exercises 5.1.5 and 10.5.5, the result is immediate. t

u
10.5.7. Show that if j is the modular invariant, then j 0 (z) satisfies

az + b ab
j0 = (cz + d)2 j 0 (z) ∀ ∈ SL2 (Z)
cz + d cd
but that j 0 (z) is not a modular form of weight 2 for the full modular group.
By Exercise 4.6.9 (or by direct verification), the first assertion is clear. j 0 (z)
is not a modular form since it is not holomorphic at infinity. t
u
A short guide for further reading
The readers will benefit much from reading Hardy’s lectures [13] and espe-
cially the chapter relating to Ramanujan’s work on the τ -function. Clearly,
this inspired Rankin’s doctoral thesis and his subsequent work on the topic.
A good introduction to the “q-world” discussed in Chapter 1 can be found
in [7].
The gentlest and most elegant introduction to the topic of modular forms
is the last section of Serre’s Course in Arithmetic [36]. There the student will
find the rudimentary notions needed to enter the topic in the case of the
full modular group. Another reasonably gentle introduction is [2]. Perhaps
the earliest books on the subject were Gunning [12] and Ogg [27]. Both of
these books are of reasonable size that the graduate student will not have
difficulty in studying the material in one semester. Zagier’s chapter in [5] is
also a short introduction to some of the modern themes.
For an amplified study from an analytic and classical perspective, Rankin’s
book [31] is an excellent source of material. Schoenberg’s book [35] also
gives a classical treatment that is accessible. A more amenable treatment
is Iwaniec [16] which assumes a bit more sophistication with analytic num-
ber theory. Lang’s monograph [20] blends some of the analytic and algebraic
themes and explores the theory of congruences from the standpoint of Ga-
lois representations. Miyake’s book [22] is also detailed and reasonably self-
contained for the advanced student. Sarnak [34] highlights some analytic
applications of the theory of modular forms but assumes the reader has a
modest understanding of the subject. Koblitz’s book [19] is also readable
and has the advantage of giving a quick introduction to the topic of half-
integral weight modular forms, a topic that was untouched in this book.
A geometric approach to the subject of modular forms is highlighted in
[9] as well as the online notes of Milne [21]. For the study of Hilbert mod-
ular forms, we recommend Garrett [10] as well as Bruinier’s chapter in [5].
For the topic of Siegel modular forms, a topic which we have not touched
in this book, we suggest the reader begin with van der Geer’s chapter [5].
Also, Siegel’s classic lectures notes from the Tata Institute [39] are an excel-
in Mathematics, DOI 10.1007/978-981-10-2651-5
286 A short guide for further reading
lent repository of a wealth of topics that span several areas of mathematics.

Shimura’s new book [38] is for the advanced reader and blends the theory of
half-integral weight forms with integral weight forms and also gives a good
introduction to the theory of Siegel modular forms. The serious reader can
also profit from a careful study of Shimura’s classic [37].
References
1. G.E. Andrews, A simple proof of Jacobi’s triple product identity. Proc. Amer. Math.
Soc., 16 (1965), 333-334.
2. T. Apostol, Modular functions and Dirichlet series in number theory, Springer-
Verlag, 1976.
3. R. Askey, Ramanujan’s extensions of the Gamma and Beta functions. Amer. Math.
Monthly, 87 (1980), 346-359.
4. B. C. Berndt. Ramanujan’s congruences for the partition function modulo 5, 7, and
11. Int. J. Number Theory, 3 (2007), 349–354.
5. J. H. Bruinier, G. van der Geer, G. Harder, and D. Zagier, 1-2-3 of Modular Forms,
Springer, 2008.
6. J. H. Bruinier, K. Ono, and R. C. Rhoades. Differential operators for harmonic weak
Maass forms and the vanishing of Hecke eigenvalues. Math. Ann., 342(3):673–693,
2008.
7. Hei-Chi Chan, An invitation to q-series, World Scientific, 2011.
8. P. Deligne, La conjecture de Weil: I, Publications Mathématiques de l’IHES, 43
(1974), 273-307.
9. F. Diamond and J. Shurman. A first course in modular forms, Volume 228, Graduate
Texts in Mathematics, Springer-Verlag, New York, 2005.
10. P. Garrett, Holomorphic Hilbert modular forms, Wadsworth and Brooks/Cole
Mathematics Series, Chapman and Hall, CRC, 1989.
11. E. Grosswald, Representations of integers as sums of squares, Springer-Verlag,
New York, 1985.
12. R. Gunning, Lectures on modular forms, Princeton University Press, 1962.
13. G. H. Hardy. Ramanujan. Twelve lectures on subjects suggested by his life and
work, Cambridge University Press, Cambridge, England, 1940.
14. E. Hecke, Analytsche Arithmetik der Positiven Quadratischen Formen in Mathe-
matische Werke, pp. 789-918.
15. M. D. Hirschhorn, A simple proof of Jacobi’s two square theorem, Amer. Math.
Monthly, 92 (1985), 579-80.
16. H. Iwaniec, Topics in classical automorphic forms, Graduate Studies in Mathemat-
ics, Volume 17, American Mathematical Society,1997.
17. H. Iwaniec, Spectral methods in automorphic forms, Graduate Studies in Mathe-
matics, Volume 53, American Mathematical Society, 2002.
18. L. Kilford, Modular forms: a classical and computational introduction, Imperial
College Press, 2008.
19. N. Koblitz, Introduction to elliptic curves and modular forms, Springer, 1984.
20. S. Lang, Introduction to modular forms, Springer-Verlag, 1976.
21. J. Milne, Modular functions and modular forms, online course notes available at
http://www.jmilne.org/math/CourseNotes/mf.html.
22. T. Miyake, Modular forms, Springer-Verlag, Berlin, New York, 1989, 335 pp.
23. L.J. Mordell, On Mr. Ramanujan’s empirical expansions of modular functions, Pro-
ceedings of the Cambridge Philosophical Society, 19 (1917), 117-124.
References 287
24. M. R. Murty. Problems in analytic number theory, Volume 206, Graduate Texts in
Mathematics, Springer, New York, second edition, 2008. Readings in Mathematics.
25. M.R. Murty and J. Esmonde, Problems in algebraic number theory, Volume 190,
Graduate Texts in Mathematics, 2nd edition, Springer, 2005.
26. V. Kumar Murty, Ramanujan and Harish-Chandra, Math. Intelligencer, 15 (1993),
no. 2, 33-39.
27. A.P. Ogg, Modular forms and Dirichlet series, Benjamin, 1969.
28. H. Petersson. Über die Entwicklungskoeffizienten der automorphen Formen. Acta
Math., 58 (1932), 169–215.
29. S. Ramanujan, On certain arithmetical functions, Trans. Cambridge Philos. Soc., 22
(1916), 159-184.
30. R. A. Rankin, The modular group and its subgroups, The Ramanujan Institute,
Madras, 1969.
31. R.A. Rankin, Modular forms, Ellis Horwood, 1984.
32. M. Razar, Modular forms for G0 (N ) and Dirichlet series, Trans. Amer. Math. Soc.,
231 (1977), 489-495.
33. W. Rudin. Real and complex analysis, McGraw-Hill Book Co., New York, third
edition, 1987.
34. P. Sarnak, Some applications of modular forms, Cambridge University Press, 1990.
35. B. Schoenberg, Elliptic modular functions, Springer 1974.
36. J.-P. Serre, A course in arithmetic, Springer, 1970.
37. G. Shimura, Introduction to the Arithmetic Theory of Automorphic Functions,
Princeton University Press, 1971.
38. G. Shimura, Modular forms and beyond, Springer 2011.
39. C.L. Siegel, Advanced Analytic Number Theory, Tata Institute of Fundamental Re-
search, Bombay, 1980.
40. J. Silverman, The arithmetic of elliptic curves, Springer 1986.
41. J. Silverman, Advanced topics in the arithmetic of elliptic curves, Springer 1994.
42. M. Spivak, Calculus on manifolds, A modern approach to classical theorems of
advanced calculus, W. A. Benjamin, Inc., New York-Amsterdam, 1965.
43. W. Stein, Modular forms, a computational approach, Graduate Studies in Mathe-
matics, Volume 79, American Mathematical Society,2007.
44. B. Sury, The Congruence Subgroup Problem, Vol. 24, Texts and Readings in Math-
ematics, Hindustan Book Agency, India, 2003.
45. H. P. F. Swinnerton-Dyer. Congruence properties of τ (n). in Ramanujan revisited
(Urbana-Champaign, Ill., 1987), pages 289–311. Academic Press, Boston, MA, 1988.
46. A. Weil, Über die Bestimmung Dirichletscher Reihen durch Funkionalgleichungen,
Math. Annalen, 168 (1967), 149-156.
47. X. Wang and D. Pei, Modular forms with integral and half-integral weights,
Springer 2012.
Index
G \ X, 19 character, see Dirichlet character

G-equivalent, 19 Chinese remainder theorem, 22
G2 , 56 completely multiplicative, 111
Gx , 19 completely multiplicative function, 250
HatF , 114 conductor of a character, 118
Mk (Γ), 75 congruence subgroup, 17, 86
Mk (Γ0 (N ), χ), 99 converse theory, 131
Sk (Γ), 75 cotangent expansion, 55
Tn (f ), 98 cusp, 177
Xn , 95 cusp form, 75
Xn (N, S × , S + ), 95 on Γ, 75
δij , 143 on full modular group, 36
η, 58 cusps, 30
θk , 58
Γ(N ), 16 Dedekind, xiii
Γ0 (N ), 18 Dedekind η-function, 58, 81, 150, 200, 227,
Γz , 79 283
Γ∞ , 27 Dedekind zeta function, 149
f | γ(z), 36 Deligne, xiii, 64
vz0 (f ), 41 digamma function, 147
dimension formula, 47, 79, 190, 224
Abel’s lemma, 106, 246 Dirichlet, 105
abscissa of convergence, 109 Dirichlet L-series, 147
action, 19–21 functional equation, 260
addition formula, 139 Dirichlet character, 99, 240
analytic automorphism, 26 conductor, 118
analytic continuation, 66 even, 118
arithmetical function, 70, 111, 211, 249 imprimitive, 118
Atkin, xiv odd, 118
Atkin-Lehner involution, 103 primitive, 118
trivial, 118
basis of eigenforms, 235 Dirichlet convolution, 111, 249
Bass, 18 Dirichlet series, 64, 105, 205
Berndt, 68 discrete subgroup, 27
Bernoulli numbers, 40, 187, 192 double cosets, 21, 95, 169
Bernoulli polynomial, 40, 188 doubly periodic, 135
Bruinier, 69 dual basis, 144, 276
in Mathematics, DOI 10.1007/978-981-10-2651-5
290 Index
dual lattice, 144 general level, 96

dual vector space, 144 level one, 61
Hecke operators, 60, 63, 87, 90, 203, 234,
Eisenstein, xiii 235
Eisenstein series, 54, 187 higher level, 239
higher level, 73, 76, 218 Hecke subgroups, 18, 167
on full modular group, 37 Hecke’s trick, 149
Eisenstein series E2 , 53, 199 Hellegouarch, xiv
elliptic curve, 139 Hermitian, 90
elliptic function, 68, 135 Hermitian operators, 235
elliptic functions, 269 Hilbert modular form, 149
elliptic point, 79 Hilbert modular group, 148
order of, 79 Hurwitz, xiii, 54
Euler product, 246 hyperbolic metric, 180
Euler’s pentagonal number theorem, 6,
155 imprimitive Dirichlet character, 118
even character, 148, 280 invariant measure, 144
even lattice, 146 irregular cusps, 75
extended upper half-plane, 73
extended upper-half-plane, 177 j-function, 48, 141, 191, 214
Jacobi, xiii, 57
faithful action, 25 Jacobi triple product identity, 5, 7, 155
Fermat’s last theorem, xiv Jacobi’s four square theorem, 9, 81, 83, 156,
field of elliptic functions, 138 227
fixed point, 19 Jacobi’s two-square theorem, 7, 156
Fourier expansion, 74
Fourier inversion formula, 114 Khare, xiv
Fourier transform, 114, 145 Klingen, 148
fractional linear transformations, 25 Kronecker, xiii
Frey, xiv Kronecker delta, 143
Fricke, 17, 22, 172
full modular group, 161 L-series
functional equation, 66, 114, 256 twist, 129, 131
fundamental domain, 27, 176 Landau, 110
fundamental parallelogram, 135, 269 Langlands, xiv
lattice, 135, 144
g-multiplicative, 70, 211 dual, 144
Galois representations, xiv even, 146
Gamma function, 254 self-dual, 146
Gauss sum, 133, 257, 265 lattices, 276
general Dirichlet series, 245 Lazard, 18
Gram matrix, 145, 277 Legendre symbol, 117, 256
Grothendieck, xiv Lehmer’s conjecture, 69
Lehner, xiv
half-plane of absolute convergence, 109 level, 17
half-plane of convergence, 109 Lipschitz formula, 39, 77
Hardy, 64
Harish-Chandra, xiv Möbius function, 70, 212
harmonic weak Maass forms, 69 Möbius inversion formula, 70, 212
Hausdorff space, 30 Möbius transformations, 25
Hecke, xiv, 64, 66, 128 Maass, xiv
Hecke eigenforms, 92 Miller basis, 93, 237, 238
Hecke eigenvalues, 91 Milnor, 18
Hecke operator modular form
Index 291
on Γ, 75 Ramanujan τ -function, 53, 70, 199

on the full modular group, 36 Ramanujan congruences, 67, 207
of half-integral weight, 118 Ramanujan conjectures, 53, 63, 65, 205
twist of, 129 Ramanujan differential equations, 58
modular function, 49 rapidly decreasing, 145
modular group, 15, 161 Rhoades, 69
modular invariant, 71, 140, 214, 273, 283 Riemann, xiii, 105, 116
moduli space, 143 Riemann zeta function, 106, 109, 116, 248,
Monster group, 48 254
Mordell, xiii Riemann-Roch theorem, 79
multiplicative, 106 Riesz representation theorem, 144
completely, 111
multiplicity one, 103 Schwartz function, 267
Schwartz space, 114
Nebentypus, 99, 240 Schwarz lemma, 26
Nesterenko, 59 self-dual lattice, 146, 277
newforms, 102, 242 Serre, xiv, 18
non-congruence subgroup, 22, 171, 173 Serre derivative, 58, 201
Siegel, 148
odd character, 148, 280 Siegel-Klingen theorem, 148
oldforms, 102, 242 slash notation, 36
Ono, 69 special values of zeta functions, 147, 279
orbit, 19 stabilizer subgroup, 19
orbit-stabilizer formula, 20, 168 Stirling’s formula, 126
order Sturm’s bound, 80, 226
of f at i∞, 36 Swinnerton-Dyer, 67
of an elliptic point, 79
order at iγ(i∞), 75 Taniyama’s conjecture, xiv
order of a zero, 41 theta function, 282
theta series, 144, 276
partial summation, 246
topological group, 20
partition function, 67
totally real number field, 148
Petersson, 48
transitive action, 25
Petersson inner product, 85, 231
trivial character, 118
Phragmén-Lindelöf theorem, 126
twist of L-series, 129, 131
Pick, 17
twist of modular form, 129
Poincaré metric, 32, 180
Poisson summation formula, 112, 115, 251,
266 uniformization theorem, 142, 272
Poisson’s summation formula for lattices, upper half-plane, 25, 175
145
polygamma function, 147 valence formula, 42, 50, 79, 189, 224
primitive Dirichlet character, 118
principal congruence subgroup, 16 Weierstrass ℘-function, 137, 150, 272, 282
addition formula, 139
q-expansion, 35 Weil, xiv
q-exponential, 3, 153 Weil conjectures, xiii, 64
q-logarithm, 14, 159 Weil’s theorem, 132
q-series, 185 width of a cusp, 75
quasimodular form, 56 Wiles, xiv
quotient space, 20 Wilton, 64
Wintenberger, xiv
Ramanujan, xiii, 10, 58, 60, 64 Wohlfahrt, 22, 172

Dokumen - Pub - Problems in The Theory of Modular Forms 978 981 10 2651 5 9811026513 978 93 80250 72 4

Uploaded by

Copyright:

Available Formats

Dokumen - Pub - Problems in The Theory of Modular Forms 978 981 10 2651 5 9811026513 978 93 80250 72 4

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dokumen - Pub - Problems in The Theory of Modular Forms 978 981 10 2651 5 9811026513 978 93 80250 72 4

Uploaded by

Copyright:

Available Formats

HBA Lecture Notes in Mathematics

IMSc Lecture Notes in Mathematics

IMSc Lecture Notes in Mathematics

More information about this series at http://www.springer.com/series/15465

Problems in the Theory

ISSN 2509-8071 (electronic)

Library of Congress Control Number: 2016952892

Printed on acid-free paper

This Springer imprint is published by Springer Nature

2 The Modular Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 The Upper Half-Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4 Modular Forms of Level One . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.5 The j-function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5 The Ramanujan τ -function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

6 Modular Forms of Higher Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

7 The Petersson Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

8 Hecke Operators of Higher Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

9 Dirichlet Series and Modular Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

10 Special Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

1 Jacobi’s q-series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

2 The Modular Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

3 The Upper Half-Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

4 Modular Forms of Level One . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

5 The Ramanujan τ -function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

6 Modular Forms of Higher Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

7 The Petersson Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

8 Hecke Operators of Higher Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239

9 Dirichlet Series and Modular Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

10 Special Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269

A short guide for further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285

This book is based on an upper level undergraduate course given at Queen’s

which he expands as an infinite series and writes

Ramanujan computes by hand many of these coefficients and makes the

At first, Ramanujan’s conjectures are amazing. Why should coefficents of

A conjecture or a problem in mathematics acts like a muse that inspires

Kingston, Ontario Ram Murty

M. Ram Murty is Professor at the Department of Mathematics and Statistics at

1.1 The q-exponential function

It is not clear exactly how to define a q-series. Some experts humorously

and view it as the q-analog of the classical exponential function. We work

It may not be clear to the reader why this should be considered as an

Exercise 1.1.1. Show that

Exercise 1.1.2. Prove that if |q| > 1 then

Exercise 1.1.3. Show that if |q| < 1 then

Exercise 1.1.4. Prove that if |q| > 1 then

Deduce that if |q| < 1 then

1.2 Jacobi’s Triple Product Identity

We now give a simple proof of the celebrated identity of Jacobi:

Remark. According to Askey, this was also contained in some unpublished

Proof. By Exercise 1.1.3 we have, upon replacing q by q 2 and x by xq,

The sum under consideration becomes

Interchanging the sums, we get

The innermost sum is

and can be factored out. The remaining sum is

Theorem 1.2.3 (Jacobi’s formula). If |q| < 1 then

We can rewrite this sum as

1.3 Jacobi’s two-square theorem

An immediate corollary is deduced by expanding the right hand side as a

Multiply both sides by a and differentiate with respect to a, to obtain

Proof. Observe that