This action might not be possible to undo. Are you sure you want to continue?
Editorial Board
J.H. Ewing F.W. Gehring P.R. Halmos
Graduate Texts in Mathematics
1 TAKEUTIIZARING. Introduction to Axiomatic Set Theory. 2nd ed.
2 OXTOBY. Measure and Category. 2nd ed.
3 SCHAEFFER. Topological Vector Spaces.
4 HILTON/STAMMBACH. A Course in Homological Algebra.
5 MAC LANE. Categories for the Working Mathematician.
6 HUGHES/PIPER. Projective Planes.
7 SERRE. A Course in Arithmetic.
8 TAKEUTIlZARING. Axiomatic Set Theory.
9 HUMPHREYS. Introduction to Lie Algebras and Representation Theory.
10 COHEN. A Course in Simple Homotopy Theory.
11 CONWAY. Functions of One Complex Variable. 2nd ed.
12 BEALS. Advanced Mathematical Analysis.
13 ANDERSON/FuLLER. Rings and Categories of Modules.
14 GOLUBITSKy/GUILLEMIN. Stable Mappings and Their Singularities.
15 BERBERIAN. Lectures in Functional Analysis and Operator Theory.
16 WINTER. The Structure of Fields.
17 ROSENBLATT. Random Processes. 2nd ed.
18 HALMOS. Measure Theory.
19 HALMOS. A Hilbert Space Problem Book. 2nd ed., revised.
20 HUSEMOLLER. Fibre Bundles. 2nd ed.
21 HUMPHREYS. Linear Algebraic Groups.
22 BARNES/MACK. An Algebraic Introduction to Mathematical Logic.
23 GREUB. Linear Algebra. 4th ed.
24 HOLMES. Geometric Functional Analysis and its Applications.
25 HEWITT/STROMBERG. Real and Abstract Analysis.
26 MANES. Algebraic Theories.
27 KELLEY. General Topology.
28 ZARISKIISAMUEL. Commutative Algebra. Vol. I.
29 ZARISKIISAMUEL. Commutative Algebra. Vol. II.
30 JACOBSON. Lectures in Abstract Algebra I: Basic Concepts.
31 JACOBSON. Lectures in Abstract Algebra II: Linear Algebra.
32 JACOBSON. Lectures in Abstract Algebra III: Theory of Fields and Galois Theory.
33 HIRSCH. Differential Topology.
34 SPITZER. Principles of Random Walk. 2nd ed.
35 WERMER. Banach Algebras and Several Complex Variables. 2nd ed.
36 KELLEy/NAMIOKA et al. Linear Topological Spaces.
37 MONK. Mathematical Logic.
38 GRAUERT/FRITZSCHE. Several Complex Variables.
39 ARVESON. An Invitation to C*Algebras.
40 KEMENy/SNELL/KNAPP. Denumerable Markov Chains. 2nd ed.
41 ApoSTOL. Modular Functions and Dirichlet Series in Number Theory. 2nd ed.
42 SERRE. Linear Representations of Finite Groups.
43 GILLMAN/JERISON. Rings of Continuous Functions.
44 KENDIG. Elementary Algebraic Geometry.
45 LOEVE. Probability Theory I. 4th ed.
46 LOEVE. Probability Theory II. 4th ed.
47 MOISE. Geometric Topology in Dimensions 2 and 3.
continued after Indet
Tom M. Apostol
Modular Functions
and Dirichlet Series
in Number Theory
Second Edition
With 25 Illustrations
SpringerVerlag
New York Berlin Heidelberg
London Paris Tokyo Hong Kong
Tom M. Apostol
Department of Mathematics
California Institute of Technology
Pasadena, CA 91125
USA
Editorial Board
J.H. Ewing
Department of
Mathematics
Indiana University
Bloomington, IN 47405
USA
F.W. Gehring
Department of
Mathematics
University of Michigan
Ann Arbor, MI 48109
USA
P.R. Halmos
Department of
Mathematics
Santa Clara University
Santa Clara, CA 95053
USA
AMS Subject Classifications
IOA20, IOA45, 10045, IOH05, 10HIO, IOJ20, 30AI6
Library of Congress CataloginginPublication Data
Apostol, Tom M.
Modular functions and Dirichlet series in number theory/Tom M. Apostol.2nd ed.
p. cm.(Graduate texts in mathematics; 41)
Includes bibliographical references.
ISBN 0387971270 (alk. paper)
I. Number theory. 2. Functions, Elliptic. 3. Functions, Modular. 4. Series,
Dirichlet. I. Title. II. Series.
QA241.A62 1990
512'.7dc20 8921760
Printed on acidfree paper.
© 1976, 1990 SpringerVerlag New York, Inc.
All rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher (SpringerVerlag New York, Inc., 17,:, Fifth Avenue, New
York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly
analysis. Use in connection with any form of information and retrieval, electronic adaptation,
computer software, or by sirrlilar or dissimilar methodology now known or hereafter developed
is forbidden.
The use of general descriptive names, trade names, trademarks. etc., in this publication, even if
the former are not especially identified, is not to be taken as a sign that such names, as
understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely
by anyone.
Typeset by Asco Trade Typesetting Ltd., Hong Kong.
Printed and bound by Edwards Brothers, Inc., Ann Arbor, Michigan.
Printed in the United States of America.
9 8 7 6 5 432 I
ISBN 038797 J270 SpringerVerlag New York Berlin Heidelberg
ISBN 3540971270 SpringerVerlag Berlin Heidelberg New York
Preface
This is the second volume of a 2volume textbook* which evolved from a
course (Mathematics 160) offered at the California Institute of Technology
during the last 25 years.
The second volume presupposes a background in number theory com
parable to that provided in the first volume, together with a knowledge of
the basic concepts of complex analysis.
Most of the present volume is devoted to elliptic functions and modular
functions with some of their numbertheoretic applications. Among the
major topics treated are Rademacher's convergent series for the partition
function, Lehner's congruences for the Fourier coefficients of the modular
functionj(r), and Heeke's theory of entire forms with multiplicative Fourier
coefficients. The last chapter gives an account of Bohr's theory of equivalence
of general Dirichlet series.
Both volumes of this work emphasize classical aspects of a subject which
in recent years has undergone a great deal of modern development. It is
hoped that these volumes will help the nonspecialist become acquainted
with an important and fascinating part of mathematics and, at the same
time, will provide some of the background that belongs to the repertory of
every specialist in the field.
This volume, like the first, is dedicated to the students who have taken
this course and have gone on to make notable contributions to number
theory and other parts of mathematics.
T.M.A.
January, 1976
* The first volume is in the SpringerVerlag series Undergraduate Texts in Mathematics under
the title Introduction to Analytic Number Theory.
v
Preface to the Second Edition
The major change is an alternate treatment of the transformation formula for
the Dedekind eta function, which appears in a fivepage supplement to Chap
ter 3, inserted at the end of the book (just before the Bibliography). Other
wise, the second edition is almost identical to the first. Misprints have been
repaired, there are minor changes in the Exercises, and the Bibliography has
been updated.
T.M. A.
July, 1989
Chapter 1
Elliptic functions
1.1 Introduction
1.2 Doubly periodic functions
1.3 Fundamental pairs of periods
1.4 Elliptic functions
1.5 Construction of elliptic functions
1.6 The Weierstrass g;J function
1.7 The Laurent expansion of g;J near the origin
1.8 Differential equation satisfied by g;J
1.9 The Eisenstein series and the invariants 9z and 93
1.10 The numbers e
b
ez, e3
1.11 The discriminant .1
1.12 Klein's modular function J(T)
1.13 Invariance of J under unimodular transformations
1.14 The Fourier expansions of 9z(T) and 93(T)
1.15 The Fourier expansions of .1(T) and J(1:)
Exercises for Chapter 1
Chapter 2
The Modular group and modular functions
2.1 Mobius transformations
2.2 The modular group r
2.3 Fundamental regions
2.4 Modular functions
Contents
1
1
2
4
6
9
11
11
12
13
14
15
16
18
20
23
26
28
30
34
vii
2.5 Special values of J 39
2.6 Modular functions as rational functions of J 40
2.7 Mapping properties of J 40
2.8 Application to the inversion problem for Eisenstein series 42
2.9 Application to Picard's theorem 43
Exercises for Chapter 2 44
Chapter 3
The Dedekind eta function
3.1 Introduction
3.2 Siegel's proof of Theorem 3.1
3.3 Infinite product representation for .1(r)
3.4 The general functional equation for YJ(r)
3.5 Iseki's transformation formula
3.6 Deduction of Dedekind's functional equation from Iseki's
formula
3.7 Properties of Dedekind sums
3.8 The reciprocity law for Dedekind sums
3.9 Congruence properties of Dedekind sums
3.10 The Eisenstein series G
2
(r)
Exercises for Chapter 3
Chapter 4
Congruences for the coefficients of the modular function j
4.1 Introduction
4.2 The subgroup r o(q)
4.3 Fundamental region of r o(p)
4.4 Functions automorphic under the subgroup r o(P)
4.5 Construction of functions belonging to r o(P)
4.6 The behavior of fp under the generators of r
4.7 The function <p(r) = .1(qr)j.1(r)
4.8 The univalent function <I>(r)
4.9 Invariance of <I>(r) under transformations of r o(q)
4.10 The function j p expressed as a polynomial in <I>
Exercises for Chapter 4
Chapter 5
Rademacher's series for the partition function
5.1 Introduction
5.2 The plan of the proof
5.3 Dedekind's functional equation expressed in terms of F
5.4 Farey fractions
viii
47
48
50
51
53
58
61
~
62
64
69
70
74
75
76
78
80
83
84
86
87
88
91
94
95
96
97
5.5 Ford circles
5.6 Rademacher's path of integration
5.7 Rademacher's convergent series for pen)
Exercises for Chapter 5
99
102
104
110
Chapter 6
Modular forms with multiplicative coefficients
6.1 Introduction 113
6.2 Modular forms of weight k 114
6.3 The weight formula for zeros of an entire modular form 115
6.4 Representation of entire forms in terms of G4 and G6 117
6.5 The linear space M
k
and the subspace Mk,o 118
6.6 Classification of entire forms in terms of their zeros 119
6.7 The Hecke operators Tn 120
6.8 Transformations of order n 122
6.9 Behavior of Tnfunder the modular group 125
6.10 Multiplicative property of Hecke operators 126
6.11 Eigenfunctions of Hecke operators 129
6.12 Properties of simultaneous eigenforms 130
6.13 Examples of normalized simultaneous eigenforms 131
6.14 Remarks on existence of simultaneous in M 2k, 0 133
6.15 Estimates for the Fourier coefficients of entire forms 134
6.16 Modular forms and Dirichlet series 136
Exercises for Chapter 6 138
Chapter 7
Kronecker's theorem with applications
7.1 Approximating real numbers by rational numbers 142
7.2 Dirichlet's approximation 143
7.3 Liouville's approximation theorem 146
7.4 Kronecker's approximation theorem: the onedimensional
case 148
7.5 Extension of Kronecker's theorem to simultaneous
approximation 149
7.6 Applications to the Riemann zeta function 155
7.7 Applications to periodic functions 157
Exercisesfor Chapter 7 159
Chapter 8
General Dirichlet series and Bohr's equivalence theorem
8.1 Introduction 161
8.2 The halfplane of convergence of general Dirichlet series 161
8.3 Bases for the sequence of exponents of a Dirichlet series 166
ix
8.4 Bohr matrices 167
8.5 The Bohr function associated with a Dirichlet series 168
8.6 The set of values taken by a Dirichlet series f(s) on a line
U = Uo 170
8.7 Equivalence of general Dirichlet series 173
8.8 Equivalence of ordinary Dirichlet series 174
8.9 Equality of the sets Uf(uo) and Ug(u
o
) for equivalent
Dirichlet series 176
8.10 The set of values taken by a Dirichlet series in a neighborhood
of the line U = Uo 176
8.11 Bohr's equivalence theorem 178
8.12 Proof of Theorem 8.15 179
8.13 Examples of equivalent Dirichlet series. Applications of Bohr's
theorem to Lseries 184
8.14 Applications of Bohr's theorem to the Riemann zeta function 184
Exercisesfor Chapter 8 187
Supplement to Chapter 3
Bibliography
Index of special symbols
Index
x
190
196
199
201
Elliptic functions
1.1 Introduction
Additive number theory is concerned with expressing an integer n as a sum
of integers from some given set S. For example, S might consist of primes,
squares, cubes, or other special numbers. We ask whether or not a given
number can be expressed as a sum of elements of S and, if so, in how many
ways this can be done.
Letf(n) denote the number of ways n can be written as a sum of elements
of S. We ask for various properties of f(n), such as its asymptotic behavior
for large n. In a later chapter we will determine the asymptotic value of the
partition function p(n) which counts the number of ways n can be written as a
sum of positive integers ~ n.
The partition function p(n) and other functions of additive number theory
are intimately related to a class of functions in complex analysis called
elliptic modular functions. They playa role in additive number theory analo
gous to that played by Dirichlet series in multiplicative number theory. The
first three chapters of this volume provide an introduction to the theory of
elliptic modular functions. Applications to the partition function are given
in Chapter 5.
We begin with a study of doubly periodic functions.
1.2 Doubly periodic functions
A function j' of a complex variable is called periodic with period w if
f(z + w) = f(z)
whenever z and z + ware in the domain off If w is a period, so is nw for
every integer n. If W
1
and W2 are periods, so is mW
1
+ nW2 for every choice of
integers m and n.
1: Elliptic functions
Definition. A function f is called doubly periodic if it has two periods W
1
and W
2
whose ratio w
2
1w
1
is not real.
We require that the ratio be nonreal to avoid degenerate cases. For
example, if W
1
and W
2
are periods whose ratio is real and rational it is easy
to show that each of W
1
and W
2
is an integer multiple of the same period. In
fact, if w21w1 = alb, where a and b are relatively prime integers, then there
exist integers m and n such that mb + na = 1. Let w = mW
1
+ nw
2
. Then
w is a period and we have
so W
1
= bw and W
2
= aw. Thus both W
1
and W
2
are integer multiples of W.
If the ratio w
2
1w
1
is real and irrational it can be shown thatfhas arbitrarily
small periods (see Theorem 7.12). A function with arbitrarily small periods
is constant on every open connected set on which it is analytic. In fact, at
each point of analyticity offwe have
f'(z) = lim f(z + zn)  f(z),
z " ~ o Zn
where {zn} is any sequence of nonzero complex numbers tending to O. Iff
has arbitrarily small periods we can choose {zn} to be a sequence of periods
tending to O. Then j'(z + zn) = j'(z) and hence j"(z) = O. In other words,
f'(z) = 0 at each point of analyticity off, hence fmust be constant on every
open connected set in whichfis analytic.
1.3 Fundamental pairs of periods
Definition. Let JO have periods w
1
, W
2
whose ratio w
2
1w
1
is not real. The
pair (w
1
, w
2
) is called afundamental pair if every period of f is of the form
mW
1
+ nw
2
, where m and n are integers.
Every fundamental pair of periods W1' W
2
determines a network of
parallelograms which form a tiling of the plane. These are called period
parallelograms. An example is shown in Figure l.la. The vertices are the
periods w = mW
1
+ nw
2
. It is customary to consider two intersecting edges
and their point of intersection as the only boundary points belonging to the
period parallelogram, as shown in Figure l.lb.
Notation. If W
1
and W
2
are two complex numbers whose ratio is not real
we denote by 0(w
1
, W2), or simply by 0, the set of all linear combinations
mW
1
+ nw
2
, where m and n are arbitrary integers. This is called the lattice
generated by W
1
and W2'
2
1.3: Fundamental pairs of periods
o
(a)
Figure 1.1
(b)
Theorem 1.1. I}' (WI' W
2
) is a .fundamental pair oj' periods, then the triangle
with vertices 0, WI' W2 contains no further periods in its interior or on its
boundary. Conversely, any pair of periods with this property isfundamental.
PROOF. Consider the parallelogram with vertices 0, W b WI + W
2
, and W2'
shown in Figure 1.2a. The points inside or on the boundary of this parallel
ogram have the form
Z = rJ..W
1
+ f3W2'
where °:::; rJ.. :::; 1 and °:::; 13 :::; 1. Among these points the only periods are 0,
WI' W
2
, and WI + W2' so the triangle with vertices 0, W
b
W
2
contains no
periods other than the vertices.
o
(a)
o
(b)
Figure 1.2
3
1: Elliptic functions
Conversely, suppose the triangle 0, w
1
, W2 contains no periods other
than the vertices, and let Wbe any period. We are to show that W = mW
1
+
nW
2
for some integers m and n. Since W
2
/W
1
is nonreal the numbers W
1
and
W
2
are linearly independent over the real numbers, hence
where t 1 and t
2
are real. Now let [t] denote the greatest integer stand
write
Then
/
W  [t
1
]W
1
 [t
2
]W
2
= rl
w
l + r2w2.
If one of rl or r2 is nonzero, then rl W
1
+ r2 W2 will be a period lying inside
the parallelogram with vertices 0, W
b
W2' W
1
+ W
2
. But if a period w lies
inside this parallelogram then either w or W
1
+ W
2
 w will lie inside the
triangle 0, W
1
, W2 or on the diagonal joining W
1
and W2, contradicting the
hypothesis. (See Figure 1.2b.) Therefore rl = r2 = 0 and the proof is
complete. D
Definition. Two pairs of complex numbers (w
1
, (
2
) and (w
1
', w
2
'), each with
nonreal ratio, are called equivalent if they generate the same lattice of
periods; that is, ifQ(w
b
(2) = Q(w
1
', w
2
').
The next theorem, whose proof is left as an exercise for the reader,
describes a fundamental relation between equivalent pairs of periods.
Theorem 1.2. Two pairs (w
b
(
2
) and (w
1
', w
2
') are equivalent if, and only if,
there is a 2 x 2 matrix (: ~ ) with integer entries and determinant
ad  be = ±1, such that
or, in other words,
W
2
' = aW
2
+ bw
1
,
w
1
' = eW
2
+ dw
1
•
1.4 Elliptic functions
Definition. A functionf is called elliptic if it has the following two properties:
(a) f is doubly periodic.
(b) f is meromorphic (its only singularities in the finite plane are poles).
4
1.4: Elliptic functions
Constant functions are trivial examples of elliptic functions. Later we
shall give examples of nonconstant elliptic functions, but first we derive some
fundamental properties common to all elliptic functions.
Theorem 1.3. A nonconstant ellipticfunction has afundamental pair of periods.
PROOF. Iffis elliptic the set of points wheref is analytic is an open connected
set. Also, f has two periods with nonreal ratio. Among all the nonzero
periods of j' there is at least one whose distance from the origin is minimal
(otherwisefwould have arbitrarily small nonzero periods and hence would
be constant). Let W be one of the nonzero periods nearest the origin. Among
all the periods with modulus IW I choose the one with smallest nonnegative
argument and call it W
l
. (Again, such a period must exist otherwise there
would be arbitrarily small nonzero periods.) If there are other periods
with modulus Iwll besides W
l
and W
l
, choose the one with smallest
argument greater than that of W
l
and call this W2. If not, find the next
larger circle containing periods =1= nW
l
and choose that one of smallest
nonnegative argument. Such a period exists since f has two noncollinear
periods. Calling this one W2 we have, by construction, no periods in the
triangle 0, W
l
, W
2
other than the vertices, hence the pair (0)1' W
2
) is funda
mental. [J
Iff and g are elliptic functions with periods Wi and W
2
then their sum,
difference, product and quotient are also elliptic with the same periods. So,
too, is the derivative f'.
Because of periodicity, it suffices to study the behavior of an elliptic
function in any period parallelogram.
Theorem 1.4. If an elliptic function f has no poles in some period parallelogram,
then f is constant.
PROOF. Iffhas no poles in a period parallelogram, thenfis continuous and
hence bounded on the closure of the parallelogram. By periodicity, f is
bounded in the whole plane. Hence, by Liouville's theorem,fis constant. [J
Theorem 1.5. If an ellipticfunction f has no zeros in some period parallelogram,
then f is constant.
PROOF. Apply Theorem 1.4 to the reciprocaI1/!
[J
Note. Sometimes it is inconvenient to have zeros or poles on the bound
ary of a period parallelogram. Since a meromorphic function has only a
finite number of zeros or poles in any bounded portion of the plane, a period
parallelogram can always be translated to a congruent parallelogram with
no zeros or poles on its boundary. Such a translated parallelogram, with no
zeros or poles on its boundary, will be called a cell. Its vertices need not be
periods.
5
1: Elliptic functions
Theorem 1.6. The contour integral of an elliptic function taken along the
boundary of any cell is zero.
PROOF. The integrals along parallel edges cancel because of periodicity. D
Theorem 1.7. The sum of the residues of an ellipticfunction at its poles in any
period parallelogram is zero.
PROOF. Apply Cauchy's residue theorem to a cell and use Theorem 1.6. D
\
Note. Theorem 1.7 shows that an elliptic function which is not constant
has at least two simple poles or at least one double pole in each period
parallelogram.
Theorem 1.8. The number ofzeros ofan elliptic function in any period parallel
ogram is equal to the number of poles, each counted with multiplicity.
PROOF. The integral
_1 f f'(z) dz
2ni c j'(z) ,
taken around the boundary C of a cell, counts the difference between the
number of zeros and the number of poles insid'e the cell. But f' / f is elliptic
with the same periods as/, and Theorem 1.6 tells us that this integral is zero.
D
Note. The number of zeros (or poles) of an elliptic function in any period
parallelogram is called the order of the function. Every nonconstant elliptic
function has order ~ 2.
1.5 Construction of elliptic functions
We turn now to the problem of constructing a nonconstant elliptic function.
We prescribe the periods and try to find the simplest elliptic function having
these periods. Since the order of such a function is at least 2 we need a
second order pole or two simple poles in each period parallelogram. The
two possibilities lead to two theories of elliptic functions, one developed by
Weierstrass, the other by Jacobi. We shall follow Weierstrass, whose point
of departure is the construction of an elliptic function with a pole of order
2 at z = 0 and hence at every period. Near each period OJ the principal part
of the Laurent expansion must have the form
A B
2+·
(z  OJ) Z  OJ
6
1.5: Construction of elliptic functions
For simplicity we take A = 1, B = O. Since we want such an expansion near
each period w it is natural to consider a sum of terms of this type,
summed over all the periods w = mW
l
+ nW2. For fixed z =1= w this is a
double series, summed over m and n. The next two lemmas deal with con
vergence properties of double series of this type. In these lemmas we denote
by Q the set of all linear combinations mw 1 + nW2' where m and n are
arbitrary integers.
Lemma 1. If rJ. is real the infinite series
converges absolutely if, and only if, rJ. > 2.
PROOF. Refer to Figure 1.3 and let rand R denote, respectively, the minimum
and maximum distances from 0 to the parallelogram shown. If w is any of
the 8 nonzero periods shown in this diagram we have
r ~ Iwl ~ R
(for 8 periods w).
Figure 1.3
In the next concentric layer of periods surrounding these 8 we have 2 ·8 = 16
new periods satisfying the inequalities
2r:::; Iwl :::; 2R (for 16 new periods w).
In the next layer we have 3 . 8 = 24 new periods satisfying
3r :::; Iwl :::; 3R (for 24 new periods w),
7
(1)
(2)
1: Elliptic functions
and so on. Therefore, we have the inequalities
I I" for the first 8 periods w,
for the next 16 periods w,
and so on. Thus the sum S(n) = L /OJI(X, taken over the 8(1 + 2 + ... + n)
nonzero periods nearest the origin, satisfies the inequalities
8 2·8 n·8 8 2·8 n·8
R" + (2Rt + ... + (nR)" S(n) ? + (2r)" + ... + (nr)'"
or
8 n 1 8 n 1
R
(X L k(Xl S(n) ex L k(Xl·
k=l r k=l
This shows that the partial sums S(n) are bounded above by  l)jr(X if
> 2. But any partial sum lies between two such partial sums, so all of the
partial sums of the series L /OJ , (X are bounded above and hence the series
converges if > 2. The lower bound for S(n) also shows that the series
diverges if 2. D
Lemma 2. If > 2 and R > 0 the series
L 1"
Iwl > R (z  OJ)
converges absolutely and uniformly in the disk Iz I R.
PROOF. We will show that there is a constant M (depending on R and
such that, if '1, we have
1 M
<
Iz  OJI(X  IOJI(X
for all OJ with /OJ / > R and all z with Iz I R. Then we invoke Lemma 1 to
prove Lemma 2. Inequality (1) is equivalent to
Iz: WI"
To exhibit M we consider all OJ in n with IOJ I > R. Choose one whose
modulus is minimal, say IOJ I = R + d, where d > O. Then if Iz I s Rand
IOJ I R + d we have
Iz: Wi = 1
1
 1 1  R d'
8
1.6: The Weierstrass tJ function
and hence
where
(
R )a.
M= l
R
+
d
.
This proves (2) and also the lemma. D
As mentioned earlier, we could try to construct the simplest elliptic
function by using a series of the form
1
L (ZW)2'
WEn
This has the appropriate principal part near each period. However, the
series does not converge absolutely so we use, instead, a series with the
exponent 2 replaced by 3. This will give us an elliptic function of order 3.
Theorem 1.9. Let f be defined by the series
1
f(z) = L ( )3'
WEn z  OJ
Thenfis an ellipticfunction with periods OJ
1
, OJ
2
and with a pole oforder 3at
each period OJ in Q.
PROOF. By Lemma 2 the series obtained by summing over IOJ I > R converges
uniformly in the disk Iz I s R. Therefore it represents an analytic function
in this disk. The remaining terms, which are finite in number, are also
analytic in this disk except for a 3rd order pole at each period OJ in the disk.
This proves thatfis meromorphic with a pole of order 3 at each OJ in Q.
Next we show thatfhas periods OJ
1
and OJ
2
• For this we take advantage
of the absolute convergence of the series. We have
1
f(z + wd = L ( )3'
WEn z + OJ
1
 OJ
But OJ  OJ
1
runs through all periods in Q with OJ, so the series for f(z + OJ
1
)
is merely a rearrangement of the series for f(z). By absolute convergence we
have f(z + OJ
1
) = f(z). Similarly, f(z + OJ
2
) = f(z) so f is doubly periodic.
This completes the proof. 0
1.6 The Weierstrass 80 function
Now we use the function of Theorem 1.9 to construct an elliptic function
or order 2. We simply integrate the series for f(z) term by term. This gives us
a principal part  (z  OJ)  2/2 near each period, so we multiply by  2 to
9
1: Elliptic functions
get the principal part (z  w)  2. There is also a constant of integration to
reckon with. It is convenient to integrate from the origin, so we remove the
term z  3 corresponding to w = 0, then integrate, and add the term z  2.
This leads us to the function
1 fZ 2
2 + L ( )3 dt.
z 0 £1)*0 t  w
Integrating term by term we arrive at the following function, called the
Weierstrass g;J function.
Definition. The Weierstrass g;J function is defined by the series
1 {II}
.f.J(z) = 2 + L ( )2  2 .
z w*o Z  W W
Theorem 1.10. The function g;J so defined has periods W
t
and w
2
• It is analytic
except for a double pole at each period w in Q. Moreover g;J(z) is an even
function of z.
PROOF. Each term in the series has modulus
1
1 lIIW2(ZW)21IZ(2WZ)1
(z  W)2  w
2
= w
2
(z  W)2 = w
2
(z  W)2 .
Now consider any compact disk Iz I ::; R. There are only a finite number of
periods W in this disk. If we exclude the terms of the series containing these
periods we have, by inequality (1) obtained in the proof of Lemma 2,
where M is a constant depending only on R. This gives us the estimate
I
z(2w  z) I MR(2Iwl + R) MR(2 + R/lwl) 3MR
< < <
w
2
(z  W)2  Iwl
4
 Iwl
3
 Iwl
3
since R < Iw I for w outside the disk IzI::; R. This shows that the truncated
series converges absolutely and uniformly in the disk Iz I ::; R and hence
is analytic in this disk. The remaining terms give a secondorder pole at
each w inside this disk. Therefore g;J(z) is meromorphic with a pole of order 2
at each period.
Next we prove that g;J is an even function. We note that
Since  w runs through all nonzero periods with w this shows that g;J(  z) =
g;J(z), so g;J is even.
10
1.8: Differential equation satisfied by f.J
Finally we establish periodicity. The derivative of is given by
, 1
f.J (z) =  2 L ( )3·
wen Z  W
We have already shown that this function has periods WI and Thus
+ w) = for each period w. Therefore the function + w) 
is constant. But when z =  w/2 this constant is  w/2) = 0
since is even. Hence + w) = for each w, so has the required
periods. 0
1.7 The Laurent expansion of 8'J near the
origin
Theorem 1.11. Let r = min {Iwl:w =I O}. Thenfor 0 < Izi < r we have
1 00
(3) = 2: + L (2n + I)G2n+2z2n,
Z n= 1
where
(4) for n 3.
PROOF. If 0 < Izi < r then Iz/wl < 1 and we have
hence
1
1 1 ( 00 (z )n)
.,.(.)2 = 2 1 + L (n + 1)  ,
2 1 z W n=1 W
W 
w
Summing over all w we find (by absolute convergence)
1 00 1 1 00
= 2: + L (n + 1) L n+2
zn
= 2: + L (n + I)G
n
+
2
z
n
,
z n=1 w*O W Z n=1
where G
n
is given by (4). Since is an even function the coefficients G
2n
+ 1
must vanish and we obtain (3). 0
1.8 Differential equation satisfied by 8'J
Theorem 1.12. The function satisfies the nonlinear differential equation
=  60G
4
SO(z)  140G
6
.
PROOF. We obtain this by forming a linear combination of powers of and
eliminates the pole at z = O. This gives an elliptic function which has
11
1: Elliptic functions
no poles and must therefore be constant. Near z = 0 we have
2 3
8O'(z) =  3 + 6G
4
z + 20G
6
z + ...,
z
an elliptic function of order 3. Its square has order 6 since
2 4 24G
4
[80'(z)] =     80G
6
+ .. "
Z6 Z2
where + ... indicates a power series in z which vanishes at z = O. Now
4f.J3(Z) = ~ + 36G
4
+ 60G
6
+ ...
Z6 Z2
hence
2 3 60G
4
[80'(z)]  480 (z) =  2  140G
6
+ ...
z
so
[80'(z)]2  480
3
(z) + 60G
4
80(z) = 140G
6
+ ....
Since the left member has no pole at z = 0 it has no poles anywhere in a
period parallelogram so it must be constant. Therefore this constant must
be 140G
6
and this proves the theorem. D
1.9 The Eisenstein series and the invariants
92 and 93
Definition. If n ~ 3 the series
is called the Eisenstein series of order n. The invariants 92 and 93 are the
numbers defined by the relations
The differential equation for 80 now takes the form
[80'(z)]2 = 480
3
(z)  g280(z)  g3'
Since only g2 and g3 enter in the differential equation they should determine
80 completely. This is actually so because all the coefficients (2n + 1)G
2n
+ 2
in the Laurent expansion of 8O(z) can be expressed in terms of g2 and g3.
Theorem 1.13. Each Eisenstein series Gn is expressible as a polynomial in g2
and g3 with positive rational coefficients. In fact, if b(n) = (2n + l)G2n +2
we have the recursion relations
b(2) = g3/28,
12
and
n2
(2n + 3)(n  2)b(n) = 3 L b(k)b(n  1  k)
k=1
or equivalently,
for n 3,
m2
(2m + l)(m  3)(2m  1)G
2m
= 3 L (2r  1)(2m  2r  1)G2rG2m2r
r= 2
for m 4.
PROOF. Differentiation of the differential equation for gives another
differential equation of second order satisfied by
(5) =  tg2.
Now we write = z 2 + 1 b(n)z2n and equate like powers of z
in (5) to obtain the required recursion relations. D
Definition. We denote by eb e2' e3 the values of at the halfperiods,
The next theorem shows that these numbers are the roots of the cubic
 g2  g3·
Theorem 1.14. We have
 g2  g3 =    e
3
).
Moreover, the roots el' e2' e3 are distinct, hence g2
3
 27g
3
2
# o.
PROOF. Since is even, the derivative is odd. But it is easy to show that
the halfperiods of an odd elliptic function are either zeros or poles. In fact,
by periodicity we have tw) =  tw) = and since odd
we also have tw) =  Hence = 0 if is finite.
Since has no poles at tw
1
, tw
2
, t(w
1
+ (2), these points must be
zeros of But is of order 3, so these must be simple zeros of Thus
can have no further zeros in the periodparallelogram with vertices
0, WI' W2' WI + w
2
. The differential equation shows that each of these points
is also a zero of the cubic, so we have the factorization indicated.
Next we show that the numbers el' e2' e3 are distinct. The elliptic function
 el vanishes at z = tw
1
. This is a double zero since = o.
Similarly,  e2 has a double zero at tW2. If el were equal to e2, the
elliptic function  el would have a double zero at tW
I
and also a double
13
1: Elliptic functions
zero at tOJ
2
, so its order would be 4. But its order is 2, so e
1
=I e2. Similarly,
e1 =I e3 and e2 =I e3.
If a polynomial has distinct roots, its discriminant does not vanish. (See
Exercise 1.7.) The discriminant of the cubic polynomial
4x
3
 g2 X  g3
is g2
3
 27g
3
2
. When x = the roots of this polynomial are distinct so
the number g2
3
 27g
3
2
=I O. This completes the proof. 0
1.11 The discriminant
The number = g2
3
 27g3
2
is called the discriminant. We regard the
invariants g2 and g3 and the discriminant as functions of the periods WI
and W2 and we write
The Eisenstein series show that g2 and g3 are homogeneous functions of
degrees 4 and 6, respectively. That is, we have
g2(AWb A(2) = A4
g2
(W
b
(
2
) and g3(AW
1
, A(
2
) = A6
g3
(W
b
(
2
)
for any A=I O. Hence is homogeneous of degree 12,
A(2) = A (
2
).
Taking A = 1/w
1
and writing T = W2/W1 we obtain
g2(1, T) = W1
4
g2(W
1
, (
2
), g3(1, T) = W
1
6
g3
(W
1
, (
2
),
T) = W
l
12
(
2
).
Therefore a change of scale converts g2' g3 and into functions of one
complex variable T. We shall label W
1
and W2 in such a way that their ratio
T = W2/W1 has positive imaginary part and study these functions in the upper
halfplane Im(T) > O. We denote the upper halfplane Im(T) > 0 by H.
If TE H we write g2(T), g3(T) and for g2(1, T) g3(1, T) and T),
respectively. Thus, we have
+00 1
92(r) = 60 L ( )4'
m, n =  ex; m + nT
(m, n) *(0,0)
+00 1
93(r) = 140 (m + nr)6
(m, n) *(0,0)
and
= g2
3
(T)  27g
3
2
(T).
Theorem 1.14 shows that =I 0 for all Tin H.
14
(6)
1.12: Klein's modular function J(r)
1.12 Klein's modular function J(7:)
Klein's function is a combination of g2 'and g3 defined in such a way that,
as a function of the periods WI and W2' it is homogeneous of degree O.
Definition. If w
2
1w1 is not real we define
g2
3
(W
1
, (2)
J(w1, w2 ) = ~ ( ) .
W
1
,W2
Since g2
3
and i\ are homogeneous of the same degree we have J(AW1, A(2)
= J(Wb (2). In particular, if T E H we have
J(l, T) = J(Wl' (
2
).
Thus J(w
1
, (2) is a function of the ratio Talone. We write J(T) for J(l, T).
Theorem 1.15. The functions g2(T), g3(T), i\(T), and J(T) are analytic in H.
PROOF. Since i\(T) =I 0 in H it suffices to prove that g2 and g3 are analytic
in H. Both {l" and g3 are given by double series of the form
+00 1
m . n ~  o o (m + mY
(m,n)*(O,O)
·'ith ex > 2. Let T = X + iy, where y > O. We shall prove that if ex > 2 this
series converges absolutely for any fixed T in H and uniformly in every strip
S of the form
S = {x + iy:lxl ~ A,y ~ iJ > O}.
(See Figure 1.4.) To do this we prove that there is a constant M > 0, depending
only on A and on iJ, such that
1 M
<
1m + nTla  1m + nil
a
for all T in S and all (m, n) =I (0,0). Then we invoke Lemma 1.
To prove (6) it suffices to prove that
1m + nTI
2
> Kim + nil
2
for some K > 0 which depends only on A and iJ, or that
(7) (m + nx)2 + (ny)2 > K(m
2
+ n
2
).
If n = 0 this inequality holds with any K such that 0 < K < 1. If n =I 0
let q = min. Proving (7) is equivalent to showing that
(q + X)2 + y2
(8) 1 2 > K
+q
15
1: Elliptic functions
     +     +     +       ~ x
A
Figure 1.4
A
for some K > O. We will prove that (8) holds for all q, with
b
2
K=
1 + (A + b)2
if Ix I s A and y ~ b. (This proof was suggested by Christopher Henley.)
If IqI s A + b inequality (8) holds trivially since (q + X)2 ~ 0 and
y2 ~ b
2
. Iflql > A + b then Ix/ql < Ixl/(A + b) s A/(A + b) < 1 so
1
1 + ~ I ~ 1  I ~ I > 1 _ _ A = _b
q q A+b A+b
hence
qb
Iq + xl ~ A + b
and
(q + X)2 + y2 b
2
q2
(9) 1 + q2 > (A + 15)2 1 + q2·
Now q2/(1 + q2) is an increasing function of q2 so
q2 (A + b)2
>
1 + q2  1 + (A + b)2
when q2 > (A + b)2. Using this in (9) we obtain (8) with the specified K. D
1.13 In\rariance of J under unimodular
transformations
If WI' W2 are given periods with nonreal ratio, introduce new periods
WI" w
2
' by the relations
16
1.13: Invariance of J under unimodular transformations
where a, b, c, d are integers such that ad  bc = 1. Then the pair (WI" W
2
')
is equivalent to (WI' W
2
); that is, it generates the same set of periods Q.
Therefore g2(W
I
', w
2
') = g2(W
I
, W2) and g3(W
I
', W2') = g3(Wb W2) since g2
and g3 depend only on the set of periods Q. Consequently, ~ ( W I " w
2
') =
~ ( W I ' W2) and J(WI', W2') = J(w
l
, w
2
)·
The ratio of the new periods is
aW2 + bW
I
CW2 + dW
I
aT + b
CT + d'
where T = W2/WI. An easy calculation shows that
, (aT + b) ad  be Im(T)
Im(T) = 1m C't" + d = ICT + dl
2
Im(T) = leT + dl
2
'
Hence T' E H if and only if T E H. The equation
, aT + b
T =
CT + d
is called a unimodular transformation if a, b, c, d are integers with ad  bc = 1.
The set of all unimodular transformations forms a group (under composition)
called the modular group. This group will be discussed further in the next
chapter. The foregoing remarks show that the function J(T) is invariant
under the transformations of the modular group. That is, we have:
Theorem 1.16. If T E H and a, b, c, d are integers with ad  bc = 1, then
(aT + b)/(CT + d) E Hand
(10)
(
aT + b)
J CT + d = J(T).
Note. A particular unimodular transformation is T' = T + 1, hence (10)
shows that J(T + 1) = J(T). In other words, J(T) is a periodic function of T
with period 1. The next theorem shows that J(T) has a Fourier expansion.
Theorem 1.17. If T E H, J(T) can be represented by an absolutely convergent
FDurier series
(11)
00
J(T) = L a(n)e21tint.
n=  00
PROOF. Introduce the change of variable
Then the upper halfplane H maps into the punctured unit disk
D = {x:O < Ixl < I}.
17
1: Elliptic functions
(See Figure 1.5.) Each T in H maps onto a unique point x in D, but each
x in D is the image of infinitely many points in H. If T and T' map onto x
then e
21tit
= e
21tit
' so T and T' differ by an integer.
H
D
Figure 1.5
If xED, let
f(x) = J(T)
where T is any of the points in H which map onto x. Since J is periodic with
period 1, J has the same value at all these points so .f(x) is welldefined.
Nowfis analytic in D because
, d d dT , /dX J'(T)
f (x) = d J(T) = d J(T) d = J (T) d = 2 . 21tit'
X TXT nle
so f'(x) exists at each point in D. Since f is analytic in D it has a Laurent
expansion about 0,
00
f(x) = L a(n)x
n
,
n=  00
absolutely convergent for each x in D. Replacing x by e
21tit
we see that J(T)
has the absolutely convergent Fourier expansion in (11). D
Later we will show that a_
n
= 0 for n ~ 2, that aI = 12
3
, and that
the Fourier expansion of 12
3
J(T) has integer coefficients. To do this we first
determine the Fourier expansions of g2(T), g3(T) and ~ ( T ) .
1.14 The Fourier expansions of g2(r)
andg
3
(r)
Each Eisenstein series L(m, n) #: (0,0) (m + nT)  k is a periodic function of T of
period 1. In particular, g2(T) and g3(T) are periodic with period 1. In this
section we determine their Fourier coefficients explicitly.
We recall that
18
1
g2(T) = 60 L 4'
(m,n)#:(O,O) (m + nT)
1
g3(T) = 140 L 6·
(m,n)#:(O, 0) (m + nT)
1.14: The Fourier expansions of g2(r) and g3(r)
These are double series in m and n. First we obtain Fourier expansions for
the simpler series
+00 1 +00 1
L 4 and L 6'
m=  00 (m + nT) m=  00 (m + nT)
Lemma 3. If T E Hand n > 0 we have the Fourier expansions
+ 00 1 8n
4
00
L   L r3e21tirnt
m=  00 (m + nT)4  3 r= 1
and
+ 00 1 8n
6
00
L 6 =   L r5e21tirnt.
m=  00 (m + nT) 15 r= 1
PROOF. Start with the partial fraction decomposition of the cotangent:
1 +00 (1 1)
n cot nT =  + L   .
T m=oo T+m m
m*O
Let x = e
21tit
. If T E H then IxI < 1 and we find
cos nT e
21tit
+ 1 x + 1 (x 1)
n cot nT = n = ni. = ni =  ni + 
sin nT e
21tlt
 1 xI 1  x 1  x
=  ni( I x
r
+ I x
r
) =  ni(l + 2 I x
r
).
r=1 r=O r=1
In other words, if T E H we have
1 +00 (1 1) ( 00 )
 + L   =  ni 1 + 2 L e21tirt .
T m=oo T+m m r=1
m*O
Differentiating repeatedly we find
(12)
and
00 1 00
L =  (2ni)2 L re21tirt
m= 00 (T + m)2 r =1
m*O
+00 1 00
3! L 4 = (2ni)4 L r3e21tirt
m=_oo(T+m) r=1
Replacing T by nT we obtain Lemma 3.
o
19
1: Elliptic functions
Theorem 1.18. If T E H we have the Fourier expansions
4n4 { 00 }
g2(T) =  1 + 240 L (J3(k)e21tikt
3 k=1
and
g3('r) = {1 
where (J(J.(k) = Ldlk d(J..
PROOF. We write
+00 1
g2(r) = 60 (m + nr)4
(m, n) =#= (0,0)
{
+ 00 1 00 + 00 (1 1)}
=60 L 4 + L L 4+ 4
m=  00 m n= 1 m=  00 (m + nT) (m  nT)
m=#=O(n=O)
{
00 +00 1 }
= 60 2((4) + (m + nr)4
{
2n
4
16n4 00 00 }
= 60  +  L L r
3
x
nr
90 3 n= 1 r= 1
where x = e
21tit
. In the last double sum we collect together those terms for
which nr is constant and we obtain the expansion for g2(T). The formula
for g3(T) is similarly proved. 0
1.15 The Fourier expansions of and J(r)
Theorem 1.19. 1fT E H we have the Fourier expansion
00
= (2n)12 L T(n)e21tint
n=1
where the coefficients T(n) are integers, with T(1) = 1 and T(2) =  24.
Note. The arithmetical function T(n) is called Ramanujan's tau function.
Some of its arithmetical properties are described in Chapter 4.
PROOF. Let
00
A = L (J3(n)x
n
,
n=1
00
B = L (J5(n)x
n
.
n=1
Then
64n
12
= g/(r)  27g/(r) = {(1 + 240A)3  (1  504B)2}.
20
1.15: The Fourier expansions of L\(r) and J(r)
Now A and B have integer coefficients, and
(1 + 240A)3  (1  504B)2 = 1 + 720A + 3(240)2A
2
+ (240)3A
3
 1
+ 1008B  (504)2B
2
= 12
2
(5A + 7B)
+ 12
3
(100A
2
 147B
2
+ 8000A
3
).
But
00
5A + 7B = L {5a3(n) + 7a
s
(n)}x
n
n=l
and
so
5d
3
+ td
s
== 0 (mod 12).
Hence 12
3
is a factor of each coefficient in the power series expansion of
(1 + 240A)3  (1  504B)2 so
= 64n
12
{12
3
f r(n)e
21tinr
} = (2n)12 f r(n)e
21tinr
27 n=1 n=1
where the r(n) are integers. The coefficient of x is 12
2
(5 + 7), so r(l) = 1.
Similarly, we find r(2) =  24. D
Theorem 1.20. Ifr E H we have the Fourier expansion
00
12
3
J(r) = e
21tir
+ 744 + L c(n)e21tinr,
n= 1
where the c(n) are integers.
PROOF. We agree to write I for any power series in x with integer coefficients.
Then if x = e
21ti
t: we have
g2
3
(r) = + 240x + 1)3 = + 720x + I),
=  24x + I)}
and hence
g2
3
(r) 1 + 720x + I 1
J(r) = = 123x(1 _ 24x + 1) = 123x (1 + 720x + 1)(1 + 24x + 1)
so
1 00
12
3
J(r) =  + 744 + L c(n)x
n
,
x n= 1
where the c(n) are integers.
D
21
1: Elliptic functions
Note. The coefficients c(n) have been calculated for n ~ 100. Berwick
calculated the first 7 in 1916, Zuckerman the first 24 in 1939, and Van
Wijngaarden the first 100 in 1953. The first few are repeated here.
c(O) = 744
c(l) = 196, 884
c(2) = 21, 4 9 ~ , 760
c(3) = 864, 299, 970
c(4) = 20, 245, 856, 256
c(5) = 333, 202, 640, 600
c(6) = 4, 252, 023, 300, 096
c(7) = 44, 656, 994, 071, 935
c(8) = 401, 490, 886, 656, 000
The integers c(n) have a number of interesting arithmetical properties. In
1942 D. H. Lehmer [20] proved that
(n + l)c(n) == 0 (mod 24) for all n ~ 1.
In 1949 Joseph Lehner [23] discovered divisibility properties of a different
kind. For example, he proved that
c(5n) == 0 (mod 25),
c(7n) == 0 (mod 7),
c(lln) == 0 (mod 11).
He also discovered congruences for higher powers of 5, 7, 11 and, in a later
paper [24] found similar results for the primes 2 and 3. In Chapter 4 we will
describe how some of Lehner's congruences are obtained.
An asymptotic formula for c(n) was discovered by Petersson [31] in 1932.
It states that
e47t.Jl1
c(n) '" M as n ~ 00.
v 2 n
3
/
4
This formula was rediscovered independently by Rademacher [37] in
1938.
The coefficients r(n) in the Fourier expansion of ~ ( r ) have also been
extensively tabulated by D. H. Lehmer [19] and others. The first ten entries
in Lehmer's table are repeated here:
r(l) = 1
r(2) = 24
r(3) = 252
r(4) = 1472
r(5) = 4830
r(6) = 6048
r(7) =  16744
r(8) = 84480
r(9) = 113643
r(10) = 115920.
Lehmer has conjectured that r(n) i= 0 for all n and has verified this for all
n < 214928639999 by studying various congruences satisfied by r(n). For
papers on r(n) see Section F35 of [27].
22
Exercises for Chapter 1
Exercises for Chapter 1
1. Given two pairs of complex numbers (Wi' W
z
) and (Wi" W
z
') with nonreal ratios
WZ/w
l
and WZ'/w
l
'. Prove that they generate the same set of periods if, and only if,
there is a 2 x 2 matrix (: :) with integer entries and determinant ±1 such that
2. Let S(O) denote the sum of the zeros of an elliptic function f in a period parallelo
gram, and let S( (0) denote the sum of the poles in the same parallelogram. Prove
that S(O)  S( (0) is a period of I [Hint: Integrate
3. (a) Prove that p(u) = p(v) if, and only if, U  v or u + v is a period of p.
(b) Let ai' ... , an and b1, ... , b
m
be complex numbers such that none of the numbers
p(ai)  p(bj) is zero. Let
f(z) = }), [p(z)  p(ak)] III [p(z)  p(b,)].
Prove that f is an even elliptic function with zeros at ai' ... , an and poles at
bl ,···, bm •
4. Prove that every even elliptic function f is a rational function of p, where the
periods of p are a subset of the periods of I
5. Prove that every elliptic function f can be expressed in the form
where R1 and Rz are rational functions and p has the same set of periods as f
6. Let f and 9 be two elliptic functions with the same set of periods. Prove that there
exists a polynomial P(x, y), not identically zero, such that
P[f(z), g(z)] = C
where C is a constant (depending on f and g but not on z).
7. The discriminant of the polynomial f(x) = 4(x  xd(x  xz)(x  X3) is the
product 16{(xz  Xd(X3  XZ)(X3  xd}z. Prove that the discriminant of f(x) =
4x
3
 ax  b is a
3
 27b
Z
•
8. The differential equation for p shows that p'(z) = 0 if z = w
l
/2, wz/2 or
(Wi + w
z
)/2. Show that
p"( = 2(el  e2)(el  e3)
and obtain corresponding formulas for p"(w
z
/2) and p"((w
l
+ w
z
)/2).
23
1: Elliptic functions
9. According to Exercise 4, the function gu(2z) is a rational function of gu(z). Prove
that, in fact,
(2z) = + + 293 = _ +
  93 4 (z)
10. Let 01
1
and W2 be complex numbers with nonreal ratio. Letf(z) be an entire function
and assume there are constants a and b such that
f(z + wd = af(z), f(z + w
2
) = bf(z),
for all z. Prove that f(z) = Ae
Bz
, where A and B are constants.
11. If k 2 and r E H prove that the Eisenstein series
G
2k
(r) = L (m + nr)2k
(m, n) *(0,0)
has the Fourier expansion
12. Refer to Exercise 11. If r E H prove that
G
2k
(  l/r) = r
2k
G
2k
(r)
and deduce that
G
2k
(i/2) = ( 4lG
2k
(2i) for all k 2,
G
2k
(i) = 0 if k is odd,
G2k(e21ti/3) = 0 if k =1= 0 (mod 3).
13. Ramanujan's tau function r(n) is defined by the Fourier expansion
00
L\( r) = (2n)12 L r(n)e
21tint
,
n=1
derived in Theorem 1.19. Prove that
where fog denotes the Cauchy product of two sequences,
n
(f 0 g)(n) = L f(k)g(n  k),
k=O
and O"a(n) = Ldln d
a
for n 1, with 0"3(0) = 2lo, 0"5(0) = 
[Hint: Theorem 1.18.]
14. A series of the form 1 f(n)x
n
/( 1  x
n
) is called a Lambert series. Assuming
absolute convergence, prove that
00 xn 00
L f(n) n = L F(n)x
n
,
n=l Ix n=1
where
F(n) = L f(d).
dIn
24
Exercises for Chapter 1
Apply this result to obtain the following formulas, valid for Ix I < 1.
00 j1(n)xn
(a) L 1n = X.
n=l  X
00 na.xn 00
(c) L n = L O"a.(n)x
n
.
n= 1 1  x n= 1
00 <p(n)x
n
x
(b) L 1 _ n = (1 _ )2'
n=l X X
00 A(n)xn 00
(d) L n = L x
n2
.
n= 1 1  x n= 1
(e) Use the result in (c) to express g2(r) and g3(r) in terms of Lambert series in
x = e
21tir
•
Note. In (a), j1(n) is the Mobius function; in (b), <p(n) is Euler's totient; and in (d),
A(n) is Liouville's function.
15. Let
00 n5xn
G(x) = " ~ 1 1  x"'
and let
F(x) = L
n=l
(n odd)
n
5
x
n
1 + X
n
(a) Prove that F(x) = G(x)  34G(x
2
) + 64G(x
4
).
(b) Prove that
31
504
(c) Use Theorem 12.17 in [4] to prove the more general result
::G n
4k
+ 1 2
4k
+ I  1
2: 1 + e
n7r
= 8k + 4 B4k + 2'
n = 1
(nodd)
25
The modular group
and modular functions
2.1 Mobius transformations
In the foregoing chapter we encountered unimodular transformations
, ar + b
r=
cr + d
where a, b, c, d are integers with ad  bc = 1. This chapter studies such
transformations in greater detail and also studies functions which, like
J(r), are invariant under unimodular transformations. We begin with some
remarks concerning the more general transformations
(1)
f(z) = az + b
cz + d
where a, b, c, d are arbitrary complex numbers.
Equation (1) definesj(z) for all z in the extended complex number system
C* = C U {oo} except for z = dlc and z = 00. We extend the definition
ofjto all of C* by defining
and
a
j(oo)=,
c
(2)
with the usual convention that zlO = 00 if z i= O.
First we note that
f(w)  f(z) = (ad  bc)(w  z)
(cw + d)(cz + d)'
which shows thatjis constant if ad  bc = O. To avoid this degenerate case
we assume that ad  bc i= O. The resulting rational function is called a
26
2.1: Mobius transformations
Mobius transformation. It is analytic everywhere on C* except for a simple
pole at z = die.
Equation (2) shows that every Mobius transformation is onetoone on
C*. Solving (1) for z in terms off(z) we find
df(z)  b
z=
ef(z) + a'
so f maps C* onto C*. This also shows that the inverse function f  1 is a
Mobius transformation.
Dividing by w  z in (2) and letting w ~ z we obtain
, ad  be
f(z) = (cz + df'
hence f'(z) =I °at each point of analyticity. Therefore f is conformal every
where except possibly at the pole z = die.
Mobius transformations map circles onto circles (with straight lines
being considered as special cases of circles). To prove this we consider the
equation
(3) Azz + Bz + Bz + C = 0,
where A and C are real. The points on any circle satisfy such an equation
with A =I 0, and the points on any line satisfy such an equation with A = 0.
Replacing z in (3) by (aw + b)/(ew + d) we find that w satisfies an equation
of the same type,
A'ww + B'w + B'w + C' = °
where A' and C' are also real. Hence every Mobius transformation maps a
circle or straight line onto a circle or straight line.
A Mobius transformation remains unchanged if we multiply all the
coefficients a, b, e, d by the same nonzero constant. Therefore there is no loss
in generality in assuming that ad  be = 1.
For each Mobius transformation (1) with ad  be = 1 we associate the
2 x 2 matrix
Then det A = ad  be = 1. If A and B are the matrices associated with
Mobius transformations f and g, respectively, then it is easy to verify
that the matrix product AB is associated with the composition fog, where
(fog)(z) = f(g(z)). The identity matrix I = ( ~ ~ ) is associated with the
identity transformation
1z + °
f(z) = z = 02 + l'
27
2: The modular group and modular functions
and the matrix inverse
AI = ( d b)
c a
is associated with the inverse off,
dz  b
fl(Z) = 
cz + a
Thus we see that the set of all Mobius transformations with ad  bc = 1
forms a group under composition. This chapter is concerned with an impor
tant subgroup in which the coefficients a, b, c, d are integers.
2.2 The modular group r
The set of all Mobius transformations of the form
, aT + b
T =
CT + d'
where a, b, c, d are integers with ad  bc = 1, is called the modular group and
is denoted by r. The group can be represented by 2 x 2 integer matrices
A = G~ ) with det A = 1,
provided we identify each matrix with its negative, since A and  A represent
the same transformation. Ordinarily we will make no distinction between
the matrix and the transformation. If A = (: ~ ) we write
aT + b
AT = CT + d·
The first theorem shows that r is generated by two transformations,
TT = T + 1
and ST =
Theorem 2.1. The modular group r is generated by the two matrices
T = ( ~ ~ )
and
s = ( ~
1)
o·
That is, every A in r can be expressed in theform
A = T"lST"2S ... ST"k
where the ni are integers. This representation is not unique.
28
2.2: The modular group r
PROOF. Consider first a particular example, say
A = C;
We will express A as a product of powers of Sand T. Since S2 = I, only the
first power of S will occur.
Consider the matrix product
AT" = 2DG = C; 1;::
Note that the first column remains unchanged. By a suitable choice of n
we can make 111n + 25/ < 11. For example, taking n =  2 we find
lIn + 25 = 3 and
AT
2
= (4 1).
11 3
Thus by multiplying A by a suitable power of T we get a matrix ( : ; ) with
IdI < /C /. Next, multiply by S on the right:
AT
2
S = C; = G
This interchanges the two columns and changes the sign of the second column.
Again, multiplication by a suitable power of T gives us a matrix with
Id I < /C /. In this case we can use either T
4
or T
3
. Choosing T
4
we find
AT
2
ST
4
= (3
1
4)(1 4) (1 0)
11 0 1 = 3 1·
Multiplication by S gives
AT
2
ST
4
S =
Now we multiply by T
3
to get
Solving for A we find
A = ST
3
ST
4
ST
2
.
At each stage there may be more than one power of T that makes IdI < Ie/
so the process is not unique.
To prove the theorem in general it suffices to consider those matrices
A = (: ;) in r with c O. We use induction on c.
29
2: The modular group and modular functions
If e = 0 then ad = 1 so a = d = ±1 and
A = (±~ ± ~ ) = ( ~ ± ~ ) = T±b.
Thus, A is a power of T.
If e = 1 then ad  b = 1 so b = ad  1 and
A = ( ~ ad d 1) = ( ~ ~ ) ( ~  ~ ) ( ~ ~ ) = raST
d
•
Now assume the theorem has been proved for all matrices A with lower
lefthand element < e for some e ~ 1. Since ad  be = 1 we have (e, d) = 1.
Dividing d by e we get
d = eq + r, where 0 < r < e.
Then
and
ATqS = (: a
q
; b ) ( ~  ~ ) = (a
q
; b =:)
By the induction hypothesis, the last matrix is a product of powers of S
and T, so A is too. This completes the proof. D
2.3 Fundamental regions
Let G denote any subgroup of the modular group r. Two points rand r'
in the upper halfplane H are said to be equivalent under G if r' = Ar for
some A in G. This is an equivalence relation since G is a group.
This equivalence relation divides the upper halfplane H into a disjoint
collection of equivalence classes called orbits. The orbit Gr is the set of all
complex numbers of the form Ar where A E G.
We select one point from each orbit; the union of all these points is
called a fundamental set of G. To deal with sets having nice topological
properties we modify the concept slightly and define a fundamental region
of G as follows.
Definition. Let G be a subgroup of the modular group r. An open subset
R
G
of H is called a fundamental region of G if it has the following two
properties:
(a) No two distinct points of R
G
are equivalent under G.
(b) If r E H there is a point r' in the closure of R
G
such that r' is equivalent
to r under G.
30
2.3: Fundamental regions
For example, the next theorem will show that a fundamental region R
r
of the full modular group r consists of all r in H satisfying the inequalities
Irl> 1, Ir+il<l.
This region is the shaded portion of Figure 2.1.
u = 1
r = u + iv, v > 0
/' It I = 1
U
1 o
1
"2
Figure 2.1 Fundamental region of the modular group
The proof will use the following lemma concerning fundamental pairs of
periods.
Lemma 1. Given WI" w
2
' with W
2
'/W
l
' not real, let
n = {mw
l
' + nW2' : m, n integers}.
Then there exists afundamental pair (WI' (2) equivalent to (WI" W2') such
that
and such that
PROOF. We arrange the elements of n in a sequence according to increasing
distances from the origin, say
n = {O, Wb W2' ...}
where
and argw
n
< argwn+l if /wn/=/Wn+l/'
31
2: The modular group and modular functions
Let WI = WI and let W
2
be the first member of this sequence that is not a
multiple of WI' Then the triangle with vertices 0, W
b
W
2
contains no element
of Q except the vertices, so (w
b
(
2
) is a fundamental pair which spans the
set Q. Therefore there exist integers a, b, c, d with ad  bc = ±1 such that
(::) = (:
If ad  bc = 1 we can replace c by c, d by d, and WI by WI and the
same equation holds, except now ad  bc = 1. Because of the way we have
chosen W
b
W2 we have
and
since WI ±W
2
are periods in Q occurring later than W
2
in the sequence. D
Theorem 2.2. If r' E H, there exists a complex number r in H equivalent to r'
under r such that
Irl 1,
and
PROOF. Let WI' = 1, w
2
' = r' and apply Lemma 1 to the set of periods
Q = {m + nr' : m, n integers}. Then there exists a fundamental pair W
b
W
2
with Iw
2
1 Iwtl, IW
t
± w
2
1 Iw
2
1. Let T = w
2
/w
t
. Then T = (:
with ad  bc = 1 and
Irl 1, Ir ± 11 Irl. o
Note. Those r in H satisfying Ir ± 11 Ir I are also those satisfying
Theorem 2.3. The open set
R
r
= {r E H : Ir I > 1, Ir + r I < I}
is a fundamental region for r. Moreover, if A E r and if Ar = r for some
r in R
r
, then A = 1. In other words, only the identity element has fixed
points in R
r
.
PROOF. Theorem 2.2 shows that if r' E H there is a point r in the closure of
R
r
equivalent to r' under r. To prove that no two distinct points of R
r
are
equivalent under r, let T' = AT where A = (: We show first that
Im(r') < Im(r) if r E R
r
and c =1= O. We have
, Im(r)
Im(T) = leT + d1 2 •
32
2.3: Fundamental regions
IfrER
r
and c =1= 0 we have
Icr + dl
2
= (cr + d)(cr + d) = c
2
rr + cd(r + i) + d
2
> c
2
 Icdl + d
2
•
If d = 0 we find Icr + d12 > c
2
2 1. If d =1= 0 we have
c
2
 Icdl + d
2
= (Icl  Idl)2 + Icdl 2 Icdl 2 1
so again Icr + d1
2
> 1. Therefore c =1= 0 implies I cr + d1
2
> 1 and hence
Im(r') < Im(r). In other words, every element A of r with c =1= 0 decreases
the ordinate of each point r in R
r
.
Now suppose both rand r' are equivalent interior points of R
r
. Then
, ar + b
r =
cr + d
and
dr'  b
r= .
cr' + a
If c =1= 0 we have both Im(r') < Im(r) and Im(r) < Im(r'). Therefore c = 0
so ad = 1, a = d = ±1, and
A=(: ~ ) = ( ± ~ ± ~ ) = T ± b .
But then b = 0 since both rand r' are in R
r
so r = r'. This proves that no
two distinct points of R
r
are equivalent under r.
Finally, if Ar = r for some r in R
r
, the same argument shows that c = 0,
a = d = ±1, so A = I. This proves that only the identity element has fixed
points in R
r
. 0
Figure 2.2 shows the fundamental region R
r
and some of its images under
transformations of the modular group. Each element of r maps circles into
circles (where, as usual, straight lines are considered as special cases of
circles). Since the boundary curves of Rr are circles orthogonal to the real
T
2
1
1
'!
o
1
"2
2
Figure 2.2 Images of the fundamental region R
r
under elements of r
33
2: The modular group and modular functions
axis, the same is true of every image f(R
r
) under the elements.f of r. The
set of all images f(R
r
), where fEr, is a collection of nonoverlapping open
regions which, together with their boundary points, cover all of H.
2.4 Modular functions
Definition. A functionfis said to be modular if it satisfies the following three
conditions:
(a) f is meromorphic in the upper halfplane H.
(b) f(Ar) = f(r) for every A in the modular group r.
(c) The Fourier expansion off has the form
00
f(r) = L a(n)e21tint.
n= m
Property (a) states thatfis analytic in H except possibly for poles. Property
(b) states that f is invariant under all transformations of r. Property (c) is
a condition on the behavior offat the point r = ioo. If x = e
21tit
the Fourier
series in (c) is a Laurent expansion in powers of x. The behavior offat ioo is
described by the nature of this Laurent expansion near O. If m > 0 and
a(  m) =1= 0 we say that f has a pole of order m at ioo. If m ~ 0 we say f is
analytic at ioo. Condition (c) states that f has at worst a pole of order m at
ioo.
The function J is a modular function. It is analytic in H with a first order
pole at ioo. Later we show that every modular function can be expressed as
a rational function of J. The proof of this depends on the following property
of modular functions.
Theorem 2.4. Iff is modular and not identically zero, then in the closure of the
fundamental region R
r
, the number of zeros off is equal to the number of
poles.
Note. This theorem is valid only with suitable conventions at the boundary
points of R
r
. First of all, we consider the boundary of R
r
as the union of
four edges intersecting at four vertices p, i, P + 1, and ioo, where p = e
21ti
/
3
(see Figure 2.3). The edges occur in equivalent pairs (1), (4) and (2), (3).
Iffhas a zero or pole at a point on an edge, then it also has a zero or pole
at the equivalent point on the equivalent edge. Only the point on the leftmost
edge (1) or (2) is to be counted as belonging to the closure of Rr.
The order of the zero or pole at the vertex p is to be divided by 3; the order
at i is to be divided by 2; the order at ioo is the order of the zero or pole at
x = 0, measured in the variable x = e
21tit
•
34
2.4: Modular functions
ooi
  ~  
(1)
p
./
"
/
/
(2) (3)
Figure 2.3
......
"
,
(4)
p + 1
PROOF. Assume first that j has no zeros or poles on the finite part of the
boundary of R
r
. Cut R
r
by a horizontal line, Im(r) = M, where M > 0 is
taken so large that all the zeros or poles ofj are inside the truncated region
which we call R. [Ifjhad an infinite number of poles in R
r
they would have
an accumulation point at ioo, contradicting condition (c). Similarly, since j
is not identically zero, j cannot have an infinite number of zeros in R
r
.]
Let oR denote the boundary of the truncated region R. (See Figure 2.4.)
Let Nand P denote the number of zeros and poles ofjinside R. Then
1 i f'(r) 1 {f f f f f}
N  P =   dr =  + + + +
2ni oR f(r) 2ni (1) (2) (3) (4) (5)
where the path is split into five parts as indicated in Figure 2.5. The integrals
along (1) and (4) cancel because of periodicity. They also cancel along (2)
and (3) because (2) gets mapped onto (3) with a reversal of direction under
} + iM ,.., ! + iM
p
R
Figure 2.4
t
p + 1
35
2: The modular group and modular functions
(5)
(2)
~
(3)
~
Figure 2.5
the mapping u = S(r) = l/r, or r = Sl U = S(u). The integrand remains
unchanged because f[S(u)] = f(u) implies f'[S(u)]S'(u) = f'(u) so
f'(r) dr = f'[S(u)] S'(u) du = f'(u) duo
f(r) f[S(u)] f(u)
Thus we are left with
N  P = _1 f f'(r) dr.
2ni (5) f(r)
We transform this integral to the xplane, x = e
21tit
• As r varies on the
horizontal segment r = u + iM, t ~ u ~ t, we have
so x varies once around a circle K of radius e 21tM about x = 0 in the negative
direction. The points above this segment are mapped inside K, so f has no
zeros or poles inside K, except possibly at x = O. The Fourier expansion
gives us
(
am (
f r) = ;n + ... = F x),
x
say, with
Hence
f'(r) = F'(x) ~ : '
f'(r) dr = F'(x) dx.
f(r) F(x)
N  P = ~ f f'(r) dr =  ~ , [ F'(x) dx = (N
F
 P
F
) = P
F
 N
F
,
2nl (5) f(r) 2nl JK F(x)
where NF and P
F
are the number of zeros and poles of F inside K.
36
2.4: Modular functions
If there is a pole of order m at x = °then m > 0, NF = 0, PF = m so
PF  N F = m, and
N = P + m.
Thereforejtakes on the value °in R
r
as often as it takes the value 00.
If there is a zero of order n at x = 0, then m =  n so PF = 0, N F = n,
hence
N + n = P.
Again,jtakes the value °in R
r
as often as it takes the value 00. This proves
the theorem ifJhas no zeros or poles on the finite part of the boundary of R
r
.
Ifjhas a zero or a pole on an edge but not at a vertex, we introduce detours
in the path of integration so as to include the zero or pole in the interior of R,
as indicated in Figure 2.6. The integrals along equivalent edges cancel as
before. Only one member of each pair of new zeros or poles lies inside the new
region and the proof goes through as before, since by our convention only
one of the equivalent points (zero or pole) is considered as belonging to the
closure of R
r
.
1
Figure 2.6
If j has a zero or pole at a vertex p or i we further modify the path of
integration with new detours as indicated in Figure 2.7. Arguing as above we
find
1 {(II) 1 J,1
/
2+iM}j"(7:)
N  P =  + + +  d7:
2ni Ct C3 C2 1/2 + iM j'(7:)
= _1 {(I + 1)+ 1}J'(7:) d7: + m
2ni Ct C3 C2 j(7:) ,
where x
m
is the lowest power of x occurring in the Laurent expansion near
x = 0, x = e
21tit
.
37
2: The modular group and modular functions
1 + iM ,.., t + iM
1
'.
p + 1
Figure 2.7
Near the vertex p we write
f(r) = (r  p)kg(r), where g(p) # o.
The exponent k is positive iffhas a zero at p, and negative iffhas a pole at p.
On the path C1 we write r  p = re
i8
where r is fixed and rx ~ e~ nl2
where rx depends on r. Then
f'(r) = _k_ + g'(r)
f(r) r  p g(r)
and
1 f f'(r) 1 fa (k g'(p + re
iO
)) iO'
 dr =  . + . re 1 de
2ni Ct f(r) 2ni 1t/2 re
'O
g(p + re
'O
)
_ krx' r fa g'(p + re
i8
) i6 , _ ~ _
  +  '0 e dO, where ex  2 ex.
2n 2n 1t/2 g(p + ref )
As r + 0, the last term tends to 0 since the integrand is bounded. Also,
rx' + n/3 as r + 0 so
. 1 f f'(r) k
hm dr = .
r ~ O 2ni Ct f(r) 6
Similarly,
lim _1 f f'(r) dr =  ~
r ~ O 2ni C3 f(r) 6
38
2.5: Special values of J
so
lim
1
(J. + J. )f'(r) dr =  ~
r+O 2ni Ct C3 f( r) 3·
Similarly, near the vertex i we write
f(r) = (r  i)'h(r), where h(i) # 0
and we find, in the same way,
lim _1 J. f'(r) dr =  i.
r'O 2rci C2 f( r) 2
Therefore we get the formula
k I
N  P = m "3  2·
Iffhas a pole at x = 0, and zeros at p and i, then m, k and I are positive and
we have
k I
N +"3 + 2= P + m.
The left member counts the number of zeros offin the closure of R
r
(with the
conventions agreed on at the vertices) and the right member counts the
number of poles. Iff has a zero of order n at x = 0 then m =  n and the
equation becomes
k I
N + n + 3+ 2= P.
Similarly, if f has a pole at p or at i the corresponding term k/3 or 1/2 is
negative and gets counted along with P. This completes the proof. D
Theorem 2.5. If f is modular and not constant, then for every complex c the
functionf  c has the same number of zeros as poles in the closure of R
r
.
In other words,ftakes on every value equally often in the closure of R
r
.
PROOF. Apply the previous theorem tof  c.
Theorem 2.6. Iff is modular and bounded in H then f is constant.
PROOF. Sincefis bounded it omits a value sofis constant.
D
D
2.5 Special values of J
Theorem 2.7. The function J takes every value exactly once in the closure of
R
r
. In particular, at the vertices we have
J(p) = 0, J(i) = 1, J(ioo) = 00.
There is afirst order pole at ioo, a triple zero at p, and J(r)  1has a double
zero at r = i.
39
2: The modular group and modular functions
PROOF. First we verify that 92(P) = 0 and 93(i) = o. Since p
3
= 1 and
p2 + P + 1 = 0 we have
~ (p) _ " ,,1 _ ~ " 1
60 92  tn (m + np)4 tn (
mp
3 + np)4  p4 tn (
mp
2 + n)4
1 1 1 1 1
= PJ,:n (n  m  mpt = PM ~ N (N + Mp)4 = 60p g2(P),
so g2(P) = O. A similar argument shows that g3(i) = O. Therefore
J( ) = 92
3
(P) = 0 and J(.) = 92
3
(i) = 1
p ~ ( p ) 1 92
3
(i) .
The multiplicities are a consequence of Theorem 2.4.
2.6 Modular functions as rational functions
of J
D
Theorem 2.8. Every rational function of J is a modular function. Conversely,
every modular function can be expressed as a rational function of J.
PROOF. The first part is clear. To prove the second, suppose f has zeros at
Z l' ... , Zn and poles at PI' ... , Pn with the usual conventions about multi
plicities. Let
g(7:) = Ii J(7:)  J(Zk)
k= 1 J (r)  J (pk)
where a factor 1 is inserted whenever Zk or Pk is 00. Then 9 has the same zeros
and poles asfin the closure of R
r
, each with proper multiplicity. Therefore
f /9 has no zeros or poles and must be constant, so f is a rational function
ofJ. D
2.7 Mapping properties of J
Theorem 2.7 shows that J takes every value exactly once in the closure
of the fundamental region R
r
. Figure 2.8 illustrates how R
r
is mapped by
J onto the complex plane.
The left half of R
r
(the shaded portion of Figure 2.8a) is mapped onto the
upper halfplane (shaded in Figure 2.8b) with the vertical part of the boundary
mapping onto the real interval (  00, 0]. The circular part of the boundary
maps onto the interval [0, 1], and the portion of the imaginary axis v > 1,
u = 0 maps onto the interval (1, +(0). Points in R
r
symmetric about the
imaginary axis map onto conjugate points in J(R
r
). The mapping is con
formal except at the vertices r = i and r = p where angles are doubled
and tripled, respectively.
40
2.7: Mapping properties of J
J
..
/
P
"
I
"
/
\
I
\
I
\
\
u
1 0 1
0= J(p) 1 = J(i)
(a) (b)
Figure 2.8
These mapping properties can be demonstrated as follows. On the
imaginary axis in R
r
we have r = iv hence x = e
21tit
= e 21tv > 0, so the
Fourier series
shows that. J(iv) is real. Since J(i) = 1 and J(iv) ~ +00 as v ~ +00 the
portion of the imaginary axis 1 ~ v < +00 gets mapped onto the real axis
1 ~ J(r) < +00.
On the left boundary of R
r
we have r = 1 + iv, hence x = e
21tit
=
e21tve1ti = _e
21tv
< O. For large v (small x) we have J( 1 + iv) < 0 so
J maps the line u = 1 onto the negative real axis. Since J(p) = 0 and
J( (0) = 00, the left boundary of R
r
is mapped onto the line  00 < J(r) ~ O.
As the boundary of R
r
is traversed counterclockwise the points inside R
r
lie on the left, hence the image points lie above the real axis in the image
plane.
Finally, we show that J takes conjugate values at points symmetric about
the imaginary axis, that is,
J(r) = J(i).
To see this, write r = u + iv. Then
x = e
21tit
= e
21ti
(u + iv) = e  21tv
e
21tiu
and
Thus rand  i correspond to conjugate points x and X, but the Fourier
series for J has real coefficients so J(r) and J(  i) are complex conjugates.
41
2: The modular group and modular functions
In particular, on the circular arc Ti = 1 we have  i =  liT, hence
J(  i) = J( liT) = J(T) so J is real on this arc.
2.8 Application to the inversion problem for
Eisenstein series
In the Weierstrass theory of elliptic functions the periods Wt, W2 determine
the invariants g2 and g3 according to the equations
(4)
A fundamental problem is to decide whether or not the invariants g2 and g3
can take arbitrary prescribed values, subject only to the necessary condition
g2
3
 27g3
2
=1= O. This is called the inversion problem for Eisenstein series
since it amounts to solving the equations in (4) for W
t
and W
2
in terms of g2
and g3. The next theorem shows that the problem has a solution.
Theorem 2.9. Given two complex numbers a2 and a3 such that a2
3
 27a3
2
=1= O.
Then there exist complex numbers W
t
and W
2
whose ratio is not real such
that
and
PROOF. We consider three cases: (1) a2 = 0; (2) a3 = 0; (3) a2a3 =1= O.
Case 1. If a2 = 0 then a3 =1= 0 since a2
3
 27a32 =1= O. Let W
t
be any
complex number such that
and let W
2
= PWt, where p = e
21ti
/
3
. We know that g3(1, p) =1= 0 because
g2(1, p) = 0 and ~ ( 1 , p) = g2
3
 27g
3
2
=1= O. Then
1
g2(W
t
, w
2
) = g2(W
b
wtp) = 4g2(1, p) = 0 = a2
W
t
and
Case 2. If a3 = 0 then a2 =1= 0 and we take W
t
to satisfy
42
(5)
2.9: Application to Picard's theorem
and let W2 = iw
t
. Then
1
g2(W
b
w
2
) = g2(Wb iw
t
) = 4g2(1, i) = a2
W
t
and
1
g3(W
b
w
2
) = g3(Wb iw
t
) = 6g3(1, i) = 0 = a3'
W
t
Case 3. Assume a2 =1= 0 and a3 =1= O. Choose a complex r with 1m r > 0
such that
a2
3
J(r) = 3 27 2'
a2  a3
Note that J(r) =1= 0 since a2 =1= 0 and that
J(r)  1 27a32
J(r) ~ .
For this r choose W
t
to satisfy
and let W
2
= rw
t
. Then
g2(W
t
, w
2
) _ W
t
4
g2
(1, r) _ 2 g2(1, r) _ a2
g3(W
t
,w
2
)  W
t
6
g3
(1, r)  W
t
g3(1,r)  a3'
so
(6)
But we also have
J(r)  1
J(r)
27g
3
2
(W
t
, w
2
)
g2
3
(W
b
W2)
27(a3/a2)2
g2
2(W
b
w
2
)
g2
3
(Wb W2)
Comparing this with (5) we find that g2(W
t
, W2) = a2 and hence by (6) we
also have g3(Wb W2) = a3' This completes the proof. D
2.9 Application to Picard's theorem
The modular function J can be used to give a short proof of a famous theorem
of Picard in complex analysis.
Theorem 2.10. Every nonconstant entire function attains every complex value
with at most one exception.
Note. An example is the exponential function f(z) = e
Z
which omits
only the value O.
43
2: The modular group and modular functions
PROOF. We assume f is an entire function which omits two values, say a
and b, a =1= b, and show thatfis constant. Let
( )
_ f(z)  a
g Z  b .
a
Then g is entire and omits the values 0 and 1.
The upper halfplane H is covered by the images of the closure of the
fundamental region R
r
under transformations of r. Since J maps the closure
of R
r
onto the complex plane, J maps the halfplane H onto an infinite
sheeted Riemann surface with branch points over the points 0, 1 and 00
(the images of the vertices p, i and 00, respectively). The inverse function J
1
maps the Riemann surface back onto the closure of the fundamental region
R
r
. Since J'(r) =1= 0 if r =1= p or r =1= i and since J'(p) == J'(i) == 0, each single
valued branch of J  1 is locally analytic everywhere except at 0 == J(p),
1 = J(i), and 00 = J( (0). For each singlevalued branch of J
1
the composite
function
h(z) = J
1
[g(z)]
is a singlevalued function element which is locally analytic at each finite
z since g(z) is never 0 or 1. Therefore h is arbitrarily continuable in the entire
finite zplane. By the monodromy theorem, the continuation of h exists as a
singlevalued function analytic in the entire finite zplane. Thus h is an entire
function and so too is
qJ(z) = eih(z).
But 1m h(z) > 0 since h(z) E H so
IqJ(z) I == eImh(z) < 1.
Therefore qJ is a bounded entire function which, by Liouville's theorem,
must be constant. But this implies h is constant and hence g is constant since
g(z) = J[h(z)]. Thereforefis constant sincef(z) == a + (b  a)g(z). D
Exercises for Chapter 2
In these exercises, r denotes the modular group, Sand T denote its gen
erators, S(r) = l/r, T(r) = r + 1, and I denotes the identity element.
1. Find all elements A of r which (a) commute with S; (b) commute with ST.
2. Find the smallest integer n > 0 such that (ST)" = I.
3. Determine the point r in the fundamental region R
r
which is equivalent to
(a) (8 + 60/(3 + 20; (b) (10i + 11)/(6i + 12).
4. Determine all elements A of r which leave i fixed.
5. Determine all elements A of r which leave p = e
21ti
/
3
fixed.
44
Exercises for Chapter 2
QUADRATIC FORMS AND THE MODULAR GROUP
The following exercises relate quadratic forms and the modular group r. We
consider quadratic forms Q(x, y) = ax
2
+ bxy + cy2 in x and y with real
coefficients a, b, c. The number d = 4ac  b
2
is called the discriminant of
Q(x, y).
6. If x and yare subjected to a unimodular transformation, say
(1) x = lXX' + Py', y = yx' + 8y', where (; ~ ) E r,
prove that Q(x, y) gets transformed to a quadratic form Ql(X', y') having the same
discriminant. Two forms Q(x, y) and Ql(X', y') so related are called equivalent. This
equivalence relation separates all forms into equivalence classes. The forms in a given
class have the same discriminant, and they represent the same integers. That is, if
Q(x, y) = n for some pair of integers x and y, then Ql(X', y') = n for the pair of
integers x', y' given by (1).
In Exercises 7 thru 10 we consider forms ax
2
+ bxy + cy2 with d > 0,
a > 0, and c > O. The associated quadratic polynomial
f(z) = az
2
+ bz + c
has two complex roots. The root T with positive imaginary part is called the
representative of the quadratic form Q(x, y) = ax
2
+ bxy + cy2.
7. (a) If d is fixed, prove that there is a onetoone correspondence between the set
of forms with discriminant d and the set of complex numbers r with Im(r) > o.
(b) Prove that two quadratic forms with discriminant d are equivalent if and only if
their representatives are equivalent under r.
Note. A reduced form is one whose representative TERr. Thus, two
reduced forms are equivalent if and only if they are identical. Also, each
class of equivalent forms contains exactly one reduced form.
8. Prove that a form Q(x, y) = ax
2
+ bxy + cy2 is reduced if, and only if, either
a < b s a < cor 0 S b s a = c.
9. Assume now that the form Q(x, y) = ax
2
+ bxy + cy2 has integer coefficients
a, b, c. Prove that for a given d there are only a finite number of equivalence classes
with discriminant d. This number is called the class number and is denoted by h(d).
Hint: Show that 0 < a S Jdi3 for each reduced form.
10. Determine all reduced forms with integer coefficients a, b, c and the class number
h(d) for each d in the interval 1 s d s 20.
CONGRUENCE SUBGROUPS
The modular group r has many subgroups of special interest in number
theory. The following exercises deal with a class of subgroups called con
gruence subgroups. Let
and
45
2: The modular group and modular functions
be two unimodular matrices. (In this discussion we do not identify a matrix
with its negative.) If n is a positive integer write
A == B (mod n) whenever a == rJ., b == {3, c == y and d == {) (mod n).
This defines an equivalence relation with the property that
implies
AlB
l
== A
2
B
2
(mod n)
Hence
and
and A
l

l
== A
2
1 (mod n).
A == B (mod n) if, and only if, AB
l
== I (mod n),
where I is the identity matrix. We denote by r(n) the set of all matrices in r
congruent modulo n to the identity. This is called the congruence subgroup
of level n (stufe n, in German).
Prove each of the following statements:
11. r(n) is a subgroup of r. Moreover, if BE r(n) then A1BA E r(n) for every A in r.
That is, r<n) is a normal subgroup of r.
12. The quotient group r /r<n) is finite. That is, there exist a finite number of elements of
r, say A1, ... , A
k
, such that every B in r is representable in the form
B = AiB<n) where 1 sis k and B<n) E r<n).
The smallest such k is called the index of r<n) in r.
13. The index of r<n) in r is the number of equivalence classes of matrices modulo n.
The following exercises determine an explicit formula for the index.
14. Given integers a, b, c, d with ad  bc == 1 (mod n), there exist integers Ct, [3, y, b
such that Ct == a, [3 == b, y = c, b == d (mod n) with ab  [3y = 1.
15. If (m, n) = 1 and A E r there exists A in r such that
A == A (mod n) and A == I (modm).
16. Letf(n) denote the number of equivalence classes of matrices modulo n. Thenfis a
multiplicative function.
17. If a, b, n are integers with n ~ 1 and (a, b, n) = 1 the congruence
ax  by == 1 (mod n)
has exactly n solutions, distinct mod n.(A solution is an ordered pair (x, y) of integers.)
18. For each prime p the number of solutions, distinct mod pr, of all possible congruences
of the form
ax  by == 1 (mod pr), where (a, b, p) = 1,
is equal to f(pr).
19. If p is prime the number of pairs of integers (a, b), incongruent mod pr, which satisfy
the condition (a, b, p) = 1 is p2r 2(p2  1).
20. f(n) = n
3
Ldln f.l(d)/d
2
, where f.l is the Mobius function.
46
The Dedekind eta function
3.1 Introduction
In many applications of elliptic modular functions to number theory the
eta function plays a central role. It was introduced by Dedekind in 1877
and is defined in the halfplane H = {r: Im(r) > O} by the equation
(1)
00
1J(r) = e1tit/12 Il (1  e21tint).
n=1
The infinite product has the form Il (1  x
n
) where x = e
21tit
. If r E H then
Ix I < 1 so the product converges absolutely and is nonzero. Moreover,
since the convergence is uniform on compact subsets of H, 1J(r) is analytic
onH.
The eta function is closely related to the discriminant L\(r) introduced
in Chapter 1. Later in this chapter we snow that
L\(r) = (2n)121J24(r).
This result and other properties of 1J(r) follow from transformation formulas
which describe the behavior of 1J(r) under elements of the modular group r.
For the generator Tr = r + 1 we have
00
(2) 1J(r + 1) = e
1ti
(t+ 1)/12 Il (1  e
21tin
(t+ 1») = e
1ti
/
12
1J(r).
n=1
Consequently, for any integer b we have
(3) 1J(r + b) = e1tib/121J(r).
Equation (2) also shows that 1J24(r) is periodic with period 1.
47
3: The Dedekind eta function
For the other generator ST = l/T we have the following theorem.
Theorem 3.1. If T E H we have
(4) '1( ~
1
) = ( h:)1/2'1(r).
Note. We choose that branch of the square root function Z1/2 which is
positive when z > o.
This chapter gives two different proofs of (4). The first is a short proof of
C. L. Siegel [48] based on residue calculus, and the second derives (4) as a
special case of a more general functional equation which relates
r , ( ~ )
CT + d
to r,(T) when
(: ~ ) E rand c > O.
(See Theorem 3.4.) A third proof, based on interchange of summation in a
conditionally convergent iterated series, is outlined in the exercises.
3.2 Siegel's proof of Theorem 3.1
First we prove (4) for T = iy, where y > 0, and then extend the result to all
Tin H by analytic continuation. If T = iy the transformation formula becomes
r,(i/y) = yl/2r,(iy), and this is equivalent to
log r,(i/y)  log r,(iy) = ! log y.
Now
ny 00
log '1(iy) =  12 + log JI(1  e
2nny
)
ny 00
  + L log(l  e
21tny
) =
12 n=1
ny

12
00 00
L L
n= 1 m= 1
e 21tmny
m
ny 00 1 e 21tmy ny 00 1 1
 12  m ~ l ; ; ; 1  e
2nmy
12 + m ~ l ; ; ; 1  e
2nmy
'
Therefore we are to prove that
00 1 1 00 lIn( 1) 1
(5) m ~ 1 ;;; 1  e
2nmy
 m ~ 1 ;;; 1  e
2nm
/
y
 12 Y  Y =  2log y.
This will be proved with the help of residue calculus.
For fixed y > 0 and n = 1,2, ... , let
1 nNz
Fn(z) =   cot niNz cot 
8z y ,
48
3.2: Siegel's proof of Theorem 3.1
y o
i
Figure 3.1
y
where N = n + !. Let C be the parallelogramjoining the vertices y, i,  y,  i
in that order. (See Figure 3.1.) Inside C, F
n
has simple poles at z = ik/N and
at z = ky/N for k = ± 1, ±2, ... , ±n. There is also a triple pole at z = 0
with residue i(y  Y 1)/24. The residue at z = ik/N. is
1 nik
8nk cot y.
Since this is an even function of k we have
But
n
L
k= n
k,*O
n 1 nik
Res Fn(z) = 2 L 8 k cot .
z = ik/ N k = 1 n Y
.0 _ cos iO _ . e 8 + e
8
cot 1  • .0  1 8 8
sIn 1 e  e
e
28
+ 1 1( 2)
 i e28 _ 1 = i 1  1 _ e28 •
Using this with () = nk/y we get
n 1 n 1 1 n 1 1
L Res F (z) =  L    L ,,
k= n z=ik/N n 4ni k= 1 k 2ni k= 1 k 1  e
21tk
/
Y
•
k,*O
Similarly
n in 1 i n1 1
L Res F (z) =  L    L   ~
k= n z=ky/N n 4n k= 1 k 2n k= lk 1  e
21tkY
•
k,*O
Hence 2ni times the sum of all the residues of Fn(z) inside C is an expression
whose limit as n ~ 00 is equal to the left member of(5). Therefore, to complete
the proof we need only show that
lim fFn(z) dz = ! log y.
n+ 00 C
49
3: The Dedekind eta function
On the edges of C (except at the vertices) the function zFn(z) has, as
n + 00, the limit k on the edges connecting y, i and  y,  i, and the limit t
on the other two edges. Moreover, Fn(z) is uniformly bounded on C for all n
(because N = n + ! and y > 0). Hence by Arzehi's bounded convergence
theorem (Theorem 9.12 in [3]) we have f
lim fFn(z) dz = f lim zFn(z) dz
n+oo C C n+oo Z
+fi}dZ
8 I y J
i
_y Z
= { fY + Ji} dz
4 I Y Z
1 {( ni) (ni )}
= 4:  log y + 2 + 2  log y =
This completes the proof.
1
 2log y.
D
3.3 Infinite product representation for
In this section we express the discriminant L\(T) in terms of 1J(T) and thereby
obtain a product representation of L\(T). The result makes use of the following
property of L\(T).
Theorem 3.2. If (: E r then
d(:: : = (CT + d)
12
d(T).
In particular,
L\(T + 1) = L\(T) and
PROOF. Since L\(w
l
, w
2
) is homogeneous of degree 12 we have
L\(w
b
w
2
) = WI  12L\(1, T) = WI  12L\(T),
where T = W2/Wl. Also,
L\(W
b
w
2
) = L\(w
l
', w
2
')
if (Wb w
2
) and (WI" w
2
') are equivalent pairs of periods. Taking WI = 1,
W
2
= T, WI' = CT + d, w
2
' = aT + b, we find
(
aT + b)
L\(T) = L\(w
b
w
2
) = L\(CT + d, aT + b) = (CT + d)12L\ 1, . D
CT + d
50
3.4: The general functional equation for l1(r)
Theorem 3.3. If r E H and x = e
21tit
we have
(6)
Consequently,
00
L\(r) = (2n)121]24(r) = (2n)12
x
n(1  X
n
)24.
n=1
00 00
(7) Lr(n)x
n
= x n(1  X
n
)24 whenever Ix I < 1
n= 1 n= 1
where r(n) is Ramanujan's taufunction.
PROOF. Let f(r) = L\(r)/1] 24(r). Then f(r + 1) = f(r) and f( l/r) = f(r),
so f is invariant under every transformation in r. Also, f is analytic and non
zero in H because L\ is analytic and nonzero and 1] never vanishes in H.
Next we examine the behavior of f at ioo. We have
00 00
1]24(r) = e
21tit
n(1  e21tint)24 = x n(1  X
n
)24 = x(l + I(x)),
n=1 n=1
where I(x) ,denotes a power series in x with integer coeffil ients. Thus, 1]24(r)
has a first order zero at x = 0. By Theorem 1.19 we also have the Fourier
expansion
(8)
00
L\(r) = (2n)12 Lr(n)x
n
= (2n)12
x
(1 + I(x)).
n=1
Thus, near ioo the function f has the Fourier expansion
(
9) f( ) = L\(r) = (2n)12
x
(1 + I(x)) = (2 )12(1 I())
r 1]24(r) x(l + I(x)) n + x,
so f is analytic and nonzero at ioo. Therefore f is a modular function which
never takes the value 0, so f must be constant. Moreover, (9) shows that
this constant is (2n)12, hence L\(r) = (2n)121]24(r). This proves (6), and (7)
follows from (8). 0
3.4 The general functional equation for 1J(r)
Extracting 24th roots in the relation
L\(ar + b) = (cr + d)12L\(r)
cr + d
and using (6) we find that
11(:: : ~ ) = e(cr + d)1/2
11
(r),
where e
24
= 1. For many applications of 1](r) we require more explicit
information concerning e. This is provided in the next theorem.
51
3: The Dedekind eta function
Theorem 3.4 (Dedekind's functional equation). If (: :) E r, c > 0, and
,. E H, we have
(10) '1(:: : ~ ) = s(a, b, c, d){  i(cr + dW/
2
'1(r)
where
s(a, b, c, d) = exp{ n{al;cd + s( d, C))}
and
(11)
kl r (hr [hrJ 1)
s(h, k) = L       .
r= 1 k k k 2
(13)
Note. The sum s(h, k) in (11) is called a Dedekind sum. Some of its properties
are discussed later in this chapter.
We will prove Theorem 3.4 through a sequence of lemmas. First we note
that Dedekind's formula is a consequence of the following equation, obtained
by taking logarithms of both members of (10),
(
aT + b) (a + d )
(12) log '1 cr + d = log '1(r) +ni u;; + s(  d, c) +1 log{ i(cr + d)}.
From the definition of l1(T) as a product we have
niT 00 • niT 00
log l1(T) =  + L log(1  e
21tlnt
) =   L A( inT),
12 n=l 12 n=l
where A(X) is defined for Re(x) > 0 by the equation
(14)
00 e21tmx
A(X) = log(1  e
21tx
) = L .
m=l m
Equations (12) and (13) give us
Lemma 1. Equation (12) is equivalent to the relation
(15)
00 00 ( aT + b) ni ( aT + b)
LA(inT) = LA in +  T 
n =1 n =1 cT + d 12 CT + d
(
a + d )
+ ni u;; + s( d, c) +! log{  i(cr + d)}.
We shall prove (15) as a consequence of a more general transformation
formula obtained by Sh6 Iseki [17] in 1957. For this purpose it is convenient
to restate (15) in an equivalent form which merely involves some changes
in notation.
52
3.5: Iseki's transformation formula
Lemma 2. Let z be any complex number with Re(z) > 0, and let h, k and H be
any integers satisfying (h, k) = 1, k > 0, hH == 1 (mod k). Then Equation
(15) is equivalent to the formula
(16)
f (z  ih)} = f  iH)}
n= 1 k n= 1 k z
1 n ( 1)
+ 2 log z  12k z  + nis(h, k).
PROOF. Given (: in r, with c > 0, and given r with Im(r) > 0, choose
z, h, k, and H as follows:
k = c, h = d, H = a, z = i(cr + d).
Then Re(z) > 0, and the condition ad  bc = 1 implies  hH  bk = 1, so
(h, k) = 1 and hH == 1 (mod k). Now b = (hH + l)jk and iz = cr + d,
so
iz  d iz + h
r==
c k
and hence
b
iz + h hH + 1 iz ( i)
ar + = H k  k = k H + ; .
Therefore, since cr + d = iz, we have
ar + b = (H +
cr + d k z
Consequently
ar + b 1 i ( 1) a + d i ( 1)
r  cr + d = k(h  H) + k z  =  c + k z 
so
(r ::: = n{a
1
;C
d
)  (z D·
Substituting these expressions in (15) we obtain (16). In the same way we
find that (16) implies (15). D
3.5 Iseki's transformation formula
Theorem 3.5 (Iseki's formula). If Re(z) > °and ° (1 1, ° f3 1, let
00
(17) 1\((1, {3, z) = L {A((r + (1)z  i{3) + A((r + 1  (1)z + i{3)}.
r=O
53
3: The Dedekind eta function
Then if either 0 :::; a :::; 1 and 0 < f3 < 1, or 0 < a < 1 and 0 :::; f3 :::; 1,
we have
Note. The sum on the right of (18), which contains Bernoulli polynomials
Bn(x), is equal to
PROOF. First we assume that 0 < a < 1 and 0 < f3 < 1. We begin with the
first sum appearing in (17) and use (14) to write
(19)
00 00 00 e21timp
L,1((r + a)z  if3) = L L e
21tm
(r+a)z.
r=O r=Om=l m
Now we use Mellin's integral for e
x
which states that
(20)
1 fC+ ooi
e
x
= 2. r(s)x
S
ds,
nl c ooi
where c > 0 and Re(x) > O. This is a special case of Mellin's inversion formula
which states that, under certain regularity conditions, we have
f
oo 1 fC + ooi
q>(s) = XS1tjJ(x) dx if, and only if, tjJ(x) = 2. q>(s)x
S
ds.
o nl c ooi
In this case we take q>(s) to be the gamma function integral,
ns) = LX' x
S
 le
X
dx
and invert this to obtain (20). (Mellin's inversion formula can be deduced
from the Fourier integral theorem, a proof of which is given in [3]. See also
[49], p. 7.) Applying (20) with x = 2nm(r + a)z and c = 3/2 to the last
exponential in (19) and writing f(c) for f ~ = ~ ~ we obtain
00 00 00 e21timP 1 f
L ,1((r + a)z  if3) = L L 2' r(s){2nm(r + a)z} s ds
r=O r=O m=l m nl (3/2)
1 f r(s) 00 1 00 e21timp
= . S L S L r+s ds
2nl (3/2) (2nz) r=O (r + a) m=l m
1 f r(s)
= 2. (2)S '(s, a)F(f3, 1 + s) ds.
nl (3/2) nz
54
3.5: Iseki's transformation formula
Here ((s, a) is the Hurwitz zeta function and F(x, s) is the periodic zeta function
defined, respectively, by the series
oc 1
((s, tX) = L ( + )"
r=O r a
and
00 e21timx
F(x, s) = L S
m=l m
where Re(s) > 1, 0 < a S 1, and x is real. In the same way we find
oc 1 f r(s)
L A((r + 1  a)z + i{3) = 2. (2)S ((s, 1  a)F(1  {3, 1 + s) ds,
r= 0 1tl (3/2) 1rZ
so (1 7) becomes
(21)
where
A(a, {3, z) = 2
1
. f z  s<I>(a, {3, s) ds,
1tl (3/2)
r(s)
(22) <1>(tX, 13, s) = (211:)' {((s, tX)F(f3, 1 + s) + ((s, 1  tX)F(1  13, 1 + s)}.
Now we shift the line of integration from c = ~ to c = 1. Actually, we
apply Cauchy's theorem to the rectangular contour shown in Figure 3.2,
! + iT ! + iT
~
~
0
1
..
t  iT !  iT
Figure 3.2
and then let T . 00. In Exercise 8 we show that the integrals along the
horizontal segments tend to 0 as T . 00, so we get
I3/2)  I3/2) + R
where R is the sum of the residues at the poles of the integrand inside the
rectangle. This gives us the formula
A(a, {3, z) = 2
1
. f zs<I>(a, {3, s) ds + R.
1rl ( 3/2)
55
(23)
3: The Dedekind eta function
In this integral we make the change of variable u =  s to get it back in
the form of an integral along the 1line. This gives us
1\(0:, {3, z) = 2
1
. f zU<1>(o:, {3,  u) du + R.
n1 (3/2)
Now the function <1> satisfies the functional equation
(24) <1>(0:, {3,  s) = <1>( 1  {3, 0:, s).
This is a consequence of Hurwitz's formula for ((s, 0:) and a proof is outlined
in Exercise 7. Using (24) in (23) we find that
(25) 1\(0:, {3, z) = 1\( 1  {3, 0:, Z  1) + R.
To complete the proof of Iseki's formula we need to compute the residue
sumR.
Equation (22) shows that <1>(0:, {3, s) has a first order pole at each of the
points s = 1, °and 1. Denoting the corresponding residues by R(1),
R(O) and R(  1) we find
r( 1) 1 oc (2Trinp  2Trin
p
)
R(1) = 2 {F({3, 2) + F(1  {3,2)} = 2 L ~ + _e2
nz nz n= 1 n n
1 oc e2Trinp 1  (2n0
2
n
= 2 L 2 = 2 2' B
2
({3) =  B
2
({3),
nz n =  oc n nz. z
n*O
where we have used Theorem 12.19 of [4J to express the Fourier series as a
Bernoulli polynomial.
To calculate R(O) we recall that ((0,0:) = 1 0:. Hence ((0, 1  0:) = 0:  1
so
oc e2Trinp _ e  2Trinp
R(O) = ((0, o:)F({3, 1) + ((0, 1  o:)F(l  {3, 1) = (1  0:) L 
n= 1 n
00 e2Trinp oc e2Trinp
= (1  0:) L  = B1(0:) L  = 2niB
1
(0:)B
1
({3),
n=oo n n=oc n
n*O n*O
where again we have used Theorem 12.19 of [4]. To calculate R( 1) we
write
R( 1) = Res zs<1>(o:, {3, s) = lim (s + 1)zs<1>(0:, {3, s)
s=l s ~  l
= lim(  s + 1)zs<1>(0:, {3,  s).
s ~ l
Using the functional equation (24) we find
R(1) = lim(1  s)zs<1>(1  {3, 0:, s) = Res zS<1>(1  {3, 0:, s).
s= 1
56
3.5: Iseki's transformation formula
Note that this is the same as R(I) == Res
s
= 1 zS<l>(o:, {3, s), except that z is
replaced by  z  1, 0: by 1  {3, and {3 by 0:. Hence we have
R( 1) ==  nzB
2
(0:).
Thus
R = R( 1) + R(O) + R(l) = nznto G)(iZ)nB1n(IX)Bi{3).
This proves Iseki's formula under the restriction a < 0: < 1, a < {3 < 1.
Finally, we use a limiting argument to show it is valid if a ~ 0: ~ 1 and
a< {3 < 1, or if a~ {3 ~ 1 and a< 0: < 1. For example, consider the series
00 ex.; oc e2nimf3
L A((r + ex)z  i{3) == L L e
2nm
(r+a)z
r=O r=Om=l m
ex 2nimf3 00
== I _e__ e2nmaz L e2nmrz
m=l m r=O
say, where
1 e 2nmaz
.fa(m) ==  1  2nmz·
m e
As m ~ 00, .fa(m) ~ auniformly in 0: if a s 0: S 1. Therefore the series
00
L e2nimf3fa(m)
m=l
converges uniformly in 0: if a :::; 0: :::; 1, provided a < {3 < 1, so we can pass
to the limit 0: ~ a+ term by term. This gives us
00 ex.;
lim L A((r + o:)z  i{3) == L A(rz  i{3).
aO+ r=O r=O
Therefore, if a < {3 < 1 we can let 0: ~ a+ in the functional equation. The
other limiting cases follow from the invariance of the formula under the
following replacements:
0: ~ 1  0:, { 3 ~ 1  { 3
0: ~ {3,
1
{3 ~ 1  0:, z ~ 
Z
0: ~ 1  {3, {3 ~ 0:,
1
D z ~  .
z
57
(26)
(27)
3: The Dedekind eta function
3.6 Deduction of Dedekind's functional
equation from Iseki's formula
Now we use Iseki's formula to prove Equation (16) of Lemma 2. This, in turn,
will prove Dedekind's functional equation for 11(r).
Equation (16) involves integers hand k with k > O. First we treat the case
k = 1 for which Equation (16) becomes
I A{n(z  ih)} = I  iH)} + log z  (z 
n= 1 n= 1 Z 2 12 z
Since A(X) is periodic with period i this can be written as
00 00 (n) 1 re ( 1)
L A(nz) = L A  +  log z   z   .
n=1 n=1 Z 2 12 z
We can deduce this from Iseki's formula (18) by taking f3 = 0 and letting
a + 0+. Before we let a + 0+ we separate the term r = 0 in the first term
of the series on the left of (18) and in the second term of the series on the right
of (18). The difference of these two terms is A(az)  A(ia). Each of these tends
to 00 as a + 0+ but their difference tends to a finite limit. We compute this
limit as follows:
. 1 
A(CXZ)  A(iCX) = log(l  e
Z
"IlX)  log(l  ez"",Z) = log 1 _ ez"",z'
By L'Hopital's rule,
1  2ni(X 2··
1
·  e 1· rel l
1m _ 2n(Xz = 1m  = 
(X+O + 1  e (X+O 2rez z
so
lim (A(CXZ)  A(icx)) = log = ni  log Z.
(X+O+ Z 2
Now when a + 0+ the remaining terms in each series in (18) double up
and we obtain, in the limit,
(28)
rei 00 00 (r) rez re rei
  log z + 2 L A(rz) = 2 L A    +  + .
2 r = 1 r = 1 Z 6 6z 2
This reduces to (27) and proves (16) in the case k = 1.
Next we treat the case k > 1. We choose rational values for a and f3 in
Iseki's formula (18) as follows. Take
Jl
a = k' where 1 Jl k  1
58
3.6: Deduction of Dedekind's functional equation from Iseki's formula
and write
hJ1 = qk + v, where 1 ~ v ~ k  1.
Now let
Note that v == hJ1 (mod k) so  Hv ==  HhJ1 == J1 (mod k), and therefore
Hv/k == J1/k (mod 1). Hence lJ. = J1/k == Hv/k (mod 1) and f3 = v/k ==
hJ1/k (mod 1). Substituting in Iseki's formula (18) and dividing by 2 we get
Rewrite this as follows:
= ~ 00 { I(rk + V)G  iH)) (rk + k V)G  iH))}
2 Jo A\ k + A k
Now sum both sides on J1 for J1 = 1, 2, ... , k  1 and note that
{rk + J1:r = 0,1,2, ... ; J1 = 1,2, ... ,k  I} = {n:n =1= °(modk)}
and similarly for the set of all numbers rk + k  J1. Also, since v == hJ1 (mod k),
as J1 runs through the numbers 1, 2, ... , k  1 then v runs through the same
59
3: The Dedekind eta function
set of values in some other order. Hence we get
CX) (n .) CX) (n (1 .)) n(1 )k 1 J12
L A  (z  zh) = L A :   zH +    z L 2
n= 1 k n= 1 k z 2 z p.= 1 k
n 0 (mod k) n 'i 0 (mod k)
n (1 )k 1 J1 n (1 )
    z L  +    z (k  1)
2 Z p.=1 k 12 z
k  1J1 (V 1) ni k  1V ni
+ ni L       L  +  (k  1)
p.= 1 k k 2 2 p.= 1 k 4
(
(k  1) (2k  1) ) k  1 J1 (V 1)
X  3(k  1) + (k  1) + ni L    
k p.= 1 k k 2
I  iH)) + (z   + l/f !!. 
n= 1 k z 12 z k p.= 1 k k 2
n$.O (mod k)
But V was defined by the equation hJ1 = qk + V, so we have
Therefore
hJ1 V
k = q + k'
= hJ1 _ [h
J1
]
k k k·
k1 J1 (V 1) h1 J1(h
J1
[h
J1
] 1)
L     = L       = s(h, k).
p.= 1 k k 2 p.= 1 k k k 2
Therefore we have proved that
(29)
A(kn (z' ih)) = f:  iH))
n= 1 k z
n 1 0 (mod k) n 1= 0 (mod k)
Add this to Equation (27) which corresponds to the case k = 1:
CX) CX) (m) n( 1) 1
L A(mz) = L A    z   + log z.
m= 1 m= 1 Z 12 z 2
This accounts for the missing terms in (29) with n == 0 (mod k), if we write
n = mk. When (27) is combined with (29) we get
f (z  ih)) = f  iH))  (z  + log z + nis(h, k).
n =1 k n = 1 k z 12k z 2
60
(30)
3.7: Properties of Dedekind sums
This proves (16) which, in tum, completes the proof of Dedekind's functional
equation for Tl(T). For alternate proofs see p. 190 and [18], [35], and [45]. D
3.7 Properties of Dedekind sums
The Dedekind sums s(h, k) which occur in the functional equation for YJ(r)
have applications to many parts of mathematics. Some of these are described
in an excellent monograph on Dedekind sums by Rademacher and Grosswald
[38]. We conclude this chapter with some arithmetical properties of the sums
s(h, k) which will be needed later in this book. In particular, Theorem 3.11
plays a central role in the study of the invariance of modular functions under
transformations of certain subgroups of r, a topic discussed in the next
chapter.
Note. Throughout this section we assume that k is a positive integer and
that (h, k) = 1.
Dedekind sums are defined by the equation
s(h, k) = kt
1
!. (hr  [hrJ _~ ) .
r= 1 k k k 2
First we express these sums in terms of the function ((x)) defined by
( )
_{x  [x] ! if xis not an integer,
(x)  0 ·f··
I X IS an Integer.
This is a periodic function of x with period 1, and ((  x)) =  ((x)). Actually,
((x)) is the same as the Bernoulli periodic function B1(x) discussed in [4],
Chapter 12. Since ((x)) is periodic and odd we find that
L ( ( ~ ) )  0
r mod k k
and, more generally,
L ((hr))  0 for (h, k) = 1.
r mod k k
Since
the Dedekind sums can now be represented as follows:
(31)
s(h,k) = r m ~ k ((i))((h:)).
This representation is often more convenient than (30) because we can exploit
the periodicity of ((x)).
61
3: The Dedekind eta function
Theorem 3.6
(a) If h' == ±h (mod k), then s(h', k) = ±s(h, k), lvith the same sign as
in the congruence. Similarly, lve have:
(b) If hh == ±1 (mod k) then s(h, k) = ±s(h, k).
(c) ~ f h
2
+ 1 == 0 (mod k), then s(h, k) = O.
PROOF. Parts (a) and (b) follow at once from (31). To prove (c) we note that
h
2
+ 1 == 0 (mod k) implies h == h (mod k), where h is the reciprocal of
h mod k, so from (a) and (b) we get s(h, k) = s(h, k) = O. D
For small values of h the sum s(h, k) can be easily evaluated from its
definition. For example, when h = 1 we find
k  1 r(r 1) 1k  1 2 1k  1
s(l, k) = L     = 2 L r   L r
r = 1 k k 2 k r = 1 2k r = 1
(k  1)(2k  1) k  1 (k  1)(k  2)

6k 4 12k
Similarly, the reader can verify that
(
2 k) = (k  1)(k  5) if k is odd.
s , 24k
In general there is no simple formula for evaluating s(h, k) in closed form.
However, the sums satisfy a remarkable reciprocity law which can be used
as an aid in calculating s(h, k).
3.8 The reciprocity law for Dedekind sums
Theorem 3.7 (Reciprocity law for Dedekind sums). If h > 0, k > 0 and
(h, k) = 1 we have
12hks(h, k) + 12khs(k, h) = h
2
+ k
2
 3hk + 1.
PROOF. Dedekind first deduced the reciprocity law from the functional
equation for log '1(7:). We give an arithmetic proof of Rademacher and
Whiteman [39], in which the sum L ~ = 1 ((hr/k))2 is evaluated in two ways.
First we have
(32)
62
k ((hr))2 ((hr))2 ((r))2 kl (r 1)2
L  = L  = L  = L 
r = 1 k r mod k k r mod k k r = 1 k 2 .
3.8: The reciprocity law for Dedekind sums
We can also write
I ((hr))2 = kil (hr _ [hrJ _ ~ ) 2
r= 1 k r= 1 k k 2
= ki
1
( h 2 ~ 2 + [hrJ2 + ~ _ hr + [hrJ _ 2hr [hrJ)
r= 1 k k 4 k k k k
= 2h
k
f !:. (hr _ [hrJ _ ~ )
r= 1 k k k 2
k1 [hrJ([hrJ ) h
2
k1 2 1 k1
+L   +1 2Lr +2:1.
r= 1 k k k r= 1 4 r= 1
Comparing this with (32) and using (30) we obtain
k  1 [hrJ ([hrJ ) h
2
+ 1 k  11
k
 1
(33) 2hs(h,k) + ~ k k + 1 = k
2
 ~ r
2
 k ~ r.
r1 r1 r1
In the sum on the left we collect those terms for which [hr/k] has a fixed value.
Since 0 < r < k we have 0 < hr/k < h and we can write
(34)
[ ~ J= v  1, where v = 1, 2, ... , h.
For a given v let N(v) denote the number of values of r for which [hr/k] =
v  1. Equation (34) holds if, and only if
hr
v1<<v
k '
or
k(v  1) kv
h < r < h'
equality being excluded since (h, k) = 1 and 0 < r < k. Therefore, if
1 ::; v ::; h  1, Equation (34) holds when r ranges from [k(v  1)/h] + 1
to [kv/h], and hence
N(v) = [k; J[k(V ~ 1)J if 1 ::; v ::; h  1.
But when v = h the quotient kv/h = k and since r = k is excluded we have
N(h) = k _ 1_[k(h; 1)]
63
3: The Dedekind eta function
Hence
(35) :t: [h{] ([h{] + 1) = vt (v  l)vN(v)
vt(v  l)v([k:]  [k(V 1)])  h(h  1)
h 1 [k ]
= : {(v  l)v  v(v + I)}
+ kh(h  1)  h(h  1)
h1 [k ]
= 2 L v + h(h  1)(k  1).
v= 1 h
Now we also have
h 1 (kV [kVJ 1) h 1 [kV] 2k h 1 2 h 1
2hs(k, h) = 2 L v      = 2 L v  +  L v  L V
\'=1 h h 2 v=1 h h v=1 v=1
so (35) becomes
k  1 [h ] ([h ]) 2k h 1 h 1
{ { + 1 = 2hs(k,h)  h + + h(h  l)(k  1).
We use this in (33) and multiply by 6k to obtain the reciprocity law. D
3.9 Congruence properties of Dedekind sums
Theorem 3.8. The number 6ks(h, k) is an integer. Moreover, if0 = (3, k) we have
(a) 12hks(k, h) == 0 (mod Ok)
and
(b) 12hks(h, k) == h
2
+ 1 (mod Ok).
PROOF. From (30) we find
6h k  1 k  1 [hrJ k  1
(36) 6ks(h, k) = k  6 k  3
Since 6 t r
2
= k(k  1) (2k  1) each term on the right of (36) is an
integer. Moreover, (36) shows that
6ks(h, k) == h(k  1)(2k  1) (mod 3)
so we have
(37) 12ks(h, k) == 2h(k  1)(2k  1) == h(k  1)(k + 1) (mod 3).
64
3.9: Congruence properties of Dedekind sums
If 31 k then 3%h and (37) implies
12ks(h, k) ==  h =1= 0 (mod 3).
If 3,rk then 31 (k  1)(k + 1) and (37) implies
(38) 12ks(h, k) == 0 (mod 3).
In other words, 12ks(h, k) == 0 (mod 3) if, and only if, 3%k. Hence, inter
changing hand k, we have
12hs(k, h) == 0 (mod 3) if, and only if, 3%h.
If 0 = 3 this implies (a) since (h, k) = 1. If 0 = 1, (a) holds trivially. Part (a),
together with the reciprocity law, gives (b) since k
2
 3hk == 0 (mod Ok).
o
Note. Theorems 3.8(b) and 3.6(c) show that
s(h, k) = 0 if, and only if, h
2
+ 1 == 0 (mod k).
Theorem 3.9. The Dedekind sums satisfy the congruence
[
2hrJ
(39) 12ks(h, k) == (k  1)(k + 2)  4h(k  1) + 4 I  (mod 8).
r<k/2 k
If k is odd this becomes
(40) 12ks(h, k) == k  1 + 4 I (mod 8).
r<k/2
PROOF. From (36) we obtain
kl [h J
12ks(h, k) = 2h(k  1)(2k  1)  :  3k(k  1)
kl [h J
= 2h(k  1) + 4hk(k  1)  :
+ k(k  1)  4k(k  1).
Now we reduce the right member modulo 8. Since 4k(k  1) == 0 (mod 8)
this gives us
kl [h J
12ks(h, k) == 2h(k  1)  : + k(k  1) (mod 8)
kl [hrJ
== (k  1)(k  2h)  4 I  (mod 8)
r= 1 k
r odd
k  1 [hrJ [2hrJ
== (k  1)(k  2h)  4 L: k + 4 L: k (mod 8).
r= 1 r<k/2
65
3: The Dedekind eta function
The next to last term is equal to
k1 [hrJ k1 ((hr)) k1 hr k1
4 L  = 4 L   4 L  + 2 L 1
r=1 k r=1 k r=1 k r=1
= 0  2h(k  1) + 2(k  1) = (k  1)(2  2h).
Since
(k  1)(k  2h) + (k  1)(2  2h) = (k  1)(k + 2)  4h(k  1)
this proves (39).
When k is odd we have 4h(k  1) == 0 (mod 8) and
(k  1)(k + 2) = k
2
+ k  2 == k'  1 (mod 8)
since k
2
== 1 (mod 8). Hence (39) implies (40). D
Theorem 3.10. If k = 2).k 1 where A > 0 and k 1 is odd, then for odd h z 1
we have
(41) 12hks(h, k) == h
2
+ k
2
+ 1 + 5k  4k L [ 2 ~ V J (mod 2).+ 3).
v<h/2
PROOF. Since h is odd we can apply (40) to obtain, after multiplication by k,
[
2kV]
12hks(k,h) == k(h  1) + 4k L h (mod 2,\+3).
v <h/2
By the reciprocity law we have
12hks(h, k) = h
2
+ k
2
 3hk + 1  12hks(k, h)
== h
2
+ k
2
 3hk + 1  k(h  1)  4k L [ 2 ~ V ] (mod 2).+3)
v<h/2
== h
2
+ k
2
+ 1 + k  4hk  4k L [2kVJ (mod 2).+3).
v <h/2 h
Since h is odd we have 4k(h + 1) == 0 (mod 2'\ + 3) hence k  4hk == 5k
(mod 2)' + 3) and we obtain (41). D
Finally, we obtain a property of Dedekind sums which plays a central
role in the study of the invariance of modular functions under transforma
tions of certain subgroups of the modular group. This will be needed in
Chapter 4.
Theorem 3.11. Let q = 3, 5, 7 or 13 and let r = 24/(q  1). Given integers
a, b, c, d with ad  bc = 1 such that c = c
1
q, where C
1
> 0, let
{
a + d} { a + d}
<5 = s(a, c)  ~  s(a, Cl)  12cl .
Then rb is an even integer.
66
3.9: Congruence properties of Dedekind sums
PROOF. Taking k = c in Theorem 3.8(b) we find
12ac{S(a, c)  al;cd} == a
2
+ 1  a(a + d) == bc (mod ec),
where 0 = (3, c). The same theorem with k = Cl = c/q gives, after multi
plication by q,
12ac{S(a, Ct)  == qa
2
+ q  qa(a + d) == qbc (mod etc),
where 0l = (3, c
l
). Note that OliO so both congruences hold modulo 0lC.
Subtracting the congruences and multiplying by r we find
12acrb == r(q  1)bc (mod OlC).
But r(q  1)bc = 24bc == 0 (mod 0lC) so this gives
12acrb == 0 (mod OlC).
Now (a, c) = 1 since ad  bc = 1. Also, 12cb is an integer so we can cancel
a in the last congruence to get
(42) 12crb == 0 (mod OlC).
Next we show that we also have
(43) 12crb == 0 (mod 3c).
Assume first that q > 3. In this case 0 = (3, qc l) = (3, Cl) = 01 so (42)
becomes
12crb == 0 (mod Oc).
If 0 = 3 this gives (43). But if 0 = 1 then so and (38) implies
12cs(a, c) == 0 (mod 3) and 12cs(a, Cl) == 0 (mod 3). Hence
12crb == r(q  1)(a + d) = 24(a + d) == 0 (mod 3),
which, together with (42), implies (43).
Now assume that q = 3 so r = 12. Then 0 = 3 and 0l is 1 or 3. IfO
l
= 3
we get (43) by the same argument used above, so it remains to treat the case
0
1
= 1. In this case so (38) implies 12c
l
s(a, c
l
) == 0 (mod 3), hence
12cs(a, Cl) == 0 (mod 9).
Also,
12cb = 12cs(a, c)  (a + d)  12cs(a, Cl) + 3(a + d)
== 12cs(a, c) + 2(a + d) (mod 9),
so
(44) 12racb = 12racs(a, c) + 2r(a
2
+ ad) (mod 9).
67
3: The Dedekind eta function
But Theorem 3.8(b) gives us 12aes(a, e) == a
2
+ 1 (mod 9) since 31 e. Hence
(44) becomes
12raeb == r(a
2
+ 1) + 2ra
2
+ 2rad (mod 9)
== 3ra
2
+ r + 2r(1 + be) == 3r + 2rbe == 0 (mod 9)
since r = 12 and 9112e. This shows that
12raeb == 0 (mod 9).
Now 3 ~ a since (a, e) = 1 so we can cancel a to obtain 12reb == 0 (mod 9)
which, with (42), implies (43).
Our next goal is to show that we also have
(45) 12erb == 0 (mod 24e)
since this implies rb is even and proves the theorem. To prove (45) we treat
separately the cases e odd and e even.
Case 1: e odd. Apply (40) with k = e to obtain
{
a + d}
12e s(a, e)  ~ == e  1 + 4T(a, e)  (a + d) (mod 8)
where we have written
T(a, e) = L [2aVJ.
v<cj2 e
We only need the fact that T(a, e) is an integer. Applying (40) again with
k = e
1
= e/q and multiplying by q we have
12C{S(a, c
l
)  a l ; C ~ } == C  q + 4qT(a, cd  q(a + d) (mod 8).
Subtracting the last two congruences and multiplying by r we find
12erb == r(q  1) + r(q  1)(a + d) == 0 (mod 8)
since r(q  1) = 24 and 4r == 0 (mod 8). Combining this with (43) we obtain
(45) and the theorem is proved for odd e.
Case 2: e even. Write e = 2
A
y with y odd. Now a is odd since (a, e) = 1 so
if a ~ 1 we can apply Theorem 3.10 with k = e and h = a to obtain
12aC{S(a, c)  al;Cd} == a
2
+ c
2
+ 1
+ 5e  4eT(e, a)  a(a + d) (mod 2
A
+ 3)
== e
2
+ 5e  be  4eT(e, a) (mod 2
A
+
3
)
since ad  be = 1. Similarly,
12aC{S(a, c
l
)  a l ~ l d } = CC
l
+ 5c  qbc  4cT(c
l
, a) (mod 2.<+3).
68
(47)
(48)
3.LO: The Eisenstein series G
2
(r)
Subtract, multiply by r and use the congruence 4cr == 0 (mod 2). + 3) to obtain
12carc5 == rcc1(q  1) + r(q  l)bc == 0 (mod 2
A
+
3
).
Since a is odd we can cancel a to obtain
(46) 12crc5 == 0 (mod 2
A
+ 3).
Now (43) states that 12crc5 == 0 (mod 3 ·2
A
y) which, together with (46)
implies (45) and proves the theorem for a 1.
To prove it for a < 0, write c5 = c5(a) to indicate the dependence on a.
If a' = a + tc, where t is an integer, an easy calculation shows that
c5(a')  c5(a) = t(q  1)/12 since s(a, c) = s(a', c) and s(a', c
1
) = s(a, c
1
). There
fore rc5(a')  rc5(a) = 2t, an even integer. Choosing t so that a' 1 we know
rc5(a') is even by the above argument, so rc5(a) is also even. This completes
the proof. D
3.10 The Eisenstein series G2( r)
If k is an integer, k 2, and if r E H the Eisenstein series
1
G
2k
(r) = L 2k
(m,n)*(O,O) (m + nr)
converges absolutely and has the Fourier expansion
2(2ni)2k 00 .
G2k(r) = 2((2k) + (2k _ I)!
where, as usual, (Jex(n) = Ldln d
ex
. The cases k = 2 and k = 3 were worked out
in detail in Chapter 1, and the same argument proves (48) for any k 2. If
k = 1 the series in (47) no longer converges absolutely. However, the series
in (48) does converge absolutely and can be used to define the function G
2
(r).
Definition. If r E H we define
(49)
00
G
2
(r) = 2'(2) + 2(2ni)2 L(J(n)e21rinr.
n=l
If x = e
21rir
the series on the right of (49) is an absolutely convergent
power series for Ix I < 1 so G
2
(r) is analytic in H. This definition also shows
that G
2
(r + 1) = G
2
(r).
Exercises 1through 5 describe the behavior of G2 under the other generator
of the modular group. They show that
(50) G
2
(
1
) = r
2
G
2
(r)  2nir,
a relation which leads to another proof of the functional equation YJ(  l/r) =
( ir)1/2YJ(r).
69
3: The Dedekind eta function
Exercises for Chapter 3
1. If r E H prove that
(51)
x
G
2
(r) = 2((2) + L
n=  x
n*O
x 1
L 2'
m=x (m + nr)
Hint: Start with Equation (12) of Chapter 1, replace r by nr, where n > 0, and sum
over all n 2 1.
2. Use the series in (51) to show that
(52)
(
1) oc x 1
r
2
G
2
= = 2((2) + L L 2'
r m=cx n=x (m + nr)
n*O
the iterated series in (52) being the same as that in (51) except with the order of sum
mation reversed. Therefore, proving (50) is equivalent to showing that
(53) L: L: 1 = L: I 1 _ 2ni
m=  x n=  x (m + nr)2 n=  x m=  x (m + nr)2 r
n*O n*O
3. (a) In the gamma function integral r(z) = ett
Z
 1 dt make the change of variable
t = where > O. to obtain the formula
(54) Il(zf'(z) = L"e·uu
z

I
du,
and extend it by analytic continuation to complex with > O.
(b) Take z = 2 and =  2ni(m + nr) in (54) and sum over all 11 2 1 to obtain the
relation
OC 1 IX
L 2 =  8n
2
cos(2nmu)gr(U) du,
n=x (nr + m) 0
n*O
where
x
giu) = LI L e2rrinru if u > 0
n=l
and
4. (a) Use Exercise 3 to deduce that
where
X
f(t) = Lgr(t + k).
k=O
70
Exercises for Chapter 3
(b) The series on the right of (55) is a Fourier series which converges to the value
±{f(O+) + f(1 )}. Show that
1 ex
f(O+) = 2. + Igr(k)
nlr k= 1
and that
ex ex
f(1 ) = I gr(k) = I a(n)e21tinr,
k= 1 n= 1
and then use (55) to obtain (50).
5. (a) Use the product defining 1](r) to show that
d
4ni  log 1](r) = G
2
(r).
dr
(b) Show that (50) implies
d (1) dId
 log 1]  = log 1](r) +  log( ir).
dr r dr 2 dr
Integration of this equation gives 1]( l/r) = C(  ir)1/21](r) for some constant C.
Taking r = i we find C = 1.
6. Derive the reciprocity law for the Dedekind sums s(h, k) from the transformation
formula for log 1](r) as given in Equation (12).
Exercises 7 and 8 describe properties of the function
r(s)
<I>((J(, 13, s) = (2n)' {((s, (J()F(f3, 1 + s) + ((s, 1  (J()F(l  13,1 + s)}
which occurs in the proof of Iseki's formula (Theorem 3.5). The properties
follow from Hurwitz's formula (Theorem 12.6 of [4]) which states that
r(s) . /2 . /2
W  s, a) = (2n)' F(a, s) + e
1tlS
F( a, s)}.
7. (a) If 0 < a < 1 and Re (s) > 1, prove that Hurwitz's formula implies
r(1  s) '( 1 )/2 . 2
F(a s) = {e
1t1
s ((1  s a) + e
1tI
(Sl)/ /(1  s 1  a)}
, (2n)1  S , s,·
(b) Use (a) to show that <J>(a, Ii, s) can be expressed in terms of Hurwitz zeta functions
by the formula
<I>(a, = e
1tis
/
2
{((s, a)(( s, 1  {3) + ((s, 1  a)(( s, fJ))
r(s)r( s)
+ e
1tis
/
2
{(( s, I  f3K(s, 1  a) + (( 5, {3)((s, a)}
and deduce that <I>((X, {3, s) = <1>( 1  {3, (X,  s).
8. This exercise gives an estimate for the modulus of the function z  s<I>((X, {3, s) which
occurs in the integral representation of A((X, fi, s) in the proof of Iseki's formula
(Theorem 3.5).
71
3: The Dedekind eta function
(a) Show that the formula of Exercise 7(b) implies
nz
s
.
zs<I>(a, [3, s) = . {e
1tls
/
2
[((s, a)(( s, [3) + ((s, 1  a)(( S, 1  [3)J
s SIn ns
+ e
1tis
/
2
[((s, a)((  s, 1  [3) + ((s, 1  a)((  s, [3)J}.
(b) For fixed zwith Iarg zI < n12, choose J > 0 so that Iarg zI nl2  J, and show
that if s = (J + it where (J we have
Izsi = O(e
1tl
(1t/2b»),
where the constant implied by the Osymbol depends on z.
(c) If s = (J + it where (J and It I 1, show that
1 (e
1t1tl
)
Is sin nsl = 0 ItI '
and that
Ie
1tiS
/
2
1 = O(e1tltl/2), Ie
1tis
/
2
1 = O(e1tltl/2).
(d) If (J and It I 1 obtain the estimate I((s, a) I = O( ItiC) for some c > 0
(see [4J, Theorem 12.23) and use (b) to deduce that
Izs<I>(a, [3, s)1 = O(/tI2Cle/tlt)).
This shows that the integral of [3, s) along the horizontal segments of the
rectangle in Figure 3.2 tends to 0 as T 00.
PROPERTIES OF DEDEKIND SUMS
9. If k 1 the equation
s(h, k) = k ((i))((h:))
is meaningful even if h is not relatively prime to k and is sometimes taken as the
definition of Dedekind sums. Using this as the definition of s(h, k) prove that
s(qh, qk) = s(h, k) if q > O.
10. If p is prime prove that
p 1
(p + l)s(h, k) = s(ph, k) + L s(h + mk, pk).
m=O
11. For integers r, h, k with k 1 prove that we have the finite Fourier expansion
((
hr)) I k  1 . 2nhrv nv
 =   L SIncot
k 2k \'= 1 k k
and derive the following expression for Dedekind sums:
1 k 1 nhr nr
s(h, k) =  L cot  cot.
4k r= 1 k k
12. This exercise relates Dedekind sums with the sequence {lI(Il)] of Fibonacci numbers
1, 1,2,3,5,8, ... , in which u(l) = u(2) = 1 and u(n + 1) = u(n) + u(n  1).
(a) If h = u(2n) and k = u(2n + 1) prove that s(h, k) = o.
(b) If h = u(2n  1) and k = u(2n) prove that 12hks(h, k) = h
2
+ k
2
 3hk + 1.
72
Exercises for Chapter 3
FORMULAS FOR EVALUATING DEDEKIND SUMS
The following exercises give a number of formulas for evaluating Dedekind
sums in closed form in special cases. Assume throughout that (h, k) = 1,
k ~ 1, h ~ 1.
13. If k == r (mod h) prove that the reciprocity law implies
12hks(h, k) = k
2
 {12s(r, h) + 3}hk + h
2
+ 1.
Use the result of Exercise 13 to deduce the following formulas:
14. If k == 1 (mod h) then 12hks(h, k) = (k  l)(k  h
2
 1).
15. If k == 2 (mod h) then 12hks(h, k) = (k  2)(k  !(h
2
+ 1)).
16. If k == 1 (mod h) then 12hks(h, k) = k
2
+ (h
2
 6h + 2)k + h
2
+ 1.
17. If k == r (mod h) and if h == t (mod r) where r ~ 1 and t = ± 1, then
h
2
 t(r  l)(r  2)h + r
2
+ 1
12hks(h,k) = k
2
 k + h
2
+ 1.
r
This formula includes those of Exercises 14 and 15 as special cases.
18. Show that the formula of Exercise 17 determines s(h, k) completely when r = 3
and when r = 4.
19. If k == 5 (mod h) and if h == t (mod 5), where t = ±1 or ±2, then
h k) k
2
h
2
+ 4t(t  2)(t + 2)h + 26 2
12hk s(, =  5 k + h + 1.
20. Assume 0 < h < k and let ro, r
1
, ... , r
n
+ 1 denote the sequence of remainders in the
Euclidean algorithm for calculating the gcd (h, k), so that
Prove that
1 n ~ 1 { "+ 1 r/ + rj_1
2
+ I} (1)" + 1
s(h,k)= ~ (I)J .
12 j =1 rjrj1 8
This also expresses s(h, k) as a finite sum, but with fewer terms than the sum in the
original definition.
73
Congruences for the coefficients
of the modular function j
4.1 Introduction
The function j(r) = 12
3
J(r) has a Fourier expansion of the form
1 00 •
j(r) =  + L c(n)x
n
, (x = e
2m
!)
x n=O
where the coefficients c(n) are integers. At the end of Chapter 1 we mentioned
a number of congruences involving these integers. This chapter shows how
some of these congruences are obtained. Specifically we will prove that
c(2n) == 0 (mod 2
11
),
c(3n) == 0 (mod 3
5
),
c(5n) == 0 (mod 52),
c(7n) == 0 (mod 7).
The method used to obtain these congruences can be illustrated for the
modulus 52. We consider the function
00
f5(r) = L c(5n)x
n
n=1
obtained by extracting every fifth coefficient in the Fourier expansion of j.
Then we show that there is an identity of the form
(1)
where the ai are integers and <I>(r) has a power series expansion in x = e
21ti
!
with integer coefficients. By equating coefficients in (1) we see that each
coefficient of fs(r) is divisible by 25.
Success in this method depends on showing that such identities exist.
How are they obtained?
74
4.2: The subgroup r o(q)
Theorem 2.8 tells us that every modular function f is a rational function
of j. Sometimes this rational function is a polynomial in j with integer co
efficients, giving us an identity of the form
f(r) = a1j(r) + a2j2(r) + ... + akjk(r).
However, the function fs(r) is not invariant under all transformations of
the modular group r and cannot be so expressed in terms of j(r). But we
shall find that fs(r) is invariant under the transformations of a certain
subgroup of r, and the general theory enables us to express fs(r) as a poly
nomial in another basic function <l>(r) which plays the same role as j(r)
relative to this subgroup. This representation leads to an identity such as
(1) and hence to the desired congruence property.
The subgroup in question is the set of all unimodular matrices (: : )
with c == 0 (mod 5). More generally we shall consider those matrices in r
with c == 0 (mod q), where q is a prime or a power of a prime.
4.2 The subgroup r o(q)
Definition. If q is any positive integer we define r o(q) to be the set of all
matrices (: ~ ) in r with c == 0 (mod q).
It is easy to verify that r o(q) is a subgroup of r. The next theorem gives a
way of representing each element of r in terms of elements of r o(p) when
p is prime. In the language of group theory it shows that r o(p) is of finite
index in r.
Theorem 4.1. Let Sr = l/r and Tr = r + 1 be the generators of the full
modular group r, and let p be any prime. Then for every V in r, V ~ r o(P),
there exists an element P in r o(P) and an integer k, 0 ~ k < p, such that
V = PST
k
•
PROOF. Given V = ( ~ ~ ) where C =1= 0 (mod p). We wish to find
p = G:). with c == 0 (mod p),
and an integer k, 0 ~ k < p, such that
75
4: Congruences for the coefficients of the modular function j
All matrices here are nonsingular so we can solve for ( : ) to get
(:
k to be that solution of the congruence
kC == D (mod p) with 0 k < p.
This is possible since C =1= 0 (mod p). Now take
c = kC  D, a = kA  B, b = A, d = C.
Then c == 0 (mod p) so PEro(p). This completes the proof. o
4.3 Fundamental region of r o(p)
As usual we write Sr = l/r and Tr = r + 1, and let R
r
denote the funda
mental region of r.
Theorem 4.2. For any prime p the set
pl
R
r
u USTk(R
r
)
k=O
is a fundamental region of the subgroup r o(p).
This theorem is illustrated for p = 3 in Figure 4.1.
PROOF. Let R denote the set
pl
R = R
r
u USTk(R
r
).
k=O
We will prove
(i) if r E H, there is a V in r o(p) such that Vr belongs to the closure of R, and
(ii) no two distinct points of R are equivalent under r o(p).
To prove (i), choose r in H, choose r 1 in the closure of R
r
and choose
A in r such that Ar = r
1
. Then by Theorem 4.1 we can write
AI = PW
where PEro(p) and W = I or W = ST
k
for some k, 0 k p  1. Then
P = A
I
W
1
and p
I
= WA. Let V = P
1
• Then VEro(p) and
Vr = WAr = Wr
i
.
Since W = I or W = STk'l this proves (i).
76
4.3: Fundamental region of r o(p)
1 o
1
2"
T
Figure 4.1 Fundamental region for r 0(3)
Next we prove (ii). Suppose!1E R, ! 2E R and v!1 = ! 2 for some V in
r o(P). We will prove that!1 = ! 2' There are three cases to consider:
(a) !1 ER
r
'!2 ERr· In this case!1 =!2 since VEr.
(b) ! 1 ERr, ! 2 ESTk(R
r
).
(c) ! 1 E STkl(R
r
), ! 2 E ST
k
2(R
r
).
In case (b), !2 = ST
k
!3 where !3 ERr. The equation
implies
k (0
V = ST = 1
This contradicts the fact that V E r o(p).
Finally, consider case (c). In this case
1)
k .
and
where !1' and !2' are in R
r
. Since V!1 = !2 we have VST
k
l!I' = ST
k
2!2' so
VSTkl = ST
k
2,
77
4: Congruences for the coefficients of the modular function i
Since VE ro(p) this requires k
2
== k
i
(mod pl. But both k
l
, k
2
are in the
interval [0, p  IJ, so k
2
= k
i
. Therefore
V = SToS = S2 = I
and r I = r 2. This completes the proof.
D
We mention (without proof) the following theorem of Rademacher [34]
concerning the generators of r o(p). (This theorem is not needed in the
later work.)
Theorem 4.3. For any prime p> 3 the subgroup ro(p) has 2[p112J + 3
generators and they may be selected from the following elements:
where Tr = r + 1, Sr = l/r, and
k k' ( k'
v" = ST ST S = (kk' + 1)
where kk' == 1 (mod pl. The subgroup r
o
(2) has generators T and V I ~
the subgroup r 0(3) has generators T and V
2
.
Here is a short table of generators:
p 2 3 5 7 11 13 17 19
Generators: T T T T T T T T
VI
V
2
V
2
V
3
V
4
V
4
V
4
V
s
V
3
V
s
V
6
V
s
V
7
V
8
V
8
V
g
V
I2
V
IO
V
I3
V
I3
4.4 Functions automorphic urlder the
subgroup r o(p)
We recall that a modular function f is one which has the following three
properties:
(a) f is meromorphic in the upper halfplane H.
(b) f(Ar) = f(r) for every transformation A in the modular group r.
(c) The Fourier expansion of f has the form
00
f(r) = L ane21tint.
n= m
78
4.4: Functions automorphic under the subgroup r o(p)
If property (b) is replaced by
(b') f(Vr) = f(r) for every transformation V in r o(P),
thenfis said to be automorphic under the subgroup ro(p). We also say that
f belongs to r o(P)·
The next theorem shows that the only bounded functions belonging to
r o(p) are constants.
Theorem 4.4. If f is automorphic under r o(p) and bounded in H, then f is
constant.
PROOF. According to Theorem 4.1, for every V in r there exists an element
P in r o(P) and an integer k, °~ k ~ p, such that
V = PA
k
,
where A
k
= ST
k
if k < p, and A
p
= I. For each k = 0, 1, ... , p, let
r
k
= {PA
k
: PEro(p)}.
Each set r k is called a right coset of r o(P). Choose an element ~ from the
coset r
k
and define a function fk on H by the equation
fk(r) = f ( ~ r).
Note that fp(r) = f(Pr) = f(r) since P E ro(P) and f is automorphic under
ro(P). The function value A(r) does not depend on which element ~ was
chosen from the coset r k because
A(r) = f ( ~ r ) = f(PAkr) = f(Akr)
and the element A
k
is the same for all members of the coset r
k
•
How does fk behave under the transformations of the full modular group?
If V E r then
fk(Vr) = f ( ~ Vr).
Now ~ V E r so there is an element Qin r o(P) and an integer m, °~ m ~ p,
such that
Therefore we have
Moreover, as k runs through the integers 0, 1, 2, ... , p so does m. In other
words, there is a permutation (J of {O, 1,2, ... , p} such that
fk(Vr) = fa(k)(r) for each k = 0,1, ... ,p.
Now choose a fixed w in H and let
p
cp(r) = n{fk(r)  f(w)}.
k=O
79
4: Congruences for the coefficients of the modular function j
Then if V E r we have
p p
cp(Vr) = n {fk(Vr)  f(w)} = n {f(j(k)(r)  f(w)} = cp(r),
k=O k=O
so cp is automorphic under the full group r. Now cp is bounded in H (since
each fk is). T h e r e f o r e ~ cp omits some value hence, by Theorem 2.5, cp is
constant, so cp(r) = cp(w) for all r. But cp(w) = 0 because
p
cp(w) = n {fk(W)  f(w)}
k=O
and the factor with k = p vanishes since fp = f Therefore cp(r) = 0 for all r.
Now take r = i. Then
p
o= n {A(i)  f(w)}
k=O
hence some factor is o. In other words, f(w) = A(i) for some k. But w was
arbitrary so f can take only the values fo(i), ... ,fp(i). This implies that f is
constant. 0
4.5 Construction of functions belonging
to ro(p)
This section shows how to construct functions automorphic under the
subgroup r o(p) from given functions automorphic under r.
Theorem 4.5. If f is automorphic under r and if p is prime, let
fir) =! Pff(r + A).
p ),=0 P
Thenfp is automorphic under r o(p). Moreover, iff has the Fourier expansion
00
f(r) = L a(n)e21tinr
n= m
then fp has the Fourier expansion
00
fp(r) = L a(np)e21tinr.
n=  [m/p]
PROOF. First we prove the statement concerning Fourier expansions. We have
1 p 1 00
fp(r) =  L L a(n)e
21tin
(r+).)/P
p ),=0 n=m
1 00 p 1
=  L a(n)e21tinr/p Le
21tin
)./p.
p n= m ),=0
80
4.5: Construction of functions belonging to r o(p)
But
so
pI {O
L e
21tin
;"/p =
;"=0 P
if p,(n
if pin
00 00
fp('r) = L a(n)e21tinr/p = L a(np)e21tinr.
n= m n= [m/p]
pin
This shows that fp has the proper behavior at the point r = ioo. Also,
fp is clearly meromorphic in H because it is a linear combination of functions
meromorphic in H.
Next we must show that
fp(Vr) = fp(r) whenever V E r o(p).
For this we use a lemma.
Lemma 1. If V E ro(p) and if °~ A ~ p  1, let T;.,r = (r + A)jp. Then
there exists an integer j1, 0 ~ j1 ~ p  1 and a transformation ~ in
r O(p2) such that
Moreover, as Aruns through a complete residue system modulo p, so does j1.
First we use the lemma to complete the proof of Theorem 4.5, then we
return to the proof of the lemma.
If V E r o(p) we have
1
P

1
(Vr+A) 1
P

1
fp(Vr) =  L f =  L f(T;., Vr).
P ;"=0 P, P ;"=0
Now we use the lemma to write the last sum as
1 p I 1 p I
 L f ( ~ ~ r) =  L f ( ~ r) = fp( r).
PJL=o PJL=o
This proves that fp is invariant under all transformations in r o(p), so fp is
automorphic under r o(p). D
PROOF OF LEMMA 1. Let V = (: :). where c == 0 (mod p), and let A. be
given, 0 ~ A ~ p  1. We are to find an integer j1, 0 ~ j1 ~ p  1 and a
transformation ~ = ( ~ ~ ) such that ~ E r O(P2) and
T;.,V = WJLTJL.
81
4: Congruences for the coefficients of the modular function j
Since TA. = G~ ) we must satisfy the matrix equation
or
(
a + AC b + Ad) = (A AJl + BP)
pc pd C CJl + Dp
with C == 0 (mod p2). Equating entries we must satisfy the relations
{
A = a + AC
(2)
C = pc
(3) {AJl + Bp = b + Ad
CJl + Dp = pd
with
C == 0 (mod p2) and AD  BC = 1.
Now (2) determines A and C. Since pic, we have C == 0 (mod p2). Substi
tuting these values in (3) we must satisfy
(4) {(a + AC)Jl + Bp = b + Ad
CpJl + Dp = pd.
Choose Jl to be that solution of the congruence
Jla == b + Ad (mod p)
which lies in the interval 0 ~ Jl ~ P  1. This is possible because ad  bc = 1
and pic imply p,ta. Note that distinct values of Amod p give rise to distinct
values of Jl mod p. Then, since pic we have
Jla +JlAc == b + Ad (mod p)
or
(a + AC)Jl == b + Ad (mod p).
Therefore there is an integer B such that
(a + AC)Jl + Bp = b + Ad.
Therefore the first relation in (4) is satisfied. The second relation requires
D = d  CJl. Thus, we have found integers Jl, A, B, C, D such that
Clearly AD  BC = 1 since all matrices in this equation have determinant
1 or p. This completes the proof of the lemma. D
82
or
4.6: The behavior under the generators of r
4.6 The behavior off
p
under the generators
ofr
Let Tr = r + 1 and Sr = l/r be the generators of r. Since T E r o(p) we
have fp(Tr) = fp(r). The next theorem gives a companion result for fp(Sr).
Theorem 4.6. If f is automorphic under r and if p is prime, then
( = f, (r) + f(pr) 
p r p p p P
To prove this we need another lemma.
Lemma 2. Let T). r = (r + A)/p. Thenfor each Ain the interval 1 S ASP  1
there exists an integer J.1 in the same interval and a transformation V in
r o(p) such that
T).S = VT
l1
•
Moreover, as Aruns through the numbers 1,2, ... ,p  1, so does J.1.
PROOF OF LEMMA 2. We wish to find (: in ro(p) such that
G = G
G = (: )
Take a = A, c = p and let J.1 be that solution of the congruence
,1,J.1 == 1 (mod p)
in the interval 1 S J.1 S P  1. This solution is unique and J.1 runs through a
reduced residue system mod p with ,1,. Choose b to be that integer such that
aJ.1 + bp =  1, and take d =  J.1. Then CJ.1 + dp = 0 and the proof is
complete. D
PROOF OF THEOREM 4.6. We have
Pfp(  = Pi! f(sr + A) = f(sr) + Pf f(T;.Sr)
r ),=0 p P).=1
= f( + Pi! f(VT
I1
r) = f(rp) + Pi! f(T
IL
r) 
rp 11=1 11=0 P
= f(rp) + pfp(r)  fG) 0
83
4: Congruences for the coefficients of the modular function.i
4.7 The function <p(r) =
The number of poles of an automorphic function in the closure of its fun
damental region is called its valence. A function is called univalent on a
subgroup G if it is automorphic under G and has valence 1. Such a function
plays the same role in G that J plays in the full group r.
It can be shown (using Riemann surfaces) that univalent functions exist
on G if and only if the genus of the fundamental region R
G
is zero. [This is
the topological genus of the surface obtained by identifying congruent edges
of R
G
• For example, the genus of R
r
is zero because R
r
is topologically
equivalent to a sphere when its congruent edges are identified.]
Our next goal is to construct a univalent function on the subgroup
r o(p) whenever the genus of r o(p) is zero. This will be done with the aid of
the discriminant = g2
3
 27g3
2
•
We recall that is periodic with period 1and has the Fourier expansion
(Theorem 1.19)
00
= (2n)12 L r(n)e21tint
n=1
where the r(n) are integers with r(1) = 1 and r(2) = 24. However, is
not invariant under all transformations of r. In fact we have
: = (cr + if (: E r.
In particular,
+ 1) = and 1) =
Even though is not invariant under r it can be used to construct functions
automorphic under the subgroup r o(q) for each integer q.
Theorem 4.7. For afixed integer q, let
q>(r) = ifr E H.
Then qJ is automorphic under ro(q). Moreover, the Fourier expansion ofqJ
has theform
where the b
n
are integers and x = e
21tit
•
84
4.7: The function <p(r) =
PROOF. First we obtain the Fourier expansion. We have
= (2n)12 r(n)x
n
= (2n)12
x
{ 1 + r(n + 1)x
n
}.
where x = e
21tit
• Hence
so
q>(r) = = x
q

1
1+ 1r(n + 1)x
nq
= x
q
 1(1 + f b x
n
)
r) 1 + 1 r(n + 1)x
n
n = 1 n
where the b
n
are integers.
Now qJ is clearly meromorphic in H, and we will prove next that qJ is
invariant under r o(q).
If V = (: E ro(q) then c = c
1
q for some integer c
1
• Hence
Vr) = (cr + d) 12 r) = (c 1qr + d) 12 r).
On the other hand,
aT + b a(qr) + bq
qVr = qd = () d = W(qr),
cr + C
1
qr +
where
w = (a b
q
).
C
1
d
But WE r because det W = ad  bc
1
q = ad  bc = 1. Hence
= = (c
1
(qr) +
so
(Vr) = = (c1qr + = (r).
qJ Vr) (c1qr + d) 12 r ) qJ
This completes the proof. D
Now qJ has a zero of order q  1 at 00 and no further zeros in H. Next we
show that qJ does not vanish at the vertex r = 0 of the fundamental region
of r o(q). In fact, we show that qJ(r) + 00 as r + O.
Theorem 4.8. Ifr E H we have
(
1) 1
qJ = q12
qJ
(r)·
Hence qJ(r) + 00 as r + O.
85
(5)
4: Congruences for the coefficients of the modular function j
PROOF. Since = we have
:r) =
so
4.8 The univalent function <1>(r)
The function cp has a zero of order q  1 at 00 and no further zeros so its
valence is q  1. We seek a univalent function automorphic under r o(q)
and this suggests that we consider cpa, where r:J. = 1/(q  1). The Fourier
expansion of cpa need not have integer coefficients, since
= x(1 + bnxnr
On the other hand we have the product representation
00
= (2n)12
x
n(1  X
n
)24
n=1
so
where the coefficients dq(n) are integers. Therefore if a = 1/(q  1) we have
(
00 )24a
= x 1 + din)x
n
and the Fourier series for cpa(r) will certainly have integer coefficients if
24a is an integer, that is, if q  1 divides 24. This occurs when q = 2, 3, 4, 5,
7, 9, 13, and 25.
Definition. If q  1 divides 24 let a = 1/(q  1) and r = 24a. We define the
function <D by the relations
a (l1(qr))r
<I>(r) = q> (r) = = 1](r) ·
86
(6)
4.9: Invariance of <1>( r) under transformations of r o(q)
The function <I> so defined is analytic and nonzero in H. The Fourier series
for <I> in (5) shows that <I> has a first order zero at 00 and that
1 1
<I>(r) = + I(x),
where I(x) is a power series in x with integer coefficients.
Since qJ is automorphic under r o(q) we have qJ(Vr) = qJ(r) for every
element V of r o(q). Hence, extracting roots of order q  1, we have
<1>( Vr) = E<I>( r )
where E
q
 1 = 1. The next theorem shows that, in fact, E = 1 whenever
24/(q  1) is an even integer and q is prime. This occurs when q = 2, 3, 5, 7,
and 13. For these values of q the function <I> is automorphic under r o(q).
4.9 Invariance of <l>(r) under transformations
of r o(q)
The properties of Dedekind sums proved in the foregoing chapter lead to a
simple proof of the invariance of the univalent function <I>(r).
Theorem 4.9. Let q = 2,3,5,7, or 13, and let r = 24/(q  1). Then thefunction
<1>( r) = (11(qr))r
11(r)
is automorphic under the subgroup r o(q).
PROOF.Ifq = 2wehaver = 24 and <I>(r) =
was already proved in Theorem 4.7. Therefore we shall assume that q 3.
Let V = (: be any element of r o(q). Then ad  be = 1 and
c == 0 (mod q). We can suppose that c O. If c = 0 then V is a power of
the translation Tr = r + 1, and since 11(r + 1) = e
1ti
/
12
11(r) we find
<I>(r + 1) = (11(qr + q))r = e"ir(Q
1
l/12<1>(r) = <l>(r).
11(r + 1)
Therefore we can assume that c > 0 and that c = c1q, where C
1
> O.
Dedekind's functional equation for 11(r) gives us
(7)
where
(8)
87
4: Congruences for the coefficients of the modular function.i
We also have
(
a(qr) + b
q
)
I](qVr) = I] c1(qr) + d = I](V1qr)
where
= (a b
q
)
I CI d·
Since VI E r we have
which, together with (7), gives us
<D(Vr) =
But (8) shows that (r.(V
1
)/r.(V)t = e where
,{a+d } {a+d }
(j = + s(d,c)  12c
1
+ s(d,c
1
) .
Since ad  bc = 1 we have ad == 1 (mod c) and ad == 1 (mod c
l
) so
s( d, c) = s(a, c) and s( d, c
l
) = s(a, CI), and Theorem 3.11 shows that
rb is an even integer. Therefore = 1 and <I>(Vr) = <I>(r). D
4.10 The function j p expressed as a
polynomial in <l>
If p is prime and if f is automorphic under r, we have shown that the
function
fp(r) = Pf f(r + A)
p ),=0 P
is automorphic under r o(p), and its Fourier coefficients consist of every pth
coefficient of f To obtain divisibility properties of the coefficients of jp(r)
we shall express j p as a polynomial in the function <1>.
In deriving the differential equation for the Weierstrass g;J function we
formed a linear combination of g;J, g;J2 and g;J3 which gave a principal part
near z = 0 equal to that of [g;J'(z)] 2. The procedure here is analogous.
Both functions jp and <I> have a pole at the vertex r = 0 of the fundamental
region of ro(p). We form a linear combination of powers of <I> to obtain a
principal part equal to that of jp.
88
4.10: The function j p expressed as a polynomial in <1>
To obtain the order of the pole of jp(r) at r = 0 we use Theorem 4.6
which gives us the relation
j ( ~ ) = j (r) + ~ j(pr)  ~ j ( ~ )
p r p p p p
valid for prime p. Replacing r by pr in this formula we obtain
Theorem 4.10. If p is prime and r E H then
j ( ~ ) = j (pr) + ~ j(p2r)  ~ j(r).
p pr p p p
Hence if x = e
21tit
we have the Fourier expansion
pj ( ~ ) = x
p2
 XI + I(x),
p pr
where I(x) is a power series in x with integer coefficients.
PROOF. We have
j(r) = XI + c(O) + c(1)x + c(2)x
2
+ ... ,
jp(r) = c(O) + c(p)x + c(2p)x
2
+ ... ,
pjp(pr) = pc(O) + pc(p)x
P
+ pc(2p)x
2p
+ ... ,
and
so
pjp(  ;r) = pjp(pr) + j(p2
r
)  j(r)
= x
p2
 XI + I(x).
Now we can express j p as a polynomial in <1>.
Theorem 4.11. Assume p = 2,3,5,7 or 13, and let
(
l1(Pr))r 24
<I>(r) = '1(r) , where r = p _ 1.
Then there exist integers aI' ... , a
p
2 such that
PROOF. By Theorem 4.10 we have
pjp(  ;r) = x
p2
 x
1
+ I(x),
o
89
4: Congruences for the coefficients of the modular function j
and, since 12a = r/2, Theorem 4.8 gives us
pr/2cI>(_ ~ ) = _1_ = XI + I(x).
pr cI>(r)
Let ljJ(r) = pr/2<D( l/(pr)). Then the difference
pj (_ ~ )  {1jJ(r)}p
2
p pr
has a pole of order :s p2  1 at x = 0, and the Laurent expansion near x = 0
has integer coefficients. Hence there is an integer b1 such that
has a pole of order :s p2  2 at x = 0, and the Laurent expansion near x = 0
has integer coefficients. In p2 steps we arrive at a function
(
1) . ( 1) 2 21
f   = Pl    {1jJ(r)}P b
1
{IjJ(r)}P  ...  b 2_ ljJ(r)
pr P pr P 1
which is analytic at x = 0 and has a power series expansion with integer
coefficients. Moreover, all the numbers b
1
, ... , b
p
2_1 are integers. Replacing
r by 1/(pr) we obtain
f(r) = pjp(r)  {pr/2cI>(r)}P
2
 b
1
{pr/2cI>(r)}p2
1
 ...  b
p
2_ 1{pr/2cI>(r)}.
Now f(r) is automorphic under r o(p) and analytic at each point r in H.
The function f is also analytic at the vertex r = 0 (by construction). Therefore
f is bounded in H so f is constant. But this constant is pc(O) since cI>(r)
vanishes at 00. Thus we find
pjp(r) = {pr/2cI>(r)}p2 + b
1
{pr/2cI>(r)}p2
1
+ ... + b
p
2_1 {pr/2cI>(r)} + pc(O)
so j p(r) is expressible as indicated in (9). D
Theorem 4.12. The coefficients in the Fourier expansion of j(r) satisfy the
following congruences:
c(2n) == 0 (mod 2
11
)
c(3n) == 0 (mod 3
5
)
c(5n) == 0 (mod 52)
c(7n) == 0 (mod 7).
PROOF. The previous theorem shows that for p = 2, 3, 5, 7 and 13 we have
c(pn) == 0 (mod p(r/2)I),
where r = 24/(p  1). Therefore we simply compute (r/2)  1 to obtain the
stated congruences. Note that (r/2)  1 = 0 when p = 13 so we get a trivial
congruence in this case. D
90
Exercises for Chapter 4
Note. By repeated application of the foregoing ideas Lehner [24] derived
the following more general congruences, valid for lJ. ~ 1:
c(2
Cl
n) == 0 (mod 2
3Cl
+8)
c(3
Cl
n) == 0 (mod 3
2Cl
+
3
)
C(5Cl
n
) == 0 (mod 5Cl+I)
c(7
Cl
n) == 0 (mod 7
Cl
).
Since it is known that c(13) is not divisible by 13, congruences of the above
type cannot exist for 13. In 1958 Morris Newman [30] found congruences
of a different kind for 13. He showed that
(
13n)
c(13np) + c(13n)c(13p) + p1c p == 0 (mod 13),
where pIp == 1 (mod 13) and c(x) = 0 if x is not an integer. The congruences
of Lehner and Newman were generalized by Atkin and O'Brien [5J in 1967.
Exercises for Chapter 4
1. This exercise relates the Dedekind function 1J(r) to the Jacobi theta function 9(r)
defined on H by the equation
ex:: ex'
t9(r) = 1 + 2 L e1tin2t = L e1tin2t.
n=l
The definition shows that 9 is analytic in H and periodic with period 2.
Jacobi's triple product identity (Theorem 14.6 in [4J) states that
ex:: ex::
n(1  x
2n
)(1 + X
2n

1
z2)(1 + x
2n
 1z 2) = L X
rn2
Z
2rn
n=l
if z =I 0 and Ix I < 1.
(a) Show that x and z can be chosen to give the product representation
00
:+(r) = n(1  e21tint)(1 + e(2n1)1tit)2.
n=l
This implies that t9( r) is never zero in H.
(b) If r E H prove that
Sr + 1 = B(r + 1). and
( q2(' ; 1)
ti(r) = .
l1(r + 1)
(c) Prove that :)( l/r) = ( ir)1/2[)(r).
Hint: If Sr = l/r, find elements A and B of r such that
Sr ; 1 = A(r ; 1)
91
4: Congruences for the coefficients of the modular function j
Vr + 1 = B(r + 1). and
2. Let G denote the subgroup of r generated by the transformations Sand T
2
, where
Sr = l/r and Tr = r + 1.
(a) If (; E G prove that a == d (mod 2) and b == c (mod 2).
(b) If V E G prove that there exist elements A and B of r such that
Vr 2+ 1= A(T ; 1)
(c) If (; E G and c > 0 prove that
(
ar + b)
9 cr + d = e(a, b, c, d){ i(cr + d)}1/29(r),
where Ie(a, b, c, d) I = 1. Express e(a, b, c, d) in terms of Dedekind sums.
Exercises 3 through 8 outline a proof (due to Mordell [28]) of the multiplica
tivity of Ramanujan's function r(n). We recall that
00 00
L r(n)e21tint = = e
21tit
n(1  e21timt)24.
n = 1 m= 1
3. Let p be a prime and let k be an integer, 1 k p  1. Show that there exists an
integer h such that
and that h runs through a reduced residue system mod p with k.
4. If p is a prime, define
1 p1 (r + k)
Fir) = +  L  .
P k=O P
Prove that:
(
1) 12
(b) F
p
r = r Fp(r).
Note: Exercise 3 will be helpful for part (b).
5. Prove that Fir) = where r(p) is Ramanujan's function.
6. Use Exercises 4 and 5 to deduce the formulas
(a) r(pn+1) = r(p)r(pn)  p11
r
(pn1) for n 2 1.
(b) r(pa
n
) = r(p)r(pa1
n
)  p11
r
(pa2
n
) for rt. 2 2 and (n, p) = 1.
7. If rt. is an integer, rt. 2 0, and if (n, p) = 1, let
g(rt.) = r(pa
n
)  r(pa)r(n).
Show that g(rt. + 1) is a linear combination of g(rt.) and g(rt.  1) for rt. 2 2 and deduce
that g(rt.) = °for all rt..
92
Exercises for Chapter 4
8. Prove that
r(m)r(n) = L
dl(rn,n)
In particular, when (m, n) = 1 this implies r(m)r(n) = r(mn).
9. If r E H and x = e
21tit
prove that
{S04.tCTs(n)x'r= {j(r)  12
3
} r(n)x',
where (J 5(0) =  1/504. Equate coefficients of x
n
to obtain the identity
n nl
(504)2 L (Js(k)(Js(n  k) = r(n + 1)  984r(n) + L c(k)r(n  k).
k=O k=1
10. Use Exercise 9 together with Exercise 10 of Chapter 6 to prove that
65520 nl
{(Jll(n) r(n)} = r(n+ 1)+ 24r(l1)+ L c(k)r(n k).
691 k=1
This formula, due to Lehmer [20], can be used to determine the coefficients c(n)
recursively in terms of r(n). Since the right member is an integer, the formula also
implies Ramanujan's remarkable congruence
r(n) == (J 11 (n) (mod 691).
93
Rademacher's series for
the partition function
5.1 Introduction
The unrestricted partition function p(n) counts the number of ways a positive
integer n can be expressed as a sum of positive integers ~ n. The number of
summands is unrestricted, repetition is allowed, and the order of the sum
mands is not taken into account.
The partition function is generated by Euler's infinite product
(1)
00 1 00
F(x) = nm = L p(n)x
n
,
m= 1 1  x n=O
where p(O) = 1. Both the product and series converge absolutely and repre
sent the analytic function F in the unit disk Ix I < 1. A proof of (1) and other
elementary properties of p(n) can be found in Chapter 14 of [4J. This chapter
is concerned with the behavior of p(n) for large n.
The partition function p(n) satisfies the asymptotic relation
e K ~
p(n) '" r; as n + 00,
4n
y
3
where K = n(2/3)1/2. This was first discovered by Hardy and Ramanujan [13J
in 1918 and, independently, by J. V. Uspensky [52] in 1920. Hardy and
Ramanujan proved more. They obtained a remarkable asymptotic formula
of the form
(2)
94
~
p(n) = L Pk(n) + O(n
1
/
4
),
k<a...jn
5.2: The plan of the proof
where LI. is a constant and P 1(n) is the dominant term, asymptotic to
e
K
y'rJI(4nJ"3). The terms P2(n), P3(n), ... are of similar type but with smaller
constants in place of K in the exponential. Since p(n) is an integer the finite
sum on the right of (2) gives p(n) exactly when n is large enough to insure
that the error term is less than 1/2. This is a rare example of a formula which
is both asymptotic and exact. As is often the case with asymptotic formulas
of this type, the infinite sum
(3)
diverges for each n. The divergence of (3) was shown by D. H. Lehmer [21]
in 1937.
Hans Rademacher, while preparing lecture notes in 1937 on the work of
Hardy and Ramanujan, made a small change in the analysis which resulted
in slightly different terms Rk(n) in place of the Pk(n) in (2). This had a profound
effect on the final result since, instead of (2), Rademacher obtained a con
vergent series,
(4)
00
p(n) = L Rk(n).
k= 1
The exact form of the Rademacher terms Rk(n) is described below in Theorem
5.10. Rademacher [35] also showed that the remainder after N terms is
O(n  1/4) when N is of order .)ii, in agreement with (2).
This chapter is devoted to a proof of Rademacher's exact formula for
p(n). The proof is of special interest because it represents one of the crowning
achievements of the socalled "circle method" of Hardy, Ramanujan and
Littlewood which has been highly successful in many asymptotic problems
of additive number theory. The proof also displays a marvelous application
of Dedekind's modular function 1J(r).
5.2 The plan of the proof
This section gives a rough sketch of the proof. The starting point is Euler's
formula (1) which implies
F(x) = ~ p(k)x
k
·fO I I 1
n+ 1 ~ n+ 1 1 < X < ,
X k=O X
for each n ~ O. The last series is the Laurent expansion of F(x)/x
n
+ 1 in the
punctured disk 0 < Ix I < 1. This function has a pole at x = 0 with residue
p(n) so by Cauchy's residue theorem we have
1 JF(x)
p(n) = 2. n+i dx,
1rl eX
95
5: Rademacher's series for the partition function
where C is any positively oriented simple closed contour which lies inside
the unit circle and encloses the origin. The basic idea of the circle method is
to choose a contour C which lies near the singularities of the function F(x).
The factors in the product defining F(x) vanish whenever x = 1, x
2
= 1,
x
3
= 1, etc., so each root of unity is a singularity of F(x). The circle method
chooses a circular contour C of radius nearly 1 and divides C into arcs
C
htk
lying near the roots of unity e21rih/k, where °::; h < k, (h, k) = 1, and
k = 1,2, ... , N. The integral along C can be written as a finite sum of integrals
along these arcs,
J
N kl J
C k ~ l h ~ O Ch.k·
(h,k)= 1
On each arc Ch,k the function F(x) in the integrand is replaced by an elemen
tary function t/Jh,k(X) which has essentially the same behavior as F near the
singularity e21rih/k. This elementary function t/Jh,k arises naturally from the
functional equation satisfied by the Dedekind eta function l1(r). The functions
F and 11 are related by the equation
F(e
21rit
) = e1rit/12/11(r),
and the functional equation for 11 gives a formula which describes the behavior
of F near each singularity e21rih/k. The replacement of F by t/Jh,k introduces
an error which needs to be estimated. The integrals of the t/Jh,k along Ch,k are
then evaluated, and their sum over hproduces the term Rk(n) in Rademacher's
series.
In 1943 Rademacher [38] modified the circle method by replacing the
circular contour C by another contour in the rplane, where x = e
21rit
. This
new path of integration simplifies the estimates that need to be made and
clarifies the manner in which the singularities contribute to the final formula.
The next section expresses Dedekind's functional equation in terms of F.
Sections 5.5 and 5.6 describe the path of integration used by Rademacher,
and Section 5.7 carries out the plan outlined above.
5.3 Dedekind's functional equation expressed
in terms of F
Theorem 5.1. Let F(t) = l / n ~ = 1 (1  t
rn
) and let
(5)
(
2nih 2nz)
x = exp k  J:2 '
, (2niH 2n)
x = exp k  ~ ,
where Re(z) > 0, k > 0, (h, k) = 1, and hH ==  1 (mod k). Then
96
(6)
(
Z)1/2 ( n nz )
F(x) = e
1ris
(h
t
k)  exp    F(x')
k 12z 12k
2
.
/
5.4: Farey fractions
Note. If Iz I is small, the point x in (5) lies near the root of unity e21rih/k,
whereas x' lies near the origin. Hence F(x') is nearly F(O) = 1, and Equation
(6) gives the behavior of F near the singularity e21rih/k. Aside from a constant
factor, for small Iz I, F behaves like
Zl/2 expC;z)
PROOF. If (: E r with c > 0, the functional equation for 1Jet) implies
(7) 1J:
T
) = {i(CT + d)p/2 exp{n{at;c
d
+ S(d,C»)}.
where T' = (aT + b)/(CT + d). Since F(e
21rit
) = e1rit/12/1J(T), (7) implies
2
. 2 ., (ni(T  T')) 1/2
(8) F(e 1rlt) = F(e 1rlt) exp 12 { i(CT + d)}
xexp{n{a + S(d,C»)}
Now choose
a = H, C = k, d = h, b =
Then
hH + 1
k
and
iz + h
T=k·
, iz
1
+ H
T =
k
and (8) becomes
( (
2nih 2nz)) ( (2niH 2n)) 1/2
F exp    = F exp    z
k k k kz
{
n nz }
x exp 12kz  12k + nis(h, k) .
When z is replaced by z/k this gives (6).
5.4 Farey fractions
o
Our next task is to describe the path of integration used by Rademacher.
The path is related to a set of reduced fractions in the unit interval called'
Farey fractions. This section describes these fractions and some of their
properties.
97
5: Rademacher's series for the partition function
Definition. The set of Farey fractions of order n, denoted by F
n
, is the set of
reduced fractions in the closed interval [0, IJ with denominators ~ n ,
listed in increasing order of magnitude.
EXAMPLES
F
1
: ¥, t
F
2
: ¥, t, t
F3: ¥, t, t, i, t
F4: ¥, t, t, t, i, i, t
F5: ¥,!, t, t, ~ , t, ~ , i, i, ~ , t
F6: ¥, i, !, !, t, ~ , t, ~ , i, i, ~ , i, t
F7: ¥ , ~ , i, !, t, ~ , t, i, ~ , !, 4, ~ , i, ~ , i, ~ , i, ~ , t
These examples illustrate some general properties of Farey fractions.
For example, Fn c Fn+ l' so we get Fn+ 1 by inserting new fractions in Fn.
If (alb) < (cld) are consecutive in F
n
and separated in F
n
+ b then the fraction
(a + c)/(b + d) does the separating, and no new ones are inserted between
alb and cld. This new fraction is called the mediant of alb and cld.
Theorem 5.2. If(alb) < (cld), their mediant (a + c)/(b + d) lies between them.
PROOF
a + c a bc  ad
b + d  b = b(b + d) > 0
and
c a + c bc  ad
d  b + d = d(b + d) > O. 0
The above examples show that t and ~ are consecutive fractions in F
n
for n = 5, 6, and 7. This illustrates the following general property.
Theorem 5.3. Given 0 ~ alb < cld ~ 1. If bc  ad = 1 then alb and cld are
consecutive terms in Fnfor the following values of n:
max(b, d) ~ n ~ b + d  1.
PROOF. The condition be  ad = 1 implies that alb and eld are in lowest
terms. If max(b, d) s n then b ~ nand d ~ n so alb and cld are certainly in
F
n
. Now we prove they are consecutive if n ~ b + d  1. If they are not
consecutive there is another fraction hlk between them, alb < hlk < cld.
But now we can show that k ~ b + d because we have the identity
(9) k = (bc  ad)k = b(ck  dh) + d(bh  ak).
But the inequalities alb < hlk < cld show that ck  dh ~ 1and bh  ak ~ 1
so k ~ b + d. Thus, any fraction hlk that lies between alb and cld has
denominator k ~ b + d. Therefore, if n ~ b + d  1, then alb and cld must
be consecutive in F
n
• This completes the proof. D
98
5.5 : Ford circles
Equation (9) also yields the following theorem.
Theorem 5.4. Given 0 ~ alb < cld ~ 1 with be  ad = 1, let hlk be the
mediant of alb and cld. Then alb < hlk < cld, and these fractions satisfy
the unimodular relations
bh  ak = 1, ck  dh = 1.
PROOF. Since hlk lies between alb and cld we have bh  ak ~ 1 and
ck  dh ~ 1. Equation (9) shows that k = b + d if, and only if, bh  ak =
ck  dh = 1. D
The foregoing theorems tell us how to construct F
n
+ 1 from F
n
•
Theorem 5.5. The set F
n
+
1
includes F
n
• Eachfraction in F
n
+
1
which is not in
F
n
is the mediant of a pair of consecutive fractions in F
n
• Moreover, if
alb < cld are consecutive in any Fn' then they satisfy the unimodular
relation bc  ad = 1.
PROOF. We use induction on n. When n = 1 the fractions Oil and III are
consecutive and satisfy the unimodular relation. We pass from F 1 to F2 by
inserting the mediant 1/2. Now suppose alb and cld are consecutive in F
n
and satisfy the unimodular relation bc  ad = 1. By Theorem 5.3, they will
be consecutive in Fm for all m satisfying
max(b, d) ~ m ~ b + d  1.
Form their mediant hlk, where h = a + c, k = b + d. By Theorem 5.4
we have bh  ak = 1 and ck  dh = 1 so hand k are relatively prime.
The fractions alb and cld are consecutive in Fm for all m satisfying
max(b, d) ~ m ~ b + d  1, but are not consecutive in F
k
since k = b + d
and hlk lies in F
k
between alb and cld. But the two new pairs alb < hlk and
hlk < cld are now consecutive in Fk because k = max(b, k) and k = max(d, k).
The new consecutive pairs still satisfy the unimodular relations bh  ak = 1
and ck  dh = 1. This shows that in passing from F
n
to F
n
+ 1 every new
fraction inserted must be the mediant of a consecutive pair in Fn' and the new
consecutive pairs satisfy the unimodular relations. Therefore Fn+ 1 has these
properties if Fn does. 0
5.5 Ford circles
Definition. Given a rational number hlk with (h, k) = 1. The Ford circle
belonging to this fraction is denoted by C(h,' k) and is that circle in the
complex plane with radius 1/(2k
2
) and center at the point (hlk) + il(2k
2
)
(see Figure 5.1).
Ford circles are named after L. R. Ford [9J who first studied their
properties in 1938.
99
5: Rademacher's series for the partition function
. 1
radIus = 2k
2
h
k
Figure 5.1 The Ford circle C(h, k)
Theorem 5.6. Two Ford circles C(a, b) and C(c, d) are either tangent to each
other or thev do not intersect. They are tangent if, and only if, bc  ad =
± 1. In particular, Ford circles of consecutive Farey fractions are tangent
to each other.
PROOF. The square of the distance D between centers is (see Figure 5.2)
a
b
Figure 5.2
whereas the square of the sum of their radii is
c
d
(
1 1 )2
(r + R)2 = 2b
2
+ 2d2 .
The difference D
2
 (r + R)2 is equal to
(
ad  bc)2 (1 1)2 (1 1)2
D
2
 (r + R)2 = bd + 2b2  2d2  2b2 + 2d2
(ad  bC)2  1
 >0
 b
2
d
2
'
Moreover, equality holds if, and only if (ad  bC)2 = 1.
100
D
5.5: Ford circles
Theorem 5.7. Let h
1
/k
1
< h/k < h
2
/k
2
be three consecutive Farey fractions.
The points of tangency of C(h, k) with C(h
1
, k
1
) and C(h
2
, k
2
) are the points
h k
1
i
(Xl(h, k) = k  k(k
2
+ k
1
2
) + k
2
+ k12
and
h k
2
i
(X2(h,k) = k + k(k
2
+ k
2
2
) + k
2
+ k/"
Moreover, the point of contact a
1
(h, k) lies on the semicircle whose diameter
is the interval [h1/k
b
h/k].
PROOF. We refer to Figure 5.3. Write a
1
for a
1
(h, k). The figure shows that
1
2k
2
h
k
b
a
,
,
,
,
"'::.    
"
"
,
,
,
,
,
,
,
,
"
,
"
1
2k
1
2
Figure 5.3
To determine a and b we refer to the similar right triangles and we get
1 k
1
2
 k
2
b = 2k2 k2 + k1 2 "
so
so
1
a 2ki k
1
2
~  1 1  k
2
+ k1
2
'
k k
1
2k
2
+ 2k
1
2
Similarly, we find
1 1
b 2k2  ~ _ k
1
2
 k
2
1 1 1  k
1
2
+ k
2
'
2k
2
2k
2
+ 2k
1
2
These give the required formula for a
b
and by analogy we get the correspond
ing formula for a
2
•
101
5: Rademacher's series for the partition function
To obtain the last statement, it suffices to show that the angle (J in Figure
5.3 is n12. For this it suffices to show that the imaginary part of rJ.
I
(h, k) is
the geometric mean of a and a', where
o
, h hi 1
a =     a =   a.
k k
l
kk 1
and
a = k(k
2
+ kI 2)
(See Figure 5.4.) Now
, k
l
( 1 k
l
)
aa = k(k2 + k12) kk
l
 k(k
2
+ k
1
2
)
k
l
( k
2
) 1
= k
2
(k
2
+ kI 2) kI (k
2
+ kI 2) = (k
2
+ kI 2)2 '
and this completes the proof.
a' a
Figure 5.4
5.6 Rademacher's path of integration
For each integer N we construct a path P(N) joining the points i and i + 1
as follows. Consider the Ford circles for the Farey series Fv. If h1/k
1
< h/k
< h
2
1k
2
are consecutive in F
N
, the points of tangency of C(h
t
, k
1
), C(h, k),
and C(h
2
, k
2
) divide C(h, k) into two arcs, an upper arc and a lower arc.
P(N) is the union of the upper arcs so obtained. For the fractions 0/1 and 1/1
we use only the part of the upper arcs lying above the unit interval [0, 1].
EXAMPLE. Figure 5.5 shows the path P(3).
Because of Theorem 5.7, the path P(N) always lies above the row of semi
circles connecting adjacent Farey fractions in FN.
The path P(N) is the contour used by Rademacher as a path of integration.
It is convenient at this point to discuss the effect of a certain change of variable
on each circle C(h, k).
Theorem 5.8. The transformation
102
5.6: Rademacher's path of integration
i + 1
o
1
"3
1
"2
2
"3
Figure 5.5 The Rademacher path P(3)
maps the Ford circle C(h, k) in the tplane onto a circle K in the zplane of
radius t about the point z = ! as center (see Figure 5.6). The points of
contact ltl(h, k) and lt
2
(h, k) of Theorem 5.7 are mapped onto the points
k
2
kk
zl(h,k) = k2 + k/ + i k2 + lk/
and
The upper arc joining lt
1
(h, k) with lt
2
(h, k) maps onto that arc of K which
does not touch the imaginary zaxis.
o
1
!
zplane
Figure 5.6
\
\
103
(10)
(11)
5: Rademacher's series for the partition function
PROOF. The translation T  (h/k) moves C(h, k) to the left a distance h/k,
and thereby places its center at i/(2k
2
). Multiplication by  ik
2
expands the
radius to 1/2 and rotates the circle through n/2 radians in the negative
direction. The expressions for z1(h, k) and z2(h, k) follow at once. 0
Now we obtain estimates for the moduli of z1 and z2.
Theorem 5.9. For the points z1 and z2 of Theorem 5.8 we have
k k
I
Z1(h, k)1 = , I
Z
2(h, k)1 = .
~ k 2 + k
1
2
~ k 2 + k
2
2
Moreover, ifz is on the chord joining ZI and Z2 we have
fik
I z l < ~ ,
if h
1
/k 1 < h/k < h
2
/k
2
are consecutive in FN. The length of this chord does
not exceed 2fik/N.
PROOF. For IZ112 we have
k
4
+ k
2
k 2 k2
Iztl
2
= (k
2
+ kJ)2 k
2
+ k
1
2
"
There is a similar formula for Iz21
2
. This proves (10). To prove (11) we note
that if z is on the chord, then IzI ::::; max( Iz1I, Iz21 ), so it suffices to prove that
fik fik
(12) I
Z
ll < "N and I
Z
21 < "N.
For this purpose we use the inequality relating the arithmetic mean and the
root mean square:
k + k (k
2
+ k 2)1/2
__1< 1
2  2 .
This gives us
(
k
2
+ k 2)1/2 > k + k
1
> N + 1 > !i
1  fi  fi fi'
so (10) and (12) imply (11). The length of the chord is ::::; Iz11 + Iz21. D
5.7 Rademacher's convergent series for pen)
Theorem 5.10. If n ~ 1 the partition function p(n) is represented by the
convergent series
1 (X) d
p(n) = M LAk(n)jk d
n
y
2 k= 1 n
104
5.7: Rademacher's convergent series for p(n)
where
A (n)
~ eftis(h,k)2ftinh/k.
k = LJ
O ~ h < k
(h,k)= 1
PROOF. We have
(13)
1 I F(x)
p(n) = 2. n+f dx
'Ttl eX
00 00
where F(x) = n(1  X
m
)l = L p(n)x
n
;
m=l n=O
C is any positively oriented closed curve surrounding x = 0 and lying inside
the unit circle. The change of variable
maps the unit disk Ix I ~ 1 onto an infinite vertical strip of width 1 in the
rplane, as shown in Figure 5.7. As x traverses counterclockwise a circle of
xplane
o
tplane
Figure 5.7
radius e 2ft with center at x = 0, the point r varies from i to i + 1 along a
horizontal segment. We replace this segment by the Rademacher path P(N)
composed of the upper arcs of the Ford circles formed for the Farey series
FN. Then (13) becomes
i+ 1
p(n) = r F(e
2nh
)e 2ninr dr = r F(e
2nit
)e 2nint dr.
J
i
Jp(N)
In this discussion the integer n is kept fixed and the integer N will later be
allowed to approach infinity. We can also write
fP(N) ktl O " ~ < k L(h'k) = t,.;, l(h'kl
(h,k)= 1
where y(h, k) denotes the upper arc of the circle C(h, k), and Lh,k is an
abbreviation for the double sum over hand k.
105
5: Rademacher's series for the partition function
Now we make the change of variable
so that
h iz
t = k + k2 •
Theorem 5.8 shows that this maps C(h, k) onto a circle K of radius! about
z = ! as center. The arc y(h, k) maps onto an arc joining the points zl(h, k)
and z2(h, k) in Figure 5.6. We now have
p(n) = LJ
Z2
(h. k) F(ex
p
(21rih _ 2 ~ Z ) ) ~ e 2ninh/k e2nnz/k2 dz
h,k zt{h,k) k k k
= L ik2e2ninh/k J
Z
2(h.k) e2nnZ/k2F(exp(21rih _ 2 ~ Z ) ) dz.
h,k zt{h,k) k k
Now we use the transformation formula for F (Theorem 5.1) which states
that
(
Z)1/2 ( 1r 1rZ )
F(x) = w(h, k) k exp 12z  12k
2
F(x'),
where
and
(
21rih 21rZ)
x = exp k  k2 '
, (21riH 21r)
x = exp k  ; ,
w(h, k) = enis(h,k), hH == 1 (mod k), (h, k) = 1.
Denote the elementary factor ZI/2 exp[,,/(12z) 1rz/{12k
2
)J by \fJk{Z) and
split the integral into two parts by writing
F(x') = 1 + {F(x')  I}.
We then obtain
p(n) = L ik
5
/
2
W(h, k)e2ninh/k(11(h, k) + 1
2
(h,k))
h, k
where
J
Z 2(h'k)
1
1
(h, k) = qJk(z)e2nnz/k2 dz
zdh,k)
and
J
Z
2(h, k) {( (21riH 21r)) }
I 2(h, k) = qJk(Z) F exp k    1 e2nnz/k2 dz.
zt{h, k) Z
We show next that 1
2
is small for large N. The path of integration in the
zplane can be moved so that we integrate along the chord joining z1(h, k)
and z2(h, k). (See Figure 5.8.) We have already estimated the length of this
106
5.7: Rademacher's convergent series for p(n)
o
K
Figure 5.8
chord; it does not exceed 2J2k/N. On the chord itself we have Izi
max{lzll, IZ21} < J2k/N. Note also that the mapping w = l/z maps
the disk bounded by K onto the halfplane Re(w) 1. Inside and on the
circle K we have 0 < Re(z) 1 and Re(l/z) 1, while on K itself we have
Re(l/z) = 1.
Now we estimate the integrand on the chord. We have
= Iz1
1
/
2
ex
p
{ ReG)  Re(Z)}
x e2nltRe<zl/k21 p(m)e2ltiHm/ke 2ltm/z I
Iz1
1
/
2
I: p(m)e2ltmRe(1/z)
12 z m=l
00
< Izll/2e2n1t L p(m)e
21t
(m
O
/24»Re
O
/z)
m=l
00
Izll/2e2n1t L p(m)e
21t
(m
O
/24»
m=l
00
= Izll/2e2n1t L p(m)e
21t
,(24m1)/24
m=l
oc
< Izll/2e2n1t L p(24m  1)e
21t
(24m1)/24
m=l
oc
= Izll/2e2n1t L p(24m  1)y24ml (where y = e
21t
/
24
)
m=l
107
5: Rademacher's series for the partition function
where
00
c = e
2nn
I p(24m  1)y24m 1.
m=l
The number c does not depend on z or on N. (It depends on n, but n is fixed
in this discussion.) Since z is on the chord we have Iz I < k/N so the
integrand is bounded by c2
1
/
4
(k/N)1/2. The length of the path is less than
so altogether we find
11
2
(h,k)1 < Ck
3
/
2
N
3
/
2
for some constant C, and therefore
I
I ikS/2W(h,k)e21tinh/kI2(h,k)1 < f I Ck
1
N
3
/
2
h, k k= 1 0 h< k
(h,k)= 1
N
CN
3
/
2
I 1 = CN
1
/
2
.
k=l
This means we can write
N
(14) p(n) = I I ikS/2w(h,k)e2ninh/kl1(h,k) + O(N
1
/
2
).
k=l
(h,k)= 1
Next we deal with I
1
(h, k). This is an integral joining zl(h, k) and z2(h, k)
along an arc of the circle K in Figure 5.8. We introduce the entire circle K
as path of integration and show that the error made is also O(N
1
/
2
). We have
1 f
Zdh' k) JO 1
1
1
(h,k) =   = J
1
J
2
,
K() 0 z2(h,k) K()
where K(  ) denotes that the integration is in the negative direction along K.
To estimate IJ 1 I we note that the length of the arc joining 0 to z1(h, k) is less
than
Since Re(1/z) = 1 and 0 < Re(z) 1 on K the integrand has absolute value
I\P (z)e2nnz/k21 = e2nnRe(z)/k2IzI1/2 _ _ 1r_ Re(Z)}
k 12 z 12k
2
e2nn21/4k1/2en/12
<
 N
1
/
2
so that
108
5.7: Rademacher's convergent series for p(n)
where C1 is a constant. A similar estimate holds for IJ21 and, as before,
this leads to an error term O(N
1
/
2
) in the formula for p(n). Hence (14)
becomes
p(n) = f L ik 5/2
W
(h, k)e  2ninh/k r '1\(z)e2nnz/k2 dz + O(N
1
/
2
).
k =1 0 h < k JK(  )
(h,k)= 1
Now we let N + 00 to obtain
where
Ak(n) = L e
1tis
(h, k)  21tinhlk.
(h,k)= 1
The integral can be evaluated in terms of Bessel functions. The change of
variable
gives us
1
W= ,
z
1
dz =  2dw,
W
1
00 51 + ooi {2 ( 1) I}
_  512  512 nw n
p(n)  i 1 ooi W exp 12 + k2 n  24 dw.
Now put t = nw/12 and the formula becomes
(
n)3
/
2 00 1fC+ooi {n
2
( 1 ) I}
p(n) = 2n 12 2ni cooi C
5
/
2
exp t + 6k2 n  24 t dt
where c = n/12. Now on page 181 of Watson's treatise on Bessel functions
[53] we find the formula
(
1 )V fC+ ooi
Iv(z) = 2:
z
. tVlet+(Z2/4t) dt (if c > 0, Re(v) > 0),
2nl c ooi
where I v(z) = i vJv(iz). Taking
and v = 3/2 we get
(
1 )3
/
4
(
n)3
1
2 00 n
3/2
n  24 (n 1 ))
p(n) = (2n) 12 63/4k3/2 13/ 2 kV"3\.n
24
.
(
1 )3
/
4
(2n) n  24 00 1 (n 1 ))
(24)3/4 Ak(n)k 1
3
/
2
k V"3 \.n  24 ·
109
5: Rademacher's series for the partition function
But Bessel functions of half odd order can be reduced to elementary functions.
In this case we have
( )
_ ftz ~ (sinh z)
13/ 2 Z  d .
n z z
Introducing this in the previous formula we finally get Rademacher's
formula,
1 00 d
p(n) =  L A
k
(n)k
1
/
2

nJ2k=l dn
Exercises for Chapter 5
D
1. Two reducedfractionsa/b andc/daresaid to besimilarlyorderedif(c  a)(d  b) ~ O.
Let al/b
t
< a2/b2 < ... denote the Farey fractions in F
n
•
(a) Prove that any two neighbors ai/bi and ai+ l/bi+ 1 are similarly ordered.
(b) Prove also that any two second neighbors ai/bi and ai+ 2/bi+ 2 are similarly ordered.
Note: Erdos [8J has shown that there is an absolute constant c > 0 such that the
kth neighbors a/bi and ai+k/bi+k in F
n
are similarly ordered if 11 > ck.
2. If a, b, c, d are positive integers such that alb < c/d and if Aand J.1 are positive integers,
prove that the fraction
(}=Aa+J.1c
Ab + J.1d
lies between alb and c/d, and that (c  d(})/((}b  a) = )../J.1. When A= J.1, () is the
mediant of alb and c/d.
3. If bc  ad = 1 and n > max(b, d), prove that the terms of the Farey sequence F
n
between alb and c/d are the fractions of the form (Aa + J.1c)/(Ab + J.1d) for which )..
and J.1 are positive relatively prime integers with Ab + J.1d ~ n. Geometrically, each
pair (A, J.1) is a lattice point (with coprime coordinates) in the triangle determined by
the coordinate axes and the line bx + dy = n. Neville [29] has shown that the
number of such lattice points is
3 n
2
2 b + O(n log n).
n d
This shows that for a given n, the number of Farey fractions between alb and c/d is
asymptotically proportional to l/(bd), the length of the interval [alb, c/d].
Exercises 4 through 8 relate Farey fractions to lattice points in the plane.
In these exercises, n ~ 1 and T" denotes the set of lattice points (x, y) in the
triangular region defined by the inequalities
110
1 ::::; x ::::; n, 1 ::::; y ::::; n, n + 1 ::::; x + y ::::; 2n.
Exercises for Chapter 5
Also, 1;,' denotes the set of lattice points (x, y) in 1;, with relatively prime
coordinates.
4. Prove that alb and c/d are consecutive fractions in the Farey sequence Fn if, and only
if, the lattice point (b, d) E
5. Prove that L(b.d)E r;, 1/(bd) = 1. Hint: Theorem 5.5.
6. Assign a weight f(x, y) to each lattice point (x, y) and let Sn be the sum of all the
weights in Tn,
Sn = L f(x, y).
(x, y) E Tn
(a) By comparing the regions T,. and T,. _1 for r 2 show that
r1 rl
Sr  Sr1 = f(r, r) + L {f(k, r) + f(r, k)}  L f(k, r  k),
k=1 k=l
and deduce that
n n r 1 n rl
Sn = L f(r, r) + L L{f(k, r) + f(r, k)}  L L f(k, r  k).
r=1 r=2 k=1 r=2k=1
Note: If f(x, y) = 0 whenever (x, y) > 1 this reduces to a formula of J. Lehner
and M. Newman [25],
(15)
r1
L f(x, y) = f(l, 1) + L L {f(k, r) + f(r, k)  f(k, r  k)}.
r=2 k=1
(k,r)= 1
This relates a sum involving Farey fractions to one which does not.
7. Let
1
S = L
n (b, d) E bd(b + d)
(a) Use Exercise 5 to show that 1/(2n  1) S Sn S 1/(n + 1).
(b) Choose f(x,y) = 1/(xy(x + y)) in (15) and show that
3 n r 1
Sn =   2 L L 2 .
2 r =1 k =1 r (r + k)
(k,r)= 1
When n CJJ this gives a formula of Gupta [12],
00 r 1 3
L L 2 =.
r= 1 k= 1 r (r + k) 4
(k,r)= 1
8. Exercise 7(a) shows that Sn 0 as n Cf:.;. This exercise outlines a proof of the
asymptotic formula
(16)
_ 12 log 2 (log n)
Sn  2 + 0 2
n n n
obtained by Lehner and Newman in [25].
111
5: Rademacher's series for the partition function
Let
so that
A =
r
r 1
L
k=l r
2
(r + k)
(k,r)= 1
±I /(d) ,
k= 1 dl(r,k) r (r + k)
(a) Show that
and deduce that
r>n
=" ~ _d_J.l(r_/d_)
Ar L. L. 3
dl r h = 1 r (h + d)
qJ(r) (1 )
A
r
= log 2 3 + 0 3 L IJ.l(d) I .
r r dlr
(b) Show that L ~ = l Ldlr IJ.l(d)/ = O(n log n) and deduce that
1 (lOg n)
L 3LIJ.l(d)/ = 0 2 .
r> n r dlr n
(c) Use the formula Lr ~ n qJ(r) = 3n
2
/n
2
+ O(n log n) (proved in [4J, Theorem 3.7) to
deduce that
(d) Use (a), (b), and (c) to deduce (16).
112
Modular forms with
multiplicative coefficients
6.1 Introduction
The material in this chapter is motivated by properties shared by the discrimi
nant L\(r) and the Eisenstein series
1
G2k(r) = L 2k'
(m,n):¢:(O, 0) (m + nr)
where k is an integer, k 2. All these functions satisfy the relation
(1) f(:: : = + d)'
where r is an integer and (: is any element of the modular group r.
The function satisfies (1) with r = 12, and G
2k
satisfies (1) with r = 2k.
Functions satisfying (1) together with some extra conditions concerning
analyticity are called modular forms. (A precise definition is given in the
next section.)
Modular forms are periodic with period 1 and have Fourier expansions.
For example, we have the Fourier expansion,
00
= (2n)12 L r(n)e21tint,
n=l
where r(n) is Ramanujan's function, and
2(2ni)2k 00 •
= 2(2k) + (2k _ 1)! (n)e
2
,,,nt,
where (Ja(n) is the sum of the cxth powers of the divisors of n.
113
6: Modular forms with multiplicative coefficients
Both r(n) and O"(X(n) are multiplicative arithmetical functions; that is, we
have
(2) r(m)r(n) = r(mn) and
They also satisfy the more general multiplicative relations
(3) r(m)r(n) = L
dl(m,n)
and
(4) O"(X(m)O"(X(n) = L
dl(m,n)
for all positive integers m and n. These reduce to (2) when (m, n) = 1.
The striking resemblance between (3) and (4) suggests the problem of
determining all modular forms whose Fourier coefficients satisfy a multi
plicative property encompassing (3) and (4). The problem was solved by
Hecke [16J in 1937 and his solution is discussed in this chapter.
6.2 Modular forms of weight k
In this discussion k denotes an integer (positive, negative, or zero), H denotes
the upper halfplane, H = {r: Im(r) > O}, and r denotes the modular group.
Definition. A function f is said to be an entire modular form of weight k if it
satisfies the following conditions:
(a) f is analytic in the upper halfplane H.
(b) f(:: : :) = (cr + d)kf(r) whenever (: :) E r.
(c) The Fourier expansion of f has the form
00
f(r) = L c(n)e21tint.
n=O
Note. The Fourier expansion of a function of period 1 is its Laurent
expansion near the origin x = 0, where x = e
21tit
. Condition (c) states that
the Laurent expansion of an entire modular form contains no negative
powers of x. In other words, an entire modular form is analytic everywhere
in H and at ioo.
The constant term c(O) is called the value of f at ioo, denoted by f(ioo).
If c(O) = 0 the function f is called a cusp form (" Spitzenform" in German),
and the smallest r such that c(r) i= 0 is called the order of the zero of f at i 00.
It should be noted that the discriminant is a cusp form of weight 12 with
a first order zero at i 00. Also, no Eisenstein series G2k vanishes at i 00.
114
6.3: The weight formula for zeros of an entire modular form
Warning. Some authors refer to the weight k as the "dimension  k"
or the "degree k." Others write 2k where we have written k.
In more general treatments a modular form is allowed to have poles in
H or at i00. This is why forms satisfying our conditions are called entire
forms. The modular function J is an example of a nonentire modular form of
weight 0 since it has a pole at ioo. Also, to encompass the Dedekind eta func
tion there are extensions of the theory in which k is not restricted to integer
values but may be any real number, and a factor e(a, b, c, d) of absolute
value 1 is allowed in the functional equation (b). This chapter treats only
entire forms of integer weight with multiplier e = 1.
The zero function is a modular form of weight k for every k. A nonzero
constant function is a modular form of weight k only if k = O. An entire
modular form of weight 0 is a modular function (as defined in Chapter 2) and
since it is analytic everywhere in H, including the point ioo, it must be constant.
Our first goal is to prove that nonconstant entire modular forms exist
only if k is even and ~ 4. Moreover, they can all be expressed in terms of the
Eisenstein series G4 and G6. The proof is based on a formula relating the
weight k with the number of zeros of f in the closure of the fundamental
region of the modular group.
6.3 The weight formula for zeros of an entire
modular form
We recall that the fundamental region R
r
has vertices at the points p, i,
P + 1 and ioo. If f has a zero of order r at a point p we write r = N(p).
Theorem6.1. Let f be an entire modularform ofweight k which is not identically
zero, and assume f has N zeros in the closure of the fundamental region
R
r
, omitting the vertices. Then we have the formula
(5) k = 12N + 6N(i) + 4N(p) + 12N(ioo).
PROOF. The method of proof is similar to that of Theorem 2.4 where we
proved that a modular function has the same number of zeros as poles in
the closure of R
r
. Since f has no poles we can write
N = _1 r f'(7:) d7:.
2ni JaR f(7:)
The integral is taken along the boundary of a region R formed by truncating
the fundamental region by a horizontal line y = M with sufficiently large M.
The path 8R is along the edges of R with circular detours made around the
vertices i, p and p + 1 and other zeros which might occur on the edges. By
115
(6)
6: Modular forms with multiplicative coefficients
calculating the limiting value of the integral as M + 00 and the circular
detours shrink to their centers we find, as in the proof of Theorem 2.4,
k 1 1
N = 12 "2 N(i)  3N(p)  N(ioo).
The only essential difference between this result and the corresponding
formula obtained in the proof of Theorem 2.4 is the appearance of the term
k112. This comes from the weight factor (cr + d)k in the functional equation
f(A(r)) = (cr + d)kf(r),
where A(r) = (ar + b)/(cr + d). Differentiation of this equation gives us
f'(A(r))A'(r) = (cr + d)kf'(r) ,+ kc(cr + d)klfer)
from which we find
f'(A(r))A'(r) = f'(r) + ~ .
f(A(r)) fer) cr + d
Consequently, for any path y not passing through a zero we have
_1 f f'(u) du = _1 f f'(r) dr + _1 f ~ dr.
2ni A(y) feu) 2ni y f(r) 2ni y cr + d
Therefore the integrals along the arcs (2) and (3) in Figure 2.5 do not cancel
as they did in the proof of Theorem 2.4 unless k = O. Instead, they make a
contribution whose limiting value is equal to
k Ii dr k k (ni 2ni) k
2ni p ~ = 2ni (log ilog p) = 2ni 2  3 = 12"
The rest of the proof is like that of Theorem 2.4 and we obtain (6), which
implies (5). D
From the weight formula (5) we obtain the following theorem.
Theorem 6.2
(a) The only entire modular forms of weight k = 0 are the constant
functions.
(b) Ifk is odd, ifk < 0, or ifk = 2, the only entire modular form of weight k
is the zero function.
(c) Every nonconstant entire modular.form has weight k ~ 4, where k is even.
(d) The only entire cusp form of weight k < 12 is the zero function.
PROOF. Part (a) was proved earlier. To prove (b), (c) and (d) we simply refer
to the weight formula in (5). Since each integer N, N(i), N(p) and N(ioo) is
nonnegative, k must be nonnegative and even, with k ~ 4 if k =1= O. Also,
if k < 12 then N(ioo) = 0 so f is not a cusp form unless f = O. D
116
6.4: Representation of entire forms in terms of G
4
and G
6
6.4 Representation of entire forms in terms
ofG
4
and G
6
In Chapter 1 it was shown that every Eisenstein series G
k
with k > 2 is a
polynomial in G
4
and G
6
. This section shows that the same is true of every
entire modular form. Since the discriminant is a polynomial in G
4
and
G
6
,
L\ = g2
3
 27g
3
2
= (60G
4
)3  27(140G
6
)2,
it suffices to show that all entire forms of weight k can be expressed in terms
of Eisenstein series and powers of The proof repeatedly uses the fact that
the product fg of two entire forms f and g of weights WI and W2' respectively,
is another entire form of weight WI + W2' and the quotient f /g is an entire
form of weight WI  W2 if g has no zeros in H or at ioo.
Notation. We denote by Mkthe set of all entire modular forms of weight k.
Theorem 6.3. Let f be an entire modular form of even weight k 0 and define
Go(r) = 1 for all r. Then f can be expressed in one and only one way as a
sum of the type
(7)
[k/12]
f = L
r=O
k12r*2
where the a
r
are complex numbers. The cusp forms of even weight k are
those sums with ao = O.
PROOF. If k < 12 there is at most one term in the sum and the theorem can
be verified directly. If f has weight k < 12 the weight formula (5) implies
N = N(ioo) = 0 so the only possible zeros of f are at the vertices p and i.
For example, if k = 4 we have N(p) = 1 and N(i) = O. Since G
4
has this
propertY,f/G
4
is an entire modular form of weight 0 and therefore is constant,
so f = a
O
G
4
. Similarly, we find f = aOG
k
if k = 6,8 or 10. The theorem
also holds trivially for k = 0 (since f is constant) and for k = 2 (since the
sum is empty). Therefore we need only consider even k 12.
We use induction on k together with the simple observation that every
cusp form in Mk can be written as a product L\h, where hEMk12'
Assume the theorem has been proved for all entire forms of even weight
<k. The form G
k
has weight k and does not vanish at ioo. Hence if
c = f(ioo)/Gk(ioo) the entire form f  cG
k
is a cusp form in M
k
so
f  cG
k
= L\h, where hEMk 12' Applying the induction hypothesis to h
we have
[(k12)/12]
h = L =
r=O
k 12  12r* 2
[k/12]
L
r= 1
k12r*2
117
6: Modular forms with multiplicative coefficients
Therefore f = cG
k
+ is a sum of the type shown in (7). This proves, by
induction, that every entire form of even weight k has at least one representa
tion of the type in (7). To show there is at most one such representation we
need only verify that the products are linearly independent. This
follows easily from the fact that = °but G
2r
(ioo) =1= 0. Details are left
as an exercise for the reader. D
Since both and G
2r
can be expressed as polynomials in G
4
and G
6
,
Theorem 6.3 also shows that f is a polynomial in G4 and G6. The exact
form of this polynomial is described in the next theorem.
Theorem 6.4. Every entire modular form f of weight k is a polynomial in G
4
and G
6
of the type
(8) f = L ca,bG4aG6b
a,b
where the c
a
,bare complex numbers and the sum is extended over all integers
a 0, b °such that 4a + 6b = k.
PROOF. If k is odd, k < °or k = 2 the sum is empty and f is 0. If k = 0, f is
constant and the sum consists of only one term, co, 0. If k = 4, 6, 8 or 10
then each of the respective quotients f1G
4
, f1G
6
, flG
4
2
and fl(G
4
G
6
) is an
entire form of weight °and hence is constant. This proves (8) for k < 12 or
k odd. To prove the result for even k 12 we use induction on k.
Assume the theorem has been proved for all entire forms of weight < k.
Since k is even, k = 4m or k = 4m + 2 = 4(m  1) + 6 for some integer
m 3. In either case there are nonnegative integers rand s such that
k = 4r + 6s. The form g = G
4
rG
6
s
has weight k and does not vanish at ioo.
Hence if c = f(ioo)lg(ioo) the entire form f  cg is a cusp form in M
k
so
f  cg = where hEM
k

12
. By the induction hypothesis, h can be
expressed as a sum as in (8), taken over all a 0, b °such that 4a + 6b =
k  12. Multiplication by gives a sum of the same type with 4a + 6b = k.
Hence f = cg + is also a sum of the required type and this proves the
theorem. D
6.5 The linear space M
k
and the
subspace Mk,o
The results of the foregoing section can be described in another way. Let
Mk denote the set of all entire forms of weight k. Then Mk is a linear space
over the complex field (since Mk is closed under addition and under multi
plication by complex scalars). Theorem 6.3 shows that Mk is finitedimen
sional with a finite basis given by the set of products occurring
in the sum (7). There are [k/12] + 1 terms in this sum if k =1= 2 (mod 12),
118
(9)
6.6: Classification of entire forms in terms of their zeros
and one less term if k == 2 (mod 12). Therefore the dimension of the space
M
k
is given by the formulas
[ :2J if k == 2 (mod 12),
dim M
k
=
[ lk
2
J+ 1 if k "# 2 (mod 12).
Another basis for M
k
is the set of products G
4
aG
6
b
where a 2 0, b 2 °and
4a + 6b = k (see Exercise 6.12).
The set of all cusp forms in Mk is a linear subspace of Mk which we denote
by Mk, o. The representation in Theorem 6.3 shows that
(10) dim Mk,o = dim M
k
 1
since the cusp forms are those sums in (7) with ao = 0.
We also note that if k 2 12, f E M
k
, 0 if and only if f = I1h, where
hEMk 12· Therefore the linear transformation T : Mk 12 + Mk, 0 defined by
T(h) = ~ h
establishes an isomorphism between Mk,o and M
k

12
. Consequently, if
k 2 12 we have
(11 )
The two formulas (11) and (10) imply
dim Mk = 1 + dim Mk  12
if k ~ 12. This equation, together with the fact that dim M
k
= 1,0, 1, 1, 1, 1
when k = 0, 2, 4, 6, 8, 10, gives another proof of (9).
EXAMPLES. Formula (9) shows that
dim Mk = 1 if k = 4, 6, 8, 10, and 14.
Corresponding basis elements are G
4
, G
6
, G
4
2
, G
4
G
6
, and G
4
2
G
6
.
Formulas (11) and (9) together show that
dim Mk,o = 1 if k = 12, 16,18,20,22, and 26.
Corresponding basis elements are 11, I1G
4
, I1G
6
, I1G
4
2
, I1G
4
G
6
, and I1G
4
2G
6
•
6.6 Classification of entire forms in terms of
their zeros
The next theorem gives another way of expressing all entire forms in terms
of G
4
, G
6
, 11 and Klein's modular invariant J.
Theorem 6.5. Let f be an entire form of weight k and let z 1, ... , ZN denote the
N zeros o.f f in the closure o.f R
r
(omitting the vertices) with zeros of order
119
6: Modular forms with multiplicative coefficients
N(p), N(i) and N(ioo) at the vertices. Then there is a constant c such that
N
(12) 1(1:) = n {J(r)  J(Zk)}'
k=1
PROOF. The product
N
g(r) = n {J(r)  J(Zk)}
k=1
is a modular function with its only zeros in the closure of R
r
at Z1, ... , ZN
and with a pole of order N at ioo. Since has a firstorder zero at ioo, the
product is an entire modular form of weight 12N which, in the closure
of R
r
, vanishes only at ZI' ... , ZN' Therefore the product
h =
has exactly the same zeros asfin the closure of R
r
. Also, h is an entire modular
form having the same weight as f since
k = 4N(p) + 6N(i) + 12N(ioo) + 12N.
Therefore f /h is an entire form of weight 0 so f /h is constant. This proves (12).
D
6.7 The Heeke operators Tn
Heeke determined all entire forms with multiplicative coefficients by intro
ducing a sequence of linear operators T", n = 1, 2, ... , which map the linear
space Mk onto itself. Heeke's operators are defined as follows.
Definition. For a fixed integer k and any n = 1, 2, ... , the operator T" is
defined on Mk by the equation
(13)
In the special case when n is prime, say n = p, the sum on d contains
only two terms and the definition reduces to the formula
(14)
(Tpf)('r) = pkI f(pr) + Pi
l
f(r + b).
P b=O P
The sum on b is the operator encountered in Chapter 4. It maps functions
automorphic under r onto functions automorphic under the congruence
subgroup r o(p).
We will show that T" maps each f in M
k
onto another function in M
k
•
First we describe the action of T" on the Fourier expansion of f
120
6.7: The Heeke operators T,.
Theorem 6.6. Iff E M
k
and has the Fourier expansion
00
f(1:) = L c(m)e21timt,
m=O
then 1;,f has the Fourier expansion
(15)
where
(16)
00
(1;, f) (1:) = L Yn(m)e21timt,
m=O
~ kl (mn)
Yn(m) = i..J d c J2 .
dl(n,m)
PROOF. From the definition in (13) we find
dl 00
(T
n
f)(1:) = n
k

1
L d
k
L L c(m)e21tim(nt+bd)/d2
din b=O m=O
00 ( )kl 1 dl
= L L ~ c(m)e21timnt/d2  L e21timb/d.
m=O din d d b=O
The sum on bis a geometric sum which is equal to d if dim, and is 0 otherwise.
Hence
00 ( )k 1
(1;,f)(1:) = L L ~ c(m)e21timnt/d2.
m= 0 din, dim d
Writing m = qd we have
(T,J)(r) = f L ( ~ ) k  1c(qd)e21<iQnt/d,
q=O din d
In the sum on d we can replace d by nld to obtain
(T"f)(r) = f L dklc(qn)e21tiqdt.
q=O din d
If x = e
21tit
the last sum contains powers of the form x
qd
. We collect those
terms for which qd is constant, say qd = m. Then q = mid and dim so
00 kl (mn) m
(T" f)(r) = L L d c (j2 x ,
m=O dln,dlm
which implies (16). o
Our next task is to prove that 1;, maps Mk into itself. For this purpose
we note that the definition of 1;, f can be written in a slightly different form.
We write n = ad and let
a1: + b
A1:=d·
121
6: Modular forms with multiplicative coefficients
Then (13) takes the form
1
(17) (T"f)(r) = n
k

I
L d
k
f(Ar) =  L a
k
f(Ar).
a ~ I , ad = n n a ~ I , ad = n
O ~ b < d O ~ b < d
The matrix ( ~ ~ ) which represents A has determinant ad = n. To deter
mine the behavior of T"f under transformations of the modular group r
we need some properties of transformations with determinant n. These are
described in the next section.
6.8 Transformations of order n
Let n be a fixed positive integer. A transformation of the form
ar + b
Ar = er + d'
where a, b, e, d are integers with ad  be = n, is called a transformation of
order n. It can be represented by the 2 x 2 matrix
A = G~ )
where, as usual, we identify each matrix with its negative.
We denote by r(n) the set of all transformations of order n. The modular
group r is r(l).
Two transformations Al and A
z
in r(n) are called equivalent, and we
write AI I'"V A
z
, if there is a transformation V in r such that
Al = VAz·
The relation I'"V is obviously reflexive, symmetric, and transitive, and hence
is an equivalence relation. Consequently, the set r(n) can be partitioned
into equivalence classes such that two elements of r(n) are in the same class
if, and only if, they are equivalent. The next theorem describes a set of
representatives.
Theorem6.7. In every equivalence class of r(n) there is a representative of
triangular form
(
a b) where d > O.
o d'
PROOF. Let A = (: ~ ) be an arbitrary element of r(n). If c = 0 there is
nothing more to prove. If e # 0 we reduce the fraction  ale to lowest terms.
That is, we choose integers rand s such that sir = ale and (r, s) = 1.
122
6.8: Transformations of order n
Next we choose two integers p and q such that ps  qr = 1 and let
v = ( ~ ~ }
Then V E rand
VA = (p q) (a b) (pa + qc pb + qd).
r s c d = ra + sc rb + sd
Since ra + sc = 0 and det(VA) = det V det A = n we see that VA E r(n) so
VA ~ A. Hence VA or its negative is the required representative. D
Theorem 6.8. A complete system qf nonequivalent elements in r(n) is given by
the set qf transformations ql' triangular form
(18)
and
where d runs through the positive divisors of n and, for each fixed d,
a = njd, and b runs through a complete residue system modulo d.
PROOF. Theorem 6.7 shows that every element in r(n) is equivalent to one
of the transformations in (18). Therefore we need only show that two such
transformations, say
_(a
1
b
1
)
At  0 d
l
are equivalent if, and only if,
(19) and
If(19) holds then b
z
= b
1
+ qd
1
for some integer q and we can take
V = ( ~ ~ }
Then VAl = A
z
so Al ~ A
z
.
Conversely, if A 1 ~ A
z
there is an element
in r such that A
z
= VA1. Therefore
(20) (a
z
b
z
) = (p q)(a
1
b
1
) = (pal pb
1
+ qd
1
).
o dz r sOd1 ra1 rb1 + sd1
Equating entries we find ra
1
= 0 so r = 0 since a
1
# 0 because a1d1 = n ~ 1.
Now ps  qr = 1 so ps = 1 hence both p and s are 1 or both are 1. We can
assume p = s = 1 (otherwise replace V by  V). Equating the remaining
entries in (20) we find a
z
= aI' d
z
= d
1
, b
2
= b
1
+ qd
1
, so b
z
== b
1
(mod d
1
).
This completes the proof. D
123
(21)
6: Modular forms with multiplicative coefficients
Note. The sum in (17) defining 1;, f can now be written in the form
1
(1;,f)(7:) =  L a
k
f(A7:),
n A
where A runs through a complete set of nonequivalent elements in r(n) of
the form described in Theorem 6.8. The coefficient a
k
is the kth power of
the entry in the first row and first column of A.
Theorem 6.9. If Al E r(n) and VI E r, then there exist matrices A
2
in r(n)
and V
2
in r such that
(22)
and
Moreover, if
Ai = ( ~ ~ : )
for i = 1, 2, then we have
(23) al(Y2A27: + £5
2
) = a2(YI7: + £5
1
).
PROOF. Since det(A
I
VI) = det Al det VI = n, the matrix Al VI is in r(n) so,
by Theorem 6.7, there exists A
2
in r(n) and V
2
in r such that (22) holds. To
verify (23) we first note that A1 VI has the form .
and that
Equating entries in the second row we find
d
1
d
2
1'1 d
2
Y2 = = Yl
n al
and
124
and
6.9: Behavior of ~ f under the modular group
since at dt = n. Hence
atY2 = d
2
Yt
and we obtain
at(Y2 A
2
r + £5
2
) = at Y2 A
2
r + at£5
2
which proves (23).
6.9 Behavior of Tn/under the modular group
Theorem6.10. If f E Mk and V = ( ~ ~ ) E r then
(24) (1;,f)(Vr) = (yr + £5)k(1;,f)(r).
PROOF. We use the representation in (21) to write
1
(T,. f)(r) =  L at k f(A
t
r)
n Al .
D
(25)
(
at ht) .
where At = 0 d
t
and At runs through a complete set ofnoneqUlvalent
elements in r(n). Replacing r by Vr we find
1
(T,.f)(Vr) =  L atkf(A
t
Vr).
nAt
By Theorems 6.7 and 6.9, there exist matrices
A
2
= ( ~ ~ : ) in r(n) and V
2
= G: ~ : ) in r
such that
and
Therefore
atkf(A
t
Vr) = at
k
f(V
2
A
2
r) = atk(Y2A2 r + £5
2
)kf(A
2
r)
= a2
k
(yr + £5)k f(A
2
r)
since f E M
k
• Now as At runs through a complete set of nonequivalent
elements of r(n) so does A
2
• Hence (25) becomes
1
(T,.f)(Vr) =  (yr + £5)k L a2
k
f(A
z
r) = (yr + £5)k(T,.f)(r). D
n A2
125
6: Modular forms with multiplicative coefficients
The next theorem shows that each Hecke operator 1;. maps M
k
into M
k
and also maps Mk,o into Mk,o.
Theorem 6.11. If f E M
k
then 1;.f E M
k
. Moreover, if f is a cusp form then
1;. f is also a cusp form.
PROOF. If f E M
k
the definition of 1;. shows that T"f is analytic everywhere
in H. Theorem 6.6 shows that 1;.f has a Fourier expansion of the required
form and that 1;.f is analytic at ioo. And Theorem 6.10 shows that 1;.f has
the proper behavior under transformations of r. Finally, if f is a cusp form,
the Fourier expansion in Theorem 6.6 shows that T" f is also a cusp form. D
6.10 Multiplieative property of Heeke
operators
This section shows that any two Hecke operators 1;. and T
m
defined on M
k
commute with each other. This follows from a multiplicative property of the
composition T
m
1;.. First we treat the case in which m and n are relatively
prime.
Theorem 6.12. If (m, n) = 1 we have the composition property
(26)
PROOF. If f E M
k
we have
1
(1;.f)(r) =  L a
k
f(Ar),
n 1,ad=n
where A = :). Applying T
m
to each member we have
1 1
{T
m
(1;.(f))}(r)= L rt
k
 L akf(BAr),
m IX 1 , IXlJ = m n a 1, ad = n
where B = This can be written as
1
(27) {(T
m
T,,)(f)} (r) =  L L (rta)k f(Cr),
mn IX 1 , IXlJ =m a 1, ad =n
where
C = BA = =
As d and b run through the positive divisors of nand m, respectively, the
product db runs through the positive divisors of mn since (m, n) = 1. The
126
6.10: Multiplicative property of Hecke operators
linear combination rxb + f3d runs through a complete residue system
mod db as band f3 run through complete residue systems mod d and b,
respectively. Therefore the matrix C runs through a complete set of non
equivalent elements of r(mn) and we see that (27) implies (26). D
The next theorem extends the composition property in (26) to arbitrary
mand n. For convenience in notation we write T(n) in place of 1;,.
Theorem 6.13. Any two Hecke operators T(n) and T(m) defined on M
k
commute
with each other. Moreover, we have the composition formula
(28) T(m)T(n) = L d
k

1
T(mnjd
2
).
dl(m,n)
PROOF. Commutativity follows from (28) since the right member is symmetric
in m and n. If (m, n) = 1 formula (28) reduces to (26). Therefore, to prove
(28) it suffices to treat the case when mand n are powers of the same prime p.
First we consider the case m = p and n = pr, where r 1. In this case we
are to prove that
(29)
(30)
We use the representation in (17) and note that the divisors of pr have the
form pt where 0 t r. Hence we have
{T(pr)f}(r) = pr L p(rt)kf(prtrt+ ht).
O:::;t::;r P
O:::;bt<pt
By (14) we have
pl (r + b)
{T(p)g}(r)=pkl
g
(pr)+pl
Lg
,
b=O P
so when we apply T(P) to each member of (30) we find
{T(p)T(pr)f} (r) = pklr L p(rt)k f(pr+ + Pht)
O:::;t:::;r P
O:::;bt<pt
pl (rt b b t)
+  1  r " (r  t)k "f p r + t + P
P i...J P i...J t+l .
O:::;t::;r b=O P
O:::;bt<pt
In the second sum the linear combination b
t
+ bpt runs through a complete
residue system mod pt+ 1. Since r  t = (r + 1)  (t + 1) the second sum,
together with the term t = 0 from the first sum, is equal to {T(pr+ 1 )f} (r). In
the remaining terms we cancel a factor p in the argument of f, then transfer
the factor pk to each summand to obtain
{T(p)T(pr)f} (r) = {T(pr+ 1 )f} (r) + plr L p(r+ It)k ht).
1 :::;t:::;r P
O:::;bt<pt
127
6: Modular forms with multiplicative coefficients
Dividing each b
t
by pt 1 we can write
b
t
= qtpt1 + r
n
where 0 ~ r
t
< pt1 and qt runs through a complete residue system mod p.
Since f is periodic with period 1 we have
(
prt
r
+ bt) = (prt
r
+ rt)
f t 1 f t 1 '
P P
so as qt runs through a complete residue system mod p each term is repeated
p times. Replacing the index t by t  1 we see that the last sum is pk  1 times
the sum defining {T(pr  1) f} (r). This proves (29).
Now we consider general powers of the same prime, say m = pS and
n = pro Without loss of generality we can assume that r ~ S. We will use in
duction on r to prove that
(31)
for all r and all s ~ r. When r = 1, (31) follows for all s ~ 1 from (29).
Therefore we assume that (31) holds for r and all smaller powers and all
s ~ r, and prove it also holds for r + 1 and all s ~ r + 1.
By (29) we have
T(P) T(p') T(pS) = T(pr + 1) T(pS) + pk  1T(pr  1) T(pS),
and by the induction hypothesis we have
r
T(p)T(pr)T(ps) = L pt(kl)T(p)T(pr+s2t).
t=O
Equating the two expressions, solving for T(pr+ 1)T(pS) and using (29) in the
sum on t we find
r ,
T(pr+ 1)T(pS) = L pt(k1)T(pr+s+ 1 2t) + L p(t+ 1)(k1)T(p,+s12t)
t=O t=O
By the induction hypothesis the last term cancels the second sum over t
except for the term with t = r. Therefore
r
T(pr+1)T(pS) = Lpt(k1)T(pr+s+12t) + p(r+1)(k1)T(ps1r)
t=O
r+ 1
= L pt(k1)T(pr+1+s2t).
t=O
This proves (31) by induction for all r and all s ~ r, and also completes the
proof of (28). D
128
6.11: Eigenfunctions of Hecke operators
6.11 Eigenfunctions of Hecke operators
In Theorem 6.6 we proved that if f E M
k
and has the Fourier expansion
(32)
00
f(1:) = L c(m)x
m
,
m=O
where x = e
21tit
, then 1;,f has the Fourier expansion
(33)
where
00
(1;,f)(r) = L yn(m)x
m
,
m=O
(34)
(35)
~ kl (mn)
Yn(m) = i..J d C d
2
•
dl(n,m)
When m = 0 we have (n, 0) = n so the constant terms of f and 1;,f are
related by the equation
Yn(O) = Ldk1C(0) = (Tkl(n)C(O)
din
for all n 2 1. Similarly, when m = 1 we find
(36)
Yn(l) = c(n)
for all n 2 1.
The sum on the right of (34) resembles that which occurs in the multi
plicative property of Ramanujan's function 1:(n) and the divisor functions
(Ja(n). These examples suggest we seek those formsffor which the transformed
function 1;, f has Fourier coefficients
(37)
Yn(m) = c(n)c(m)
since this would imply the multiplicative property
c(n)c(m) = L d
k
 l C ( ; ~ ) .
dl(n,m)
The relation (37) is equivalent to the identity
T,. f = c(n) (
for all n 2 1. A nonzero function f satisfying a relation of the form
(38) 1;,f = A(n)f
for some complex scalar A(n) is called an eigenfunction (or eigenform) of the
operator Tn' and the scalar A(n) is called an eigenvalue of 1;,. If f is an
eigenform so is cf for every c # o.
129
6: Modular forms with multiplicative coefficients
EXAMPLES. If a linear operator T maps a Idimensional function space V
into itself, then every nonzero function in Vis an eigenfunction of T. Formula
(9) shows that
dim Mk = 1 if k = 4, 6, 8, 10 and 14,
so each Hecke operator T" has eigenforms in Mk for each of these values of k.
For example, the respective Eisenstein series G
4
, G
6
, G
s
, G
IO
and G
I4
are
eigenforms for each T".
Similarly, formula (11) implies that
dim Mk,o = 1 if k = 12,16,18,20,22 and 26,
so each T" has eigenforms in Mk, 0 for each of these values of k. The respective
cusp forms t1, t1G
4
, t1G
6
, t1G
s
, t1G
IO
and t1G
14
are eigenforms for each T".
If f is an eigenform for every Hecke operator T", n 1, then f is called a
simultaneous eigenform. All the examples just mentioned are simultaneous
eigenforms.
6.12 Properties of simultaneous eigenforms
Theorem 6.14. Assume k is even, k 4. If the space M
k
contains a simultaneous
eigenform f with Fourier expansion (32), then c(l) # O.
PROOF. The coefficient of x in the Fourier expansion of T"f is Yn(l) = c(n).
Since f is a simultaneous eigenform this coefficient is also equal to A(n)c(l), so
c(n) = A(n)c(l)
for all n 1. If c(l) = 0 then c(n) = 0 for all n 1 and f(1:) = c(O). But then
c(O) = 0 since k 4, hence f = 0, contradicting the definition of eigenform.
This proves that c(l) # O. D
An eigenform with c(l) = 1 is said to normalized. If M
k
contains a simul
taneous eigenform then it also contains a normalized eigenform since we
can always make c(l) = 1 by multiplying f by a suitable nonzero constant.
It is easy to characterize all cusp forms which are simultaneous eigenforms.
Since the zero function is the only cusp form of weight < 12 we need consider
only k 12.
Theorem 6.15. Assume f E Mk, 0 where k is even, k 12. Then f is a simul
taneous normalized eigenform if, and only if, the coe.fficients in the Fourier
expansion (32) the multiplicative property
(39) c(m)c(n) = L d
k
 I
dl<n,m)
for all m 1, n 1, in which case the coefficient c(n) is an eigenvalue ofT".
130
6.13: Examples of normalized simultaneous eigenforms
PROOF. The equation T" f == A(n).f is equivalent to the relation
(40) Yn(m) == A(n)c(m)
obtained by equating coefficients of x
m
in the corresponding Fourier expan
sions. Since f is a cusp form so is T" f hence (40) is to hold for all m ~ 1
and n ~ 1. Now Yn(1) == c(n) so (40) implies A(n) == c(n) if c(1) == 1, and
hence Yn(m) == c(n)c(m). On the other hand, Equation (34) shows that (40)
is equivalent to (39) if c( 1) == 1. Therefore f is a normalized simultaneous
eigenform if, and only if, (39) holds for all m ~ 1, n ~ 1. D
6.13 Examples of normalized simultaneous
eigenforms
The discriminant L\ is a cusp form with Fourier expansion
00
L\(r) == (2n)12 L r(m)x
m
m=1
where r(1) == 1. Therefore (2n) 12 L\(r) is a normalized eigenform for each
T" with corresponding eigenvalue r(n). This also proves that Ramanujan's
function r(n) satisfies the multiplicative property in (3).
The next theorem shows that the only simultaneous eigenforms in M2k
which are not cusp forms are constant multiples of the Eisenstein series G
2k
.
Theorem 6.16. Assume that.1' E M2k, where k 2 2, and that f is not a cusp
form. Then f is a normalized simultaneous eigeYl:form (f, and only if,
(41)
(2k  I)!
f(r) = 2(2ni)2k G2k(r).
PROOF. In the Fourier expansion (32) we have c(O) # 0 since f is not a cusp
form. The relation
(42)
is equivalent to the relation
(43)
T" f == A(n)f
Yn(m) == A(n)c(m)
obtained by equating coefficients of x
m
in the corresponding Fourier expan
sions. When m == 0 this becomes
Yn(O) == A(n)c(O).
On the other hand, (35) implies Yn(O) = 0"2k1(n)c(0) since f EM2k. But
c(O) # 0, so Equation (42) holds if, and only if,
A(n) = 0"2k1(n).
131
6: Modular forms with multiplicative coefficients
D
Using this in (43) we find that
Yn(m) = (J 2k  1(n)c(m).
When m = 1 this relation, together with (36), gives us
c(n) = (J2k  1(n)c(I).
Therefore,! is a normalized simultaneous eigenform in M 2k if, and only if,
c(n) = (J2k  1(n)
for all n ~ 1. Since the Eisenstein series G
2k
has the Fourier expansion
2(2ni)2k 00
G2k(r) = 2((2k) + (2k _ I)! m ~ l 0"2k_l(m)x
m
,
the function in (41) is normalized and its Fourier expansion is given by
(2k  I)! 00
f(r) = (2 yk ((2k) + L 0"2k l(m)x
m
.
nl m= 1
(44)
Note. Since
k+ 1 (2n)2k
((2k) = (1) 2(2k)! B2k
where B
k
is the kth Bernoulli number defined by
x ~ Bk k
~ 1 = i..J k'x,
e k=O •
the constant term in (44) is equal to B
2k
/(4k). (See [4J, Theorem 12.17.)
We can also write
Since the eigenvalue 2(n) in (42) is (J 2k _1(n), Theorem 6.16 shows that the
divisor functions (Ja(n) satisfy the multiplicative property in Equation (4)
when (X = 2k  1. Actually, they satisfy (4) for all real or complex (x, but
O'a(n) is the nth coefficient of an entire form only when (X is an odd integer ~ 3.
EXAMPLES. The problem of determining all entire noncusp forms whose
coefficients satisfy the multiplicative property (39) has been completely
settled by Theorem 6.16. For the cusp forms the problem has been reduced
by Theorem 6.15 to that of determining simultaneous normalized eigenforms
of even weight 2k ~ 12. We have already noted that the function (2n) 12~ ( r )
is the only simultaneous normalized eigenform of weight 2k = 12. Also
there is exactly one simultaneous normalized eigenform for each of the
weights
2k = 16, 18, 20, 22, and 26
132
6.14: Remarks on existence of simultaneous eigenforms in M2k,O
since dim M2k, 0 = 1 for these weights. The corresponding normalized
eigenforms are given by
G ( ) 00 n{ 2(2k  12) 00 m}
(
2 ) 12L\( ). 2 k  12 r ,,() 1 "( )
n r 2((2k12)  B
2k

12
.
We define r(O) = 0 and (J2k1(0) =  B
2k
/(4k). Then the coefficients c(n) of
these eigenforms are given by the Cauchy product
4k  24 n
c(n) =  B L r(m)0"2k13(n  m).
2k12 m=O
They satisfy the multiplicative property
c(m)c(n) = L
dl(m,n)
for all m 1, n 1.
6.14 Remarks on existence of simultaneous
eigenforms in M2k,O
Let K = dim M2k, 0 where 2k 12. Then we' have
[
2
1
k
2
J 1
if 2k == 2 (mod 12)
K=
[
2
1
k
2
J
if 2k ¢ 2 (mod 12).
Let e(k) denote the number of linearly independent simultaneous eigenforms
in M
2k
,O. Clearly, e(k) K. We have shown that e(k) = 1 when K = 1.
Hecke showed that e(k) = 2 when K = 2, and later Petersson [32] showed
that e(k) = K in all cases. He did this by introducing an inner product
(f, g) in M2k, 0 defined by the double integral
(J, g) = ff j(r)g(r)v
2k
 2 du dv
Rr
extended over the fundamental region R
r
in the r = u + iv plane. Relative
to the Petersson inner product the Hecke operators are Hermitian, that is,
they satisfy the relation
(T" f, g) = (f, T"g)
for any two cusp forms in M2k, o. Therefore, by a well known theorem of
linear algebra (see [2], Theorem 5.4) for each T" there exist K eigenforms
which form an orthonormal basis for M2k, o. These need not be simultaneous
eigenforms for all the T,.. However, since the T,. commute with each other,
another theorem of linear algebra (see [10], Ch. IX, Sec. 15) shows that
133
6: Modular forms with multiplicative coefficients
M2k, 0 has an orthonormal basis consisting of K simultaneous eigenforms.
Each of these can be multiplied by a constant factor to get a new basis of
simultaneous normalized eigenforms. (The new basis will be orthogonal
but need not be orthonormal.) Since the 1;. are Hermitian, the corresponding
eigenvalues are real. Details of the proofs of these statements can be found
in references [32], [26], or [11].
6.15 Estimates for the Fourier coefficients of
entire forms
Assume f is an entire form with Fourier expansion
(45)
00
f(r) = Lc(n)x
n
,
n=O
(46)
where x = e
21tit
. Write r = u + iv so that x = e21tve21tiu. For fixed v > 0,
as u varies from 0 to 1 the point x traces out a circle C(v) of radius e 21tv
with center at x = O. By Cauchy's residue theorem we have
1 I f(r) II ·n
c(n) = 2. ,,:tT dx = f(u + lV)X, duo
nl C(v) x 0
We shall use this integral representation to estimate the order of magnitude
of Ic(n) I. First we consider cusp forms of weight 2k.
Theorem 6.17. Iff E M
2k
,O we have
c(n) = O(n
k
).
PROOF. The series in (45) converges absolutely if Ix I < 1. Since c(O) = 0 we
can remove a factor x and write
II(r)1 = IXI!J1c(n)xn11 ~ IxIJllc(n)lIxlnl.
If r is in R
r
, the fundamental region of r, then r = u + iv with v ~ fi/2
> 1/2, so Ixl = e
21tv
< e
1t
. Hence
If(r)1 ~ A Ixl = Ae
21tv
where
00
A = LIc(n)le(nl)1t.
n=1
This implies
(47)
Now define
g(r) = t Ir  rI = v
134
6.15: Estimates for the Fourier coefficients of entire forms
ifrEH. Then
g(Ar) = ler + dl
2
g(r)
if A = (: r, so gk(Ar) = Icr + dl
2k
l(r). Therefore the product
qJ(r) = If (r )Igk(r) = If (r) Iv
k
is invariant under the transformations of r. Moreover, qJ is continuous
in R
r
, and (47) shows that qJ(r) + 0 as v + +00. Therefore qJ is bounded
in R
r
and, since qJ is invariant under r, qJ is also bounded in H, say
IqJ(r) I ::; M
for all r in H. Therefore
If(r)1 ::; Mv
k
for all r in H. Using this in (46) we find
Ic(n) I :0:;; f If(u + iv)x
n
Idu :0:;; Mvklxl
n
= Mvke21Cnv.
This holds for all v > O. When v = lin it gives us
Ic(n)1 s Mn
k
e
21t
= O(n
k
).;
Theorem 6.18. If f E M2k and f is not a cusp form, then
o
(48) c(n) = O(n
2k
 1).
PROOF. If f = G
2k
each coefficient c(n) is of the form CllT
2k
l(n) where Cl is
independent of n. Hence
Now
lT2kl(n) = = n
2k

1
:0:;; n
2k

1
f = O(n
2k

1
),
din d= 1
so (48) holds if f = G
2k
·
For a general noncusp form in M2k, let 2 = f(ioo )/G
2k
(ioo). Thenf  2G
2k
is a cusp form so
f = 2G
2k
+ g
where gEM2k, o. The Fourier coefficients of f are the sum of those of 2G
2k
and g so they have order of magnitude
o
135
6: Modular forms with multiplicative coefficients
Note. For cusp forms, better estimates for the order of magnitude of the
c(n) have been obtained by Kloosterman, Salie, Davenport, Rankin, and
Selberg (see [46]). It has been shown that
c(n) = O(n
k
(1/4)+C)
for every 8 > 0, and it has been conjectured that the exponent can be further
improved to k  ! + 8. For the discriminant Ramanujan conjectured the
sharper estimate
Ir(p) I :::; 2
p
ll/2
for primes p. This was recently proved by P. Deligne [7].
6.16 Modular forms and Dirichlet series
Hecke found a remarkable connection between each modular form with
Fourier series
(49)
and the Dirichlet series
(50)
00
f(r) = c(O) + L c(n)e21tint
n= 1
qJ(S) = f C(7)
n= 1 n
(51)
(52)
formed with the same coefficients (except for c(O)). If f EM2k then c(n) =
O(n
k
) iffis a cusp form, and c(n) = O(n
2k

1
) iffis not a cusp form. Therefore,
the Dirichlet series in (50) converges absolutely for (J = Re(s) > k + 1 if f is
a cusp form, and for (J > 2k if f is not a cusp form.
Theorem 6.19. If the coefficients c(n) the multiplicative property
c(m)c(n) = L
dl<m,n)
the Dirichlet series will have an Euler product representation of the form
1
qJ(s) = n1 ( ) 5 + 2kl 25'
p cpp P p
absolutely convergent with the Dirichlet series.
PROOF. Since the coefficients are multiplicative we have (see [4], Theorem
11.7)
(53)
136
6.16: Modular forms and Dirichlet series
whenever the Dirichlet series converges absolutely. Now (51) implies
c(p)c(pn) = c(pn+ 1) + p2k1
c
(pn1)
for each prime p. Using this it is easy to verify the power series identity
(1 c(P)x + p2kl X2)(1 +J/(pn)x
n
) = 1
for all Ix I < 1. Taking x = p s, we find that (53) reduces to (52). 0
EXAMPLE. For the Ramanujan function we have the Euler product represen
tation
00 r(n) 1
Jl ;:;s = I] 1  r(p)pS + pll2s
for (J > 7 since r(n) = O(n
6
).
Hecke also deduced the following analytic properties of cp(s).
Theorem 6.20. Let <p(s) be the function defined for (J > k by the Dirichlet
series (50) associated with a modular form f(r) in M
k
having the Fourier
series (49), where k is an even integer ~ 4. Then <p(s) can be continued
analytically beyond the line (J = k with the following properties:
(a) If c(O) = 0, <p(s) is an entire function of s.
(b) If c(O) i= 0, <p(s) is analytic for all s except for a simple pole at s = k '
with residue
( 1)k/2
c
(0)(2n)k
r(k)
(c) The ,function <p satisfies the functional equation
PROOF. From the integral representation for r(s) we have
if (J > O. Therefore if (J > k we can multiply both members by c(n) and sum
on n to obtain
(2n)sr(s)q>(s) = fO{f(iy)  c(O)}ySl dy.
137
6: Modular forms with multiplicative coefficients
Since f is a modular form in Mk we have f(ijy) = (iy)k f(iy) so
(2n)sr(s)qJ(s) = {>J{f(i
Y
)  c(O)}ySl dy + f{(iy)kfG)  c(O)}ySl dy
= foo {f(iy)  c(O)}ySl dy + i
k
foo f(iw)W
k

s

1
dw _ c(O)
lIS
= lOO{f(iy)  c(O)}ySl dy
+ (It/
2
lOO {f(iw)  c(O)}W
k

S

1
dw
+ (1)k/2
C
(0) fOOW
k

s

1
dw _ c(O)
1 S
= fOO {f(iy)  c(O)}(yS + (1)k/2l
S
) dy
1 y
(
1 (_I)k/2)
 c(O)
Although this last relation was proved under the assumption that (J > k, the
right member is meaningful for all complex s. This gives the analytic continua
tion of qJ(s) beyond the line (J = k and also verifies (a) and (b). Moreover,
replacing s by k  s leaves the right member unchanged except for a factor
(_I)k/2 so we also obtain (c). D
Hecke also proved a converse to Theorem 6.20 to the effect that every
Dirichlet series qJ which satisfies a functional equation of the type in (c),
together with some analytic and growth conditions, necessarily arises from
a modular form in M
k
• For details, see [15].
Exercises for Chapter 6
Exercises 1 through 6 deal with arithmetical functions f satisfying a relation
of the form
(54)
f(m)f(n) = L
dl(m,n)
for all positive integers m and n, where rx is a given completely multiplicative
function (that is, rx(l) = 1 and rx(mn) = rx(m)rx(n) for all m and n). An arith
metical function satisfying (54) will be called rxmultiplicative. We write
f = 0 if f(n) = 0 for all n.
1. Assume f is and f =1= o. Prove that f( 1) = 1. Also prove that cf is
amultiplicative if, and only if, c = 0 or c = 1.
138
Exercises for Chapter 6
2. If f and g are exmultiplicative, prove that f + g is exmultiplicative if, and only if,
f = 0 or g = O.
3. Let II' ... , fk be k distinct nonzero exmultiplicative functions. If a linear combination
k
f = I Cih
i=1
is also exmultiplicative, prove that:
(a) The functions II' ... , Ik are linearly independent.
(b) Either all the Ci are 0 or else exactly one of the Ci is 1 and the others are O. Hence
either f = 0 or f = h for some i. In other words, linear combinations of ex
multiplicative functions are never exmultiplicative except for trivial cases.
4. If I is exmultiplicative, prove that
a(n)f(m) = I Jl(d)f(mnd)f(!!).
din d
5. If f is multiplicative, prove that f is exmultiplicative if, and only if,
(55)
for all primes p and all integers k ~ 1.
6. The recursion relation (55) shows that f(pn) is a polynomial in f(p), say
f(pn) = Qn(f(P)).
The sequence {Qn(x)} is determined by the relations
Ql(X) = x, Q2(X) = X
Z
 ex(p), Qr+ l(X) = xQr(x)  ex(p)Qr1(X) for r 2: 2.
Show that
Qi2ex(p)I/Z
X
) = a(p)"/z Un(x),
where Un(x) is the Chebyshev polynomial of the second kind, defined by the relations
U
1
(x) = 2x, Uz(x) = 4x
z
 1, U
r
+
1
(x) = 2xU
r
(x)  U
r

1
(x) for r 2: 1.
7. Let £Zk(r) = tG
zk
(r)/((2k). If x = e
Z1tit
verify that the Fourier expansion of £Zk(r)
has the following form for k = 2, 3, 4, 5, 6, and 7:
00
£4(r) = 1 + 240 I0"3(n)x
n
,
n=l
00
£6(r) = 1  504 IO"s(n)x
n
,
n= 1
00
£8(r) = 1 + 480 I0"7(n)x
n
,
n= 1
00
£10(r) = 1  264 I0"9(n)x
n
,
n=1
00
£14(r) = 1  24 IO"I3(n)x
n
.
n=1
139
6: Modular forms with multiplicative coefficients
Derive each of the identities in Exercises 8, 9, and 10 by equating coefficients
in appropriate identities involving modular forms.
n1
8. 0"7(n) = 0"3(n) + 120 I 0"3(m)0"3(n  m).
m=l
n1
9. 110"9(n) = 210"5(n)  100"3(n) + 5040 I 0"3(m)0"5(n  m).
m=1
65 691 691 n1
10. r(n) = 0"11(n) + 0"5(n)   I 0"5(m)0"5(n  m).
756 756 3 m= 1
Show that this identity implies Ramanujan's congruence
r(n) == 0"11(11) (mod 691).
11. Prove that the products G k  1 2 r ~ r which occur in Theorem 6.3 are linearly inde
pendent.
12. Prove that the products G
4
Q
G
6
b
are linearly independent, where a and b are non
negative integers such that 4a + 6b = k.
13. Show that the Dirichlet series associated with the normalized modular form
(2k  1)! Y(2k ~ ( ) 21timt
f (r) = . 2k S ) + ~ 0" 2k  1 m e
(2nl) m= 1
is q>(s) = ((s)((s + 1  2k).
14. Aquadratic polynomial 1  Ax + Bx
2
with real coefficients A and B can be factored
as follows:
1  Ax + Bx
2
= (1  r
1
x)(1  r2x).
Prove that '1 = 'Y. + if1 and '2 = }'  ip, where rJ., {3, I are real and {3()'  rJ.) = O.
Hence, if {3 # 0 the numbers, 1 and r2 are complex conjugates.
Note. For the quadratic polynomial occurring in the proof of Theorem
6.19 we have
where
and
Petersson conjectured that '1 and '2 are always complex conjugates. This
implies
and
When c(n) = r(n) this is the Ramanujan conjecture. The Petersson conjecture
was proved recently by Deligne [7].
140
(56)
(57)
Exercises for Chapter 6
15. This exercise outlines Riemann's derivation of the functional equation
n
S
/
2
rG}(s) = n(SI)/
2r
C~ s)W  s)
from the functional equation (see Exercise 4.1)
.9( ~
1
) = (_ ir)I/2.9(r)
satisfied by Jacobi's theta function
00
:J(r) = 1 + 2 Lerrin2t.
n=l
(a) If (J > 1 prove that
S/2
r
(S) s foo rrn
2
x s/2
1
d
n n = e x x
2 0
and use this to derive the representation
n
S
/
2
rG)c(s) = faror/J(x)x
s
/
2

1
dx,
where 21/J(x) = 9(x)1.
(b) Use (a) and (57) to obtain the representation
nS/2r(:)((s) = __1_ + joo(XS/21 + X(1s)/21)I/J(X) dx
2 s(s  1) 1
for (J > 1.
(c) Show that the equation in (b) gives the analytic continuation of ((s) beyond the
line (J = 1 and that it also implies the functional equation (56).
141
Kronecker's theorem
with applications
7.1 Approximating real numbers by rational
numbers
Every irrational number fJ can be approximated to any desired degree of
accuracy by rational numbers. In fact, if we truncate the decimal expansion
of fJ after n decimal places we obtain a rational number which differs from
fJ by less than 10
n
• However, the truncated decimals might have very large
denominators. For example, if
fJ = n  3 = 0.141592653 ...
the first five decimal approximations are 0.1, 0.14, 0.141, 0.1415, 0.14159.
Written in the form alb, where a and b are relatively prime integers, these
rational approximations become
1 7 141 283 14159
10' 50' 1000' 2000' 100,000·
On the other hand, the fraction 1/7 = 0.142857 ... differs from fJ by less than
2/1000 and is nearly as good as 141/1000 for approximating fJ, yet its denomi
nator 7 is very small compared to 1000.
This example suggests the following type of question: Given a real
number fJ, is there a rational number h/k which is a good approximation to
fJ but whose denominator k is not too large?
This is, of course, a vague question because the terms" good approxima
tion" and "not too large" are vague. Before we make the question more
precise we formulate it in a slightly different way. If fJ  h/k is small, then
(kfJ  h)/k is small. For this to be small without k being large the numerator
kfJ  h should be small. Therefore, we can ask the following question:
142
7.2: Dirichlet's approximation theorem
Given a real number fJ and given e > 0, are there integers hand k such that
IkfJ  hi < e?
The following theorem of Dirichlet answers this question in the affirma
tive.
7.2 Dirichlet's approximation theorem
Theorem7.1. Given any real fJ and any positive integer N, there exist integers
hand k with 0 < k ~ N such that
(1)
1
IkfJ  hi <
N·
(2)
PROOF. Let {x} = x  [x] denote the fractional part of x. Consider the
N + 1 real numbers
0, {fJ}, {2fJ}, ... , {NfJ}.
All these numbers lie in the half open unit interval 0 ~ {mfJ} < 1. Now
divide the unit interval into N equal halfopen subintervals of length liN.
Then some subinterval must contain at least two of these fractional parts,
say {afJ} and {bfJ}, where 0 ~ a < b ~ N. Hence we can write
1
I{bfJ}  {afJ} I < N·
But
{bfJ}  {afJ} = bfJ  [bfJ]  afJ + [afJ] = (b  a)fJ  ([bfJ]  [afJ]).
Therefore if we let
k=ba
inequality (2) becomes
and h = [bfJ]  [afJ]
1
IkfJ  hi < N' with 0 < k ~ N.
This proves the theorem. o
Note. Given e > 0 we can choose N > lie and (1) implies IkfJ  hi < E.
The next theorem shows that we can choose hand k to be relatively
prime.
Theorem 7.2. Given any real fJ and any positive integer N, there exist relatively
prime integers hand k with 0 < k ~ N such that
1
IkfJ  hi < N·
143
7: Kronecker's theorem with applications
PROOF. By Theorem 7.1 there is a pair h', k' with 0 < k' ~ N satisfying
I
h'l 1
(3) fJ  k < Nk'·
Let d = (h', k'). If d = 1 there is nothing to prove. If d > 1 write h' = hd,
k' = kd, where (h, k) = 1 and k < k' ~ N. Then 11k' < 11k and (3) becomes
Ie  ~ I < ~ k ' < ~ k '
from which we find IkfJ  hi < liN. 0
Now we restate the result in a slightly weaker form which does not involve
the integer N.
Theorem 7.3. For every real fJ there exist integers hand k with k > 0 and
(h, k) = 1 such that
PROOF. In Theorem 7.2 we have 1/(Nk) ~ 1/k
2
because k ~ N. D
Theorem 7.4. If fJ is real, let S(fJ) denote the set of all ordered pairs of integers
(h, k) with k > 0 and (h, k) = 1 such that
Ie  ~ I < :2'
Then S(fJ) has the following properties:
(a) S(fJ) is nonempty.
(b) If fJ is irrational, S(fJ) is an infinite set.
(c) When S(fJ) is infinite it contains pairs (h, k) with k arbitrarily large.
(d) If fJ is rational, S(fJ) is a finite set.
PROOF. Part (a) is merely a restatement of Theorem 7.3. To prove (b), assume
() is irrational and assume also that S(fJ) is finite. We shall obtain a contra
diction. Let
. IfJ hi
a = mIn   .
(h, k) E 5(6) k
Since fJ is irrational, a is positive. Choose any integer N > 11a, for example,
N = 1 + [l/a]. Then liN < a. Applying Theorem 7.2 with this N we obtain
a pair of integers hand k with (h, k) = 1 and 0 < k ~ N such that
I
fJ  ~ I <_1
k kN·
144
7.2: Dirichlet's approximation theorem
Now l/(kN) ~ 1/k
2
so the pair (h, k) E S(fJ). But we also have
1 1
 <  < ~
kN N '
so
contradicting the definition o f ~ . This shows that S(fJ) cannot be finite if fJ is
irrational.
To prove (c) assume that all pairs (h, k) in S(fJ) have k ~ M for some M.
We will show that this leads to a contradiction by showing that the number
of choices for h is also bounded. If (h, k) E S(fJ) we have
1
IkfJ  hi < k ~ 1,
so
Ihi = Ih  kfJ + kfJ I ~ Ih  kfJ I + IkfJ I < 1 + IkfJ I ~ 1 + MIfJ I·
Therefore the number of choices for h is bounded, contradicting the fact that
S(fJ) is infinite.
To prove (d), assume fJ is rational, say fJ = alb, where (a, b) = 1 and b > o.
Then the pair (a, b) E S(fJ) because fJ  alb = O. Now we assume that S(fJ)
is an infinite set and obtain a contradiction. If S(fJ) is infinite then by part (c)
there is a pair (h, k) in S(fJ) with k > b. For this pair we have
o< I ~ ~ I < ~
b k k
2
'
from which we find 0 < Iak  bh I < b/k < 1. This is a contradiction because
ak  bh is an integer. 0
Theorem 7.4 shows that a real number fJ is irrational if, and only if,
there are infinitely many rational numbers h/k with (h, k) = 1 and k > 0
such that
Ie  ~ I < :2'
This inequality can be improved. It is easy to show that the numerator 1
can be replaced by ! (see Exercise 7.4). Hurwitz replaced t by a smaller
constant. He proved that fJ is irrational if, and only if, there exist infinitely
many rational numbers h/k with (h, k) = 1 and k > 0 such that
Ie  ~ I < flk2 '
Moreover, the result is false if 1/.)5 is replaced by any smaller constant. (See
Exercise 7.5.) We shall not prove Hurwitz's theorem. Instead, we prove a
theorem of Liouville which shows that the denominator k
2
cannot be re
placed by k
3
or any higher power.
145
7: Kronecker's theorem with applications
7.3 Liouville's approximation theorem
Theorem 7.5. Let fJ be a real algebraic number ofdegree n ~ 2. Then there is a
positive constant C(fJ), depending only on fJ, such that for all integers hand
k with k > 0 we have
(4)
I
fJ  ~ I > C(fJ)
k k
n
'
(5)
PROOF. Since fJ is algebraic of degree n, fJ is a zero of some polynomial f(x)
of degree n with integer coefficients, say
n
f(x) = L arx
r
,
r=O
where f(x) is irreducible over the rational field. Since f(x) is irreducible it
has no rational roots so f(hlk) =1= 0 for every rational hlk.
Now we use the mean value theorem of differential calculus to write
f ( ~ ) = f ( ~ )  f(O) = f ' ( ~ ) ( ~  0)'
where ~ lies between fJ and hlk. We will deduce (4) from (5) by getting an upper
bound for I f ' ( ~ ) 1 and a lower bound for If(hlk)l. We have
(
h) n (h)r N
f k = r ~ o Q
r
k = k"
where N is a nonzero integer. Therefore
(6)
which is the required lower bound. To get an upper bound for If ' ( ~ ) I we let
If d > 1 then (4) holds with C(fJ) = 1, so we can assume that d < 1. (We
cannot have d = 1 since fJ is irrational.) Since ~ lies between fJ and hlk and
d < 1 we have I~  fJ I < 1 so
I ~ I = IfJ + ~  fJl ~ IfJl + I ~  fJl < IfJl + 1.
Hence
If ' ( ~ ) I ~ A(fJ) < 1 + A(fJ),
where A(fJ) denote the maximum value of If'(x) Iin the interval Ix I ~ IfJ I + 1.
Using this upper bound for If ' ( ~ ) I in (5) together with the lower bound in (6)
we obtain (4) with C(fJ) = 1/(1 + A(fJ)). D
146
7.3: Liouville's approximation theorem
A real number which is not algebraic is called transcendental. A simple
counting argument shows that transcendental numbers exist. In fact, the
set of all real algebraic numbers is countable, but the set of all real numbers
is uncountable, so the transcendental numbers not only exist but they form
an uncountable set.
It is usually difficult to show that some particular number such as e or n
is transcendental. Liouville's theorem can be used to show that irrational
numbers that are sufficiently well approximated by rationals are necessarily
transcendental. Such numbers are called Liouville numbers and are defined
as follows.
Definition. A real number () is called a Liouville number if for every integer
r ~ 1 there exist integers h
r
and k
r
with k
r
> 0 such that
(7)
I
hrl 1
0< () <
k
r
k/'
Theorem 7.6. Every Liouville number is transcendental.
PROOF. If a Liouville number () were algebraic of degree n it would satisfy
both inequality (7) and
for every r ~ 1, where C(()) is the constant in Theorem 7.5. Therefore
1
or 0 < C(()) < k rn'
r
The last inequality gives a contradiction if r is sufficiently large. 0
EXAMPLE. The number
00
() = L 10m!
m=1
is a Liouville number and hence is transcendental. In fact, for each r ~ 1 we
can take k
r
= lor! and
r 1
h
r
= k
r
L 10m!'
m=1
Then we have
h 00 1 1 00 1
o< ()  k
r
= L 10m! ~ 10(r+l)! L 10m
r m=r+l m=O
10/9 1 10/9 1
==<
10(r+ I)! k/ lor! k/
so (7) is satisfied.
147
7: Kronecker's theorem with applications
Note. The same argument shows that L ~ = l a
m
l0
m
! is transcendental if
am = 0 or 1 and am = 1 for infinitely many m.
We turn now to a generalization of Dirichlet's theorem due to Kronecker.
7.4 Kronecker's approximation theorem:
the onedimensional case
Dirichlet's theorem tells us that for any real fJ and every e > 0 there exist
integers x and y, not both 0, such that
IfJx + yl < e.
In other words, the linear form fJx + y can be made arbitrarily close to 0 by a
suitable choice of integers x and y. If fJ is rational this is trivial because we
can make fJx + y = 0, so the result is significant only if fJ is irrational.
Kronecker proved a much stronger result. He showed that if fJ is irrational
the linear form fJx + y can be made arbitrarily close to any prescribed real
number r:J.. We prove this result first for rx in the unit interval. As in the proof
of Dirichlet's theorem we make use of the fractional parts {nfJ} = nfJ  [nfJ].
Theorem 7.7. IffJ is a given irrational number the sequence of numbers {nfJ}
is dense in the unit interval. That is, given any r:J., °~ rx ~ 1, and given any
e > 0, there exists a positive integer k such that
r {kfJ}  rxl < e.
Hence, ifh = [kfJ] we have IkfJ  h  rxl < e.
Note. This shows that the linear form fJx + y can be made arbitrarily
close to r:J. by a suitable choice of integers x and y.
PROOF. First we note that {nfJ} =1= {mfJ} if m =1= n because fJ is irrational.
Also, there is no loss of generality if we assume 0 < fJ < 1 since nfJ =
n[fJ] + n{fJ} and {nfJ} = {n{fJ}}.
Let e > 0 be given and choose any r:J., 0 ~ rx ~ 1. By Dirichlet's approxima
tion theorem there exist integers hand k such that IkfJ  hi < e. Now
either kfJ > h or kfJ < h. Suppose that kfJ > h, so that 0 < {kfJ} < e. (The
argument is similar if kfJ < h.) Now consider the following subsequence of
the given sequence {nfJ}:
{kfJ}, {2kfJ}, {3kfJ}, ....
We will show that the early terms of this sequence are increasing. We have
kfJ = [kfJ] + {kfJ}, so mkfJ = m[kfJ] + m{kfJ}.
Hence
{mkO} = m{kO} if, and only if, {kO} < ~ .
m
148
7.5: Extension of Kronecker's theorem to simultaneous approximation
Now choose the largest integer N which satisfies {kO} < liN. Then we have
1 1
N + 1 < {kO} < N'
Therefore {mkO} = m{kO} for m = 1, 2, , N, so the N numbers
{kO}, {2kO}, , {NkO}
form an increasing equallyspaced chain running from left to right in the
interval (0, 1). The last member of this chain (by the definition of N) satisfies
the inequality
N
1 < {NkO} < 1,
N+
or
1
1  N + 1 < {NkO} < 1.
Thus {NkO} differs from 1 by less than 1/(N + 1) < {kO} < G. Therefore the
first N members of the subsequence {nkO} subdivide the unit interval into
subintervals of length < e. Since rx lies in one of these subintervals, the
theorem is proved. 0
The next theorem removes the restriction 0 :5; rx ~ 1.
Theorem 7.8. Given an)' real rx, any irrational 8.. and an)' f. > 0.. there exist
integers hand k with k > 0 such that
IkO  h  rxl < e.
PROOF. Write rx = [rx] + {rx}. By Theorem 7.7 there exists k > 0 such that
I{kO}  {rx} I < e. Hence
IkO  [kO]  (rx  [rxJ) 1< e
or
IkO  ([kO]  [rxJ)  rxl < e.
Now take h = [kO]  [rx] to complete the proof.
o
7.5 Extension of Kronecker's theorem to
simultaneous approximation
We turn now to a problem of simultaneous approximation. Given n irrational
numbers 0b O
2
, ••• , On' and n real numbers rxb rx2' ... , rx
n
, and given e > 0,
we seek integers hI' h
2
, ••• , h
n
and k such that
IkO
i
 hi  rxd < e for i = 1,2, ... , n.
149
7: Kronecker's theorem with applications
It turns out that this problem cannot always be solved as stated. For example,
suppose we start with two irrational numbers, say e
i
and 2e
b
and two real
numbers ~ I and ~ 2 , and suppose there exist integers h
b
h
2
and k such that
Ike1  hI  ~ 1 I < 8
and
12ke
i
 h
2
 ~ 2 1 < 8.
Multiply the first inequality by 2 and subtract from the second to obtain
12h
I
 h
2
+ 2 ~ 1  ~ 2 1 < 38.
Since 8, ~ I and ~ 2 are arbitrary and h
b
h
2
are integers, this inequality cannot
in general be satisfied. The difficulty with this example is that e
i
and 2e
I
are
linearly dependent and we were able to eliminate e
i
from the two inequalities.
Kronecker showed that the problem of simultaneous approximation can
always be solved if e1, ... , en are linearly independent over the integers;
that is, if
n
LCiei = 0
i= 1
with integer multipliers C1, ... , C
n
implies C1 = ... = c
n
= O. This restriction
is compensated for, in part, by removing the restriction that the e
i
be
irrational. First we prove what appears to be a less general result.
Theorem 7.9 (First form of Kronecker's theorem). If ~ I ' ••• , ~ n are arbitrary
real numbers, if e
b
... , en are linearly independent real numbers, and if
8 > 0 is arbitrary, then there exists a real number t and integers h
b
... , h
n
such that
Ite
i
 hi  ~ d < 8 for i = 1, 2, ... ,n.
Note. The theorem exhibits a real number t, whereas we asked for an
integer k. Later we show that it is possible to replace t by an integer k, but
in most applications of the theorem the real t suffices.
The proof of Theorem 7.9 makes use of three lemmas.
Lemma 1. Let {An} be a sequence of distinct real numbers. For each real t
and arbitrary complex numbers Co, ... , eN define
N
f(t) = L creitAr.
r=O
Then for each k we have
1 JT .
Ck = lim  f(t)e ltAk dt.
T+oo T 0
150
7.5: Extension of Kronecker's theorem to simultaneous approximation
PROOF. The definition of f(t) gives us
N
f(t)e  it).k = L crei().r ).k)t.
r=O
Hence
from which we find
Now let T ~ 00 to obtain the lemma.
Lemma 2. If t is real, let
D
(8)
n
F(t) = 1 + L e21ti(tlJrcxr),
r= 1
where CX l' ... , CX
n
and 0
1
, ... , On are arbitrary real numbers. Let
L= sup IF(t)l.
oo<t<+oo
Then the following two statements are equivalent:
(a) For every G > 0 there exists a real t and integers h
1
, ••• , h
n
such that
ItOr  CX
r
 h
r
I < G for r = 1, 2, ... , n.
(b) L = n + 1.
PROOF. The idea of the proof is fairly simple. Each term of the sum in (8)
has absolute value 1 so IF(t) I :::; n + 1. If (a) holds then each number tOr  CX
r
is nearly an integer hence each exponential in (8) is nearly 1 so IF(t) Iis nearly
n + 1. Conversely, if (b) holds then IF(t)1 is nearly n + 1 for some thence
every term in (8) must be nearly 1 since no term has absolute value greater
than 1. Therefore each number tOr  CX
r
is nearly an integer so (a) holds.
Now we transform this idea into a rigorous proof.
First we show that (a) implies (b). If (a) holds take e = 1/(2nk), where
k 2 1, and let t
k
be the corresponding value of t given by (a). Then
2n(t
k
Or  cx
r
) differs from an integer multiple of 2n by less than 11k so
1
cos 2n(t
k
Or  CX
r
) 2 cos k.
Hence
n 1
IF(t
k
) I 2 1 + L cos 2n(t
k
O
r
 CX
r
) 2 1 + n cos
r= 1 k
151
(9)
7: Kronecker's theorem with applications
and therefore L :2:: IF(t
k
) I :2:: 1 + n cos(l/k). Letting k + 00 we find L :2:: n + 1.
Since L ::::; n + 1 this proves (b).
Now we assume (a) is false and show that (b) is also false. If(a) is false there
exists an G > 0 such that for all integers hI' ... , h
n
and all real t there is a k,
1 ::::; k ::::; n, such that
G
ItOk  rtk  hkl :2:: 2n·
(We can also assume that G ::::; n/4 because if (a) is false for G it is also false for
every smaller G.) Let X
r
= tOr  rt
r
 hr. Then (9) implies 12nxkl :2:: Gso the
point 1 + e21tixk lies on the circle of radius 1 about 1 but outside the shaded
sector shown in Figure 7.1.
o
Figure 7.1
Now 11 + e
if
I < 2 so 11 + e
if
I = 2  b for some b > O. Hence
11 + e21tixk I ::::; 11 + e
if
I = 2  b,
so
IF(t) I = 11 +rtl e21tiXr IS 11 + e21tixk I +rt
1
Ie21tixr I
r¢k
::::; (2  b) + (n  1) = n + 1  b.
Since this is true for all t we must have L ::::; n + 1  b < n + 1, contra
dicting (b). D
Lemma 3. Let g = g(Xl' ... , x
n
) be the polynomial in n variables given by
and write
(10)
152
g
p = 1 + " a x rl ..• X r
l1
1..J rl, ... ,r
l1
1 n ,
7.5: Extension of Kronecker's theorem to sin\ultaneous approximation
where p is a positive integer. Then the coefficients a
rt
.....
r
" are positive
integers such that
(11) 1 + Lart ..... r,1 = (1 + n)P,
and the number of terms in (10) is at most (p + l)n.
PROOF. Since 1 + L a
rt
.....
r
" = gP(I, 1, ... , 1) = (1 + n)P this proves (11).
Let 1 + N be the number of terms in (10). We shall prove that
(12)
1 + N ~ (p + It
by induction on n. For n = 1 we have
(1 + xl)P = 1 + ( ~ ) x l + ( ~ ) x / + ... +x/
and the sum on the right has exactly p + 1 terms. Thus (12) holds for n = 1.
Ifn> 1 we have
gp = {(I + Xl + + x
n

l
) + xn}P
= (1 + Xl + + xnI)P + (i)(l + ... + xn_I)PI
xn
+ ... + x/,
so if there are at most (p + 1t 1 terms in each group on the right there will
be at most (p + l)n terms altogether. This proves (12) by induction. D
PROOF OF KRONECKER'S THEOREM. Choosing F(t) as in Lemma 2 we have
n
F(t) = 1 + L e21ti(tOrcxr).
r =: 1
By Lemma 2, to prove Kronecker's theorem it suffices to prove that
L= sup IF(t)l=n+l.
oo<t<+oo
The pth power of F(t) is a sum of the type discussed in Lemma 1,
(13)
N
f(t) = FP(t) = 1 + L creit).r,
r= 1
with A
o
= 1 and A
r
replaced by 2n(rl0l + ... + rnO
n
) if r ~ 1. The numbers
A
r
are distinct because the Oi are linearly independent over the integers. The
coefficients C
r
in (13) are the integers a
rt
.....
r
" of Lemma 3 multiplied by a
factor of absolute value 1. Hence (11) implies
(14)
N
1 + L Icrl = 1 + La
rt
.....
r
" = (1 + n)P.
r= 1
By Lemma 1 we have
(15)
1 f.T .
C
r
= lim  FP(t)elt).r dt.
T ~ o o T 0
153
7: Kronecker's theorem with applications
Now IF(t)1 ~ L so IFP(t) I ~ LP for all t, hence
I ~ L
T
FP(t)eit.l.
r
dtl :::;; ~ L
T
U dt = U.
Hence (15) implies lerl ~ LP for each r, and (14) gives us
(1 + n)P ~ (N + I)LP ~ (p + l)nLP
by Lemma 3. Therefore
from which we find
IOg(n 1
1
) :::;; ~ log(p + 1).
Now let p + 00. The last inequality becomes log[(n + 1)/L] ~ 0, so L 2
n + 1. But L ~ n + 1 hence L = n + 1, and this proves Kronecker's
theorem. []
The next version of Kronecker's theorem replaces the real number t by an
integer k.
Theorem 7.10 (Second form of Kronecker's theorem). If Ct
1
, ••• , Ct
n
are
arbitrary real numbers, if01, . .. ,On' 1 are linearly independent real numbers,
and if £ > 0 is given, then there exists an integer k and integers m
1
, •. , m
n
sueh that
PROOF. We apply the first form of Kronecker's theorem to the system
Ct
1
, ••• , Ct
n
, 0 and {Ol}' {02}, ... , {On}' 1, with £/2 instead of £, where £ < 1.
Then there exists a real t and integers h1, ... , h
n
+ 1 such that
£
It{OJhiCtil<2 fori=I,2, ... ,n
and
(16)
The last inequality shows that t is nearly equal to the integer h
n
+ 1. Take
k = h
n
+
1
• Then (16) implies
Ik{OJ  hi  Ctd = It{OJ  hi  Ct
i
+ (k  t){OJI
~ It{OJ  hi  Ctd + Ik  t I < £.
154
7.6: Applications to the Riemann zeta function
Hence, writing {Oi} = 0i  [Oa, we obtain
Ik(Oi  [Oa)  hi  ad < G
or, what is the same thing,
IkO
i
 (hi + k[Oa)  (Xii < G.
Putting mi = hi + k[Oa we obtain the theorem. o
7.6 Applications to the Riemann zeta
function
With the help of Kronecker's theorem we can determine the least upper
bound and greatest lower bound of I'((J + it) Ion any fixed line (J = constant,
(J > 1.
Definition. For fixed (J, we define
m((J) = infl '((J + it) I
t
and M((J) = sup I'((J + it) I,
t
where the infimum and supremum are taken over all real t.
Theorem 7.11. For eachfixed (J > 1 we have
and
'(2(J)
m(O") = (0")'
PROOF. For (J > 1 we have I'((J + it)l ~ '((J) so M((J) = '((J), the supremum
being attained on the real axis. To obtain the result for m((J) we estimate the
reciprocal 11/((s) I. For (J > 1we have
(
17) 1_11 nil  si < n(1 a)  '((J)
(s)  p p  p + P  (20")'
Hence I'(s) I ~ '(2(J)/'((J) so m((J) 2 '(2(J)/'((J).
Now we wish to prove the reverse inequality m((J) ~ '(2(J)/'((J). The idea
is to show that the inequality
11  psi ~ 1 + pa
used in (17) is very nearly an equality for certain values of t. Now
1  ps = 1  pait = 1  paeitlo
gp
= 1 + pa
e
i(tlo
gp
1t),
so we need to show that  t log p  'It is nearly an even multiple of 2n for
certain values of t. For this we invoke Kronecker's theorem. Of course,
there are infinitely many terms in the Euler product for 1/'(s) and we cannot
expect to make  t log p  n nearly an even multiple of 2n for all primes p.
But we will be able to do this for enough primes to obtain the desired
inequality.
155
7: Kronecker's theorem with applications
Choose any c, 0 < c < nf2, and choose any integer n ~ 1. We apply
Kronecker's theorem to the numbers
1
Ok = 2;" log Pk' k = 1,2, ... , n,
where Pl' ... , Pn are the first n primes. The Oi are linearly independent
because
n
L ai log Pi = 0 implies log(p1at ... Pnan) = 0
i= 1
so Plat ... Pn
an
= 1 hence each ai = O. We also take lJ.
l
= lJ.2 = ... = lJ.
n
= 1.
Then by Theorem 7.9 there is a real t and integers h
l
, ... , h
n
such that
ItO
k
 lJ.k  h
k
I < cf(2n), which means
(18) It log Pk  n  2nh
k
l < c.
For this t we have
1  Pk s = 1  Pk O'eitlogpk = 1 + Pk O'e
i
(tlo
gpk
1t)
= 1 + PkO'cos(tlogpk  n) + ipkO'sin(tlogpk  n),
so
11  Pk  S I ~ 1 + Pk  0' cos(  t log Pk  n).
But (18) implies
coslt log Pk  nl = coslt log Pk  n  2nh
k
l > cos c,
since the cosine function decreases in the interval [0, nf2]. Hence
11  Pk  S I > 1 + Pk  0' cos c.
Now consider any partial product of the Euler product for If'(s). For
a given c and n there exists a real t (depending on c and on n) such that
(19) Il
l
1 (1  Pk ')1 = kD1 11  Pk 'I > kD1 (1 + Pk a cos e).
Now
and hence, by the Cauchy condition for convergent products, there is an
no such that n ~ no implies
I
fI 11  Pk 'I  11 < c
k=n+l
or
00
1  c < n 11  Pk si < 1 + c.
k=n+l
156
7.7: Applications to periodic functions
Using (19) with n ~ no we have
1 noon
Ws)l = }]1,1  Pk
s
'JI
I
11  Pksi > (1  e) }]I (1 + Pk
a
cos e),
This holds for n ~ no and a certain t depending on n and on c. Hence
1 lin
() = , f 1(( . )1 = sup 1(( . )1 ~ (l  e) n(1 + Pk a cos e).
m (J In t (J + It t (J + It k =1
Letting n + 00 we find
1 00
() ~ (1  c) n (1 + Pk (1 cos c).
m (J k= 1
We will show in a moment that the last product converges uniformly for
o:::;; c :::;; n/2. Therefore we can let c + 0 and pass to the limit term by term
to obtain
1 00 (1 '((J)
m(O") ~ }]I (1 + Pk ) = ((20")'
This gives the desired inequality m((J) :::;; '(2(J)/'((J).
To prove the uniform convergence of the product, we use the fact that
a product n(1 + fn(z)) converges uniformly on a set if, and only if, the
series L fn(z) converges uniformly on this set. Therefore we consider the
series LPk (1 cos c. But this is dominated by LPk (1 :::;; Ln (1 = '((J) so the
convergence is uniform in the interval 0 :::;; c :::;; n/2, and the proof is complete.
D
7.7 Applications to periodic functions
We say that n complex numbers Wl' W
2
, ••• , W
n
are linearly independent
over the integers if no linear combination
alw
l
+ a2 W
2
+ ... + an(Qn
with integers coefficients is 0 except when a
l
= a
2
= ... = an = O. Other
wise the numbers Wl' .. , W
n
are called linearly dependent over the integers.
Elliptic functions are meromorphic functions with two linearly indepen
dent periods. In this section we use Kronecker's theorem to show that there
are no meromorphic functions with three linearly independent periods
except for constant functions.
Theorem 7.12. Let W
l
and W
2
be periods off such that the ratio W2/Wl is
real and irrational. Then f has arbitrarily small nonzero periods. That is,
given c > 0 there is a period Wsuch that 0 < IWI < c.
PROOF. We apply Dirichlet's approximation theorem. Let () = W
2
/W
l
'
Since () is irrational, given any c > 0 there exist integers hand k with k > 0
such that
c
1kO  hi < 1WI I'
157
7: Kronecker's theorem with applications
Multiplying by I w
1
1 we find
IkW2  hw1 I < c.
But w = kW
2
 hW
1
is a period of f with Iwl < c. Also, w =I 0 since w21w
1
is irrational. 0
Theorem 7.13. Iff has three periods W
1
, W2, W
3
which are linearly independent
over the integers, thenf has arbitrarily small nonzero periods.
PROOF. Suppose first that w21w
1
is real. If w
2
1w
1
is rational then W
1
and
W2 are linearly dependent over the integers, hence W
b
W
2
, W3 are also depen
dent, contradicting the hypothesis. IfW2 IW
I
is irrational, thenfhas arbitrarily
small nonzero periods by Theorem 7.12.
Now suppose w21w
1
is not real. Geometrically, this means that W
1
and
W2 are not collinear with the origin. Hence W3 can be expressed as a linear
combination of W
1
and W2 with real coefficients, say
W
3
= + pw
2
, where and pare real.
Now we consider three cases:
(a) Both and prational.
(b) One of prational, the other irrational.
(c) Both and pirrational.
Case (a) implies w
1
, W
2
, W3 are dependent over the integers, contradicting
the hypothesis.
For case (b), assume is rational, say = alb, and {3 is irrational. Then
we have
so
This gives us two periods bW
3
 aWl and bW2 with irrational ratio, hence f
has arbitrarily small periods. The same argument works, of course, if p is
rational and is irrational.
Now consider case (c), both and p irrational. Here we consider two
subcases.
(c
l
) Assume and p are linearly dependent over the integers. Then
there exist integers a and b, not both zero, such that + bP = O. By sym
metry, we can assume that b =I O. Then p= and
a
W
3
= 
so bW
3
=  awz)·
Again we have two periods bW3 and bW
I
 aW2 with irrational ratio, so f
has arbitrarily small nonzero periods.
(cz) Assume and pare linearly independent over the integers. Then by
Kronecker's theorem, given any c > 0 there exist integers h
b
h
2
and k
158
Exercises for Chapter 7
such that
G
Ikf3  h
2
1 < 1 I I I "
+ W
l
+ W
2
Multiply these inequalities by Iwll, Iw
2
1, respectively, to get
Glw
l
'
IklXW1  hI WI I < 1 I I I I'
+ W
l
+ W
2
Since kW
3
= krJ.w
l
+ kf3W2 we find, by the triangle inequality,
G(lwll + Iw
2
1)
Ikw3hlWlh2W21<1 I I I ,<G.
+ Wt + W2
Thus kW3  hlw
l
 h2w2 is a nonzero period with modulus <G. D
Note. In Chapter 1 we showed that a function with arbitrarily small
nonzero periods is constant on every open connected subset in which it is
analytic. Therefore, by Theorem 7.13, the only meromorphic functions
with three independent periods are constant functions.
Further applications of Kronecker's theorem are given in the next chapter.
Exercises for Chapter 7
1. Prove the following extension of Dirichlet's approximation theorem.
Given n real numbers 8
1
, ... , 8
n
and given an integer N ~ 1, there exist integers
h
l
, ... , h
n
and k, with 1 ~ k ~ N
n
, such that
1
Ik8
i
 hi I < N for i = 1, 2, ... , n.
2. (a) Given n real numbers 8
1
" .• ,8
n
, prove that there exist integers h
l
, ... , h
n
and
k > 0 such that
I
h'l 1
8
i
 i < k
l
+ l/n for i = 1,2, ... , n.
(b) If at least one of the 8
i
is irrational, prove that there is an infinite set of ntuples
(hl/k, ... , hn/k) satisfying the inequalities in (a).
3. This exercise gives another extension of Dirichlet's approximation theorem. Given
m linear forms,
i = 1,2, ... , m,
in n + m variables Xl' ... , X
n
, Yl, ... ,Ym, prove that for each integer N > 1 there
exists integers Xl' ... , X
n
, Yl" .. , Ym such that
1
IL
i
I < N for i = 1, 2, ... , m
159
7: Kronecker's theorem with applications
and 0 < max{ lXII, ... , Ix
n
!} ~ N
m
/
n
• Hint: Let M
j
= aj1Xl + ... + ajnXn and
examine the points ({MIL ... , {M
m
}) in the unit cube in mspace, where {M
j
} =
M
j
 [M
j
].
4. Let {} be irrational, 0 < () < 1. Then () lies between two consecutive Farey fractions,
say
~ < () < ~ .
b d
(a) Prove that either 0  a/b < 1/(2b
2
) or c/d  () < 1/(2d
2
).
(b) Deduce that there exist infinitely many fractions h/k with (h, k) = 1 and k > 0
such that
1
( ) _ ~ 1 < _ 1 .
k 2k
2
5. Let a = (1 + )5)/2. This exercise shows that the inequality
(20)
has only a finite number of solutions in integers hand k with k > 0 if 0 < c < 1/)5.
(a) Let f3 = C/..  )5 so that a and f3 are roots of the equation x
2
 xI = O.
Show that for any integers hand k with k > 0 we have
and deduce that
(b) If (20) has infinitely many solutions h/k with k > 0, say hl/k l' h2/k  , ... , show that
k
n
+ X'; as n + x and use part (a) to prove that c :2: 1/)5.
6. In Lemma 2, define
L = lim sup IF(t) I instead of L = sup IF(t) I.
t ~ + oc oo<t<oo
Prove that the equation L = n + 1 is equivalent to the following statement: For
every G > 0 and every T > 0 there exists a real t > T and integers hi' ... , h
n
such that
It()i  hi  ad < G for every i = 1, 2, ... , n.
7. Prove that the multiplier t in the first form of Kronecker's theorem can be taken
positive and arbitrarily large. That is, under the hypotheses of Theorem 7.9, if T > 0
is given there exists a real t > T satisfying the n inequalities It()i  hi  ad < G.
Show also that the integer multiplier k in the second form of Kronecker's theorem
can be taken positive and arbitrarily large.
160
General Dirichlet series and
Bohr's equivalence theorem
8.1 Introduction
This chapter treats a class of series, called general Dirichlet series, which
includes both power series and ordinary Dirichlet series as special cases.
Most of the chapter is devoted to a method developed by Harald Bohr [6]
in 1919 for studying the set of values taken by Dirichlet series in a halfplane.
Bohr introduced an equivalence relation among Dirichlet series and showed
that equivalent Dirichlet series take the same set of values in certain half
planes. The theory uses Kronecker's approximation theorem discussed in
the previous chapter. At the end of the chapter applications are given to the
Riemann zeta function and to Dirichlet Lfunctions.
8.2 The halfplane of convergence of general
Dirichlet series
Definition. Let {A(n)} be a strictly increasing sequence of real numbers such
that A(n) ~ +00 as n ~ 00. A series of the form
00
L a(n)eSA(n)
n=l
is called a general Dirichlet series. The numbers A(n) are called the
exponents of the series, and the numbers a(n) are called its coefficients.
As usual, we write s = (J + it where (J and t are real.
Note. When A(n) = log n then eSA(n) = n
S
and we obtain the ordinary
Dirichlet series La(n)n
s
• When A(n) = n the series becomes a power series
in x, where x = e
S
•
161
8: General Dirichlet series and Bohr's equivalence theorem
A general Dirichlet series is analogous to the Laplace transform of a
function, Sg:: f(t)e
st
dt. As a matter of fact, both Dirichlet series and Laplace
transforms are special cases of the LaplaceStieltjes transform, e st drx(t).
When rx(t) has a continuous derivative rx'(t) = f(t) this gives the Laplace
transform of f When rx is a step function with jump a(n) at the point A(n)
the integral becomes the general Dirichlet series L a(n)eSA(n). Much of
what we do here can be extended to LaplaceStieltjes transforms, but we
shall not deal with these generalizations.
As is the case with ordinary Dirichlet series, each general Dirichlet
series has associated with it an abscissa (Jc of convergence and an abscissa
(Ja of absolute convergence. We could argue as in Chapter 11 of [4] to
prove the existence of (Jc and (Ja. Instead we give a different method of proof
which also expresses (Jc and (Ja in terms of the exponents A(n) and the
coefficients a(n).
Theorem 8.1. Assume that the series L a(n)eSA(n) converges for some s with
positive real part, say for s = So with (Jo > O. Let
L
I.
 1m sup '( ) .
noo An
Then L ::; (J 0 • Moreover, the series converges in the halfplane (J > L, and
the convergence is uniform on every compact subset of the halfplane
(J > L.
PROOF. First we prove that L ::; (Jo. Let A(n) denote the partial sums of the
coefficients,
n
A(n) = L a(k).
k=l
Note that A(n) > 0 for all sufficiently large n. If we prove that for every
e > 0 we have
(1) logIA(n)1 < ((Jo + e)A(n)
for all sufficiently large n, then it follows that
logIA(n)1
A(n) < (Jo + e
for these n, so L ::; (Jo + e, hence L ::; (Jo. Now relation (1) is equivalent to the
inequality
(2)
To prove (2) we introduce the partial sums
n
S(n) = L a(k)eSOA(k).
k=l
162
8.2: The halfplane of convergence of general Dirichlet series
The S(n) are bounded since the series Lk= 1 a(k)eSOA(k) converges. Suppose
that IS(n) I < M for all n. To express A(n) in terms of the S(n) we use partial
summation:
n n
A(n) = L a(k) = L a(k)eSOA(k)eSOA(k)
k=1 k=1
n
= L {S(k)  S(k  1)}e
SoA
(k),
k=1
provided 8(0) = O. Thus
n nl
A(n) = L S(k)eSOA(k)  L S(k)eSOA(k+ 1)
k=1 k=1
nl
= L S(k) {esoA(k)  esoA(k+ 1)} + S(n)eSOA(n).
k=1
Hence
nl
IA(n)1 < M L leSOA(k)  e
SOA
(k+1)1 + Me(]OA(n).
k=1
But
nl n
1
1 jA(k+1) I nl jA(k+1)
L IesoA(k)  esoA(k+ 1) I = L So e
Sou
du ISo I L e(]OU du
k=1 k= 1 A(k) k= 1 A(k)
= ISol jA(nlel1OUdU = _ eI1OA(1l) <
A(I) 0'0 0'0
Thus
Now A(n) 00 as n 00 so
eeA(nl > M(l + 1;:1)
if n is sufficiently large. Hence for these n we have IA(n) I < e(UO+E:)A(n), which
proves (2) and hence (1). This proves that L 0'0.
Now we prove that the series converges for all S with (J > L. Consider
any section of the series La(n)eSA(n), say We shall use the Cauchy
convergence criterion to show that this section can be made small when
a and b are sufficiently large. We estimate the size of such a section by using
163
8: General Dirichlet series and Bohr's equivalence theorem
partial summation to compare it to the partial sums A(n) = 1 a(k). We
have
b b
L a(n)eSA(n) = L {A(n)  A(n  l)}e
SA
(n)
n=a n=a
b
= L A(n) {eSA(n)  eSA(n+ I)} + A(b)eSA(b+ 1)
n=a
 A(a  l)e
SA
(a).
This relation holds for any choice of s, a and b. Now suppose s is any complex
number with (J > L. Let e = !«(J  L). Then e > 0 and (J = L + 2e. By the
definition of L, for this e there is an integer N(e) such that for all n N(e)
we have
logl A(n) I
A(n) < L + e.
We can also assume that A(n) > 0 for n N(e). Hence
IA(n) I < e(L+f)A(n) for all n N(e).
If we choose b a > N(e) we get the estimate
The last two terms are e
fA
(b+1) + efA(a) since L + e  (J = c. Now we
estimate the sum by writing
I
f
A(n+ 1) I fA(n+ 1)
leSA(n)  eSA(n+ 1)1 = s e
su
du lsi e
au
du
A(n) A(n)
so
b b fA(n+ 1)
L e(L+f)A(n) IeSA(n)  eSA(n+ 1) I Is I L e(L+E)A(n) e
au
du
n=a n=a A(n)
b fA(n+ 1) b fA(n+ 1)
lsi L eaUe(L+f)U du = lsi L e
fU
du
n=a A(n) n=a A(n)
f
A(b+ 1) I I
= lsi efudu =  e
fA
(b+l»).
A(a) e
Thus we have
I
t a(n)eSA(n) I ::s; (e·A(a)  e'A(b+ 0) + e'A(b+ 1) + e·A(a).
n=a e
164
8.2: The halfplane of convergence of general Dirichlet series
Each term on the right tends to 0 as a + 00, so the Cauchy criterion shows
that the series converges for all s with (J > L. This completes the proof.
Note also that this proves uniform convergence on any compact subset of
the halfplane (J > L. D
Theorem 8.2. Assume the series La(n)eSA(n) convergesfor some s with (J > 0
but diverges for all s with (J < o. Then the number
_ 1· log IL ~ = 1 a(k) I
L  1m sup '( )
noo A n
is the abscissa of convergence of the series. In other words, the series
converges for all s with (J > L and diverges for all s with (J < L.
PROOF. We know from Theorem 8.1 that the series converges for all s with
(J > L and that L cannot be negative. Let S be the set of all (J > 0 such that
the series converges for some s with real part (J. The set S is nonempty and
bounded below. Let (Jc be the greatest lower bound of S. Then (Jc > 0, Each
(J in S satisfies L :::; (J hence L :::; (Jc. If we had (Jc > L there would be a (J in
the interval L < (J < (Jc. For this (J we would also have convergence for all
s with real part (J (by Theorem 8.1) contradicting the definition of (Jc. Hence
(Jc = L. But the definition of (Jc shows that the series diverges for all s with
o:::; (J < L. By hypothesis it also diverges for all s with (J < O. Hence it
diverges for all s with (J < L. This completes the proof. D
As a corollary we have:
Theorem 8.3. Assume the series L a(n)eSA(n) converges absolutely for some s
with (J > 0 but diverges for all s with (J < o. Then the number
_ 1· log L ~ = 1 Ia(k) I
(Ja  1m SUp '( )
noo An
is the abscissa of absolute convergence of the series.
PROOF. Let A be the abscissa of convergence of the series L la(n)leSA(n).
Then, by Theorem 8.2,
_ 1· log L ~ = 1 Ia(k) I
A  1m sup '( ) .
noo A n
We wish to prove that L la(n)leaA(n) converges if (J > A and diverges if
(J < A. Clearly if (J > A then the point s = (J is within the halfplane of
convergence of L Ia(n) IeSA(n) so L Ia(n) IeGA(n) converges.
Now suppose L la(n)leGA(n) converges for some (J < A. Then the series
L Ia(n) IeSA(n) converges absolutely for each s with real part (J so, in particular
it converges for all these s, contradicting the fact that A is the abscissa of
convergence of L Ia(n) IeSA(n) . D
165
8: General Dirichlet series and Bohr's equivalence theorem
8.3 Bases for the sequence of exponents of a
Dirichlet series
The rest of this chapter is devoted to a detailed study of Harald Bohr's
theory with applications to the Riemann zetafunction and Dirichlet's
Lseries. The first notion we need is that of a basis for the sequence of
exponents of a Dirichlet series.
Definition. Let A = {)w(n)} be an infinite sequence of distinct real numbers. By
a basis of the set A we shall mean a finite or countably infinite sequence
B = {f3(n)} of real numbers satisfying the following three conditions:
(a) The sequence B is linearly independent over the rationals. That is,
for all m ~ 1, if
with rational multipliers rk' then each rk = O.
(b) Each A(n) is expressible as a finite linear combination of terms of B,
say
q(n)
A(n) = L r
n
,kf3(k)
k=1
where the rn, k are rational and the number of summands q(n) depends
on n. (By condition (a), if A(n) i= 0 this representation is unique.)
(c) Each f3(n) is expressible as a finite linear combination of terms of A,
say
m(n)
f3(n) = L tn,kA(k)
k=1
where the tn,k are rational and m(n) depends on n.
EXAMPLE 1. Let A be the set of all rational numbers. Then B = {I} is a basis.
EXAMPLE 2. Let A = {log n}. Then B = {log Pn} is a basis, where Pn is the
nth prime. It is easy to verify properties (a), (b) and (c). For independence
we note that
q
Lrk log Pk = 0
k=l
implies rl = ... = r
q
= O.
To express each A(n) in terms of the basis elements we factor n and compute
log n as a linear combination of the logarithms of its prime factors. Property
(c) is trivially satisfied since B is a subsequence of A.
166
8.4: Bohr matrices
Theorem 8.4. Every sequence A has a subsequence which is a basis for A.
PROOF. Construct a basis as follows. For the first basis element take ,1(nl),
the first nonzero A (either ,1(1) or ,1(2)), and call this f3(I). Now delete the
remaining elements of A that are rational multiples of f3(I). If this exhausts
all of A take B = {f3( I)}. If not, let ,1(nz) denote the first remaining A, take
f3(2) = ,1(nz), and strike out the remaining elements of A which are rational
linear combinations of f3(I) and f3(2). Continue in this fashion to obtain a
sequence B = (f3(I), f3(2), ...) = (,1(nl)' ,1(nz), ...). It is easy to verify that B
is a basis for A. Property (a) holds by construction, since each f3 was chosen
to be independent of the earlier elements. To verify (b) we note that every Ais
either an element of B or a rational linear combination of a finite number of
elements of B. Finally, (c) holds trivially since B is a subsequence of A. D
Note. Every sequence A has infinitely many bases.
8.4 Bohr matrices
It is convenient to express these concepts in matrix notation. We display the
sequences A and B as column matrices, using an infinite column matrix for A
and a finite or infinite column matrix for B, according as B is a finite or
infinite sequence.
We also consider finite or infinite square matrices R = (rij) with rational
entries. If R is infinite we require that all but a finite number of entries in each
row be zero. Such rational square matrices will be called Bohr matrices.
We define matrix addition and multiplication of two infinite Bohr
matrices as for finite matrices. Note that a sum or product of two Bohr
matrices is another Bohr matrix. Also, the product RB of a Bohr matrix R
with an infinite column matrix B is another infinite column matrix r.
Moreover, we have the associative property (R
1
R
z
)B = R1(RzB) if R
1
and R
z
are Bohr matrices and B is an infinite column matrix.
In matrix notation, the definition of basis takes the following form. B is
called a basis for A if it satisfies the following three conditions:
(a) If RB = 0 for some Bohr matrix R, then R = o.
(b) There exists a Bohr matrix R such that A = RB.
(c) There exists a Bohr matrix T such that B = TA.
The relation between two bases Band r of the same sequence A can be
expressed as follows:
Theorem 8.5. If A has two bases Band r, then there exists a Bohr matrix
A such that r = AB.
PROOF. There exist Bohr matrices Rand T such that r = TA and A = RB.
Hence r = T(RB) = (TR)B = AB where A = TR. D
167
8: General Dirichlet series and Bohr's equivalence theorem
Theorem 8.6. Let Band r be two bases for A, and write r = AB, A = RBB,
A = Rrr, where A, R
B
, R
r
a r ~ Bohr matrices. Then R
B
= RrA.
Note. If we write AlB for R
B
, Air for R
r
and riB for A, this last equation
states that
PROOF. We have A = RBB and A = Rrr = RrAB. Hence RBB = RrAB,
so (R
B
 RrA)B = o. Since R
B
 RrA is a Bohr matrix and B is a basis, we
must have R
B
 RrA = O. D
8.5 The Bohr function associated with a
Dirichlet series
To every Dirichlet series f(s) = L ~ = 1 a(n)eSA(n) we associate a function
F(Zb Z2' ...) of countably many complex variables Zl' Z2, ... as follows.
Let Z denote the column matrix with entries Zb Z 2, .... Let B = {p(n)}
be a basis for the sequence A = {A(n)} of exponents, and write A = RB,
where R is a Bohr matrix.
Definition. The Bohr function F(Z) = F(z b Z2' ...) associated with f(s),
relative to the basis B, is the series
00
F(Z) = L a(n)e(RZ)n,
n=l
where (RZ)n denotes the nth entry of the column matrix RZ.
In other words, if
q(n)
A(n) = L rn,kp(k)
k=l
then
00
F(Zl' Z2' ...) = L a(n)e(r
n
,l::l +ooo+rn,q(n)=q(n».
n=l
Note that the formal substitution Zm = spm gives Z = sB, RZ = sRB =
sA, so (RZ)n = sA(n) and hence
00
F(sB) = L a(n)eSA(n) = f(s).
n=l
In other words, the Dirichlet series f(s) arises from F(Z) by a special choice
of the variables Zb Z 2, .... Therefore, if the Dirichlet series f(s) converges
for s = (J + it the associated Bohr series F(Z) also converges when Z = sB.
168
8.5: The Bohr function associated with a Dirichlet series
Moreover, if the Dirichlet series f(s) converges absolutely for s = a + it
then the Bohr series F(Z) converges absolutely for any choice of Z b Z2' ...
with Re Zn = af3(n) for all n. To see this we note that if Re Zn = af3(n) then
Re Z = aB so
00 00 00
L la(n)e(RZ)nl = L la(n)lea(RB)n = L la(n)lea).(n).
n=l n=l n=l
To emphasize the dependence of the Bohr function on the basis B we
sometimes write A = RBB and
00
FB(Z) = L a(n)e(RBZ)n.
n=l
Bohr functions FBand Fr corresponding to different bases are related by
the following theorem.
Theorem 8.7. Let Band r be two bases for A and write r = AB for some
Bohr matrix A. Then
PROOF. By Theorem 8.6 we have
A = RBB = Rrr, where R
B
= RrA.
Hence
00 00
FB(Z) = L a(n)e(RBZ)J1 = L a(n)exp{ (RrAZ)n} = Fr(AZ). D
n=l n=l
Definition. Assume the Dirichlet series f(s) = 1 a(n)es).(n) converges
absolutely for some s = a + it. We define Uf(a; B) to be the set of values
taken on by the associated Bohr function, relative to the basis B, when
Re Z = aB. Thus,
Uf(a; B) = {F(Z): Re Z = aB}.
The next theorem shows that this set is independent of the basis B.
Theorem 8.8. rrBand r are two basesfor A then Uf( a; B) = Uf(a; r).
PROOF. Choose any value FB(Z) in Uf(a; B), so that Re Z = aB. By Theorem
8.7 we have FB(Z) = Fr(AZ), where r = AB. But
ReAZ = ARe Z = AaB = aAB = ar
so FB(Z) E Uf(a; r). This proves Uf(a; B) Uf(a; r), and a similar argument
Uf(a;B). D
169
8: General Dirichlet series and Bohr's equivalence theorem
Note. Since U/(0'; B) is independent of the basis B we designate the set
U/(0'; B) simply by U/(0').
8.6 The set of values taken by a Dirichlet
series f(s) on a line (J = (J 0
This section relates the set U/(0'0) with the set of values taken by the Dirichlet
series f(s) on the line a = 0'0.
Definition. If the Dirichlet series f(s) = L:=1 a(n)e  sA(n) converges absolutely
for a = 0'0 we let
V/(ao) = {f(ao + it): 00 < t < +oo}
denote the set of values taken by f(s) on the line a = 0'0.
Since f(s) can be obtained from its Bohr function F(Z) by putting Z = aB,
it follows that V/(ao) ~ U/(0'0). Now we prove an inclusion relation in the
other direction. r
Theorem 8.9. Assume 0'0 > aa' where aa is the abscissa ofabsolute convergence
of a Dirichlet series f(s). Then the closure of V/(ao) contains U/(0'0). That
is, we have
V/(a
o
) ~ U/(0'0) ~ V/(a
o
), and hence U/(0'0) = V/(a
o
).
PROOF. The closure V/(a
o
) is the set of adherent points of V/(ao). We are
to prove that every point u in U/(0'0) is an adherent point of V/(a
o
). In
other words, given u in U/(0'0) and given e > 0 we will prove that there exists
a v in V/(a
o
) such that lu  vi < e. Since v = f(ao + it) for some t, we are
to prove that there exists a real t such that
If(ao + it)  ul < e.
Since u E U/(0'0) we have u = F(Zl' Z2' ...) where Zn = a
o
f3(n) + iYn. Hence
Z = 0'0 B + i1': RZ = 0'0 RB + iRY = 0'0 A + iR Y,
so
say. Therefore
00
u = L a(n)eaoA(n)eiJln.
n=l
On the other hand, we have
00
f(ao + it) = L a(n)eaoA(n)eitA(n),
n=l
170
8.6: The set of values taken by a Dirichlet series I(s) on a line (J = (Jo
hence
00
f((1o + it)  u = I a(n)eUQA,(n)(eitA,(n)  e
iJln
).
n=1
The idea of the proof from here on is as follows: First we split the sum into
two parts, I ~ =1 + I:=N + l' We choose N so the second part I:=N + 1 is
small, say its absolute value is <!8. This is possible by absolute convergence.
Then we show that the first part can be made small by choosing t properly.
The idea is to choose t to make every exponential e  itA,(n) very close to e  iJln
simultaneously for every n = 1,2., ... ., N. Then each factor eiti.(n)  e
iJln
will be small, and since there are only N terms., the whole sum will be small.
Now we discuss the details. For the given 8, choose N so that
I
f a(n)eao;'(n)(eit;'(n)  e
illn
) I < ~ .
n=N+1 2
Then we have
This holds for any choice of t. We wish to choose t to make the first sum
<!8. Since IeitA,(n) I = 1 we can rewrite the sum in question as follows:
IntIa(n)eao;'(n)(e it;'(n)  e i
ll n
) I= IntIe it;'(n)a(n)eao;'(n)(l  ei(t;'/JlIln») I
N
~ I Ia(n) IeUQA,(n) lei(tA,(n)Jln)  11.
n=1
Let M = 1 + I ~ = 1 /a(n)/eUQA,(n). For the given 8 there is a <5 > 0 such
that
(3)
/e
ix
_ 11 < _8_ iflxl < <5.
2M
Suppose we could choose a real t and integers k
1
, ••• , k
N
such that
(4)
tA(n)  Jln = 2nk
n
+ X
n
where IX
n
I < <5 for n = 1, 2, ... , N. Then for this t we would have
By (3), this would give us
l
ei(tA,(n)Jln) _ 11 < _8_
2M'
171
8: General Dirichlet series and Bohr's equivalence theorem
and hence
N . 8 N 8
L la(n)leUoA(n)lel(tA(n)Jln)  11 <  L la(n)leUoA(n) <.
n= I 2M n= I 2
Thus, the proof will be complete if we can find t and integers k
l
, ... , k
N
to
satisfy (4). If the A(n) were linearly independent over the integers we could
apply Kronecker's theorem to A(I), ... A(N) and obtain (4). However the
A(n) are not necessarily independent so instead we apply Kronecker's
theorem to the following system:
f) b f) 2 , . . . , f)Q' a b a2, . . . , aQ'
where
f) = {3(n)
n 2nD'
Yn
an = 2nD·
The {3(n) are the elements of the basis B used to define F(Z), and the Yn are
the imaginary parts of the numbers Zn which determine u. The integers Q and
D are determined as follows. We express A in terms of B by writing
A(n) = r
n
, 1{3(I) + + r
n
,q(n){3(q(n)).
Then Q is the largest of the integers q(I), , q(N), and D is the least common
multiple of the denominators of the rational numbers ri, j that arise from the
A(n) appearing in the sum. There are at most q(l) + ... + q(N) such numbers
ri, j. The numbers f)n are linearly independent over the integers because B
is a basis.
By Kronecker's theorem a real t and integers hI, ... , h
Q
exist such that
b
Itf)k  ak  h
k
I < 2nDA '
where
N q(n)
A = L L Irn,jl.
n=1 j=l
For this t we have 12nDtf)k  2nDa
k
 2nDh
k
l < bfA, or
b
It{3(k)  Yk  2nDh
k
l < A·
Therefore t{3(k)  Yk = 2nDh
k
+ b
k
, where Ibkl < b/A. Now we can write
q(n) q(n)
tA(n)  Jln = t L rn, j{3U)  L r
ll
, jYj
j= I j= I
q(n) q(n)
= I r
n
,i
t
{3(j)  y) = L rn,i2nDhj + b)
j= I j= I
q(n) q(n)
= 2n L hjDrn,j + L bjrn,j
j= I j= 1
172
8.7: Equivalence of general Dirichlet series
where k
n
is an integer and Ixnl < (bjA) L J ~ ) 1 Irn,jl < b. But this means we
have found a real t and integers kJ, ... , k
N
to satisfy (4), so the proof is
complete. D
8.7 Equivalence of general Dirichlet series
Consider two general Dirichlet series with the same sequence of exponents
A, say
00
L a(n)e  s).(n)
n=1
and
00
L b(n)es).(n).
n=1
Let B = {f3(n)} be a basis for A and write A = RB, where R is a Bohr matrix.
Definition. We say the two series are equivalent, relative to the basis B, and
we write
W 00
L a(n)es).(n) "" L b(n)es).(n)
n=1 n=1
if there exists a finite or infinite sequence of real numbers Y = {Yn} such
that
b(n) = a(n)e
ixn
where X = {x
n
} = R Y.
In other words, if we write
q(n)
).(n) = L r
n
,kf3(k),
k=1
equivalence means that for some sequence {Yn} we have
(
q(n) )
b(n) = a(n) exp ik'f/n,kYk '
Theorem 8.10. Two equivalent Dirichlet series have the same abscissa of
absolute convergence. Moreover, the relation "" just defned is independent
of the basis B.
PROOF. Equivalence implies Ib(n) I = la(n)1 so the series have the same
abscissa of absolute convergence.
Now let Band r be two bases for A, and assume that two series are equiv
alent with respect to B. We will show that they are also equivalent with
respect to r.
Write A = RBB. Then there is a sequence Y = {Yn} such that b(n) =
a(n)e
ixn
, where X = {x
n
} = R
B
Y. Now write A = Rrr. If we show that for
some sequence V = {v
n
} we have X = R
r
V then the two series will be
equivalent relative to r. The sequence
V = AY
173
8: General Dirichlet series and Bohr's equivalence theorem
has this property, where A is the Bohr matrix such that r = AB. In fact,
we have R
r
V = RrA Y = R
B
Y = X, since RrA = R
B
. This completes the
proof. D
Theorem 8.11. The relation defined in the foregoing definition is an equiv
alence relation. That is, it is reflective, symmetric, and transitive.
PROOF. Every series is equivalent to itself since we may take each Yn = O.
The corresponding X
n
will then be zero.
If b(n) = a(n)e
ixn
then a(n) = b(n)e
ixn
. Since X = R
B
Y we have X =
R
B
(  Y) so the relation is symmetric.
To prove transitivity we may use the same basis throughout and assume
that b(n) = a(n)e
ixn
, where X = R
B
Y for some 1': and that a(n) = c(n)e
iun
,
where U = R
B
V for some V Then b(n) = c(n)ei(Xn+u
n
) where
X + U = RB Y + RB V = RB( Y + V).
This completes the proof.
8.8 Equivalence of ordinary Dirichlet series
Theorem 8.12. Two ordinary Dirichlet series
D
f
n= 1 n
and
f
n= 1 n
are equivalent if, and only there exists a completely multiplicativefunction
f such that
(a) b(n) = a(n)f(n) for all n 1, and
(b) I f(P)1 = 1 whenever a(n) i= °and p is a prime divisor ofn.
PROOF. For ordinary Dirichlet series the sequence of exponents A = {A(n)}
is {log n} and for a basis we may use the sequence B = {log Pn}, where Pn
denotes the nth prime. In fact, if we use the primepower decomposition
(5)
00
n = npkan,k
k=l
where each exponent an, k 0, we he
00
log n = I an,k log Pk,
k=l
so the integer powers may be used as entries in the Bohr matrix R
B
for which
A = RBB. In the sum and product only a finite number of the an,k are
nonzero.
We note that, because of the fundamental theorem of arithmetic, the
numbers an, k defined by (5) have the property
(6)
174
8.8: Equivalence of ordinary Dirichlet series
Now let A(s) = La(n)n
S
, B(s) = I b(n)n
S
• Suppose that A(s) '" B(s).
Then there exists a real sequence {Yk} such that
(7) b(n) = a(n)eXP{iJlan,kYk}
where the integers an, k are determined by equation (5). Define a function f
by the equation
f(n) = eXP{iJl
an
,kYk}
Property (6) implies that f(mn) = f(m)f(n) for all m and n, so f is completely
multiplicative. Equation (7) states that b(n) = a(n)f(n), and the definition of
f shows that If(n) I = 1 for all n, so conditions (a) and (b) of the theorem are
satisfied.
Now we prove the converse. Assume there exists a completely multi
plicative function f satisfying conditions (a) and (b). We must show that
there is a real sequence {Yk} satisfying (7) for all n. First we consider those n
for which a(n) = O. Property (a) implies b(n) = 0, so equation (7) holds for
such n since both sides are zero no matter how we choose the real numbers
Yk. We shall now construct the sequence {Yk} so that equation (7) also holds
for those n for which a(n) =1= O.
Assume, then, that n is such that a(n) =1= O. We use the primepower
decomposition (5) and the completely multiplicative property of f to write
(8)
where
00
f(n) = ng(n, k),
k=l
g(n, k) = {f(pk)an'k if Pk In,
1 otherwIse.
Condition (b) implies that If(Pk) I = 1for each prime divisor Pk ofn. Therefore
for such primes we may write
where Yk = argf(Pk)· The real numbers Yk have been defined for those k such
that the prime Pk divides some n with a(n) =1= O. For the remaining k (if any)
we define Yk = o. Thus, Yk is welldefined for every integer k ~ 1 and we
have
g(n, k) = exp(ia
n
, kYk)
for every k ~ 1. Equation (8) now becomes
f(n) = e x p L ~ l an,kYk}
175
8: General Dirichlet series and Bohr's equivalence theorem
This, together with property (a), shows that (7) holds for those n for which
a(n) i= O. Thus, (7) holds for all n so A(s) '" B(s). This completes the proof of
the theorem. D
8.9 Equality of the sets U/((Jo) and Ug((Jo) for
equivalent Dirichlet series
Theorem8.I3. Let .f(s) and g(s) be equivalent general Dirichlet series, each of
which converges absolutely for (J = (Jo. Then
Uf((Jo) = Ug((Jo).
PROOF. Let B = {f3(n)} be a basis for the sequence A of exponents. If
f(s) = La(n)eSA(n) and g(s) = Lb(n)eSA(n) then there is a real sequence
{Yk} such that
{
q(n) }
b(n) = a(n) exp .
The Bohr series of f and g are given by
00 {q(n)}
and
ex; {q(n)}
Expressing the b(n) in terms of the a(n) we find
G(Zt, Z2"") = a(n) ex
p
{  + iYk)} = F(zl + iYI' Z2 + iY2'" .).
Since the real part of Zn + iYn is the real part of Zn, both series take the same
set of values on the lines X
n
= (Jo f3(n). Hence Uf((JO) = Ug((Jo), as asserted.
D
8.10 The set of values taken by a Dirichlet
series in a neighborhood of the
line (J = (Jo
Definition. Let f(s) be a general Dirichlet series which converges absolutely
for (J > (Ja. Given lJ > 0 and (J0 such that (J 0  lJ > (Ja' we define the
set W
f
((J0; lJ) as follows:
Wf((Jo;lJ) = {f(s):(Jo lJ < (J < (Jo + lJ, 00 < t < +oo}.
176
8.10: The set of values taken by a Dirichlet series in a neighborhood of the line (J = (J0
That is, Wf(O"o; £5) is the set of values taken by f(s) in the strip
0"0  £5 < 0" < 0"0 + £5.
Also, if 0"0 > 0"a we define
Wf(O"o) = n Wf(O"o; b).
o<c5<Q'oQ'a
Thus, Wf(O"0) is the intersection of the sets of values taken by f(s) in all
such strips.
It is clear that Vf(O" 0) S; Wf(O"0) since every value taken by f(s) on the line
0" = 0"0 is also taken in every strip containing this line. Of course, it may
happen that Vf(O"o) = Wf(O"o) or that Vf(O"o) =1= Wf(O"o).
In general, we have:
Theorem 8.14. Vf(O"o) S; Wf(O"o) S; Vf(O"o), hence Vf(O"o) = Wf(O"o).
PROOF. We remark that this proof is entirely functiontheoretic and has
nothing to do with the concept of a basis.
We are to prove that every point in W
f
(0"0) is in the closure of l/f(0"0).
We will show that if WE Wf(O"o) then W is an adherent point of Vf(O"o). In
fact, we will prove that
W = lim f(O"o + it
n
)
noo
for some real sequence {t
n
}.
Since W E W
f
(0"0) this means that W E W
f
(0"0; £5) for all £5 > 0 such that
£5 < 0"0  O"a. In particular, WE Wf(O"o; lin) for all n ~ no for some no.
This means that for n ~ no we have W = f(sn) where Sn = O"n + it
n
and
0"0  (lin) < O"n < 0"0 + (lin). Using the numbers t
n
so determined, consider
the difference
~ v  f(O"o + it
n
) = f(O"n + it
n
)  f(O"o + it
n
)
where n ~ no. We shall express this difference in terms of the derivative
f'(s). Now just as in the case of ordinary Dirichlet series, the function f(s)
defined by
ex:,
f(s) = L a(n)eSA(n)
n=l
is analytic within its halfplane of absolute convergence. In fact in the proof
of Theorem 8.1 we showed that the series converges uniformly on every
compact subset of the halfplane 0" > O"c. Therefore the sum is analytic in
the halfplane 0" > O"c. Moreover, we can calculate the derivative f'(s) by
termbyterm differentiation, so
00
f'(s) =  L a(n)A(n)eSA(n).
n=l
177
8: General Dirichlet series and Bohr's equivalence theorem
Hence if (J ~ (J0 then s is the halfplane of absolute convergence and we get
00 00
If'(s)1 s L la(n)IIA(n)leUA(n) = L la(n)leuO'A(n)IA(n)le(truO')A(n)
n=1 n=1
where (Ja < (Jo' < (Jo. Now IA(n)le(trtro')A(n) ~ 0 as n ~ 00 so, in particular,
this factor is less than 1 for large enough n. Hence
00
If'(s)1 s L la(n)letro'A(n). K
n=1
for some K, which shows that If'(s) I is uniformly bounded in the region
(J ~ (Jo'. Let (Jo' = (Jo  11no and let M be an upper bound for If'(s) I in the
region (J ~ (Jo'. Then
Iw  f((1o + it
n
)I = If((1n + it
n
)  f((1o + it
n
)I = If:"1'((1 + itn) d(1!
M
S MI(Jn  (Jol s
n
if n ~ no. Hence lim
n
+
oo
f((Jo + it
n
) = w, so w is an adherent point of
Vf((J0). This completes the proof. D
8.11 Bohr's equivalence theorem
We have just shown that Wf((Jo) ~ Vf((Jo). The next theorem shows that this
inclusion is actually equality.
Theorem 8.15. We have
The proof of Theorem 8.15 is lengthy and appears in Section 8.12. In
this section we show how Theorem 8.15 leads to Bohr's equivalence theorem.
Theorem 8.16 (Bohr's equivalence theorem). Let f and g be equivalent
Dirichlet series with abscissa of absolute convergence (Ja. Then in any
open half plane (J > C 1 ~ (Ja the functions f(s) and g(s) take the same set
ofvalues.
PROOF. Let Sf((J 1) be the set of values taken by f(s) in the halfplane (J > (J 1.
Then
Sf( (J 1) = U V
f
((J0)·
UO>U1
Now we prove that
Sf((J 1) = U Wr((Jo)·
UO>U1
178
8.12: Proof of Theorem 8.15
First of all, we have Sf(0"1) ~ U
t1
0>
t1
1 Wf(O"o) because Vf(O"o) ~ Wf(O"o). To
get inclusion in the other direction, assume WE U
t1
0>
t1
1 Wf(O"o). Then
w E W
f
(0"0) for some 0"0 > 0" 1. Hence W E W
f
(0"0; b) for all <5 satisfying
o< b < 0"0  O"a. In other words, f(s) takes the value w in every strip
0"0  b < 0" < 0"0 + b if 0 < b < 0"0  O"a. In particular, when b = 0"0  O"b
we have 0"0  <5 = 0"1 so f(s) = w for some s with 0" > 0"1. Hence WE Sf(0"1).
This proves that U
t1
0>
t1
1 Wf(O"o) ~ Sf(0"1), so the two sets are equal. Therefore,
we also have
Sg(O" 1) = U ~ ( O " 0)·
t10>t11
To prove Bohr's theorem it suffices to prove that
Wf(O"o) = ~ ( O " o )
whenever f and g are equivalent. But f '" g implies
Uf(O" 0) = Ug(O"0).
Hence Uf(O"o) = Ug(O"o). But, in view of Theorem 8.9, this means
Vf(O"o) = Vg(O"o)·
But Theorem 8.15 states that Vf(O"o) = Wf(O"o) and ~ ( O " o ) = ~ ( O " o ) so Bohr's
equivalence theorem is a consequence of Theorem 8.15. D
8.12 Proof of Theorem 8.15
To complete the proof of Bohr's equivalence theorem we need to prove
Theorem 8.15, which means we must establish the inclusion relation.
(9)
The proof of (9) makes use of two important theorems of analysis which we
state as lemmas:
Lemma 1 (Helly selection principle). Let {em, n} be a double sequence of real
numbers which is bounded, say
lem,nl < A for all m, n.
Then there exists a subsequence of integers nl < n
z
< ... with n
r
~ 00
as r ~ 00, and a sequence {en} of real numbers such that for every
m = 1,2, ... , we have
lim em, n
r
= em.
r+ 00
Note. The important point is that one subsequence {n
k
} works for every m.
To show the true import of the Lemma, let us see what we can deduce in a
trivial fashion. Display the double sequence as an infinite matrix. Consider
179
8: General Dirichlet series and Bohr's equivalence theorem
the first row: {0l,n}:==1. This is a bounded infinite sequence so it has an
accumulation point, say 01. Hence there is a subsequence {n
r
} such that
limr+(X) 01,nr = 01. Similarly, for the second row there is an accumulation
point O
2
and a subsequence nr' such that lim
n
+
oo
02,nr' = 02, and so on. The
subsequence {n/} needed for O
2
may be quite different from that needed for
01. Helly's principle says that one subsequence works simultaneously for all
rows.
PROOF OF LEMMA 1. Let °1 be an accumulation point of the first row and
suppose the subsequence {n
r
(1)} has the property that
lim 0l,nr(l) = 01.
r+ 00
In the second row, consider only those entries 02. n
r
(1). This is a bounded
sequence which has a convergent subsequence with limit °
2
, say. Thus,
lim 02,n
r
(2) = O
2
r+ 00
where {n
r
(2)} is a subsequence of {n
r
(1)}. Repeat the process indefinitely. At
the mth step we have a subsequence {nr(m)} which is a subsequence of all
earlier subsequences and a number Om such that
lim Om, nr(m) = Om.
r+ 00
Now define a sequence {n
r
} by the diagonal process:
That is, n1 is the first integer used in the first row, n2 the second integer used
in the second row, etc. Look at the mth row and consider the sequence
{Om,n
r
}. We assert that
r+ 00
Since n
r
= nr(r), after the mth term in this row we have r > m so every integer
nr(r) occurs in the subsequence n/
m
), so from this point on {n
r
} is a sub
sequence of {nr(m)} hence 0m,n
r
~ Om' as asserted. D
Lemma 2 (Rouche's theorem). Given two functions f(z) and g(z) analytic
inside and on a closed circular contour C. Assume
Ig(z)1 < If(z)1 on C.
Then f(z) and f(z) + g(z) have the same number of zeros inside C.
PROOF OF LEMMA 2. Let m=inf{lf(z)Ilg(z)l:zEC}. Then m>O
because C is compact and the difference If(z) I  Ig(z) I is a continuous
function on C. Hence for all real t in the interval 0 :::; t :::; 1 we have
If(z) + tg(z) I ~ If(z)1  Itg(z) 1~ If(z)1  Ig(z)1 ~ m > O.
180
8.12: Proof of Theorem 8.15
If 0 ~ t ~ 1 define a number cp(t) by the equation
cp(t) = _1 [f'(z) + tg'(z) dz.
2ni J
c
f(z) + tg(z)
This number cp(t) is an integer, the number of zeros minus the number of
poles of the function f(z) + tg(z) inside C. But there are no poles, so cp(t) is
the number of zeros of f(z) + tg(z) inside C. But cp(t) is a continuous function
of t on [0, 1]. Since it is an integer, it is constant: cp(O) = cp(I). But cp(O) is the
number of zeros of f(z), and cp(l) is the number of zeros of f(z) + g(z). This
proves Rouche's theorem. 0
PROOF OF RELATION (9). Vf(O"o) £ Wf(O"o). Assume v E Vf(O"o). Then either
v E Vf(O"o) or v is an accumulation point of Vf(O"o). If v E Vf(O"o) then v E Wf(O"o)
since Vf(O"o) £ Wf(O"o). Hence we can assume that v is an accumulation point
of Vf(O"o), and v $ Vf(O"o). This means there is a sequence {t
n
} of real numbers
such that
v = lim f(O"o + itn)'
n+ 00
We wish to prove that v E W
f
(0" 0)' This means we must show that v E W
f
(0" 0; b)
for every b satisfying 0 < b < 0"0  O"a' In other words, if 0 < b < 0"0  O"a
we must find an s = 0" + it in the strip
0"0  b < 0" < 0"0 + b
such that f(s) = v. Therefore we are to exhibit an s in this strip such that
f(s) = lim f(O"o + itn)'
n+ 00
Let us examine the numbers f(O"o + it
m
) for the given sequence {t
n
}. We
have
00
f(O"o + it
m
) = L a(n)etTOA(n). eitmA(n).
n= 1
The products tmA(n) form a double sequence. There exists a double sequence
of real numbers 8
n
,m such that
8
n
,m = tmA(n) + 2nk
m
,n' with 0 ~ 8
n
,m < 2n,
where km,n is an integer. If we replace tmA(n) by 8
n
,m in the series we don't
alter the terms, hence
00
f(O"o + it
m
) = L a(n)etToA(n)eiOn,m.
n=l
By Lemma 1, there is a subsequence of integers {n
r
} and a sequence of real
numbers {8
m
} such that
(10)
lim 8
m
,nr = Om'
r+ 00
181
8: General Dirichlet series and Bohr's equivalence theorem
Use this sequence {Om} to form a new Dirichlet series
00
g(s) = L b(n)es).(n)
n=1
where
This has the same abscissa of absolute convergence as f(s). Now consider
the following sequence of functions:
f,.(s) = f(s + it
n
)
where {n
r
} is the subsequence for which (10) holds. We assert that
(a) f,.(s) + g(s) uniformly in the strip 0'0  b < a < 0'0 + b, hence, in
particular, in the circular disk Is  0'0 I < b.
(b) g(O'o) = v.
(c) There is a d, 0 < d < b, and an R such that fR(S)  v and g(s)  v have
the same number of zeros in the open disk Is  0'01 < d.
If we prove (b) and (c) then fR(S)  v has at least one zero in the disk because
g(O'o) = v. But fR(S) = f(s + it
nR
) and s + it
nR
is in the strip if s is in the disk,
so this proves the theorem. Now we prove (a), (b) and (c).
Proof of (a). We have
If,.(s)  g(s) I = IJla(n)eSA(nl(ei8n,n,  e
i8n
)1
00
s L la(n)letr).(n)leiOn,nr  eiOnl
n=1
N
S L la(n)le(trob»).(n)leiOn,nr  eiOnl
n=1
00
+ 2 L la(n)le(trob»).(n).
n=N+ 1
Now if 8 > 0 is given there is a number N = N(8) such that
00 8
2 L la(n)le(trob»).(n) < ,
n=N+ 1 2
because the series L:= 1 Ia(n) Ie(trob»).(n) converges. In the finite sum L ~ = 1
we use the inequality
182
8.12: Proof of Theorem 8.15
to write
l
eiOn,n
r
 eiOnl < 10  0 I
 n, n
r
n .
But if M(b) = 1 + L ~ = l la(n)le(tTOb)A(n), there is an integer ro = ro(8)
such that for every n = 1, 2, ... , N we have
8
/
0  0 I <  if r ~ ro.
n, n
r
n 2M(b)
Therefore, if r ~ ro we have
8 N 8 8 8
I/"(s)  g(s) I ~  L la(n)/e(tTOb)A(n) +  <  +  = 8.
2M(b) n = 1 2 2 2
Since ro depends only on 8 and on b this shows that /,.(s) ~ g(s) uniformly
in the strip 0'0  b < a < 0'0 + b as r ~ 00. This proves (a).
Proof of (b). We use (a) to write
g(ao) = lim};.(a
o
) = limf(ao + itnJ = v.
roo roo
Proof of (c). Assume first that g is not constant. Since g(a0) = v there is a
positive d < b such that g(s) =I v on the circle
C = {s: Is  0'01 = d}.
Let M be the minimum value of Ig(s)  vIon C. Then M > O. Now choose R
so large that IfR(s)  g(s) I < M on C. This is possible by uniform convergence
of the sequence {fR(S)}, since the circle C lies within the strip 1(1  0'0 I < b.
Then on C we have
/fR(s)  g(s) I < M ~ Ig(s)  vi.
If G(s) = fR(s)  g(s) and F(s) = g(s)  v we have /G(s) / < /F(s) / on C with
F(s), G(s) analytic inside C. Therefore, by Rouche's theorem the functions
F(s) + G(s) and F(s) have the same number of zeros inside C. But F(s) + G(s)
= fR(s)  v, so fR(S)  v has the same number of zeros inside C as g(s)  v.
Now g(ao) = v so g(s)  v has at least one zero inside C. Hence fR(s)  v
has at least one zero inside C. As noted earlier, this completes the proof if g
is not constant.
To complete the proof we must consider the possibility that g(s) is constant
in the halfplane of absolute convergence. Then g'(s) = 0 for all s in this half
plane, which means
00
g'(s) =  L A(n)b(n)eSA(n) = O.
n= 1
But as in the case of ordinary Dirichlet series, if a general Dirichlet series
has the value 0 for a sequence of values of s with real parts tending to +00
then all the coefficients must be zero. (See [4], Theorem 11.3.) Hence
183
8: General Dirichlet series and Bohr's equivalence theorem
A(n)b(n) = 0 for all n 2 1. Therefore b(n) = 0 with at most one exception,
say b(n
1
), in which case A(n
1
) = O. Therefore, since a(n) = b(n)e
iOn
, we must
have a(n) = 0 with at most one exception, say a(nl), and then A(nl) = O.
Hence the series for f (s) consists of only one term, f (s) = a(n l)e  sA(nd = a(n1),
so f(s) itself is constant. But in this case the theorem holds trivially. D
8.13 Examples of equivalent Dirichlet series.
Applications of Bohr's theorem to
Lseries
Theorem 8.17. Let k 2 1 be a given integer, and let Xbe any Dirichlet character
modulo k. Let L:= 1 a(n)n  S be any Dirichlet series whose coefficients
have the following property:
a(n) :1= 0 implies (n, k) = 1.
Then
i a(7) ~ i a ( n ) ~ ( n ) .
n= 1 n n= 1 n
PROOF. Since these are ordinary Dirichlet series we may use"Theorem 8.12
to establish the equivalence. In this case we take f(n) = x(n). Then f is
completely multiplicative and condition (a) is satisfied. Now we show that
condition (b) is satisfied. We need to show that If(p) I = 1 if a(n) :1= 0 and
pIn. But a(n) :1= 0 implies (n, k) = 1. Since pin we must have (p, k) = 1 so
If (P) I = IX(p) I = 1 since X is a character. Therefore the two series are
equivalent. D
Theorem 8.18. For a given modulus k, let ·Xl' ... , Xq>(k) denote the Dirichlet
characters modulo k. Then in any halfplane of the form (J > (J 1 2 1 the
set of values taken by the Dirichlet Lseries L(s, X;) is independent of i.
PROOF. Applying the previous theorem with a(n) = Xl(n) we have
f X l ~ n ) ~ i Xl(n);(n)
n=l n n=l n
for every character Xmodulo k. Here we use the fact that Xl (n) :1= 0 implies
(n, k) = 1. Thus each Lseries L(s, X) is equivalent to the particular Lseries
L(s, Xl). Therefore, by Bohr's theorem, L(s, X) takes the same set of values as
L(s, X1) in any open halfplane within the halfplane of absolute convergence.
D
8.14 Applications of Bohr's theorem to the
Riemann zeta function
OUf applications to the Riemann zeta function require the following identity
involving Liouville's function A(n) which is defined by the relations
A(I) = 1, A(PI al ... Pr
ar
) = (_1)a
l
+ ... +a
r
•
184
8.14: Applications of Bohr's theorem to the Riemann zeta function
The function )..(n) is completely multiplicative and we have (see [4J, p. 231)
I ),(n) = ((2s) if (J > 1
n=l n
S
((s) .
Theorem 8.19. Let A(n) denote Liouville's function and let
C(x) = L A(n).
n:5x n
Then if (J > 1 we have
((2s) = Joc C(x) dx.
(s  1)((s) 1 X
S
PROOF. By Abel's identity (Theorem 4.2 in [4J) we have
L A(n) 2 = C(x) + s JX C(t) dt.
n:5 x n n
S
X
S
1 t
S
+ 1
Keep (J > 0 aRd let x ~ 00. Then
C(x) = o ( ~ L !) = O(IOg x) = 0(1) as x~ 00,
X
S
x(1 n:Sx n x(1
so we find
~ A(n) _
~ S+ 1  S
n= 1 n
f
x C(t)
S+I dt, for (J > O.
1 t
Replacing s by s  1 we get
I A(7) = (s  1) foo C ~ ) dt for (J > 1.
n=l nIt
Since the series on the left has sum ((2s)/((s) the proof is complete. D
Now we prove a remarkable theorem discovered by P. Tunin [50] in
1948 which gives a surprising connection between the Riemann hypothesis
and the partial sums of the Riemann zeta function in the halfplane (J > 1.
Theorem 8.20. Let
n 1
(n(s) = L k
s
'
k=l
If there exists an no such that (n(s) :1= 0 for all n ~ no and all (J > 1, then
((s) :1= 0 for (J > !.
PROOF. First we note that the two Dirichlet series L ~ =1 k S and L ~ =1 A(k)k  S
are equivalent because ).. is completely multiplicative and has absolute
185
8: General Dirichlet series and Bohr's equivalence theorem
value 1. by Bohr's theorem, (n(s) =1= 0 for (J > 1 implies that
1 A(k)k s =1= 0 for (J > 1. But for s real we have
lim i = ,1(1) = 1.
s + 00 k = 1 k
Hence for all real s > 1 we must have = 1 A(k)k  s > O. Letting s 1+
we find
A(k) > 0
i...J if n no·
k=1 k 
In other words, the function
(11 )
C(x) = L ,1(n)
n:S;x n
is nonnegative for x no. Now we use the identity of Theorem 8.19,
valid for (J > 1. Note that the denominator (s  1)'(s) is nonzero on the
real axis s > t, and '(2s) is finite for real s > t. Therefore, by the integral
analog of Landau's theorem (see Theorem 11.13 in [4]) the function on the
left is analytic everywhere in the halfplane (J > t. This implies that '(s) =1= 0
for (J > t, and the proof is complete. D
Turan's theorem assumes that the sum C(x) in (11) is nonnegative for all
x no. In 1958, Haselgrove [14] proved, by an ingenious use of machine
computation, that C(x) is negative for infinitely many values of x. Therefore,
Theorem 8.20 cannot be used to prove the Riemann hypothesis. Subse
quently, Tunin [51] sharpened his theorem by replacing the hypothesis
C(x) 0 by a weaker inequality that cannot be disproved by machine
computation.
Theorem 8.21 (Turan). Let C(x) = Ln:s;x A(n)jn. rr there exist constants
r:t. > 0, C > 0 and no such that
(12)
loga x
C(x) > c
fi
for all x no, then the Riemann hypothesis is true.
PROOF. If G > 0 is given there exists an nl no such that c loga x xl: for
all x nl so (12) implies
C(x) >  xl: 1/2 •
186
Exercises for Chapter 8
Let A(x) = C(x) + xf, 1/2, where G is fixed and 0 < G < t. Then A(x) > 0
for all x 2 n
1
• Also, for a > 1 we have
f
OO A(x) d  foo C(x) d foo f,sI/2 d
x x+ x X
1 X
S
1 x
S
1
'(2s) 1 = f( )
(
+ 1 S,
(s  1)' s) s  2  G
say. Arguing as in the proof of Theorem 8.20, we find that the function f(s)
is analytic on the real line s > t + G. By Landau's theorem it follows that
f(s) is analytic in the halfplane a > t + G. This implies that '(s) =1= 0 for
a > t + G, hence '(s) =1= 0 for a > t since G can be arbitrarily small. D
Note. Since each function 'n(s) is a Dirichlet series which does not vanish
identically there exists a halfplane (J > 1 + an in which 'n(s) never vanishes.
(See [4], Theorem 11.4.) The exact value of an is not yet known. In his 1948
paper [50] Turan proved that, for all sufficiently large n, 'n(s) =1= 0 in the
halfplane a > 1 + 2(1og log n)jlog n, hence an ~ 2(log log n)jlog n for large
n. In the other direction, H. L. Montgomery has shown that there exists a
constant c > 0 such that for all sufficiently large n, 'n(s) has a zero in the half
plane a > 1 + c(log log n)jlog n, hence an 2 c(1og log n)jlog n for large n.
The number 1 + an is also equal to the abscissa of convergence of the
Dirichlet series for the reciprocal 1j'n(s). If a > 1 + an we can write
1 _ f JIn(k)
'n(s)  k= 1 ~ '
where /In(k) is the Dirichlet inverse of the function un(k) given by
{
1 if k ~ n,
un(k) = 0 ifk > n.
The usual Mobius function /l(k) is the limiting case of /In(k) as n + 00.
Exercises for Chapter 8
1. IfL a(n)eSA(n) has abscissa of convergence (Jc < 0, prove that
. loglLk=n a(k) I
(Jc = lIm sup A .
n+oo (n)
2. Let (Jc and (Ja denote the abscissae of convergence and absolute convergence of a
Dirichlet series. Prove that
. log n
o~ (Ja  (Jc ~ lIm sup A(n)'
n+ 00
This gives 0 ~ (Ja  (Jc ~ 1 for ordinary Dirichlet series.
187
8: General Dirichlet series and Bohr's equivalence theorem
3. If log njA(n) + 0 as n + 00 prove that
. 10gla(n)1
aa = a
c
= hm sup .
n+ ex: A(n)
What does this imply about the radius of convergence of a power series?
4. Let {J(n)} be a sequence of complex numbers. Let A denote the set of all points
s = a + it for which the series I a(n)eSA(n) converges absolutely. Prove that A
is convex.
Exercises 5 ~ 6 ~ and 7 refer to the seriesf'(s) == L:::= 1 a{n)es).(n) with exponents
and coefficients given as follows
n 1 2 3 4 5
A(n) 1  log 2 1 log2 1 + log 2 0
a(n)
1 1 1 1 1
8 2" 4
8
2"
n 6 7 8 9 10
A(n) 1  log 2 log 2 1 log 3 1 + log 2
a(n)
1 1 1 3 1
8
4
2"
4 8
Also, a(n + 10) == 1 2
n
and A(n + 10) == (n + 1) log 3 for n ~ 1.
5. Prove that aa =  (log 2)jlog 3.
6. Show that the Bohr function corresponding to the basis B = (1, log 2, log 3) is
1  2e
Z3
F(Zb Z2' Z3) = cos(izd  !i sin(i
z
2)(1 + cos(izd) + 2 Z3 '
e
if X3 > log 2, Z1' 22 arbitrary.
7. Determine the set Uf(O). Hint: The points 1, 1 + i, 1  i are significant.
8. Assume the Dirichlet series f(s) = I ~ = 1 a(n)es).(n) converges absolutely for a > aa'
If a > aa prove that
1 IT, {a(n) if A = A(n)
lim  e
A
(rT+lt)f(a + it) dt =
T+ + y 2T  T 0 if A=1= A(I), A(2), ....
9. Assume the series f(s) = L ~ = 1 a(n)eSA(n) converges absolutely for (J > aa > O. Let
v(n) = eA(n).
(a) Prove that the series g(s) = L ~ = 1 a(n)eSv(n) converges absolutely if a > O.
(b) If a > aa prove that
ns)f(s) = fCg(t)t
S

1
dt.
188
Exercises for Chapter 8
This extends the classic formula for the Riemann zeta function,
f(s)((s) = foc ~ dt.
o e  1
Hint: First show that f(s)eSA(n) = So etv(n)t
s

1
dt.
189
(2)
Supplement to Chapter 3
Alternate proof of Dedekind' s functional equation
This supplement gives an alternate proof of Dedekind' s functional equation
as stated in Theorem 3.4:
Theorem. IfA = (; ~ ) E r and c > 0, thenfor every T in H we have
(
aT + b) { }112
(1) TJ CT + d == e(A)  i(CT + d) TJ(T) ,
where
e(A) = eX
P
{1Ti(a
1
;C
d
 s(d, C»)}
and s(d, c) is a Dedekind sum.
The alternate proof was suggested by Basil Gordon and is based on the fact
that the modular group r has the two generators TT == T + 1 and ST =
 lIT. In Theorem 2.1 we showed that every A in r can be expressed in
the form
where the n, are integers. But T = ST1ST1S, so every element of r also
has the form STtnlS · · · STtnk for some choice of integers m
l
, ••• , m
k
. The
idea of the proof is to show that if the functional equation (1) holds for a
particular transformation A = ( ~ ~ ) in r with c > 0 and with e(A) as
specified in (2), then it also holds for the products ATtn and AS for every
190
PROOF. We have ATtn ==
Supplement to Chapter 3
integer m. (See Lemma 3 below.) In Theorem 3.1 we proved that it holds
for S. Therefore, because every element of r has the form STmtS · · · STmk,
it follows that the functional equation (1) holds for every A with e > o.
The argument is divided into three lemmas that show that the general
functional equation is a consequence of the special case in Theorem 3. 1
together with three basic properties of Dedekind sums derived in Sections
3.7 and 3.8. The first two lemmas relate e(A) with e(AT
m
) and e(AS), where
T and S are the generators of the modular group.
Lemma 1. If A = ( ~ ~ ) E rand e > 0, then for every integer m we
have
(
a b)(1 m) == (a am + b),
c dOl c em + d so
s(AT
m
) = exp{ 7Ti(a+ ~ ; : . + d  seem + d, c»)}.
But s(em + d, c) == s(d, e) by Theorem 3.5(a), and hence, we obtain
Lemma 1. 0
Lemma 2. If A
PROOF. We have
( ~ ~ ) E rand e > 0, then we have
{
e 7Ti/4
e
(A) if d > 0,
e(AS) == e
7T1
/
4
e(A) ifd < O.
== (a b)(O 1) == (b a)
AS c d 1 0 d c·
If d > 0, we represent the transformation AS by the matrix
AS = ( ~
a),
e
but if d < 0, then  d > 0 and we use the representation AS
(
b a).
d e
For d > 0, we have
(3) seAS) = exp{ 7Ti(b
1
;/  s(  e, d))}
= eX
P
{7Ti(b
1
;/ + s(e, d))}
191
Supplement to Chapter 3
because s(  c, d) =  s(c, d) by Theorem 3.5(a). The reciprocity law for
Dedekind sums implies
c d 1 1
s(c, d) + s(d, c) = 12d + 12c  4 + 12cd·
We replace the numerator in the last fraction by ad  bc and rearrange
terms to obtain
bc a+d 1
Ud + s(c, d) = ~  s(d, c)  4·
Using this in (3), we find e(AS) = e  7Ti/4e(A).
(
b
If d < 0, we use the representation AS = _ d
;) to obtain
(4)
In this case,  d > 0 and we use the reciprocity law in the form
c dIad  bc
s(c,  d) + s(  d, c) = _ 12d  12c  4  12cd·
Rearrange terms and use s(  d, c) =  s(d, c) to obtain
b+c a+d 1
12d  s(c, d) = ~  s(d, c) + 4·
Using this in (4), we find that e(AS)
of Lemma 2.
e
7Ti
/
4
e(A). This completes the proof
o
Lemma 3. If Dedekind's functional equation
(5)
is satisfied for some A = ( ~ ~ ) in r with c > 0 and seA) given by (2),
then it is also satisfied for AT
In
and for AS. That is, (5) implies
(6)
and
(7)
whereas
(8)
192
Supplement to Chapter 3
PROOF. Replace T by TinT in (5) to obtain
'Y](ATmT) = e(A){  i(cTmT + d)}1I2'Y](T
m
T)
= e(A){  i(CT + mc + d)}1/2
e
1Tim/12'Y](T).
Using Lemma 2 we obtain (6).
Now ~ e p l a c e T by ST in (5) to get
'Y](AST) = e(A){  i(CST + d)}1/2'Y](ST).
Using Theorem 3.1 we can write this as
(9) 'Y](AST) = e(A){  i(CST + d)}1/2 {  iT}I/2'Y](T).
If d > 0, we write
C dT  C
CST + d =  + d = _.
T T '
hence,
 i(dT  c)
 i(CST + d) = . e
1Til2
,
IT
and therefore, { i(CST + d)}112 { iT} 1/2 = e
1T1
/
4
{  i(dT  C)}1I2. Using this
in (9) together with Lemma 3, we obtain (7).
If d < 0, we write
_
~ + d __  dT + C
CST + d =
T T
so that in this case we have
 i(  dT + c) 1Til2
 i(CST + d) = _ iT e
and therefore, { i(CST + d)}112 { iT}1I2 = e
1Ti
/
4
{  i(  dT + c)} 112 . Using
this in (9) together with Lemma 3, we obtain (8). D
Remark on the root of unity B(A)
Dedekind's functional equation (1), with an unspecified 24th root of unity
e(A), follows immediately by extracting 24th roots in t ~ e functional equation
for d(T). Much of the effort in this theory is directed at showing that the
root of unity e(A) has the form given in (2). It is of interest to note that a
simple argument due to Dedekind gives the following theorem:
Theorem. If (1) holds whenever A = (: ~ ) E rand c # 0, then
B(A) = eX
P
{7Ti(a};C
d
 f(d. C))}
for some rational number f(d, c) depending only on d and c.
193
Supplement to Chapter 3
PROOF. Let
aT + b
AT=
CT + d
and
, a'T + b'
AT= cT+d
be two transformations in r having the same denominator CT + d. Then
adbc=1 and a'd  b'c = 1,
so both pairs a, b and a', b' are solutions of the linear Diophantine equation
xd  yc = 1.
Consequently, there is an integer n such that
Hence,
a' = a + nc, b' = b + nd.
(a + nC)T + (b + nd) aT + b A
A'T = d = d + n = T + n.
CT+ CT+
Therefore, we have
'11(A'T) = '11(AT + n) = e1Tin/12'11(AT) = e1Tin/12e(A){  i(CT + d)}1/2'11(T),
because of (1). On the other hand, (1) also gives us
'11(A'T) = e(A'){  i(CT + d)}1I2'11(T).
Comparing the two expressions for '11(A'T), we find e(A')
n = (a'  a)/c, so
(
tri(a'  a))
e(A') = exp 12c e(A),
or
(
.') (. )
tria, Tria
exp  12c e(A) = exp  12c e(A).
e1Tin/12e(A). But
This shows that the product ex
p
( ~ ~ ; ) e ( A ) depends only on cand d. There
fore, the same is true for the product
(
tri(a + d))
exp 12c e(A).
This complex number has absolute value 1 and can be written as
(
tri(a + d))
exp 12c e(A) = exp( trif(d, c))
194
Supplement to Chapter 3
for some real number f(d, c) depending only on c and d. Hence,
Because 8
24
= 1, it follows that 12cf(d, c) is an integer, so f(d, c) is a
rational number. D
195
Bibliography
1. Apostol, Tom M. Sets of values taken by Dirichlet's Lseries. Proc. Sympos. Pure
Math., Vol. VIII, 133137. Amer. Math. Soc., Providence, R.I., 1965. MR 31
# 1229.
2. Apostol, Tom M. Calculus, Vol. II, 2nd Edition. John Wiley and Sons, Inc. New
York, 1969.
3. Apostol, Tom M. Mathematical Analysis, 2nd EditioQ. AddisonWesley Publishing
Co., Reading, Mass., 1974.
4. Apostol, Tom M. Introduction to Analytic Number Theory. Undergraduate Texts
in Mathematics. SpringerVerlag, New York, 1976.
5. Atkin, A. O. L. and O'Brien, J. N. Some properties ofp(n) and c(n) modulo powers
of 13. Trans. Amer. Math. Soc. 126 (1967),442459. MR 35 # 5390.
6. Bohr, Harald. Zur Theorie der allgemeinen Dirichletschen Reihen. Math. Ann.
79 (1919), 136156.
7. Deligne, P. La conjecture de Weil. I. Inst. haut. Etud sci., Publ. math. 43 (1973),
273307 (1974). Z. 287, 14001.
8. Erdos, P. A note on Farey series. Quart. J. Math., Oxford Ser. 14 (1943), 8285.
MR 5, 236b.
9. Ford, Lester R. Fractions. Amer. Math. Monthly 45 (1938), 586601.
10. Gantmacher, F. R. The Theory of Matrices, Vol. 1. Chelsea Publ. Co., New York,
1959.
11. Gunning, R. C. Lectures on Modular Forms. Annals of Mathematics Studies, No.
48. Princeton Univ. Press, Princeton, New Jersey, 1962. MR 24 # A2664.
12. Gupta, Hansraj. An identity. Res. Bull. Panjab Univ. (N.S.) 15 (1964), 347349
(1965). MR 32 # 4070.
13. Hardy, G. H. and Ramanujan, S. Asymptotic formulae in combinatory analysis.
Proc. London Math. Soc. (2) 17 (1918),75115.
14. Haselgrove, C. B. A disproof of a conjecture of P6lya. Mathematika 5 (1958),
141145. MR 21 #3391.
15. Heeke, E. Uber die Bestimmung Dirichletscher Reihen durch ihre Funktional
gleichung. Math. Ann. 112 (1936), 664699.
196
Bibliography
16. Hecke, E. Ober Modulfunktionen und die Dirichlet Reihen mit Eulerscher Produkt
entwicklung. I. Math. Afll1.114 (1937),128; II. 316351.
17. Iseki, Shoo The transformation formula for the Dedekind modular function and
related functional equations. Duke Math. J. 24 (1957),653662. MR 19, 943a.
18. Knopp, Marvin I. Modular Functions in Analytic Number Theory. Markham
Mathematics Series, Markham Publishing Co., Chicago, 1970. MR 42 #198.
19. Lehmer, D. H. Ramanujan's function r(n). Duke Math. J. 10 (1943), 483492.
MR 5, 35b.
20. Lehmer, D. H. Properties of the coefficients of the modular invariant J(r). Amer.
J. Math. 64 (1942), 488502. MR 3, 272c.
21. Lehmer, D. H. On the HardyRamanujan series for the partition function. J.
London Math. Soc. 12 (1937), 171176.
22. Lehmer, D. H. On the remainders and convergence of the series for the partition
function. Trans. Amer. Math. Soc. 46 (1939),362373. MR 1, 69c.
23. Lehner, Joseph. Divisibility properties of the Fourier coefficients of the modular
invariantj(r). Amer. J. Math. 71 (1949),136148. MR 10, 357a.
24. Lehner, Joseph. Further congruence properties of the Fourier coefficients of the
modular invariarttj(r). Amer. J. Math. 71 (1949), 373386. MR 10, 357b.
25. Lehner, Joseph, and Newman, Morris. Sums involving Farey fractions. Acta
Arith. 15 (1968/69),181187. MR 39 # 134.
26. Lehner, Joseph. Lectures on Modular Forms. National Bureau of Standards,
Applied Mathematics Series, 61, Superintendent of Documents, U.S. Government
Printing Office, Washington, D.C., 1969. MR 41 # 8666.
27. LeVeque, William Judson. Reviews in Number Theory, 6 volumes. American Math.
Soc., Providence, Rhode Island, 1974.
28. Mordell, Louis J. On Mr. Ramanujan's empirical expansions of modular functions.
Proc. Cambridge Phil. Soc. /9(1917), 117124.
29. Neville, Eric H. The structure of Farey series. Proc. London Math. Soc. 51 (1949),
132144. MR 10, 681f.
30. Newman, Morris. Congruences for the coefficients of modular forms and for the
coefficients ofj(r). Proc. Amer. Math. Soc. 9 (1958),609612. MR 20 # 5184.
31. Petersson, Hans. Uber die Entwicklungskoeffizienten der automorphen formen.
Acta Math. 58 (1932),169215.
32. Petersson, Hans. Ober eine Metrisierung der ganzen Modulformen. Jher. Deutsche
Math. 49 (1939),4975.
33. Petersson, H. Konstruktion der samtlichen L6sungen einer Riemannscher
Funktionalgleichung durch Dirichletreihen mit Eulersche Produktenwicklung. I.
Math. Ann. 116 (1939), 401412. Z. 21, p. 22; II. 117 (1939),3964. Z. 22, 129.
34. Rademacher, Hans. Ober die Erzeugenden von Kongruenzuntergruppen der
Modulgruppe. Abh. Math. Seminar Hamburg, 7 (1929), 134148.
35. Rademacher, Hans. Zur Theorie der Modulfunktionen. J. Reine Angew. Math.
167 (1932), 312336.
36. Rademacher, Hans. On the partition function pen). Proc. London Math. Soc. (2)
43 (1937), 241254.
37. Rademacher, Hans. The Fourier coefficients of the modular invariant j( r). Amer.
J. Math. 60 (1938), 501512.
38. Rademacher, Hans. On the expansion of the partition function in a series. Ann. of
Math. (2) 44 (1943),416422. MR 5, 35a.
197
Bibliography
39. Rademacher, Hans. Topics in Analytic Number Theory. Die Grundlehren der
mathematischen Wissenschaften, Bd. 169, SpringerVerlag, New YorkHeidel
bergBerlin, 1973. Z. 253.10002.
40. Rademacher, Hans and Grosswald, E. Dedekind Sums. Carus Mathematical
Monograph, 16. Mathematical Association of America, 1972. Z. 251. 10020.
41. Rademacher, Hans and Whiteman, Albert Leon. Theorems on Dedekind sums.
Amer. J. Math. 63 (1941),377407. MR 2, 249f.
42. Rankin, Robert A. Modular Forms and Functions. Cambridge University Press,
Cambridge, Mass., 1977. MR 58 #16518.
43. Riemann, Bernhard. Gessamelte Mathematische Werke. B. G. Teubner, Leipzig,
1892. ErHiuterungen zu den Fragmenten XXVIII. Von R. Dedekind, pp. 466
478.
44. Schoeneberg, Bruno. Elliptic Modular Functions. Die Grundlehren der mathema
tischen Wissenschaften in Einzeldarstellungen, Bd. 203, SpringerVerlag, New
YorkHeidelbergBerlin, 1974. MR 54 #236.
45. Sczech, R. Ein einfacher Beweis der TransformationsformeI fur log T1(Z). Math.
Ann. 237 (1978), 161166. MR 58 #21948.
46. Selberg, Atle. On the estimation of coefficients of modular forms. Proc. Sympos.
Pure Math., Vol. VIII, pp. 115. Amer. Math. Soc., Providence, R.I., 1965. MR 32
#93.
47. Serre, JeanPierre. A Course in Arithmetic. Graduate Texts in Mathematics, 7.
SpringerVerlag, New YorkHeidelbergBerlin, 1973.
48. Siegel, Carl Ludwig. A simple proofofl1( l/r) == l1{r)JTTi. Mathematika 1 (1954),
4. MR 16, 16b.
49. Titchmarsh, E. C./ntroduction to the Theory ofFourier Integrals. Oxford, Clarendon
Press, 1937.
50. Tunin, Paul. On some approximative Dirichlet polynomials in the theory of the
zetafunction of Riemann. Danske Vid. Selsk. Mat.Fys. Medd. 24 (1948), no. 17,
36 pp. MR 10, 286b.
51. Tunin, Paul. Nachtrag zu meiner Abhandlung •. On some approximative Dirichlet
polynomials in the theory of the zetafunction of Riemann." Acta Math. Acad.
Sci. Hungar. 10 (1959), 277298. MR 22 # 6774.
52. Uspensky, J. V. Asymptotic formulae for numerical functions which occur in the
theory of partitions [Russian]. Bull. Acad. Sci. URSS (6) 14 (1920),199218.
53. Watson, G. N. A Treatise on the Theory ofBessel Functions, 2nd Edition. Cambridge
University Press, Cambridge, 1962.
198
Index of special symbols
n(W
t
, W2) lattice generated by W
t
and W2, 2
p(z) Weierstrass pfunction, 10
G
n
Eisenstein series of order n, n 3, 12
G
2
Eisenstein series of order 2, 69
g2,g3
invariants, 12
eb
e
2,e3
values of f.J at the halfperiods, 13
w
2
), discriminant  27 , 14
H upper halfplane Im('r) > 0, 14
J(r) Klein's modular function 15
r(n) Ramanujan tau function, 20
Gcz(n) sum of the ath powers of divisors of n, 20
r modular group, 28
S, T generators of r, 28
R
G
fundamental region of subgroup G of r, 30
R fundamental region of r, 31
l1(r) Dedekind eta function, 47
s(h, k) Dedekind sum, 52
A(X)
log(1  e
21tx
), 52
1\((1, {3, z) Iseki's function, 53
'(s, a)
Hurwitz zeta function, 55
F(x, s) periodic zeta function, 55
j(r) 12
3
J(r), 74
ro(q) congruence subgroup of r, 75
lPl (, + A)
fir)
If,
80
Pl=O P
<1>( r)
86
,
9(r) Jacobi theta function, 91
p(n) partition function, 94
199
Index of special symbols
F(x)
F
n
M
k
Mk,o
~
r(n)
K
E
2k
(r)
F(Z)
Vf(O"O)
'n(S)
200
generating function for p(n),
set of Farey fractions of order n,
linear space of entire forms of weight k,
subspace of cusp forms of weight k,
Hecke operator,
set of transformations of order n,
dim M 2k,O,
normalized Eisenstein series,
Bohr function associated with Dirichlet series,
set of values taken by Dirichlet series f(s) on line 0" = 0"0,
partial sums Lk
s
,
kSn
94
98
117
119
120
122
133
139
168
170
185
A
Abscissa, of absolute convergence, 165
of convergence, 165
Additive number theorv, 1
Apostol, Tom M., 196
Approximation theorem, of Dirichlet, 143
of Kronecker, 148, 150, 154
of Liouville, 146
Asymptotic formula for p(n), 94, 104
Atkin, A. O. L., 91,196
Automorphic function, 79
B
Basis for sequence of exponents, 166
Bernoulli numbers, 132
Bernoulli polynomials, 54
Berwick, W. E. H., 22
Bessel functions, 109
Bohr, Harald, 161, 196
Index
Bohr, equivalence theorem, 178
function of a Dirichlet series, 168
matrix, 167
c
Circle method, 96
Class number of quadratic form, 45
Congruence properties, of coefficients of
j(r), 22, 90
of Dedekind sums, 64
Congruence subgroup, 75
Cusp form, 114
D
Davenport, Harold, 136
Dedekind, Richard, 47
Dedekind function y/(r), 47
Dedekind sums, 52, 61
Deligne, Pierre, 136, 140, 196
201
Index
Differential equation for p(z), 11
Dirichlet, Peter Gustav Lejeune, 143
Dirichlet's approximation theorem, 143
Dirichlet Lfunction, 184
Dirichlet series, 161
Discriminant L\(r), 14
Divisor functions a:x(n), 20
Doubly periodic functions, 2
E
e
t
, e
2
, e
3
, 13
Eigenvalues of Hecke operators, 129
Eisenstein series Gn' 12
recursion formula for, 13
Elliptic functions, 4
Entire modular forms, 114
Equivalence, of general Dirichlet series,
173
of ordinary Dirichlet series, 174
of pairs of periods, 4
of points in the upper halfplane H, 30
of quadratic forms, 45
Estimates for coefficients of modular
forms, 134
Euler, Leonhard, 94
Euler products of Dirichlet series, 136
Exponents of a general Dirichlet
series, 161
F
Farey fractions, 98
Ford, L. R., 99, 196
Ford circles, 99
Fourier coefficients ofj(r), 21,74
divisibility properties of, 22, 74, 91
Functional equation, for r,(r), 48, 52
for 9(r), 91
for '(s), 140
for A(ex, {3, z), 54
for <I>(ex, {3, s), 56, 71
Fundamental pairs of periods, 2
Fundamental region, of modular group
r,31
of subgroup r o(P), 76
202
G
92,93,12
General Dirichlet series, 161
Generators, of modular group r, 28
of congruence subgroup r o(p), 78
Grosswald, Emil, 61, 198
Gupta, Hansraj, Ill, 196
H
Halfplane H, 14
Halfplane, of absolute convergence, 165
of convergence, 165
Hardy, Godfrey Harold, 94, 196
HardyRamanujan formula for p(n), 94
Haselgrove, C. B., 186, 196
Hecke, Erich, 114, 120, 133, 196, 197
Hecke operators Tn' 120
HeIly, Eduard, 179
Helly selection principle, 179
Hurwitz, Adolf, 55, 145
Hurwitz approxiniation theorem, 145
Hurwitz zeta function, 55, 71
I
Invariants 92, 93' 12
Inversion problem for Eisenstein
series, 42
Iseki, Sho, 52, 197
Iseki's transformation formula, 53
J
j(r), J(r), 74,15
Fourier coefficients of, 21
Jacobi, Carl Gustav Jacob, 6, 91, 141
Jacobi theta function, 91, 141.
Jacobi triple product identity, 91
K
Klein, Felix, IS
Klein modular invariant J(r), 15
Kloosterman, H. D., 136
Knopp, Marvin I., 197
Kronecker, Leopold, 148
Kronecker approximation theorem, 148,
150,154
L
Lambert, Johann Heinrich, 24
Lambert series, 24
Landau, Edmund, 186
Lehmer, Derrick Henry, 22, 93, 95, 197
Lehmer conjecture, 22
Lehner, Joseph, 22, 91, Ill,}97
LeVeque, William Judson, 197
Linear space M
k
of entire forms, 118
Linear subspace Mk,o of cusp forms, 119
Liouville, Joseph, 5, 146, 184
Liouville approximation theorem, 146
Liouville function ).. (n), 25, 184
Liouville numbers, 147
Littlewood, John Edensor, 95
M
Mapping properties of J(r), 40
Mediant,98
Mellin, Robert Hjalmar, 54
Mellin inversion formula, 54
Mobius, Augustus Ferdinand, 24, 27, 187
Mobius function, 24, 187
Mobius transformation, 27
Modular forms, 114
and Dirichlet series, 136
Modular function, 34
Modular group r, 28
subgroups of, 46, 75
Montgomery, H. L., 187
Mordell, Louis Joel, 92, 197
Multiplicative property, of coefficients of
entire forms, 130
of Hecke operators, 126, 127
of Ramanujan tau function, 93, 114
Index
N
Neville, Eric Harold, 110, 197
Newman, Morris, 91, Ill, 197
Normalized eigenform, 130
o
O'Brien, J. N., 91, 196
Order of an elliptic function, 6
p
pfunction of Weierstrass, 10
Partition function p(n), 1, 94
Period, 1
Period parallelogram, 2
Periodic zeta function, 55
Petersson, Hans, 22, 133, 140, 197
Petersson inner product, 133
PeterssonRamanujan conjecture, 140
Picard, Charles Emile, 43
Picard's theorem, 43
Product representation for L\(r), 51
Q
Quadratic forms, 45
R
Rademacher, Hans, 22, 62, 95, 102,
104, 197
Rademacher path of integration, 102
Rademacher series for p(n), 104
Ramanujan, Srinivasa, 20, 92, 94,
136,191
Ramanujan conjecture, 136
Ramanujan tau function, 20, 22, 92,
113, 131, 198
Rankin, Robert A., 136, 198
Reciprocity law for Dedekind sums, 62
Representative of quadratic form, 45
203
Index
Riemann, Georg Friedrich Bernhard,
140, 155, 185, 198
Riemann zeta function, 20, 140, 155,
]85,189
Rouche, Eugene, 180
Rouche's theorem, 180
s
Salie, Hans, 136
Schoeneberg, Bruno, 198
Sczech, R., 61, 198
Selberg, Atle, 136, 198
Serre, JeanPierre, 198
Siegel, Carl Ludwig, 48, 198
Simultaneous eigenforms, 130
Spitzenform, 114
Subgroups of the modular groups, 46, 75
T
Tau function, 20, 22, 92, 113, 131
Theta function, 91,141
Transcendental numbers, 147
Transformation of order n, 122
Transformation formula, of Dedekind,
48, 52
of Iseki, 54
Tunln, Paul, 185, 198
Tunin's theorem, 185, 186
u
Univalent modular function, 84
Uspensky, J. V., 94, 198
v
Valence of a modular function, 84
Van Wijngaarden, A., 22
Values, of J(r), 39
of Dirichlet series, 170
Vertices of fundamental region, 34
204
w
Watson, G. N., 109, 198
Weierstrass, Karl, 6
Weierstrass pfunction, 10
Weight of a modular form, 114
Weight formula for zeros of an entire
form, 115
Whiteman, Albert Leon, 62, 198
z
Zeros, of an elliptic function, 5
Zeta function, Hurwitz, 55
periodic, 55
Riemann, 140, 155, 185,189
Zuckerman, Herbert S., 22
Graduate Texts in Mathematics
com"",(',/ from P ( / ~ ( ' "
48 SACHS/WU. General Relativity for Mathematicians.
49 GRUENBERG/WEIR. Linear Geometry. 2nd ed.
50 EDWARDS. Fermat's Last Theorem.
51 KLINGENBERG. A Course in Differential Geometry.
52 HARTSHORNE. Algebraic Geometry.
53 MANIN. A Course in Mathematical Logic.
54 GRAVER/WATKINS. Combinatorics with Emphasis on the Theory of Graphs.
55 BROWN/PEARCY. Introduction to Operator Theory I: Elements of Functional Analysis.
56 MASSEY. Algebraic Topology: An Introduction.
57 CROWEUJFox. Introduction to Knot Theory.
58 KOBLITZ. padic Numbers, padic Analysis, and ZetaFunctions. 2nd ed.
59 LANG. Cyclotomic Fields.
60 ARNOLD. Mathematical Methods in Classical Mechanics. 2nd ed.
61 WHITEHEAD. Elements of Homotopy Theory.
62 KARGAPOLOV/MERZIJAKOV. Fundamentals of the Theory of Groups.
63 BOLLABAs. Graph Theory.
64 EDWARDS. Fourier Series. Vol. I. 2nd ed.
65 WELLS. Differential Analysis on Complex Manifolds. 2nd ed.
66 WATERHOUSE. Introduction to Affine Group Schemes.
67 SERRE. Local Fields.
68 WEIDMANN. Linear Operators in Hilbert Spaces.
69 LANG. Cyclotomic Fields II.
70 MASSEY. Singular Homology Theory.
71 FARKAS/KRA. Riemann Surfaces.
72 STILLWELL. Classical Topology and Combinatorial Group Theory.
73 HUNGERFORD. Algebra.
74 DAVENPORT. Multiplicative Number Theory. 2nd ed.
75 HOCHSCHILD. Basic Theory of Algebraic Groups and Lie Algebras.
76 lITAKA. Algebraic Geometry.
77 HECKE. Lectures on the Theory of Algebraic Numbers.
78 BURRIS/SANKAPPANAVAR. A Course in Universal Algebra.
79 WALTERS. An Introduction to Ergodic Theory.
80 ROBINSON. A Course in the Theory of Groups.
81 FORSTER. Lectures on Riemann Surfaces.
82 BOTT/Tu. Differential Forms in Algebraic Topology.
83 WASHINGTON. Introduction to Cyclotomic Fields.
84 IRELAND/RoSEN. A Classical Introduction to Modem Number Theory.
85 EDWARDS. Fourier Series: Vol. II. 2nd ed.
86 VAN LINT. Introduction to Coding Theory.
87 BROWN. Cohomology of Groups.
88 PIERCE. Associative Algebras.
89 LANG. Introduction to Algrebraic and Abelian Functions. 2nd ed.
90 BR0NDSTED. An Introduction to Convex Polytopes.
91 BEARDON. On the Geometry of Discrete Groups.
92 DIESTEL. Sequences and Series in Banach Spaces.
93 DUBROVIN/FoMENKO/NoVIKOV. Modem GeometryMethods and Applications Vol. 1.
94 WARNER. Foundations of Differentiable Manifolds and Lie Groups.
95 SHIRYAYEV. Probability, Statistics, and Random Processes.
96 CONWAY. A Course in Functional Analysis.
97 KOBLITZ. Introduction to Elliptic Curves and Modular Forms.
98 BROCKER/TOM DIECK. Representations of Compact Lie Groups.
99 GRovE/BENSON. Finite Reflection Groups. 2nd ed.
100 BERG/CHRISTENSEN/RESSEL. Harmonic Anaylsis on Semigroups: Theory of Positive Definite
and Related Functions.
101 EDWARDS. Galois Theory.
102 VARADARAJAN. Lie Groups, Lie Algebras and Their Representations.
103 LANG. Complex Analysis. 2nd ed.
104 DUBROVIN/FoMENKO/NoVIKOV. Modem GeometryMethods and Applications Vol. II.
105 LANG. SLiR).
106 SILVERMAN. The Arithmetic of Elliptic Curves.
107 OLVER. Applications of Lie Groups to Differential Equations.
108 RANGE. Holomorphic Functions and Integral Representations in Several Complex Variables.
109 LEHTO. Univalent Functions and Teichmiiller Spaces.
110 LANG. Algebraic Number Theory.
III HUSEMOLLER. Elliptic Curves.
112 LANG. Elliptic Functions.
113 KARATZAS/SHREVE. Brownian Motion and Stochastic Calculus.
114 KOBLITZ. A Course in Number Theory and Cryptography.
115 BERGERIGosTIAUX. Differential Geometry: Manifolds, Curves, and Surfaces.
116 KELLEy/SRINIVASAN. Measure and Integral, Volume 1.
117 SERRE. Algebraic Groups and Class Fields.
118 PEDERSEN. Analysis Now.
119 ROTlAAN. An Introduction to Algebraic Topology.
120 ZIEMER. Weakly Differentiable Functions: Sobolev Spaces and Functions of Bounded
Variation.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.