Springer Monographs in Mathematics

Tsuneo Arakawa
Tomoyoshi Ibukiyama
Masanobu Kaneko

Bernoulli
Numbers
and Zeta
Functions
Springer Monographs in Mathematics

For further volumes:

http://www.springer.com/series/3733
Tsuneo Arakawa • Tomoyoshi Ibukiyama
Masanobu Kaneko

Bernoulli Numbers and Zeta

Functions

with an appendix by Don Zagier

123
Tsuneo Arakawa Tomoyoshi Ibukiyama (emeritus)
(deceased 2003) Osaka University
Osaka, Japan
Masanobu Kaneko
Kyushu University Don Zagier (Appendix)
Fukuoka, Japan Max Planck Institute for Mathematics
Bonn, Germany

ISSN 1439-7382 ISSN 2196-9922 (electronic)

ISBN 978-4-431-54918-5 ISBN 978-4-431-54919-2 (eBook)
DOI 10.1007/978-4-431-54919-2
Springer Tokyo Heidelberg New York Dordrecht London
Library of Congress Control Number: 2014938983

Mathematics Subject Classification: 11B68, 11B73, 11M06, 11L03, 11M06, 11M32, 11M35

© Springer Japan 2014

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Preface

Two subjects are treated in this book. The main subject is the theory of Bernoulli
numbers, which are a series of rational numbers that appear in various contexts of
mathematics, and the other subject is the related theory of zeta functions, which are
very important in number theory. We hope that these are enjoyable subjects both for
amateur mathematics lovers and for professional researchers. There are easy parts
as well as difficult parts in this book, but we believe that, according to the taste of
the readers, they can enjoy at least some parts of this attractive mathematics. Since
the logical relations between the chapters are rather loose and not a straight course
from the beginning to the end, it would still be worthwhile for readers who pick
only some chapters which fit their tastes and background knowledge. As far as we
know, books whose main subject is Bernoulli numbers are rare. Some parts of this
book consist of rather standard number theory, but our expositions on these subjects
are not always so standard, and some parts are completely new and have not been
written in any reference before.
Now, there are many numbers in the world which have names, but what is a
Bernoulli number? In high-school we learn the formula for the sum of the integers
from 1 to n. Or maybe the formula of the sum of squares 12 C 22 C    C n2 is
also written in the standard textbooks for high schools. But if we make the powers
bigger, for example, if we take

16 C 26 C    C n6 ;

then what kind of formula can we have? It seems that many people hit upon this kind
of question and it often happens that ambitious high-school students try and find the
solution. In fact, the formula was already known for any power at the beginning
of the eighteenth century. (See Chap. 1, p. 2.) Bernoulli introduced some special
numbers to express this formula. These are the Bernoulli numbers. (For a more
precise history, see Chap. 1.) In this book, this formula of the sum of powers appears
in various places. In fact, in this book, four alternative proofs of this formula will be
given in different places (cf. Sects. 1.2, 4.3, 5.2, 8.3).

v
vi Preface

Well, is it so interesting to give this formula of the sum of powers? Indeed it is.
But if the meaning of Bernoulli numbers were only this, we would not have written
this book. The real meaning of Bernoulli numbers lies in a different place. Let’s
look at the following formulas:

1 1 1 1 1 2
1C 2
C 2 C 2 C 2 C 2 C D ;
2 3 4 5 6 6
1 1 1 1 1 4
1C 4
C 4 C 4 C 4 C 4 C D ;
2 3 4 5 6 90
1 1 1 1 1 1 
1 C  C  C  D :
3 5 7 9 11 13 4
(You can find the proofs in Corollary 4.12, p. 61, and Theorem 9.6, p. 148.) At
first glance, these kinds of formulas could give you a mysterious impression, since
a regular sum of simple rational numbers suddenly changes into a transcendental
number like . The left-hand sides of these formulas are values at some special
points of so-called zeta functions, or L-functions, which are another important
subject of the book. An example of zeta functions is given by

1 1 1 1 1
.s/ D 1 C C s C s C s C s C :
2s 3 4 5 6
Here the sum is regarded as a function with respect to the complex variable s. This
function is called the Riemann zeta function. The left-hand sides of the first two
formulas above are the values of .s/ at s D 2 and s D 4, namely .2/ and .4/. The
right-hand sides are related to Bernoulli numbers and generalized Bernoulli numbers
(see below). The true reason that Bernoulli numbers are indispensable for number
theory lies in this type of relation with special values (values at special points) of
zeta functions. A huge number of functions called zeta functions appear in number
theory, and they always contain very important information. We can even say that
there appears a different zeta function for each important number theoretical object.
The zeta functions which are related to Bernoulli numbers are just a part of them.
But still they have something to do with many number theoretical phenomena. For
example, they are connected with a kind of volume in non-Euclidean geometry, with
dimension formulas of spaces of modular forms, and with the number of equivalence
classes of quadratic forms (class number formula), and they appear as constant terms
of Eisenstein series, which are important in the theory of automorphic forms. Also
they are related to the number theory of cyclotomic fields, that is, the systems of
numbers obtained by adjoining the roots of unity to the rational numbers. If we
wanted to explain all of this, we would need several more books, so it is not possible
to do it here, but we have tried to give some hints about those relations.
Now, for the reader’s convenience, we sketch the content of each chapter.
Comments below written in parentheses will not be explained in this book in detail.
Preface vii

In Chap. 1, after explaining the history of how Bernoulli numbers were intro-
duced and giving the definition of Bernoulli numbers using recurrence relations,
we prove the formula for the sum of powers of consecutive natural numbers.
In Chap. 2, we give a formula to write down Bernoulli numbers in a simple
way. There we use Stirling numbers, whose definition is comparatively simple. In
Chap. 3, we will prove the Clausen–von Staudt theorem which gives information
on the denominators of Bernoulli numbers, and also Kummer’s congruence which
gives a congruence relation between Bernoulli numbers. (This latter theorem is
needed when we define p-adic zeta functions.) In Chap. 4, we define generalized
Bernoulli numbers associated to Dirichlet characters. (Through this extension,
the connection with quadratic fields and cyclotomic fields will appear.) Also the
Bernoulli polynomials (a system of polynomials similar to the generating function
of Bernoulli numbers) are defined, and as an application, we give a second proof
of the formula of the sum of powers. The formula expressing the values of the
Riemann zeta function at even positive integers by Bernoulli numbers is also given
here. In Chap. 5, we give the Euler–Maclaurin summation formula, which is very
useful to evaluate sum of values of functions at integers. In the coefficients of this
formula, Bernoulli numbers or Bernoulli polynomials appear. As an application, the
analytic continuation of the Riemann zeta function to the whole plane is proved
and the values of the Riemann zeta function at non-positive integers are beautifully
expressed by Bernoulli numbers. In Chap. 6, the relation between ideals of a
quadratic field and quadratic forms is given and a relation between the generalized
Bernoulli numbers and the class numbers of positive definite quadratic forms is
explained. In Chap. 7, a congruence relation between the class number of imaginary
quadratic fields and the Bernoulli numbers will be given. In Chap. 8, Gauss sums
are introduced and then formulas to describe various sums of the roots of unity with
characters by Gauss sums and generalized Bernoulli numbers are given. Here we
get a third proof of the formula for sums of powers. In Chap. 9, L-functions will
be introduced by modifying the usual definition of the Riemann zeta function by
adjoining characters; then their functional equation is shown by using a contour
integral expression. We will also give a formula to describe values at non-positive
integers of L-functions in terms of generalized Bernoulli numbers. This formula
gives a “raison d’être” for the generalized Bernoulli numbers. In Chap. 10, among
the zeta functions of prehomogeneous vector spaces, we pick an example which
apparently does not seem so simple but can be shown to be a very easy function.
(This is an example connected with the appearance of Bernoulli numbers in the
dimension formula for modular forms.) In Chap. 11, we give an alternative proof of
Kummer’s congruence by considering p-adic measures. (This has a deep connection
with p-adic L-functions.) In Chap. 12, we review the fact that Bernoulli numbers
appear in the Taylor expansion of the (co)tangent function, and ask what would
emerge if we replace the (co)tangent with elliptic functions (a well-worn device
to generalize phenomena on trigonometric functions), and explain the theory of
Hurwitz numbers. In Chap. 13, we show the analytic continuation of the Barnes
multiple zeta function, which is a generalization of the Hurwitz zeta function, and
explain relations between their special values and Bernoulli polynomials. We also
viii Preface

show the functional equation of the double zeta functions in the same way as in
Chap. 9. In Chap. 14, we define poly-Bernoulli numbers, which are obtained by
replacing the usual generating function of Bernoulli numbers by polylogarithm
functions. In an Appendix by Don Zagier, some “curious and exotic” identities for
Bernoulli numbers are given.
As explained above, Bernoulli numbers are related with a great many things.
But Bernoulli numbers do not have so many of their own problems to be solved,
except for a few very difficult problems. So, maybe there exist few mathematicians
who study only Bernoulli numbers. Since the situation is like that, there are not
many books similar to this one, and it was not so easy to have a perspective on
the wide range of relations for Bernoulli numbers. Although the present authors
are specialists of number theory, we dare not say we are specialists of Bernoulli
numbers. This book consists partly of quotations from the papers by authors
who occasionally encountered Bernoulli numbers from their different interests,
and partly of various memoranda (namely these are observations which are not
necessarily very original results worthy to be written as careful scientific papers,
but rather lightly checked, interestingly interpreted, or calculated for fun and worth
being taken note of so as not to be forgotten). And this kind of connection is
indeed the usual common connection of mathematicians to the Bernoulli numbers.
In this sense, we believe that this book is useful for any reader who is interested in
mathematics.
The original edition of this book was published in Japanese by Makino Publisher
in Japan in 2001. We would like to give our sincere thanks to Professor Don B.
Zagier for recommending us enthusiastically to write this English version based
on that Japanese book, for recommending this book to Springer-Verlag to be
published, and also for writing the nice Appendix. We would like to thank Springer-
Verlag for publishing this book, in particular Dr. Joachim Heinze of Springer
and Ms. Chino Hasebe and Mr. Masayuki Nakamura of Springer Tokyo for their
valuable help for the arrangement, and Dr. Alexander Weisse for his help with TeX.
We would also like to thank anonymous referees of the English version for a lot
of valuable comments. We would like to give our deep thanks to Mr. Suenobu
Makino, the owner of Makino Publishers, who originally suggested us the theme
of Bernoulli numbers and recommended to write the book in Japanese. Without his
recommendation, the Japanese version would not have appeared. Tsuneo Arakawa,
one of the co-authors for the Japanese version, passed away on October 3, 2003 at
the age of 54 before this English project started. Since some parts of this book are
based on his Japanese manuscript, we would like to keep his name as one of the
authors. The second- and the third-named authors express their profound regret and
sadness over the early death of Tsuneo Arakawa.

November 2013 The Authors

Contents

1 Bernoulli Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.1 Definitions: Introduction from History . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.2 Sums of Consecutive Powers of Integers
and Theorem of Faulhaber .. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 6
1.3 Formal Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 13
1.4 The Generating Function of Bernoulli Numbers . . . . . . . . . . . . . . . . . . . 20
2 Stirling Numbers and Bernoulli Numbers . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
2.1 Stirling Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
2.2 Formulas for the Bernoulli Numbers Involving
the Stirling Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 34
3 Theorem of Clausen and von Staudt, and Kummer’s
Congruence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41
3.1 Theorem of Clausen and von Staudt . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41
3.2 Kummer’s Congruence .. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
3.3 Short Biographies of Clausen, von Staudt and Kummer . . . . . . . . . . . 46
4 Generalized Bernoulli Numbers . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 51
4.1 Dirichlet Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 51
4.2 Generalized Bernoulli Numbers . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 53
4.3 Bernoulli Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 55
5 The Euler–Maclaurin Summation Formula and the
Riemann Zeta Function .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 65
5.1 Euler–Maclaurin Summation Formula . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 65
5.2 The Riemann Zeta Function . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 67
6 Quadratic Forms and Ideal Theory of Quadratic Fields .. . . . . . . . . . . . . . 75
6.1 Quadratic Forms .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 75
6.2 Orders of Quadratic Fields . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 77
6.3 Class Number Formula of Quadratic Forms . . . .. . . . . . . . . . . . . . . . . . . . 87

ix
x Contents

7 Congruence Between Bernoulli Numbers

and Class Numbers of Imaginary Quadratic Fields . . . . . . . . . . . . . . . . . . . 95
7.1 Congruence Between Bernoulli Numbers and Class Numbers . . . . 95
7.2 “Hurwitz-integral” Series . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 97
7.3 Proof of Theorem 7.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 99
8 Character Sums and Bernoulli Numbers . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 103
8.1 Simplest Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 104
8.2 Gaussian Sum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 107
8.3 Exponential Sums and Generalized Bernoulli Numbers . . . . . . . . . . . 110
8.4 Various Examples of Sums . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 118
8.5 Sporadic Examples: Using Functions . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 121
8.6 Sporadic Examples: Using the Symmetry.. . . . . .. . . . . . . . . . . . . . . . . . . . 122
8.7 Sporadic Example: Symmetrize Asymmetry . . .. . . . . . . . . . . . . . . . . . . . 127
8.8 Quadratic Polynomials and Character Sums . . . .. . . . . . . . . . . . . . . . . . . . 132
8.9 A Sum with Quadratic Conditions . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 133
9 Special Values and Complex Integral Representation
of L-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 139
9.1 The Hurwitz Zeta Function . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 139
9.2 Contour Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 141
9.3 The Functional Equation of .s; a/ . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 145
9.4 Special Values of L-Functions and the Functional Equations . . . . . 148
10 Class Number Formula and an Easy Zeta Function
of the Space of Quadratic Forms . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 155
10.1 Ideal Class Groups of Quadratic Fields . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 155
10.2 Proof of the Class Number Formula of Imaginary
Quadratic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 162
10.3 Some L-Functions Associated with Quadratic Forms .. . . . . . . . . . . . . 173
11 p-adic Measure and Kummer’s Congruence . . . . . . . .. . . . . . . . . . . . . . . . . . . . 183
11.1 Measure on the Ring of p-adic Integers and the Ring
of Formal Power Series . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 183
11.2 Bernoulli Measure .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 196
11.3 Kummer’s Congruence Revisited. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 198
12 Hurwitz Numbers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 203
12.1 Hurwitz Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 203
12.2 A Short Biography of Hurwitz.. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 207
13 The Barnes Multiple Zeta Function . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 209
13.1 Special Values of Multiple Zeta Functions
and Bernoulli Polynomials .. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 210
13.2 The Double Zeta Functions and Dirichlet Series . . . . . . . . . . . . . . . . . . . 212
13.3 .s; ˛/ and Continued Fractions .. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 218
Contents xi

14 Poly-Bernoulli Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 223

14.1 Poly-Bernoulli Numbers . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 223
14.2 Theorem of Clausen and von Staudt Type . . . . . .. . . . . . . . . . . . . . . . . . . . 227
14.3 Poly-Bernoulli Numbers with Negative Upper Indices.. . . . . . . . . . . . 233
Appendix (by Don Zagier): Curious and Exotic Identities for
Bernoulli Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 239
A.1 The “Other” Generating Function(s) for the Bernoulli
Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 240
A.2 An Application: Periodicity of Modified Bernoulli Numbers .. . . . . 244
A.3 Miki’s Identity .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 246
A.4 Products and Scalar Products of Bernoulli Polynomials . . . . . . . . . . . 249
A.5 Continued Fraction Expansions for Generating
Functions of Bernoulli Numbers . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 256

References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 263

Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 269
Chapter 1
Bernoulli Numbers

1.1 Definitions: Introduction from History

In his posthumous book Ars Conjectandi [16] published in 1713 (the law of large
numbers in probability theory is stated in this book), Jakob Bernoulli1 introduced
the Bernoulli numbers in connection to the study of the sums of powers of
consecutive integers 1k C 2k C    C nk . After listing the formulas for the sums
of powers

X  
n.n C 1/ X 2 n.n C 1/.2n C 1/ X 3
n n n
n.n C 1/ 2
iD ; i D ; i D ;:::
i D1
2 i D1
6 i D1
2

up to k D 10 (Bernoulli expresses the right-hand side without factoring), he gives

a general formula involving the numbers which are known today as Bernoulli
numbers. Bernoulli then explains how these numbers are determined inductively,
and emphasizes how his formula ((1.1) below) is useful for computing the sum
of powers. He claims that he did not take “a half of a quarter of an hour” to
compute the sum of tenth powers of 1 to 1;000, which he computed correctly as
91409924241424243424241924242500.
Using modern notation, his formula is written as
! 0 ! 1
X
n X k
k nkC1j @ 1 X kC1
k
kC1j A
i D k
Bj D Bj n ; (1.1)
i D1 j D0
j kC1j k C 1 j D0 j

1
Born on December 27, 1654 in Basel, Switzerland—died on August 16, 1705 in Basel,
Switzerland. Jakob is the eldest among the mathematicians in the famous Bernoulli family. It
is said that Jakob, his younger brother Johann (born on July 27, 1667 in Basel, Switzerland—
died on January 1, 1748 in Basel, Switzerland), and his second son, Daniel (born on February 8,
1700 in Groningen, Netherlands—died on March 17, 1782 in Basel, Switzerland) are the most
distinguished among them.

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 1

in Mathematics, DOI 10.1007/978-4-431-54919-2__1, © Springer Japan 2014
2 1 Bernoulli Numbers

k 
where j
is the binomial coefficient
!
k k.k  1/.k  2/    .k  j C 1/
D ;
j jŠ

and Bj is the number determined by the recurrence formula

!
X k
kC1
Bj D k C 1; k D 0; 1; 2; : : : :
j D0
j

It is this Bj that is subsequently called a Bernoulli number,2 and is the main theme
of this book.
Bernoulli does not give symbols for B0 and B1 , and he writes A; B; C; D; : : :
for B2n . As we will see later, B2nC1 D 0 except for B1 . In his book, the right-hand
side of (1.1) is written as

1 1 c c:c  1:c  2 c3

ncC1 C nc C Anc1 C Bn C
cC1 2 2 2:3:4
c:c  1:c  2:c  3:c  4 c:c  1:c  2:c  3:c  4:c  5:c  6
C nc5 C Dnc7
2:3:4:5:6 2:3:4:5:6:7:8
C ;

where c D k.
We would like to mention here that, in the book called Katsuyo Sanpo (“essentials
of the art of calculation”) by the outstanding Japanese mathematician Takakazu
Seki,3 published also posthumously, in 1712 (and thus 1 year before Bernoulli!), the
formula for the sums of powers and the inductive definition of the Bernoulli numbers
are given. His formula and definition are completely the same as Bernoulli’s.

2
It was apparently de Moivre (Abraham, born on May 26, 1667 in Vitry-le-Francois, Champagne,
France—died on November 27, 1754 in London, England) who first called this number a Bernoulli
number in the book Miscellanea analytica de seriebus et quadraturis (London, 1730). De Moivre
is famous for de Moivre’s formula in trigonometry.
3
Born in 1642(?) in Kohzuke(?), Japan—died on October 24, 1708 in Edo, Japan. He is usually
considered Japan’s greatest mathematician of the Edo period (1600–1857, when the country
was closed to essentially all foreign contact). Hardly anything is known about his mathematical
education, and he seems to have been largely self-taught. He served under the shoguns Tsunashige
Tokugawa and Ienobu Tokugawa, occupying the post of Controller of the Treasury Office, and
wrote several treatises in higher mathematics. Apart from discovering Bernoulli numbers simul-
taneously with or before Bernoulli, he discovered determinants and the rules for calculating them
simultaneously with or before Leibniz, solved extraordinarily difficult problems of elimination
theory for systems of polynomial equations in many variables, and began a study of calculation
procedures for the arc of a circle that was continued and completed by his disciple Katahiro Takebe
with the discovery of infinite series expansions for various trigonometric functions.
1.1 Definitions: Introduction from History 3

Seki refers to B0 ; B1 ; B2 ; : : : as Shusuu (which means “numbers to be taken”) of the

first order, second order etc., labeling odd indexed B2nC1 as well. It is not widely
known that Seki independently found the Bernoulli numbers, but the collected works
of Seki [84] have been published and the volume contains an English translation
of each article. A reproduction of Seki’s table giving the formula for the sums of
powers in terms of binomial coefficients and “Seki–Bernoulli numbers”, together
with a translation into modern notation, is given as Figs. 1.1 and 1.2.
Neither Seki nor Bernoulli explains in much detail how to deduce the formula
(1.1). Prior to their work, the formula for the sums of powers was discussed in
4
Academia Algebrae Pnby Faulhaber. Pnsaying that, when k is
He obtained the result
odd, the quantity i D1 i is a polynomial in n.nC1/
k
D i D1 i , (the first example
Pn 3 Pn 2 2
Pn
being i D1 i D i D1 i ), whereas when k is even,
k
i D1 i is divisible
n.nC1/.2nC1/ P n
PD
2
by 6 i D1 i as polynomials in n, and the quotient is again a
polynomial in niD1 i . But apparently he did not reach the Bernoulli numbers.
These facts discovered by Faulhaber were rediscovered by Jacobi5 [52], who gave
a rigorous proof. We shall provide a proof of the formula (1.1) in the next section.
As for the results of Faulhaber, we only formulate them and do not give proofs,
leaving them to the literature. We mention here that a very extensive bibliography
on Bernoulli numbers, compiled by Karl Dilcher, is available online [28].
Let us now give the definition of the Bernoulli numbers again. We follow Seki
and Bernoulli, and define them using a recurrence formula. The Bernoulli numbers
may also be defined by using a generating function. We will see in the next section
that these definitions are equivalent.
Definition 1.1 (Bernoulli numbers). Define Bn .n D 0; 1; 2; : : : / inductively by
the formula
!
X n
nC1
Bi D n C 1 .n D 0; 1; 2; : : : /: (1.2)
i D0
i

Let us compute Bn for small n. For n D 0 we have

B0 D 1:

Putting n D 1 in the formula (1.2) we have

B0 C 2B1 D 2;

from which we obtain

1 1
B1 D .2  B0 / D :
2 2

4
Johann Faulhaber (born on May 5, 1580 in Ulm, Germany—died in 1635 in Ulm, Germany).
5
Carl Gustav Jacob Jacobi (born on December 10, 1804 in Potsdam, Prussia (now Germany)—died
on February 18, 1851 in Berlin, Germany).
4 1 Bernoulli Numbers

Fig. 1.1 Katsuyou Sanpou, Seki’s tabular presentation of the power-sum formula, written not in
Chinese characters but using the notation of Sangi or counting rods. The table (which is reproduced
here with a 90ı rotation) contains each of the ingredients of the formula in parentheses in Eq. (1.1):
The number designated by “power” in the translation overleaf is the k of this formula, the binomial
coefficients are tabulated in the “Pascal’s triangle” in the right part of the table (D left part in the
translation), the Bernoulli numbers Bj are given on the left (the right in the English translation),
 
the missing coefficient kC1
kC1
in (1.1) corresponds to the crossed out “1” at the beginning of each
column of the Pascal triangle, and the “denominator” is the number k C 1 by which the whole
expression has to be divided at the end of the calculation
5

1
1 12 level 12 void
1 11 66 level 11 take 5/66 and add
1 10 55 220 level 10 void
1 9 45 165 495 level 9 take 1/30 and subtract
1 8 36 120 330 792 level 8 void
1 7 28 84 210 462 992e level 7 take 1/42 and add
1 6 21 56 126 252 462 792 level 6 void
1 5 15 35 70 126 210 330 495 level 5 take 1/30 and subtract
1 4 10 20 35 56 84 120 165 220 level 4 void

.4  B0  4B1  6B2 / D 0:
1a 3 6 10 15 21 28 36 45 55 66 level 3 take 1/6 and add

B0 C 4B1 C 6B2 C 4B3 D 4;

.3  B0  3B1 / D :
1
6
1b 2 3 4 5 6 7 8 9 10 11 12 level 2 take 1/2 and add

B0 C 3B1 C 3B2 D 3;
1 1 1 1 1 1 1 1 1 1 1 1 level 1 all
0 1 2 3 4 5 6 7 8 9 10 11 power
1c 2d 4 5 6 7 8 9 10 11 12 denominator

Similarly, putting n D 3 in (1.2), we have

a: The crossed-out numbers are to be omitted.
b: This number should also be crossed out.
1.1 Definitions: Introduction from History

c: This number should be 2.

1
3
d: This number should be 3.

B2 D

1
4
Putting n D 2 in (1.2) we have
e: This number should be 924.

B3 D
(English translation, reflected
with respect to the original)
Fig. 1.2 Katsuyou Sanpou

from which we obtain

which gives
6 1 Bernoulli Numbers

Table 1.1 Bernoulli numbers

n 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Bn 1 12 1
6
0  1
30
0 1
42
0  30
1 5
0 66 0  2730
691
0 7
6
0  3617
510
0 43867
798

Putting n D 4 in (1.2), we have

B0 C 5B1 C 10B2 C 10B3 C 5B4 D 5;

which gives

1 1
B4 D .5  B0  5B1  10B2  10B3 / D  ;
5 30
and so on. Table 1.1 is the list of Bernoulli numbers Bn up to n D 18. It follows from
the definition that Bn is a rational number. Mathematical software packages such as
Mathematica and Maple include the algorithm for computing Bernoulli numbers,
and it is not hard to obtain several hundred Bn ’s in a moment.
Remark 1.2. There is another convention on Bernoulli numbers: B1 D  12 and
everything else is the same. We adopt the first definition because this is the original
definition of Seki and Bernoulli for one thing, and it is better suited to the special
values of the Riemann zeta function for another, that is, the formula in Theorem 5.4
is valid also for m D 1. Although the difference is minuscule, we call readers’
attention to it.
Since Bn D 0 for every odd n greater than 1 (we prove this later), it is not difficult
to translate formulas from one definition to the other by replacing Bn by .1/n Bn .

1.2 Sums of Consecutive Powers of Integers

and Theorem of Faulhaber

In this section we first prove the Seki–Bernoulli formula (1.1) for the sum

X
n
Sk .n/ WD ik (k; n integers, k  0; n  1):
i D1

Then we will state Faulhaber’s theorem and give a recurrence formula for Bernoulli
numbers at the end of this section.
Before going into the proof, let us give several explicit formulas for Sk .n/:

S0 .n/ D n;
n2 n
S1 .n/ D C
2 2
1.2 Sums of Consecutive Powers of Integers and Theorem of Faulhaber 7

1
D n.n C 1/;
2
n3 n2 n
S2 .n/ D C C
3 2 6
1
D n.n C 1/.2n C 1/;
6
n4 n3 n2
S3 .n/ D C C
4 2 4
1 2
D n .n C 1/2 ;
4
n5 n4 n3 n
S4 .n/ D C C 
5 2 3 30
1
D n.n C 1/.2n C 1/.3n2 C 3n  1/;
30
n6 n5 5n4 n2
S5 .n/ D C C 
6 2 12 12
1 2
D n .n C 1/2 .2n2 C 2n  1/;
12
n7 n6 n5 n3 n
S6 .n/ D C C  C
7 2 2 6 42
1
D n.n C 1/.2n C 1/.3n4 C 6n3  3n C 1/:
42

For the proof, we first see immediately that

S0 .n/ D n:

Let k  1. From the binomial theorem we obtain

!
X k
kC1 j
.m C 1/kC1  mkC1 D m :
j D0
j

Putting m D 1; 2; : : : ; n, and summing over all m, we obtain

!
X k
k C 1
.n C 1/kC1  1 D Sj .n/:
j D0
j
8 1 Bernoulli Numbers

From this we have

⎧ ! ⎫
1 ⎨ X ⎬
k1
k C 1
Sk .n/ D .n C 1/kC1  1  Sj .n/ :
kC1⎩ j D0
j ⎭

Putting k D 1; 2; : : : in this formula, we obtain

n2 n n3 n2 n
S1 .n/ D C ; S2 .n/ D C C ;::::
2 2 3 2 6
By induction we see that
1
Sk .n/ is a polynomial of degree k C 1 in n, whose leading term is nkC1 .
kC1
Our objective is to write down this polynomial explicitly.
Consider the polynomial Sk .x/.D kC1 1
x kC1 C    / obtained by replacing
the variable n by x. In general, two polynomials f .x/; g.x/ are identical if
f .n/ D g.n/ for all positive integers n. By definition we have

Sk .n C 1/  Sk .n/ D .n C 1/k .n D 1; 2; 3; : : : /:

Thus, we obtain

Sk .x C 1/  Sk .x/ D .x C 1/k :

Putting x D 0, and using Sk .1/ D 1, we have

Sk .0/ D 0:

This yields the constant term of Sk .x/. The other coefficients may be obtained
.j /
from the derivatives Sk .0/; 1  j  k. Taking the derivative of the relation
Sk .x C 1/  Sk .x/ D .x C 1/k , we obtain

Sk0 .x C 1/  Sk0 .x/ D k.x C 1/k1 : (1.3)

Putting x D 0; 1; 2; : : : ; n  1, and adding them all, we have

Sk0 .n/  Sk0 .0/ D kSk1 .n/:

This holds for any positive integer n. If we let Sk0 .0/ D bk (b0 D 1), then we have
the relation

Sk0 .x/ D kSk1 .x/ C bk :

1.2 Sums of Consecutive Powers of Integers and Theorem of Faulhaber 9

Taking the derivatives of both sides, we have

Sk00 .x/ D kSk1

0
.x/: (1.4)

Putting x D 0, we have

Sk00 .0/ D kbk1 :

Taking the derivative of (1.4) once again, and using (1.4), we have
Sk000 .x/ D kSk1
00 0
.x/ D k.k  1/Sk2 .x/:

Putting x D 0, we have
Sk000 .0/ D k.k  1/bk2 :
Similarly, taking the derivatives successively, we obtain
.j /
Sk .0/ D k.k  1/    .k  j C 2/bkj C1 .2  j  k C 1/:

Finally, we have

X
kC1 .j /
Sk .0/ j
Sk .x/ D x
j D0
jŠ
!
X
kC1
1 kC1 .0/
D bkj C1 x j .Sk .0/ D 0/
j D1
k C 1 j
!
1 X kC1
k
D bj x kC1j :
k C 1 j D0 j

Since Sk .1/ D 1, we obtain, by putting x D 1 in the above formula, the recurrence

formula
!
X k
kC1
kC1D bj ;
j D0
j

which is nothing but the recurrence for the Bernoulli numbers. We therefore
conclude bj D Bj . In view of the identity
! !
1 kC1 1 k
D ;
kC1 j kC1j j

we obtain (1.1). t
u
10 1 Bernoulli Numbers

Remark 1.3. If we set Sk0 .x  1/ D Bk .x/ (Bk .x/ D x k C    ), it follows from

(1.3) and (1.4) that

Bk .x C 1/  Bk .x/ D kx k1 ; Bk0 .x/ D kBk1 .x/:

This Bk .x/ is called the kth Bernoulli polynomial, which leads to various appli-
cations. We will give its definition and some properties in Sect. 4.3 (p. 55). The
formula for sums of consecutive powers may be obtained easily from the Bernoulli
polynomial. We will show it in Sect. 4.3.
Let us prove the fact that Bk D 0 for odd k greater than 1 using the polynomial
Sk .x/. We will prove this fact later more easily using the generating function, but
we prove it here because we need to use it before that.
Proposition 1.4. If n is an odd integer greater than or equal to 3, then Bn D 0.
As a consequence, .1/n Bn D Bn for all positive integers n except for n D 1.
Proof. Suppose k  1. Putting x D 1 in the formula Sk .xC1/Sk .x/ D .xC1/k
and using the fact Sk .0/ D 0, we obtain Sk .1/ D 0. Thus, by putting x D 1 in
the formula
!
X k
kC1
.k C 1/Sk .x/ D Bj x kC1j ;
j D0
j

we obtain
!
X k
kC1
.1/j Bj D 0:
j D0
j

If we subtract this from the recurrence formula in the definition of Bernoulli

numbers
!
X k
kC1
Bj D k C 1;
j D0
j

then only odd indexed terms remain and we have

Πk1 !
X 2 
kC1
2 B2j C1 D k C 1;
j D0
2j C 1

where Œx stands for the greatest integer less than or equal to x. Since B1 D 1=2, the
term for j D 0 in the left-hand side cancels with the right-hand side, and we obtain
1.2 Sums of Consecutive Powers of Integers and Theorem of Faulhaber 11

Πk1 !
X 2 
kC1
B2j C1 D 0 .k  3/:
j D1
2j C 1

Putting k D 3; 5; 7; : : : , we obtain Bk D 0 inductively for all odd k  3. t

u
Next, we state Faulhaber’s theorem.
Theorem 1.5. Let k  1 and set

u D n.n C 1/ . D 2S1 .n//; v D n.n C 1/.2n C 1/ . D 6S2 .n//:

Then we have

u2 X .k/ k1i
k1
S2kC1 .n/ D A u
2k C 2 i D0 i

and

v X k1
.k/
S2k .n/ D .k C 1  i /Ai uk1i ;
.2k C 1/.2k C 2/ i D0

.k/
where Ai is a number determined by the formula

i 1
!
.k/ .k/ 1 X kC1j .k/
A0 D 1; Ai D Aj .1  i  k  1/:
k C 1  i j D0 k C j  2i

 kC1j 
(For k C j  2i < 0, we put 2i C12j
D 0.)
For the proof we refer to [30] and [60]. We encourage readers to attempt to prove
it by themselves.
Before ending this section we prove an amusing recurrence formula for Bernoulli
numbers using Sk .x/ [55].
In the defining recurrence formula for Bernoulli numbers (1.2), if we move the
right-hand side to the left, and use the fact that .1/i Bi D Bi for i ¤ 1 and
B1 D 1=2, then we obtain
!
X
n
nC1
.1/ i
Bi D 0 (1.5)
i D0
i

fn D .n C 1/Bn . Then, we have

for n  1. Now, we define B
12 1 Bernoulli Numbers

Theorem 1.6. For n  1, we have

!
X
A
n
n C 1
.1/i BnCi D 0:
i D0
i

e
Remark 1.7. Though this formula is similar to (1.5), it requires only half the number
of terms to compute B2n (and thus B2n ).
Proof. From the binomial theorem we have
!
X
nC1
n C 1 nCi C1
x nC1
.1  x/ nC1
D .1/ i
x :
i D0
i

Taking the derivatives of both sides with respect to x, we have

!
 n  X
nC1
n C 1
.n C 1/ x .1  x/nC1  x nC1 .1  x/n D .1/i .n C i C 1/x nCi :
i D0
i

From this we have

.n C 1/x n .1  x/n .1  2x/ D .1/nC1 .2n C 2/x 2nC1

!
X n
i nC1
C .1/ .n C i C 1/x nCi : (1.6)
i D0
i

Putting x D .1 C y/=2 in this formula, we have

 n  n
1Cy 1y
l.h.s. D .n C 1/ y
2 2
D 22n .n C 1/y.1  y 2 /n
!
X
n
n 2i C1
2n
D 2 .n C 1/ .1/ i
y :
i D0
i

We put x D 1; 2; : : : ; m in (1.6), add them up, and use the equations

X
m X
m
y 2i C1 D .2x  1/2i C1 D S2i C1 .2m/  22i C1 S2i C1 .m/
xD1 xD1

to obtain
1.3 Formal Power Series 13

!
X n
 
2n i n
 2 .n C 1/ .1/ S2i C1 .2m/  22i C1 S2i C1 .m/
i D0
i
!
X n
i nC1
D .1/ .2n C 2/S2nC1.m/ C
nC1
.1/ .n C i C 1/SnCi .m/:
i D0
i

Since this holds for all positive integers m, we can use the same argument as before
to conclude that the formula obtained by replacing m by x,
!
X n
 
2n i n
 2 .n C 1/ .1/ S2i C1 .2x/  22i C1 S2i C1 .x/
i D0
i
!
X n
i nC1
D .1/ .2n C 2/S2nC1 .x/ C
nC1
.1/ .n C i C 1/SnCi .x/;
i D0
i

holds as a formula for polynomials. We take the derivative of both sides of this
with respect to x and set x D 0. The desired formula follows from the facts
that Sk0 .0/ D Bk (previous section), that the term for i D 0 in the left-hand side
vanishes due to the fact 2S10 .0/  2S10 .0/ D 0, and that Bk D 0 for odd k  3
(Proposition 1.4). t
u

1.3 Formal Power Series

We have defined Bernoulli numbers by a recurrence formula. However, it is also

common to define Bernoulli numbers using the generating function
1
X tn
te t
D Bn : (1.7)
et  1 nD0
nŠ

For the other definition of Bn we mentioned in Remark 1.2 in Sect. 1.1, we need to
t
t
replace the left-hand side by et 1 D ett e1  t. In fact, the generating function is very
useful for the study of various properties of Bernoulli numbers.
In this section we will explain fundamental facts about formal power series for
those who are not familiar with such objects. In the next section we will prove the
above formula as Theorem 1.12. Those who are not familiar with abstract algebra
are advised to take a glance at the next section and see how computations go before
reading this section.
Let R be a commutative integral domain with unit (written 1). (Although we
write in this general fashion, readers may think of R as the rational number field
Q for the time being. In this book R is either Q, or its finite extension, except
for Chap. 11, where we need p-adic numbers. A commutative integral domain is
14 1 Bernoulli Numbers

a commutative ring with the property that ab D 0 implies a D 0 or b D 0.)

A formal sum
1
X
an t n D a0 C a1 t C a2 t 2 C a3 t 3 C   
nD0

with an indeterminate t and coefficients in R is called a formal power series with

coefficients in R, and the set of all such formal power series is denoted by RŒŒt.
Two formal power series are defined to be equal if and only if all coefficients of t n
coincide. Typical examples which will appear in this book are (taking R D Q)
1 n
X t t t2 t3
et D D1C C C C
nD0
nŠ 1Š 2Š 3Š

and
X1
tn t2 t3
log.1 C t/ D .1/n1 D t  C     :
nD1
n 2 3

These are Taylor expansions of complex functions e x and log.1 C x/ around x D 0,

regarded as formal sums. We use the notation e t ; log.1 C t/, but these are nothing
but the formal power series of the right-hand sides, and we will not consider them
as functions in t.
The sum and the product of two formal power series are defined to be
1
X 1
X 1
X
an t C
n
bn t D
n
.an C bn /t n
nD0 nD0 nD0

and
X
1  X
1  X1 X
n
an t n  bn t n D cn t n ; cn D ai bni :
nD0 nD0 nD0 i D0

The product is defined using the distributive law formally, and the first few terms of
it are

.a0 C a1 t C a2 t 2 C a3 t 3 C    /  .b0 C b1 t C b2 t 2 C b3 t 3 C    /
D a0 b0 C .a0 b1 C a1 b0 /t C .a0 b2 C a1 b1 C a2 b0 /t 2
C .a0 b3 C a1 b2 C a2 b1 C a3 b0 /t 3 C    :

It is readily verified that RŒŒt is a commutative ring with unit. The zero element
and the unit are the formal power series

0 C 0  t C 0  t2 C 0  t3 C   
1.3 Formal Power Series 15

and

1 C 0  t C 0  t2 C 0  t3 C    ;

respectively. They are denoted simply by 0 and 1, respectively. The set of formal
power series all of whose coefficients are 0 except for the constant term is naturally
identified with R. Furthermore, regarding a polynomial P .t/ with coefficients in
R as a formal power series whose coefficients an are 0 when n is greater than the
degree of P .t/, we can consider RŒt as a subring of RŒŒt. The definitions of the
sum and the product of RŒŒt are natural generalizations of those of polynomials.
Since we assume that R is an integral domain, RŒŒt is also an integral domain.
Namely, we have
Proposition 1.8. The ring RŒŒt does not have a zero divisor (the product of
two non-zero elements is again non-zero). In particular, if A ¤ 0 and AB D
AC .A; B; C 2 RŒŒt/, then B D C .
Proof. Let A D a0 C a1 t C a2 t 2 C a3 t 3 C    and B D b0 C b1 t C b2 t 2 C b3 t 3 C
   be non-zero elements, and let ak ; bl be the first non-zero coefficients. Then the
coefficient ckCl of t kCl in A  B D c0 C c1 t C c2 t 2 C    is

ckCl D a0 bkCl C a1 bkCl1 C    C ak bl C    C akCl1 b1 C akCl b0 D ak bl :

Since R is an integral domain, ckCl D ak bl is not 0. Thus A  B is not 0.

If AB D AC , then A.B  C / D 0, which implies A D 0 or B  C D 0. Thus, if
A ¤ 0, then we have B D C . t
u
P1
Proposition 1.9. A formal power series nD0 an t n is invertible in RŒŒt if and
only if the constant term a0 is invertible.
P1 n
Proof. If
P nD0 an t P is invertible, P i.e., if there exists a formal power series
1 1 1
nD0 b n t n
such that . nD0 an t n
/  . nD0 bn t / D 1, then the constant term a0 b0
n

of the product must be 1, and thus a0 is invertible. Conversely, suppose that a0 is

invertible. We define bn 2 R .n  0/ by
X
n
b0 D a01 and bn D a01  ai bni .n  1/;
i D1
P1 P1
Pncompute the product . nD0 an t /  . nD0 Pbn t n /. Then, we have a0 b0 D 1
n
and we
and i D0 Pai bni D 0 for n  1. This shows that 1 n
nD0 bn t is the multiplicative
1 n
inverse of nD0 an t . t
u
We denote by A.0/ the constant term of a formal power series A.t/. As we
mentioned before, a formal power series is not a function and we do not generally
replace t by any special value. However, replacing t by 0 is always valid, and we
use this notation exceptionally for this case.
16 1 Bernoulli Numbers

Next, we show that we can define an infinite sum of elements of RŒŒt under
P1 .k/ i
certain conditions. Let fAk .t/g1
kD1 ; Ak .t/ D i D0 ai t be an infinite family of
formal power series. Suppose that these series satisfy the condition
.k/
for each i , there exist only finitely many k such that ai ¤ 0.
.k/ .k/
In other words, if we write Ak .t/ D ank t nk C .higher-order terms/ .ank ¤ 0/, we
have nk ! 1 as k ! 1. Then we define A1 C A2 C A3 C    to be
1
X
A1 C A2 C A3 C    D ai t i ;
i D0

where
1
X .k/
ai D ai :
kD1

In short, we take the sum of the coefficients in a common degree, and the above
condition ensures that each sum is a finite sum.
As a special case, we can define the “substitution” of a formal power series
P1 vanishing
with term into another formal power series. Namely, if A.t/ D
constant P
1
i D0 ai t i
, and B.t/ D i D1 bi t .B.0/ D b0 D 0/, then we can give a meaning
i

to the expression
1
X
A.B.t// D ai .B.t//i ;
i D0

because, from b0 D 0, the polynomial

ai .B.t//i begins at least in the degree i term,
i 1
and the family ai .B.t// i D0 satisfies the above condition. As an easy example,
we have
.t/ .t/2 .t/3
e t D 1 C C C C
1Š 2Š 3Š
t t2 t3
D 1 C  C
1Š 2Š 3Š
and
.t/2 .t/3
log.1  t/ D log.1 C .t// D t  C  
2 3
t2 t3
D t     :
2 3
The identities

e log.1Ct / D 1 C t
1.3 Formal Power Series 17

and

log.1 C .e t  1// D t

can be considered as the identities obtained by substituting one formal power series
to the other. (We will give proofs of these identities after Remark 2.7 on p. 33.)
We can also compute the reciprocal of a formal power series with constant term 1
by “substitution”. First we remark that the reciprocal of 1  t is 1 C t C t 2 C    :
1
D 1 C t C t2 C    :
1t

We can see this by computing the product .1t/.1Ct Ct 2 C   /. Now, let 1CB.t/
be a formal power series with constant term 1. Then B.t/ is a formal power series
without constant term (B.0/ D 0), and thus the reciprocal of 1 C B.t/ can be
computed by
1 1
D D 1 C .B.t// C .B.t//2 C    :
1 C B.t/ 1  .B.t//
As an example, we compute the first few Bernoulli numbers using (1.7). Since
t2 t3 t4 t5
et  1 D t C C C C C
2Š 3Š 4Š 5Š
 
t t2 t3 t4
D t 1C C C C C ;
2 6 24 120
we have
 
1 t t2 t3 t4
  D 1 C C C C
1C t
C t2
C t3
C t4
C 2 6 24 120
2 6 24 120
 2
t t2 t3 t4
C C C C C
2 6 24 120
 3
t t2 t3 t4
 C C C C
2 6 24 120
 4
t t2 t3 t4
C C C C C C
2 6 24 120
t t2 t4
D 1 C C 0  t3  C :
2 12 720
18 1 Bernoulli Numbers

From this we have

te t 1
D et   
et  1 t2
1 C 2 C 6 C 24
t t3 t4
C 120 C
  
t2 t3 t4 t t2 t4
D 1Ct C C C C 1 C C0t 
3
C
2 6 24 2 12 720
t t2 t4
D 1C C C 0  t3  C :
2 12 720
Thus, we have B0 D 1; B1 D 12 ; B2 D 16 ; B3 D 0; B4 D  30 1
;:::.
In this “substitution” operation, the following proposition gives the “inverse”.
Proposition 1.10. For a formal power series A.t/ D a0 C a1 t C a2 t 2 C    , there
exists a formal power series B.t/ such that
B.0/ D 0; A.B.t// D t

if and only if

a0 D 0 and a1 is invertible.

In this case B.t/ is unique, and we have B.A.t// D t. In other words A.t/ and B.t/
are inverse to each other with respect to composition.
Proof. If B.t/ D b1 t C b2 t 2 C    exists and satisfies A.B.t// D t, then by
comparing the constant terms and the degree 1 terms, we have a0 D 0 and a1 b1 D 1.
This shows that the condition is necessary.
Conversely, suppose that A.t/ satisfies a0 D 0 and a1 is invertible. We would
like to determine the coefficients of B.t/ D b1 t C b2 t 2 C b3 t 3 C    satisfying
A.B.t// D t. We first see from the coefficient of t that a1 b1 D 1. Since a1 is
invertible, we put b1 D a11 . For n  2, the coefficient of t n in A.B.t// equals the
coefficient of t n in

a1 .B.t// C a2 .B.t//2 C a3 .B.t//3 C    C an .B.t//n

since there is no term t n after this due to the fact that B.0/ D 0. This can be
written as

a1 bn C .polynomial in a2 ; a3 ; : : : ; an ; b1 ; b2 ; : : : ; bn1 /:

If b1 ; b2 ; : : : ; bn1 are already determined, then we can determine bn uniquely from

the fact that the above formula equals 0 and a1 is invertible. This proves the existence
as well as the uniqueness of B.t/.
1.3 Formal Power Series 19

The B.t/ obtained in this way satisfies B.0/ D 0, and b1 is invertible. We thus
see that there is a C.t/ (C.0/ D 0) such that

B.C.t// D t:

Then we substitute C.t/ for t in the formula t D A.B.t//, and use the fact that
B.C.t// D t to get

C.t/ D A.B.C.t/// D A.t/:

Thus, we have

B.A.t// D t:

t
u
The formulas we mentioned earlier,

e log.1Ct /  1 D t and log.1 C .e t  1// D t;

can be interpreted as saying log.1 C t/ and e t  1 are inverse to each other with
respect to composition. P1  
n d P1 n
The
P1derivative of formal power series nD0 an t , written dt nD0 an t
(or . nD0 an t n /0 ), is defined formally by term-by-term differentiation:
1
! 1
d X X
an t n
D nan t n1 D a1 C 2a2 t C 3a3 t 2 C 4a4 t 3 C    :
dt nD0 nD1

For example, .e t /0 D e t , and .log.1 C t//0 D 1  t C t 2  t 3 C    D 1Ct 1

.
This definition satisfies the usual rule of derivatives regarding sum and product.
We givePa proof of the productP rule .f .t/g.t//0 D f 0 .t/g.t/ C f .t/g 0 .t/. Let
f .t/ D nD0 an t , and g.t/ D 1
1 n n
nD0 bn t . Then

1
! !0 1
!
X X
n X X
n
0
.f .t/g.t// D ai bni t n
D n ai bni t n1
nD0 i D0 nD1 i D0
1
!
X X
n X
n
D i ai bni C .n  i /ai bni t n1
nD1 i D0 i D0
1
! 1
!
X X
n X X
n1
D i ai bni t n1
C ai .n  i /bni t n1
nD1 i D1 nD1 i D0
0 0
D f .t/g.t/ C f .t/g .t/:
20 1 Bernoulli Numbers

Z tX
1
Also, when R  Q, the integral an t n dt is defined formally by term-by-term
0 nD0
integration:
Z tX
1 1
X t nC1 t2 t3 t4
an t dt D
n
an D a0 t C a1 C a2 C a3 C    :
0 nD0 nD0
nC1 2 3 4

In order to treat the generating function of Bernoulli numbers, it is convenient

to generalize formal power series to Laurent6 series (with a finite number of terms
with negative powers).
Definition 1.11. The set of formal Laurent series with coefficients in R,
1
X
ai t i .for some integer N /;
i DN

is denoted by R..t//.
The sum and the product in R..t// are defined in the same way as in RŒŒt and with
these operations R..t//
P is a commutative integral domain that contains RŒŒt as a
subdomain. Also, 1 i
i DN ai t is invertible in R..t// if and only if the first non-zero
coefficient aN is invertible. These can be proved in the same way as in the proof of
Propositions 1.8 and 1.9. In particular, if R is a field, then so is R..t//. Furthermore,
the formal derivative and integral (when R  Q) are also defined, and the quotient
rule .f .t/=g.t//0 D .f 0 .t/g.t/  f .t/g 0 .t//=g.t/2 holds. This follows from the
product rule. (Put f .t/=g.t/ D h.t/, and take the derivative of f .t/ D g.t/h.t/.)

1.4 The Generating Function of Bernoulli Numbers

Theorem 1.12. Let Bn .n D 0; 1; 2; : : : / be the Bernoulli numbers. We have the

following formula in Q..t//:
1
X tn
te t
D Bn :
et  1 nD0
nŠ

As a matter of fact, the right-hand side belongs to QŒŒt, but we consider the
quotient of the formal power series te t by e t  1 in Q..t//.

6
Pierre Alphonse Laurent (born on July 18, 1813 in Paris, France—died on September 2, 1854 in
Paris, France).
1.4 The Generating Function of Bernoulli Numbers 21

P
Proof. It is sufficient to show . 1 tn
nD0 Bn nŠ /.e  1/ D te . The formula in question
t t

is obtained by dividing both sides by e  1. From the definition of the product of

t

formal power series we have

1
! 1
! 1 !
X tn X Bn n X tn
Bn .e  1/ D
t
t
nD0
nŠ nD0
nŠ nD1
nŠ

1
!
X X
n1
Bi 1
D tn
nD1 i D0
i Š .n  i /Š

1
! !
X X
n1
n tn
D Bi :
nD1 i D0
i nŠ

P n
From the recurrence formula (1.2) we have in1
D0 i Bi D n for all n  1. Thus
we have
1
! ! 1
X X
n1
n tn X tn
Bi D D te t ;
nD1 i D0
i nŠ nD1
.n  1/Š

which concludes the proof. t

u
Remark 1.13. Conversely, if we define Bn by the formula in the above theorem,
then we have
1
!
X tn
Bn .e t  1/ D te t :
nD0
nŠ

Expanding the left-hand side as in the above proof, and comparing it with the right-
hand side, we obtain the recurrence formula (1.2). This shows that Definition 1.1
and the definition using the generating function in Theorem 1.12 are equivalent.
Using the generating function, we can give a simpler proof of Proposition 1.4.
Proposition 1.14 (Proposition 1.4 revisited). If n is an odd integer greater than
or equal to 3, then Bn D 0.
t et
Proof. It suffices to show that the formal power series et 1
 t
2
does not have any
odd-degree terms. Since we have

te t t t.e t  1 C 1/ t t t
 D  D t C
e 1 2
t e 1
t 2 e 1 2
and
.t/e t .t/ t t t t
 D C D t C ;
e t  1 2 1  et 2 e 1 2
22 1 Bernoulli Numbers

t et
et 1
 2t is invariant under the substitution t ! t. This shows that the coefficients
of odd-degree terms are all 0. t
u
Looking at Bn for even n in Table 1.1, we see that they are non-zero and, from
B2 on, the signs alternate. This P
may be seen from the formula for the value of the
Riemann zeta function .s/ D 1 1
nD1 ns at positive even integers (replace k by 2k
on p. 61) (4.4):

.1/k1 .2/2k
.2k/ D B2k .k  1/:
2 .2k/Š

But here, we give a direct proof using the following recurrence formula for the
Bernoulli numbers due to Euler7 [33].
Proposition 1.15.
!
X
n1
2n
.2n C 1/B2n D  B2m B2.nm/ .n  2/:
mD1
2m

Proof. By the previous proposition, if we subtract the term of degree 1 from the
generating function of the Bernoulli numbers, we obtain the generating function of
even indexed B2n :

X 1
te t t t 2n
 D B2n :
e 1 2
t
nD0
.2n/Š

Let f .t/ be the left-hand side. Taking the derivative of f .t/ as was explained on
p. 19, we have

t2
f .t/  tf 0 .t/ D f .t/2  :
4
P1 t 2n
Substituting nD0 B2n .2n/Š for f .t/, we have

1 1
! !
X t 2n X Xn
2n t 2n t2
.1  2n/B2n D B2m B2.nm/  :
nD0
.2n/Š nD0 mD0
2m .2n/Š 4

7
Leonhard Euler (Born on April 15, 1707 in Basel, Switzerland—died on September 18, 1783 in
St. Petersburg, Russia).
1.4 The Generating Function of Bernoulli Numbers 23

Comparing the coefficients of both sides, we have

!
Xn
2n
.1  2n/B2n D B2m B2.nm/
mD0
2m

for n  2. Since the terms for m D 0 and m D n yield 2B2n , we subtract them from
both sides. Multiplying by 1, we obtain the formula in the proposition. t
u
Corollary 1.16. For n  1, we have .1/n1 B2n > 0.
Proof. We prove this by induction. For n D 1, B2 D 16 > 0. Suppose that the
proposition is true for all integers less than n. By multiplying the formula in the
proposition by .1/n1 , we get
!
X
n1
2n
.2n C 1/  .1/ B2n D
n1
.1/m1 B2m .1/nm1 B2.nm/ :
mD1
2m

The right-hand side is positive by assumption. Thus, .1/n1 B2n > 0, which settles
the proof. t
u
To conclude this chapter, we give expansions of tan x and cot x as alternative
generating functions for Bernoulli numbers.
Proposition 1.17. The Taylor expansion of tan x and the Laurent expansion of cot x
around x D 0 are

X1
x 2n1
tan x D .1/n1 .22n  1/22n B2n ;
nD1
.2n/Š
1
1 X x 2n1
cot x D C .1/n 22n B2n :
x nD1 .2n/Š

The right-hand sides converge for jxj < 

2
and 0 < jxj < , respectively.
Proof. The formal power series f .t/ in Proposition 1.15 can be written as f .t/ D
et Cet
2 coth. 2 /, where coth t D et et is the hyperbolic cotangent. Replacing t by x, we
t t

have

x x  X 1
x 2n
coth D B2n ;
2 2 nD0
.2n/Š

which is regarded as the Laurent expansion of coth. x2 /. Replacing x by 2ix .i D

p
1/, dividing both sides by x, and using the fact i coth.ix/ D cot x, we obtain
the Laurent expansion of cot x. From the duplication formula for cot x which reads

1
cot 2x D .cot x  tan x/ ;
2
24 1 Bernoulli Numbers

we have

tan x D cot x  2 cot.2x/:

This shows that the Taylor expansion of tan x can be obtained from the expansion of
cot x. As for the radii of convergence, we see it from the fact that the poles nearest
to 0 are located at ˙=2 and ˙, respectively. t
u
x 2n1
Remark 1.18. The coefficient of .2n1/Š in the Taylor expansion of tan x,

B2n
Tn WD .1/n1 .22n  1/22n ;
2n
is sometimes called the tangent number. It is a positive integer. The fact that it is
positive is seen from Corollary 1.16. We will prove that it is an integer in Remark 7.6
on p. 98. Knuth and Buckholtz [61] give another proof of this fact and a method to
compute Bernoulli numbers using Tn . See also Exercise 3.6 in Chap. 3.
Exercise 1.19. Compute the Bernoulli numbers Bn up to n D 12 by using
Theorem 1.6.
Exercise 1.20. Guess any pattern of prime numbers appearing in denominators of
Bernoulli numbers. (This is one of the topics covered in Chap. 3.)
Exercise 1.21. Compute the first several terms of the formal power series e log.1Ct /
and log.1 C .e t  1// using the definition of composition.
Exercise 1.22. Prove the quotient rule

.f .t/=g.t//0 D .f 0 .t/g.t/  f .t/g 0 .t//=g.t/2

of formal Laurant series.

Exercise 1.23. Compute first several tangent numbers Tn , and check that they are
positive integers.
Chapter 2
Stirling Numbers and Bernoulli Numbers

In this chapter we give a formula that describes Bernoulli numbers in terms of

Stirling numbers. This formula will be used to prove a theorem of Clausen and
von Staudt in the next chapter. As an application of this formula, we also introduce
an interesting algorithm to compute Bernoulli numbers.
We first summarize the facts on Stirling numbers. Those Stirling numbers we
need in this chapter are of the second kind, but we also introduce Stirling numbers
of the first kind, which will be needed later.1

2.1 Stirling Numbers

We first define the Stirling numbers of the second kind, restricting ourselves to the
case where a combinatorial meaning can be given. We adopt Knuth’s notation [59].
Definition 2.1 (Stirling numbers of the second kind (Stirling’s subset
numbers)).
For positive integers n and m, define


n
WD the number of ways to divide a set of n elements into m
m nonempty sets.

(Knuth proposes to read this as “n subset m”.)

1
According to Knuth [59], Stirling (James, born in May, 1692 in Garden, Scotland—died on
December 5, 1770 in Edinburgh, Scotland) first introduced the second kind. The names “first kind”
and “second kind” are due to Nielsen (Niels, born on December 2, 1865 in Orslev, Denmark—died
on September 16, 1931 in Copenhagen, Denmark), who first used these names in his book on the
Gamma function [73, §26].

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 25

in Mathematics, DOI 10.1007/978-4-431-54919-2__2, © Springer Japan 2014
26 2 Stirling Numbers and Bernoulli Numbers

For example, there are seven ways to divide the set {1, 2, 3, 4} into two non-
empty sets:

f1g [ f2; 3; 4g; f2g [ f1; 3; 4g; f3g [ f1; 2; 4g; f4g [ f1; 2; 3g;
f1; 2g [ f3; 4g; f1; 3g [ f2; 4g; f1; 4g [ f2; 3g:
4
Thus, 2 D 7. 
From the definition, mn D 0 if m > n. Also, from the definition we have the
recurrence formula:




nC1 n n
D Cm : (2.1)
m m1 m

This formula can be proved in the  same

 way n as in the
 n  combinatorial proof of the
relation of binomial coefficients: nC1 m
D m1
C m
. Suppose we divide a set
of n C 1 elements into m sets. We look at one particular element. If this element
n
forms a set by itself, there are m1 ways to divide the remaining n elements into

m  1 sets. If this element belongs to a set with other elements, there are mn ways
to divide other n elements into m sets, and there are  n m ways to put this particular
element into one of these sets, and
thus there are m m ways altogether.
We now proceed to define mn by the recurrence formula (2.1) for any integers
m and n.
Definition 2.2 (Stirling numbers

 of the second kind (general case)). For any
n
integers n and m, define by the recurrence formula (2.1) with the initial


m

0 n 0
conditions D 1, and D D 0 . n, m ¤ 0 /.
0 0 m
  n 
From (2.1), if we know two out of three values nC1 , m1 , and mn , we can
m n
n for the case m D 0, where we cannot divide m m by m.
determine the third except
In this case the value 0 D 0 is already given in the definition. We give the values of
n n
m for 7  m; n  7 in Table 2.1 on p. 27. It is easy to see how all values of m
are determined from the initial conditions, considering how the positions .nC1; m/,
.n; m  1/, and .n; m/ arelocated
in the table.
The new definition of mn coincides with the previous combinatorial definition
 
when n; m > 0. To see this, it suffices to verify n1 D 1 .n  1/ and m1 D
0 .m  2/ under the new definition. Since these values are obvious under the old
definition, and the recurrence formulas are the same for both definitions, the claim
follows. We put m D 1 in (2.1) to obtain




nC1 n n
D C :
1 0 1
2.1 Stirling Numbers 27


n
Table 2.1
m
mnn 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7
7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
6 0 0 0 0 0 0 0 0 0 0 0 0 0 1 21
5 0 0 0 0 0 0 0 0 0 0 0 0 1 15 140
4 0 0 0 0 0 0 0 0 0 0 0 1 10 65 350
3 0 0 0 0 0 0 0 0 0 0 1 6 25 90 301
2 0 0 0 0 0 0 0 0 0 1 3 7 15 31 63
1 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1
0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
2 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0
3 0 0 0 0 1 3 2 0 0 0 0 0 0 0 0
4 0 0 0 1 6 11 6 0 0 0 0 0 0 0 0
5 0 0 1 10 35 50 24 0 0 0 0 0 0 0 0
6 0 1 15 85 225 274 120 0 0 0 0 0 0 0 0
7 1 21 175 735 1624 1764 720 0 0 0 0 0 0 0 0

 
Putting n D 0 in this formula, we obtain 11 D 1. Also, since n0 D 0, we have
nC1 n  
1 D 1 , and thus n1 D 1 .n  1/. Similarly, m1 D 0 .m  2/ can be verified
by putting n D 0 in (2.1).
Remark 2.3. Logically, we only need Definition 2.2. However, in order to make the
meaning clearer, we began with Definition 2.1.
We now define the Stirling numbers of the first kind.
Definition 2.4 (Stirling numbers of the first kind (Stirling’s cycle numbers)).
For positive integers n and m, define

n
WD number of permutations of n letters .elements of the sym-
m metric group of degree n/ that consist of m disjoint cycles.

.We read “ n cycle m”./

An element of the symmetric group of degree n (the group of all permutations
(bijections) of the set f1; 2; : : : ; ng) can be decomposed uniquely into the product
of disjoint cycles. (See any textbook of group theory, for example [42, Chap. I,
Th. 6.3].) Among the permutations of n letters, we count the number of permutations
that
decompose
 into m disjoint cycles (including the cycles of length 1), and that
is mn . For example, the permutations of four letters that decompose into two
disjoint cycles are
28 2 Stirling Numbers and Bernoulli Numbers

.1/.2 3 4/; .1/.2 4 3/; .2/.1 3 4/; .2/.1 4 3/;

.3/.1 2 4/; .3/.1 4 2/; .4/.1 2 3/; .4/.1 3 2/;
.1 2/.3 4/; .1 3/.2 4/; .1 4/.2 3/:
h i
Thus, 4
D 11.
2



It follows immediately from the definition that nn D 1 (only the identity), n1 D
.n  1/Š (number of n-cycles).
The Stirling numbers of the first kind satisfy the recurrence formula
  
nC1 n n
D Cn : (2.2)
m m1 m

We can see this in the same way as before. A permutation of n C 1 letters may fix
 n C 1. In this case the letter n C 1 forms a cycle by itself, and there are

then letter
m1
permutations of the remaining n letters that decompose into m  1 disjoint
cycles. A permutation that moves the letter n C 1 can be obtained by inserting the
letter n C 1 in one of the m disjoint cycles in the decomposition of permutations
of n letters. There are j ways to form a cycle of length j C 1 out of a cycle of
length j by inserting another letter.
Therefore,
 there are n ways to insert the letter
n C 1, and thus there are in total n mn permutations that move the letter n C 1. This
shows (2.2).

Just as we did for the Stirling numbers of the second kind, we define mn for any
m and n as follows.
Definition 2.5 (Stirling numbers
 of the first kind (general case)). For any
n
integers n and m, define by the recurrence formula (2.2) with the initial
  m 
0 n 0
conditions D 1, and D D 0 . n, m ¤ 0 /.
0 0 m


Table 2.2 shows the values of mn in the same range as Table 2.1. Readers will
certainly notice the remarkable relations between these two tables. As Knuth asserts,
there is only one kind of Stirling number, and the recurrence formulas (2.1) and (2.2)
are essentially the same formula (see the next proposition).
We list some formulas which Stirling numbers satisfy. There are more formulas,
but we only list the ones we need later in this book. For others, see [38] for example.
Proposition 2.6.


n m
.1/ D :
m n

Xn

n m
.2/ xn D x .n  0/;
mD0
m
2.1 Stirling Numbers 29

hni
Table 2.2
m
mnn 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7
7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1
6 0 0 0 0 0 0 0 0 0 0 0 0 0 1 21
5 0 0 0 0 0 0 0 0 0 0 0 0 1 15 175
4 0 0 0 0 0 0 0 0 0 0 0 1 10 85 735
3 0 0 0 0 0 0 0 0 0 0 1 6 35 225 1624
2 0 0 0 0 0 0 0 0 0 1 3 11 50 274 1764
1 0 0 0 0 0 0 0 0 1 1 2 6 24 120 720
0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
2 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0
3 0 0 0 0 1 3 1 0 0 0 0 0 0 0 0
4 0 0 0 1 6 7 1 0 0 0 0 0 0 0 0
5 0 0 1 10 25 15 1 0 0 0 0 0 0 0 0
6 0 1 15 65 90 31 1 0 0 0 0 0 0 0 0
7 1 21 140 350 301 63 1 0 0 0 0 0 0 0 0

where x m stands for

x m D x.x  1/.x  2/    .x  m C 1/ .m > 0/; x 0 D 1:

X
n 
n m
.3/ x n D .1/n .1/m x :
mD0
m

  Xn
  
d n n m d m
.4/ x D x .n  1/:
dx mD1
m dx

X

n l
.5:1/ For m, n  0, .1/l D .1/m ım;n :
l m
l0

X 

l n l
.5:2/ For m, n  0, .1/ D .1/m ım;n :
l m
l0

Here, ım;n stands for Kronecker’s delta; i.e., ım;n D 1 for m D n and ım;n D 0 for
m ¤ n. The sums on the left-hand sides of .5.1/ and .5.2/ are finite sums.
30 2 Stirling Numbers and Bernoulli Numbers

 !
.1/m X
m
n l m
.6/ For m, n  0, D .1/ l n:
m mŠ l
lD0

X1
 n
.e t  1/m n t
.7/ D .m  0/:
mŠ nDm
m nŠ

X1

tm n n
.8/ D t .m  1/:
.1  t/.1  2t/    .1  mt/ nDm
m

X1  n
. log.1  t//m n t
.9/ D .m  0/:
mŠ nDm
m nŠ

Proof. (1) Write an;m D m . It suffices to verify that an;m satisfies the same
n

recurrence formula with the same initial conditions as those for mn . It is easy
for the initial conditions. In the recurrence formula (2.1), replacing m by n
and n by m, we have




m C 1 m m
D n :
n n  1 n

Adding n m
n to both sides, we have




m C 1 m m
Cn D :
n n n  1

This implies an;m1 C nan;m D anC1;m , which is nothing but the recurrence
formula (2.2). P
(2) We use the same strategy. Namely, write x n D nmD0 an;m x m , and verify that
an;m satisfies the same recurrence formula as mn within the range m; n  0
and the same initial conditions. It is easy for the initial conditions (put a0;m D 0
.m > 0/). From x mC1 D x m .x  m/ and x m  x D x mC1 C mx m , we have

X
nC1
anC1;m x m D x nC1 D x n  x
mD0

X
n
D an;m x m  x
mD0
2.1 Stirling Numbers 31

X
n
D an;m .x mC1 C mx m /
mD0

X
nC1 X
n
D an;m1 x m C man;m x m :
mD1 mD0

Comparing the coefficients of x m of both sides, we obtain the recurrence

formula (2.1). (Here, we put an;1 D an;nC1 D 0.)
(3) is similarly proved and we omit it.
(4) We prove it by induction. For n D 1, both sides equal x dx
d
. Suppose the formula
is true for n. Then
    n
  
d nC1 d X n m d m
x D x x
dx dx mD1 m dx
n

X  m  mC1 !
n d d
Dx mx m1
Cx m

mD1
m dx dx

X
n
  m X n
  
n m d n mC1 d mC1
D m x C x
mD1
m dx mD1
m dx

nC1 

X
  m
n n d
D m C xm
mD1
m m  1 dx

nC1

X   m
nC1 d
D xm :
mD1
m dx
 n 
(To prove the next-to-last equality, we use nC1 D 0 and n0 D 0.) This shows
that the formula is true for n C 1.
(5) We prove it using (2) and (3). For (5.1), we put the formula

X
l 
l
x l D .1/l .1/m xm
mD0
m

obtained by replacing n by l in (3) into the formula

n

X n l
x D
n
x ;
l
lD0
32 2 Stirling Numbers and Bernoulli Numbers

which is obtained by replacing m by l in (2). Then we have

n

X Xl 
n l
xn D .1/l .1/m xm
l mD0
m
lD0

X
n X
n

n l
D .1/ m
.1/ l
xm:
mD0
l m
lDm

Comparing the coefficients on both sides, we obtain (5.1). The formula (5.2)
can be obtained by substituting (2) into (3).
(6) We prove
it by verifying that the right-hand side satisfies the recurrence formula
for mn . Denote the right-hand side by an;m once again. It is easy to see that
0;0 D 1 and an;0 D 0 formn  1. (Define 0 D 1.) If n D 0, and m  1, then
0
a
P m l m
lD0 .1/ l D .1  1/ D 0. Thus, we have a0;m D 0. As for the recurrence
formula, we verify it as follows:

man;m C an;m1
! !
.1/m X .1/m1 X
m m1
l m l m1
D .1/ l C
n
.1/ ln
.m  1/Š l .m  1/Š l
lD0 lD0
 ! !
.1/m X
m
m m1
D .1/l  ln
.m  1/Š l l
lD0
!
m X m
.1/ l l m n
D .1/ l
.m  1/Š m l
lD0
!
.1/m X
m
l m
D .1/ l nC1
mŠ l
lD0

D anC1;m :

One can prove (6) more naturally using the difference calculus. See for example
[53, §58].
(7) First, note that the right-hand
 side of (7) can be expressed as the sum from
n D 0. (If n < m, then mn D 0.) Substitute (6) into the right-hand side and
simplify:

1
 !
X .1/m X
m
m tn
r.h.s D .1/l ln
nD0
mŠ l nŠ
lD0
! 1 !
.1/m X X .lt/n
m
l m
D .1/
mŠ l nD0
nŠ
lD0
2.1 Stirling Numbers 33

!
.1/m X
m
l m
D .1/ e lt
mŠ l
lD0
m
.1/
D .1  e t /m
mŠ
.e t  1/m
D
mŠ
D l.h.s.
P
It can also be proved by writing .e t  1/m =mŠ D 1
n
a t , and verifying
 n nDm n;m nŠ
that an;m satisfies the same recurrence formula as m .

(8) Denote the right-hand side by fm . From the recurrence formula (2.1) for mn ,
n
we have (noting that m D 0 for n < m)

1

X n
fm D tn
nDm
m
1 

X 

n1 n1
D Cm tn
nDm
m1 m

X1
 X1

n n n
Dt t C mt
n
t
nDm1
m1 nDm
m

D tfm1 C mtfm :

Thus, fm D 1mtt
fm1 . From n1 D 1 .n  1/, we havef1 D 1t t
, which shows
that fm is equal to the left-handP
side of (8).
(9) Writing . log.1  t//m=mŠ D 1 tn
nDm an;m nŠ , and taking the derivatives of both
sides, we have

X1
. log.1  t//m1 1 t n1
 D an;m :
.m  1/Š 1t nDm
.n  1/Š

From this we have

1
X X1
tn t n1
an;m1 D .1  t/ an;m :
nDm1
nŠ nDm
.n  1/Š

n
This shows that an;m satisfies the same recurrence formula as m
. It is easy to
verify the initial conditions. t
u
Remark 2.7. Sometimes the Stirling numbers are defined by the formulas (2) and
(3) in this proposition.
34 2 Stirling Numbers and Bernoulli Numbers

Using (5.1) and (7) in this proposition, we give a proof of the formulas that were
left unproved in the previous chapter (p. 17). We give a proof of

log.1 C .e t  1// D t:

Once this is proved, the converse

e log.1Ct /  1 D t

follows from Proposition 1.10. (Of course, we can prove it similarly using (9) in this
proposition.)
1
X .e t  1/m
log.1 C .e t  1// D .1/m1
mD1
m
1
X 1
 n
X n t
D .1/m1 .m  1/Š
mD1
m nŠ nD0

X1 X1 
! n   
m1 m n t m
D .1/ .m  1/Š D
nD0 mD1
1 m nŠ 1

X1
tn
D .1/n1 ı1;n
nD0
nŠ
D t:

2.2 Formulas for the Bernoulli Numbers Involving

the Stirling Numbers

A formula expressing the Bernoulli numbers as a finite sum involving the Stirling
numbers of the second kind has been known at least since Kronecker2 [62]. (See
also an account of von Staudt in Sect. 3.3.) By Proposition 2.6 (6), the Stirling
numbers can be expressed as a finite sum involving only the binomial coefficients.
Therefore, the Bernoulli numbers can be expressed as an elementary (i.e., involving
only binomial coefficients and simple polynomials) double sum. Gould [36] studied
various historical sources for analogous formulas, and he concluded that, at least
until around 1970, the very fact that there are closed finite formulas (without
involving infinite sums and integrals) for the Bernoulli numbers was not widely
known. He also conjectures that there is no “elementary” (in the above sense)

2
Leopold Kronecker (born on December 7, 1823 in Liegnitz, Prussia (now Legnica, Poland)—died
on December 29, 1891 in Berlin, Germany).
2.2 Formulas for the Bernoulli Numbers Involving the Stirling Numbers 35

formula for the Bernoulli numbers expressed as a finite single sum. As far as we
know, such a formula has not been discovered.
We prove the following formula.
Theorem 2.8.

Xn
.1/m mŠ mn
Bn D .1/ n
.n  0/:
mD0
mC1

Proof. Rewriting the generating function of Bn as

te t t  log .1  .1  e t //
D D ;
et  1 1  e t 1  e t
P
and substituting 1  e t .D t C    / for t in  log.1  t/ D 1 mD1
tm
m
, we have
1
X tn  log .1  .1  e t //
Bn D
nD0
nŠ 1  e t

X1
.1  e t /m1
D
mD1
m

X1
.1/m .e t  1/m
D
mD0
mC1

X1 1

.1/m mŠ X n .t/n
D (Proposition 2.6 (7))
mD0
m C 1 nDm m nŠ
1  !
X Xn
.1/m mŠ mn tn
D .1/ n
:
nD0 mD0
mC1 nŠ

tn
Comparing the coefficients of on both sides, we obtain the formula. t
u
nŠ
Remark 2.9. By Proposition 2.6 (6), we can also write
!
Xn
1 X
m
l m
Bn D .1/ n
.1/ l n:
mD0
m C 1 l
lD0

Proposition 2.10 (A variation of Theorem 2.8).

 nC1
X
n
.1/m mŠ mC1
Bn D .n  0/:
mD0
mC1
36 2 Stirling Numbers and Bernoulli Numbers

Proof. We can prove this by computing the generating function on the right-hand
side. Here, we useTheorem2.8.
From the recurrence formula (2.1) for the Stirling
nC1 n
numbers we have mC1 D mn C .m C 1/ mC1 , and thus
 n  n 
X
n
.1/m mŠ m
C .m C 1/ mC1
r.h.s. D :
mD0
mC1
h i
mC1
On the other hand, since 1 D mŠ, it follows from Proposition 2.6 (5.1) that

n 

X
n
.1/m mŠ.m C 1/ mC1
X
n
mC1 n
D .1/m D ı1;n :
mD0
mC1 mD0
1 mC1

Hence, the right-hand side of the proposition equals


Xn
.1/m mŠ mn
C ı1;n D .1/n Bn C ı1;n :
mD0
m C 1

(Theorem 2.8.) For n D 1, this formula equals B1 C 1 D 12 D B1 . For n ¤ 1, it

equals .1/n Bn , which coincides with Bn due to Proposition 1.4 on p. 10. u
t
This proposition can be interpreted as the following algorithm, which first
appeared in the study of multiple zeta values at non-positive integers by S. Akiyama
and Y. Tanigawa [2]. First, define the 0th row by 1, 12 , 13 , 14 , 15 ; : : : . (We count the
number starting from 0 instead of 1 for convenience.) Then, we define the first row
by 1  .1  12 /, 2  . 12  13 /, 3  . 13  14 /; : : : . In general, if we denote by an;m the
mth entry of the nth row, define the mth entry of the .n C 1/st row anC1;m by
.m C 1/  .an;m  an;mC1 /. In other words, we take the difference of the adjacent
entries and multiply by (position of the entry C1), and place them in the next row.
Then, the leftmost entries of the rows, 1, 12 , 16 , 0,  30 1
; : : : , are nothing but the
Bernoulli numbers.
Taking Proposition 2.10 into account, it suffices to show the following proposi-
tion to prove this fact.
Proposition 2.11. Given a0;m .m D 0; 1; 2; : : : /, define an;m .n  1/ by

an;m D .m C 1/  .an1;m  an1;mC1 / .n  1; m  0/:

Then,

X
n

nC1
an;0 D .1/ mŠ m
a0;m :
mD0
mC1
2.2 Formulas for the Bernoulli Numbers Involving the Stirling Numbers 37

1 1 1 1 1 1 1 1 1
2 3 4 5 6 7 8 9

1 1 1 1 1 1 1 1
2 3 4 5 6 7 8 9

1 1 3 2 5 3 7
6 6 20 15 42 28 72

0 1 1 2 5 5
30 20 35 84 84

1 1 3 1
− 30 −
30
−
140
−
105
0

0 − 1 − 1 − 4
42 28 105

1 1 1
42 42 140

0 1
30

1
− 30

Proof. We use the generating function

1
X
gn .t/ D an;m t m :
mD0

By the recurrence formula for an;m , we have for n  1

1
X
gn .t/ D .m C 1/.an1;m  an1;mC1 /t m
mD0
1 1
d X d X
D . an1;m t mC1 /  . an1;mC1 t mC1 /
dt mD0 dt mD0

d d
D .tgn1 .t//  .gn1 .t/  an1;0 /
dt dt
38 2 Stirling Numbers and Bernoulli Numbers

d
D gn1 .t/ C .t  1/ .gn1 .t//
dt
d
D ..t  1/gn1 .t//:
dt

Thus, if we put .t  1/gn .t/ D hn .t/, we have

d
hn .t/ D .t  1/ .hn1 .t// .n  1/:
dt
From this we obtain
 
d n
hn .t/ D .t  1/ .h0 .t//:
dt

It follows from Proposition 2.6 (4) (with x replaced by t  1) that

Xn
  m
n d
hn .t/ D .t  1/ m
h0 .t/:
mD0
m dt

Putting t D 0, we have
n

X n
an;0 D .1/m mŠ.a0;m1  a0;m /
mD0
m
n1

X  Xn

n n
D .1/mC1 .m C 1/Ša0;m  .1/m mŠa0;m
mD0
mC1 mD0
m

X
n 


n n
D .1/ mŠa0;m .m C 1/
m
C
mD0
m C 1 m

X
n

nC1
D .1/ mŠ m
a0;m :
mD0
mC1

This proves the formula in the proposition. t

u
Remark 2.12. We note that, if we use instead the algorithm

anC1;m D .m C 1/  .an;mC1  an;m /;

we obtain Bernoulli numbers with the other convention (B1 D 12 ).

Exercise 2.13. Prove the following.

(1) mn D 0 when mn < 0.
2.2 Formulas for the Bernoulli Numbers Involving the Stirling Numbers 39

 1
(2) m D .m  1/Š for m D 1; 2; 3; : : :.
nC1  n
(3) n
D n1 D n.nC1/2
for n D 0; 1; 2; : : :.
Exercise 2.14. Formulate the statements corresponding to the previous exercise for
the Stirling numbers of the first kind, and prove them.
Exercise 2.15. Prove the formula e log.1Ct /  1 D t for the composition of formal
power series by using Proposition 2.6 (9).
Exercise 2.16. Prove the following formula for the sum of powers:

 !
X
n X
n
k n C 1
ik D jŠ :
i D1 j D0
j j C1

Exercise 2.17. Compute the first several Bn using the formula in Theorem 2.8.
Exercise 2.18. In the Akiyama–Tanigawa triangle (on p. 37), what are the numbers
next to the leftmost entries?
Chapter 3
Theorem of Clausen and von Staudt,
and Kummer’s Congruence

3.1 Theorem of Clausen and von Staudt

The denominators of the Bernoulli numbers can be completely determined. This is

due to Clausen1 [26] and von Staudt2 [96]. More precisely, the “fractional part” of
Bn is given by the following theorem. This result gives a foundation for studying
p-adic properties of the Bernoulli numbers. It also plays a fundamental role in the
theory of p-adic modular forms through the Eisenstein3 series [82].
Theorem 3.1. For n D 1 and for any even integer n  2, Bn can be written as
X 1
Bn D  C Cn .Cn is an integer/:
pWprime
p
p1jn

Here, the sum runs over all the prime numbers p such that p  1 divides n. (For
integers a and b, the symbol a j b means that a divides b, and a − b means that a
does not divide b.)
Proof. The statement is clearly true for n D 1. Let n be an even  integer greater
than or equal to 2. Since the numerator of each term .1/m mŠ mn =.m C 1/ in the
right-hand side of the formula in Theorem 2.8 is an integer, the only prime numbers

1
Thomas Clausen (born on January 16, 1801 in Snogebaek, Denmark—died on May 23, 1885 in
Dorpat, Russia (now Tartu, Estonia)).
2
Karl Georg Christian von Staudt (born on January 24, 1798 in Imperial Free City of Rothenburg
(now Rothenburg ob der Tauber, Germany)—died on June 1, 1867 in Erlangen, Bavaria (now
Germany).
3
Ferdinand Gotthold Max Eisenstein (born on April 16, 1823 in Berlin, Germany—died on October
11, 1852 in Berlin, Germany).

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 41

in Mathematics, DOI 10.1007/978-4-431-54919-2__3, © Springer Japan 2014
42 3 Theorem of Clausen and von Staudt, and Kummer’s Congruence

that contribute to the denominator of Bn are divisors of m C 1. It follows from this

that a prime number p such that p > n C 1 cannot be a divisor of the denominator
of Bn .
First suppose that m C 1 is a composite number, and let m C 1 D ab, 1 < a; b <
m. If a ¤ b, then ab divides mŠ, and thus .1/m mŠ mn =.m C 1/ is an integer.
If a D b and 2a  m, then a and 2a divide mŠ, and thus a2 D m C 1 divides mŠ.
This implies that .1/m mŠ mn =.m C 1/ is an integer. If a D b and 2a > m, then
2a  m C 1, and thus m C 1 D a2  2a  m C 1. This implies a2 D 2a, or a D 2.
In this case m D 3, and this term appears when n  3. In this case (n even), from
Proposition 2.6 (6) on p. 30 we have

 X3
!
n l 3
m
.1/ mŠ D .1/ ln
m l
lD0

D 0  3 C 3  2n  3n
 1  .1/n  0 mod 4:

(For integers a; b and a natural number4 n, if nj.a  b/, then we say that a is
congruent to b modulo n, and we write a  b mod n. If n − .a  b/, then we
write a 6 b mod n.) This shows that .1/m mŠ mn =.m C 1/ is an integer. We have
thus shown that if m C 1 is a composite number, .1/m mŠ mn =.m C 1/ is an integer.
Next, suppose that m C 1 equals a prime number p. Since p  1  n, it follows
from Proposition 2.6 (6) that

 Xp1
!
n l p1
m
.1/ mŠ D .1/ ln
m l
lD0

X
p1
p1
 l n mod p (since l
 .1/l mod p)
lD0

1 mod p if p  1 divides n,
 (3.1)
0 mod p if p  1 does not divide n.

The last congruence can be seen as follows. If p  1 divides n, then the fact that
P p1 n
lD0 l  p  1  1 mod p follows from Fermat’s5 Little Theorem; i.e.,
l p1
 1 mod p if l and p are relatively prime. If p  1 does not divide n,
then choose c such that c n 6 1 mod p (for example, a primitive root modp).
When l runs from 1 through p  1, cl mod p also
Pp1 runs from the classes of 1
Pp1
through p  1, and thus we have lD0 .cl/n  n
lD0 l . Therefore, we have

4
Throughout the book, the term “natural number” means a positive integer.
5
Pierre de Fermat (born on August 17, 1601 in Beaumont-de-Lomagne, France—died on January
12, 1665 in Castres, France).
3.2 Kummer’s Congruence 43

Pp1 Pp1
.c n  1/ lD0 l n  0 mod p, and lD0 l n  0 mod p. We thus see that if
n
m C 1 D p, .1/m mŠ m =.m C 1/ is an integer  if p  1 does not divide n. If, on
the other hand, p  1 divides n, then .1/m mŠ mn =.m C 1/ is not an integer but the
above congruence shows that it becomes an integer if we add p1 to it. This completes
the proof. t
u
The theorem of Clausen and von Staudt completely describes the denominators
of Bernoulli numbers. Then how about the numerators? Not much is known about
this. Here we briefly explain the notion of (ir)regular primes, which are related to
the numerators of Bernoulli numbers.
A prime p is said to be regular if p does not divide any of the numerators of the
Bernoulli numbers B2 ; B4 ; : : : ; Bp3 . Otherwise p is called irregular. According
to Kummer, p is regular if and only if the class number of the pth cyclotomic field
Q.p / is not divisible by p (cf. Washington [100, §5.3]). It is well known that this
work of Kummer originated from his attempt to solve Fermat’s last theorem. For
this topic, see [31] or [79].
For instance, there are only three irregular primes less than 100: 37; 59; 67.
Kummer determined (by hand!) all irregular primes less than or equal to 163, and
moreover determined the indices of Bn whose numerators are divisible by a given
irregular prime. The next few irregular primes after 67 are 101; 103; 131; 149; 157,
and p D 157 is the first prime which divides two of the numerators of
B2 ; B4 ; : : : ; Bp3 (see the table in [100] for irregular primes p less than 4,001
and indices of Bernoulli numbers divisible by p).

3.2 Kummer’s Congruence

The goal of this section is to prove the following theorem, which plays a crucial role
in the arithmetic of cyclotomic fields and the theory of p-adic L-functions. For a
prime number p, we denote by Z.p/ the ring of rational numbers whose denominator
is prime to p. Using such a ring, it is easy to describe how many times p divides the
numerator of a rational number, or decide whether or not two rational numbers are
congruent modulo a power of p. The set of invertible elements, Z .p/ , of the ring Z.p/
is the set of rational numbers whose numerator and denominator are both relatively
prime to p. For two elements x and y, we say that the congruence relation x  y
mod p a holds if the numerator of x  y is divisible by p a .
Theorem 3.2. Let p be an odd prime.
(1) Suppose n is a positive even integer not divisible by p  1. Then we have

Bn
2 Z.p/ :
n
44 3 Theorem of Clausen and von Staudt, and Kummer’s Congruence

(2) Let a  1 be a natural number and m; n be two even integers that satisfy
a C 1  m  n. Suppose that m and n are not divisible by p  1, and that
n  m mod .p  1/p a1 . Then we have
Bn Bm
 mod p a :
n m

If we remove the condition m  a C 1, then the above congruence relation

does not hold in general (except for the case a D 1, where m  a C 1 gives no
restriction) and a certain correction is necessary (see Theorem 11.6 on p. 198 for a
more precise statement). This corrected version is what is known today as Kummer’s
congruence. However, the version that Kummer6 himself gave was different (except
when a D 1) from these theorems we are presenting (see [64]). The formulation of
Kummer’s congruence seems to have been changed since a natural interpretation
of the congruence by p-adic integrals was discovered. Kummer’s own proof is
elementary, and we can derive the above congruence relation using his method. The
complete form of the congruence will be proved using a p-adic integral in Chap. 11.
We first prove the following lemma.
Lemma 3.3. Let p be an odd prime number, and let '.t/ 2 Z.p/ ŒŒt; r; s 2 Z.p/ .
Develop the formal power series obtained by substituting e rt  e st D .r  s/t C
.r 2 s 2 / 2
2 t C    for t in '.t/ W
1
X tn
'.e rt  e st / D An :
nD0
nŠ

Then we have An 2 Z.p/ , and for any m; a .m  a  1/, the congruence

AmC.p1/pa1  Am mod p a
holds.
P1
Proof. Write '.t/ D kD0 ak t
k
. Then, we have
1
X
'.e rt  e st / D ak .e rt  e st /k
kD0

1
!
X X
k
k rt .kh/ st h
D ak .1/h e e
h
kD0 hD0

1 X
!
X k
k .r.kh/Csh/t
D .1/ akh
e :
h
kD0 hD0

6
Ernst Eduard Kummer (born on January 29, 1810 in Sorau, Brandenburg, Prussia (now
Germany)—died on May 14, 1893 in Berlin, Germany).
3.2 Kummer’s Congruence 45

We obtain Am by taking the derivative m times with respect to t, and putting t D 0:

1 X
!
X k
k
Am D h
.1/ ak .r.k  h/ C sh/m :
h
kD0 hD0

Since .e rt  e st /k begins with the term t k , the sum with respect to h in the above
formula equals 0 if k > m. Thus, the above sum with respect to k is a finite sum.
Replacing m by m C .p  1/p a1 , and subtracting Am , we have

AmC.p1/pa1  Am
1 X
!
X k
k a1
D h
.1/ ak .r.k  h/ C sh/m ..r.k  h/ C sh/.p1/p  1/:
h
kD0 hD0

If r.k  h/ C sh is divisible by p, then since m  a, .r.k  h/ C sh/m is divisible

a1
by p a . If r.k  h/ C sh is not divisible by p, then .r.k  h/ C sh/.p1/p 1

is divisible by p a since the order of .Z=p a Z/ is .p  1/p a1 . In either case the
right-hand side is divisible by p a , and thus

AmC.p1/pa1  Am mod p a

holds. t
u
Proof (of Theorem 3.2). For a natural number c ¤ 1 relatively prime to p, consider

1 c
'.t/ D  :
t .1 C t/c  1

Since c 2 Z
.p/ , the expression

.1 C t/c  1 c1 2
DtC t C
c 2

belongs to Z.p/ ŒŒt, and it is invertible in Z.p/ ..t// (at the end of Sect. 1.3 on p. 20).
From

c 1 c1 c2  1
D  C t C ;
.1 C t/  1
c t 2 12

we have '.t/ 2 Z.p/ ŒŒt. Since

X tn 1
t te t t
D  t D 1  C Bn ;
et  1 et  1 2 nD2 nŠ
46 3 Theorem of Clausen and von Staudt, and Kummer’s Congruence

we have
 
1 c 1 t ct
'.e  1/ D t
t
 D 
e  1 e ct  1 t e t  1 e ct  1
1  
1c X Bn t n1
D C .1  c n /  :
2 nD2
n .n  1/Š

Thus it follows from the lemma (r D 1; s D 0) that .1 c n / Bnn 2 Z.p/ , and if n  m

mod .p  1/p a1 , then

Bn Bm
.1  c n /  .1  c m / mod p a :
n m
Here, we choose c to be a primitive root modulo p (c mod p is a generator of
.Z=pZ/ ). If n is not divisible by p  1, then m is not divisible by p  1, and
we have 1  c n ; 1  c m 2 .Z=p a Z/ with 1  c n  1  c m mod p a . This gives
n ; m 2 Z.p/ and the congruence in the theorem. t
u
Bn Bm

3.3 Short Biographies of Clausen, von Staudt and Kummer

Thomas Clausen, an astronomer and a mathematician, was born on January 16, 1801
in Snogbæk, Denmark.7 He was educated by a local pastor Georg Holst before he
got a position as an assistant at the Altona observatory, where Schumacher8 was the
head. He then moved to the Joseph von Utzschneider Optical Institute in Munich as
the successor to Fraunhofer,9 who is famous for “Fraunhofer lines” in physics and
optics. In 1842, he was appointed observer at the Dorpat observatory in Russia, and
there he remained until his retirement in 1872. The Copenhagen Academy awarded
Clausen a prize for his work on the determination of the orbit of the 1770 comet
(published in Astronomische Nachrichten in 1842). Bessel10 highly praised this

7
The description in this section is based on Biermann [17, 18], Noether [74], Hensel [40], and
Lampe [65].
8
Heinrich Christian Schumacher (born on September 3, 1780 in Bramstedt, Germany—died in
1850). His first degree was in law, and after that he studied astronomy under Gauss and became
an astronomer. He launched the journal Astronomische Nachrichten, which is famous for the
contributions of Abel and Jacobi on the theory of elliptic functions, as the managing editor.
Incidentally, Abel once met Clausen during his stay in Hamburg (cf. [19]).
9
Joseph von Fraunhofer (born on March 6, 1787 in Straubing, Germany—died on June 7, 1826 in
Munich, Germany).
10
Friedrich Wilhelm Bessel (born on July 22, 1784 in Minden, Westphalia (now Germany)—died
on March 17, 1846 in Königsberg, Prussia (now Kaliningrad, Russia)), an astronomer, who is
famous for the Bessel function.
3.3 Short Biographies of Clausen, von Staudt and Kummer 47

work. In 1854 he received through Gauss11 a corresponding membership from the

Göttingen Academy. In 1856 he received the same class of membership from the St.
Petersburg Academy. Clausen published approximately 150 papers on a multitude
of subjects. Still today one sees his name in such terms as Clausen’s identity12 in
the theory of hypergeometric series [23], and Clausen’s function (a variant of the
dilogarithm function) [24]. Also, he computed  up to 250 decimal places,13 and
6
found the prime factorization of the sixth Fermat number 22 C 1.14 The actual
6
factorization is 22 C 1 D 274177  67280421310721 and, as Clausen commented,
the second factor of this is the largest prime number known to that day (letter to
Gauss on January 1, 1855). He wrote to his friend and astronomer C. A. F. Peters
in Königsberg that he had found a new method of factoring a large number, and as
an example he wrote .1017  1/=9 D 2071723  5363222357, which was not known
before and had taken some interest in connection to the period length of the periodic
decimal of 1=p. However, he never published his “new method” and we have no
idea what kind of method he had found.
Clausen never married, and died on May 23, 1885 in Dorpat, now Tartu in
Estonia.
Karl Georg Christian von Staudt was born on January 24, 1798 in Rothenburg-
ob-der-Tauber, Germany, as a son of a wealthy family. From 1818 to 1822, von
Staudt studied mathematics at Göttingen under Gauss. He was also interested in
astronomy (at that time Gauss was also the head of the observatory), and, as early
as in 1820, he computed the positions of Mars and the asteroid Pallas, and in 1821
he computed the orbit of a comet. The accuracy of the computation was highly
praised (described as an “outstanding proficiency”) by his master Gauss. He had
been holding positions as a high-school teacher at Würzburg and Nürnberg until he
was appointed professor at Erlangen in 1835, where he remained throughout his life.
His main contribution in mathematics is in Geometrie der Lage (1847, Geometry
of Position), which is the title of his book and is now called projective geometry. As
for Bernoulli numbers, he discovered not only the theorem introduced in Sect. 3.1
[96, found independently of Clausen], but also, according to Noether15 [74], several
other results including the formula in Theorem 2.8 (p. 35) and even Kummer’s

11
Johann Carl Friedrich Gauss (born on April 30, 1777 in Brunswick, Duchy of Brunswick (now
Germany)—died on February 23, 1855  in Göttingen, Hanover
 (now Germany)).
2˛;2ˇ;˛Cˇ
12
2 F1 .˛; ˇ; ˛ C ˇ C 12 I x/2 D 3 F2 ˛CˇC 12 ;2˛C2ˇ
Ix
13
It was correct up to 248 decimal places. This was the world record from 1847 to 1853. For a
history of computation of , see for example [13].
n
14
Fermat conjectured that all the numbers of the form 22 C 1 (which are called Fermat numbers)
are prime. This is true for n  4, but Euler showed in 1732 that this is not the case for n D 5 by
n
giving the factorization. As of November 2013, it is known that every Fermat number 22 C 1 with
5  n  32 is not prime. (Complete factorization is known up to n D 11.)
15
Max Noether (born on September 24, 1844 in Mannheim, Germany—died on December 13,
1921 in Erlangen, Germany).
48 3 Theorem of Clausen and von Staudt, and Kummer’s Congruence

congruence (modulo prime number case). He presented these results in two papers16
submitted to Erlangen University on the occasion of becoming a member of the
senate and the faculty. These results, however, caught no attention because, as
Noether put it, von Staudt was so modest and did not care about the circulation of
the papers (cf. [88] for the contents of these two papers). He also published a paper
[97] on a proof of the “fundamental theorem on algebra” of Gauss (Clausen also
wrote a paper on the same subject [25]). He wrote up a paper one week before his
death, while suffering from asthma. He died on June 1, 1867. Noether quotes from
the funeral address the passage “his endurance, prepared for the lifelong effort, not
for months or years, to pursue a distinct goal”, which he thinks describes well the
character of von Staudt.
Ernst Eduard Kummer was born on January 29, 1810 in Sorau, Germany
(now Zary, Poland). Kummer entered the Gymnasium in Sorau in 1819 and
the University of Halle in 1828. He soon gave up his original study, Protestant
theology, under the influence of the mathematics professor Scherk17 and applied
himself to mathematics. In 1831 he received a prize for his essay on the question
posed by Scherk: “De cosinuum et sinuum potestatibus secundum cosinus et sinus
arcuum multiplicium evolvendis”.18 Kummer taught from 1832 until 1842 at the
Gymnasium in Liegnitz (now Legnia, Poland), mainly mathematics and physics.
His students during this period included Kronecker and Joachimsthal,19 both of
whom became interested in mathematics through Kummer’s encouragement and
stimulation. In particular, the influence of Kummer upon Kronecker was enormous
and they became lifelong friends. One can see the fragments of their friendship in
their correspondence collected in Kummer’s Werke.
The subjects of Kummer’s mathematical study are rather clearly divided accord-
ing to the periods. The first is in function theory, the second in number theory
and the third in geometry. His period of Gymnasium teaching coincided with his
creative period in function theory. The most important was the famous paper on
the hypergeometric series [63]. Kummer sent this paper to Jacobi, which led to his
scientific connection with Jacobi and with Dirichlet.20 In 1839, through Dirichlet’s
proposal, Kummer became a member of the Berlin Academy of Sciences, and in
1840, he married Ottilie Mendelssohn, a cousin of Dirichlet’s wife.21 (She died in
1848 and later Kummer married Bertha Cauer.) On the recommendation of Dirichlet
and Jacobi, Kummer was appointed full professor at the University of Breslau (now

16
Under the same title “De numeris Bernoullianis”, Erlangen, 1845.
17
Heinrich Ferdinand Scherk (born on October 27, 1798 in Poznan, Poland—died on October 4,
1885 in Bremen, Germany), he also wrote a paper on Bernoulli numbers [81].
18
“On expansions of powers of cosine and sine by cosine and sine with their arguments multiplied.”
19
Ferdinand Joachimsthal (born on March 9, 1818 in Goldberg, Prussian Silesia (now Zlotoryja,
Poland)—died on April 5, 1861 in Breslau, Germany (now Wroclaw, Poland)).
20
Johann Peter Gustav Lejeune Dirichlet (born on February 13, 1805 in Düren, French Empire
(now Germany)—died on May 5, 1859 in Göttingen, Hanover (now Germany)).
21
Dirichlet’s wife Rebecca Mendelssohn is a younger sister of the composer Felix Mendelssohn.
3.3 Short Biographies of Clausen, von Staudt and Kummer 49

Wroclaw, Poland) in 1842. His second period of research, dominated especially by

number theory and lasting approximately 20 years, began about this time.
In 1855, Gauss died and Dirichlet left Berlin to succeed him at Göttingen. In the
same year Kummer was appointed professor at Berlin to succeed Dirichlet.
In his third period, starting from around 1860, he studied geometry, his most
famous work being the study of the quartic surface known today as the Kummer
surface. (He also published papers on the arithmetic of cyclotomic fields until the
mid 1870s.)
Kummer’s popularity as a professor, based not only on the clarity of his lectures
but on his charm and sense of humor as well, attracted a great number of students.
Kummer was first Gutachter for 39 dissertations at Berlin. Of his doctoral students,
seventeen later became university teachers, several of them famous mathematicians:
Fuchs,22 du Bois-Reymond,23 Gordan,24 Bachmann,25 Schwarz26 (Kummer’s son
in law), Cantor,27 Schönflies.28 Kummer was also second Gutachter for thirty
dissertations at Berlin.
He retired in 1883 and was succeeded by Fuchs. Kummer died on May 14 in
1893, 10 years after his retirement.
Exercise 3.4. Compute the integer Cn in Theorem 3.1 for several n. What is the
first n such that Cn ¤ 1? (Note that limn!1 jCn j D 1 because Cn is very close
to Bn .)
Exercise 3.5. Suppose n is even and n  4. Prove the following.
(1) 2Bn 2 Z.2/ .
(2) The congruence 2Bn  1 mod 4 holds. Hint: Use the formula in Theorem 2.8
and look at each term on the right modulo 4.
Exercise 3.6. Use Theorems 3.1 and 3.2 (1) to prove that the tangent number Tn at
the end of Chap. 1 is an integer.

22
Lazarus Immanuel Fuchs (born on May 5, 1833 in Moschin, Prussia (now Poznan, Poland)—died
on April 26, 1902 in Berlin, Germany).
23
Paul David Gustav du Bois-Reymond (born on December 2, 1831 in Berlin, Germany—died on
April 7, 1889 in Freiburg, Germany).
24
Paul Albert Gordan (born on April 27, 1837 in Breslau, Germany (now Wroclaw, Poland)—died
on December 21, 1912 in Erlangen, Germany).
25
Paul Gustav Heinrich Bachmann (born on June 22, 1837 in Berlin, Germany—died on March
31, 1920 in Weimar, Germany).
26
Karl Herman Amandus Schwarz (born on January 25, 1843 in Hermsdorf, Silesia (now
Poland)—died on November 30, 1921 in Berlin, Germany).
27
Georg Ferdinand Ludwig Philipp Cantor (born on March 3, 1845 in St. Petersburg, Russia—died
on January 6, 1918 in Halle, Germany).
28
Arthur Moritz Schönflies (born on April 17, 1853 in Landsberg an der Warthe, Germany (now
Gorzów, Poland)—died on May 27, 1928 in Frankfurt am Main, Germany).
Chapter 4
Generalized Bernoulli Numbers

In this chapter we introduce generalized Bernoulli numbers and Bernoulli

polynomials. Generalized Bernoulli numbers are Bernoulli numbers twisted by
a Dirichlet character, which we define at the beginning of the first section. Bernoulli
polynomials are generalizations of Bernoulli numbers with an indeterminate. These
two generalizations are related, and they will appear in various places in the
following chapters.

4.1 Dirichlet Characters

Let us define a Dirichlet character as a map from the set of integers Z to the set of
complex numbers C.
Definition 4.1. Let f be a natural number. A function  W Z ! C is called a
Dirichlet character modulo f if it satisfies the following three conditions:
(i)  is multiplicative; i.e. .ab/ D .a/.b/ for any a; b 2 Z.
(ii) The value of .a/ only depends on a mod f ; i.e. .a C bf / D .a/ for any
a; b 2 Z.
(iii) If .a; f / D 1, then .a/ ¤ 0, and if .a; f / ¤ 1, then .a/ D 0. Here, .a; f /
stands for the greatest common divisor of a and f .
In particular, the function 0 defined by

1 .a; f / D 1;
0 .a/ D
0 .a; f / ¤ 1

satisfies the three conditions. We call this the trivial character (modulo f ).

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 51

in Mathematics, DOI 10.1007/978-4-431-54919-2__4, © Springer Japan 2014
52 4 Generalized Bernoulli Numbers

For a natural number f , let '.f / be the number of integers from 1 to f which
are relatively prime to f . This is called the Euler (totient) function. If a is an integer
relatively prime to f , then we have

a'.f /  1 mod f:

This is known as Euler’s theorem in elementary number theory. For a proof see for
example Ireland and Rosen [50].
A Dirichlet character  modulo f satisfies the following properties.
Lemma 4.2. (1) If a  1 mod f , then .a/ D 1.
(2) If .a; f / D 1, then .a/'.f / D 1.
(3)

'.f / if  D 0 (trivial character),

.a/ D
0 if  ¤ 0 .
a mod f

The summation is taken over all representatives a modulo f .

Proof. (1) Putting b D 1 in condition (i), we have .a/ D .a/.1/. If a 
1 mod f , then we have .a/ ¤ 0 from (iii). Thus, we have .1/ D 1, which in
turn implies .a/ D 1 from (ii).
(2) Suppose .a; f / D 1. From Euler’s theorem we have a'.f /  1 mod f . Thus,

.a/'.f / D .a'.f / / D .1/ D 1:

(3) Since '.f / is the number of a mod f such that .a; f / D 1, the assertion
holds if  is the trivial character. If  is not thePtrivial character, then there
exists b such
P that .b/ ¤ 0; 1. If we
P put S D a mod f .a/, then we have
.b/S D a mod f .b/.a/ D a mod f .ba/. When a runs through all
the representatives modf , so does ba since .b; f / D 1. Thus, the sum on the
right equals S . This implies .1.b//S D 0 and hence S D 0, as 1.b/ ¤ 0.
t
u
We note that a Dirichlet character defines naturally by definition a character of
the multiplicative group .Z=f Z/ .
Now suppose f 0 j f , and let 0 be a Dirichlet character modulo f 0 . If  satisfies

.a/ D 0 .a/ if .a; f / D 1;

then we say that  is defined mod f 0 . For a Dirichlet character  modulo f , the
smallest f 0 such that  is defined modf 0 is called the conductor of , and is
written as f . By definition, f is a divisor of f .
Definition 4.3 (primitive character). A Dirichlet character  modulo f is called
primitive if f D f .
4.2 Generalized Bernoulli Numbers 53

Let  be a Dirichlet character modulo f . Since .1/2 D ..1/2 / D .1/ D 1,

we see that .1/ D ˙1.
A Dirichlet character with .1/ D 1 is called an even character. A character
with .1/ D 1 is called an odd character. Given a Dirichlet character , we
define the complex conjugate character  by

.a/ D .a/;

where .a/ stands for the complex conjugate of .a/. We have f D f .

Example 4.4. Define j W Z ! C .j D 0; 1; 2; 3/ as follows.
If .a; 2/ ¤ 1 .i.e. 2ja/, then j .a/ D 0 .j D 0; 1; 2; 3/. When a is odd, for each
a mod 8, .a/ is given in the following table:

na 1 mod 8 3 mod 8 5 mod 8 7 mod 8

0 .a/ 1 1 1 1
1 .a/ 1 1 1 1
2 .a/ 1 1 1 1
3 .a/ 1 1 1 1

Then, j .j D 0; 1; 2; 3/ are all the Dirichlet characters modulo 8. The character 0

is the trivial character with f0 D 1 and 1 , 2 are primitive characters modulo 8,
but 3 .D 1 2 / is not primitive, with f3 D 4.

4.2 Generalized Bernoulli Numbers

Given a Dirichlet character  modulo f , the generalized Bernoulli numbers Bn;

are defined by using the generating function

Xf
X1
.a/te at tn
D Bn; :
aD1
ef t  1 nD0
nŠ

In the following we assume f > 1 and  is not the trivial character unless otherwise
stated.
Since we have the expansion
 
te at 1 a 1
D C  t
ef t  1 f f 2
 2   3 
a f t2 a 3a2 a f t3
C aC C  C C ;
f 6 2 f 2 2 6
54 4 Generalized Bernoulli Numbers

it follows from Lemma 4.2 (3) that

B0; D 0;

1 X
f
B1; D .a/a;
f aD1

1 X X
f f
B2; D .a/a2  .a/a;
f aD1 aD1

1 X 3X f X
f f f
B3; D .a/a 
3
.a/a C
2
.a/a:
f aD1 2 aD1 2 aD1

A similar formula for Bn; is given as

X
f
Bn; D f n1
.a/Bn .a=f / ; (4.1)
aD1

where Bn .x/ is the Bernoulli polynomial, which will be defined in the next section.
(See Remark 4.10 (2) in the next section.)
We have seen that all the Bernoulli numbers with odd indices greater than 1 are 0.
For generalized Bernoulli numbers we have the following.
Proposition 4.5. Let  be a non-trivial character. Then, for any n satisfying
.1/n1 D .1/, we have Bn; D 0. In other words, if  is an even character,
then Bn; with odd indices n are 0; if  is an odd character, then Bn; with even
indices n are 0.
Proof. Since  is non-trivial, we can rewrite the generating function as follows:

f 1 f 1
X .a/te at X .f  a/te .f a/t
D
aD1
ef t  1 aD1
ef t  1
f 1
X .a/te at
D .1/
aD1
1  e f t
f 1
X .a/.t/e a.t /
D .1/ :
aD1
e f .t /  1

It follows immediately from this that the generating function is an even function if
.1/ D 1, and an odd function if .1/ D 1. t
u
4.3 Bernoulli Polynomials 55

Remark 4.6. For any n satisfying .1/n D .1/, it is known that Bn; is non-zero.
This can be seen as follows. From Theorem 9.6 on p. 148, Bn; is non-zero if and
only if L.n; / is non-zero. We can prove L.n; / ¤ 0 using the Euler product
formula for the L-function with n > 1. The fact that L.1; / ¤ 0, i.e., B1; ¤ 0 for
all , is essentially equivalent to Dirichlet’s theorem on prime numbers in arithmetic
progressions. Dirichlet proved his theorem using the class number formula. See
Dirichlet [29] or Serre [83, Chap. 6].

4.3 Bernoulli Polynomials

Bernoulli polynomials are often defined by means of a generating function. Here,

we define them in a different manner.1 Their generating function will be given in
Proposition 4.9 (5).
First, define a Q-linear map I W QŒx ! QŒx from the polynomial ring QŒx to
itself by
Z xC1
I.f / D f .y/ dy .f .x/ 2 QŒx/
x

(difference of the primitive function). Since we see immediately that

1  
I.x n / D .x C 1/nC1  x nC1 D x n C .lower terms/;
nC1

the matrix representing I with respect to the basis 1; x; x 2 ; x 3 ; : : : of QŒx is an

upper triangular matrix with the diagonal components 1. In particular, I is invertible.
We thus define the Bernoulli polynomials as follows.
Definition 4.7. Define the Bernoulli polynomials Bn .x/ (n D 0; 1; 2; : : : ) by

Bn .x/ WD I 1 .x n /:

In other words, Bn .x/ is the unique polynomial satisfying

Z xC1
I.Bn .x// D Bn .y/ dy D x n :
x

Because of the form of I.x/, Bn .x/ is a monic polynomial (a polynomial in which

the coefficient of the highest-degree term is 1) of degree n with coefficients in Q.

1
This was suggested by Don Zagier.
56 4 Generalized Bernoulli Numbers

Example 4.8.

1 1 3 1
B0 .x/ D 1; B1 .x/ D x  ; B2 .x/ D x 2  x C ; B3 .x/ D x 3  x 2 C x;
2 6 2 2
1 5 5 1
B4 .x/ D x 4  2x 3 C x 2  ; B5 .x/ D x 5  x 4 C x 3  x:
30 2 3 6

Proposition 4.9. (1) Bn .1/ D Bn .n  0/. Also, if n ¤ 1 , then Bn .0/ D

Bn .1/ D Bn .
(2) Bn .x C 1/  Bn .x/ D nx n1 .n  0/.
(3) Bn0 .x/ D nBn1 .x/ .n  1/.
(4) Bn .1  x/ D .1/n Bn .x/ .n  0/.
(5) (Generating function)
1
X tn te xt
Bn .x/ D t : (4.2)
nD0
nŠ e 1

(6) (Explicit formula with Bernoulli numbers)

!
X n
j n
Bn .x/ D .1/ Bj x nj :
j D0
j

(7)

X
k1  i
Bn x C D k 1n Bn .kx/ .k  1/:
i D0
k

Z xC1
Proof. (2) It is obtained by differentiating the defining equation Bn .y/ dy D
x
x n with respect to x.
(3) It follows from (2) that
Z xC1
Bn0 .y/ 1
dy D .Bn .x C 1/  Bn .x// D x n1 :
x n n

Thus, it follows from the definition that

Bn0 .x/
D Bn1 .x/;
n
or

Bn0 .x/ D nBn1 .x/:

4.3 Bernoulli Polynomials 57

(4) Since we have

Z xC1 Z x Z 1x
Bn .1y/ dy D  Bn .y/ dy D Bn .y/ dy D .x/n D .1/n x n ;
x 1x x

it follows from the definition that .1/n Bn .1  x/ D Bn .x/.

Applying I to the coefficient of each term of the left-hand side of (4.2), we have
(5) P
nD0 x t =nŠ D e as formal power series in t. On the other hand, regarding
n n xt

the right-hand side of (4.2) as a formal power series, and applying I term by
term, we have I.te xt =.e t  1// D tI.e xt /=.e t  1/ because of the linearity.
Furthermore, since we have
Z xC1
e .xC1/t  e xt e t  1 xt
I.e xt / D e yt dy D D e ;
x t t

we see I.te xt =.e t  1// D e xt . The injectivity of I implies that both sides are
equal.
(1) The first half of (1) can be seen by putting x D 1 in (5) and by using the
generating function of Bernoulli numbers (Theorem 1.12 on p. 20). Putting
x D 0 in (2), we see that Bn .1/ D Bn .0/ for n ¤ 1. The second half of the
assertion follows from this and the first half.
(6) We compute the product t=.e t 1/e xt on the right-hand side of the generating
function (4.2). Since
X 1
t .t/e t tn
D D .1/ n
Bn ;
et  1 e t  1 nD0
nŠ

we have
1
! 1 !
t X tn X .xt/n
e D
xt n
.1/ Bn
et  1 nD0
nŠ nD0
nŠ
0 ! 1
X1 Xn
@ .1/j n tn
D Bj x nj A :
nD0 j D0
j nŠ

This proves the formula (6).

(7) We have
Z xC1 X
k1   k1 Z
X  
y i xC1
yCi
Bn C dy D Bn dy
x i D0
k k i D0 x k
k1 Z
X xC1Ci
 
k
yCi
D Bn .y/ kdy k
!y
xCi
i D0 k
58 4 Generalized Bernoulli Numbers

Z x
k C1
Dk Bn .y/ dy
x
k
 x n
Dk D k 1n x n :
k
From this we have

X
k1  
x i
Bn C D k 1n Bn .x/:
i D0
k k

The formula is obtained by letting x ! kx. t

u
Remark 4.10. (1) As we mentioned in Remark 1.3 on p. 10, it is easy to prove the
formula for the sum of powers ((1.1) on p. 1) using the Bernoulli polynomials.
Replacing n by k C 1 in Proposition 4.9 (2), we have

1
xk D .BkC1 .x C 1/  BkC1 .x// :
kC1

Putting x D 1; 2; : : : ; n, and adding them all up, we have

X
n
1
ik D .BkC1 .n C 1/  BkC1 .1// :
i D1
kC1

We then use Proposition 4.9 (1), (4), (6) to obtain

BkC1 .n C 1/  BkC1 .1/ D .1/kC1 BkC1 .n/  BkC1

!
X
kC1
j kC1
D .1/ kC1
.1/ Bj .n/kC1j  BkC1
j D0
j
!
X k
kC1
D Bj nkC1j :
j D0
j

  k  1
1 kC1
Since kC1 j
D j kC1j , the formula (1.1) follows.
(2) The formula (4.1) which expresses generalized Bernoulli numbers in terms of
Bernoulli polynomials is obtained as follows. In (4.2) replace x by a=f and t
by f t. Then we obtain
1
te t a 1 X .f t/n
D Bn .a=f / :
ef t  1 f nD0 nŠ
4.3 Bernoulli Polynomials 59

Thus, we have

1
X X X 1
1 X
f f
tn .a/te t a .f t/n
Bn; D D .a/ Bn .a=f /
nD0
nŠ aD1
e 1
f t
aD1
f nD0 nŠ
1 X
X f
f n1 t n
D .a/Bn .a=f / :
nD0 aD1
nŠ

Comparing coefficients, we obtain

X
f
Bn; D f n1 .a/Bn .a=f /:
aD1

Next we take up a beautiful formula involving Bernoulli polynomials. For any

real number x, we denote by Œx the greatest integer less than or equal to x. This is
sometimes called the Gauss symbol .
Theorem 4.11. Let k be a natural number. The formula

kŠ X e 2 i nx
Bk .x  Œx/ D  (4.3)
.2 i /k nk
n2Z
n¤0

holds for all real numbers x if k  2; it holds for all real numbers x 62 Z, if k D 1.
Here, the sum is taken for all integers different from 0. If k D 1, the infinite sum on
the right-hand side should be understood as

X
N
e 2 i nx
lim :
N !1 n
nDN
n¤0

If k  2, then the right-hand side of (4.3) converges absolutely and uniformly with
respect to x.
Proof. Let 0 < x < 1. Decompose the function

e x
f ./ D
e 1
into partial fractions as a meromorphic function in . This f ./ has poles at
 D 2 i n .n 2 Z/ and all of them are of order 1. The residue at  D 2 i n is
given by

e x.t C2 i n/
lim .  2 i n/f ./ D lim t  D e 2 i nx :
!2 i n t !0 e t C2 i n  1
60 4 Generalized Bernoulli Numbers

Let N be a sufficiently large natural number, and put R D 2.N C1=2/. Let CN be a
square path passing through four corner points RCiR; RCiR; RiR; RiR
in this order in the -plane. If t is a point inside CN such that t ¤ 2 i n .n 2 Z/,
then it follows from the residue theorem that
Z !
f ./ X
N
e 2 i nx
d  D 2 i f .t/ C
CN   t nDN
2 i n  t
!
1 XN
e 2 i nx
D 2 i f .t/   :
t t  2 i n
nDN
n¤0

As N ! 1, one can show that

Z
f ./
d  ! 0:
CN t

Therefore, if 0 < jtj < 2, we have

1 X
N
e 2 i nx
f .t/ D C lim
t N !1 t  2 i n
nDN
n¤0

1 X
N
e 2 i nx
D  lim  
t N !1 2 i n 1  2ti n
nDN
n¤0

X 1  k1
e 2 i nx X
N
1 t
D  lim :
t N !1 2 i n 2 i n
nDN kD1
n¤0

Since the sum

X1 X1
e 2 i nx k1
t
nD1
.2 i n/k
kD2
n¤0

converges absolutely, the above formula becomes

X
N 1 X
X 1
1 e 2 i nx e 2 i nx k1
f .t/ D  lim  t :
t N !1 2 i n nD1 .2 i n/k
nDN kD2
n¤0 n¤0
4.3 Bernoulli Polynomials 61

Since the generating function (4.2) of Bernoulli polynomials can be regarded as the
Taylor expansion of te xt =.e t  1/ at t D 0 for each real number x, we have
1
X t k1
f .t/ D Bk .x/ :
kŠ
kD0

Thus, comparing the coefficients, we obtain

X
N
e 2 i nx
B1 .x/ D  lim ;
N !1 2 i n
nDN
n¤0
1
X e 2 i nx
Bk .x/ D kŠ .k  2/:
nD1
.2 i n/k
n¤0

Since the sum on the right-hand side converges uniformly over the closed interval
Œ0; 1 if k  2, and since Bk .x/ is continuous in the same interval, the above formula
holds in Œ0; 1. Furthermore, the right-hand side is periodic with respect to x with
period 1. Thus, (4.3) is obtained by extending the left-hand side. t
u
From this theorem we can obtain the values of the Riemann zeta function at
positive even integers. (The Riemann zeta function will be treated in the next
section.)
Corollary 4.12. Let k be an even integer greater than or equal to 2. Then we have

X1
1 1 Bk
k
D .2 i /k : (4.4)
nD1
n 2 kŠ

For example,

X1 X1 X1 X1
1 2 1 4 1 6 1 8
D ; D ; D ; D ; etc:
nD1
n2 6 nD1
n4 90 nD1
n6 945 nD1
n8 9450

Proof. It suffices to put x D 0 in the theorem. The left-hand side is equal to Bk from
Proposition 4.9 (1). Since the right-hand side is twice the sum over the positive n
since k is even, the formula in question holds. t
u
Remark 4.13. The formula in the theorem can be rewritten as follows.
(1) If k is even,

X1
cos 2 nx
Bk .x  Œx/ D 2.1/k=21 kŠ :
nD1
.2 n/k
62 4 Generalized Bernoulli Numbers

(2) If k is odd,
X1
sin 2 nx
Bk .x  Œx/ D 2.1/.kC1/=2kŠ : (4.5)
nD1
.2 n/k

Here, we assume x is not an integer if k D 1.

Theorem 4.11 has an interesting application.2 Let us consider the special values
of the following zeta function. Let a, b be integers different from 0, and let p, q, r
be positive even integers. Define
X 1
S.p; q; rI a; b/ D :
mp nq .am C bn/r
m¤0; n¤0
amCbn¤0

Here, the sum runs over all integers m; n satisfying m ¤ 0; n ¤ 0; am C bn ¤ 0.

Proposition 4.14. Let p; q; r be positive integers. Then, we have

 pqr S.p; q; rI a; b/ 2 Q:

Proof. Define l D .am C bn/. We can express S.p; q; rI a; b/ as an integral as

follows (we use the notation e.w/ D e 2 i w ):
X 1
S.p; q; rI a; b/ D
mp nq l r
m; n; l2Z
m; n; l¤0; amCbnClD0

X Z 1
e..am C bn C l/x/
D dx
0 mp nq l r
m; n; l2Z
m; n; l¤0
Z X
1
e..am C bn C l/x/
D dx
0 mp nq l r
m; n; l2Z
m; n; l¤0
Z 1 X e.amx/ X e.bnx/ X e.lx/
D dx:
0 mp nq lr
m2Z n2Z l2Z
m¤0 n¤0 l¤0
P
Here, we used the fact that the series m¤0 e.amx/=mp converges uniformly over
the closed interval Œ0; 1, because p; q; r  2. Using the formula (4.3), we have

2
This is based on a lecture of Don Zagier at Kyushu University in 1999.
4.3 Bernoulli Polynomials 63

pCqCr2 Z 1
.1/ 2 .2/pCqCr
S.p; q; rI a; b/ D Bp .axŒax/Bq .bxŒbx/Br .x/ dx:
pŠqŠrŠ 0

Thus, the value of the integral is a rational number (recall that the Bernoulli
polynomial is a polynomial with rational coefficients), and the assertion of the
proposition follows. t
u
Example 4.15. (p D q D r D 2, a D b D 1.) From the above formula,
Z 1
.2/6
S.2; 2; 2I 1; 1/ D B2 .x/3 dx:
8 0

Integration by parts using B10 .x/ D 1; B20 .x/ D 2B1 .x/ gives
Z 1 Z 1
B2 .x/3 dx D B10 .x/B2 .x/3 dx
0 0
Z

1 1
D B1 .x/B2 .x/3 0  6B1 .x/2 B2 .x/2 dx
0
Z 1  0
D B23  2 B1 .x/3 B2 .x/2 dx
0
Z

1 8 1  0
D B23  2 B1 .x/3 B2 .x/2 0 C B1 .x/5 B2 .x/ dx
5 0
Z
1 8
1 16 1
D  B22 C
B23 B1 .x/5 B2 .x/ 0  B1 .x/6 dx
2 5 5 0
1 1 16
1
D B23  B22 C B2  B1 .x/7 0
2 10 35
1
D 2 3 :
2 3 57
Thus, we have

2 6
S.2; 2; 2I 1; 1/ D :
33  5  7

Exercise 4.16. For a non-trivial character  modulo f > 1, write B4; and B5; by
Pf
aD1 .a/a with i D 1, 2, 3, 4, 5.
i

Exercise 4.17. Prove that the Bernoulli polynomial Bn .x/ for odd n > 1 is always
divisible by x.x  12 /.x  1/.
Chapter 5
The Euler–Maclaurin Summation Formula
and the Riemann Zeta Function

Aside from the formula for the sum of powers, the most basic topics in which
Bernoulli numbers appear is the Euler–Maclaurin1 summation formula and the
values of the Riemann zeta function at integer arguments. In this chapter we survey
these topics.

5.1 Euler–Maclaurin Summation Formula

The Euler–Maclaurin summation formula gives a very effective tool for evaluating
a sum of values of a function at integers.
Theorem 5.1. Let a and b be integers satisfying a  b, and let M be a natural
number. Suppose f .x/ is an M times continuously differentiable function2 over
Œa; b. Then, we have

X
b Z b X BkC1 M 1
1
f .n/ D f .x/dxC .f .a/Cf .b// C .f .k/ .b/  f .k/ .a//
nDa a 2 .k C 1/Š
kD1
M Z b
.1/
 BM .x  Œx/f .M / .x/dx:
MŠ a

Here, BkC1 is the Bernoulli number and BM .x/ is the Bernoulli polynomial, and
the sum on the right-hand side is understood to be 0 if M D 1.

1
Colin Maclaurin (born in February 1698 in Kilmodan, Cowal, Argyllshire, Scotland—died on
June 14, 1746 in Edinburgh, Scotland).
2
Differentiable M times and the M th derivative is continuous.

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 65

in Mathematics, DOI 10.1007/978-4-431-54919-2__5, © Springer Japan 2014
66 5 The Euler–Maclaurin Summation Formula and the Riemann Zeta Function

Remark 5.2. (1) Regard f .n/ as the area of a rectangle whose base is the interval
Œn  12 ; n C 12  and whose height is f .n/. Then the sum of the first two terms
on the right-hand side is the first approximation, so to say. The next term can
be seen as the higher approximation in terms of the data at the end points of the
interval.
(2) If we use the values of the Riemann zeta function at negative integers .k/ D
BkC1 =.k C 1/, which will be proved later, then the formula looks like

X
b Z b X .k/ M 1
1
f .n/ D f .x/dxC .f .a/Cf .b//  .f .k/ .b/f .k/ .a//
nDa a 2 kŠ
kD1
Z b
.1/M
 BM .xŒx/f .M / .x/dx:
MŠ a

Proof. Let g.x/ be an M times continuously differentiable function on Œ0; 1. Since
B1 .x/ D x  1=2, we have B10 .x/ D 1. Thus, using integration by parts, we have
Z 1 Z 1
g.x/dx D ŒB1 .x/g.x/10  B1 .x/g 0 .x/ dx
0 0
Z 1
1
D .g.1/ C g.0//  B1 .x/g 0 .x/ dx:
2 0

0
Iterating integration by parts using Bk .x/ D BkC1 .x/=.k C 1/ (Proposition 4.9 (3)
on p. 56), we obtain
Z Z
1
1 1
1 1 1
g.x/dx D .g.1/ C g.0//  B2 .x/g 0 .x/ 0 C B2 .x/g 00 .x/ dx
0 2 2 2 0
1 1
1
D .g.1/ C g.0//  B2 .x/g 0 .x/ 0
2 2
Z 1
1
1 1
C B3 .x/g 00 .x/ 0  B3 .x/g 000 .x/ dx
23 23 0
D 
X
M 1
1 .1/k
1
D .g.1/ C g.0// C BkC1 .x/g .k/ .x/ 0
2 .k C 1/Š
kD1
M Z 1
.1/
C BM .x/g .M / .x/ dx:
MŠ 0

Now, if k  2, then Bk .0/ D Bk .1/ D Bk (Proposition 4.9 (1)) and .1/k Bk D Bk

(Proposition 1.4 on p. 10). Thus, we have
5.2 The Riemann Zeta Function 67

Z 1 X
M 1
1 BkC1
g.x/dx D .g.1/ C g.0//  .g .k/ .1/  g .k/ .0//
0 2 .k C 1/Š
kD1
M Z 1
.1/
C BM .x/g .M / .x/ dx:
MŠ 0

Therefore,
Z 1 X
M 1
1 BkC1
.g.1/ C g.0// D g.x/dx C .g .k/ .1/  g .k/ .0//
2 0 .k C 1/Š
kD1
M Z 1
.1/
 BM .x/g .M / .x/dx:
MŠ 0

Now, we take f .x C n/ (a  n  b  1) for the function g.x/. Since the function

BM .x  Œx/, which is obtained by extending BM .x/ from the interval Œ0; 1 to the
entire real line by periodicity, satisfies BM .x  n/ D BM .x  Œx/ for any integer n
and n  x  n C 1, we obtain
Z nC1 X
M 1
1 BkC1
.f .n C 1/ C f .n// D f .x/dx C .f .k/ .n C 1/  f .k/ .n//
2 n .k C 1/Š
kD1
M Z nC1
.1/
 BM .x  Œx/f .M / .x/dx:
MŠ n

The formula in the theorem is obtained by adding up the above formula from n D a
to n D b  1 and then adding 12 .f .a/ C f .b// to both sides. t
u

5.2 The Riemann Zeta Function

The exponential function e z with the complex variable z is defined by the formula

z2 zn
ez D 1 C z C C  C C  :
2Š nŠ
The power series on the right-hand side converges absolutely for any z 2 C, and
thus it is a holomorphic function over the entire z-plane. From the formula for the
product of two power series, we have the exponential law

e zCw D e z e w :

In particular, for x, y 2 R, we have Euler’s formula

e xCiy D e x e iy D e x .cos y C i sin y/:

68 5 The Euler–Maclaurin Summation Formula and the Riemann Zeta Function

For any positive real number a, the complex power as for s 2 C is defined by

as D e s log a :

Its absolute value is given by

jas j D e  log a ;

where  D Re.s/.
Let s be a complex number. Consider the sum

XN
1 1 1
s
D 1C s C  C s:
nD1
n 2 N

Here, ns is a complex power of n just defined above. Let f W .0; 1/ ! C be the

function defined by

f .x/ D x s :

We apply the Euler–Maclaurin summation formula to this f .x/ with a D 1; b D N .

The k-th derivative of f is given as

f .k/ .x/ D .s/.s  1/    .s  k C 1/x sk

D .1/k s.s C 1/    .s C k  1/x sk :

Also, if s ¤ 1, then we have

Z N  
s 1 1
x dx D 1  :
1 s1 N s1

It follows from Theorem 5.1 that for s ¤ 1

XN  
1 1 1
D 1 
nD1
ns s  1 N s1
  M 1
! 
1 1 X BkC1 sCk1 1
C 1C s C  1  sCk
2 N kC1 k N
kD1
!Z
sCM 1 N
 BM .x  Œx/x sM dx; (5.1)
M 1

 
where xi is the binomial coefficient x.x  1/    .x  i C 1/= i Š, and we used the
RN
fact .1/k BkC1 D BkC1 . If s D 1, then it follows from 1 x 1 dx D log N that
5.2 The Riemann Zeta Function 69

XN   MX1  
1 1 1 BkC1 1
D log N C 1C C 1  1Ck
nD1
n 2 N kC1 N
kD1
Z N
 BM .x  Œx/x 1M dx: (5.2)
1

As an application of these formulas, we obtain the following elementary result.

X1
1
Proposition 5.3. (1) The series s
converges absolutely if Re.s/ > 1, and the
nD1
n
1
X 1
series diverges.
nD1
n
X
N
1
(2) Let faN g be the sequence defined by aN D log N . Then faN g converges.
nD1
n
The limit  D lim aN is called Euler’s constant.3 For any natural number M ,
N !1
we have
X
M 1 Z 1
BkC1 BM .x  Œx/
D  dx:
kC1 1 x M C1
kD0

In particular, setting M D 1, we have

Z 1
x  Œx
 D1 dx: (5.3)
1 x2

XN ˇ ˇ
X
ˇ1ˇ N
1
Proof. (1) Let Re.s/ D . Since ˇ ˇ
ˇ ns ˇ D n
, by putting s D , M D 1
nD1 nD1
in formula (5.1), we obtain

XN     Z N
1 1 1 1 1
D 1   1
C 1 C  B1 .x Œx/x  1 dx:
nD1
n   1 N 2 N 
1

Since  > 1, both 1=N  1 and 1=N  converge to 0 as N ! 1. As for the

integral in the last term, first note that the function BM .x  Œx/ is bounded
since 0  x  Œx  1. Thus, if
Z 1 Z N
x  1 dx D lim x  1 dx (5.4)
1 N !1 1

3
It is unknown whether  D 0:5772156649015328606065120900824    is an irrational number
or not.
70 5 The Euler–Maclaurin Summation Formula and the Riemann Zeta Function

converges, then so does

Z N
BM .x  Œx/x  1 dx
1

as N ! 1. The integral (5.4) converges since if  > 0, we have

Z N  
1 1 1
lim x  1 dx D lim 1  D :
N !1 1 N !1  N 
X1
1
Summing it all up, we see that converges absolutely if  > 1. Also, the
nD1
ns
X1
1
fact that diverges follows immediately from (5.2).
nD1
n
(2) It follows from the formula (5.2) that

XN   MX1  
1 1 1 BkC1 1
aN D  log N D 1C C 1  1Ck
nD1
n 2 N kC1 N
kD1
Z N
 BM .x  Œx/x 1M dx:
1

Similarly to (1), the last integral converges, and thus faN g converges.
Letting N ! 1, we obtain

X
M 1 Z 1
1 BkC1
 D lim aN D C  BM .x  Œx/x 1M dx
N !1 2 kC1 1
kD1

X
M 1 Z 1
BkC1
D  BM .x  Œx/x 1M dx:
kC1 1
kD0

(Here we used B1 D 1=2.) If M D 1, then we have B1 .x  Œx/ D x  Œx  1

2 and
Z 1 
dx 1 1
D  D 1:
1 x2 x 1
Thus, we have
Z 1
1 x  Œx  12
D  dx
2 1 x2
Z 1
x  Œx
D1 dx:
1 x2
t
u
5.2 The Riemann Zeta Function 71

The famous Riemann4 zeta function .s/ is defined for all complex numbers s
satisfying Re.s/ > 1 by
X1
1
.s/ D s
:
nD1
n

As we have already seen, the infinite series on the right-hand side converges
absolutely for Re.s/ > 1. If  is a positive number greater than 1, then this series
converges uniformly in Re.s/  , and thus it is a holomorphic function of s in
Re.s/ > 1. Let us study properties of .s/, using the summation formula (5.1).
Suppose Re.s/ > 1, and let N ! 1 in (5.1). Then,

1 1
!
X 1 1 1 X
M
BkC1 sCk1
D C C  (5.5)
nD1
ns s1 2 kC1 k
kD1
!Z
1
sCM 1
 BM .x  Œx/x sM dx:
M 1

Now, take the difference between (5.1) and (5.5). Then, for Re.s/ > 1, we have
1
!
X N
1 1 1 1 X
M
BkC1 sCk1 1
.s/  D  s1  C   sCk
nD1
n s s  1 N 2N s k C 1 k N
kD1
!Z
1
sCM 1
 BM .x  Œx/x sM dx: (5.6)
M N

Since the integral of the right-hand side converges absolutely in the region Re.s/ >
1  M , the formula (5.6) gives an analytic continuation of .s/ to a meromorphic
function in the region Re.s/ > 1  M . In other words, .s/ can be extended to
Re.s/ > 1  M through the formula (5.6). Since M can be any natural number,
.s/ has an analytic continuation to the entire s-plane as a meromorphic function.
It also follows from (5.6) that its only pole is a simple pole located at s D 1, and the
residue at s D 1 equals 1.
We summarize all this, and find the values of .s/ at integers less than or equal
to 0.
Theorem 5.4. (1) The function .s/ has an analytic continuation to the entire
s-plane as a meromorphic function, is holomorphic if s ¤ 1, and has a pole
of order 1 at s D 1 with residue 1.

4
Georg Friedrich Bernhard Riemann (born on September 17, 1826 in Breselenz, Germany—died
on July 20, 1866 in Selasca, Italy).
72 5 The Euler–Maclaurin Summation Formula and the Riemann Zeta Function

(2) Let m be a positive integer. The value of .s/ at s D 1  m is given by

Bm
.1  m/ D  :
m

Proof. Since we have already seen (1), we prove (2). If s D 1  m, taking M in

(5.6) with M  m, the binominal coefficient in front of the integral on the right-
hand side vanishes. This implies that .1  m/ is expressed as a finite sum. Our goal
is to express it in a simpler form. In (5.6), let s D 1  m, and M D m. Then we
have
X
N
N m N m1 X BkC1
m1
.1  m/  nm1 D   C
nD1
m 2 .k C 1/Š
kD1

 .1  m/.2  m/    .k  m/  N mk1 : (5.7)

If m D 1, then there is no term of summation, and

X
N
1 1
.0/ D 1N  D  D B1 :
nD1
2 2

Suppose m  2. Now,
  by using the
m1formulas
 .1  m/.2  m/    .k  m/ D
.1/k kŠ m1
k , 1 m1
kC1 k D 1
mk1 kC1 if k < m  1, and .1/k Bk D Bk for
k  2, we have

X
m1
BkC1
 .1  m/.2  m/    .k  m/  N mk1
.k C 1/Š
kD1
!
X
m1
k BkC1 m  1
D .1/ N mk1
kC1 k
kD1
!
X
m2
m  1 N mk1 Bm
D .1/k BkC1 C .1/m1
kC1 mk1 m
kD1
!
X
m1
k1 m  1 N mk Bm
D .1/ Bk C .1/m1
k mk m
kD2
!
X m1
m1
N mk Bm
D Bk  :
k mk m
kD2

Therefore, the right-hand side of (5.7) equals

!
X m1
m1
N mk Bm
 Bk 
k mk m
kD0
5.2 The Riemann Zeta Function 73

(since B0 D 1; B1 D 1=2). On the other hand, from the formula for the sum of
powers (1.1) on p. 1, we have
!
X
N X
m1
m1 N mk
m1
n D Bk : (5.8)
nD1
k mk
kD0

Thus, from (5.7) we have the formula

Bm
.1  m/ D  :
m
t
u
Remark 5.5. The value in (2) was first given by Euler ([32], see also Weil5 [103]).
The values of the Riemann zeta function at positive even integers in Corollary 4.12
(p. 61)

.1/k1 B2k
.2k/ D .2/2k .k 2 N/
2 .2k/Š
were also computed first by Euler (ibid.). There is not much known about the values
at positive odd integers. It is conjectured that all .2k C 1/ are transcendental
numbers.6 Apéry7 [5] proved that .3/ is an irrational number, and Rivoal [80]
proved that there are infinitely many irrational numbers among .2k C 1/ .k D
1; 2; 3; : : :/ (For more recent results, see [99].)
Remark 5.6. The formula (5.8) for the sum of powers can also be obtained
P usingm1
the
Euler–Maclaurin summation formula (Theorem 5.1). Indeed, the sum N nD1 n
can be obtained by letting f .x/ D x m1 , a D 1, and b D N in the summation
formula.
Exercise 5.7. Use (5.6) (N D M D 1) and (5.3) to prove
 
1
lim .s/  D :
s!1 s1

Exercise 5.8. Use (5.6) (N D M D 1) to deduce the formula

Z 1
x  Œx  12
.s/ D s dx
0 x sC1

5
André Weil (born on May 6, 1906 in Paris, France—died on August 6, 1998 in Princeton, USA).
6
Since  is a transcendental number [71], .2k/ are all transcendental numbers.
7
Roger Apéry (born on November 14, 1916 in Ruen, France—died on December 18, 1994 in Caen,
France).
74 5 The Euler–Maclaurin Summation Formula and the Riemann Zeta Function

which is valid for 1 < Re.s/ < 0.

Exercise 5.9. Use the previous exercise and the formula (proved in Chap. 4, (4.5))
X1
1 sin 2 nx
x  Œx  D
2 nD1
n

to deduce the functional equation of the Riemann zeta function

s
.s/ D 2s s s1 .s/ sin .1  s/:
2
(This way of proof is rigorously described in Chap. 2.1 in [90]. We give a different
proof of (the symmetric form of) the functional equation in Chap. 9.)
Exercise 5.10. Carry out the computation suggested in Remark 5.6.
Chapter 6
Quadratic Forms and Ideal Theory
of Quadratic Fields

There are close relations between Bernoulli numbers and quadratic fields or
quadratic forms. In order to explain those, we give a survey of the ideal theory
of quadratic fields and quadratic forms. Since Gauss, it is well known that there is a
deep relation between the ideal theory of quadratic fields (i.e. quadratic extensions
of the rational number field) and integral quadratic forms. This is obvious for
specialists, but textbooks which explain this in detail are rare. In most textbooks,
they treat the correspondence of ideals only in the case of maximal orders (the
ring of all integers) for which the description is simple and easy. This is very
disappointing. We cannot see the whole picture of the theory of quadratic forms by
such a restricted treatment and sometimes it causes misunderstanding. We need the
full description of this kind of relation when we explain later the relation between
L-functions of prehomogeneous vector spaces and the Bernoulli numbers. So it
would be a good chance to try to explain the full relation here.

6.1 Quadratic Forms

For integers a, b, c and variables x, y, we call the homogeneous polynomial

Q.x; y/ D ax 2 C bxy C cy 2

of degree two an integral binary quadratic form (or a quadratic form for short, since
we treat only the integral and binary case in this book). In particular, in case a, b, c
are mutually coprime (namely the greatest common divisor, g.c.d., of a, b, c is one),
we say Q.x; y/ is primitive. If we define the matrix S by
 
a b=2
SD ;
b=2 c

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 75

in Mathematics, DOI 10.1007/978-4-431-54919-2__6, © Springer Japan 2014
76 6 Quadratic Forms and Ideal Theory of Quadratic Fields

then we can write

 
x
Q.x; y/ D .x; y/S :
y

Often this expression helps easy understanding. A symmetric matrix like S whose
diagonal components are integers and the other components are half-integers1 is
called a half-integral symmetric matrix. We denote by L the set of all half-integral
symmetric matrices of order two:

  
a b=2
L D I a; b; c 2 Z :
b=2 c

This set corresponds bijectively to the set of all integral

 (binary) quadratic forms.
x
When the corresponding quadratic form .x; y/S is primitive, we say that S is
y
primitive. We consider the following groups of integral matrices:

GL2 .Z/ D fg 2 M2 .Z/I det.g/ D ˙1g;

SL2 .Z/ D fg 2 M2 .Z/I det.g/ D 1g;

where M2 .Z/ is the set of all integral 2  2 matrices. The group SL2 .Z/ is called the
(full) modular group and this is the subgroup of GL2 .Z/ of index two. The group
GL2 .Z/ (and hence also SL2 .Z/) acts on L from the right by

L 3 S 7! det. / t S  2 L . 2 GL2 .Z//:

Two elements of L or the corresponding quadratic forms are said to be SL2 .Z/-
equivalent, or just equivalent, when they belong to the same SL2 .Z/-orbit under this
action. Two elements of L are said to be GL2 .Z/-equivalent when they belong to
the same GL2 .Z/ orbit by the above action. (In some books, SL2 .Z/-equivalence
is called a proper equivalence, but we do not add the word “proper” for this in this
book). Usually we call the number b 2  4ac D  det.2S / the discriminant of the
quadratic form. We denote this number by D.S /. It is obvious that D.S /  0
or 1 mod 4. On the other hand, if D is an integer such that D  0 or 1 mod 4,
there exists S 2 L such that D.S / D D. The discriminant is invariant by the
action of GL2 .Z/. The cases where D is a square (i.e. a square of integers, e.g.
0, 1, 4, 9 . . . ) are exceptional, so in this chapter, we assume that D is not a square
unless otherwise stated. We fix such a non-square integer D  0 or 1 mod 4 and put
L .D/ D fS 2 L I D.S / D D; S primitiveg. Then it is known that both SL2 .Z/-
equivalence classes and GL2 .Z/-equivalence classes in L .D/ are finite. (The proof

1
Here half-integers mean their doubles are integers.
6.2 Orders of Quadratic Fields 77

of the finiteness of the class number and an explicit formula for the class number
will be given later.) When D.S / > 0, we call these numbers of equivalence classes
the class number and the class number in the wide sense of quadratic forms of
discriminant D for SL2 .Z/-equivalence and for GL2 .Z/ equivalence, respectively.
When D.S / < 0, we have ac > 0, and this means that S is positive definite
or negative definite. We denote by LC .D/ the subset of positive definite quadratic
forms of L .D/ and by L .D/ negative definite forms. When D.S / < 0, we say
that the number of SL2 .Z/ equivalence classes in LC .D/ the class number of the
quadratic forms of discriminant D. In this case, the number of GL2 .Z/ equivalence
classes in L .D/ and the number of SL2 .Z/ equivalence classes in LC .D/ are the
same. Indeed this is proved as follows. Any S 2 L .D/ is GL2 .Z/ equivalent to
 
 1 0
an element of LC .D/ since for  D , we have det. /t S  D  t S  2
0 1
LC .D/. If S1 and S2 2 LC .S / are GL2 .Z/ equivalent and S2 D det. /t S1  for
 2 GL2 .Z/, then since S2 and t S1  are positive definite, we have det. / > 0,
hence det. / D 1, and S1 and S2 are SL2 .Z/ equivalent.

6.2 Orders of Quadratic Fields

Now we review the theory of quadratic fields, but there are many books describing
this theory, so here we write only briefly. Also, we describe the theory mostly from
the viewpoint of quadratic forms. In some parts this makes our explanation more
complicated. But this is still useful since this kind of explanation is seldom found in
books nowadays. p p
When m is a non-square integer, we say that K D Q. m/ D Q C Q m is a
quadratic field. When m > 0, we call K a real quadratic field and when m < 0, we
call it an imaginary quadratic field. If m has a square factor, we can bring this factor
outside the square root, so we can assume that m is an integer without square factor.
We assume this from now on. An element ˛ in this field K satisfying ˛ 2 Cb˛Cc D 0
for some rational integers b, c (i.e. integers in the usual sense) is called an integer
of K. The set Omax of all integers in K form a ring. If we put
⎧ p
⎨ 1C m if m  1 mod 4;
2
!D p (6.1)
⎩ m otherwise,

then one can show Omax D ZpC Z!. The ring Omax is called the maximal order (or
the ring of all integers) of Q. m/.
If we define the linear map ˛ ! ˛ of ˛ 2 K to ˛ 2 K as Q-vector space by
p p
a C b m D a  b m (a, b 2 Q), which is called a conjugation, then this is an
automorphism as a field. For any element ˛ of K, we write N.˛/ D ˛˛, T r.˛/ D
˛ C ˛, the former is called the norm and the latter the trace of the element ˛.
78 6 Quadratic Forms and Ideal Theory of Quadratic Fields

We put DK D .!  !/2 D T r.!/2  4N.!/ and call it the discriminant, or the

fundamental discriminant of K. More concretely we have DK D m if m  1 mod 4
and DK D 4m otherwise.
The maximal order Omax is neither a principal ideal domain, nor a unique
factorization domain in general, so it is known that we cannot define a useful prime
factor decomposition of elements of Omax . But if we consider ideals of Omax instead
of elements, any ideal of Omax is factored uniquely into a product of prime ideals
of Omax . This fact is a direct consequence of the fact that Omax is a Dedekind
domain. But sometimes it is useful to treat subrings of Omax consisting only of a
part of integers. For example, it is inevitable to treat such rings when we consider the
correspondence with quadratic forms. But such subrings are not Dedekind domains
in general, and we must keep in mind that the prime ideal decomposition does not
hold for such rings in general. Now any subring O of Omax containing 1 is a free
Z-module since any subgroup of a finitely generated free abelian group is free by
the fundamental theorem of finitely generated abelian groups. When the rank of O
is two, we call it simply an order. Let us describe all such rings.
Lemma 6.1. Let ! be as in (6.1). For an arbitrary natural number f , we put

Of D Z C f !Z:
p p
Then this is an order of Q. m/. Conversely, for any order O of Q. m/, there exists
a natural number f such that O D Of .
Proof. To show that Z C f !Z is an order, we must show that this is a subring. This
is shown by the following calculation.
⎧
⎨ m1 2
f .f !/ C f if m  1 mod 4;
.f !/2 D 4
⎩ f 2m otherwise.

Conversely, if O is an order, then since it is of rank 2, the ring O contains an element

which does not belong to Z. So there is an integer a and non-zero integer b such
that a C b! 2 O. Since we have 1 2 O by definition, we have a 2 O so b! 2 O.
We take the smallest positive integer f among those b such that b! 2 O. All the
other such b are divisible by f . Indeed, for b such that b! 2 O, we divide b by f
and write b D f x C b0 ; x 2 Z; 0  b0 < f . Then we have b0 ! D .b  f x/! D
b!  x.f !/ 2 O. By our choice of f , we must have b0 D 0, and b is divisible
by f . Hence we have O D Z C f !Z D Of . t
u
Orders Of for different f are different rings. If f and f 0 are natural numbers, then
it is easy to see that Of 0 Of if and only if f jf 0 . This ring can be also written as

Of D Z C f Omax :
6.2 Orders of Quadratic Fields 79

We call f the conductor of the order Of . The conductor of an order is 1 if and

only if it is maximal. We define the discriminant of the order Of by DK f 2 . It is
conventional to call non-zero ideal a of Of simply an ideal of Of . Since an ideal a
is a submodule of Of , it is a free Z-module, but since it is also a module over Of ,
the rank is two. So the ring Of =a is a finite ring. The size of this finite ring is called
the norm of a and denoted by N.a/. We describe the free basis over Z of the ideal
a of Of more concretely. We say that an ideal a is primitive if it cannot be written
as a D lb for any integer l > 1 and any ideal b of Of .
Lemma 6.2. For any ideal a of Of , there exist positive integers a, e, d with
a=2  d < a=2 such that

a D Za C Z.d C ef !/:

Here the integers a, b, d are uniquely determined by a. If a is a primitive ideal, then

we have e D 1 and the norm of a is a.
Proof. For a while, we take an arbitrary ideal a of Of and do not assume that it is
primitive. Since the module Of =a is a finite module, there is some natural number
m such that m1 2 a and we have a\Z ¤ 0. So we take the smallest positive rational
integer a in a. Then it is clear that aZ D a \ Z. On the other hand, we denote by
e the smallest positive integer among those y such that x C yf ! 2 a for some
integer x. We take d 2 Z such that d C ef ! 2 a. Then we can conclude that

a D Za C Z.d C ef !/:

This can be seen as follows. First, if x C yf ! 2 a for some x, y 2 Z, then y is

divisible by e. Indeed, if y D eq C r with 0  r < e and 0  q, then we have
x C yf !  q.d C ef !/ D .x  qd / C rf ! 2 a. Here if r ¤ 0, it contradicts
the definition of e, so we have r D 0. Hence there is an integer z such that .x C
yf !/  z.d C ef !/ D x  zd 2 Z \ a, which is a multiple of a. Hence we have
a D Za C Z.d C ef !/. By adding a multiple of a to d C ef ! if necessary, we
may assume that d satisfies a=2  d < a=2. If we take a, e and d as above,
then it is clear by our choice that the integers a, d , e are uniquely determined by
the ideal a. Now we show that if a is primitive in addition, then we have e D 1
in the above. Indeed, since a is an ideal of Of , we have f !a a and hence
af ! 2 a. So we have af ! D xa C y.d C f e!/ for some integers x, y. This
means a D ye and xa C yd D 0. So a=e D y 2 Z and d=e D x 2 Z.
So if we put b D Z.a=e/ C Z.d=e C f !/, then we have a D eb. Since a is an
ideal of Of , we have e.˛b/ D ˛a a D eb, and hence we have ˛b b. So
b is also an ideal of Of . Since we assumed that a is primitive, we have e D 1.
Since Of D Z C Zf ! D Z C Z.d C f !/ and a D Za C Z.d C f !/, we have
Of =a Š Z=aZ and the norm of the primitive ideal a is a. t
u
We call .a; d C ef !/ in the above lemma the standard basis of the ideal a.
Although we described the standard basis of an ideal above, we are not claiming
80 6 Quadratic Forms and Ideal Theory of Quadratic Fields

that the module M D Za C Z.d C f !/ is always an ideal for any random choice
of a, d 2 Z. The module M is an Of -ideal only when a, d , f satisfy some special
conditions. We shall describe these conditions below.
Lemma 6.3. The module M D Za C Z.d C f !/ is an ideal of Of if and only if

ajN.d C f !/:

This is equivalent to saying that there exists an integer c such that

b 2  f 2 DK D 4ac;

where we put b D T r.d C f !/.

Proof. All we need to do is to examine the condition f !M M . We have f ! 
a D .d /a C a.d C f !/ 2 M . As for f !.d C f !/, we have

f !.d C f !/ D .d C T r.f !//.d C f !/  N.d C f !/: (6.2)

So this belongs to M if and only if N.d C f !/ is divisible by a. Since we have

b 2  f 2 DK D .2d C f T r.!//2  f 2 .T r.!/2  4N.!//

D 4.d 2 C df T r.!/ C f 2 N.!//
D 4N.d C f !/;

this is equivalent to say that

b 2  f 2 DK D 4ac

for some c 2 Z. t
u
Remark 6.4. (1) Notation being the same as in the last lemma, we have
p
bC D
d C f! D ;
2

where D D f 2 DK : the discriminant of Of .

(2) The norm a of a primitive ideal p
and the conductor f might not be coprime. For
example, the module Z5pC Z5 2 is a primitive ideal of the order O5 of the
quadraticpfield K D Q. p2/ with conductor
p 5 and its norm is 5. We can write
Z5 C Z5 2 D 5.Z C Z 2/, but Z C Z 2 is not an ideal (nor a subset) of O5 .
For an ideal a of Of , the ring fx 2 KI xa ag is a subring of Omax which
contains Of , so it is Of 0 for some divisor f 0 of f . When this coincides with Of ,
that is, when f D f 0 , we say that a is a proper Of -ideal. (In the usual ring theory,
6.2 Orders of Quadratic Fields 81

the term proper ideal is often used to indicate the ideal which is not the whole
ring itself. Note that our usage here is different from that.) Next, we shall see the
condition that a primitive ideal of Of is a proper Of -ideal.
Lemma 6.5. For a primitive ideal a of Of , we write a D Za C Z.d C f !/ by the
standard basis and define integers b, c as in Lemma 6.3. Then the ideal a is a proper
Of -ideal if and only if gcd of a, b, c is one.
Proof. Assume that a is an Of 0 module for a divisor f 0 of f . If a is an Of 0
module, then af 0 ! 2 a, hence we have af 0 ! D xa C y.d C f !/. Comparing
the coefficients, we see a D y.f =f 0 / and xa C yd D 0. But since a > 0,
we have y ¤ 0 and so d D x.f =f 0 /. Hence we have a; d 2 .f =f 0 /Z. So
bf 0 =f D df 0 =f C f 0 T r.w/ 2 Z and b 2 .f =f 0 /Z. Multiplying the Eq. (6.2) in
the proof of Lemma 6.3 by f 0 =f , we have

f 0 !.d C f !/ D .f 0 d=f C f 0 T r.!//.d C f !/  acf 0 =f:

Since f 0 !.d C f !/ 2 a, we have cf 0 =f 2 Z, i.e. c 2 f =f 0 Z. So if f 0 < f ,

then the g.c.d. of a, b, c is bigger than one. Conversely, assume that n is the greatest
common divisor of a, b, c. We first show that n always divides f . Indeed, write
a D na0 , b D nb0 , c D nc0 . Then b 2  4ac D n2 .b02  4a0 c0 / D f 2 DK . If DK
is odd, then DK does not contain any square factor, so we have njf . If DK is even,
then DK D 4m, where m  2 or 3 mod 4 and m is square-free. If n is odd, then
by the same reason as before, we have njf . Assume n is even and n D 2n0 . Then
we have n0 jf . If f =n0 is odd, then .f =n0 /2 m  2 or 3 mod 4 but this is equal to
b02  4a0 c0 which is 0 or 1 mod 4 and we have a contradiction. So f =n0 is even
and we have njf . So if we assume that n > 1 and define f 0 by nf 0 D f , then
f 0 < f and a, b, c 2 .f =f 0 /Z. Then we also have d 2 .f =f 0 /Z. Indeed we have
.f =f 0 /jb D T r.d C f !/ D 2d C f T r.!/ and .f =f 0 /2 jac D N.d C f !/ D
d 2 C 2df T r.!/ C f 2 N.!/. So we see that .f =f 0 /2 divides

.2d C f T r.!//2  2.d 2 C 2df T r.!/ C f 2 N.!/ D 2d 2 C f 2 .T r.!/2  2N.!//:

Since f =f 0 divides f , we have .f =f 0 /2 j2d 2 . So we have .f =f 0 /jd . So we have

a, b, c, d 2 .f =f 0 /Z. Hence we see that af 0 ! D .af 0 =f /f ! D a.df 0 =f / C
.af 0 =f /.d C f !/ 2 a and f 0 !.d C f !/ D .df 0 =f / C f 0 T r.!//.d C f !/f ! 
ac.f 0 =f / 2 a, so a is an ideal of Of 0 . t
u
Lemma 6.6. If the discriminant of a quadratic form ax 2 C bxy C cy 2 is not a
square, then there exists a quadratic field K and a natural number f such that
b 2  4ac D f 2 DK .
Proof. We have b 2  4ac  0 mod 4 or 1 mod 4. So we can write b 2  4ac D
2e f02 m0 for some odd natural number f0 , some odd number m0 with no square
factor and some non-negative integer e. If e D 0 then m0  1 mod 4. Since
p
we assumed that b 2  4ac is not a square, the field K D Q. m0 / is quadratic
and DK D m0 . Then DK and f D f0 satisfy the demand. If e ¤ 0, then
82 6 Quadratic Forms and Ideal Theory of Quadratic Fields

p
b 2  4ac  0 mod 4, so we have e  2. If e is odd, then put K D Q. 2m0 /.
Then we have DK D 8m0 and f D 2.e3/=2 f0 satisfies the demand. If e is even, put
p
K D Q. m0 /. If m0  1 mod 4 then we have DK D m0 , and then put f D 2e=2 f0 .
If m0  3 mod 4 then we have DK D 4m0 and put f D 2.e2/=2 f0 . These satisfy
the demand of the lemma. t
u
This lemma suggests that there is a deep relation between binary quadratic forms
and quadratic fields. For each proper primitive ideal a of an order of a quadratic
field, using a, b, c of Lemma 6.3, we define a quadratic form Qa by

Qa .x; y/ D ax 2 C bxy C cy 2 :

We can also write this as

p ! p !
1 bC D bC D
Qa .x; y/ D N.a/ N xa C y D aN xCy :
2 2a
p
Here we put D D b 2  4ac, and a and bC2 D D d C f ! form a basis of a.
Now in order to see relations between equivalence classes of quadratic forms
with some equivalent classes of ideals of a quadratic order, we give the following
definitions. We say that two proper Of -ideals a and b are equivalent in the wide
sense if there exists ˛ 2 K  such that a D .˛/b. If N.˛/ > 0 in the above in
addition, we say that a and b are equivalent in the narrow sense. Here the condition
N.˛/ > 0 is satisfied automatically if K is an imaginary quadratic field. If K is a
real quadratic field, then taking ˛ instead of ˛ if necessary, we may replace the
condition N.˛/ > 0 by the condition ˛ > 0 and ˛ > 0. Actually this is the usual
definition which fits into the general theory. The number of equivalence classes in
the wide (narrow) sense of proper Of -ideals is called the class number in the wide
(narrow) sense of Of . It is common to call the number of equivalence classes in the
wide sense just the class number. If K is an imaginary quadratic field, the definition
in the narrow sense and that in the wide sense are the same, as we saw already.
Now let K be a quadratic field and let DK be the fundamental discriminant of K.
Let f be a natural number. We define a mapping from the set of proper primitive
ideals a of Of to the set of quadratic forms by setting Qa .x; y/ D ax 2 Cbxy Ccy 2
by using the standard basis such that a D Za C Z.d C f !/, where we define b, c
by b D 2d C T r.!/, N.d C f !/ D ac. Since the standard basis is unique, this is
a well-defined mapping from ideals.
This mapping does not give a bijective mapping of proper primitive ideals to
primitive quadratic forms. But we can prove the following claim.
Theorem 6.7. When K is an imaginary quadratic field, that is, if DK < 0, then the
above mapping induces a bijective correspondence between the set of equivalence
classes in the narrow sense of proper Of ideals to the set of SL2 .Z/-equivalence
classes of positive definite primitive quadratic forms with discriminant f 2 DK . If
K is a real quadratic field, that is, if DK > 0, then this gives a bijection from
6.2 Orders of Quadratic Fields 83

the set of equivalence classes in the narrow (wide) sense of proper Of ideals to
the set of SL2 .Z/-equivalence (GL2 .Z/-equivalence) classes of indefinite primitive
quadratic forms with discriminant f 2 DK .
Proof. For simplicity, we write L.D/ D L .D/ if D > 0 and L.D/ D LC .D/
if D < 0. We denote by L.D/=
the set of GL2 .Z/ equivalence classes in L.D/
and by L.D/=  the set of SL2 .Z/ equivalence classes. Since any Of ideal is
equivalent (both in the wide sense and narrow sense) to the primitive ideal, it is
sufficient to consider primitive ideals. For a given proper primitive Of ideal, we
define a, b, c as before. Then since .a; b; c/ D 1, the quadratic form Qa .x; y/ D
ax 2 C bxy C cy 2 is primitive and the discriminant of Qa is f 2 DK . By our choice
of the standard basis of a, we have a > 0. So if D.S / < 0, then Qa .x; y/ is
positive definite. So the mapping a to Qa .x; y/ induces mappings to L.D/=

and to L.D/= . We first show that this mapping is surjective. Take a primitive
quadratic form Q.x; y/ D ax 2 Cbxy Ccy 2 . If this is positive definite, then we have
a > 0 automatically. If this is indefinite, then we might have a < 0, but changing the
quadratic form by SL2 .Z/ equivalence, we may assume that a > 0. This is proved
as follows. Assume that a < 0. For a given x, y, we have 4a.ax 2 C bxy C cy 2 / D
.2ax C by/2 C .4ac  b 2 /y 2 . We have 4ac  b 2 < 0 by the assumption
p that the
quadratic form is indefinite. We fix an integer y such that jaj= b 2  4ac < y.
Choosing x suitably, we can assume that 2ax C by is in any set of representatives
modulo 2a. So we can assume that j2ax C byj  jaj. So we have .2ax C by/2 
.b 2  4ac/y 2 < 0 for some x. This means that there exists .x; y/ 2 Z2 such
that ax 2 C bxy C cy 2 > 0. This property still holds if we divide x, y by the
g.c.d. of x and y, so we may assume that x and y are coprime. We can take an
 A of SL2 .Z/ such that .x; y/ is the first row of A. So the (1,1) component
element
a b=2 t
of A b=2 c A is positive. This proves the assertion. We assume a > 0 from
now on. Let D be the discriminant of Q, that is, D D b 2  4ac D f 2 DK . Since we
have .f T r.!//2 D f 2 .DK C 4N.!//, we have .f T r.!//2  f 2 DK  b 2 mod 4
and we have b  f T r.!/ mod 2. So if we define d by b D 2d C f T r.!/, then
p
bC D
a D Za C Z.d C f !/ D Za C Z
2

is a proper primitive ideal of Of since we assumed gcd.a; b; c/ D 1. This d might

not satisfy the condition a=2  d < a=2, but changing b to b0 D b  2ak for
some k, we see that d0 D d  ak satisfies the condition. So we take k 2 Z such that
a=2  d0 D d  ak < a=2. If we define b0 D b  2ak and c0 D c  bk C ak 2 ,
then b02  4ac0 D f 2 Dk and we have ac0 D N.d0 C f !/, b0 D 2d0 C f T r.!/.
So we have a D Za C Z.d0 C f !/. We have

Qa .x; y/ D a0 x 2 C b0 xy C c0 y 2
D a.x  ky/2 C b.x  ky/y C cy 2 ;
84 6 Quadratic Forms and Ideal Theory of Quadratic Fields

so this is SL2 .Z/ equivalent (a fortiori, GL2 .Z/-equivalent) to Q.x; y/. So the
mapping to L.D/=
or L.D/=  is surjective. Next we see that these mappings
to L.D/=
and L.D/=  induce well defined mappings from ideal classes in the
wide sense and in the narrow sense, respectively.pWrite two proper primitive Of
ideals by standard bases as ai D Zai C Z.bi C D/=2,pwhere D D bi2  4ai ci

(i D 1, 2), and assume that a2 D p a1 ˛ for some ˛ 2 Q. D/ p . Then there exists
A 2 GL2 .Z/ such that .a2 ; .b2 C D/=2/ D .a1 ˛; .b1 C D/˛=2/A. Since we
have
p ! p ! 
a2 b2 C2p D a1 b1 C2p D ˛0
D A;
a2 b2  D
2
a1 b1  D 2
0˛

we have a2 D a1 N.˛/ det.A/. Writing .X; Y / D .x; y/ t A, we also have

p !
b2 C D
a2 .a2 x C b2 xy C c2 y / D N
2 2
a2 x C y
2

D a1 N.˛/.a1 X 2 C b1 X Y C c1 Y 2 /
D a2 det.A/.a1 X 2 C b1 X Y C c1 Y 2 /:

Since ai > 0 for i D 1, 2, we have N.˛/ det.A/ > 0. So we are done. Next
we see that the mapping is injective. For two primitive proper Of ideals a1 D
Za1 C Z.d1 C f !/ and a2 D Za2 C Z.d2 C f !/, we define

Qa 1 .x; y/ D a1 x 2 C b1 xy C c1 y 2 D a11 N.a1 x C .d1 C f !/y/ and

Qa 2 .x; y/ D a2 x 2 C b2 xy C c2 y 2 D a21 N.a2 x C .d2 C f !/y/;

p y/ D det.A/Qa 1 ..x; y/A/ for some A 2 GL2 .Z/.

as before. Assume that Qa 2 .x;
We define !1 , !2 2 K D Q. DK / by
p !
b1 C D t
.!1 ; !2 / D .a1 ; d1 C f !/ A D a1 ;
t
A:
2

and put ˛ D a2 =!1 . We will show that a2 D ˛a   > 0. Since

 1 and det.A/N.˛/
a1 a1 !1
a1 x C .d1 C f !/y D .x; y/ , and A D , we have
d1 C f ! d1 C f ! !2
  
a1
Qa 1 ..x; y/A/ D a11 N .x; y/A D a11 N.x!1 C y!2 /
d1 C f !
D a11 .x 2 N.!1 / C xyT r.!1 !2 / C y 2 N.!2 //:
6.2 Orders of Quadratic Fields 85

Since this is equal to det.A/Qa1 .x; y/, we have

N.!1 / D det.A/a1 a2 and T r.!1 !2 / D det.A/a1 b2 :

So we have

N.˛/ det.A/ D det.A/a22 =N.!1 / D a2 =a1 > 0: (6.3)

Now, since A 2 GL2 .Z/, we have a1 D Z!1 C Z!2 . So we have a1 ˛ D

Za2 C Z.a2 !2 =!1 /. So it is sufficient to show that a2 !2 =!1 D d2 C f !. We have
a2 !2 =!1 D a2 !2 !1 =N.!1 / D det.A/!2 !1 =a1 . We must calculate !2 !1 . Taking
the conjugate of the definition of !i , we have
   
!1 ; !2 a1 ; d1 C f ! t
D A:
!1; ! 2 a1 ; d1 C f !

So taking the determinant, we have

!1 ! 2 C !2 ! 1 D det.A/a1 f .!  !/:

So adding !1 ! 2 C!1 !2 D det.A/b2 a1 D det.A/a1 .2d2 CT r.f !// to the both sides
and dividing by 2, we have !2 ! 1 D det.A/a1 .d2 C f !/ D .N.!1 /=a2 /.d2 C f !/.
So we have a2 D a1 ˛. Hence a1 and a2 are equivalent in the wide sense, and if
det.A/ D 1, then equivalent in the narrow sense. Hence the mapping is injective.
t
u
When DK > 0, the equivalences of proper Of -ideals in the wide sense and in
the narrow sense are the same if and only if there exists " 2 Of with N."/ D 1, as
we can see easily. We give the corresponding fact in the case of symmetric matrices
(or quadratic forms) in the following proposition.
Proposition 6.8. Assume that D D f 2 DK is positive and is not a square. Assume
that for S1 , S2 2 L .D/ we have S2 D  t BS1 B for some B 2 GL2 .Z/ with
det.B/ D 1. Then there exists C 2 SL2 .Z/ such that S2 D t CS1 C if and only if
there exists " 2 O
f such that N."/ D 1.

Proof. We may assume that S1 and S2 are primitive. First we assume the existence
of C . Then if we write A D BC 1 2 GL2 .Z/, then we have det.A/ D 1 and S1 D
det.A/ t AS1 A. We have shown in the proof of the last theorem that S1 is SL2 .Z/
equivalent to a matrix S in L.D/ corresponding to a quadratic form Qa .x; y/ for
some primitive Of proper ideal a D Za C Z.d C f !/. Now we put .!1 ; !2 / D
.a; d Cf !/A and ˛ D a=!1 , and apply the argument of the proof of the last theorem
for a1 D a2 D a. Then by (6.3) we have N.˛/ < 0 and we have a˛ D a. Since a is a
proper Of -ideal, we have ˛ 2 Of . We also have a D a˛ 1 , so ˛ 1 2 Of and N.˛/
and N.˛ 1 / are integers. So we have N.˛/ D 1 and ˛ 2 Of . Conversely, if there
exists
2 Of with N.
/ D 1, then we have a
D a. If we define A 2 GL2 .Z/ by

.a; d C f !/
D .a; d C f !/A;
86 6 Quadratic Forms and Ideal Theory of Quadratic Fields

then det.A/ D N.
/ D 1 and Qa ..x; y/A/ D det.A/Qa .x; y/, that is, AS t A D
det.A/S . Assuming that S1 D PS t P for P 2 SL2 .Z/, C D PAP 1 satisfies the
condition in the proposition. t
u
p p
Example 6.9. (1) If we pput K D Q. m/ and a D Zm C Z m, then Qa .x; y/ D
jmj1 N.jmjx C y pm/ D jmjx 2  sgn.m/y 2 . p
(2) We takep K D Q. 5/. For an ideal2 a D 3Z C .12 C 5/Z of Omax D
Z C Z 5, we have Qa .x; y/ D 3x C 2xy C 2y . On the other hand, for
Omax D .1/ itself, we have Q.1/ D x 2 C 5y 2 . Any primitive quadratic form
with discriminant 20 is equivalent to one of these two.
Remark 6.10. In the above, we did not treat negative definite quadratic forms. But
this causes no problem since all these can be obtained by multiplying 1 to the
positive definite ones. On the other hand, we excluded primitive quadratic forms
with square discriminant. These quadratic forms are (as far as discriminant is not
zero) very special quadratic forms in various senses. (For example, the volume of
the fundamental domain with respect to the unit group (the automorphism group)
is not finite etc.) We would like to omit such details, but at least we try to give a
classification of equivalence classes below.
Let Q.x; y/ D ax 2 C bxy C cy 2 be a primitive quadratic form and assume that
b 2  4ac D f 2 for a positive integer f . First we see that we may assume a D 0 by
replacing Q by an equivalent quadratic form. Indeed, if a ¤ 0, then noting that the
equation ax 2 C bx C c D 0 has a solution x D .b ˙ f /=2a, we denote by e the
g.c.d. of b C f and 2a and define x0 , y0 by b C f D x0 e and 2a D y0 e. Then
x0 and y0 are coprime and there exist w0 ; z0 2 Z such that x0 w0  y0 z0 D 1. Since
 2  
we have a xy00 C b xy00 C c D 0, we get ax02 C bx0 y0 C cy02 D 0 and

     2   
x0 y0 a b=2 x0 z0 ax0 C bx0 y0 C cy02  0
D D :
z0 w0 b=2 c y0 w0   

So we can assume that Q.x; y/ D bxy C cy 2 up to SL2 .Z/-equivalence. If the

discriminant of Q is 0, then b is also 0, and by the assumption that it is primitive,
we have c D ˙1. It is easy to see that the quadratic forms y 2 and y 2 are not
SL2 .Z/-equivalent but and GL2 .Z/-equivalent. So we classified the case where the
discriminant is zero. From  now on,
 we assume that the discriminant is not zero. We
˛ˇ
consider an element A D 2 GL2 .Z/ which maps bxy C cy 2 to quadratic
 ı
forms whose coefficient of x 2 is 0. Since
   
0 b=2 b˛ C c 2 
t
A AD ;
b=2 c  

we should have  D 0 or b˛ C c D 0. Since b and c are coprime, we have z,

w 2 Z such that bz C cw D ˙1. So such an A is given either by
6.3 Class Number Formula of Quadratic Forms 87

   
c z 1 z
AD˙ or ˙ :
b w 0 ˙1

with det.A/ D ˙1. For the former one, if we take bz C cw D 1, then we have
det.A/ D 1 and
! ! !
t 0 b=2 0 b.bz C cw/=2 0 b=2
A AD D det.A/ ;
b=2 c b.bz C cw/=2 bzw C cw2 b=2 w

so we can choose a representative of SL2 .Z/ equivalence classes such that b > 0.
On the other hand, we have
     
10 0 b=2 1 z 0 b=2
D
z1 b=2 c 01 b=2 c C bz

so we can take c among representatives modulo b. We have

     
1 0 0 b=2 1 0 0 b=2
 D :
0 1 b=2 c 0 1 b=2 c

So representatives of SL2 .Z/-equivalence classes and GL2 .Z/-equivalence classes

with discriminant f 2 ¤ 0 .f > 0/ can be taken in the set of the following matrices:
 
0 f =2
f =2 c

with 0  c  f  1 and 0  c  f =2, respectively. The fact that these are not
equivalent with each other can be seen easily by the above calculation. In particular,
the number of SL2 .Z/- or GL2 .Z/- equivalence classes of quadratic forms with a
fixed square discriminant is finite. The number of SL2 .Z/ equivalence classes is f ,
and if we restrict to the primitive classes, the number is '.f /, where '.f / is the
Euler function. The case where the discriminant is not a square will be explained in
the next section.

6.3 Class Number Formula of Quadratic Forms

In this section, first we prove the finiteness of the class number (the number of
equivalence classes) of primitive quadratic forms. As we mentioned in the last
section, this is equivalent to the finiteness of the class number (in the narrow sense or
in the wide sense) of quadratic orders Of , which are not necessarily maximal. Next
we state a formula of the class numbers (in the wide sense) of quadratic orders. (The
proof is postponed until Chap. 10.) As we saw in the last section, the class number
88 6 Quadratic Forms and Ideal Theory of Quadratic Fields

in the narrow sense is easily obtained from the one in the wide sense. Actually, once
we know the class number formula for the maximal order, the class number of an
arbitrary order is determined. The reason for this can be explained if we use group-
theoretic language, but in Chap. 10, we will give a different direct proof using zeta
functions.
Proposition 6.11. The number of SL2 .Z/ equivalence classes of binary primitive
quadratic forms of a given discriminant is finite. In particular, the class number of
binary primitive quadratic forms with discriminant f 2 DK is finite.
Proof. We already explained the case where the discriminant is square. So, here
we treat only those with discriminant f 2 DK , where K is a quadratic field and DK
is the fundamental discriminant of K. We fix f 2 DK now and we consider only
those quadratic forms which have this number as discriminant. We say first that any
primitive quadratic form is equivalent to ax 2 C bxy C cy 2 with jbj  jaj  jcj.
Indeed, we take a quadratic form ax 2 C bxy C cy 2 so that jaj, the absolute value
of the coefficient of x 2 , is the smallest among those equivalent to the original
one. Then by the transformation .x; y/ ! .y; x/, the coefficient x 2 becomes c,
so by our choice of a, we have jaj  jcj. For some integer m 2 Z, we have
jaj  b  2ma  jaj. So, changing the quadratic form to the equivalent one by the
transformation .x; y/ ! .x  my; y/, we have a.x  my/2 C b.x  my/y C cy 2 D
ax 2 C .b  2am/xy C .c  bm C am2 /y 2 , we may assume that jaj  b  jaj
from the first. So we can assume that jbj  jaj  jcj. (In particular, if DK < 0, we
may assume that the quadratic form is positive definite, so we have jbj  a  c.)
On the other hand, we have b 2  4ac D f 2 DK by definition, so we have

4jacj  b 2  j4ac  b 2 j D f 2 jDK j:

p since b  a , a  jacj, we have 3a  4jacj  b  f jDK j. So jaj 

2 2 2 2 2 2
But
f jDK j=3. It is obvious that there is only a finite number of such integers a.
2

Hence the number of b is also finite by jbj  jaj and c D .b 2  f 2 DK /=4a is also.
Hence the assertion is proved. t
u
Let DK be the fundamental discriminant of a quadratic field and put D D f 2 DK .
We denote by h.D/ the class number in the wide sense of the order Of of the
quadratic field K D Q C Q!. To describe the class number formula, we introduce
  symbol. For an odd prime p and an integer a, the quadratic
the quadratic residue
residue symbol pa with respect to p is defined as follows.
⎧
  ⎪
⎪
⎨1 if a 62 pZ and there exists x 2 Z such that x 2  a mod p,
a
D 1 if a 62 pZ and there is no x 2 Z such that x 2  a mod p,
p ⎪
⎪
⎩0 if a 2 pZ:

The following three relations are known as the reciprocity law of the quadratic
residue symbols.
6.3 Class Number Formula of Quadratic Forms 89

(1) For odd primes p, q, we have

  
q p
D .1/.p1/.q1/=4 :
p q

(2) For an odd prime p, we have

 
1
D .1/.p1/=2 :
p

(3) For an odd prime p, we have

 
2
D .1/.p 1/=8 :
2

The proof will be omitted here. (See e.g. [83].)

For the fundamental discriminant DK of a quadratic field and a prime p, we
define the notation K .p/ by
⎧ D 
⎪
⎪
K
if p is an odd prime,
⎪
⎪
p
⎨
1 if p D 2 and DK  1 mod 8,
K .p/ D
⎪
⎪ 1 if p D 2 and DK  5 mod 8,
⎪
⎪
⎩
0 if p D 2 and DK  0 mod 4:

We also define K .1/ D 1 and


1 if DK > 0,
K .1/ D
1 if DK < 0:

Moreover, if u is an integer and if u D ˙p1e1    prer is a prime factor decomposition

of u, we put K .u/ D K .˙1/K .p1 /e1    K .pr /er . Actually, by using the
reciprocity law and other things, one can show that K .u/ depends only on
u mod DK . We omit the details of the proofs of this fact (cf. [107, §5]). The function
  Dirichlet character modulo jDK j. It is
K prolonged in this way is a primitive
common to write this character as DuK . So in this book, we sometimes use this
notation instead of K .
Now we  can
 write down the class number formula by using the notation

K .u/ D DuK . When DK > 0, by Dirichlet, it is known that Omax D f˙1g 

f"n I n 2 Zg for some " 2 Omax with " ¤ ˙1. This " is called the fundamental unit
of K.
90 6 Quadratic Forms and Ideal Theory of Quadratic Fields

Theorem 6.12. (1) The class number of the quadratic field K with discriminant
DK is given by
⎧ jDK j1  
⎪
⎪
⎪ w X DK
⎪ 
⎨ DK u if DK < 0,
u
h.DK / D uD1
K 1  
⎪
⎪
DX
DK
⎪
⎪
⎩ log."/ log.sin.u=DK // if DK > 0.
uD1
u

Here w is the half of the number of units of Omax and when DK > 0, we denote
by " the fundamental unit of K.
(2) The class number of the order Of of K with discriminant D D f 2 DK is
given by
  
h.DK /f Y 1 DK
h.D/ D  1 :
ŒOmax W Of  p p
pjf

Here for any ring R, the notation R means the group of all units of R.
As we mentioned before, the class number in the narrow sense can be easily
obtained by this formula, and the class number of primitive quadratic forms also.
The proof of the class number formula will be given in Chap. 10 when K is an
imaginary quadratic field. When K is a real quadratic field, the proof will be omitted
from this book.
The class numbers of imaginary quadratic
 fields have a relation to the Bernoulli
numbers. Indeed, if we put .u/ D DuK , then by Sect. 4.2, we have

jDK j1
1 X
B1; D .u/u; (6.4)
DK uD1

so if DK < 0, we see

h.DK / D wB1; : (6.5)

In particular, if p is a prime such that p  3 mod 4 and p > 3, then it is shown

by using (6.4) and the reciprocity law of quadratic residues that the class number
p
h.p/ of Q. p/ is given by

p1  
1X u
h.p/ D  u: (6.6)
p uD1 p

By the above formula, we see that h.D/ D 1 for D D 3, 4, 7, 8,
11, 19, 43, 67 and 163. It is known that there are no other (fundamental
6.3 Class Number Formula of Quadratic Forms 91

discriminants of) imaginary quadratic fields such that h.DK / D 1. Even if we take
the non-maximal orders of imaginary quadratic fields into account, there are only
four more orders with class number one, namely D D 12, 16, 27 and 28. In
contrast to this, it is conjectured that there are infinitely many real quadratic fields
with class number one,2 but no proof is known.
Remark 6.13. For S1 and S2 2 L .D/, we can also define an equivalence class by
S2 D t AS1 A for some A 2 GL2 .Z/. If S1 and S2 are SL2 .Z/ equivalent in the sense
we defined before, then of course these are equivalent in the above sense. But the
equivalences S2 D t A1 S1 A1 and S2 D  t A2 S1 A2 for some A1 and A2 2 GL2 .Z/
with det.A1 / D det.A2 / D 1 are different conditions in general. For example, for
two symmetric matrices S1 , S2 2 L .12/ given by
   
3 0 1 0
S1 D and S2 D ;
0 1 0 3

there do not exist A 2 GL2 .Z/ such that S2 D t AS1 A. This can be seen from the
fact that the diophantine equation a2  3b 2 D 1 does not have integer solutions.
On the other hand, each p correspondspto the symmetric
p matrix
p obtained
p from the
standard basis of 3Z C 3Z or Z C Z 3. Since 3.Z C Z 3/ D 3Z C 3Z, these
ideals are equivalent in the wide sense. In fact, we have
     
01 3 0 01 1 0
 D ;
10 0 1 10 0 3
 
01
so we have S2 D det.A/t ASA for A D 2 GL2 .Z/. We will give one more
10
example. Take S3 , S4 2 L .23/ given by
   
3 1=2 3 1=2
S3 D and S4 D :
1=2 2 1=2 2

Then we have
   
1 0 1 0
S4 D S3
0 1 0 1

but there exists no A 2 GL2 .Z/ such that S4 D det.A/ t AS3 A. Indeed if there exists
such A, then since S3 and S4 are positive definite, we should have det.A/ > 0.
Then comparing diagonal components of both sides, we see that we should have
A D ˙12 . This is a contradiction since .1; 2/ components do not coincide for
A D ˙12 . We explain shortly the difference between SL2 .Z/ equivalence and the
equivalence in the above sense. For any S1 2 L .D/, define

2
In Section 304 of his book on number theory, Disquisitiones Arithmeticae [35], Gauss stated his
speculation on this and also his reasonable insight on the relation to the size of fundamental units.
92 6 Quadratic Forms and Ideal Theory of Quadratic Fields

   
1 0 1 0
S2 D S1 :
0 1 0 1
If S2 is SL2 .Z/ equivalent to S1 , then S1 (or the corresponding SL2 .Z/ equivalence
class, or ideal or class in the narrow sense) is called ambig. In this case, for any
S 2 L .D/ such that S D t AS1 A for some A 2 GL2 .Z/, we see that S
is SL2 .Z/ equivalent to S1 . So there is no difference between two equivalences.
In the case when S2 is not SL2 .Z/ equivalent to S1 , then the GL2 .Z/ equivalence
class which contains S1 contains two different SL2 .Z/ equivalent classes. To count
the equivalence classes of L .D/ up to the equivalence that S2 D t AS1 A for
A 2 GL2 .Z/, we must count ambig classes, which is a part of the genus theory
of Gauss. Since this is complicated in general, we do not explain this theory in this
book. (See [29]. See also Exercises 6.21 and 10.25.)
integer m, show that 1 and ! in (6.1) gives a basis
Exercise 6.14. For a square-free p
of the maximal order Omax of Q. m/.
Exercise 6.15. Show that the quadratic form Q.x; y/ D x 2 C bxy C cy 2 with
discriminant D is equivalent to the following quadratic form.

x 2  D4 y 2 if b is even,
x 2 C xy C 1D 2
4 y if b is odd.
p
Exercise 6.16. (1) Let a D Za C Z bC2 D be a proper primitive ideal of order Of
of discriminant D D f 2 DK , where DK is a discriminant of a quadratic field
K. We assume a > 0. Show that a is a principal ideal of Of (i.e. a D Of ˛
for some element ˛ 2 Of ) if and only if there exists integers x and y such that
ax 2 C bxy C cy 2 D ˙1.
(2) Show that the class of quadratic forms in the wide sense corresponding to a
principal ideal a contains the one given in Exercise 6.15.
(3) Notation being as above, give an example of a principal ideal a such that there
exists no x, y 2 Z with ax 2 C bxy C cy 2 D 1.
(4) Give an example of an equivalence class of quadratic forms Q.x; y/ in the
narrow sense corresponding to a principal ideal (in the wide sense) such that
Q.x; y/ is not SL2 .Z/ equivalent to any form given in Exercise 6.15.
Exercise 6.17. Fix a positive discriminant D D f 2 DK , where K is a real quadratic
field, and consider an integral quadratic form Q.x; y/ D Ax 2 C Bxy C Cy 2 whose
discriminant is D.
(1) Let jaj be minimum among coefficients of x 2 of quadratic forms which are
p Q2 .x; y/ D ax C
2
equivalent to Q. Show that there exists
p a quadratic form
bxy C y equivalent to Q such that D  2jaj < b < pD and jaj
2
p jcj.
(2) Taking Q2 as in (1), show that ac < 0 and 0
p p < b. (Use . D C b/.p D  b/ D
4ac.) In particular, show that 0 < b < D, D  b < 2jaj < D C b.
6.3 Class Number Formula of Quadratic Forms 93

Exercise 6.18. Show that for any integral quadratic form Q, there exists a quadratic
form SL2 .Z/ equivalent to Q such that the coefficient of x 2 is positive. (See the
proof of Theorem 6.7, for example.)
Exercise 6.19. (1) Show that the following pairs of quadratic forms in (i) and
(ii) are SL2 .Z/ equivalent respectively.
(i) ax 2 C bxy C ay 2 and ax 2  bxy C ay 2 :
(ii) ax 2 C axy C cy 2 and ax 2  axy C cy 2 :
(2) Assume that D D f 2 DK < 0 and a positive definite quadratic form
Q.x; y/ D ax 2 C bxy C cy 2 has discriminant D. As shown in the proof
of Proposition 6.11, by replacing SL2 .Z/ equivalence, we may assume that
jbj  a  c. Such quadratic forms are called reduced. Conversely, assume that
there are two SL2 .Z/ equivalent positive definite quadratic forms Qi .x; y/ D
ai x 2 C bi xy C ci y 2 with discriminant D with jbi j  ai  ci (1  i  2).
Show that .a1 ; b1 ; c1 / D .a2 ; b2 ; c2 / except for the following pairs.
(i) a1 D c1 and .a2 ; b2 ; c2 / D .a1 ; b1 ; a1 /.
(ii) a1 D b1 and .a2 ; b2 ; c2 / D .a1 ; a1 ; c1 /.
Hint: Write down the transformation explicitly as a2 D a1 ˛ 2 Cb1 ˛ Cc1  2 and
assuming that a2  a1 and using the condition of a1 , b1 , c1 , show that j˛ j  1.
Exercise 6.20. (1) Show that h.D/ D 1 for D D 3, 4, 7, 8, 11, 12,
16, 19, 27, 28, 43, 67 and 163.
(2) Show that h.20/ D 2 and give a complete set of representatives of the ideal
classes.
(3) Show that h.40/ D 2 and give a complete set of representatives of the ideal
classes.
 
a b=2
Exercise 6.21. Fix a discriminant D D f 2 DK and consider S D 2
b=2 c
L .D/. Show that the following conditions on S are equivalent (the class to which
such an S belongs is called an ambig class).
(1) There exists A 2 GL2 .Z/ with det.A/ D 1 such that t ASA D S .
(2) The matrix S is SL2 .Z/ equivalent to
 
a b=2
:
b=2 c
 
a0 b 0 =2
(3) There exists S 0 D 2 L .D/ which is SL2 .Z/ equivalent to S
b 0 =2 c 0
such that b 0 is divisible by a0 . (Here we may take b 0 D 0 or b 0 D a0 .)
 
10
Hint: If t ASA D S with det.A/ D 1, then show that T r.A/ D 0 and A2 D .
01
Then reduce the case when A is upper triangular.
Chapter 7
Congruence Between Bernoulli Numbers
and Class Numbers of Imaginary
Quadratic Fields

In this chapter, we prove a congruence relation between Bernoulli numbers and

class numbers of imaginary quadratic fields (Theorem 7.1). For that purpose, we
study in Sect. 7.2 a certain type of power series which Hurwitz1 introduced (we call
this “Hurwitz-integral” series). Although one may prove Theorem 7.2 without using
this, we nevertheless introduce this notion because of its elegance and usefulness.

7.1 Congruence Between Bernoulli Numbers and Class

Numbers

For a prime p satisfying p  3 mod 4, recall that h.p/ denotes the class number
of the imaginary quadratic field of discriminant p.
Theorem 7.1. Let p > 3 be a prime, p  3 mod 4. We have the congruence2

h.p/  2B pC1 mod p:

2

Remark 7.2. (1) By the theorem of Clausen and von Staudt (Theorem 3.1), the
denominator of B pC1 is prime to p.
2
(2) Using Euler’s formula .1  k/ D Bk =k (Theorem 5.4 (2), p. 72), we have

1
Adolf Hurwitz (born on March 26, 1859 in Hildesheim, Germany—died on November 18, 1919
in Zürich, Switzerland).
2
This is due to Augustin Louis Cauchy (born on August 21, 1789 in Paris, France—died on May 23,
1857 in Sceaux, France). He expressed the left-hand side by a difference of numbers of quadratic
residues and non-residues in the interval .0; p=2/. That this is equal to the class number is nothing
but the class number formula of Dirichlet.

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 95

in Mathematics, DOI 10.1007/978-4-431-54919-2__7, © Springer Japan 2014
96 7 Congruence Between Bernoulli Numbers and Class Numbers of Quadratic Fields

  B pC1
1p
 D pC1
2
 2B pC1 mod p
2 2
2

and hence the theorem can be written as

 
1p
h.p/   mod p:
2

The class number formula (6.6) at the end of the last chapter gives
ˇp1   ˇ
ˇX u ˇ 1X
p1
p1
1 ˇ ˇ
1  h.p/ D ˇ uˇ  uD :
p ˇ p ˇ p 2
uD1 uD1

This shows that the class number h.p/ is uniquely determined by its value
mod p. Therefore, the above congruence asserts that the value ..1  p/=2/ of
the Riemann zeta function “knows completely” the class number h.p/.
(3) For a more advanced interpretation of this theorem using p-adic modular forms,
see [82].
(4) An analogous result in the real quadratic case is known as the following
Ankeny3–Artin4 –Chowla5 congruence [4]: Let p be a prime number congruent
p
to 1 modulo 4, and " D .t C u p/=2 be the fundamental unit. Then, we have

u
h.p/  B p1 mod p:
t 2

Example 7.3. Let us compute several class numbers by using the theorem.
(1) For p D 7,
 
1 1
2B 7C1 D 2B4 D 2  D  1 mod 7;
2 30 15
so h.7/ D 1:
(2) For p D 11,
 
1 1 1
2B 11C1 D 2B6 D 2 D D  1 mod 11;
2 42 21 1  22

so h.11/ D 1:

3
Nesmith Cornett Ankeny (born in 1927 in Walla Walla, USA—died on August 4, 1993 in Seattle,
USA).
4
Emil Artin (born on March 3, 1898 in Vienna, Austria—died on December 20, 1962 in Hamburg,
Germany).
5
Sarvadaman D. S. Chowla (born on October 22, 1907 in London, England—died on December
10, 1995 in Laramie, USA).
7.2 “Hurwitz-integral” Series 97

(3) For p D 19,

 
5 5 5
2B 19C1 D 2B10 D 2 D D  1 mod 19;
2 66 33 5  38
so h.19/ D 1:
(4) For p D 23,
 
691 691 1
2B 23C1 D 2B12 D 2  D   3 mod 23;
2 2730 1365 8
so h.23/ D 3:

7.2 “Hurwitz-integral” Series

Definition
P1 7.4. A formal power series with rational number coefficients of the form
n
c
nD0 n t =nŠ such that all cn are integers is said to be a Hurwitz-integral series.
The set of all Hurwitz-integral series is denoted by H.
Note that P .t/ 2 H if and only if P .0/; P 0 .0/; P 00 .0/; P .3/ .0/; : : : are all in Z.
Typical examples of elements of H are e t and log.1 C t/.
Proposition 7.5. (1) H is an integral domain. Z t
d
(2) H is closed under term-by-term differentiation and integration .
dt 0

(3) The set of invertible elements of H is H D fP .t/ 2 HjP .0/ D ˙1g:
Proof. (1) We only need to show that the set H is closed under P multiplication,
1
other properties
P1 being trivial to verify. For two series P .t/
P D n
nD0 cn t =nŠ,
1
Q.t/ D n
nD0 dn t =nŠ, write the product P .t/Q.t/ as
n
nD0 en t =nŠ. The
numbers en are given by
!
X n
n
en D ci dni
i D0
i

Pintegers nif ci ; dni are all integers.

and hence are
(2) For P .t/ D 1 nD0 cn t =nŠ 2 H, we have

1
X X 1
0 t n1 tn
P .t/ D cn D cnC1 ;
nD1
.n  1/Š nD0
nŠ
Z t 1
X X1
t nC1 tn
P .t/dt D cn D cn1 :
0 nD0
.n C 1/Š nD1
nŠ

Both of these are clearly in H.

98 7 Congruence Between Bernoulli Numbers and Class Numbers of Quadratic Fields

P
(3) In order that P .t/ D 1 nD0 cn t =nŠ is invertible in QŒŒt, the condition c0 ¤ 0
n

is necessary (Proposition 1.9 on P p. 15).nThis being so, write the reciprocal

P .t/1 in QŒŒt as P .t/1 D 1 nD0 b n t =nŠ. To have P .t/ 1
2 H, we need
b0 D c01 2 Z and thus c0 D ˙1. If this is satisfied, other coefficients bn
automatically belong to Z, because bn is the constant term of the nth derivative
 .n/
P .t/1 of P .t/1 and the derivative is of the form

 .n/ polynomial in P .t/; P 0 .t/; P 00 .t/; : : : with integer coefficients

P .t/1 D ;
power of P .t/

and so, if c0 D P .0/ D ˙1, then the value of this at t D 0 is an integer since
P .0/; P 0 .0/; P 00 .0/;    2 Z. We therefore have P .t/ 2 H . Conversely, if
P .t/ 2 H , then from P .t/1 jt D0 D c01 2 Z we have c0 D ˙1.
t
u
Remark 7.6. The series sin.t/; cos.t/ (Taylor expansions of sin.x/; cos.x/ at the
origin, viewed as formal power series in x D t) are both in H, and cos.t/ D
1  t 2 =2Š C t 4 =4Š C    is invertible by proposition (3) above. Hence the series
tan.t/ D sin.t/= cos.t/ is Hurwitz-integral. This shows that the tangent numbers Tn
in Remark 1.18 (p. 24) are integers.
Proposition 7.7. Suppose P .t/ 2 H; P .0/ D 0. Then for any N 2 N we have

P .t/N
2 H:
NŠ
Proof. We proceed by induction on N . The case N D 1 is trivial. Assume the
0
proposition is valid up to N  1. Using P .t/N =N Š D P .t/N 1 P 0 .t/=.N  1/Š
and P .0/ D 0, we have
Z
P .t/N t
P .t/N 1 0
D P .t/dt:
NŠ 0 .N  1/Š

The induction hypothesis and Proposition 7.5 (1), (2) show the right-hand side
belongs to H. t
u
Definition 7.8. For series P .t/.¤ 0/; Q.t/ 2 H, we write P .t/jH Q.t/ or
Q.t/ H 0 mod P .t/ if the condition Q.t/=P .t/ 2 H holds. Also we write
Q1 .t/ H Q2 .t/ mod P .t/ if P .t/jH .Q1 .t/  Q2 .t//.
In particular, when m is an integer (this is also an element in H), the congruence
P
Q1 .t/ H Q2 .t/ mod m means, if we write Q1 .t/ D 1
.1/ n
nD0 cn t =nŠ, Q2 .t/ D
P1 .2/ n .1/ .2/
nD0 cn t =nŠ, that the congruence cn  cn mod m holds for every n.
7.3 Proof of Theorem 7.1 99

7.3 Proof of Theorem 7.1

Put .p  1/=2 D m. We have m > 1 because of our assumption p > 3. In the class
number formula ((6.6) on p. 90)
p1  
1X u
h.p/ D  u;
p uD1 p

  the sum into two parts according to u < p=2 and u > p=2, and note
divide
1
p D 1 because p  3 mod 4. Then we obtain

X u X p  u
ph.p/ D  u .p  u/
p p
0<u<p=2 0<u<p=2

X u X u
D 2 uCp : (7.1)
p p
0<u<p=2 0<u<p=2

On the other hand, if we divide the sum according to the parity of u, then we have
X  2u  X  p  2u 
ph.p/ D  .2u/  .p  2u/
p p
0<u<p=2 0<u<p=2

X  2u  X  2u 
D 4 uCp :
p p
0<u<p=2 0<u<p=2

From this we have

  X u X u
2
ph.p/ D 4 uCp : (7.2)
p p p
0<u<p=2 0<u<p=2

Subtract (7.2) from two times (7.1) to obtain

   X u
2
2 ph.p/ D p
p p
0<u<p=2

and hence
X m  
1 a
h.p/ D   :
2 2
aD1
p
p
 
Applying Euler’s criterion a
p  am mod p (cf. [50, Proposition 5.1.2]) here, we
have
1 X m
m
h.p/  a mod p:
2  2m aD1
100 7 Congruence Between Bernoulli Numbers and Class Numbers of Quadratic Fields

Hence, the proof of the theorem h.p/  2BmC1 mod p boils down to showing

X
m
am  2.2  2m /BmC1 mod p:
aD1

Further, by 2  mC1
1
mod p and .2  2m /  .22mC1  2m /  2m .1  2mC1 /
mod p, this reduces to showing
X
m
BmC1
am  2m .1  2mC1 / mod p:
aD1
mC1

P1 n
First, by t
et 1 D1 t
2 C nD2 Bn tnŠ , we have

1 1 2
D t  2t
et C1 e 1 e 1
1
! 1
!!
1 t X tn 2t X .2t/n
D 1 C Bn  1 C Bn
t 2 nD2 nŠ 2 nD2
nŠ
1
1 X t n1
D C .1  2n /Bn
2 nD2 nŠ
1
1 X BnC1 t n
D C .1  2nC1 /  :
2 nD1 n C 1 nŠ

Replacing t 7! 2t and multiplying both sides by 2, we obtain

X1 n
2 nC1 BnC1 t
D 1 C 2 nC1
.1  2 / : (7.3)
e 2t C 1 nD1
n C 1 nŠ

Now since
   2  3
2 1 1  e 2t 1  e 2t 1  e 2t
D D1C C C C
e C1
2t
1  1e2
2t
2 2 2
 P 
and D 1
1e2t
2 nD1 2
n1 t n
nŠ
2 H, we have e2t2C1 2 H. Also, by Proposition 7.7,
 2t N  2t N
we have 1e2 H 0 mod N Š. In particular, if N  2, we have 1e2 H
tn 1e2t
0 mod 2. By this, combined with the fact that the coefficient of nŠ in 2 is
divisible by 2 if n  2, we have

2
H 1 C t mod 2: (7.4)
e 2t C1
7.3 Proof of Theorem 7.1 101

Therefore, by (7.3), we have

BnC1
2n .1  2nC1 / 2Z
nC1

if n  2. Next observe

X
m X
m
 4at X
 2mC1
.1 C e 2t / e 4at D e C e .4aC2/t D e 2jt
aD0 aD0 j D0
p 1
X X .2jt/n 1 X
X p 
tn
D D .2j / n

j D0 nD0
nŠ nD0 j D0
nŠ
1
X X
p 
tn
DpC1C 2 n
j n

nD1 j D1
nŠ
X t n
H 1  2n mod p:
n1
nŠ
p1jn

Pp
(Here we have used (3.1) on p. 42, i.e., j D1 j n  1 mod p if p  1jn and
Pp
j D1 j  0 mod p otherwise.) Multiplying both sides by e2t C1 .2 H/, we have
n 2

X
m  X tn 
2
2 e 4at
H 1 2n mod p:
aD0
e 2t C 1 n1
nŠ
p1jn

tm p1
Comparing the coefficients of mŠ on both sides, we obtain by (7.3) (note m D 2 <
p  1)

X
m
BmC1
2 .4a/m  2mC1 .1  2mC1 / mod p:
aD0
mC1

Dividing both sides by 2 and using 4m D 2p1  1 mod p, we have

X
m
BmC1
am  2m .1  2mC1 / mod p:
aD0
mC1

This is what we wanted to show. t

u
Exercise 7.9. Compute the class number h.31/ by using Theorem 7.1 and
Table 1.1 in Chap. 1.
102 7 Congruence Between Bernoulli Numbers and Class Numbers of Quadratic Fields

Exercise 7.10. Let p be a prime congruent to 3 modulo 4. Prove that the numerator
of B pC1 can never be divisible by p.
2

Exercise 7.11. The Euler number En is defined by the generating series

1
X tn
2
D En :
e t C e t nD0
nŠ

Prove that all En are integers. Moreover, prove that E2n is odd for all n  1. (E2nC1
is easily seen to be 0.) Hint: Use (7.4).
Chapter 8
Character Sums and Bernoulli Numbers

We would like to explain arithmetic identities between Bernoulli numbers and the
root of unity. Namely we shall explain certain small collections of formulas between
exponential sums or character sums and Bernoulli numbers. We often encounter
such formulas when we compare the dimension formulas of modular forms obtained
by the Riemann–Roch theorem and by the trace formula. Often, the exponential
sums appear in the first method and the Bernoulli numbers appear in the second
method. But sometimes it is not easy to prove these relations in an elementary way,
and there are several cases that no elementary proofs are known. It seems that there
is no general elementary way to study these relations. Our elementary method in
this section is very restrictive and clearly not enough to attack difficult problems,
but still, it should give us some general feeling of what are easy exponential sums.
Before going to easy things treated in this section, we give here one example for
which no elementary proof is known.
Let p be a prime such that p  3 mod 4. We put  D e 2 i=p . Then the following
formula holds.
X  
.abc/ p pC1 1
D  p B1; C B3; :
.1   a /.1   b /.1   c / 4 6
.a;b;c/2S

 
Here we put S D .a; b; c/ 2 .F p / I ab C bc C ca D 0 ; Fp is the prime field
3

  with characteristic p, is the quadratic residue character defined by .x/ D

Z=pZ
x
p
, and Bn; is the generalized Bernoulli number.
Motivated by dimension formulas, this relation is conjectured in [69], and later
proved in [47]. The proof there is a remote indirect proof. It is natural to ask if there
exists a more elementary and direct proof. But no elementary alternative proof is
currently known. Assuming that such a direct proof exists, we have no idea if such
a proof is simple or complicated before we really prove it.

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 103
in Mathematics, DOI 10.1007/978-4-431-54919-2__8, © Springer Japan 2014
104 8 Character Sums and Bernoulli Numbers

The results we shall give below are much easier compared with this example.
But we see that even researchers sometimes fail to recognize that these are easy,
so we think it is of some use to explain such elementary formulas here. To fix our
standpoint, we assume here that we are satisfied if an exponential sum or a character
sum can be written by the generalized Bernoulli numbers.

8.1 Simplest Examples

We start from simple examples which can be treated easily.PFor a prime p, we put
p D e 2 i=p . As one of our examples, we consider the sum aD1 .1  pa /1 . In fact
p1

this sum is equal to

X
p1
1 p1
D : (8.1)
aD1
1  pa 2

Similar sums obtained by attaching the quadratic residue symbols or character

values of a to the numerator of 1=.1  pa / can be expressed by using Bernoulli
numbers and Gaussian sums. We call loosely this kind of sum obtained by
combination of roots of unity and characters an exponential sum or a character
sum. As for the first example (8.1) we gave above, the argument is very simple.
For similar sums with some characters too, this kind of formula can be obtained by
fairly easy calculation.
Proof 1. We give the first proof. Since pa (1  a  p  1) are mutually different
p  1 roots of the equation X p  1 D 0 which is not equal to 1, we have

Xp  1 Y
p1
D X p1 C X p2 C    C 1 D .X  pa /:
X 1 aD1

Take the logarithm of both sides and compare their derivatives with respect to X .
Then the left-hand side is
Pp1
d aX a
log.X p1 C X p2 C    C 1/ D p1aD1 ;
dX X CC1
and the right-hand side is

d X X
p1 p1
1
log.X  pa / D :
dX aD1 aD1
X  pa

Pp1
Here if we put X D 1, then we get the relation to be proved since p 1 aD1 aD
.p  1/=2. t
u
8.1 Simplest Examples 105

Proof 2. The next proof might appear in an exercise of the elementary algebra.
Here we use a little tool of algebra (Galois theory of cyclotomic fields). We never
use this argument later, so if the readers are not familiar with this tool, they are
recommended to skip this and to go directly to Proof 3. We call the field generated by
the nth roots of unity over the rational number field a cyclotomic field. Gauss studied
cyclotomic fields deeply [35, Chapter 7]. The cyclotomic field Q.p / is a Galois
extension of Q, and the set of the mappings a : p 7! pa for integers a with 1 
a  p1 gives the Galois group of Q.p / over Q. So the sum which appears in (8.1)
Pp1
is equal to the trace of x D 1=.1p / 2 Q.p / over Q, that is, the sum aD1 a .x/,
where a .x/ is the action of the element a in the Galois group on x. This trace is
invariant by each element a of the Galois group, so by the general Galois theory,
we see this is a rational number. The degree of extension of Q.p / over Q (namely
the dimension as a vector space over Q) is p  1, but since we have p D 1  1=x, x
also generates Q.p /, so the minimal polynomial of x over Q is also of degree p 1.
The minimal polynomial of p is 1 C X C X 2 C    C X p1 D .X p  1/=.X  1/,
so the minimal polynomial of 1=x D p  1 is given by ..X C 1/p  1/=X . If
we put X D 1=Y , then this is written as Y 1p ..Y  1/p  Y p /, so the minimal
polynomial of x whose coefficient of the highest degree is one is given by

1 p
.Y  .Y  1/p /:
p

The trace of x is 1 times the coefficient of Y p2 , so by the binomial theorem,

we get

1 p.p  1/ p1
 D :
p 2 2

Hence we prove (8.1). t

u
Proof 3. Proof 2 is somewhat clumsy and similar methods cannot be used so easily
in other cases as imagined. Here we explain a more general method. Let t be
an indeterminate (i.e. a variable, or an element transcendental over the field in
question), and consider the rational function

1
:
1  p t

By expanding this as a formal power series along t D 0, we have

1 1 p1 Pp1
1 X XX c c
cD0 p t
D pi t i D pc t cCpj D :
1  p t i D0 j D0 cD0
1  tp
106 8 Character Sums and Bernoulli Numbers

If jtj < 1, this calculation is valid also as a convergent power series. If we take here
t ! 1, then by the definition of differentiation (or the theorem of l’Hôpital1 ), we
easily see that

1X c
p1
1
D c :
1  p p cD1 p

p
In the above calculation, we used only the properties that p D 1 and p ¤ 1, so
for any a prime to p, we can apply the same calculation to pa instead of p , so we
get also

1 X ac
p1
1
D  c :
1  pa p cD1 p

Here if we take the summation over a, we have

X
p1
X
p1

1 if c  6 0 mod p,
pca D pca  1 D
p  1 if c  0 mod p,
aD1 aD0

so we get

X
p1
1 1X
p1
1 p.p  1/ p1
D  .c/ D  D :
aD1
1  pa p cD1 p 2 2

Hence we prove (8.1). t

u
An advantage of Proof 3 is that we can calculate mechanically without any
special insight on the result beforehand.
An alternative proof for the expression for .1   a /1 that we obtained in Proof 3
above will be given also in Lemma 8.5. By the way, we see of course without any
calculation that the element .1   a /1 of Q.p / is a linear combination of the
p1
basis p , . . . , p of Q.p / over Q. In principle, for any element of a finite field
extension K over k, a concrete expression of the inverse by a basis can be obtained
by Euclidean algorithm on polynomials. But we cannot expect in general that such
a linear combination is written by a simple formula. Since we have an accidentally
easy expression in the above case, we could use it.

1
Guillaume François Antoine de l’Hôpital, (born in 1661 in Paris, France—died on February 2,
1704 in Paris, France).
8.2 Gaussian Sum 107

8.2 Gaussian Sum

Before we start explaining the relation between exponential sums and the general-
ized Bernoulli numbers, we explain about Gaussian sums. The simplest definition
of a Gaussian sum is given by

X
p1
2 2 i
g.p/ D px p D e p

xD0

for an odd prime p. This number was calculated by Gauss and the result is very
beautiful. Namely we have
q
g.p/ D .1/.p1/=2 p :
p
Here the choice of the square root is given as follows. If p  1 mod 4, we take p
p
as a positive number and if p  3 mod 4 we take p so that its argument is =2.
As we shall see later, it is very easy to prove that g.p/2 D .1/.p1/=2 p, but it is
considerably difficult to determine the sign of the square root in g.p/. It is said that
Gauss spent 4 years getting the proof. 2
At the present time, there are many known ways to determine this sign, but it
would be beyond the scope or different from the aim of this book to carry those
proofs here. So we would like to ask readers to refer to some other books on the
determination of the sign. (For example, many proofs are introduced in [15].)
Now, this Gaussian sum can be also expressed in a slightly different way. Let 
be the quadratic residue symbol modulo p defined on p. 88 (i.e. .n/ D pn )
and put

2
Gauss wrote in his letter (Collected Works X, p. 24) to Olbers dated September 3, 1805: “Since
four years ago, a week has seldom passed when I had not made one or another attempt to solve this
difficulty, especially lively now also again in the latest period. But all pondering, all search have
been in vain, and each time sadly I must have laid down my pen again. At last a couple of days
ago it succeeded, not for my wearisome search, but only through the grace of God, I would like
to say. As the lightning strikes, the problem has been solved;    Strangely enough, the solution
of this problem appears now easier than many others which did not well keep me so many days as
these years, and certainly no one will, when I give a lecture on this matter some day, get suspicious
of the long dilemma in which it brought me.” Incidentally, Olbers (Heinrich Wilhelm Matthias
Olbers, 1758–1840) was a medical doctor and an astronomer. The biggest asteroid, Ceres, which
made Gauss leap to fame, was discovered by Italian astronomer Giuseppe Piazzi (1746–1826) on
New Year’s day in 1801, but became lost through the difficulty of observation, then was accurately
located by Gauss by calculation. It was rediscovered by Olbers on New Year’s day in 1802. Also,
he newly discovered the asteroid Pallas in 1802 and Vesta in 1807. Olbers was a friend of Gauss
whom he visited most often.
108 8 Character Sums and Bernoulli Numbers

X
p1
g. / D .n/pn :
nD0

For each n, the number of representatives mod p of x 2 Z such that n  x 2 mod p

is given by 1 C .n/, so we get

X
p1
X
p1
X
p1
g.p/ D .1 C .n//pn D pn C .n/pn :
nD0 nD0 nD0

Here the first sum is obviously 0, so this coincides with g. /.

Here we shall explain more general Gaussian sums which will be needed later.
If we change x 2 to a quadratic form of several variables and integral vectors in
the above definition of Gaussian sum g.p/, we can define the “Gaussian sum of
quadratic forms”. This plays an important role in the transformation formula of
theta functions for example, and it is possible to calculate this as the above formula
of Gauss, though we do not go into this direction here. On the other hand, if we
change into a general Dirichlet character in the definition of g. /, then we can
define a Gaussian sum associated with a character. We treat this in this book.
Definition 8.1. Let  be a primitive Dirichlet character with conductor f and put
f D e 2 i=f . We write

f 1
X
g./ D .n/fn
nD0

and we call this the Gaussian sum associated with character .

In such a general case too, the absolute value of g./ is a simple quantity and
given by g./g./ D f . (The proof will be given later.) But for general , there is
no formula of the Gaussian sum itself which is simple and explicit. When 2 D 1,
there are some formulas which will be explained later.
Lemma 8.2. Let  be a primitive Dirichlet character with conductor f and let a
be an arbitrary integer. Then the following formula holds.

X
f
.a/g./ D .n/fan :
nD1

In particular, we have g./ D .1/g./.

A good point of this relation is that we are not assuming that a is coprime to f
and this is often useful for calculation of exponential sums.
8.2 Gaussian Sum 109

Proofs of Lemma. If a is coprime to f , the relation is obtained trivially by replacing

n by an in the definition of the Gaussian sum g./. If a is not coprime to f , then the
left-hand side is of course 0 by definition of a Dirichlet character, but the right-hand
side is also 0 by the following reasoning. If .a; f / D d > 1, then fan depends only
on n mod f =d , but elements 1  n  f such that 1 mod f =d form a subgroup in
.Z=f Z/ and so for each fixed n0 coprime to f , we have
X X
.n/ D .n0 / .n/ D 0:
nn0 mod f =d n1 mod f =d

The second relation is obtained by putting a D 1 and taking the conjugate of both
sides in the first relation. Namely we have .1/g./ D g./ and since .1/2 D
.1/ D 1, we have .1/1 D .1/. t
u
Since the Gaussian sum is never 0 as we see below, the above formula can be
written also as

X
f
.a/ D g./1 .n/fan : (8.2)
nD1

Lemma 8.3. We have

g./g./ D f:

Proof. By Lemma 8.2, for any integer a, we have

f 1
X a.nm/
.a/.a/g./g./ D .n/.m/f :
n;mD1

Summing up both sides from a D 0,. . . ,f  1, the left-hand side becomes

'.f /g./g./ since .a/.a/ D 1 if .a; f / D 1 and 0 otherwise. Here '.f / is
the Euler function, namely the number of positive integers not more than f which
are coprime to f . By using

f 1

X a.nm/ 0 if n ¤ m,
f D
aD0 f if n D m,

the right-hand side becomes

f 1
X
f  .n/.n/ D f '.f /:
nD1

Comparing both sides, we prove the lemma. t

u
110 8 Character Sums and Bernoulli Numbers

Next we will give a formula for the Gaussian sum for a primitive Dirichlet
character with 2 D 1.
Proposition 8.4. Let  be a non-trivial primitive Dirichlet character such that

2 D 1. Then there exists a quadratic field K such that .u/ D K .u/ D DuK ,
where DK is the fundamental discriminant of K. Furthermore, for such , we have
p
g.K / D DK :

By using the Chinese remainder theorem, we decompose the sum in the definition
of g.K / into powers of primes and we reduce the formula to those in the case of
primes. In the middle of the proof, we need the reciprocity law of quadratic residues.
We omit the details here. We refer to [15]. See also Exercises 8.26 and 8.27.

8.3 Exponential Sums and Generalized Bernoulli Numbers

In this section, we consider sums similar to those in Sect. 8.1, attaching characters to
them. In fact, these new sums are far more interesting. In the previous sections, we
took only the prime power root of unity, but in this section we consider the f th root
f D e 2 i=f of unity. Since it is troublesome to write the suffix f always, we fix a
natural number f once and for all and we denote f simply by . We will explain
below how to obtain formulas for sums like those in the last section by generalized
Bernoulli numbers and simple combinatorial numbers.
First of all, when a is not divisible by f , we get the formula given below. (This
formula will be used also in Chap. 11, Sect. 11.2.)
Lemma 8.5. If f =j a, then we have

f 1
1 1 X ac
D  c :
1  a f cD1

Proof. This is obtained by the same calculation as in Proof 3 in Sect. 8.1, but here
we give more direct proof. By expanding polynomials, we have

f 1 f 1
X X
.1  X /  cX c D .f  1/X f C .c  .c  1//X c
cD1 cD1

Xf
D f X f C Xc:
cD1
8.3 Exponential Sums and Generalized Bernoulli Numbers 111

If we substitute as X D  a in the above, then the second term of the right-hand side
Pf
is zero since cD1  ac D  a .1   af /=.1   a / D 0, and we have
f 1
X
.1   a / c ac D f  af D f;
cD1

Dividing both sides by f .1   a /, we have the equality in the lemma. t

u
Now we consider a Dirichlet character  whose conductor is f (i.e. a primitive
Dirichlet character modulo f ). First, we put

X
f
.a/
Pk ./ D
aD1
.1   a /k

and try to describe this. Of course if f D 1, then the denominator becomes zero so
this has no meaning. So we assume that f > 1 from now on. As in the last section,
we define the Gaussian sum with respect to  by

X
f
g./ D .a/ a :
aD1

To treat the exponential sums slightly more generally, we put

X
f
.a/
Pk; .t/ D :
aD1
.1   a e t /k

We have Pk; .0/ D Pk ./. The next relation holds.

P1; .0/ D g./B1; : (8.3)

Proof. By the last lemma and Lemma 8.2, we have

f 1
1 X X X
f f
1
P1; .0/ D  .a/ c ac D  g./ .c/c D g./B1; :
f aD1 cD1
f cD1

In the last equality, we used the formula (4.1) for B1; in Sect. 4.2 (p. 54). t
u
In fact, P1; .t/ is almost identical to the definition of the generalized Bernoulli
numbers. Indeed, calculating the development, we have

1 1 Pf 1
X
f
X X j D0 .j /e
jt
P1; .t/ D aj jt
.a/ e D g./ .j /e jt
D g./ :
aD1 j D0 j D0
ef t  1
112 8 Character Sums and Bernoulli Numbers

(Here we are regarding e t as a formal variable and expanding it as a formal power

series. Or we may say we are expanding it in the neighborhood of t D 1.
Anyway, in the final stage, it is an equality between rational functions of e t , and the
calculation is justified regardless of any interpretation.) Since we are taking f > 1,
we have .0/ D .f / D 0. So by definition, we have
1
X t n1
P1; .t/ D g./ Bn; :
nD1
nŠ

Hence we have
d l P1; .l/ 1
.0/ D P1; .0/ D g./ BlC1; : (8.4)
dt l l C1
Next we want to give a formula for Pk; .0/. Now we put
1
Qk;a .t/ D :
.1   a e t /k
Then we have
X
f
Pk; .t/ D .a/Qk;a .t/;
aD1

and
d
Qk;a .t/ D k.QkC1;a .t/  Qk;a .t//;
dt
so inductively Qk;a .t/ is expressed by higher-order derivatives of Q1;a .t/, so Pk; .t/
is also expressed by derivatives of higher order of P1; .t/. More explicitly, by
using the Stirling number of the first kind (as for the definition, see Sect. 2.1), it
is expressed as follows.
Proposition 8.6.

k1 
X
1 k .l/
Qk;a .t/ D Q1;a .t/;
.k  1/Š l C1
lD0

k1 
X
1 k .l/
Pk; .t/ D P1; .t/:
.k  1/Š l C1
lD0

Proof. This is proved by induction with respect to k. The assertion is true for
k D 1. We assume it is true up to k. Since QkC1;a .t/ D Qk;a .t/ C k1 dt d
Qk;a .t/,
we can express QkC1;a .t/ by the linear combination of derivatives of Qk;1 .t/ by the
.l/
inductive assumption, and the coefficient of Q1;a .t/ in QkC1;a .t/ in this expression
is given by
8.3 Exponential Sums and Generalized Bernoulli Numbers 113

  
1 k 1 k
C
.k  1/Š l C1 k l

for 0  l  k. By the recurrence relation of the Stirling number of the first kind
(p. 28, Eq. (2.2)), this is equal to

1 kC1
:
kŠ l C 1

Hence we prove the assertion. t

u
Corollary 8.7.

X k1 
g./ X
f
.a/ k BlC1;
Pk; .0/ D D : (8.5)
aD1
.1   /
a k .k  1/Š l C1 l C1
lD0

Proof. In the second formula in the proposition, put t D 0 and then use (8.4). t
u
Example 8.8. For example, we have

f 1
X .a/
D g./B1; ;
aD1
.1   a /
f 1
X  
.a/ 1
D g./ B1; C B2; ;
aD1
.1   a /2 2
f 1
X  
.a/ 3 1
D g./ B1; C B2; C B3; :
aD1
.1   a /3 4 6

In the above, we assumed that the character  is primitive in order to evaluate

.l/
P1; .0/ by Bernoulli numbers in the final stage, but we did not use this condition at
any other place. So, in the above, even if we take 1 instead of .a/ for any a, we
can execute almost the same calculation. For example, we get

f 1 f 1
X X 1 f 1
Q1;a .t/ D D 
aD1 aD1
1   a et 1  ef t 1  et
 
f ef t et
D f  ft  1 t
e 1 e 1
1
X .1  f n /t n1
D f 1C Bn :
nD1
nŠ
114 8 Character Sums and Bernoulli Numbers

Hence if we put t D 0 here, we have

f 1
X 1 f 1
D (8.6)
aD1
1  a 2

since B1 D 1=2. The case f D p is the formula in Sect. 8.1.

We give similar formulas in more general cases. Noting the relation
f 1
X .l/ 1
Q1;a .0/ D .1  f lC1 /BlC1 C ıl0 .f  1/
aD1
l C1

(where ıl0 is Kronecker’s delta, that is, 1 for l D 0 and 0 otherwise), we get the
following formula.
Lemma 8.9.
f 1 f 1
X 1 X
D Qk;a .0/
aD1
.1   /
a k
aD1

k1 
X

1 k 1
D .1  f lC1
/BlC1 C .f  1/ıl0 :
.k  1/Š l C1 l C1
lD0

We give examples of this formula.

Example 8.10.

f 1
X 1 f 1
D ;
aD1
1 a 2
f 1
X 1 .f  1/.f  5/
D ;
aD1
.1   /
a 2 12
f 1
X 1 .f  1/.f  3/
D ;
aD1
.1   /
a 3 8
f 1
X 1 .f  1/.f 3 C f 2  109f C 251/
D :
aD1
.1   a /4 720

Using similar methods, we will give formulas for the following sums:

f 1
X
c k  ca :
cD0
8.3 Exponential Sums and Generalized Bernoulli Numbers 115

When f divides a, then we have  ac D 1, so this should give the formula to describe
sums of powers. Also under the assumption that f does not divide a, we give
formulas to express these sums by Qk;a .0/. This is a generalization of the formula to
express the right-hand side of Lemma 8.5 by the left-hand side. (These formulas will
be applied to calculations
  of exponential sums later.) As usual, for natural numbers
k, j , we denote by jk the binomial coefficient
!
k k.k  1/    .k  j C 1/
D :
j jŠ
k 
Here for j D 0, we put 0 D 1.
Lemma 8.11. (1) When a is divisible by f , we have
!
X
f
X
f
X k
k f kj C1
k ac
c  D c D k
Bj :
cD0 cD0 j D0
j k  j C 1

(2) When a is not divisible by f , we have

0 !
1
X
f
X
k X
kmC1 
k kj C1 A
c k  ac D .1/m .m1/ŠQm;a .0/@ .1/kj f j :
cD0 mD1 j D1
j m

The first formula is nothing but the formula (1.1) in Chap. 1.

Proof. First we assume that a is an arbitrary integer. To extract c k by differentiating
an easily handled function, we consider the following transformation, taking t as a
variable.
0 1
f 1 f 1
X d k X
c k  ca e ct D k @  ac e ct A
cD0
dt cD0
 
dk 1  ef t
D k
dt 1   a et
!
.k/
X k
k .j /
D Q1;a .t/  f kj e f t Q1;a .t/: (8.7)
j D0
j

To show the last equality, we used the Leibniz rule3 on the higher-order derivatives
of products.

3
Gottfried Wilhelm Freiherr von Leibniz (born on July 1, 1646 in Leipzig, Saxony (now
Germany)—died on November 14, 1716 in Hannover, Hanover (now Germany)).
116 8 Character Sums and Bernoulli Numbers

If f ja, then we have

X1
1 et Bn n1
Q1;a .t/ D Q1;0 .t/ D  D 1  D 1  t :
e 1
t e 1
t
nD0
nŠ

This has a pole at t D 0. So comparing the constant terms of both sides of (8.7) as
a power series expansion with respect to t and noting that

.0/ 1 B2
Q1;a .t/ D  C .1  B1 /  t  ;
t 2
.j / .1/j j Š Bj C1 Bj C2
Q1;a .t/ D    t  .j  1/;
t j C1 j C 1 .j C 2/
1
X f nt n
ef t D ;
nD0
nŠ

we get

f 1
! !
X BkC1 X k kj Bj C1 X k kj f j C1
k k
c D
k
C f C f  .1/j j Šf k
cD0
k C 1 j D0
j j C 1 j D0
j .j C1/Š
!
X
k1
k Bj C1 f kC1
D f kj C fk
j D0
j j C 1 k C 1
!
X
k1
k Bj C1 f kC1
D f kj C fk
j D0
j C1 kj kC1
!
X k
k Bj f kC1
D f kj C1 C  f k:
j D1
j k  j C 1 k C 1

Pk   1
j k
(In the second equality, we used the fact j D0 .1/ j j C1 D 1
kC1 .) So we get

!
X
f
X k
k f kj C1
c D k
Bj :
cD0 j D0
j kj C1

In this way, we obtained the formula for the sum of powers again.
.n/
Next, we assume that a is not divisible by f . Here we want to express Q1;a .t/
by Qk;a .t/. This is a converse of the formula in Proposition 8.6 and this time, it is
described by using the Stirling number of the second kind. (As for definition, see
Sect. 2.1.)
8.3 Exponential Sums and Generalized Bernoulli Numbers 117

In the formula (5.1) of Proposition 2.6 p. 28, if we substitute as m ! mC1; n !

k C 1, we get

X
kC1

kC1 l
ımC1;kC1 .D ım;k / D .1/kC1 .1/l :
l mC1
lDmC1

By virtue of this equality and Proposition 8.6, we have

.k/
X
k
.m/
Q1;a .t/ D ım;k Q1;a .t/
mD0

X
k X
kC1

kC1 l .m/
D .1/ kC1
.1/l
Q1;a .t/
mD0
l mC1
lDmC1

X
kC1
X
l1 
kClC1 k C 1 l .m/
D .1/ Q1;a .t/
l mD0
m C 1
lD1

X
kC1

kC1
D .1/kClC1 .l  1/Š Ql;a .t/:
l
lD1

So putting t D 0 in (8.7), we have

f 1
X X
kC1

kC1
c k  ac D .1/kCmC1 .m  1/Š Qm;a .0/
cD0 mD1
m
! kj C1

X k
k X kj C1
C fj .1/kj Cm.m  1/Š Qm;a .0/
j D0
j mD1
m
! kj C1

X k
k X kj C1
D fj .1/kj Cm .m  1/Š Qm;a .0/:
j D1
j mD1
m
0 !
1
X k X
kmC1 
k k  j C 1
D .1/m .m  1/ŠQm;a .0/ @ .1/kj f j A:
mD1 j D1
j m

t
u
By the above calculation, we get the following examples.
118 8 Character Sums and Bernoulli Numbers

Example 8.12.

f 1
X f
c ac D  ;
cD0
1  a
f 1
X f 2 C 2f 2f
c 2  ac D  ;
cD0
1 a .1   a /2
f 1
X f 3 C 3f 2  3f 3f 2 C 9f 6f
c 3  ac D C C :
cD0
1 a .1   /
a 2 .1   a /3

Application of these formulas will be given in the next section.

8.4 Various Examples of Sums

In this section, we treat several sporadic examples. Sometimes we can use the
similar method for sums on character values as for sums on roots of unity. We
shall see this. Let  be a primitive Dirichlet character modulo f . We consider the
following sum.
f 1
X
Sk ./ D .a1 C    C ak /a1    ak :
a1 ;:::;ak D0

Here we should note that the sum depends on the range of natural numbers ai
(namely, if we change them to other representatives modf , then the value of the
sum would change). Even for k D 2, it is not an efficient method to calculate this
kind of sum by dividing the sum into various pieces and to seek the answers by
case-by-case analysis. First of all, let us recall the following relation (cf. (8.2)).

X
f
.a/ D g./1 .n/ an :
nD1

By virtue of this relation, we get

f 1
X X
f
Sk ./ D g./1 .n/ n.a1 CCak / a1    ak : (8.8)
a1 ;:::;ak D0 nD1

But since we have

f 1
X f
a an D  ;
aD0
1  n
8.4 Various Examples of Sums 119

we obtain

X
f
.n/
Sk ./ D .f /k g./1
nD1
.1   n /k

D .f /k g./1 Pk; .0/


.1/kC1 f k X 1
k1
k
D BlC1; :
.k  1/Š l C1 l C1
lD0

For example, we have

 
1
S2 ./ D f B1; C B2; ;
2
2
 
3 1
S3 ./ D f B1; C B2; C B3; :
3
4 6

We also consider the following similar examples.

f 1
X
S.; e1 ; : : : ; ek / D .a1 C    C ak /a1e1    akek :
a1 ;:::;ak D1

The formula to describe this by Bernoulli numbers can be obtained by almost the
same method as before. Indeed, as before, writing the character values by Gaussian
sum and roots of unity, we have
0 1
f 1
X
f
Y
k X
S.; e1 ; : : : ; ek / D g./1 .n/ @ c ej  cn A
nD1 j D1 cD1

X
f
X
Dg./1 .n/ .1/e1 CCek l1 lk Cm1 CCmk f l1 CClk
nD1 1lj ej
1mj ej lj C1
1j k
! !
e1 ek
  Qm1 CCmk ;n .0/.m1  1/Š    .mk  1/Š
l1 lk



e1  l1 C 1 ek  lk C 1
 
m1 mk
! !
X
e1 CCek l1 lk l1 CClk e1 ek
D .1/ f 
1l e
l1 lk
j j
1mj ej lj C1
1j k
120 8 Character Sums and Bernoulli Numbers

.1/m1 CCmk
 .m1  1/Š    .mk  1/Š
.m1 C    C mk  1/Š



e1  l1 C 1 ek  lk C 1
 
m1 mk
!
X k 1 1  m1 C    C mk
m1 CCm
 BlC1; :
l C1 l C1
lD0

Below, we sum up the above formula again.

Proposition 8.13. We have the following formula.
!
X
e1 X
ek Y k
ej
S.; e1 ; : : : ; ek / D .1/e1 CCek C1  .f /l1 CClk
j D1
lj
l1 D1 lk D1

e1 l
X 1 C1 ek l
X k C1 Y k

.1/m1 CCmk ej  lj C 1
  .mj  1/Š
m1 D1 mk D1
.m1 C    C mk  1/Š j D1 mj

m1 CCm  !
X k 1 1 m1 C    C mk
 BlC1; :
l C1 l C1
lD0

Example 8.14. We give examples of applications of the above lemma below.

‚ …„ ƒ
k 
.1/kC1 f k X 1
k1
k
Sk ./ D S.; 1; : : : ; 1/ D BlC1; ;
.k  1/Š l C1 l C1
lD0

f 1
!
X X
e1
f el e
S.; e/ D .n/ne D BlC1; ;
nD0
l C1 l
lD0

f 1
X  
.f C 1/ 1
S.; 2; 1/ D .m C n/m2 n D f 2 f B1; C B2; C B3; :
m;nD0
2 3

We also give another similar example. Let m be a natural number which is

coprime to a fixed f and smaller than f . We consider the following sum.

X
f
Dm ./ D .a C m/a:
aD1

Then we have

X
f
N 1
Dm ./ D g./ N
a.n/ .aCm/n

a;nD1
8.5 Sporadic Examples: Using Functions 121

0 1
X
f  
f
N 1 @
D g./ N
.n/ mn
 Cf A;
.1   n /
nD1;.n;f /D1

but since we have

X
f
X
m1 X
f
X
m1
N
.n/.1   mn /.1   n /1 D N
.n/ nl
D g./
N .l/;
nD1;.n;f /D1 lD0 nD1;.n;f /D1 lD1

Pf
taking the relation N
nD1 .n/.1   n /1 D g./B
N 1; in the last section into
account, we get
!
X
m
Dm ./ D f .l/ C B1; :
lD1

We will give several similar examples in the exercise of this chapter.

8.5 Sporadic Examples: Using Functions

Lemma 8.15. Let f > 1 be an odd number and let  be a primitive character with
conductor f . Then we have

f 1
X .a/
aD1
.1   a /.1   2a /2

 
1 1 1
D g./ B3;N C .1 C .2//B2;N C .3 C 5.2//B1;N :
24 4 8

Proof. If we put

X
f
.a/.1 C  a e t /
Qk; .t/ D ;
aD1
.1   2a e 2t /k

X
f
.a/
Pk; .t/ D ;
aD1
.1   a e t /k

then we get
1
X t n1
Q1; .t/ D P1; .t/ D g./ Bn;N :
nD1
nŠ
122 8 Character Sums and Bernoulli Numbers

Since

d
Qk; .t/ D 2kQkC1; .t/ C .1  2k/Qk; .t/  .2/Pk; .2t/:
dt

we can write Qk; .0/ inductively by Pk; .0/, and then by Bernoulli numbers. For
example, if we put k D 1 in the above relation and take the derivatives, we get
0
Q1; .t/ D 2Q2; .t/  Q1; .t/  .2/P1; .2t/;

0 1 00 0 0
Q2; .t/ D .Q .t/ C Q1; .t// C .2/P1; .2t/;
2 1;
and
 
g./1
Q2; .0/ D  B2; C .1 C .2//B1; ;
2 2
   
0 g./ 1 1
Q2; .0/ D  B3; C C .2/ B2; :
2 3 2

In the same way we get

0
Q2; .t/ D 4Q3; .t/  3Q2; .t/  .2/P2; .2t/

and
1 0
Q3; .0/ D .Q .0/ C 3Q2; .0/ C .2/P2; .0//:
4 2;

Since the left-hand side of the lemma is equal to Q3; .0/, substituting this for the
above value and using the relation P2; .0/ D g./.B1; C B2; =2/, we get the
right-hand side. t
u

8.6 Sporadic Examples: Using the Symmetry

In the examples below, we denote by .n/ D . pn / the quadratic residue character

where p is an odd prime.
8.6 Sporadic Examples: Using the Symmetry 123

Example 8.16. We put

X
p1
I D .a/a:
aD1

By definition, this is equal to pB1; and we know already that B1; D 0 if

.1/ D 1 (i.e. if p  1 mod 4 by the quadratic reciprocity law). This is also
shown directly as follows. We assume that p  1 mod 4. Substituting p  a for a
and noting .a/ D .a/ D .p  a/, we get

X
p1
X
p1
I D .p  a/.p  a/ D .a/.p  a/:
aD1 aD1

Hence adding this to the original expression, we get

X
p1
X
p1
2I D .a/.a C p  a/ D p .a/ D 0:
aD1 aD1

Of course this argument cannot be used when p  3 mod 4. Actually, if p 

3 mod 4 and p > 3, then I =p is the class number of the imaginary quadratic field
p
Q. p/, so it is a more complicated value.
Now there are many cases where the above kind of substitution is effective in our
calculation. For example:
Example 8.17. The character being as above, put

X
p1
I D .l/lmn:
l;m;nD1;4lnm2 mod p

Then substituting m ! p  m, we get

X
p1
2I D .l/ln.m C p  m/
l;m;nD1;4lnm2 mod p

X
p1
Dp .l/l. .ln/ C 1/n
l;nD1

X
p1
Dp .l .n/n C .l/ln/
l;nD1

D p 3 .p  1/B1; :
124 8 Character Sums and Bernoulli Numbers

So we have

p 3 .p  1/
I D B1; :
2
Example 8.18. The next sum was first considered in connection
 with dimension
formulas of automorphic forms [92]. We put .n/ D pn , where p is a prime and

S D f.s; t; r/ 2 F3p I s.r C s/.s C t/ ¤ 0g:

Put  D e 2 i=p and define a sum I by

X .s 2  rt/
I D :
. rCs  1/. sCt  1/. s  1/
.s;t;r/2S

At first glance, it might seem difficult to calculate this sum in an elementary way,
but actually this can be easily calculated as follows. If we put a D r C s, b D s C t,
c D s, then s 2  rt D c 2  .a C c/.b C c/ D ab  bc  ca. Hence if we put
T D F3 p , then substitute a, b, c for a, b, c, we get

X .ab  bc  ca/
I D
. a  1/. b  1/. c  1/
.a;b;c/2T

X .ab  bc  ca/
D
. a  1/. b  1/. c  1/
.a;b;c/2T

X .ab  bc  ca/ a  b  c
D
.1   a /.1   b /.1   c /
.a;b;c/2T

X .ab  bc  ca/. a  1 C 1/. b  1 C 1/. c  1 C 1/

D :
. a  1/. b  1/. c  1/
.a;b;c/2T

So by expanding the numerator and transposing the same term to the left-hand
side, we get
X
2I D  .ab  bc  ca/
.a;b;c/2T

1 1 1
 C C
.1   a /.1   b / .1   b /.1   c / .1   c /.1   a /

1 1 1
   C 1 :
1  a 1  b 1  c
8.6 Sporadic Examples: Using the Symmetry 125

If we fix a and b, then we have


X  .ab/ if a C b ¤ 0 ;
.ab  .a C b/c/ D
c2F
.p  1/ .ab/ D .p  1/ if a D b ¤ 0 :
p

So if we fix a ¤ 0, then we have

X
p1
X
.ab  bc  ca/ D .p  1/  .ab/ D p:
b;cD1 b¤a; b¤0

Here by Examples 8.8 and 8.10, we have

X .ab/
D .1/g. /2 B1;2 D pB1;2 ;
.1   a /.1   b /
a;b2F
p

X 1 p.p  1/ p.p  1/.p  5/

p D C :
.1   a /.1   a / 2 12
a2F
p

So noting p  1 D p  .a2 /, we get

X 1 p.p  1/.p  5/ p.p  1/
.abbcca/ DpB1;2 C C :
.1 a /.1 b / 12 2
.a;b;c/2T

We have also
X .ab  bc  ca/ p.p  1/
D :
1  a 2
.a;b;c/2T

X
.ab  bc  ca/ D p.p  1/:
.a;b;c/2T

So as a total, we have

3 p.p  1/.p  5/ p.p  1/ 3 p.p  1/2

I D pB1;2   D pB1;2  :
2 8 2 2 8
In particular, if p  1 mod 4 then B1; D 0, and if p  3 mod 4 and p > 3,
then B1; D h.p/ (h.p/ is the class number of the imaginary quadratic field
p
Q. p/), so there exists a formula for I by class numbers.
Example 8.19. We put S D f.a; b; c/ 2 Z3 I 1  a; b; c  p  1; ab C bc C ca 
0 mod pg and
126 8 Character Sums and Bernoulli Numbers

X
I D .abc/abc:
.a;b;c/2S

The condition in the sum I is complicated and it is not easy to calculate I . But if
we assume that .1/ D 1, then I can be calculated as follows. By the relation
X
I D .abc/.p  a/.p  b/.p  c/
.a;b;c/2S

we get
X
2I D .abc/..p  a/.p  b/.p  c/ C abc/
.a;b;c/2S
X
D .abc/.p 3  p 2 .a C b C c/ C p.ab C bc C ca//:
.a;b;c/2S

But we have .abc/ D .c 2 .a C b// D .a C b/ and since the condition of the
definition of S is symmetric with respect to a, b, c, we have
X X
.abc/.ab C bc C ca/ D 3 .a C b/ab
.a;b;c/2S .a;b;c/2S
X
D3 .a C b/ab
a;b;aCb60 mod p

X
p1  
1
D3 .a C b/ab D 3p 2 B1; C B2; :
2
a;bD1

Since we assumed .1/ D 1, we have B1; D 0 now. Moreover we have

!
X X X
p1 p1
.a C b/.a C b C c/ D 3 .a C b/ a
.a;b;c/2S aD1 bD1

X
p1
D 3 .a/a D 3pB1;
aD1

and this is equal to 0. We have

X X
p1
X
p1
.a C b/ D .a C b/ D . .a// D 0:
a;b;c2S a;bD1 aD1
8.7 Sporadic Example: Symmetrize Asymmetry 127

So we have

3p 3
I D B2; :
4

Remark 8.20. When .1/ D 1, a simple formula to write down I by general-
ized Bernoulli numbers is known only conjecturally.

8.7 Sporadic Example: Symmetrize Asymmetry

Let .x/ be the quadratic residue character modulo p (p is a prime) and consider
the following special sum.

X
p1
I D .m  4n/ .m/mn:
m;nD1

A formula for this sum was first given in [91] and [39] by comparing the trace
formula and algebraic geometry, but this can be calculated in an elementary way as
well. We illustrate this calculation below.
In the above definition, if we take n instead of m  4n, then the sum is very
simple and given by
!2
X
p1
X
p1
.mn/mn D .m/m D p 2 B1;2 :
m;nD1 mD1

But the sum I we actually defined above is subtly breaking its symmetry and that
obstructs an easy calculation. We get over this by changing it to an exponential sum
and transforming it to a symmetric one. By the relation

X
p1
X
p1
1 1
.m/ D g. / .a/ am
.m  4n/ D g. / .b/ b.m4n/ ;
aD1 bD1

we have

X
p1
X
p1
I D g. /2 .ab/ mn amCb.m4n/ :
a;bD1 m;nD1

Pp1 p
But for c 62 pZ, we have mD1 m
cm
D  1 c , hence we have
128 8 Character Sums and Bernoulli Numbers

0 1
B X p1
.ab/ X p1
C
I D g. /2 @p 2 C .1/ mn 4nb A
a;bD1
.1   4b /.1  aCb /
m;n;bD1
a6b mod p

X
p1
.ab/ p.p  1/2
D p .1/  ;
a;bD1
.1   4b /.1  aCb / 4
aCb¤p

using g. /2 D .1/p: In order to calculate I from the above relation, it is

sufficient to calculate

X
p1
.ab/
J D .1/ :
a;bD1
.1   4b /.1   aCb /
aCb¤p

Now a direct calculation of J seems difficult partly because the sum in the definition
of J is not symmetric with respect to a and b. So we try to rewrite this in a
symmetric way. If we exchange a and b, this is just a change of notation, so we
have

X
p1
.ab/
J D .1/ :
a;bD1
.1   4a /.1
  bCa /
aCb¤p

Now we will add this to the original expression and take the average of both. To
calculate this, we use the following equality.

1 1 2   4a   4b 2 4aC4b   4a   4b

C D D
1   4a 1   4b .1   4a /.1   4b / .1   4a /.1   4b /
 4aC4b  1 C .1   4a /.1   4b /
D
.1   4a /.1   4b /
1   4aC4b
D 1 :
.1   4a /.1   4b /

After all, if we put

X
p1
.ab/
J1 D ;
a;bD1
1   aCb
aCb¤p

X
p1
.ab/.1   4aC4b /
J2 D ;
a;bD1
.1   4a /.1   4b /.1   aCb /
aCb¤p
8.7 Sporadic Example: Symmetrize Asymmetry 129

then we have

.1/
J D .J1  J2 /:
2
Here we have

1X X
p1 p1
J1 D  .ab/k k.aCb/
p a;bD1
kD1
aCb¤p

1 p.p  1/ .1/.p  1/
D  .g. /2  .1/.p  1//  D :
p 2 2
Also we have

X
p1
1   4aC4b
J2 D .ab/
a;bD1
.1   4a /.1  4b /.1   aCb /
aCb¤p

p1 P3
X .ab/ l.aCb/ X
p1
4 .1/
D lD0
 4a /.1   4a /
:
.1   4a /.1   4b / aD1
.1  
a;bD1

Now, by virtue of Example 8.10, we have

p1  
X
p1
1 X 1 1
D 
aD1
.1   4a /.1   4a / aD1
1   4a .1   4a /2

p1 .p  1/.p  5/ p2  1
D C D :
2 12 12
Furthermore we have

X
p1
.a/ la 1 X
p1
1 X
p1
D  .a/c 4acCal
D  g. / .4c C l/c:
aD1
1   4a p a;cD1 p cD1

Pp1
Hence if we put S.l/ D cD1 .4c C l/c, then we have

.1/ X
3
.1/.p 2  1/
J2 D S.l/2  :
p 3
lD0

Summing up, we have

1 X 1 X
3 3
p1 p2  1 .p  1/.2p  1/
J D C  S.l/2 D  S.l/2 :
4 6 2p 12 2p
lD0 lD0
130 8 Character Sums and Bernoulli Numbers

Next the following quantity should be evaluated.

!2 !2
X
3 X
p1
X
p1
S.l/ D2
.4c/c C .4c C 1/c
lD0 cD1 cD1
!2 !2
X
p1
X
p1
C .4c C 2/c C .4c C 3/c : (8.9)
cD1 cD1

There are many ways to evaluate this, but here we calculate it just directly. (As for
more general method of calculation, see [45].) First we have

X
p1
. .2c/c C .2c C 1/c/
cD1
!
1 X X
p1 p1
D . .2c/  .2c/ C .2c C 1/.2c C 1//  .2c C 1/
2 cD0 cD0
!
1 X
2p1
D .c/c
2 cD1

1X X
p1 p1
D .c/.c C c C p/ D .c/c:
2 cD1 cD1

Hence we have

X
p1
X
p1
.2c C 1/c D .1  .2// .c/c:
cD1 cD1

In the same way, we have

X
p1
1 X
4p1
. .4c/c C .4c C 1/c C .4c C 2/c C .4c C 3/c/ D .c/c
cD1
4 cD1

X
p1
D .c/c:
cD1

This implies that

X
p1
X
p1
. .4c C 1/ C .4c C 3//c D .1  .2// .c/c:
cD1 cD1

Now we consider the case p  1 mod 4 and p  3 mod 4 separately.

8.7 Sporadic Example: Symmetrize Asymmetry 131

First if we assume p  3 mod 4, the .1/ D 1. So we have .4c C 3/ D

 .4.p  c/  3/ D  .4.p  1  c/ C 1/, and

X
p1
X
p1
X
p1
.4c C 3/c D  .4c C 1/.p  1  c/ D .4c C 1/c:
cD0 cD0 cD0

Hence in this case we have

X
p1
X
p1
1  .2/ X
p1
.4c C 1/c D .4c C 3/c D .c/c:
cD1 cD1
2 cD1

So we have
  X !2
X
3
.1  .2//2
p1
S.l/ D 1 C .1  .2//2 C
2
.c/c
2 cD1
lD0

D .4  3 .2//p 2 B1;2 :

By the assumption p  3 mod 4, if p ¤ 3 then we have B1; D h.p/, where

p
h.p/ is the class number of the imaginary quadratic field Q. p/ (p. 90, (6.5)).
Next we assume that p  1 mod 4. In this case we have B1; D 0, so we cannot
Pp1 Pp1
separate cD1 .4c C 1/c and cD1 .4c C 3/c in the above expression in the
same way. In order to separate these two, we introduce a Dirichlet character ı with
conductor 4. Here for x  3 mod 4 we have ı.x/ D 1 and we see that ı is a
primitive Dirichlet character with conductor 4p. By definition, we have

X
4p1
X
p1
4pB1; ı D .c/ı.c/c D . .4c C 1/.4c C 1/  .4c C 3/.4c C 3//
cD0 cD0

D 4.S.1/  S.3// D 8S.1/ D 8S.3/:

By virtue of the fact B1; D 0, we have S.0/ D S.2/ D 0. So we get

p2
S.1/2 C S.3/2 D .B1; ı /2 :
2
Also we have h.p/ D B1; ı . Summing up all the above results, we get
⎧
⎪
⎪7=2 if pD3 ;
p.p1/.p2/ ⎨ 2
I D  p  h.p/2 =4 if p  1 mod 4;
12 ⎪
⎪
⎩.4  3 .2//p 2  h.p/2 =2 if p  3 mod 4 and p ¤ 3:

By the way, as for further generalization of the results in this section, see [45].
132 8 Character Sums and Bernoulli Numbers

8.8 Quadratic Polynomials and Character Sums

The example given below is not related to Bernoulli numbers so much, but since the
method is similar, we take this opportunity
  to explain it. Here too, we assume that
p is an odd prime and .x/ D p is the quadratic residue character modulo p.
x
Pp1
Let P .x/ be a polynomial in a variable x. If we consider the sum xD0 .P .x//,
then this is related with the number of pairs .x; y/ 2 F2p of integral solutions of
y 2 D P .x/, so related with the congruence zeta function of algebraic curves and
a complicated quantity. There is no simple formula for general P .x/. But in the
special case where P .x/ is a quadratic polynomial, it corresponds to a conic. In this
case the structure of the curve itself is simple and there is a simple formula for the
sum as above. Let us practice the calculation. We write a quadratic polynomial of
x as P .x/PD ax 2 C bx C c (a, b, c 2 Z, a 6 0 mod p). It is not so difficult to
calculate x mod p .P .x//. Since we take p to be an odd prime, we consider the
finite field Fp and rewrite P .x/ D a..x  d /2 C e/. Then we have
X X
.P .x// D .a/ .x 2 C e/:
x mod p x mod p

So this right-hand side is the point.

Lemma 8.21.

X 1 if e 6 0 mod p;
.x C e/ D
2

x mod p p1 if e  0 mod p:

Proof. If we write  D e 2 i=p , then we have

1 X
p1
.k/ k.x Ce/ ;
2
.x 2 C e/ D
g. /
kD0

where g. / is the Gaussian sum. So we have

X
p1
1 X
p1
X
p1
 k.x Ce/ :
2
.x C e/ D
2
.k/
xD0
g. / xD0 kD0

Pp1 2
But we have xD0  kx D .k/g. /, so we get the following result.

1

X X
p1
1 if e 6 0 mod p,
2 ke
.k/  D  ke
D
kD0 kD1 p1 if e  0 mod p.

t
u
8.9 A Sum with Quadratic Conditions 133

By the way, the result above has of course a relation to the number of pairs
.x; y/ 2 F2p such that x 2 Ce D y 2 . Namely, for each x 2 Fp , there are 1C .x 2 Ce/
number of y which satisfy this equation. So p  1 pairs of .x; y/ are solutions of
this equation if e 6 0 mod p.
In general, also for a quadratic form Q.x/ of n variables, that is, a homogeneous
polynomial of variables
P xi of x D .x1 ; x2 ; : : : ; xn / of degree two, we sometimes
consider the sum x mod p  Q.x/ and the results are well known, but this is remote
from Bernoulli numbers and we omit it here.
Pp1 Pp1
Next we treat the sum xD0 .x 2 C e/x. (Note that the sum xD0 .P .x//x
for a general quadratic polynomial P .x/ does not reduce to this sum.) Here we have

X
p1
X
p1
X
p1
.x 2 C e/x D .p  x/ ..p  x/2 C e/ D .p  x/ .x 2 C e/;
xD1 xD1 xD1

so we have
!
X
p1
X
p1
X
p1
2 .x C e/x D p
2
.x C e/ D p
2
.x C e/  .e/ :
2

xD1 xD1 xD0

Hence we have

X
p1
 .1C 2.e//p if e 6 0 mod p;
.x 2 C e/x D p.p1/
xD0 2 if e  0 mod p:

As an application, when e 62 pZ, we can show the following formula.

X
p1
.1 C .e//p 2 .p  1/
.x 2 C ey 2 /xy D  :
x;yD0
4

We omit the proof here.

8.9 A Sum with Quadratic Conditions

For a polynomial f .x; y; z/, we put

M.f / D f.a; b; c/ 2 Z3 I 1  a; b; c  p  1I f .a; b; c/  0 mod pg:

Proposition 8.22. For an integer e 6 0 mod p, we write he .x; y; z/ D exz  y 2 .

Then we have
134 8 Character Sums and Bernoulli Numbers

X .a/ p1
D .1 C .e//g. /B1; :
.1   a /.1   b /.1   c / 4
.a;b;c/2M.he /

Proof. Denote the left-hand side by I . In the definition of I , we may replace b by

b. Noting that

.1   b /1 C .1   b /1 D 1;

and taking integers aN and eN such that aaN  e eN  1 mod p, we have

X
p1
.a/
2I D
a;bD1
.1   a /.1   aN eb
N 2/

!
X
p1
.a/ 1 X
p1
D  m aN eb
N 2m
.1   /
a p mD1
a;bD1

1 X
p1
.a/
D m. .aN em/g.
N /  1/
p a;mD1 .1   a /

p1 p1
D N
.e/g. /B1;  g. /B1;
2 2
p1
D .1 C .e//g.
N /B1; :
2
t
u
As an application of this formula, we can show the following.
Corollary 8.23.

X p 3 .p  1/
.n/lmn D .1 C .e//B1; :
4
.l;m;n/2M.he /

Proof. For a while, for any rational number ˛ whose denominator is not divisible
by p, we understand that we define  ˛ by regarding ˛ as an element of Fp . Then
we have

X
p1
X
p 3 I D lmn .a/ alCbmCcn
l;m;nD1 .a;b;c/2M.he /

X
p1
X
p1
ae m2 X
p1
n ae m 2
D lmn .a/ al   4n  ae .bC 2n /

l;m;nD1 aD1 bD1

8.9 A Sum with Quadratic Conditions 135

X
p1
lmn al .a/ aem
2 =4n 2 =4n
D . .n=.ae//g. /   aem /
l;m;n;aD1

X
p1
lmn al . .ne/ aem
2 =4n
D g. /  .a//
l;m;n;aD1

X
p1
X
D .g. / .ne/  .l/g. //lmn C pg. / .n/lmn:
l;m;nD1 4lnem2 mod p

Hence we have
0 1
g. / @ p 2 .p  1/2 X
I D .1 C .e//pB1; p .n/lmnA :
p3 4
.l;m;n/2M.h4=e /

Pp1 2
In the above calculation, we used xD0  cx D .c/g. /. Here in the final
expression of I and in the first term of the right-hand side above, the number e
appears only as .e/. Since we have .4=e/ D .e/, we can replace the condition
M.h4=e / of the sumPon the right-hand side by M.he /. By these considerations, we
get the formula for .l;m;n/2M.he / .n/lmn. t
u
Several such sums with quadratic conditions have something to do with special
values of zeta functions of prehomogeneous vector spaces. But there are a lot of
mysteries around this in general. It has not been clarified when such formulas for
exponential sums can exist or what essential reasons there are for that. For some
special quadratic conditions, we have several conjectures but proofs are not known.
It seems premature to treat this problem here, so we omit the concrete description of
conjectures or other things, but it would be a very interesting problem to study these
sums from the viewpoint of relations between exponential sums and special values
of zeta functions of prehomogeneous vector spaces.
Exercise 8.24. Give an example of non-trivial primitive Dirichlet character  such
that 3 is trivial and for that , calculate the Gaussian sum g./ explicitly.
Exercise 8.25. (1) Assume that the cardinality of a finite group G is odd. Then
show that any character  such that 2 D 1 (the identity character) is a trivial
character.
(2) Assume that G is a finite cyclic group of even order. Then show that the non-
trivial character  of G such that 2 D 1 exists uniquely.
Exercise 8.26. (1) Let p be an odd prime. For any natural number r, it is known
that .Z=p r Z/ Š Z=p r1 Z  Z=.p  1/Z. By using this, show that if a non-
trivial Dirichlet character  modulo p r satisfies .a/2 D 1 for any a with
.a; p/ D 1, then the conductor of  is p.
136 8 Character Sums and Bernoulli Numbers

(2) Let r be a natural number with r  2. Itis known that .Z=2r Z/ Š Z=2Z 
Z=2r2 Z (or Š ..1 C 2Z/=.1 C 4Z//  .1 C 4Z/=.1 C 2r2 Z/ , the latter is
cyclic generated by 5.) Show that if a non-trivial Dirichlet character  modulo
2r satisfies .a/2 D 1 for any a with .a; 2/ D 1, then the conductor of  is
either 4 or 8.
(3) Show that if  is any non-trivial Dirichlet character  such that 2 is trivial, then
the conductor of  is jDK j for a fundamental discriminant of some quadratic
field K.
(4) Show that if  is any non-trivial primitive Dirichlet character  such that 2
is trivial, then  D K for some quadratic field K, where K is defined as in
Sect. 6.3.
Exercise 8.27. Let  be a primitive Dirichlet character with conductor f . Assume
that f D m1 m2 with .m1 ; m2 / D 1. We regard  as a character of .Z=f Z/ Š
.Z=m1 Z/  .Z=m2 Z/ and denote by i the restriction to Z=mi Z for i D 1, 2.
We also denote by i the Dirichlet character modulo mi associated with i .

 i is mi .
(1) Prove that the conductor of
(2) For any n, put n D exp 2n i . We take integers x, y with xm1 C y m2 D 1.
Show that for any a 2 Z, we have

fa D m
ay ax
 :
1 m2

(3) Show the following relation between Gaussian sums.

g./ D 1 .m2 /2 .m1 /g.1 /g.2 /:

(4) For each primitive Dirichlet character  with conductor 4 or 8 defined in

Sect. 4.1, calculate the Gaussian sum g./.
(5) Let K be p the Dirichlet character modulo
p DK defined in Sect. 6.3. Show that
g.K / D DK by using g.p/ D .1/.p1/=2 p for odd primes p and the
reciprocity law of the quadratic residue symbol.
Exercise 8.28. Prove Example 8.14 using Proposition 8.13.
Exercise 8.29. Show that the following relations hold.
(1) For any primitive character  modulo f , we have

f 1
X .a/ a 1
D  g./B2; :
aD1
.1   /
a 2 2
f 1
X .a/ a .1 C  a / 1
D  g./B3; :
aD1
.1   a /3 3
8.9 A Sum with Quadratic Conditions 137

(2) If we denote by .n/ D . pn / the quadratic residue character, we have

X
p1  
1 p.p  1/.7p C 1/
.m C n/ .m/mn D p 2 B1;2  :
m;nD0
2 12

(3) We have

X
p1
.lm C mn C nl/ D .1/p.p  1/:
l;m;nD0

X
p1
p 2 .p  1/
.lm C mn C nl/l D .1/ :
2
l;m;nD0

X
p1
p 2 .p  1/.2p  1/
.lm C mn C nl/l 2 D .1/ :
6
l;m;nD0

X
p1
p 3 .p  1/2
.lm C mn C nl/l 3 D .1/ :
4
l;m;nD0

X
p1
p 2 .p 2  1/
.lm C mn C nl/lm D .1/ :
6
l;m;nD0

X
p1
p 3 .p 2  1/
.lm C mn C nl/l 2 m D .1/ :
12
l;m;nD0

X
p1
3p 3 p 3 .p  1/
.lm C mn C nl/lmn D  .B1; /2 C .1/ :
2 4
l;m;nD0

Pp1
(Hint: nD0 .n.l C m/ C lm/ is 0 if l C m 6 0 mod p. In the last relation,
substitute l, m, n by p  l, p  m, p  n and add that to the original expression.)
Chapter 9
Special Values and Complex Integral
Representation of L-Functions

As a continuation of Chaps. 4 and 5, we study here properties of Hurwitz

zeta functions and Dirichlet L-functions such as their analytic continuation and
functional equation, and calculate their special values at negative integers. There
are various proofs for the functional equation; here we explain the method using a
contour integral. Although there would be a viewpoint that it would be too much to
introduce a contour integral, it is interesting for its own sake and useful too, so we
venture to derive the functional equation from a contour integral by a method to cut
out the path of the integral.

9.1 The Hurwitz Zeta Function

For a positive real number a, we define a zeta function .s; a/ by

1
X
.s; a/ D .n C a/s :
nD0

Since the series on the right-hand side converges absolutely for Re.s/ > 1, .s; a/
is defined in this range. This function .s; a/ is called a Hurwitz zeta function.
(Hurwitz [43], Whittaker1 and Watson2 [104, Part II]). In particular, when a D 1,
.s; 1/ is nothing but the Riemann zeta function .s/. In this section, we shall give
an integral representation of .s; a/.
We denote by .s/ the gamma function. This is defined by

1
Edmund Taylor Whittaker (born on October 24, 1873 in Southport, England—died on March 24,
1956 in Edinburgh, Scotland).
2
George Neville Watson (born on January 31, 1886 in Devon, England—died on February 2, 1965
in Warwickshire, England).

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 139
in Mathematics, DOI 10.1007/978-4-431-54919-2__9, © Springer Japan 2014
140 9 Special Values and Complex Integral Representation of L-Functions

Z 1
.s/ D e t t s1 dt (9.1)
0

for complex numbers s such that Re.s/ > 0. The integral on the right-hand side
converges absolutely and uniformly on any compact sets in Re.s/ > 0, and so we
see that .s/ is a holomorphic function for Re.s/ > 0. By integration by parts, we
get the most important property

.s C 1/ D s .s/ (9.2)

of the gamma function. Through this equality (9.2), .s/ can be continued
analytically to a meromorphic function on the whole s-plane. More concretely, if
we rewrite (9.2) as

.s C 1/
.s/ D ;
s

then since the right-hand side is a meromorphic function for Re.s/ > 1, .s/ is
continued analytically to Re.s/ > 1 and s D 0 is a pole of order 1 with residue 1.
We also have

.s C 2/
.s/ D ;
s.s C 1/

and repeating this, we get

.s C M /
.s/ D (9.3)
s.s C 1/    .s C M  1/

for arbitrary natural number M . Through (9.3), .s/ is continued analytically to

a meromorphic function on Re.s/ > M . Since M is arbitrary, .s/ is continued
analytically to a meromorphic function on the whole s-plane and s D 0; 1; 2; : : :
are poles of order 1, and the residues can be calculated by (9.3).
Changing the variable as t ! .n C a/t .n C a > 0/ in (9.1), we get
Z 1
1 1
D e .nCa/t t s1 dt:
.n C a/ s .s/ 0

Formal calculation leads us to

1
X Z 1 1
X
1 1
D e .nCa/t t s1 dt;
nD0
.n C a/s .s/ 0 nD0
9.2 Contour Integral 141

and for the range Re.s/ > 1, we see that the right-hand side converges absolutely, so
the above calculation is justified. Hence, for the range Re.s/ > 1, we get an integral
representation of .s; a/ as
Z 1 Z 1
1 e at s1 1 te .1a/t s2
.s; a/ D t dt D t dt: (9.4)
.s/ 0 1  e t .s/ 0 et  1

This integral representation suggests us that .s; a/ has something to do with

Bernoulli polynomial Bk .a/.

9.2 Contour Integral

We transform the integral representation (9.4) to a complex integral in order to

apply to it the residue theorem of complex analysis. We regard the interval Œ0; 1/
of integration as a path of a complex integral, and then expanding this a little, we
consider the following contour I."; 1/ (" > 0). We define I."; 1/ by a curve
' W .1; 1/ ! C given by
⎧
⎨ u  
u < ";
I."; 1/ W '.u/ D " exp  i uC" "  u  ";
⎩ "
u u > ":

In the definition of I."; 1/, the parts for u < " and u > " overlap, but we interpret
it that for u < " we take the path above the real axis and for u > " below the real
axis. This path is illustrated in Fig. 9.1.
Now consider the complex curvilinear integral
Z
te .1a/t s2
t dt:
I.";1/ et  1

Since we have to treat t s2 on I."; 1/, we shall choose a complex power t s for
s 2 C. Denoting the argument of t by arg t, we can define a single-valued function
log t on C  fz D x C iy j y D 0; x  0g by

log t D log jtj C i arg t .0 < arg t < 2/:

0 ε

Fig. 9.1 Path of I."; 1/

142 9 Special Values and Complex Integral Representation of L-Functions

Using this, a single-valued analytic function t s on Cfz D x C iy j y D 0; x  0g

is defined by

t s D e s log t :

We divide the contour I."; 1/ into three pieces as follows.

C1 : the part of the real axis from 1 to ",
I."/: the circle of radius " with center at the origin (oriented counter-clockwise),
C2 : the part of the real axis from " to 1.
We have I."; 1/ D C1 C I."/ C C2 . For t on C1 , we put arg t D 0, and on C2 we
put arg t D 2. Then the integral on C1 is given by
Z Z "
te .1a/t s2 te .1a/t s2
t dt D t dt
C1 et  1 1 et  1
Z 1
te .1a/t s2
D t dt
" et  1

and on C2 by
Z Z 1
te .1a/t s2 te .1a/t .s2/.log t C2 i /
t dt D e dt
C2 et  1 " et  1
Z 1 .1a/t
te
De 2 i.s2/
t s2 dt:
" et  1

Noting e 2 i.s2/ D e 2 i s , together we get

Z Z Z
t e .1a/t s2   1
t e .1a/t s2 e .1a/t s1
t dt D e 2 i s 1 t dt C t dt:
I.";1/ e 1
t
" e t 1 I."/ e t 1
(9.5)

The circle I."/ is parametrized as t D "e i .0    2/, so on I."/ we

have t s1 D "s1 e i.s1/ , and the absolute value of the integral is estimated from
above as
ˇZ ˇ Z 2 ˇˇ .1a/"ei ˇ
ˇ
ˇ e .1a/t ˇ ˇ e i ˇ
ˇ t s1
dt ˇ ˇ " s1 i.s1/
e i "e ˇ d
ˇ ˇ ˇ e "ei  1 ˇ
I."/ e  1
t
0

Z 2 ˇˇ .1a/"ei ˇ
ˇ
ˇe i s ˇ
D" Re.s/
ˇ "ei e ˇ d: (9.6)
0 ˇe 1 ˇ
9.2 Contour Integral 143

"
We see is bounded as a function of " and , so if we take the limit " ! 0
1
e "ei
then, for Re.s/ > 1, by (9.6) we get
Z
e .1a/t s1
lim t dt D 0: (9.7)
"!0 I."/ et  1

On the other hand, in (9.5), by Cauchy’s integral formula, the curvilinear integral
on the left-hand side does not depend on the choice of sufficiently small positive
number " (" < 2), for if we take another "0 > 0, then in the region surrounded by
two paths, the integrand is a single-valued holomorphic function. Now we assume
that Re.s/ > 1. In (9.5), since the left-hand side does not depend on " > 0, we fix
" on the left-hand side, and on the right-hand side we take a limit as " ! 0. Then
using (9.7), we get
Z Z
te .1a/t s2   1
te .1a/t s2
t dt D e 2 i s  1 t dt: (9.8)
I.";1/ e 1
t
0 et  1

Thus we obtain the following curvilinear integral representation of .s; a/.

Proposition 9.1. Let " be a positive number with " < 2. If Re.s/ > 1, then the
Hurwitz zeta function .s; a/ .a > 0/ has the contour integral representation
Z
1 te .1a/t s2
.s; a/ D t dt:
.s/.e 2 i s  1/ I.";1/ et  1

Through this formula, .s; a/ is continued analytically to a meromorphic function

on the whole s-plane and has a unique pole at s D 1 of order 1 with residue 1.
Proof. The first half is immediately obtained by the integral representation (9.4) and
(9.8) of .s; a/. We shall see the latter half below in turn. We prepare the following
lemma.
Z
te .1a/t s2
Lemma 9.2. The integral t dt converges absolutely for arbitrary
I.";1/ e  1
t

s 2 C and gives a holomorphic function on the whole s-plane.

Z
te .1a/t s2
Proof. The curvilinear integral t dt (j D 1; 2) converges absolutely
Cj e  1
t

for arbitrary s 2 C, so it is holomorphic there. Also the curvilinear integral

Z
te .1a/t s2
t dt converges absolutely for arbitrary s by (9.6) since " > 0
I."/ e  1
t

is fixed, and is differentiable with respect to s, so this is also holomorphic on the

whole s-plane. t
u
144 9 Special Values and Complex Integral Representation of L-Functions

1
As is well known in complex analysis, is a holomorphic function on the
.s/
whole s-plane, so by Proposition 9.1 and Lemma 9.2, poles of .s; a/ result from
zeros of the function e 2 i s  1. Since .s; a/ converges absolutely for Re.s/ > 1
and is holomorphic there, possible poles are s D 1 or s D 1  m (m 2 N). Among
these, at s D 1  m the function

1
.s/.e 2 i s  1/

is holomorphic. Hence .s; a/ has a unique pole s D 1 and its order is 1. The residue
is given by
Z Z
.s  1/ te .1a/t s2 1 te .1a/t 1
lim t dt D t dt
s!1 .s/.e 2 i s  1/ I.";1/ e t  1 2 i I.";1/ e t  1
Z  .1a/t 
1 te .1a/t 1 e
D t dt D Res D 1:
2 i I."/ e  1 t t D0 et  1

The above calculation of the last integral is due to the residue theorem. The only
e .1a/t
pole of the integrand t inside I."/ is t D 0 and it is of order 1, so it is enough
e 1
to calculate its residue.
We have finished the proof of the latter half of Proposition 9.1. t
u
Next, we calculate the special values of .s; a/ at non-positive integers s D 1m
(m 2 N) by using the contour integral representation.
Proposition 9.3. Let m be a natural number. Then for a > 0, we have

Bm .a/
.1  m; a/ D  :
m

Here Bm .a/ is the Bernoulli polynomial defined in Sect. 4.3 (p. 55).
Proof. By Proposition 9.1 we have
 Z
1 te .1a/t m1
.1  m; a/ D lim t dt:
s!1m .s/.e 2 i s  1/ I.";1/ et  1

te .1a/t m1
In the last curvilinear integral, the integrand t becomes a single-valued
et  1
holomorphic function on C n f0g. Notation being as before, in the integral for the
contour I."; 1/ D C1 C I."/ C C2 , the curvilinear integrals on C1 and C2 cancel
each other. That is,
9.3 The Functional Equation of .s; a/ 145

Z Z
te .1a/t m1 te .1a/t m1
t dt C t dt D 0:
C1 et  1 C2 et  1

The calculation of the curvilinear integral on I."/ is done by calculation of the

residues. By virtue of (5) and (4) of Proposition 4.9 on Bernoulli polynomials, the
.1a/t
residue of the integrand t eet 1 t m1 at t D 0 is given by
 
te .1a/t m1 Bm .1  a/ Bm .a/
Res t D D .1/m :
t D0 et  1 mŠ mŠ

On the other hand, the limit is calculated by (9.3) as follows.

1 s.s C 1/    .s C m  1/
lim D lim  
s!1m .s/.e 2 i s  1/ s!1m .s C m/ e 2 i.sCm1/  1

.1/.2/    .1  m/
D
2 i
.1/m1 .m  1/Š
D :
2 i
So we have
 
te .1a/t m1
.1  m; a/ D .1/ m1
.m  1/Š Res t :
t D0 et  1

By this, Proposition 9.3 is proved. t

u

9.3 The Functional Equation of .s; a/

We fix t0 < 0, and denote by C.t0 / the line on the t-plane defined by Re.t/ D t0
oriented upwards:

C.t0 / W t D t0 C iy .1 < y < 1/:

We assume Re.s/ < 0. In the contour integral representation of the Hurwitz zeta
function .s; a/ (Proposition 9.1), we cut out the contour I."; 1/ as in Fig. 9.2, and
shift it over the imaginary axis to the line C.t0 /. Here we are taking 0 < a  1.
When it goes beyond the imaginary axis, it passes the pole t D 2 i n(n 2 Z;
te .1a/t s2
n ¤ 0) of the integrand t t , so the residues appear there. The assumptions
e 1
Re.s/ < 0 and 0 < a  1 are conditions for the integral to converge. Under the
condition 0 < a  1, the curvilinear integral along C.t0 / converges absolutely.
Since we can transform as
146 9 Special Values and Complex Integral Representation of L-Functions

−C(t0)

=⇒
0 t0 0

Fig. 9.2 Shift of the contour

.s; a/

1 X e .1a/2 i n Z te .1a/t s2
D .2 i /  t dt
.s/.e 2 i s  1/ .2 i n/1s C.t0 / e  1
t
n2Z
n¤0
 1
1 X i 3 i
e 2 i na e 2 .s1/ C e 2 i na e 2 .s1/
D .2 i /
.s/.e 2 i s  1/ nD1
.2 n/1s
Z
te .1a/t s2
 t dt ;
C.t0 / et  1

taking the limit t0 ! 1, we have

 1 1

. i /.2/s1  i s X e 2 i na  is X e
2 i na
.s; a/ D e 2 e 2 : (9.9)
.s/ sin s nD1
n1s nD1
n1s

Here substituting 1  fag for a, we get

 1 1

. i /.2/s1  i s X e 2 i na  is X e
2 i na
.s; 1  fag/ D e 2 e 2 ; (9.10)
.s/ sin s nD1
n1s nD1
n1s
9.3 The Functional Equation of .s; a/ 147

where the notation fag denotes the fractional part of a: 0  fag < 1, a  fag 2 Z.
This transformation is valid for Re.s/ < 0. By these relations (9.9), (9.10), we get
1
is is .2/s X e 2 i na
e 2 .s; a/ C e  2 .s; 1  fag/ D ; (9.11)
.s/ nD1 n1s

and since .s; a/ is analytically continued to a meromorphic function on the whole

X1
e 2 i na
s-plane, is also continued analytically to a meromorphic function.
nD1
n1s
Summing up, we get the following theorem.
Theorem 9.4. We assume that 0 < a  1. The Hurwitz zeta function .s; a/ is
continued analytically to a meromorphic function on the whole s-plane. It has a
unique pole at s D 1. It is of order 1 and the residue is 1. Moreover, the following
functional equation is satisfied.
 1 1

. i /.2/s1  i s X e 2 i na  is X e
2 i na
.s; a/ D e 2 e 2 :
.s/ sin s nD1
n1s nD1
n1s

If we put s D 1  m here and use (4.3) on p. 59, we obtain Proposition 9.3.

Furthermore if we put a D 1 in (9.11), noting that .s; 1/ D .s/, we get

.2/s .2/s
.s/ D s  .1  s/ D  .1  s/:
2 .s/ cos 2 2 .s/ sin .sC1/
2

Then if we use the duplication formula of the gamma function

s  
1=2 s1 sC1
.s/ D  2
2 2

and the reflection formula


.s/ .1  s/ D
sin s
(for these formulas, see [104, Chapter 11]), we get
s  .2/s
.s/ D  1=2 21s  sC1  .1  s/
2 2 2 sin .sC1/ 2
 
1s
D  s1=2 .1  s/:
2
148 9 Special Values and Complex Integral Representation of L-Functions

Adjusting a little, this gives the usual functional equation of the Riemann zeta
function.
Theorem 9.5 (Functional equation of Riemann zeta function). The Riemann
zeta function .s/ can be continued analytically to a meromorphic function on the
whole s-plane. If we put
s 
.s/ D  s=2 .s/;
2
it satisfies the functional equation

.s/ D .1  s/:

9.4 Special Values of L-Functions and the Functional

Equations

In this section, our aim is to define a Dirichlet L-function and to look for its
functional equation and special values.
Let  be a Dirichlet character modulo f . The Dirichlet L-function associated
with the character  is defined by

X1
.n/
L.s; / D :
nD1
ns

The series on the right-hand side converges absolutely for Re.s/ > 1 and it is
holomorphic there.
First, we consider the special values at positive integers.
The Gaussian sum g./ associated with  is defined by

X
f
g./ D .a/e 2 i a=f
aD1

(Definition 8.1, p. 108). If we assume that  is a primitive character modulo f , then

by virtue of Lemma 8.2, we have

X
f
.n/g./ D .a/e 2 i an=f (9.12)
aD1

for arbitrary integer n.

9.4 Special Values of L-Functions and the Functional Equations 149

Theorem 9.6. Let  be a primitive character modulo f and let k be a natural

number. If .1/ D .1/k , then the special value of the Dirichlet L-function
L.s; / at s D k is given by

.1/k1 .2 i /k
L.k; / D g./Bk; :
2kŠf k

Example 9.7. We give examples first. Let  be the primitive character modulo 4 and
let k be an odd positive integer. Then we have  D  and g./ D 2i , so we have
kC1
.1/ 2  k
L.k; / D Bk; :
2k kŠ
We write two of them explicitly below. By the expression in (4.1) in Sect. 4.2, we
have B1; D  12 and B3; D 32 , so we get

1 1 1 
L.1; / D 1  C  C  D ;
3 5 7 4
1 1 1 3
L.3; / D 1  C  C    D :
33 53 73 32

Next, let  be the primitive character modulo 3. Since we have B1; D  13 , B3; D
p
3 , g./ D
2
3i , we get

1 1 1 1 1 
L.1; / D 1  C  C  C  D p ;
2 4 5 7 8 3 3
1 1 1 1 1 22  3
L.3; / D 1  C  C  C    D p :
23 43 53 73 83 34 3

Proof. We give a proof of the above theorem. Using (9.12), we have

X
f 1
X e 2 i an=f
L.k; /g./ D .a/ :
aD1 nD1
nk

Noting the assumption .1/ D .1/k , the right-hand side is transformed into

1X X e 2 i an=f
f
.a/ :
2 aD1 nk
n2Z
n¤0
150 9 Special Values and Complex Integral Representation of L-Functions

By virtue of Theorem 4.11 (p. 59) and (4.1) (p. 54), this value can be rewritten as

.2 i /k X
f
.2 i /k
.a/Bk .a=f / D Bk; :
2kŠ aD1 2kŠf k1

Noting g./1 D .1/g./=f by Lemmas 8.2 and 8.3 in the last section, we can
derive the theorem from this. t
u
We express L.s; / by using Hurwitz zeta functions. Hurwitz zeta functions
are very useful to deduce various properties of L-functions. Under the assumption
Re.s/ > 1, we can transform L.s; / as

Xf
X1
.a C f m/
L.s; / D
aD1 mD0
.a C f m/s

X
f 1
X 1
D .a/
aD1 mD0
.a C f m/s

X
f 1
X
s 1
Df .a/
aD1 mD0
.m C a=f /s

and the next proposition follows.

Proposition 9.8. Let  be a primitive Dirichlet character modulo f > 1. Then in
the range Re.s/ > 1, we have

X
f  
s a
L.s; / D f .a/ s; :
aD1
f

Through this expression, L.s; / is continued analytically to a holomorphic function

on the whole s-plane.
Proof. We have already shown the first half of the assertion. We use Proposition 9.1
for the latter half. By Proposition 9.1, the Hurwitz zeta function .s; a=f / is
continued analytically to a meromorphic function on the whole s-plane, its pole
being only situated at s D 1 and it is of order 1. Hence L.s; / is analytically
continued to a meromorphic function on the whole s-plane and it has at most one
pole at s D 1 of order 1. Calculating the residue at s D 1, we get

X
f
f 1 .a/ D 0
aD1

since f > 1 and  is non-trivial, so we see it is holomorphic at s D 1 too. t

u
9.4 Special Values of L-Functions and the Functional Equations 151

Let us derive the functional equation of L.s; / from the functional equation of
Hurwitz zeta functions (Theorem 9.4). We use the following notation ı D ı./,
defined by

0 if .1/ D 1;
ıD
1 if .1/ D 1:

So .1/ D .1/ı , ı 2 f0; 1g.

Theorem 9.9 (Functional equation). Let  be a primitive character modulo f .
Then L.s; / satisfies the following functional equation.
 s=2    .1s/=2  
f sCı f 1sCı
L.s; / D W ./ L.1  s; /;
 2  2

where we put

g./
W ./ D p :
f iı

Proof. By using Proposition 9.8 and Theorem 9.4, we continue L.s; / analytically
to the region Re.s/ < 0, and get

X
f  
s a
L.s; / D f .a/ s;
aD1
f
⎧ ⎫
. i /.2/s1 ⎨  i s X X .a/e 2 i n f 2 i n fa ⎬
1 f 1 f
 i s X X .a/e
a

D s e 2 e 2 :
f .s/ sin s ⎩ nD1 aD1
n1s
nD1 aD1
n1s ⎭

Here, using (9.12), we get

. i /.2/s1 g./   i s  i s


L.s; / D e 2  .1/e 2 L.1  s; /:
f s .s/ sin s

Simplifying the expression, we get

   
.s  ı/ g./ 2 s
.s/ cos L.s; / D L.1  s; /: (9.13)
2 2i ı f

By using the duplication formula of .s/ and the relation .s/ .1  s/ D ,
sin s
we can rewrite .s/ as
152 9 Special Values and Complex Integral Representation of L-Functions

s   
1=2 s1 sC1
.s/ D  2
2 2
 
sCı 
D  1=2 2s1  1sCı 
2 cos .sı/ 2 2

and then substituting this into (9.13) and simplifying the result, we get the desired
functional equation. t
u
Finally, let us look at special values of L.s; / at negative integers.
Theorem 9.10. Let  be a primitive character modulo f . For any positive integer
m, we have the following formula:

Bm;
L.1  m; / D  :
m
Proof. By Propositions 9.8 and 9.3, we have

X
f  
a
L.1  m; / D f m1
.a/ 1  m;
aD1
f

X
f
Bm .a=f /
D f m1
.a/ :
aD1
m

Bm;
By (4.1) on p. 54, this is equal to  . t
u
m
Remark 9.11. We note that Theorem 9.6 can be derived also from Theorem 9.10
and the functional equation (Theorem 9.9).
Exercise 9.12. Let a be any complex number. Put f .s/ D .s/.s  a/. Find a
functional equation of f .s/, that is, give a relation between f .s/ and f .b  s/ for a
suitably chosen number b.
Exercise 9.13. Let a be a positive integer. Determine the location of poles of the
function .s/.s  a/ and their residues for each case when a is even or odd.
Exercise 9.14. (1) Let  be the primitive character with conductor 4. Show that
L.1; / D 4 by using the integral
Z 1
1
dx
0 1 C x2

and Abel’s theorem of power series convergence.

(2) Let  be the primitive character with conductor 3. Show that L.1; / D 
p by
3 3
using the integral
9.4 Special Values of L-Functions and the Functional Equations 153

Z 1
1
dx:
0 1 C x C x2

Exercise 9.15. (1) Let  be any Dirichlet character modulo f . Show that for any
x 2 C with Re.s/ > 1, the infinite product
Y
.1  .p/p s /1
pWpri me;p−f

converges and equal to L.s; /.

(2) Show that L.s; / ¤ 0 for any s 2 C with Re.s/ > 1.
Chapter 10
Class Number Formula and an Easy
Zeta Function of the Space of Quadratic Forms

In this chapter, as an application of quadratic forms and quadratic fields, we

give an explicit formula of some simple zeta functions, related to some so-called
prehomogeneous vector spaces. We also prove a class number formula of imaginary
quadratic fields. Before that, we review the theory of multiplicative structure of
ideals of quadratic field without proof.

10.1 Ideal Class Groups of Quadratic Fields

We first review the prime ideal decomposition of the maximal order of quadratic
fields. Since there are many good books on prime ideal decompositions of the
Dedekind domain, we do not give proofs here. For details, we refer to Lang [67],
Weber1 [101]. Then we supply the theory of ideals of a quadratic order which is not
maximal. Since such orders are not Dedekind domains, the theory of ideals is not
covered by the above general theory, so we explain the difference.
We fix a quadratic field K of discriminant DK and denote by Omax the maximal
order of K. For any non-zero ideals a and b of Omax , we write
 
X
ab D ai bi I ai 2 a; bi 2 b;  is an arbitrary natural number
i D1

and call this a product of a and b. It is clear that this is also a non-zero ideal of
Omax . We give the following theorem without proof. We denote by K the Dirichlet
character associated with K defined in Sect. 6.3.

1
Heinrich Martin Weber (born on May 5, 1842 in Heidelberg, Germany—died on May 17, 1913 in
Strasbourg, Germany (now France)).

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 155
in Mathematics, DOI 10.1007/978-4-431-54919-2__10, © Springer Japan 2014
156 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

Theorem 10.1. (1) Any .non-zero/ ideal of the maximal order Omax of a quadratic
field K is a product of prime ideals. This factorization is unique. For any ideals
a and b of Omax , we have N.ab/ D N.a/N.b/.
(2) For any prime p 2 Z, the ideal pOmax is decomposed as follows.
(i) If K .p/ D 1, then pOmax D p1 p2 , where p1 and p2 are different prime
ideals related by p2 D p1 . Here we put

p1 D f˛I
N ˛ 2 p1 g

and x 7! xN is the non-trivial automorphism of K=Q.

(ii) If K .p/ D 1 then pOmax is a prime ideal.
(iii) If K .p/ D 0, namely if p is a prime factor of DK , then

pOmax D p2

for a prime ideal p.

There are no prime ideals of Omax other than those which appear in (1), (2) and (3)
above.
By this theorem, we have a relation between zeta functions as follows. For a
quadratic field K, we define the Dedekind zeta function of K by
X 1
K .s/ D :
a
N.a/s

Here the sum runs over all non-zero ideals of Omax and it converges absolutely for
Re.s/ > 1. The existence and uniqueness of the prime ideal factorization gives the
Euler product decomposition
Y 1
K .s/ D ;
p
1  N.p/s

again for Re.s/ > 1. Here the product runs over all prime ideals of Omax . Also the
decomposition of primes of Z in K (Theorem 10.1) gives

K .s/ D .s/L.s; K /:

We shall give a direct proof of this relation later.

We also give the following theorem on character, again without proof.
Proposition 10.2. Let K and DK be as before. Let  be a primitive Dirichlet
character modulo DK and we consider the following property on :

.N.a// D 1 for any ideal a of Omax which is coprime to DK :

10.1 Ideal Class Groups of Quadratic Fields 157

 
(1) The Dirichlet character K .n/ D DnK is a primitive Dirichlet character
modulo jDk j and satisfies the above property.
(2) Conversely, any non-trivial primitive Dirichlet character modulo jDK j which
satisfies the above property is equal to K .
Now in the rest of this section, we explain a multiplicative structure of ideals
including the case of non-maximal orders.
p f  1 and we consider ideals of an order Of of a quadratic
We fix an integer
field K D Q. m/. We extend the notion of ideals so that we can consider a
group of ideals. For this purpose, we introduce Of -modules, called fractional ideals,
which are not necessarily contained in Of . To define this, we first consider a free
Z-submodule L of the quadratic field K. The rank of L as a free Z-module is at
most 2. Indeed, if we assume that rankZ L  3, then there are elements !1 , !2 ,
!3 2 L which are the part of a basis over Z. If c1 !1 C c2 !2 C c3 !3 D 0 for some
ci 2 Q, then there is some non-zero r 2 Z such that rci 2 Z for all i , so we have
rci D 0 and hence ci D 0 for i D 1, 2, 3. Since L K and K is of dimension two
as a vector space over Q, this is a contradiction.
p
Definition 10.3. A free Z-submodule of rank 2 in the quadratic field K D Q. m/
is called a lattice of K.
For lattices L1 and L2 , we define their product by
 
X
L1 L2 D ai bi I ai 2 L1 ; bi 2 L2 ;  is an arbitrary natural number :
i D1

This product is a lattice.

p Indeed, for any element ˛ 2 K, there exists r 2 Z such
that r˛ 2 Z C Z m and p applying this to a basis of lattices L1pand L2 , we see
that rL1 , rL2 Z C Z pm for some integer r. But since Z C Z m is a ring, we
have r 2 L1 L2 Z C Z m. Hence by virtue of the structure theorem for finitely
generated abelian groups, L1 L2 is a free Z-module of rank 2 at most. It is obvious
that its rank is 2, because for any 0 ¤ ! 2 L1 , we have !L2 L1 L2 and !L2 is
of rank 2. This multiplication is commutative and associative.
Generalizing slightly the notion of a proper ideal, we call a lattice L of K a
fractional ideal of Of if L is an Of -module. Then we also have Of L D LOf D L,
since 1 2 Of . So Of is an identity element of the semi-group of fractional ideals
of Of . In this book, if we say just an ideal of Of , we always understand that it is a
subset of Of .
Now, in order to introduce a group structure on fractional ideals of an order Of ,
we need inverse elements. In other words, for an ideal a, we would like to find a
lattice L such that aL D Of . But such lattice does not exist in general.
p p
Example 10.4. Take K D Q.p 2/ and consider the order O5 D Z C 5 2Z of
conductor 5. Put a D 5Z C 5 2Z. Then this is an ideal of O5 , but there exists no
lattice L such that aL D O5 . The reason is as follows. If such an L exists, then p
since
5 2 a, we have 5L O5 . So any element ˛ of L is written as ˛ D a=5 C b 2
158 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

p p
(a, b 2 Z), but .5 C 5 2/˛ D .a  10b/ C .a C 5b/ 2 and since this should
p
belong to O5 , we have a C 5b 2p5Z, so a 2 5Z. So we get L Z C Z 2,
therefore we have aL 5.Z C Z 2/ ¤ O5 . Notice in this example that a is not
a proper ideal.
Lemma 10.5. Let a be an Of ideal. There exists the fractional ideal L of Of such
that aL D Of if and only if a is a proper Of ideal. If a is proper, L is uniquely
determined by a.
Proof. First assume that a is proper and we construct L such that aL D Of
explicitly. It is sufficient to give it when a is a primitive ideal. Let f1; !g be the
standard basis of the maximal order of K and express the ideal as a D Za C Z.d C
f !/ in the standard basis. We denote by x 7! x the non-trivial automorphism of
K=Q, and put M D Z C Za1 .d C f !/ D a1 a. This is obviously an Of module
and hence a fractional ideal of Of . Taking the product, we have

aM D Za C Z.d C f !/ C Z.d C f !/ C Za1 .d C f !/.d C f !/:

Here, because a is an ideal, c WD a1 .d C f !/.d C f !/ is an integer (p. 80,

Lemma 6.3). Moreover, by the assumption that a is a proper ideal, a, b WD 2d C
f T r.!/, c are mutually prime (p. 81, Lemma 6.5). But by the above expression we
have a, b, c 2 aM , so 1 2 aM . Therefore, by d C f ! 2 aM , we get f ! 2 aM ,
and we see Of aM . By the above expression, we have aM Of , so we have
aM D Of and we proved the first half. Conversely, let a be an Of ideal such that
there exists L with aL D Of . Then if ˛a a for ˛ 2 K, then ˛Of D ˛aL
aL D Of , so ˛ 2 Of . So a is a proper Of ideal. If aM D aN D Of for fractional
Of ideals M , N , then M D M Of D M.aN / D .M a/N D Of N D N , so the
uniqueness follows. t
u
We denote the Of -module M defined above by a1 and call it the inverse ideal
of a.
We call a fractional Of ideal L a proper fractional Of -ideal when

f 2 KI L Lg D Of :

For any proper ideal a of Of , a1 is also a proper fractional ideal. For any proper
fractional ideal L, we have rL Of for some integer r, and rL is obviously a
proper ideal of Of , so L has also an inverse M , which is a fractional Of ideal such
that LM D Of .
Lemma 10.6. The product L1 L2 of proper fractional ideals L1 and L2 of Of is a
proper fractional ideal.
Proof. It is sufficient to prove when L1 and L2 are (integral) ideals. Let a and b
be proper ideals and put ab D c. For ˛ 2 K, assume that ˛c c. Since a is a
proper ideal, there exists a1 and we have a1 c D a1 ab D b. Hence we have
˛b D a1 .˛c/ a1 c D b and since b is a proper ideal, we have ˛ 2 Of . t
u
10.1 Ideal Class Groups of Quadratic Fields 159

Summarizing the above, we have

Proposition 10.7. The set of all the proper fractional ideals of Of is a group by
the product defined above with the unit element Of .
The group defined in this way is called an ideal group of Of . The equivalence
of ideals is defined for fractional ideals in the same way as the usual ideals, namely
L
L0 if L D ˛L0 for some ˛ 2 K  , and the quotient group of fractional ideals
divided by this equivalence is called an ideal class group (in the wide sense). The
class number of Of is the cardinality of the ideal class group of Of .
Now we give a certain subset of ideals which, while not including all invertible
(i.e. proper) ideals, does include representatives for the whole ideal class group
of Of .
Definition 10.8. Let m be a natural number. We say that an ideal a of Of is coprime
to m if

a C mOf D Of :

Lemma 10.9. An ideal a of Of is coprime to m if and only if the norm of a is

coprime to m. Any ideal coprime to f is a proper ideal.
Proof. We may assume that a D lb for a primitive ideal b. By using the standard
basis, we write b D Za C Z.d C f !/. The module a C mOf is obviously an ideal
of Of , so a is coprime to m if and only if 1 2 a C mOf , or equivalently there exist
x, y, z, w 2 Z such that l.xa C y.d C f !// C m.z C wf !/ D 1. This implies that
lxa C lyd C mz D 1 and lyf C mwf D 0. If there exist such x, y, z, w, then by
the second equation, we see that ly is divisible by m and hence by the first equation
we see .m; la/ D 1. Since N.a/ D al 2 , we have .m; N.a// D 1. Conversely if
we assume that .m; la/ D 1, then there exist x, z 2 Z such that lax C mz D 1.
Since lax C mz 2 a C mOf , this ideal is equal to Of and we have proved the first
assertion. Since f Omax Of , for any ideal a of Of we have

a C f Of a C f Omax Of :

So if we assume that a is coprime to f , we have a C f Omax D Of . If ˛a a, then

˛ 2 Omax , so we have

˛Of D ˛a C f ˛Omax a C f Omax D Of :

Hence ˛ 2 Of and a is a proper ideal. t

u
of Of which is not coprime to f . For example a D
Therepexists a proper ideal p
25Z C 5 2Z O5 D Z C 5 2Z is not coprime to 5 but is a proper ideal of O5 .
In order to consider classes of proper Of ideals, it is enough to consider ideals
coprime to f by the following lemma.
160 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

Lemma 10.10. Let m be an arbitrary natural number. For any proper ideal a of
Of , there exists a proper ideal of Of which is equivalent to a and coprime to m.
Proof. It is enough to prove it when a is a proper primitive ideal. We write a
primitive quadratic form corresponding to the class of a by
 
a b=2
:
b=2 c

It is enough to see that we can make a prime to m by changing the quadratic form to
an equivalent quadratic form. Let S be the set of all primes which divide m and put

S1 D fp 2 S I p − ag;
S2 D fp 2 S I pja and p − cg;
S3 D fp 2 S I pja and pjcg:
Q Q
Put x D p2S2 p and y D p2S1 p. Then we have .x; y/ D 1, and if we put
l WD ax 2 C bxy C cy 2 , then l is coprime to m. Indeed, if p 2 S1 then ax 2 is
coprime to p, but pj.bxy C cy 2 /, so p − l, and if p 2 S2 then pj.ax 2 C bxy/ but
p − cy 2 , so p − l. If p 2 S3 , then since we assumed first that a, b, c are coprime, we
have p − b, and since p − xy, again l is coprime to p. Taking a matrix A 2 SL2 .Z/
whose first row is .x; y/, we have
   
a b=2 t l 
A AD ;
b=2 c 

so we are done. t
u
We denote by I0 .Of ; f / the set of ideals of Of which are coprime to f and by
I0 .Omax ; f / the set of ideals of Omax which are coprime to f .
Lemma 10.11. The mapping  W I0 .Omax ; f / ! I0 .Of ; f / defined by .a/ D
a \ Of is a bijection. This maps a product to a product and does not change the
norm.
Proof. First we show that this mapping does not change the norm. Let a be an ideal
of Omax coprime to f . By the isomorphism theorem of rings, we have Of =a \
Of Š .Of C a/=a. Since we assumed here that a is coprime to f , we have a C
f Omax D Omax , but since f Omax Of , we also have Of C a D Omax . Namely
we get

jOf =a \ Of j D jOmax =aj;

and this means that the norms are equal. In particular, we have seen that the
image of the mapping is contained in I0 .Of ; f / since the norm is coprime to f .
10.1 Ideal Class Groups of Quadratic Fields 161

The surjectivity of the mapping is proved as follows. For a0 2 I0 .Of ; f /, the Omax
ideal a0 Omax is coprime to f , since 1 2 Of D a0 C f Of a0 Omax C f Omax .
We have

a0 a0 Omax \ Of D .a0 Omax \ Of /.a0 C f Of /

a0 C a0 f Omax a0 C a0 Of D a0 :

So we have .a0 Omax / D a0 . Injectivity is proved as follows. If we write .a/ D a0

for a fixed a 2 I.f; Omax /, then as we have shown above, we have .a0 Omax / D a0 .
We also have N.a0 Omax / D N.a0 / D N.a/. Since a0 Omax .a \ Of /Omax a
and N.a/ D N.a0 Omax /, we have a D a0 Omax . This means that a is determined by
a0 D .a/. So the mapping is bijective. Now take ideals a, b 2 I0 .Omax ; f /. We put
a0 D .a/ D a \ Of and b0 D .b/ D b \ Of . We will show that .ab/ D a0 b0 .
First we show that a0 b0 is coprime to f . Since a0 is coprime to f as shown before,
we have a0 C f Of D Of , so a0 b0 C f b0 D b0 . Since b0 is also coprime to f ,
we have a0 b0 C f b0 C f Of D b0 C f Of D Of . Since f b0 f Of , we have
a0 b0 C f b C f Of D a0 b0 C f Of . So we have a0 b0 C f Of D Of . So a0 b0 is
coprime to f . Now we put c0 D a0 b0 . We will show that c0 D ab \ Of . We have
already shown that a D a0 Omax , b D b0 Omax and c0 Omax \ Of D c0 . We have

a0 b0 Omax D a0 Omax b0 Omax D ab;

so for ab 2 I0 .f; Omax /, we have .ab/ D a0 b0 . Therefore a product is mapped to

a product. t
u
So in order to consider the classes of proper Of ideals, we can lift proper Of
ideals prime to f in each class to ideals of Omax and consider these. But this does
not mean that the classes of Omax ideals and of Of ideals are the same. In fact, for
a principal ideal ˛Omax , the ideal Of \ ˛Omax is not necessarily a principal ideal
of Of and this causes the difference. Also we should note the following point.
We denote by I.Of ; f / the group of fractional proper Of ideals generated by
I0 .Of ; f /. Then a principal ideal ˛Of .˛ 2 K/ is contained in I.Of ; f / if and
only if we can write ˛ D ˇ= by numbers ˇ,  2 Of which are prime to f .
Here, we may take  as a rational integer prime to f since ˛ D ˇ =N. /. We
denote the set of such principal ideals by P .Of ; f /. In the case when ˛ 2 K is
not in Of , even if (the numerator and the denominator of) N.˛/ is prime to f , we
cannot conclude that ˛Of is in P .Of ; f /. Now, obviously an element ˇ 2 Of
is prime to f if and only if ˇ  a 2 f Omax for some integer a such that a is
prime to f . So if we denote by PZ .Omax ; f / the group generated by principal
ideals ˇOmax , where ˇ 2 Omax and ˇ  a 2 f Omax for some integer a prime
to f , then P .Of ; f / is lifted to this group, and we see that the ideal class group
of Of is isomorphic to I.Omax ; f /=PZ .Omax ; f /. If we denote by P .Omax ; f /
the group generated by integral principal Omax ideals prime to f , then by counting
P .Omax ; f /=PZ .Omax ; f /, we can calculate the difference between h.f 2 DK / and
162 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

h.DK /. This calculation is not so difficult and is explained for example in [101] or
[67], but we omit it here. In next section, we use a different method to obtain this
difference in the case of imaginary quadratic fields.

10.2 Proof of the Class Number Formula of Imaginary

Quadratic Fields

We prove that the class number of an imaginary quadratic field can be written in
terms of a generalized Bernoulli number. There are many ways to prove this; here
we use properties of theta functions.
Theorem 10.12. Let K be an imaginary quadratic field, K be the character
corresponding with K defined in Sect. 6.3 and w be the half of the number of roots
of unity in the maximal order Omax of K. Then the class number of OK is given by

h.DK / D wB1;K :

For the proof, we review the transformation formula of theta functions. Let a, b,
c be integers, and we assume that the quadratic form Q.x; y/ D ax 2 Cbxy Ccy 2 is
positive definite. We define Q1 by Q1 .x; y/ D 4.4ac b 2 /1 .cx 2 bxy Cay 2 /.
We put

H D f 2 CI the imaginary part of  is positiveg

(the upper half plane). For  2 H, we write

X
Q ./ D e  iQ.x;y/ :
x;y2Z

It is easy to see that this converges absolutely and uniformly on compact sets in H,
and is a holomorphic function on H.
Proposition 10.13. For  2 H, the following transformation formula holds.
  p
1
Q1  D 4ac  b 2 .=2i / Q ./:


Proof. A general proof can be found for example in [3, p. 25]. Here we give a direct
proof restricted to this case. We use the Poisson2 summation formula which is well

2
Siméon Denis Poisson (born on June 21, 1781 in Pithiviers, France—died on April 25, 1840 in
Sceaux, France).
10.2 Proof of the Class Number Formula of Imaginary Quadratic Fields 163

known in the theory of Fourier3 transforms. To simplify the notation, for arbitrary
complex number z 2 C, we write e.z/ D e 2 i z . For ˛ > 0,  2 H and ,   2 R, the
following formula is well known:
Z 1
e..2˛/1   2 /
e.21 ˛ 2 /e.  /d  D p :
1 i˛

Here as thepsquare root of i˛ (whose real part is positive since  2 H), we take
the branch ˛ > 0 for  D i . When we put

f .; / D e.21 Q.; //;

the Fourier transform of this function with respect to the variables .; / 2 R2 is
calculated as follows:

fO.  ;  /
Z 1Z 1
D f .; /e.    /d  d
1 1
Z 1 Z 1
D e.21 .a. C .b=2a//2 C .c  b 2 =4a/2 //e.    /d  d
1 1
Z 1 Z 1
D e.21 .a 2 C .c  b 2 =4a/2 //e.   .  .b=2a/  //d  d
1 1
1
D p p
ia i.c  b 2 =4a/
e..2a/1   2 /e..2.c  b 2 =4a//1 .  .b=2a/  /2 /
 
2i 1 2   2
D p e .a  b  C c / :
 .4ac  b 2 / 2.ac  b 2 =4/

By this result and the Poisson summation formula (e.g. [102]), we get
X X
f .; / D fO.  ;  /:
;2Z   ; 2Z

Hence we get the transformation formula of theta functions. t

u
Proof of the Class Number Formula. We show this by calculating the residue
of a zeta function in two different ways. Namely, we can derive the formula by the
relation K .s/ D .s/L.s; K / and the following calculation:

3
Jean Baptiste Joseph Fourier (born on March 21, 1768 in Auxerre, France—died on May 16, 1830
in Paris, France).
164 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

h.DK /
lim .s  1/K .s/ D p
s!1 w jDK j

and

g./ i B1; 
L.1; K / D B1;K D  p K ;
jDK j jDK j
lim .s  1/.s/ D 1:
s!1

The latter two equalities are in Theorems 5.4 and 9.6. The point here is the
evaluation of lims!1 .s 1/K .s/. We fix an ideal class of Omax and denote by C the
set of all (integral) ideals of Omax belonging to this class. We define the following
zeta function which is a part of K .s/:
X 1
.s; C / D :
a 2C
N.a/s

Lemma 10.14. Independently of the choice of C , we have


lim .s  1/.s; C / D p :
s!1 w jDK j

Proof. If we take a fractional ideal L which is equivalent to some ideal in C , then

all the other ideals in C are written as L˛(˛ 2 K). Since L˛ Omax , we have
˛ 2 L1 . We can choose L so that L1 D a Omax by dividing L by some
rational number if necessary. Here we can assume that a is a primitive ideal without
loss of generality. Also we have L˛ D Lˇ if and only if ˛ D
ˇ for some unit
of
Omax . Since we assumed that K is imaginary,
is a root of unity in K. Hence we
have

N.a/s X 1
.s; C / D :
2w ˛2a N.˛/s

We write a D Za C Z.d C !/ using the standard basis and put N.d C !/ D ac,
b D 2d C T r.!/ as usual. Then for ˛ D xa C y.d C !/ we have N.˛/ D
a.ax 2 C bxy C cy 2 /. Since N.a/ D a, we can cancel as in the numerator and the
denominator. So we have
1 X
.s; C / D .ax 2 C bxy C cy 2 /s :
2w
x;y2ZI .x;y/¤.0;0/

We have also 4ac  b 2 D jDK j. Now our aim is to calculate the residue of this
function at s D 1. We use theta functions. Since .s; C / converges absolutely
10.2 Proof of the Class Number Formula of Imaginary Quadratic Fields 165

uniformly on compact subsets of fs 2 CI Re.s/ > 1g, exchanging the summation

and the integral, and also using the following well-known formula of the gamma
function
Z 1  s Z 1
.s/ t t dt
D e D e nt t s1 dt;
ns 0 n t 0

we get
Z 1 X
1
e .a
2 CbCc2 /y
.s/.s; C / D y s1 dy
2w 0
.;/2Z2 .0;0/
s Z 1

D .Q .iy/  1/y s1 dy:
2w 0

Here Q .iy/  1 is a rapidly decreasing function for y ! 1, so the integral

Z 1
.Q .iy/  1/y s1 dy
1

converges for all s 2 C and is a holomorphic function on the whole complex

s-plane. On the other hand, if we take .0; 1/ as the interval of integration, it might
not converge around 0 when s is small. In order to deal with this problem, we use
the transformation formula of theta functions. Namely, if we substitute y 1 for y in
the integral on the interval .0; 1/, in the range Re.s/ > 1 we have
Z 1
.Q .iy/  1/y s1 dy
0
Z 1
D .Q .iy 1 /  1/y s1 dy
1
Z 1 p
D .Q1 .iy/2y jDK j1  1/y s1 dy
1
Z 1 p p
D 2.Q1 .iy/  1/y s jDK j1 C 2y s jDK j1  y s1 dy
1
Z 1
y s 2 1
D2 .Q1 .iy/  1/ p dy  p  :
1 jDK j jDK j.1  s/ s

The integral in the last expression converges for arbitrary s, so it is holomorphic

on the whole complex plane, and the rest has a pole at s D 0 and s D 1. So the
residue of .s; C / at s D 1 is given by
1 1 2 
p D p :
.1/ 2w DK w jDK j

Hence the lemma is proved.

166 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

Therefore we have

h.DK /
lim .s  1/K .s/ D p :
s!1 w jDK j

Comparing this with the value of L.1; K /, we get the class number formula. t
u
Since .s/ and L.s; / have functional equations, K .s/, which is a product of
these two, also has a functional equation and it is easy to calculate its explicit form.
But as a matter of fact, if we use the above integral representation, we can obtain the
analytic continuation and the functional equation of .s; C /, and through this we
can get an alternative proof of the analytic continuation and the functional equation
of K .s/. Namely:
Proposition 10.15. Let C be an ideal class of the maximal order Omax of a
imaginary quadratic field K, and let the other notation be the same as above.
Then .s; C / can be continued analytically to a meromorphic function on the
whole s-plane. If we put .s; C / D .2/s jDK js=2 .s/.s; C /, then it satisfies
the following functional equation.

.1  s; C / D .s; C /:

In the same way, if we put K .s/ D .2/s jDK js=2 .s/K .s/, then we have

K .1  s/ D K .s/:

Proof. For the proof, we use the integral representation mentioned before. For
Re.s/ > 1, we have
Z 1 Z 1
2
.2w/ s .s/.s; C / D .Q .iy/  1/y s1 dy C p .Q1 .iy/  1/y s dy
1 jDK j 1

2 1
p  (10.1)
jDK j.1  s/ s

but the integral on the right-hand side is holomorphic on the whole s-plane. Hence
the right-hand side is a meromorphic function on the whole complex s-plane. So the
analytic continuation is proved.
The functional equation is shown as follows. Multiplying both sides of the
integral representation before by 2s jDK js=2 and changing s ! 1  s, we get

.2w/.1  s; C /
D .2w/jDK j.1s/=2 .2/s1 .1  s/.1  s; C /
Z 1
D 2s1 jDK j.1s/=2 .Q .iy/  1/y s dy
1
10.2 Proof of the Class Number Formula of Imaginary Quadratic Fields 167

Z 1
1=2
C 2 jDK j
s
jDK j .1s/=2
.Q1 .iy/  1/y s1 dy
1

2s jDK j.1s/=2 jDK j1=2 2s1 jDK j.1s/=2

 
s 1s
 Z 1
D 2s jDK js=2 21 jDK j1=2 .Q .iy/  1/y s dy
1
Z 1 
1 21 jDK j1=2
C .Q1 .iy/  1/y s1
dy   : (10.2)
1 s 1s

Here the integrals in the parentheses are almost identical to the integral expres-
sion (10.1) of .s; C / if we exchange Q and Q1 . We defined

4
Q1 .x; y/ D .cx 2  bxy C ay 2 /;
jDK j

and the quadratic forms ax 2 C bxy C cy 2 and cx 2  bxy C ay 2 are equivalent since
it is the same if we replace .x; y/ ! .y; x/. Since we have

.Q1 .x; y//s D 22s jDK js .cx 2  bxy C ay 2 /s

we see in the same way as before that

Z 1
.2w/ .s/.s; C / D 22s  s jDK js .Q1 .iy/  1/y s1 dy:
0

Just in the same way as in the proof of (10.1), we have

Z 1 Z 1
.Q1 .iy/  1/y s1 dy D .Q1 .iy/  1/y s1 dy
0 1
Z p
1
p 1
21 jDK j 1
s
C2 jDK j .Q .iy/  1/y dy   :
1 1s s

(If we note that the discriminant of Q1 is 16jDK j1 , we see the symmetry to the
case Q.) Therefore comparing this with (10.2), we have
Z 1
.2w/.1  s; C / D 2s jDK js=2 .Q1 .iy/  1/y s1 dy
0
Z 1
D .2/s jDK js=2 22s jDK js  s .Q1 .iy/  1/y s1 dy
0

D .2w/.2/s jDK js=2 .s/.s; C / D .2w/.s; C /:

168 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

Hence the proposition is proved. The functional equation of K .s/ is obvious from
this. t
u
As for the formula for the class number h.f 2 DK / of the ideal class group of a
general order Of , it is easiest to compare it with the ideal class group of Omax by
a group theoretic method. (For example, see Weber [101, p. 366], Lang [67, p. 95],
Zagier [106].) But here we introduce a different method.
We first give the formula for the zeta functions of Of ideals for general f  1.
Here we do not assume that K is imaginary, since the theory is the same for real
quadratic fields.
We fix a natural number f and consider the zeta function of ideals of the order
Of D Z C Zf ! of K, which is defined by

X 1
.s; f 2 DK / D ;
a
N.a/s

where the sum runs over arbitrary ideals of Of , which are not necessarily proper.
Proposition 10.16. (1) For a positive integer a, let r.a/ be a number of residue
classes modulo 2a represented by integers b such that b 2  f 2 DK mod 4a.
Then we have
1
X
.s; f 2 DK / D .2s/ r.a/as :
aD1

(2) More precisely, we have the next relation.

Y 1p .mC1/.12s/ K .p/p s .1  p m.12s/ /

.s; f 2 DK / D .s/L.s; K / :
pjf
1  p 12s
p m kf

Here we denote by m the highest power of p dividing f (hence m depends

on p).
Proof. A primitive ideal of Of is written as a D Za C Z.d C f !/ (a D N.a/)
and there exists an integer c such that N.d C f !/ D ac, so if we put b D 2d C
f T r.!/ then b 2  f 2 DK mod 4a (p. 79, Lemma 6.2 and p. 80, Lemma 6.3). For
positive integers a, a0 , we have Za C Z.d C f !/ D Za0 C Z.d 0 C f !/ if and
only if a D a0 and d  d 0 mod a. So if we fix a, two primitive ideals which have
the above bases are equal if and only if b  b 0 mod 2a for b D 2d C f T r.!/
and b 0 D 2d 0 C f T r.!/. Besides, if b 2  f 2 DK mod 4a for some integer b,
then since T r.!/2  DK D 4N.!/, we have b 2  .f T r.!//2 mod 4 and there
exists an integer d such that b D 2d C f T r.!/. So we can define a primitive
ideal Za C .d C f !/Z. Taking ideals which are not primitive into account, (1) is
proved. Next we show (2). For this purpose, we should just calculate r.a/. Here
we introduce a local version of r.a/. For a positive integer a and a fixed prime p,
10.2 Proof of the Class Number Formula of Imaginary Quadratic Fields 169

we denote by e the maximum non-negative integer such that p e ja. If p is odd, we

define rp .a/ to be the number of integers b mod p e such that b 2  f 2 DK mod p e .
If p D 2, we define rp .a/ D r2 .a/ to be the number of integers b mod 2eC1 such
that b 2  f 2 DK mod 2eC2 . ObviouslyQwe have rp .a/ D rp .p eP /. By the Chinese
remainder theorem, we have r.a/ D pja rp .a/. So we have 1 aD1 r.a/a
s
D
Q P1 e es
p r
eD0 p .p /p and it is enough to calculate the above local Dirichlet series
for each prime p. First we assume that p is an odd prime. Let f D p m f0 , where
f0 is a positive integer prime to p. We consider the congruence equation b 2 
p 2m f02 DK mod p e with unknown b. If e  2m then b 2  0 mod p e . Hence if
e D 2e0 here, then b  0 mod p e0 so rp .p e / D p e =p e0 D p e0 . If e D 2e0 C 1,
then b  0 mod p e0 C1 , and rp .p e / D p e =p e0 C1 D p e0 . So we have rp .p e / D p e0
in both cases. Now we assume that 2m < e. Then we have b D p m b0 and b02 
DK f02 mod p e2m . If pjDK then p divides DK precisely once, so if e  2m  2,
there is no such b0 . Hence it can have solutions only when e  2m D 1, and in this
case, the congruence implies nothing but pjb0 , so p mC1 jb is the condition. But we
are counting b mod p 2mC1 , so we have rp .p 2mC1 / D p m , and we have rp .p e / D 0
for e > 2m C 1. If p − DK , then the congruence equation b02  DK f02 mod p e2m
has solutions only when K .p/ D 1 and the number of solutions mod p e2m is two
in that case. Since we must count these as residues mod p em , the number is 2p m .
After all, if p − DK and e > 2m, we have rp .p e / D p m .1 CK .p//, independently
of e. Summing up, we have the following formula for rp .p e /.
⎧
⎪
⎪p
Œe=2
0  e  2m;
⎪
⎪
⎨.1 C  .p//p m 2m C 1  e and p − DK ,
K
rp .p e / D (10.3)
⎪
⎪ p m
e D 2m C 1 and pjDK ,
⎪
⎪
⎩
0 2m C 2  e and pjDK .

Here Œe=2 denotes the biggest integer which does not exceed e=2, that is, we put
Œe=2 D e0 if e D 2e0 or 2e0 C 1 for some integer e0 . Also when p D 2, we
obtain the same formula (10.3) by similar consideration. For example, assume that
0  e  2m. If e D 2e0 is even, then b 2  22m f02 DK mod 2eC2 means that
b D 2e0 b0 and b02  f02 DK mod 4. This last relation imposes the condition that
b0 is even or odd. So the number of solutions b mod 2eC1 is 2e0 C1 =2 D 2e0 . If
e D 2e0 C 1 is odd, then 2e0 C 2  2m, so b D 2e0 C1 b0 . Again the relation b02 
f02 DK mod 2 tells that b0 is odd or even. So the number of solutions b mod 22e0 C2
is 22e0 C2.e0 C1/ =2 D 2e0 . The case e  2m C 1 is calculated similarly. Here we
note that if DK is odd, then K .2/ D 1 if DK  1 mod 8 and K .2/ D 1 if
DK  5 mod 8, and that the congruence equation x 2  y mod 2lC2 with l  1
for a fixed y and unknown x has no solution when y  5 mod 8 but two solutions
x mod 2lC1 when y  1 mod 8. We should also note that, by definition, if DK is
even then DK  0 mod 4, and if DK  4 mod 8 then DK =4  3 mod 4. The rest
is the same and is omitted here. By using (10.3), we see
170 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

1
X .1 C p s /  .1 C p s1 /p .12s/.mC1/
rp .p e /p es D
eD0
1  p 12s
⎧
⎪
⎨p m.12s/s if pjDK ,
C .1 C K .p//p m.12s/s
⎪
⎩ if p − DK .
1  p s

The last term written for each case pjDK or p − DK separately has a unified
expression

.1 C K .p//p m.12s/s
:
1  K .p/p s

Reducing to a common denominator, the right-hand side becomes

.1 C p s /.1  p .mC1/.12s/  K .p/p s .1  p m.12s/ //

:
.1  K .p/p s /.1  p 12s /

Hence multiplying this by the Euler p-factor of .2s/, we get

1
X 1  p .mC1/.12s/  K .p/p s .1  p m.12s/ /
.1  p 2s /1 r.p e /p es D :
eD0
.1  p s /.1  K .p/p s /.1  p 12s /

t
u
Next we put
X 1
.s; Of / D :
a
N.a/s

But this time, we define the above sum so that a runs only over proper ideals of Of .

Here we introduce the Möbius4 function .m/. This is a function of natural

numbers to f0; ˙1g, defined as follows:
(1) .1/ D 1.
(2) When the prime factor decomposition is given by m D p1    pr with different
r primes, we define .m/ D .1/r .
(3) When m has a square factor, that is, if m is divisible by a square of a prime, we
define .m/ D 0.

4
August Ferdinand Möbius (born on November 17, 1790 in Schulpforta, Saxony (now Germany)—
died on September 26, 1868 in Leipzig, Germany).
10.2 Proof of the Class Number Formula of Imaginary Quadratic Fields 171

If m1 and m2 are coprime natural numbers, then we have .m1 /.m2 / D .m1 m2 /.
The next formula is known as the Möbius inversion formula.
Lemma 10.17. Let F .n/ and G.n/ be functions defined for all natural numbers n.
We assume that, for an arbitrary natural number n, the following equality holds.
X
F .n/ D G.d /:
d jn

Then for an arbitrary natural number n, we have

X n
G.n/ D  F .d /:
d
d jn

The proof is easy and is omitted here (see [83]).

Proposition 10.18. (1) Using the Möbius function , we have
X
.s; f 2 DK / D f s e s .s; Oe /;
ejf
X
.s; Of / D f s .f =d /d s .s; d 2 DK /:
d jf

(2) The next relation holds.

.s; Of / D .s/L.s; K /
Y .1  p s /.1  K .p/p s /  p m12sm .1  p 1s /.K .p/  p 1s /
 :
pjf
1  p 12s
p m kf

Proof. The first equality of (1) comes from the facts that any ideal of Of is a
proper ideal of Oe for some ejf and that if the ideal is regarded as an ideal of
Oe , then the norm becomes f =e times the original one. The second equality follows
from the Möbius inversion formula, applying it to F .f / D f s .s; f 2 DK / and
G.e/ D e s .s; Oe /. In order to show (2), we calculate the right-hand side of
the second equality of (1) by using Proposition 10.16. Since the Möbius function
Qtmultiplicative,
is it is enough to calculate the Euler p-factors. Here when f D
mi
p
i D1 i , by the definition of the Möbius function, we can assume that d runs only
Q m j
over the numbers pi i i with ji D 0; 1. So if we write

1  p .mC1/.12s/  K .p/p s .1  p m.12s/ /

f .m; p s / D :
1  p 12s
172 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

then we have
Y
.s; Of / D .s/L.s; K / .f .m; p s /  p s f .m  1; p s // :
pjf

Calculating the latter product, we get the result. t

u
Remark 10.19. The explicit expression of the zeta function in Propositions 10.16
and 10.18 was given in [54, 106]. (See these papers for applications.)
Now, when K is an imaginary quadratic field, we obtain the class number of Of
from the above formula for .s; Of /. As before, we denote by C an ideal class of
proper ideals of Of and write

X 1
.s; C; Of / D :
a 2C
N.a/s

Let wf be half the number of roots of unity in Of . For a primitive quadratic form
Q.x; y/ D ax 2 Cbxy Ccy 2 with discriminant f 2 DK corresponding to C , we have
X
.2wf /.s; C; Of / D .ax 2 C bxy C cy 2 /s :
.x;y/2Z2 .0;0/

This is proved in the same way as in the case of the maximal order. If we use the
transformation formula of Q ./, the residue at s D 1 of this Dirichlet series is
obtained in the same way as before, independently of C , and given by

lim .s  1/.s; C; Of / D p :
s!1 wf f jDK j

By this, we have

lim .s  1/.s; Of / D h.f 2 DK / p :
s!1 wf f jDK j

On the other hand, using the Euler product expression in the formula for .s; Of /
in Proposition 10.18, we have
Y
lim .s  1/.s; Of / D .lim .s  1/K .s//  .1  K .p/p 1 /
s!1 s!1
pjf

h.DK / Y
D p .1  K .p/p 1 /:
w jDK j pjf
10.3 Some L-Functions Associated with Quadratic Forms 173

Hence we get the class number formula

h.DK / Y
h.f 2 DK / D f .1  K .p/p 1 /
w=wf
pjf

as was given in Chap. 6, Theorem 6.12 (2).

The class number formula for a real quadratic field K (or for indefinite quadratic
forms) is fairly complicated compared with this, and it is remote from our standpoint
from Bernoulli numbers, so we omit it here.

10.3 Some L-Functions Associated with Quadratic Forms

We introduce here one of the simplest example of zeta functions associated with so-
called prehomogeneous vector spaces. The content of this section is based on [47].
  that p  3 mod 4. We define a Dirichlet character
Let p be a prime. We assume
modulo p by .a/ D pa . From this, we define a function Q on the set L of
all 2  2 half-integral symmetric matrices as follows. First, for
 
a b=2
T D 2 L ;
b=2 c

if T mod p D 0 or det.T / 6 0 mod p, we define Q .T / D 0. If T mod p ¤ 0 and

det.T /  0 mod p, then there exists g 2 GL2 .Z=pZ/ such that
 
e0
t
gT g  mod p:
00

Indeed since the rank of T mod p is 1, there exists g such that the second column
of T g mod p is 0, and so is the second column of t gT g, but since this is symmetric,
the .2; 1/ component is also 0. We define .T / D .e/ for such T . This is well
defined since if we assume that
     0 
xy e0 x z e 0
D
zw 00 yw 0 0
 
0 xy
for e, e 6 0 mod p and 2 GL2 .Z=pZ/, then we have e 0 D x 2 e (and z D 0,
z w
x 2 .Z=pZ/ ).
We denote by LC the set of all positive definite half-integral symmetric matrices
in L . Recall that T1 and T2 2 L are said to be equivalent if we have t gT1 g D T2
for some g 2 SL2 .Z/. We define an L-function of LC with character Q by
174 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

X Q .T /
L.s; LC ; / D :

.T / det.T /s
T 2L
C =

Here LC =
denotes a complete set of representatives of LC up to the equivalence
and
.T / the order of the finite group fg 2 SL2 .Z/I t gT g D T g. (This L-function
was defined first by K. Hashimoto in relation to dimension formulas of automorphic
forms. This is also an example of L-functions of prehomogeneous vector spaces.
As we see later, Q can be regarded as a character of some ideal class group, so the
term “L-function with character Q ” is not strange.)
 
a b=2
Let us try to calculate an example. If T D 2 L and rank.T mod
b=2 c
p/ D 1, then Q .T / D .a/ if p − a and Q .T / D .c/ if pja. Indeed if pja, then
by the condition on the rank, we have pjb and c 2 .Z=pZ/ , and if p − a, then we
have
    !  
1 0 a b=2 1 b=2a a 0 a0
D b2  mod p:
b=2a 1 b=2 c 0 1 0 c  4a 00

In the case p D 3, for det.2T / D 3, 12, 15, 24, 27, . . . , we have

       
Q 1 1=2 Q 10 Q 21 Q 10
D 1; D 1; D 1; D 1;
1=2 1 03 12 06
     
Q 2 0 D 1; Q 1 1=2
D 1; Q 3 3=2
D 0; : : :
03 1=2 7 3=2 3
 
a a=2
We also have
.T / D 6 for T D (a D 1, 2, 3 . . . ) and
.T / D 2 for the
a=2 a
remaining T with rank.T mod 3/ D 1. From this we get
   
22s 1 1 1 1 1 1 22s
L.s; LC ; /D C s  C s  C C
63 s 3 2 6 6 2 2 2  33s
22s1 1 22s1
D C C C :
3sC1 3sC1 33s
Theorem 10.20. This L-function is essentially the Riemann zeta function. More
precisely we have

22s1 B1;
L.s; LC ; / D  .2s  1/:
ps
10.3 Some L-Functions Associated with Quadratic Forms 175

In particular, for an arbitrary natural number m, we have

p m1
L.1  m; LC ; / D B2m B1; :
22m  m

Before proving this, we prove a lemma. For an imaginary quadratic field K, its
discriminant DK and a positive integer f , we put d D f 2 DK and

 
a b=2 ;pri m
P.d / D T D 2 LC I 4ac  b D d :
2
b=2 c

;pri m
Here LC is the subset of LC consisting of primitive ones. We also put S.d / D
P.d /=
.
Lemma 10.21. We assume that d (< 0) is divisible by p.
p p X
(1) If Q. d / ¤ Q. p/, then Q .T / D 0.
T 2S.d /
p p
(2) If Q. d / D Q. p/, then Q .T / D 1 for any T 2 P.d /.
Proof. We first see that Q .T / is the value of at the norm of the ideal correspond-
ing to T if rank.T mod p/ D 1. In this case, by definition, there exist integers x,
y coprime with each other such that Q .T / D .e/ for e D ax 2 C bxy C cy 2 .
Since 4ae D .2ax C by/2 C .4ac  b 2 /y 2 and since we assumed that 4ac  b 2
is divisible by p, we have .ae/ D .2ax C by/2 . By assumption, we have
p − e, and replacing by an equivalent one if necessary, we may assume that
p − a. Hence we have .2ax C by/ ¤ 0 and .2ax C by/2 D 1, so we
have .a/ D .e/. p This is the value at the norm. Therefore, since we have
p
K D if K D Q. d / D Q. p/, we get Q .T / D 1 by Proposition 10.2
p p
and Lemma 10.11. If K D Q. d / ¤ Q. p/, then K ¤ . So Q .T / takes 1
on some elements of S.d / again by Proposition 10.2. That is, Q is a character of
the group S.d / which is not trivial, so the sum over the group becomes 0. t
u
Proof of Theorem. First, it is obvious that Q .eT / D .e/ Q .T / for any integer e
prime to p and any T 2 LC , and since
.eT / D
.T /, we reduce the L-function to
the sum over primitive quadratic forms. Namely if we put

X Q .T /
Lpri m .s; LC ; / D ;
;pri m

.T / det.T /s
T 2LC =

then we have
176 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

1
X X .e/ Q .T /
L.s; LC ; / D
eD1 T 2L;pri m =

.eT / det.eT /s
C

1
X .e/ pri m
D L .s; LC ; /
eD1
e 2s

D L.2s; /Lpri m .s; LC ; /:

Here L.2s; / is the Dirichlet L-function. Next, if we apply the previous lemma to
;pri m
LC , only the part for T with 4 det.T / D d D f 2 p (f are positive integers)
remains alive. Since
.T / has a common value for all the elements in P.d /, we
denote this by
.d /. Then we have
1
X jS.d /j
Lpri m .s; LC ; / D 22s :
p s f 2s
.d /
f D1

By virtue of the relation between primitive quadratic forms and proper primitive
ideals of an order of a quadratic field, jS.d /j is the class number h.f 2 DK / of the
p
order Of of the imaginary quadratic field K D Q. p/. By the class number
formula, we have

1  
h.DK / X 12s Y 1
Lpri m .s; LC ; / D 22s f 1  .q/ :
jO
max jp
s q
f D1
qjf
qWpri me

Here by Theorem 10.12, we have

h.DK / B1;
D s :
jO
max jp s 2p

The above sum is calculated as

1
X Y 1
 1
X Y  1

f 12s 1 .q/ D f 12s 1C .q/.q/
q q
f D1 qjf qjf f D1
qWpri me

1
X X 1
D f 12s .m/.m/:
m
f D1 mjf

If we put f D mn here, the above sum becomes

10.3 Some L-Functions Associated with Quadratic Forms 177

1
X Y
n12s m2s .m/.m/ D .2s  1/ .1  .q/q 2s /
m;nD1 qWpri me

D .2s  1/L.2s; /1 :

Combining all the above considerations together, we get the expression of

L.s; LC ; / in the theorem. As for special values, it is an immediate consequence
of the fact that the values of the Riemann zeta function at integers not more than 0
are given by .1  m/ D Bm =m (p. 72, Theorem 5.4). t
u
The above relatively easy considerations gave us a chance to start thinking that
zeta functions of prehomogeneous vector spaces consisting of symmetric matrices
would be easy objects. As things turned out, we found out that zeta functions for the
vector space of symmetric matrices of any degree are described by previously known
standard objects such as the Mellin transform of Eisenstein series of half-integral
weight and the Riemann zeta functions, and they have a decisive application for
dimension formulas of automorphic forms. See [46,48,49] for these developments.
Exercise 10.22. (1) Let DK be the discriminant
 of a quadratic field K and p an
odd prime such that K .p/ D p D 1. Show that there exists b 2 Z such
DK

b 2  DK mod 4p. Show also that for such b, the module p D Zp C

that p
bC DK
Z 2 is an ideal of Omax whose norm is p.
p
(2) Assume DK  1 mod 8 and put p D Z2 C Z 1C 2 DK . Show that this is an ideal
of Omax and N.p/ D 2. Also when DK  4 mod 8 (i.e. DK  12 mod 16),
give a prime ideal p with N.p/ D 2.
(3) Show that if a is an ideal of Omax and N.a/ is prime to DK , then K .N.a// D 1.
Exercise 10.23. (1) Let K be a quadratic field and Of is the order of conductor
f . Let a and b are proper Of ideals which are prime to f . Show that if a D b˛
for some ˛ 2 K  , then ˛ D ˇ= for some ˇ 2 Of prime to f and  2 Z
prime to f . p p p
(2) We put K D Q. 3/, O2 D Z C Z.2 3/ and ˛ D 4 C 3. Show that, although
N.˛/ D 13 is prime to 2, there is no pair ˇ,  2 O2 prime to 2 such that
˛ D ˇ= .
Exercise 10.24. Let a be a proper Of ideal. We define the norm N.a/ of a by the
cardinality of the set Of =a.
p
(1) If a D Za C Z bC2 D with a > 0, then show that N.a/ D a.
p
(2) Assume that a D Z!1 C Z!2 . Then show that N.a/ D j.!1 !2  !2 !1 /= Dj.
(3) Let ˛ 2 Of . Then show that N.˛a/ D jN.˛/jN.a/.
p
Exercise 10.25. Let a D Za C Z bC2 D
be a proper primitive Of ideal and put
p
b C D
a D Za C Z:
2
178 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

(1) Show that a is also a proper primitive Of ideal and that

aa D aOf :

(2) Denote by Q.x; y/ a representative of quadratic form corresponding to the

narrow ideal class containing a. Then show that the following conditions are
equivalent.
(i) Q.x; y/ belongs to an ambig class.
(ii) a2 D ˛Of with N.˛/ > 0.
p
Exercise 10.26. Let a D Za C Z bC2 D be a proper primitive Of ideal, where
D D f 2 DK for some fundamental discriminant DK . Assume that b  0 mod a.
Show that a2 is a principal ideal aOf .
Exercise 10.27 (Product of “einig” pair of ideals). Denote by DK the fundamen-
tal discriminant of a quadratic field K. Fix a discriminant D D f 2 DK and consider
two proper primitive Of ideals
p
bi C D
ai D Zai C Z ;
2
where ai > 0 (i D 1, 2). As is shown in Chap. 6, there exist integers c1 and
c2 such that D D b12  4a1 c1 D b22  4a2 c2 , and we have gcd.a1 ; b1 ; c1 / D
gcd.a2 ; b2 ; c2 / D 1. In this exercise, we will obtain the free basis of the product
p p p
b2 C D b1 C D b1 b2 C D C .b1 C b2 / D
a1 a2 D Za1 a2 C Za1 C Za2 CZ :
2 2 4

We assume that gcd.a1 ; a2 ; .b1 C b2 /=2/ D 1. (Here we note that b1  b2  2

since b12  D  b22 mod 4.) Such a pair of ideals is called einig by Dirichlet (see
[29]). We fix integers x, y, z such that xa1 C ya2 C z.b1 C b2 /=2 D 1. Define a
rational number b0 by the relation
p p p p
b2 C D b1 C D b1 b2 C D C .b1 C b2 / D b0 C D
xa1 C ya2 Cz D :
2 2 4 2

(1) Show that b0 is an integer.

(2) We put

b1  b2
˛Dx C c1 z;
2
b2  b1
ˇDy C c2 z;
2
 D .c2 x C c1 y/
10.3 Some L-Functions Associated with Quadratic Forms 179

Then, show the following equalities.

p ! p !
b1 C D b0 C D
a2 D a2 C a1 a2 ˛;
2 2
p ! p !
b2 C D b0 C D
a1 D a1 C a1 a2 ˇ;
2 2
p   p
b1 b2 C D C .b1 C b2 / D b1 C b2 b0 C D
D C a1 a2 :
4 2 2

(3) Show that

p
b0 C D
a1 a2 D Za1 a2 C Z
2

and N.a1 a2 / D a1 a2 .
b2  D
(4) If we put c0 D 0 , then this should be an integer. Show that
4a1 a2
 2
b1 C b2
c0 D .c1 z C b1 x/.c2 z C b2 y/ C .a1 x C a2 y/.c2 x C c1 y/  xy
2
D ˛ˇ  :

(Note, e.g. that b12  b22 D .D C 4a1 c1 /  .D C 4a2 c2 / D 4.a1 c1  a2 c2 /.)

(5) For variables x1 , x2 , y1 , y2 , put

X D .x1 C y1 ˛/.x2 C y2 ˇ/  c0 y1 y2 D x1 x2 C ˛x2 y1 C ˇx1 y2 C y1 y2 ;

b1 C b2
Y D a1 x1 y2 C a2 x2 y1 C y1 y2 :
2
Then show that

.a1 x12 C b1 x1 y1 C c1 y12 /.a2 x22 C b2 x2 y2 C c2 y22 / D a1 a2 X 2 C b0 X Y C c0 Y 2 :

Exercise 10.28 (General case). Fix a discriminant D D f 2 DK and take two

proper primitive Of ideals a1 and a2 as in the last exercise. This time, we assume
that
 
b1 C b2
gcd a1 ; a2 ; D l;
2
where l might be greater than 1. Take integers x, y, z such that
b1 C b2
xa1 C ya2 C z D l:
2
180 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

b1 Cb2
We write a1 D la10 , a2 D la20 and 2 D lB.
(1) Put

b1 b2 C D
b0 D xa1 b2 C ya2 b1 C z
2

Show directly that b0 is an integer and divisible by l.

(2) Put

b1  b2
˛Dx C c1 z;
2
b2  b1
ˇDy C c2 z;
2
 D .xc2 C yc1 /

Show the following equalities.

p ! p !
b1 C D 0 b0 C l D
a2 D a2 C la10 a20 ˛;
2 2
p ! p !
b2 C D 0 b0 C l D
a1 D a1 C la10 a20 ˇ;
2 2
p ! p ! p !
b1 C D b2 C D b0 C l D
DB C la10 a20 :
2 2 2

(3) We put m D gcd.l; ˛; ˇ;  /. Show that m D 1 by the following steps.

(i)

b1  b2
D a10 ˛  a20 ˇ 2 mZ:
2
Hence c1 z 2 mZ.
(ii)

b2  b1
B.c1 z/  a20  D c1 C Bx:
2
Hence c1 2 mZ.
(iii) We have b1 Cb
2
2
D lB 2 mZ and hence b1 2 mZ. Also we have
a1 D la10 2 mZ.
(iv) m D 1 since gcd.a1 ; b1 ; c1 / D 1.
10.3 Some L-Functions Associated with Quadratic Forms 181

(4) Show that

p p !
b0 C l D b0 = l C D
a1 a2 D Z la10 a20 CZ 0 0
D l Za1 a2 C Z ;
2 2

where b0 = l is an integer.
(5) Show that

N.a1 a2 / D N.a1 /N.a2 /:

(6) Find integers a3 , b3 , c3 , X and Y similarly as in the last exercise such that

.a1 x12 C b1 x1 y2 C c1 y12 /.a2 x22 C b2 x2 y2 C c2 y22 / D a3 X 2 C b3 X Y C c3 Y 2 :

(Hint: Write down the product of general elements of a1 and a2 by the basis
of a1 a2 and take the norms. For example, put a3 D a10 a20 , b3 D b0 = l, c3 D
b32  D
, and X D lx1 x2 C ˛x2 y1 C ˇx1 y2 C y1 y2 , Y D a10 x2 y1 C a10 x1 y2 C
4a10 a20
By1 y2 . )
Exercise 10.29. (1) We fix D D f 2 DK . Show that for a natural number m and a
proper Of ideal a, there exists a proper Of ideal b which is equivalent to a and
whose norm N.b/ is prime to m.
(2) For two proper Of ideals a1 , a2 , show that N.a1 a2 / D N.a1 /N.a2 / by using
Exercise 10.27 but not using Exercise 10.28.
p
Exercise 10.30. (1) For K D Q. 5/, let Of be the order of K of conductor f .
By using Proposition 10.18, count the number of proper Of ideals of norm 4
and 16 for f p D 1, 2, 3. Also, give them all explicitly.
(2) For K D Q. 11/ and the order O3 of K of conductor 3, give all the proper
ideals of norm 9 explicitly.
p
p 10.31. For K D Q. 7/, we take the order O3 of conductor 3. Put L D
Exercise
3Z C 7Z. Show that L is a proper O3 lattice, though it is not an integral ideal of
O3 (i.e. not contained in O3 ).
p p
Exercise 10.32. p For K D Q. 82/, let Omax D Z C Z 82 be the maximal order
and O2 D Z C 2 82Z the order of conductor 2.
p
(1) Show that a D 2Z C 82Z is an ideal of Omax .
(2) Show that a is not prime to 2 and that a \ O2 is not a proper ideal of O2 .
Exercise 10.33. Assume that a is a proper Of ideal. We write a by the standard
basis as a D l.Za C Z.d C f !// with N.d C f !/ D ac. We assume that a C
f Omax D Of . Prove that a C f Of D Of by the following steps.
(1) Two numbers l and f are coprime.
(Hint: If there exists a prime p such that pjl and pjf , then a C f Omax
pOmax , which is a contradiction.)
182 10 Class Number Formula and an Easy Zeta Function of the Space of Quadratic Forms

(2) Two numbers a and f are coprime.

(Hint: We have ac D N.d C f !/ D d 2 C df T r.!/ C f 2 N.!/. If there is a
prime p such that pja and pjf , then pjd , and a pOmax , which is again a
contradiction.)
(3) We have a C f Of D Of .
(Hint: By (1) and (2), two numbers al and f are coprime and we have x, y 2 Z
such that xal C yf D 1.)
Chapter 11
p-adic Measure and Kummer’s Congruence

In modern number theory, the p-adic method or p-adic way of thinking plays an
important role. As an example, there are objects called p-adic L-functions which
correspond to the Dirichlet L-functions, and in fact the natural setup to understand
the Kummer congruence described in Sect. 3.2 is in the context of the p-adic
L-functions. To be precise, a modified version (by a suitable “Euler factor”) of
Kummer’s congruence guarantees the existence of the p-adic L-function.
To discuss this aspect fully is beyond the scope of this book, but in this chapter we
explain the p-adic integral expression of the Bernoulli number and prove Kummer’s
congruence using it. Interested readers are advised to read books such as Iwasawa
[51], Washington [100], Lang [66].
We assume the basics of p-adic numbers. For this we refer readers to Serre [83,
Ch. 1] or Gouvea [37]. The results in this chapter are not used in other chapters.

11.1 Measure on the Ring of p-adic Integers and the Ring

of Formal Power Series

In this section we review the general correspondence between measures on the ring
of p-adic integers Zp and the ring of formal power series. We use this setup in the
next section to define the Bernoulli measure on Zp and to express Bernoulli numbers
as integrals. This expression turns out to be very useful in proving Kummer’s
congruence relation.
Let Qp be the algebraic closure of the field Qp of p-adic numbers. The p-adic
absolute value j j of Qp (normalized by jpj D 1=p) is extended uniquely to Qp . We
use the same notation j j for this extension. Then Qp is not complete with respect
to this absolute value, and the completion is denoted by Cp . The absolute value j j
also extends naturally to Cp . Let Op be the ring of integers of Cp :

Op D fx 2 Cp j jxj  1g:

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 183
in Mathematics, DOI 10.1007/978-4-431-54919-2__11, © Springer Japan 2014
184 11 p-adic Measure and Kummer’s Congruence

Remark 11.1. Like the complex number field C, the field Cp is complete and
algebraically closed. To do analysis in the p-adic setting, we need this big field.
First we review the general theory of measures on Zp .
Denote the Z-module Z=p n Z by Xn and the canonical map from XnC1 to Xn by
nC1 , so nC1 W XnC1 ! Xn is defined by

x mod p nC1 Z 7! x mod p n Z:

The system of pairs .Xn ; n / gives a projective system and we have the projective
limit lim Xn :


 Y 
lim Xn D .xn / 2 Xn j nC1 .xnC1 / D xn :

n1

The ring of p-adic integers Zp is identified with this projective limit lim Xn .


Definition 11.2 (Measure on Zp ). A set of functions  D fn g1

is called an
nD1
Op -valued measure on Zp if the following two conditions are satisfied:

(i) Each n is an Op -valued function on Xn , n W Xn ! Op .

(ii) For any n 2 N and x 2 Xn , the distribution property
X
n .x/ D nC1 .y/
y2XnC1
nC1 .y/Dx

holds.
The set of Op -valued measures on Zp is denoted by M.Zp ; Op / . This has an
Op -module structure. Further, the norm of  D fn g 2 M.Zp ; Op / is defined as

kk D sup jn .x/j:

n2N; x2Xn

Also, the Op -module of continuous Op -valued functions on Zp is denoted by

C.Zp ; Op /, and the norm k'k of an element ' 2 C.Zp ; Op / is defined by

k'k D sup j'.x/j:

x2Zp

For ' 2 C.Zp ; Op / and  D fn g 2 M.Zp ; Op / , the integral on Zp is defined by

Z pn 1
X
'.x/d.x/ D lim '.r/n .r/:
Zp n!1
rD0
11.1 Measure on the Ring of p-adic Integers and the Ring of Formal Power Series 185

(We use the abbreviated notation n .r/ for n .r mod p n /. A similar abbreviation
will be used in the following.) The convergence of the limit on the right-hand side
is guaranteed by the following estimate: when n < m, we have

ˇ pX
n 1 pm 1
X ˇ
ˇ ˇ
ˇ '.r/n .r/  '.l/m .l/ ˇ
rD0 lD0
ˇ pn 1 pmn 1 ˇ
ˇ X X ˇ
D ˇˇ '.r/n .r/  '.r C p n q/m .r C p n q/ ˇ
rD0 qD0
0 1
ˇ pX
n 1 pmn 1
X ˇ
ˇ @ ˇ
Dˇ .'.r/  '.r C p n q// m .r C p n q/A ˇ
rD0 qD0

 max j'.r/  '.r C p n q/j kk:

r; q

For each natural number k, the binomial polynomial

!
t t.t  1/    .t  k C 1/
D
k kŠ

in t is a continuous function on Zp .
To  D fn g 2 M.Zp ; Op / we associate f 2 Op ŒŒX  in the following manner.
n
Set  D Op ŒŒX , n D ..1 C X /p  1/ and consider the projective system
f.=n ; $n /g by the natural map $n W =n ! =n1 . Define fn .X / 2
=n by

pn 1 pn 1 r
! pn 1
X XX r X
fn .X / D n .r/.1 C X / D r
n .r/ X D
k
cn;k X k :
rD0 rD0 kD0
k
kD0

Here we understand that the equalities are mod n and put

pn 1
!
X r
cn;k D n .r/ :
rD0
k

Since we have
0 1
pn 1
X
.$n fn /.X / D $n @ n .r/.1 C X /r A
rD0
186 11 p-adic Measure and Kummer’s Congruence

0 1
pn1 1 p1
X X 0
D $n @ n .r 0 C p n1 l/.1 C X /r .1 C X /p A
n1 l

r 0 D0 lD0

pn1 1
X 0
D n1 .r 0 /.1 C X /r
r 0 D0

D fn1 .X /;

the system .fn / is an element in the projective limit lim =n . Now we have the

isomorphism

 Š lim =n ;  3 g 7! .gn / 2 lim =n ;

 

where, for g 2 , the system .gn / is given by gn D g mod n . Through this

isomorphism, the above ffn g corresponds to f 2  by
1
X
f .X / D cm X m ;
mD0

where
pn 1
!
X r
cm D lim n .r/
n!1
rD0
m
Z !
x
D d.x/:
Zp m

We therefore have obtained a map from M.Zp ; Op / to Op ŒŒX . An important

fact is that this map gives a natural isomorphism between M.Zp ; Op / and the ring
of formal power series Op ŒŒX , often referred to as the Iwasawa isomorphism. The
way to associate
P a measure to an element in Op ŒŒX  is described as follows.
For f D 1 mD0 cm X 2 Op ŒŒX , define  D fn g by
m

1 X r
n .r/ D  f .  1/ .r 2 Xn /; (11.1)
pn pn
 D1

the sum running over all p n -th roots  of 1. Since j  1j < 1, f .  1/ converges.
For each m  0, we have
!
1 X r 1 X X r m
m
 .  1/ D n m
 .1/mj  j
pn pn p pn j D0 j
 D1  D1
!
X m
D .1/mj :
j
0j m
j r mod pn
11.1 Measure on the Ring of p-adic Integers and the Ring of Formal Power Series 187

So this is contained in Op . In particular, if p n > r > m, then this is zero. When  is

a primitive p  -th root of 1 (  1), the equality

j  1j'.p / D jpj .' is the Euler function/
holds and hence

j.  1/m j D jp m='.p / j:
From this, we conclude that p e divides the quantity
!
X m
.1/mj
0j m
j
j r mod pn

for e D m=.p n/  n. Therefore,

0 1
1
X X
1
n .r/ D cm @ n  r .  1/m A
p pn
mD0  D1

is convergent and the value is in Op . To check the distribution property (ii) of the
measure, we need to calculate the following value:
X X
nC1 .y/ D nC1 .x C p n a/
y2XnC1 ; nC1 .y/Dx a mod p
0 1
1 X X
D @  .xCp
n a/
A f .  1/:
p nC1 a mod p
 p nC1 D1

Using the identity


X 0 if  p ¤ 1;
n
pn a
 D n
a mod p p if  p D 1

for a p nC1 -th root  of 1, we have

X X
 xp a D p  x ;
n

n
a mod p  p D1
nC1
p D1

so we have
X
nC1 .y/ D n .x/
y2XnC1
nC1 .y/Dx
188 11 p-adic Measure and Kummer’s Congruence

which is to be proved. If we define the formal power series fQ 2 Op ŒŒX 

corresponding to this measure defined as before, then the coefficients ck0 of X k of
this series are given by

pn 1
!
X r
ck0 D lim n .r/
n!1
rD0
k
1 pn 1
! !
X X r X m
D lim cm .1/mj :
n!1
mD0 rD0
k 0j m
j
j r mod p n

0
We fix k. To calculate the coefficient
 r  of cm in the expression of ck in the right-hand
side above, we fix m. We have k D 0 for k > r so we may assume that k  r.
Taking n big enough, we assume that m < p n . Then, if j  r mod p n for some j
with 0  j  m, we have j D r since we also have 0  r  p n  1 by definition.
So we may assume that k  r D j  m. So the coefficient of cm is given by
! ! ! !

X
m
r m X mk m
mk
1 if m D k;
.1/ mr
D .1/mki D
k r i k 0 if m ¤ k:
rDk i D0

Hence we have ck0 D ck . So we have fQ D f and two mappings are inverse with
each other and we see that the set M.Zp ; Op / of Op -valued measures and the space
of formal power series Op ŒŒX  are bijective.
More precisely, we can introduce a product for both spaces and show that these
are isomorphic as Op algebras, as given in the following theorem whose complete
proof is omitted (see e.g. Lang [66, Ch.4]).
For two measures ;  2 M.Zp ; Op /, we define an Op -valued function ./n
on Xn by

pn 1
X
.  /n .x/ D n .y/n .x  y/ .x 2 Xn /: (11.2)
yD0

Then    D f.  /n g becomes an element of M.Zp ; Op /. We call this a

convolution product of  and . The set M.Zp ; Op / becomes an Op algebra by
this product   .
Theorem 11.3 (Iwasawa isomorphism). Between the space M.Zp ; Op / of Op -
valued measures and the ring of formal power series Op ŒŒX , there is an Op
algebra isomorphism P W M.Zp ; Op / ! Op ŒŒX  given by

1
X
P
M.Zp ; Op / 3  D fn g 7! f .X / D cm X m 2 Op ŒŒX :
mD0
11.1 Measure on the Ring of p-adic Integers and the Ring of Formal Power Series 189

Here, cm is determined by :
Z !
x
cm D d.x/;
Zp m

and conversely n is determined by f :

1 X x
n .x/ D  f .  1/:
pn pn
 D1

For convenience of the description below, we recall Mahler’s1 theorem giving

the necessary and sufficient condition for an Op -valued function on Zp to be
continuous.
Theorem 11.4. The function ' W Zp ! Op is continuous if and only if it can be
written as
1
!
X x
'.x/ D an ; an 2 Op ; jan j ! 0:
nD0
n

If this is the case, the coefficients an are uniquely determined by ' and given by
!
X
n
n
an D .1/ nk
'.k/:
k
kD0

We omit the proof (cf. Lang [66, §4.1]).

If we use Theorem 11.4, we can understand a part of Theorem 11.3 more
intuitively as follows. Fix x0 2 Z. Denote by ' the characteristic polynomial of
x0 C p n Zp . Then by the definition of the p-adic measure, we see easily that
Z
'.x/d.x/ D n .x0 /:
Zp

P1 x 
So if we replace '.x/ by the expansion '.x/ D mD0 am m in Theorem 11.4, we
have
1 1
!
X X X
m
m
n .x0 / D am cm D cm .1/ mk
'.k/:
mD0 mD0
k
kD0

1
Kurt Mahler (born on July 26, 1903 in Krefeld, Prussian Rhineland—died on February 25, 1988
in Canberra, Australia).
190 11 p-adic Measure and Kummer’s Congruence

n
Now, for any  with  p D 1, we have
1 1
!
X X
mXm
cm .  1/ D m
cm .1/mk  k :
mD0 mD0
k
kD0

Since '.k/ D 1 if k  x0 mod p n and '.k/ D 0 otherwise, we have

1 1
!
1 X x0 X X X
m
m
 cm .  1/ D
m
cm .1/mk '.k/:
pn pn k
 D1
mD0 mD0 kD0

So we get the expression of .x/ by f in Theorem 11.3.

We describe here several useful properties of the correspondence P in Theo-
rem 11.3 between measures and formal power series. Let the maximal ideal of Op be

P D fz 2 Op j jzj < 1g:

For z 2 P, define the function .1 C z/x in x by

1
!
X x n
.1 C z/ WD
x
z :
nD0
n

By Mahler’s theorem, .1 C z/x is a continuous function of x 2 Zp . When x is a

non-negative integer, this definition of .1 C z/x coincides with the usual binomial
expansion. We have the relation
0 0
.1 C z/x .1 C z/x D .1 C z/xCx .x; x 0 2 Zp /: (11.3)

This is obvious for x; x 0 2 N, and the general case for x; x 0 2 Zp follows from the
fact that the set N of natural numbers is dense in Zp .
In the following, we list several properties of measures and corresponding power
series, which will be used later.
Property (1). Let z 2 P. If  corresponds to f (i.e. P D f ), then
Z
f .z/ D .1 C z/x d.x/:
Zp

In particular, by putting z D 0,
Z
f .0/ D d.x/:
Zp
11.1 Measure on the Ring of p-adic Integers and the Ring of Formal Power Series 191

1
X
Proof. Writing f .X / D cn X n , we have by Theorem 11.3
nD0

Z 1 Z !
X x n
.1 C z/ d.x/ D
x
z d.x/
Zp Zp nD0 n

1 Z !
X x
D zn d.x/
nD0 Zp n
1
X
D cn zn D f .z/:
nD0

t
u
We call the map from C.Zp ; Op / to Op a bounded linear functional on
C.Zp ; Op / if the following conditions (i), (ii) are satisfied:
(i) For any '; ' 0 2 C.Zp ; Op / and any a; b 2 Op ,

.a' C b' 0 / D a .'/ C b .' 0 /:

(ii) There exists a positive constant M > 0 such that for any ' 2 C.Zp ; Op /,

j .'/j  M k'k:

The norm of is defined by

j .'/j
k k D sup :
'2C.Zp ;Op / k'k
'¤0

Let be a bounded linear functional on C.Zp ; Op /. For x 2 Xn D Z=p n Z, write

the characteristic function of x C p n Zp as 'x; n . If we put

n .x/ D .'x; n /;

then  D fn g is an Op -valued measure on Zp (i.e.  2 M.Zp ; Op /). Conversely,

given  D fn g 2 M.Zp ; Op /, if we put
Z
.'/ D '.x/ d.x/;
Zp
192 11 p-adic Measure and Kummer’s Congruence

then is a bounded linear functional on C.Zp ; Op /. This correspondence between

and  is easily seen to be one to one.
Moreover, for h 2 C.Zp ; Op / and  2 M.Zp ; Op / , the map
Z
' 7! '.x/h.x/ d.x/; .' 2 C.Zp ; Op //
Zp

is a bounded linear functional on C.Zp ; Op /. Let h be the corresponding measure.

It is an interesting problem to compute the formal power series corresponding to the
measure h when  corresponds to f D P 2 Op ŒŒX . Properties (2) and (3)
below give examples of this correspondence.
For f 2 Op ŒŒX , put

1 X
.Uf /.X / D f .X /  f ..1 C X /  1/: (11.4)
p p
 D1

Since
1 X
..1 C X /  1//l 2 Zp ŒX 
p p
 D1

for non-negative integers l, we have Uf 2 Op ŒŒX .

Property (2). Let f 2 Op ŒŒX  and f be the corresponding measure. Also, let
be the characteristic function of Z
p . Then the formal power series corresponding
to the measure f is Uf , i.e., f D Uf . More precisely, we have for any
' 2 C.Zp ; Op /
Z Z
'.x/ .x/ df .x/ D '.x/ dUf .x/:
Zp Zp

This can also be written as

Z Z
'.x/ df .x/ D '.x/ dUf .x/:
Z
p Zp

Proof. Write the power series corresponding to the measure f as g. When z 2 P,

by Property (1) we have
Z
g.z/ D .1 C z/x .x/ df .x/:
Zp

Let  be a pth root of 1. Regarding also as a function on Z=pZ via .a mod

p/ D .a C pZp /, and putting
11.1 Measure on the Ring of p-adic Integers and the Ring of Formal Power Series 193

X
O ./ D 1 .a/ a ;
p
a2Z=pZ

(Fourier transform on Z=pZ) we have

X
.a/ D O ./ a
 p D1

by a simple calculation (inverse Fourier transform). Since


 p1 if  ¤ 1;
O ./ D p1
p
if  D 1;

by the definition of , we obtain

Z
g.z/ D .1 C z/x .x/ df .x/
Zp
Z X
D .1 C z/x O ./ x df .x/
Zp  p D1
X Z
D O ./ .1 C z/x  x df .x/
 p D1 Zp

X Z  x
D O ./ 1 C ..1 C z/  1/ df .x/
 p D1 Zp

X X
D O ./f ..1 C z/  1/ D f .z/  1 f ..1 C z/  1/:
p p
 p D1  D1

This shows g D Uf . (Here we define the power  x for x 2 Zp by

1
!
X x
 D .1 C   1/ D
x x
.  1/n :
nD0
n

If we choose a 2 Z so that x  a 2 pZp , we have  x D  a .) t

u
Define the differential operator D on the ring of formal power series Op ŒŒX  by

d
D D .1 C X /DX ; where DX D :
dX
194 11 p-adic Measure and Kummer’s Congruence

Property (3). For f 2 Op ŒŒX , the power series corresponding to the
measure xf is Df . Hence the power series corresponding to the measure
x k f .k natural number/ is D k f and the equalities
Z Z
x k df .x/ D dD k f .x/ D .D k f /.0/
Zp Zp

hold.
Proof. It is enough to show this when k D 1. Let g 2 Op ŒŒX  be the power series
corresponding to the measure xf . By Property (1), we have for z 2 P
Z
g.z/ D x.1 C z/x df .x/:
Zp

P1 P1
Put f .X / D nD0 an X n ; g.X / D nD0 bn X n . Using
! ! !
X X X
X D .n C 1/ Cn
n nC1 n

and Theorem 11.3, we have

Z ! Z !
x x
bn D dg .x/ D x df .x/
Zp n Zp n
Z ! Z !
x x
D .n C 1/ df .x/ C n df .x/
Zp n C 1 Zp n

D .n C 1/anC1 C nan :

On the other hand, Df is computed as

.Df /.X / D ..1 C X /DX f /.X /

D .1 C X /.a1 C 2a2 X C    C nan X n1 C    /
1
X
D ..n C 1/anC1 C nan /X n :
nD0

This gives g D Df . t
u
In general, for a power series f .X /, we define a new power series f  .Z/ in Z
by setting X D e Z  1:

f  .Z/ D f .e Z  1/: (11.5)

11.1 Measure on the Ring of p-adic Integers and the Ring of Formal Power Series 195

For example, when

1
!
X a
f .X / D .1 C X / D a
X n;
nD0
n

we have

X1
an Z n
f  .Z/ D e aZ D :
nD0
nŠ

Note the identity

.DZk f  /.0/ D .D k f /.0/ (11.6)

since

DZ f  .Z/ D .1 C X /DX f .X / D Df .X /:

The next property is the basis of the fact that the isomorphism P in Theorem 11.3
is an Op algebra isomorphism.
Property (4). Let the measures ;  correspond respectively to the power series
f; g 2 Op ŒŒX  (i.e.,  D f ;  D g ). Then the power series corresponding to
the convolution    is fg:

f  g D fg :

Proof. By Eq. (11.1), we have

1 X r
n .r/ D  f .  1/;
pn pn
 D1

1 X kCr
n .k  r/ D n  g.  1/:
p pn
 D1

Substituting this into the right-hand side of (11.2), we obtain

pn 1
X 1 X 1 X kCr
.  /n .k/ D n
 r f .  1/ n  g.  1/
p pn p pn
rD0  D1  D1

p 1 n
1 XX k 1 X
D n f .  1/g.  1/  n .=/r
p p rD0
 
196 11 p-adic Measure and Kummer’s Congruence

1 X
D f .  1/g.  1/ k
pn


D fg; n .k/:

Here  and  run through all p n -th roots of 1. From this, Property (4) follows. t
u

11.2 Bernoulli Measure

We define a specific measure called the Bernoulli measure. Recall that the first
Bernoulli polynomial is by definition equal to

1
B1 .x/ D x  :
2
In the following, p denotes an odd prime. For each natural number n and x 2 Xn D
Z=p n Z, set


x
En .x/ D B1 ;
pn

where in the right-hand side, we regard x as an integer representing x mod p n , and

for w 2 R, fwg is the real number satisfying 0  fwg < 1 and w  fwg 2 Z (the
fractional part of w). Then E D fEn g is a measure on Zp but is not Op -valued. We
modify this as follows in order to have an Op -valued measure. Take an invertible
element c in Zp (i.e. c 2 Zp ), and for x 2 Xn D Z=p Z, let
n

Ec;n .x/ D En .x/  cEn .c 1 x/:

We understand c 1 x as an element in Xn D Z=p n Z. It is easy to see that Ec D

fEc;n g is an Op -valued measure. We call this the Bernoulli measure.
Proposition 11.5. (1) The formal power series corresponding to the Bernoulli
measure Ec is given by
1 c
fc .X / D  :
X .1 C X /c  1

(2) Let k be a natural number. For c 2 Z

p with c ¤ 1, we have
k

Z
Bk .1/k
D x k1 dEc :
k 1  ck Zp

In particular, if p  1 − k, then Bk =k 2 Z.p/ .

11.2 Bernoulli Measure 197

Proof.(1) Since c 2 Z
p , we see fc 2 Zp ŒŒX , the first two terms of fc .X / being

c1 1  c2
fc .X / D C X C  :
2 12

Let  D fn g be the measure on Zp corresponding to fc by Theorem 11.3. For

r 2 Xn D Z=p n Z we have

1 X r
n .r/ D  fc .  1/
pn pn
 D1

X  
1 1 1 c
D fc .0/ C n  r  c :
pn p n  1  1
 p D1; ¤1

n
Now we use Lemma 8.5 on p. 110. For  p D 1;  ¤ 1 and f D p n , the lemma
gives

f 1
1 1 X cj
D j
c  1 f j D1

since .c; p/ D 1. By this, if we choose l so that cl  k mod p n ; 0  l < p n ,

we obtain
f 1
1 X c c X k X cj c.f  1/
 k D  j 
f n c  1 f 2 pn j D1
2f
 p D1; ¤1  D1

cl c.f  1/
D 
f 2f

1 
c k c c
Dc n
 C
p 2 2f

and by substituting this into the formula for n .r/ above and noting that fc .0/ D
.c  1/=2, we have


1  
c1 r 1 1 c r c c
n .r/ D C  C  c C 
2f pn 2 2f pn 2 2f

  
1  
r 1 c r 1
D n
 c n
 :
p 2 p 2

By the definition of the Bernoulli measure, we conclude n .r/ D Ec;n .r/, i.e.,
 D Ec and the power series corresponding to Ec is fc .
The proof of (2) goes as follows. By Property (3) and Eq. (11.6) we have
198 11 p-adic Measure and Kummer’s Congruence

Z
x k1 dEc D .D k1 fc /.0/ D .DZk1 fc /.0/:
Zp

Here by definition (11.5), we have

1 c
fc .Z/ D fc .e Z  1/ D 
e Z  1 e cZ  1
1
X Z n1
D .1  c n /.1/n Bn ;
nD1
nŠ

so we have
Bk
.DZk1 fc /.0/ D .1  c k /.1/k
k
and thus
Z
Bk
x k1 dEc D .1  c k /.1/k :
Zp k

This gives (2). t

u

11.3 Kummer’s Congruence Revisited

The “right” formulation of Kummer’s congruence is the following.

Theorem 11.6. Suppose p is an odd prime.
(1) Assume that m is a positive even integer such that p  1 − m. Then Bm =m 2
Z.p/ .
(2) Let a be a positive integer, and m and n positive even integers satisfying m 
n mod .p  1/p a1 and m 6 0 mod .p  1/. Then we have

Bm Bn
.1  p m1 /  .1  p n1 / mod p a :
m n
To prove this, we need the following integral expression of the Bernoulli number,
a refined version of Proposition 11.5 (2).
Proposition 11.7. Let k be a positive even integer and take c 2 Z
p . Then we have

Z
Bk
.1  c k /.1  p k1 / D x k1 dEc :
k Z
p
11.3 Kummer’s Congruence Revisited 199

Proof. The power series that corresponds to the Bernoulli measure Ec is fc in

Proposition 11.5. As in (11.4), define from fc a new power series g by

1 X
g.X / D Ufc .X / D fc .X /  fc ..1 C X /  1/:
p p
 D1

We have g 2 Op ŒŒX  and so we let  D g be the measure on Op obtained from g.

By Property (2) on p. 192 we have
Z Z
x k1 dEc D x k1 d:
Z
p Zp

Further, using Property (3) on p. 194 and (11.6) one sees

Z
x k1 d D .D k1 g/.0/ D .DZk1 g  /.0/:
Zp

We compute the value .DZk1 g  /.0/ . First,

 
 1 c 1 X 1 c
g .Z/ D Z    :
e  1 e cZ  1 p p e Z  1  c e cZ  1
 D1

Here, since

1 X 1 1
D p ;
p p X  1 X 1
 D1

we get
 
1 c 1 c
g  .Z/ D  cZ   cpZ
e 1 e 1
Z e 1 e
pZ 1
1
X 1
X
k Bk Bk
D .1  c /.1/
k
Z k1  .1  c k /.1/k .pZ/k1
kŠ kŠ
kD0 kD0
1
X Bk k1
D .1  c k /.1  p k1 /.1/k Z :
kŠ
kD1

Hence if k is even we have

Bk
.DZk1 g  /.0/ D .1  c k /.1  p k1 / ;
k
and the proposition is established. t
u
200 11 p-adic Measure and Kummer’s Congruence

Proof of Theorem 11.6. The first assertion is already given in Theorem 3.2, but we
give here an alternative proof for that too. Since we assumed m 6 0 mod p  1, we
can take c 2 Z such that .c; p/ D 1 and c m 6 1 mod p. For instance one may take
a primitive root mod p. From the proposition above, we have
Z
Bn
.1  c n /.1  p n1 / D x n1 dEc
n Z
p

and
Z
Bm
.1  c /.1  p
m m1
/ D x m1 dEc :
m Z
p

The assumption m  n mod .p  1/p a1 gives c nm  1 mod p a , and since we
assumed .1c m ; p/ D 1, we have also .1c n ; p/ D 1. Since Ec is an Op measure,
the above integral values are in Op and we see that Bn =n and Bm =m 2 Z.p/ . Since
x m1  x n1 mod p a if x 2 Z p , and since Ec is an Op -valued measure, we have

 
n1 Bn m1 Bm
.1  c / .1  p /
n
 .1  p / 2 p a Op :
n m

The left-hand side being contained in Zp , we conclude

Bn Bm
.1  p n1 /  .1  p m1 / 2 p a Zp :
n m
This proves the theorem. t
u
Theorem 3.2 is a corollary of Theorem 11.6. Indeed, if a < m  n, then by
Theorem 11.6, we have

Bm Bn
.1  p m1 /  .1  p n1 /
m n
 
Bm Bn Bn nm
D .1  p m1 /  C .p  1/p m1
m n n
 0 mod p a :

Since p  1 − n, we have Bn =n 2 Z.p/ by Theorem 11.6. Since a  m  1, we have

p m1 Bn =n 2 p a Z.p/ . Hence we have

Bm Bn
  0 mod p a :
m n
11.3 Kummer’s Congruence Revisited 201

Exercise 11.8. Give an example of an odd prime p and integers 2  a D m < n

such that the congruence in Theorem 3.2 does not hold. Check that for the same
choice of a, n, m and p, the congruence of Theorem 11.6 surely holds.
Hint: For example, put p D 5, a D m D 2 and n D 22 and use the following
values:
1 854513
B2 D ; B22 D :
6 138
Exercise 11.9. Show that the Bernoulli number Bn is given by the limit (p-adic
limit in Qp )

p 1
m
1 X n
lim m i :
m!1 p
i D0

(For a function f W Zp ! Qp with a suitable condition, the limit

p 1m
1 X
lim m f .i /
m!1 p
i D0

is sometimes referred to as the Volkenborn integral of f over Zp . See [94, 95] for
details.)
Chapter 12
Hurwitz Numbers

12.1 Hurwitz Numbers

In this section, we briefly introduce Hurwitz’s generalization of Bernoulli numbers,

known as the Hurwitz numbers.
As we have seen in Proposition 1.17, Bernoulli numbers appear as the coeffi-
cients of expansions of trigonometric functions. Hurwitz replaced the trigonometric
function by a “lemniscate function”, which, in modern terms, is an elliptic
p function
having complex multiplication by the ring of Gaussian integers ZΠ1, and
considered its expansion. It was Gauss who first studied the lemniscate function
or more general elliptic functions, but most of his work on them did not appear
during his lifetime. In his Disquisitiones [35], Gauss mentioned at the beginning of
Chap. 7 that his theory of cyclotomy developed there could also be applied to the
division of the lemniscate. Inspired by this comment, Abel1 and Jacobi started their
investigation on elliptic functions and laid the foundations of the theory.
Let us define Hurwitz numbers. For this, put
Z 1
dx
$ D2 p .D 2:622057 : : : /:
0 1  x4
This is one-half of the arc-length of the lemniscate defined by

r 2 D cos.2/;

and is analogous to the similar quantity for the case of the circle:
Z 1
dx
D2 p :
0 1  x2

1
Niels Henrik Abel (born on August 5, 1802 in Frindoe, Norway—died on April 6, 1829 in Froland,
Norway).

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 203
in Mathematics, DOI 10.1007/978-4-431-54919-2__12, © Springer Japan 2014
204 12 Hurwitz Numbers

p
Let }.z/ be the Weierstrass }-function with periods $ and $ 1 defined by
X  
1 1 1
}.z/ D 2 C  2 : (12.1)
z p .z  / 2
2Z$ CZ$ 1
¤0

The function }.z/ is a doublypperiodic meromorphic function on the complex

plane with pperiods $ and $ 1, having double poles at the lattice points
Z$ C Z$ 1 and holomorphic elsewhere (for the basic properties p of the elliptic
function, see [104]).
p Owing to the fact that the lattice Z$ CZ$ p1 is preserved by
multiplication by 1, the function z 7! }. z/ for any 2 ZΠ1 is expressible
0
as a rational function of }.z/ and }p .z/. By virtue of this property, }.z/ is said to
have complex multiplication by ZΠ1. In particular, it is easily seen from the
definition that }.z/ is an even function and
p
}. 1z/ D }.z/:

Also, }.z/ satisfies the differential equations

} 0 .z/2 D 4}.z/3  4}.z/

and

} 00 .z/ D 6}.z/2  2:

Write the Laurent expansion of }.z/ at z D 0 as

1
X 2n Hn zn2
1
}.z/ D C (12.2)
z2 nD2 n .n  2/Š
p
and use this to define the Hurwitz numbers Hn . From the property }. 1z/ D
}.z/, we see that Hn D 0 unless n is a multiple of 4. The analogous expansion for
the Bernoulli numbers is
1
X .1/ 2 1 2n Bn x n2n
1 1
D 2C ;
2
sin .x/ x nD2
n .n  2/Š

which can be obtained from the expansion of cot given in Proposition 1.17 by
differentiation, since cot0 .x/ D  sin21.x/ . In a similar manner to the proof of
Proposition 1.15, namely by comparing the coefficients of

} 00 .z/ D 6}.z/2  2; (12.3)

12.1 Hurwitz Numbers 205

Table 12.1 Hurwitz numbers

n 4 8 12 16 20 24 28 32
1 3 567 43659 392931 1724574159 2498907956391 1671769422825579
Hn 10 10 130 170 10 130 290 170

we have
1
H4 D ;
10
and the recursion
!
X
n1
4n
.2n  3/.4n  1/.4n C 1/H4n D3 .4i  1/.4n  4i  1/ H4i H4.ni /
i D1
4i

for n  2. In particular, we see from this that H4n is a positive rational number.
Several values are in Table 12.1.
The analogy between Hurwitz and Bernoulli numbers is more apparent when we
look at the special values of zeta functions. Namely, first we rewrite the formula
(4.4) as

X 1 2n
n1 .2/
D .1/ B2n .n  1/:
r2Z
r 2n .2n/Š
r¤0

Note the sum on the left runs over all non-zero integers. If we replace the rational
integers here by Gaussian integers, we have the formula

X 1 .2$/4n
D H4n :
p 4n .4n/Š
2ZCZ 1
¤0

To deduce this, expand the right-hand side of the definition (12.1) of }.z/ by using
2
1
.z /2
D 12 C 2z3 C 3z
4
C    and compare with (12.2).
Hurwitz proved a Clausen–von Staudt type theorem for Hn . To state this, we
need the following classical theorem of Fermat and Euler.
Theorem 12.1. Any prime number congruent to 1 modulo 4 can be written as a
sum of two squares essentially in the unique way.
Various proofs of this theorem are known since Euler. See [50] or [77] for some
of the standard proofs. For Zagier’s “One-sentence proof”, which simplifies a proof
by Heath-Brown, see [108].
Suppose now that p is a prime number congruent to 1 modulo 4 and write

p D a2 C b 2 :
206 12 Hurwitz Numbers

We may assume a is odd and b is even. We change the sign of a if necessary in

order to have

a  b C 1 .mod 4/:

(Note that the right-hand side is independent of the choice of the sign of b since b
is even.) These conditions uniquely determine a and so we denote it by ap to make
the dependence on p explicit. For instance, we have

5 D .1/2 C 22 ; 13 D 32 C 22 ; 17 D 12 C 42 ;

so

a5 D 1; a13 D 3; a17 D 1:

Hurwitz’s theorem is then stated as:

Theorem 12.2. For each n  1, we have

X
4n
1 .2ap / p1
H4n D C C Gn
2 p1 mod 4
p
p1j4n

with an integer Gn . In the sum, p runs over prime numbers congruent to 1 modulo
4 such that p  1 divides 4n.
As examples of the theorem, we have

1 1 .2/
H4 D D C ;
10 2 5
3 1 .2/2
H8 D D C  1;
10 2 5
567 1 .2/3 6
H12 D D C C C 5;
130 2 5 13
43659 1 .2/4 2
H16 D D C C C 253;
170 2 5 17
392931 1 .2/5
H20 D D C C 39299;
10 2 5
1724574159 1 .2/6 62
H24 D D C C C 13265939:
130 2 5 13
Hurwitz also proved that every Gn for n > 1 is odd and determined its class
modulo 8. The proof [44] of Theorem 12.2 is not as elementary as the original
Clausen–von Staudt theorem. It uses the complex multiplication of the }-function
12.2 A Short Biography of Hurwitz 207

as well as the notion of Hurwitz integral series (cf. Chap. 7). See [27, 58] for a
modern interpretation of the theorem and Kummer-type congruence for Hn .

12.2 A Short Biography of Hurwitz

Adolf Hurwitz was born on March 26, 1859 in Hildesheim, Germany.2 His teacher
at the Gymnasium was Schubert,3 who is famous for his enumerative geometry
(“Schubert calculus”). Impressed by Hurwitz’s genius, Schubert came every Sunday
as a house tutor and persuaded Hurwitz’s father, who was not at all affluent, to
send his son to the university to study mathematics. Already during his time at
the Gymnasium, while he was still 17, Hurwitz wrote a joint paper with Schubert
on enumerative geometry. In the spring of 1877, he entered Munich Technical
University, where Klein4 was. From the fall of the same year until the spring
of 1879, he attended the lectures of Kummer, Weierstrass, and Kronecker at
Berlin University. Returning once again to Munich, and then moving to Leipzig
together with Klein, he obtained his doctoral degree there with research on modular
functions. He became a “Privatdozent” at Göttingen University in 1882 and then, on
the recommendation of Lindemann,5 a professor at Königsberg University in 1884.
Hilbert6 and Minkowski7 were among the students there. The exchanges among the
three of them in those early days and the influence which this had on Hilbert are
vividly described in [78]. In 1892, he moved to the ETH in Zürich as a successor
of Frobenius,8 a position he was to hold until his death. At the same time he was
offered a professorship at Göttingen as the successor of Schwarz, who moved to
Berlin, but he had already accepted the position at Zürich (the director at Zürich at
that time, Bleuler, went especially to Königsberg to make the contract).
Hurwitz’s health was not good. He got typhoid fever twice and suffered from
frequent migraines. And in 1905, one of his kidneys had to be removed. In the midst
of all this, he continued to do research and published in a large number of areas.

2
The description in this section is based on Freudenthal [34], Hilbert [41], and Pólya [76].
3
Hermann Cäsar Hannibal Schubert (born on May 22, 1848 in Potsdam, Germany—died on July
20, 1911 in Hamburg, Germany).
4
Felix Christian Klein (born on April 25, 1849 in Düsseldorf, Prussia (now Germany)—died on
June 22, 1925 in Göttingen, Germany).
5
Carl Louis Ferdinand von Lindemann (born on April 12, 1852 in Hannover, Hanover (now
Germany)—died on March 6, 1939 in Munich, Germany) who proved in 1882 that  is a
transcendental number.
6
David Hilbert (born on January 23, 1862 in Königsberg, Prussia (now Kaliningrad, Russia)—died
on February 14, 1943 in Göttingen, Germany).
7
Hermann Minkowski (born on June 22, 1864 in Alexotas, Russian Empire (now Kaunas,
Lithuania)—died on January 12, 1909 in Göttingen, Germany).
8
Ferdinand Georg Frobenius (born on October 26, 1849 in Berlin-Charlottenburg, Prussia (now
Germany)—died on August 3, 1917 in Berlin, Germany).
208 12 Hurwitz Numbers

His articles have been collected and comprise two volumes. As well as the Hurwitz
zeta functions and Hurwitz numbers treated in this book, there are many other topics
which carry his name, such as the Hurwitz class number relations, Hurwitz’s genus
formula, the Hurwitz estimates for the size of automorphism groups of Riemann
surfaces and his investigations on the quaternions and other algebras. And according
to Hilbert, continued fractions were also a favorite topic of Hurwitz. (He wrote
approximately five articles on the subject.)
Hurwitz seems to have been an exceptionally modest and courteous man.
According to Pólya,9 who was the main editor of his collected works, he never
failed to tip his hat in greeting even to service people whom he met on the street.
Moreover, he was so skilled at the piano that in his youth he hesitated between a
career as a mathematician or as a pianist. He died on the 18th of November, 1919,
at the age of 60.
Exercise 12.3. Deduce the recursion of Hn given in the text from (12.3).
Exercise 12.4. Compute the residues 2Hn mod 16 for first several n, and find the
pattern. (For a proof, see [44].)

9
George Pólya (born on December 13, 1887 in Budapest, Hungary—died on September 7, 1985 in
Palo Alto, USA).
Chapter 13
The Barnes Multiple Zeta Function

In this chapter, we introduce Barnes’ multiple zeta function, which is a natural

generalization of the Hurwitz zeta function, give an analytic continuation, and then
express their special values at negative integers by using Bernoulli polynomials.
Furthermore, for double zeta functions, we take up the problem of finding a
functional equation. This problem has several interesting points. For example, a
kind of Lerch1 type zeta function appears in the functional equation, and also the
proof has the flavor of the theory of automorphic forms.
The motivation to take up Barnes’ multiple zeta function here is that this is
deeply related with the very important and interesting conjecture about constructing
class fields over real quadratic fields, or totally real algebraic number fields (Stark–
Shintani2 conjecture, see [86, 87, 89] for example). Although we do not treat it in
this book, the value at s D 0 of the first-order derivative of Barnes’ multiple zeta-
function is expressed by special values of the multiple gamma function. Shintani
saw that for the construction of class fields of real quadratic fields, special values
of the double gamma function play an important role, and he proposed a detailed
conjecture that certain special values of the double gamma functions generate the
class fields, and solved part of his conjecture (cf. [85–87]). For the construction of
class fields over imaginary quadratic fields, modular forms such as the j -invariant
play an extremely important role, but in the case of real quadratic fields, any
viewpoint from modular forms is not considered in the above papers. The Lerch
type function treated in this chapter was introduced with the intention of giving
more or less such a viewpoint to the construction problem of class fields over real
quadratic fields. For details, see [6, 7].

1
Mathias Lerch (born on February 20, 1860 in Milinov, Bohemia (now Czech Republic)—died on
August 3, 1922 in Susice, Czechoslovakia (now Czech Republic)).
2
Takuro Shintani (born on February 4, 1943 in Kita-kyushu, Japan—died on November 14, 1980
in Tokyo, Japan).

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 209
in Mathematics, DOI 10.1007/978-4-431-54919-2__13, © Springer Japan 2014
210 13 The Barnes Multiple Zeta Function

13.1 Special Values of Multiple Zeta Functions

and Bernoulli Polynomials

Let all a; w1 ; w2 ; : : : ; wr be positive real numbers. We put w D .w1 ; : : : ; wr / and

define a zeta function .s; w; a/ by
1
X 1
X
.s; w; a/ D  .a C m1 w1 C    C mr wr /s :
m1 D0 mr D0

This zeta function converges absolutely for Re.s/ > r. This kind of zeta function
was systematically studied by Barnes3 [11, 12].
For a while, we assume Re.s/ > r. We see that .s; w; a/ has an integral
representation:
1 1 Z 1
1 X X
.s; w; a/ D  t s1 e .aCm1 w1 CCmr wr /t dt
.s/ m D0 m D0 0
1 r

Z 1 1
X 1
X
1
D t s1  e .aCm1 w1 CCmr wr /t dt
.s/ 0 m1 D0 mr D0

Z 1
1 e at
D t s1 dt
.s/ 0 .1  e w1 t /    .1  e wr t /
Z 1
1 e .w1 CCwr a/t
D t s1 dt:
.s/ 0 .e w1 t  1/    .e wr t  1/

Since the series .s; w; a/ converges absolutely for Re.s/ > r, by virtue of
Lebesgue’s dominant convergence theorem for the Lebesgue4 integral, the above
calculation is justified for the same range of s. When r D 1, this is essentially the
Hurwitz zeta function. Namely, for w1 > 0, we have
 
a
.s; w1 ; a/ D ws
1  s; :
w1

Now, a multiple zeta-function .s; w; a/ as well as the Hurwitz zeta function, can be
represented by a contour integral. Indeed, taking " > 0 small enough and denoting

3
Ernest Barnes (born on April 1, 1874 in Birmingham, England—died on November 29, 1953 in
Sussex, England).
4
Henri Léon Lebesgue (born on June 28, 1875 in Beauvais (Oise), France—died on July 26, 1941
in Paris, France).
13.1 Special Values of Multiple Zeta Functions and Bernoulli Polynomials 211

by I."; 1/ the contour in Sect. 9.2, we have

Z
1 e .w1 CCwr a/t
.s; w; a/ D t s1 dt: (13.1)
.s/.e 2 i s  1/ I.";1/ .e w1 t  1/    .e wr t  1/

An advantage of this representation is that .s; w; a/ is continued analytically to a

meromorphic function on the whole s-plane by this, because the contour integral
on the right-hand side converges absolutely for arbitrary s 2 C and it defines a
holomorphic function on the whole s-plane.
A point which possibly becomes a pole of .s; w; a/ is among zeros

s D r; r  1; : : : ; 0; 1; 2; : : :

of e 2 i s  1 which are not in the range of absolute convergence. But .s/ has poles
of order 1 at integer points less than or equal to 0, so points s D 1  m .m 2 N/
cannot appear as poles and .s; w; a/ is holomorphic there.
Let us look for special values of .s; w; a/ at integer points s D 1  m .m 2 N/.
For any integer k  0, and for any fixed a and w, we define constants Ck .w; a/ by

X 1
e at t r
Qr D Ck .w; a/t k ;
i D1 .e w i t  1/
kD0

Proposition 13.1. We take m 2 N. If we write a > 0 as

a D a1 w1 C a2 w2 C    C ar wr .a1 ; a2 ; : : : ; ar 2 R/;

we have

.1  m; w; a/
0 1
X Y r
Bmj .aj /
D .1/r .m  1/Š @ A w1m1 1    wrmr 1
j D1
m j Š
m1 C  Cmr DmCr1
m1 0; :::; mr 0

D .1/r .m  1/ŠCmCr1 .w; a/:

Proof. The proof is based on the same method as the one used for looking at
special values of the Hurwitz zeta functions. In the same way as in the proof of
Proposition 9.3, we have
 Z
1 e .w1 CCwr a/t
.1  m; w; a/ D lim t m dt
s!1m .s/.e 2 i s  1/ I."/ .e w1 t  1/    .e wr t  1/
212 13 The Barnes Multiple Zeta Function

Z Y r
.1/m1 .m  1/Š e .1aj /wj t
D t m dt:
2 i I."/ j D1
e wj t  1

Now by using the generating function of Bernoulli polynomials (p. 56, (4.2)) and
Proposition 4.9 (4), the Laurent expansion of the integrand is obtained as follows:

Y r X1 X1 Y r  
m e .1aj /wj t m .wj t/mj 1
t Dt  Bmj .1  aj /
j D1
e wj t  1 m D0 m D0 j D1
mj Š
1 r

0 1
1 1 m 1
X X Y
r
.1/mj Bmj .aj /wj j
D  @ A t m1 C  Cmr rm :
m1 D0 mr D0 j D1
mj Š

The residue at t D 0 of this integrand is given by

0 1
m 1
X Y
r
Bmj .aj /wj j
@ .1/mj A:
j D1
mj Š
m1 C  Cmr DmCr1
m1 0; :::; mr 0

Using the generating function of Bernoulli polynomials in Proposition 4.9, we have

1
X Bm1 .a1 /    Bmr .ar / m1 r m1 CCmr
w1    wm
r t
m1 ;:::;mr D0
m 1 Š    m r Š

Y
r
.wi t/e ai wi t t r e at
D D .w1    wr / Qr ;
i D1
e wi t  1 i D1 .e
wi t  1/

so we get the final equality of the proposition. t

u

13.2 The Double Zeta Functions and Dirichlet Series

Now we set w D .w1 ; w2 /. It will be a very interesting problem to study what kind
of zeta functions appear if we cut out the contour integral representation (13.1) of
the Barnes double zeta function .s; w; a/, in the same way as we did when we
obtained the functional equation of the Hurwitz zeta functions (Theorem 9.4).
In order to answer this problem, we introduce a new Dirichlet series. For each
real algebraic irrational number 5 ˛, let us consider the following Dirichlet series:

5
A real algebraic number which is not a rational number.
13.2 The Double Zeta Functions and Dirichlet Series 213

X1
cot  n˛
.s; ˛/ D :
nD1
ns

Since cot.x/ is unbounded, we need to show that this converges when Re.s/ is big
enough.
By virtue of Roth’s theorem on Diophantine approximation in transcendental
number theory, the following things are known. Assume that ˛ is an algebraic
irrational number. When we take any positive number ", there exists a positive
constant C."/ such that
ˇ m ˇˇ
ˇ
ˇ˛  ˇ > C."/n2" (13.2)
n
for arbitrary integers m; n (n > 0). Now for a real number x, we use the notation
hhxii for the distance to the nearest integer from x. We have 0  hhxii  1=2.
Using this notation, for any integer n, (13.2) is expressed as

hhn˛ii > C."/n1" :

2
When 0 < x < =2, the inequality sin x > x holds, so

1 1
j cot  n˛j  <
sin hhn˛ii 2hhn˛ii

1
< n1C" :
2C."/

By this estimate, if we put  D Re.s/, we can take

1
1 X n1C"
2C."/ nD1 n

as a majorant series of .s; ˛/, and it converges for  > 2 C ". Since we can take
arbitrary ", .s; ˛/ converges for6 Re.s/ > 2, and moreover it converges uniformly
on compact sets, so it is a holomorphic function of s in that range.
We denote by SL2 .Z/the modular group which appeared before (p. 76). If we
ab
let an element V D of SL2 .Z/ act on a real algebraic irrational number ˛ by
cd

6
In fact, if we use Roth’s theorem more efficiently, we can show absolute convergence for Re.s/ >
1 (see [6]).
214 13 The Barnes Multiple Zeta Function

a˛ C b
V˛ D ;
c˛ C d

then V ˛ is again a real algebraic irrational number.

Let ! and a be positive numbers and consider the following double zeta function:
1
X
2 .s; !; a/ D .m C n! C a/s :
m; nD0

In the former notation, we have 2 .s; !; a/ D .s; .1; !/; a/. We put

.2/s
.s/ D :
.s/ sin.s=2/

The zeta function 2 .s; !; a/ converges absolutely for Re.s/ > 2, is holomorphic
there, and is continued analytically to a meromorphic function on the whole s-plane,
and its poles are s D 1; 2 and of order 1.
Using the contour integral representation (13.1), the principal part of the Laurent
expansion at s D 1 is given by
 
1C! a 1
2 .s; !; a/ D  C  : (13.3)
2! ! s1

Proposition
 13.2.
 Let ˛ > 0 be a real algebraic irrational number. We put
n 1
V D .n 2 Z/. Then for the range Re.s/ > 2, we get the following
1 0
transformation formula:

˛ s1 .s; V ˛/  .s; ˛/ D .1  ˛ s1 / cot.s=2/.s/  .s/2 .1  s; ˛; 1/: (13.4)

Proof. We start from the contour integral representation of the double zeta function
2 .s; ˛; 1/ given by
Z
1 e ˛t
2 .s; ˛; 1/ D t s1 dt: (13.5)
.s/.e 2 i s  1/ I.";1/ .e t  1/.e ˛t  1/

This expression is obtained if we put r D 2; w1 D 1; w2 D ˛; a D 1 in (13.1).

Since the right-hand side is continued analytically to a meromorphic function on the
whole s-plane, we consider the range Re.s/ < 1. We take t0 < 0 and let C.t0 / be
the vertical line Re.t/ D t0 , oriented upwards:

C.t0 / W t D t0 C iy .1 < y < 1/:

13.2 The Double Zeta Functions and Dirichlet Series 215

In the same way as in the proof of Theorem 9.4, shifting the contour I."; 1/ to
the left to C.t0 / and using the residue theorem, we can write
 Z
1 e ˛t
2 .s; ˛; 1/ D  t s1 dt
.s/.e 2 i s  1/ C.t0 / .e t  1/.e ˛t  1/

!
X .2 i n/s1 .2 i n=˛/s1 1
2 i 2
C 2 i n=˛  ;
1e i n˛ e 1 ˛
n2Zf0g

and then taking the limit t0 ! 1 and noting

1 1 1 1 1 n 1
2
D cot  n˛ C and D cot  ;
1e i n˛ 2i 2 e 2 i n=˛ 1 2i ˛ 2
we have

2 .s; ˛; 1/
   
2 i.2/s1 e  i s=2 .1 C e  i s / s 1
D  .1  s; ˛/ C ˛  1  s;
.s/.e 2 i s  1/ 2 ˛

e  i s=2 .1  e  i s / s
C .1  ˛ /.1  s/ :
2i

The transformation up to here is valid for Re.s/ < 1. Substituting 1  s for s and
using .s/ .1  s/ D = sin s, in the range Re.s/ > 2 we have

2 .1  s; ˛; 1/
   
s e  i s=2 .1  e  i s / 1
D .2/ .s/ sin s .s; ˛/ C ˛  s;
s1
1  e 2 i s ˛

i e  i s=2 .1 C e  i s /
C .1  ˛ s1 /.s/
1  e 2 i s

  
1 1
D .s; ˛/  ˛  s; 
s1
C .1  ˛ / cot.s=2/.s/ ;
s1
.s/ ˛

and simplifying this, we have

 
1
˛ s1  s;   .s; ˛/ D .1  ˛ s1 / cot.s=2/.s/  .s/2 .1  s; ˛; 1/:
˛
216 13 The Barnes Multiple Zeta Function

Therefore, noting .˛ C n; s/ D .˛; s/, we obtain the assertion of the theorem.
t
u
The transformation formula (13.4) can be extended to the following form. For
x 2 R, we denote by fxg the real number which satisfies x  fxg 2 Z, 0  fxg < 1
(fractional part of x).
 
ab
Theorem 13.3. For a real algebraic irrational number ˛ and V D 2
cd
SL2 .Z/ which satisfies c > 0 and c˛Cd > 0, the following transformation formula
holds in the range Re.s/ > 2.

s1 .s; V ˛/  .s; ˛/

X
D .1  s1 / cot.s=2/.s/  .s/ 2 .1  s; ; xj C yj /:
j mod c

Here we put  D c˛ C d and j runs over integers which represent residue classes
modulo c, and for each j mod c, we put xj D 1  fjd=cg, yj D fj=cg.
Proof. See [6] and 
[7]. 
n 1
If we put V D in this theorem, we get the transformation formula in
1 0
Proposition 13.2. t
u
From now on, we always
 assume that ˛ is a real quadratic irrational number.
ab
Then there exists V D 2 SL2 .Z/ which satisfies the conditions
cd

V ˛ D ˛; c˛ C d > 0: (13.6)

This is easily understood if we think as follows. Let F be the real quadratic field
Q.˛/. Let L D Z˛ C Z be a lattice generated by ˛ and 1 and O.L/ be an order
associated to L, that is, O.L/ D fu 2 F j uL Lg. We denote by E.L/ the group
of all totally positive units of O.L/. We take an element  of E.L/ such that  ¤ 1.
Since  2 L and ˛ 2 L, if we define a 2  2 matrix V with coefficients in Z by
     
˛ ˛ ab
 DV ; V D ;
1 1 cd

then we have V 2 SL2 .Z/ and  D c˛ C d > 0. In order to make c > 0, it is

sufficient to take  so that  > 1 if ˛ > ˛ (˛ is the conjugate of ˛), and to take  so
that 0 <  < 1 if ˛ < ˛.
Applying Theorem 13.3 to this situation, we get the following theorem.
Theorem 13.4. When a real quadratic irrational number ˛ is given, we take a
totally positive unit  of F such that
13.2 The Double Zeta Functions and Dirichlet Series 217

     
˛ ˛ ab
 DV ; V D 2 SL2 .Z/
1 1 cd

with c > 0. Then in the range Re.s/ > 2, we have

.s/ X
.s; ˛/ D  cot.s=2/.s/ C 2 .1  s; ; xj C yj /: (13.7)
1 s1
j mod c

Through this expression, .s; ˛/ is analytically continued to a meromorphic function

on the whole s-plane, and the poles are s D 1 and s D 1 C 2 log  .n 2 Z; n ¤ 0/
in

and they are of order 1. Moreover, its values at s D 2m  1 .m 2 Z; m  2/ are

expressed using Bernoulli polynomials:7

.1/m1 .2/2m1 X X Bk .xj /B2mk .yj / 2mk1

2m
.2m  1; ˛/ D  :
1  2m2 j mod c
kŠ.2m  k/Š
kD0

Proof. The expression (13.7) is immediately obtained by using V ˛ D ˛ in

Theorem 13.3. Poles can possibly appear only at s D 2, s D 1, s D 1 C 2 log  .n 2
in

Z; n ¤ 0/, and s D 0. For s D 2; 4; 6; : : : we have .s/ D 0, so we note that
these are not poles. As for s D 2, as we remarked in the footnote on p. 213, .s; ˛/
converges absolutely for Re.s/ > 1 and is holomorphic there, so this cannot be a
pole. Now we consider s D 0. Taking the result that the principal part of 2 .s; !; a/
at the pole s D 1 is given by (13.3) and the fact .0/ D  12 into consideration, the
residue of the function .s; ˛/ at s D 0 (possible pole of order at most 1) is given by

X  
1 2 1 C  xj C yj 
 1
 :
 .1   / j mod c 2 

We see easily that this value is zero. So .s; ˛/ is holomorphic at s D 0.

Lastly, the values at s D 2m1 (m 2 Z; m  2) are, by virtue of (13.7), given by

.1/m1 .2/2m1 X
.2m  1; ˛/ D 2 .2  2m; ; xj C yj /
.1  2m2 /.2m  2/Š j mod c

and by Proposition 13.1, we get the expression using Bernoulli polynomials. t

u

7
When ˛ is a unit of a real quadratic field, this kind of special value has been calculated by Berndt
[14].
218 13 The Barnes Multiple Zeta Function

13.3 .s; ˛/ and Continued Fractions

As explained in detail in Zetafunktionen und quadratische Körper by D. Zagier [107,

§13], any quadratic irrational number ˇ can be expanded into a continued fraction8
of the following form.

1
ˇ D b1  .bj 2 Z; b2 ; b3 ; : : :  2/:
1
b2 
1
b3 
b4  :
::

It is important that all the coefficients bj (j  2) except for the first one are integers
bigger than or equal to 2. Here the sequence fb1 ; b2 ; : : :g is periodic, namely, there
exist some natural numbers  and r such that for j not less than , we have

brCj D bj .j  /:

We call the smallest such natural number r the period of ˇ. We write this continued
fraction expansion as

ˇ D ŒŒ b1 ; b2 ; : : : ; b1 ; b ; : : : ; bCr1 : (13.8)

The sequence fb ; : : : ; bCr1 g is called a fundamental period. Moreover when

 D 1, that is, if it is periodic from the first term, we say that the continued fraction
expansion of ˇ is purely periodic.
For a real quadratic irrational number ˇ, We denote by ˇ the conjugate number
of ˇ. Now we assume that the continued fraction expansion of a real quadratic
irrational number ! is purely periodic and we write

! D ŒŒ n1 ; n2 ; : : : ; nr  .nj  2/:

Then we have
1
D ŒŒ nr ; nr1 ; : : : ; n1 
!
[107, p. 136, (17)]. By this expression, we have

!>1 and 0 < ! < 1: (13.9)

8
The continued fractions treated here are different from the usual continued fractions of real
quadratic irrational numbers.
13.3 .s; ˛/ and Continued Fractions 219

If a real quadratic irrational number ! satisfies condition (13.9), we say that ! is

reduced.9 Conversely, if ! satisfies this condition, we can show that the continued
fraction expansion of ! is periodic, so ! is reduced if and only if the continued
fraction expansion of ! is purely periodic.
Let F be a real quadratic field and regard F as a subfield of the real numbers R.
Moreover we define a mapping of F into R  R by

F ./ D .; / 2 R  R;

and identify  with the point .; / of R  R. Now we assume that ! 2 F  Q is

reduced and that the continued fraction expansion of ! is given by

! D ŒŒ n1 ; n2 ; : : : ; nr :

We define an integer sequence fnj gj 2Z by extending the continued fraction

expansion fn1 ; n2 ; : : : ; nr g by periodicity so that nj Cr D nj for any j 2 Z. For
a positive integer i we define a reduced number !i by

!i D ŒŒ ni ; ni C1 ; : : : ; ni Cr1 : (13.10)

Of course we have !1 D !rC1 D !. For n 2 Z, we put

 
n 1
V .n/ D 2 SL2 .Z/:
1 0

From the continued fraction expansion (13.10), we get

1
!i D ni  D V .ni /!i C1 ;
!i C1

so we have

! D V .n1 /V .n2 /    V .nr /!:

We define now a lattice L of F by

L D Z! C Z

and take a generator " of the group E.L/ of totally positive units of O.L/ so that
" > 1. We define V by the relation

9
This is different from the usual definition of reduced for the setting of ordinary continued
fractions. Usually a real quadratic irrational number ˛ is called reduced when ˛ > 1 and
0 > ˛ > 1.
220 13 The Barnes Multiple Zeta Function

     
! ! ab
" DV ; V D 2 SL2 .Z/:
1 1 cd

If we think of the stabilizer subgroup ! D fM 2 SL2 .Z/ j M! D !g of !, we

see ! is generated by V and ˙I2 (I2 is the identity matrix). Because c > 0 and r
is the period of !, we have

V D V .n1 /V .n2 /    V .nr /:

For any V0 2 SL2 .Z/ and a real algebraic irrational number ˛, we use the notation
of an automorphy factor

J.V0 ; ˛/ D c˛ C d: (13.11)

As in the case when ˛ is in the complex upper half plane, this has a property
J.V1 V2 ; ˛/ D J.V1 ; V2 ˛/J.V2 ; ˛/ for any V1 , V2 2 SL2 .Z/. Then using this
property of the automorphy factor (13.11), we have

" D J.V; !/ D J.V .n1 /; !2 /J.V .n2 /; !3 /    J.V .nr /; !/

D !2 !3    !r !:

So we have

" D !1 !2    !r : (13.12)

Now, we put 0 D 1, 1 D ! D !rC1 , 2 D !rC1 !r , : : : , r D !rC1 !r    !2 . Of

course we have r D ". Here we have

L D Zi C Zi C1 .0  i  r  1/; (13.13)

because by the relation i 1 D i  1

!rC2i
we have
 
1
Zi 1 C Zi D i Z CZ
!rC2i
 
D i Z.nrC1i  !rC1i / C Z

D Zi C Zi C1

and starting from i D 1 inductively we can show the assertion.

Now we define a double zeta function of a lattice L by
13.3 .s; ˛/ and Continued Fractions 221

X
L .s/ D  s : (13.14)
T
 2L C.1;"/

Here we denote by C.1; "/ the simplicial cone given by

C.1; "/ D f.x C y"; x C y"0 / j x > 0; y  0g:

As we shall see later, L .s/ is written as a finite sum of double zeta functions
explained before, so it converges absolutely for Re.s/ > 2 and is holomorphic in
that range. Let RC be the set of positive numbers. The group of totally positive units
E.L/ acts on RC  RC by

.u; v/ D .u; v/ . 2 E.L/; .u; v/ 2 RC  RC /:

The set C.1; "/ is a fundamental domain of RC  RC by E.L/.

Since !i > 1 for each i and since we see 1 < i < i C1 <
and
< i C1 <
i < 1, we can write

[
r
C.1; "/ D fxk1 C yk j x > 0; y  0g .disjoint union/;
kD1

so by (13.13) we get

[
r
L \ C.1; "/ D fmk1 C nk j m 2 Z>0 ; n 2 Z0 g .disjoint union/: (13.15)
kD1

This decomposition was given in Zagier [105]. Here we denote by Z>0 or Z0 the set
of positive integers, or integers not less than 0, respectively. By the decomposition
(13.15), we get

X
r
L .s/ D .k1 /s 2 .s; !rC2k ; 1/: (13.16)
kD1

By this expression, L .s/ converges absolutely for Re.s/ > 2 and is continued
analytically to a meromorphic function which is holomorphic in the whole s-plane
except for s D 1; 2.
Between the functions .s; !/ and L .s/, the following functional equation
holds.
Theorem 13.5. Let ! be a reduced number of F . Put L D Z C Z! and let " > 1
be a totally positive fundamental unit of the order O.L/ of L. Then we have

.s/
.s; !/ D  cot.s=2/.s/  L .1  s/: (13.17)
"s11
222 13 The Barnes Multiple Zeta Function

We can also show through this equation that .s; !/ can be continued analytically
to a meromorphic function of s.
Proof. Since we have now !i D V .ni /!i C1 and !i > 0, by Proposition 13.2, we
have an equality

.!rC2k /s1 .s; !rC1k /  .s; !rC2k /

 
D 1  .!rC2k /s1 cot.s=2/.s/  .s/2 .1  s; !rC2k ; 1/

for 1  k  r. Multiplying .k1 /s1 by both sides of this equality, then taking the
sum from k D 1 to k D r and using (13.12) and (13.16), we get

."s1  1/.s; !/ D .1  "s1 / cot.s=2/.s/  .s/L .1  s/:

t
u
Remark 13.6. If we cut out the contour integral representation, not only of a double
zeta function, but also of a multiple zeta function in the same way as in Sect. 13.2,
then a kind of Lerch type zeta function appears. It would be an interesting problem
to investigate what kind of modular properties it has.
Chapter 14
Poly-Bernoulli Numbers

In this chapter, we define and study a generalization of Bernoulli numbers referred to

as poly-Bernoulli numbers, which is a different generalization than the generalized
Bernoulli numbers introduced in Chap. 4.

14.1 Poly-Bernoulli Numbers

We give a definition in a down-to-earth way.

.k/
Definition 14.1 (Poly-Bernoulli number [56]). Poly-Bernoulli numbers Bn 2
Q .n  0; k 2 Z/ are defined by the generating function

X 1
Lik .1  e t / .k/ t
n
D B :
1  e t nD0
n
nŠ

X1
tn
Here, for any integer k, Lik .t/ stands for the formal power series .
nD1
nk

When k  1, the series Lik .t/ is the defining series of a “polylogarithm”

function, whereas when k  0, it is the Taylor series of the rational function
 d k  t 
t dt 1t at the origin. If k D 1, we have Li1 .t/ D  log.1  t/ and
Li1 .1  e t / D t, hence by Theorem 1.12 on p. 20, the number Bn is nothing
.1/

but the original Bernoulli number Bn . If k  1, by making a change of variables

ti 7! 1  e ti in the following iterated integral expression of Lik .t/,
Z t Z tk Z t3 Z t2
dtk dtk1 dt2 dt1
Lik .t/ D  ;
0 tk 0 tk1 0 t2 0 1  t1

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 223
in Mathematics, DOI 10.1007/978-4-431-54919-2__14, © Springer Japan 2014
224 14 Poly-Bernoulli Numbers

.k/
Table 14.1 Bn .0  k  5; 0  n  7/
k\n 0 1 2 3 4 5 6 7
0 1 1 1 1 1 1 1 1
1 1 1
2
1
6
0  30
1
0 1
42
0
2 1 1
4
 36
1
 24
1 7
450
1
40
 2205
38
 168
5

3 1 1
8
 216
11
 288
1 1243
54000
 7200
49
 3704400
75613 599
35280
4 1 1
16
 1296
49 41
3456
26291
3240000
 144000
1921 845233
1555848000
1048349
59270400
5 1 1
32
 7776
179 515
41472
 194400000
216383
 25920000
183781 4644828197
653456160000
153375307
49787136000

we have the following expression of the generating function:

Lik .1  e t /
D
1  e t
Z t Z tk Z tk1 Z t3 Z t2
1 dtk dtk1 dt2
et  t    dt1 : (14.1)
e  1 0 e tk  1 0 e tk1  1 0 0 e 1 0
t2

P
This shows that, in the case of k  1, the generating function 1
.k/ t n
nD0 Bn nŠ of
.k/
Rpoly-Bernoulli numbers Bn is obtained, starting with 1, by applying the integration
t
0 dt and division by e  1 successively k times, and finally by multiplying e .
t t

Remark 14.2. (1) Our poly-Bernoulli number is different from the “Bernoulli
numbers of higher order” defined by Nörlund [75, Chapter 6]. The generating
 t k
function of the Bernoulli numbers of higher order is et 1 and this can be
t
written as a linear combination of derivatives of et 1 . Hence the Bernoulli
number of higher order is essentially a linear combination of classical Bernoulli
.k/
numbers. As is shown below, our Bn can also be expressed in terms of classical
Bernoulli numbers, but we need to take products successively.
(2) There is a very interesting object called “multiple zeta values” (cf. [57,98,109]),
which generalize the values at positive integer arguments of the Riemann zeta
function. The poly-Bernoulli numbers are in some way related to the multiple
zeta values [9].
.k/
Table 14.1 gives some values of Bn for k  0.
.k/
The next two propositions give recurrence formulas for Bn , which can be
derived from the definition and the iterated integral expression of the generating
function.
Proposition 14.3. For any k 2 Z and n  0, we have
 !
1 X
n1
n
B.k/ D Bn
.k1/
 B.k/
:
n
nC1 mD1
m1 m

We regard the sum on the right as 0 if n  1.

14.1 Poly-Bernoulli Numbers 225

1
!
X t n
t t
Proof. We differentiate the relation Lik .1  e / D .1  e / which B.k/
n
nŠ nD0
is obtained from the definition by multiplying by 1  e t . By the relation Lik0 .t/ D
1
Lik1 .t/, we have
t

X1 X1
Lik1 .1  e t / t t .k/ t
n
t .k/ t
n1
e D e B C .1  e / B :
1  e t nD0
n
nŠ nD1
n
.n  1/Š

X 1
Lik1 .1  e t / tn
Multiplying by e t on both sides and using t
D Bn.k1/ , we have
1e nD0
nŠ

1
X X1 X1
tn tn .k/ t
n1
Bn.k1/ D B.k/ C .e t
 1/ B
nD0
nŠ nD0
n
nŠ nD1
n
.n  1/Š
1
X 1 1
t n X t l X .k/ t n1
D B.k/ C B
nD0
n
nŠ lŠ nD1 n .n  1/Š
lD1
1 1
!
X t n X X
n1
1 .k/
D C
B.k/ BmC1
tn
nD0
nŠ n
nD1 mD0
.n  m/ŠmŠ
1
! !
.k/
X X
n1
n .k/ tn
D B0 C n C
B.k/ BmC1 :
nD1 mD0
m nŠ

.k1/ .k/
Hence we have B0 D B0 if n D 0, and if n  1 we obtain
! !
X
n1
n .k/ X
n1
n
Bn.k1/ D B.k/ C B D .n C 1/Bn C
.k/
B.k/
m :
n
mD0
m mC1 mD1
m  1

This divided by n C 1 gives the proposition. t

u
Proposition 14.4. For k  1 and n  0, we have
! ! !
X
n
n .k1/ X .1/l
m
m
B.k/ D .1/ m
B Bl :
n
mD0
m nm nl C1 l
lD0

Proof. We use the iterated integral expression of the generating function, which
shows
Z t
Lik .1  e t / et Lik1 .1  e s /
t
D e s ds:
1e e 1 0
t 1  e s
226 14 Poly-Bernoulli Numbers

From this we have

1 1
!Z 1
! 1
!
X t n X t n1 t X .s/n X s n
B.k/
n D Bn Bn.k1/ ds
nD0
nŠ nD0
nŠ 0 nD0
nŠ nD0
nŠ
1
!Z 1
! !
X t n1 t X X
n
n .k1/ s n
D Bn .1/ B nm
ds
nD0
nŠ 0 nD0 mD0 m m nŠ
1
! 1 ! !
X t n1 X X
n
t nC1
nm n
D Bn .1/ Bm
.k1/
ds
nD0
nŠ nD0 mD0 m .n C 1/Š
1
! l ! !!
X Xn
Bnl n X tn
lm l
D .1/ Bm
.k1/

nD0 lD0
l C1 l mD0
m nŠ
1
! ! !!
X Xn Xn
.1/l n l tn
D .1/m Bm.k1/
Bnl :
nD0 mD0
l C1 l m nŠ
lDm

n l   n nm
Use l m D m nl and replace l ! n  l and then m ! n  m to obtain
! ! !
X
n X n
.1/l n l
.1/m Bm
.k1/
Bnl
mD0
l C1 l m
lDm
! ! !
X
n
n .k1/ X .1/l
m
m
D .1/m B Bl ;
mD0
m nm nl C1 l
lD0

which gives the proposition. t

u
Remark 14.5. In the above two recurrences, we need the numbers with different
.k1/ .k/
upper index like Bn in order to express Bn . We do not know if there exists a
recurrence formula with the single index k.
The following theorem generalizes Theorem 2.8 (p. 35), which expresses
Bernoulli numbers by Stirling numbers.
Theorem 14.6.

Xn
.1/m mŠ mn
B.k/ D .1/ n
; .n  0; k 2 Z/:
n
mD0
.m C 1/k

Proof. The proof goes in the same way as Theorem 2.8. In fact, by using
Proposition 2.6 (7) (p. 30) we have
14.2 Theorem of Clausen and von Staudt Type 227

1
X X1
tn Lik .1  e t / .1  e t /m1
B.k/ D t
D
nD0
n
nŠ 1e mD1
mk

X1 1
X 1

.1/m .e t  1/m .1/m mŠ X n .t/n
D D
mD0
.m C 1/k mD0
.m C 1/k nDm m nŠ
1  !
X Xn
.1/m mŠ mn tn
D .1/ n
:
nD0 mD0
.m C 1/k nŠ

tn
Comparing coefficients of nŠ
on both sides, we obtain the theorem. t
u

14.2 Theorem of Clausen and von Staudt Type

In this section we discuss denominators of poly-Bernoulli numbers. The denomina-

tors of the classical Bernoulli numbers are completely determined by the theorem
of Clausen and von Staudt (Theorem 3.1, p. 41). The theorem not only determines
the denominator but also describes the “fractional part” of Bn . In the case of general
.k/
Bn we only have partial results below, but when k D 2 (“di-Bernoulli numbers”),
the denominators are completely determined. Note that, by Theorem 14.6, no prime
.k/
greater than n C 1 appears in the denominator of Bn .
Theorem 14.7. Let k  2 be an integer and p be a prime number satisfying kC2 
p  n C 1.
.k/
(1) When p  1 j n, p k Bn is in Z.p/ .the set of rational numbers whose denomina-
tors are prime to p/ and

p k B.k/
n  1 mod p:

.k/
(2) When p  1 − n, p k1 Bn is in Z.p/ and
⎧

⎪
⎪ 1 n n
⎨ p p  1  2k mod p
⎪ if n  1 mod .p  1/;
p k1 B.k/
n 

⎪
⎪ .1/n1 n
⎪
⎩ mod p otherwise:
p p1

Proof. Denote by ordp .a/ the p-order of a rational number a. We have

ordp .p t / D t and both numerator and denominator of a  p ordp .a/ are prime
to p. We proceed in the same way as in the proof of Theorem 3.1, using the formula
in Theorem 14.6 and calculating the p-order of each term in the formula. Put
.k/ .1/m mŠf n g
bn .m/ D .mC1/km . We prove (1) and (2) simultaneously.
228 14 Poly-Bernoulli Numbers

.k/
Write m C 1 D ap e ; .a; p/ D 1; e  0. If e D 0, then bn .m/ 2 Z.p/
.k/
and p k1 bn .m/  0 mod p by the assumption  k  2, hence this term can be
discarded. Note that, since the Stirling number mn is an integer, we have
 
mŠ
ordp .bn.k/.m//  ordp :
.m C 1/k

.k/
First assume e  2. We show the congruence p k bn .m/  0 mod p and further
k1 .k/
P1 h m i
p bn .m/  0 mod p if p 1 − n. Using ordp .mŠ/ D j D1 pj , we estimate

  1 
X
mŠ m
ordp D  ek
.m C 1/k j D1
pj
  e
m ap  1
  ek D  ek D ap e1  1  ek
p p
 p e1  1  ek D .1 C p  1/e1  1  ek
 1 C .e  1/.p  1/  1  ek D .e  1/.p  1/  ek
 .e  1/.k C 1/  ek D k C e  1
 k C 1:
 
.k/
mŠ
From this we obtain ordp .mC1/ k  k C 1 and thus p k1 bn .m/ 2 Z.p/ .
.k/
Therefore p k bn .m/  0 mod p holds. If at least one inequality in the above
.k/
chain of inequalities is strict, we get p k1 bn .m/  0 mod p. If all the inequalities
become equalities, we have e D 2; m C 1 D p 2 , and p D k C 2. In this case, the
.k/
lemma below (the case a D p) shows p k1 bn .m/  0 mod p if p  1 − n.
Lemma 14.8. Let n and a be natural numbers. We have the congruence
⎧ !

⎪
⎪ c  1
n ⎨ mod p if n D a  1 C c.p  1/ for some c  a;
 a1
ap  1 ⎪
⎪
⎩
0 mod p otherwise:
n
We use the generating function of the Stirling numbers m (Proposition 2.6 (8)
on p. 30)

X1

n n tm
t D : (14.2)
nDm
m .1  t/ .1  2t/    .1  mt/

When m D ap  1, the right-hand side reduces modulo p to

14.2 Theorem of Clausen and von Staudt Type 229

1
! 1
!
t ap1 X a C i  1 i.p1/ X a C i  1 a1C.aCi /.p1/
Dt ap1
t D t :
.1  t p1 /a i D0
i i D0
a1

Here we used .1t/.12t/    .1.p1/t/  1t p1 mod p. Putting aCi D c,

we obtain the lemma. t
u
Next
 suppose
 e D 1 .m D ap  1/. If a  3, then p 2
j.ap  1/Š ensures
.k/
mŠ
ordp .mC1/ k > k C 1 and this gives p k1 bn .m/  0 mod p. If a D 2,
 
.k/
we have ordp .mŠ/ D 1 and ordp .mC1/mŠ
k D 1  k, and hence ordp .bn .m// D
  .k/
1  k C ordp mn . From this, we have p k bn .m/  0  mod p: If n 6 1
n
mod .p  1/, by putting a D 2 in the lemma above we have 2p1  0 mod p.
.k/
This gives us p k1 bn .m/  0 mod p if n 6 1 mod .p  1/. If n  1
mod .p  1/, by writing n D 1 C c.p  1/, we have  n c  2 (note c D 1 never
happens because n  m) and the lemma again gives 2p1  c  1  n mod p.
Hence we have
 n
.1/2p1 .2p  1/Š 2p1
p k1 bn.k/ .m/ D p k1
.2p/k
n
 k mod p:
2

(We have used Wilson’s1 theorem .p  1/Š  1 mod p, which implies

.2p  1/Š=p  ..p  1/Š/2  1 mod p.) Lastly, when a D 1 (m D p  1),
.k/ .p1/Šfp1
n
g  n
we have bn .m/ D pk
. By the lemma we have p1  0 mod p if n 6 0
k1 .k/
 n
mod .p  1/ and hence p bn .m/   p p1 mod p. If n  0 mod .p  1/,
1
 n .k/
then we have p1  1 mod p and p k bn .m/  1 mod p: Combining all
these, we obtain the theorem. t
u
Remark 14.9. Let k and p be as in the theorem, and assume n 6 0; 1 mod .p 1/.
Denote by n0 the unique integer satisfying n0  n mod p  1 and 1 < n0 < p.
Then we have

0 Bn
p k1 B.k/
n  .n  n / mod p:
n
.k/
If in particular n is odd, we have p k1 Bn  0 mod p.
This comes from the congruence

1
John Wilson (born on August 6, 1741 in Applethwaite, England—died on October 18, 1793 in
Kendal, England).
230 14 Poly-Bernoulli Numbers


.1/n1 n Bn
 .n  n0 / mod p;
p p1 n

which is a consequence of Eq. (14.3) in the proof of Theorem 14.10.

The denominators of di-Bernoulli numbers are completely determined.
.2/ .n  2/
Theorem 14.10. (1) If n is odd, we have Bn D  Bn1 : .Hence in
4
this case the determination of the denominator is essentially reduced to the
theorem of Clausen and von Staudt (Theorem 3.1).)
(2) Suppose n is even . 2/. For a prime p, denote by ord.p; n/ the p-order
.2/
of Bn . Then we have the following:
(2-1) If p > n C 1, then ord.p; n/  0 .no p appears in the denominator/.
(2-2) For p with 5  p  n C 1,
(a) ord.p; n/ D 2 if p  1 j n.
(b) Suppose p  1 − n.
Bn
(b-1) We have ord.p; n/  0 if either p j or n  n0 mod p.p  1/ for some
n
n0 with 1 < n0 < p  1, and
(b-2) ord.p; n/ D 1 otherwise.
(2-3) We have ord.3; n/  0 if n > 2 and n  2 mod 3, and ord.3; n/ D 2
otherwise.
(2-4) We have ord.2; n/  0 if n > 2 and n  2 mod 4, ord.2; n/ D 1 if n  0
mod 4, and ord.2; 2/ D 2.
Bn
Remark 14.11. The rational number in (b-1) is always an element of Z.p/
n
(Theorem 3.2 (1)).
For the proof, we need a lemma.
Lemma 14.12. Let n  2 be even, p  5 a prime, and 2p  1 D m. We have


n
.1/m mŠ 0 mod p 2 ;
m

.1/m mŠ mn
and hence 2 Z.p/ :
.m C 1/2
Proof. By Proposition 2.6 (p. 30), we have

 2p1 !
X
n l 2p  1
.1/ mŠ m
D .1/ ln
m l
lD1
 ! ! !
X
p1
2p  1 n 2pl 2p  1 p 2p  1
D .1/l
l C.1/ .2p  l/ C.1/
n
pn
l 2p  l p
lD1
14.2 Theorem of Clausen and von Staudt Type 231

 ! !
X
p1
2p  1 n l 2p  1
 .1/ l
l C.1/ .2npl n1
Cl /
n
mod p 2 :
l l 1
lD1

2p1 2p1 2p 2p1

Using l
C l1
D l l1
, we see the last sum is equal to
!
X
p1
2p  1 n1
2p.1  n/ .1/ l
l :
l 1
lD1

Combining
!
2p  1
 .1/l1 mod p
l 1

and p  1 − n  1 (since n is even and p is odd), we have

!
X
p1
2p  1 X
p1
.1/ l
l n1
 l n1  0 mod p:
l 1
lD1 lD1

t
u

Proof of Theorem 14.10. The iterated integral expression (14.1) of the generating
P
function of Bn and es s1 D 1 sl
.2/ l
lD0 .1/ Bl lŠ give

1
X Z tX
1
tn et sl
B.2/ D t .1/l Bl ds
nD0
n
nŠ e 1 0 lD0 lŠ
1
X 1
t m1 X t lC1
D Bm  .1/l Bl :
mD0
mŠ .l C 1/Š
lD0

From this we obtain

!
X
n
n Bnl Bl
B.2/ D .1/ l
:
n
l l C1
lD0

If n is odd, one of n  l and l is odd. Since Bl D 0 for odd l  3, we have for n  3

n .n  2/
n D  Bn1 B1 C B1 Bn1 D 
B.2/ Bn1 :
2 4
This also holds for n D 1, and thus (1) is proved.
232 14 Poly-Bernoulli Numbers

(2-1) is clear from the formula in Theorem 14.6. By elementary computation, one
2 in that formula is not an integer are m C 1 D
mŠ
sees that the only cases where .mC1/
8; 9; prime; 2  prime. By Lemma 14.12, one sees that p does not appear in the
.1/m mŠf n g
denominator of .mC1/2m if m C 1 D 2p; (p: prime  5). Next consider the case
of m C 1 D p (p: prime  5). In this case, we have

 X !
p1
X
p1
n l p1
.1/ mŠm
D .1/ ln  ln mod p:
m l
lD1 lD1

This reduces mod p to 1 if p  1 j n, 0 if p  1 − n. Hence the p-order of

.1/m mŠfm
n
g
.mC1/2
is 2 if p  1 j n. The other terms being in Z.p/ , this establishes (2-2)-
(a). Suppose p  1 − n. Noting the congruence
!
p1 X l
1
 .1/l C .1/l1 p mod p 2
l i D1
i

(expand .p  1/.p  2/    .p  l/), we have

!
X
p1
X
p1
X
p1
X l
l p1 1
.1/ ln  ln  p ln mod p 2 :
l i D1
i
lD1 lD1 lD1

It is known that if n is even and p  1 − n, the congruence

X
p1
l n  pBn mod p 2
lD1

holds (cf. Ireland and Rosen [50, Cor. of Prop. 15.2.2]). On the other hand, if n  n0
mod p  1; 1 < n0 < p  1, then n0 is also even and by Vandiver2 [93, (63)] (see
also Remark 14.24) we have

X
p1
X
l
1 X
p1
0
X
l
1
ln  ln mod p
i D1
i i D1
i
lD1 lD1

 Bn0 mod p:

2
Harry Schultz Vandiver (born on October 21, 1882 in Philadelphia, USA, died on January 9, 1973
in Austin, USA).
14.3 Poly-Bernoulli Numbers with Negative Upper Indices 233

This gives


n
.1/m mŠ  p .Bn  Bn0 / mod p 2 :
m

By the assumption p  1 − n, we have Bnn 2 Z.p/ and Bn0  n0 Bnn mod p

(Theorem 3.2). We therefore have (m D p  1)


n Bn
.1/m mŠ  p.n  n0 / mod p 2 ; (14.3)
m n

and (2-2)-(b) follows. As for the 3-order, the terms in the formula of Theorem 14.6
which have 3 in the denominators are
  
2Š n2 5Š n5 8Š n8
; ; :
32 62 92
Computation using Proposition 2.6 (6) (p. 30) gives (2-3). The determination of the
2-order is similar and is omitted. t
u
By noting the appearance of the condition p j Bnn in (2-2)-(b-1) of Theorem 14.10,
we can state the irregularity of a prime p in terms of the denominator of the di-
.2/
Bernoulli number Bn .
Corollary 14.13. A prime p . 5/ is irregular if and only if there exists an even n
.2/
with p C 1  n  2p  4 such that the denominator of Bn is not divisible by p.
For classical Bernoulli numbers, odd indexed ones are 0 except for B1 , denomi-
nators of even indexed ones are easily determined (Clausen and von Staudt), and
the numerators of even indexed ones are mysterious quantities that are closely
related to the arithmetic of cyclotomic fields. As for the di-Bernoulli numbers,
the odd indexed ones are essentially the classical Bernoulli numbers and the
determination of denominators of even indexed ones requires the numerators of the
classical Bernoulli numbers. Can one expect any interesting arithmetic concerning
the numerators of the di-Bernoulli numbers? We do not know.

14.3 Poly-Bernoulli Numbers with Negative Upper Indices

.k/
When k is 0 or negative, the poly-Bernoulli number Bn is a positive integer
.k/
(Table 14.2). We give yet another formula for Bn in this case, and introduce
.k/
two different combinatorial interpretations of Bn for non-positive k which were
discovered recently.
We have the following formula.
234 14 Poly-Bernoulli Numbers

.k/
Table 14.2 Bn .5  k  0; 0  n  7/
knn 0 1 2 3 4 5 6 7
0 1 1 1 1 1 1 1 1
1 1 2 4 8 16 32 64 128
2 1 4 14 46 146 454 1394 4246
3 1 8 46 230 1066 4718 20266 85310
4 1 16 146 1066 6902 41506 237686 1315666
5 1 32 454 4718 41506 329462 2441314 17234438

Theorem 14.14. For any n; k  0, we have

X
min.n;k/


nC1 kC1
Bn.k/ D .j Š/ 2
:
j D0
j C1 j C1

.k/
In particular, Bn is positive and symmetric in n and k.
Corollary 14.15.
.n/
Bn.k/ D Bk :
.k/
Proof. 3 We compute the two-variable generating function of Bn with the aid of
Theorem 14.6 and Proposition 2.6 (7), in the following way.
X 1
1 X n X1 X1 Xn

x yk n xn y k
Bn.k/ D .1/ n m
.1/ mŠ .m C 1/k
nD0 kD0
nŠ kŠ nD0 mD0
m nŠ kŠ
kD0
1
X X
n

n .mC1/y x n
D .1/n .1/m mŠ e
nD0 mD0
m nŠ
1
X 1
X
 n
n x
D .1/m mŠe .mC1/y .1/n
mD0 nDm
m nŠ
1
X .e x  1/m
D .1/m mŠe .mC1/y
mD0
mŠ
 
1
D ey
1  .e y .e x  1//
e xCy
D :
ex C e y  e xCy

Furthermore, we have

3
The following proof, which greatly simplifies the original proof in [10], is due to Hiroyuki Ochiai.
The authors would like to thank him for providing this simple proof.
14.3 Poly-Bernoulli Numbers with Negative Upper Indices 235

e xCy e xCy
D
e x C e y  e xCy 1  .e x  1/.e y  1/
1
X
D e xCy .e x  1/j .e y  1/j
j D0

1
X 1 d x d
D .e  1/j C1 .e y  1/j C1 :
j D0
.j C 1/ dx
2 dy

Applying Proposition 2.6 (7) again, we obtain

1 X
X 1
xn y k
B.k/
n
nD0 kD0
nŠ kŠ
0 1 0 1
X1 X1
 nC1 1

X  kC1
d @ n C 1 x A d @ k C 1 y A
D .j Š/2
j D0
dx nDj
j C 1 .n C 1/Š dy j C 1 .k C 1/Š
kDj

X 1
1 X
1 X


nC1 k C 1 xn y k
D .j Š/2 :
j D0 nDj kDj
j C1 j C 1 nŠ kŠ
nC1 kC1
Comparison of the coefficients on both sides gives the theorem (note j C1 j C1
D0
when j > min.n; k/). t
u
.k/
By the theorem, we also have a different generating function for Bn .
Corollary 14.16. We have
X 1
1 X 1
X
Bn.k/ x n y k D pj .x/pj .y/;
nD0 kD0 j D0

where
j Šx j
pj .x/ D :
.1  x/ .1  2x/    .1  .j C 1/x/

Proof. This is a consequence of

1

X 
nC1
pj .x/ D j Š xn
nDj
j C1

which follows from Proposition 2.6 (8). t

u
Computing this generating function in a different manner, we obtain the
following.
236 14 Poly-Bernoulli Numbers

Proposition 14.17. For any n > 0, we have

X
n
.l/
.1/l Bnl D 0:
lD0

Proof. Denote the left-hand side of Corollary 14.16 by B.x; y/. By Theorem 14.6
and Proposition 2.6 (8), we have

X 1
1 X X
n

n
B.x; y/ D .1/n
.1/ mŠ m
.m C 1/k x n y k
nD0 kD0 mD0
m
1
X 1
X

n n 1
D .1/m mŠ .1/n x
mD0 nDm
m 1  .m C 1/y
1
X mŠx m
D :
mD0
.1 C x/ .1 C 2x/    .1 C mx/ .1  .m C 1/y/

(The term with m D 0 is 1y

1
.) Here we put y D x and use Proposition 2.6 (8),
h i
mC1
mŠ D 1 , and Proposition 2.6 (5.1), to obtain B.x; x/ D 1 and the proposition
follows. t
u
Remark 14.18. This is trivial when n is odd, because of the symmetry in
Corollary 14.15. When n is even, however, this is a non-trivial statement. For
example, from Table 14.2, we have 1  2 C 1 D 0; 1  8 C 14  8 C 1 D 0; etc.
In recent years, two different combinatorial interpretations of the poly-Bernoulli
numbers with negative upper indices have been found. We briefly state these
theorems, one by Chad Brewbaker and the other by Stephan Launois. For proofs,
see their papers [20, 68].
Definition 14.19. A matrix whose entries are 0 or 1 is called “lonesum” if it is
uniquely reconstructed from its row and column sums.
Theorem 14.20 (Brewbaker [20]). Let n and k be natural numbers. The number
.k/
of n by k lonesum matrices is equal to Bn .
We denote by Sn the symmetric group of degree n, viewed as a permutation
group on the set f1; 2;    ; ng.
Theorem 14.21 (Launois [68]). Let n and k be natural numbers. The cardinality
of the set

f 2 SnCk j n  i  .i /  k; 1  8i  n C kg

.k/
is equal to Bn .
14.3 Poly-Bernoulli Numbers with Negative Upper Indices 237

We end this section by proving a special case of a theorem of Vandiver, making

use of the formulas in Theorem 14.6 and Corollary 14.15.
Theorem 14.22. For an odd prime p and an integer i with 1  i  p  2, we have

p2  
X 1 1
Bi  1C CC .m C 1/i mod p:
mD1
2 m

Proof. By Theorem 14.6 and Fermat’s little theorem,

.1/ .2p/
Bi D Bi  Bi mod p:

.i /
By Corollary 14.15, the right-hand side is equal to Bp2 . Applying Theorem 14.6
again, we obtain

X
p2

p2
Bi   .1/m mŠ .m C 1/i mod p:
mD0
m

p2
It therefore suffices to show the following lemma (note 0 D 0).
Lemma 14.23. If 1  m  p  2, then


p2 1 1
.1/ m1
mŠ  1C CC mod p:
m 2 m

Write .1/m1 mŠ p2
m
D bm . The recursion of the Stirling numbers (2.1) (p. 26)
gives


p1
.1/ m1
mŠ D m.bm1 C bm / .m  2/:
m
On the other hand, by Proposition 2.6 (6) (p. 30), we have

 !
p1 Xm
l m
.1/ m1
mŠ D .1/ l p1
m l
lD1
!
Xm
l m
 .1/ mod p
l
lD1

 1 mod p

and hence bm  bm1 C m1 mod p:


This together with b1 D p2
1
D 1 proves the lemma and thus settles the
theorem. u
t
238 14 Poly-Bernoulli Numbers

Remark 14.24. For 1 < i  p  2, the right-hand side of the theorem is congruent
modulo p to

p1  
X 1 1
1C CC mi :
mD1
2 m

Vandiver’s original theorem (a special case of [93, (63)]) states that this is congruent
modulo p to Bi .
Exercise 14.25. Prove that if mŠ
.mC1/2
is not an integer, then m C 1 D
8; 9; prime; or 2  prime.
.k/
Exercise 14.26. Define the numbers Cn by the generating function

X 1
Lik .1  e t / .k/ t
n
D C :
et  1 nD0
n
nŠ

(1) Prove the formula

 nC1
X
n
.1/m mŠ mC1
Cn.k/ D .1/ n
:
mD0
.m C 1/k

(This is regarded as a generalization of Proposition 2.10.)

(2) Prove the duality

.k/ .n/
Cn1 D Ck1 .n; k  1/:

.k/
Exercise 14.27. Prove that the number Bn is always even when n; k  1. Hint:
Use Theorem 14.14.
Exercise 14.28. List all n by k lonesum matrices for small n and k, and check
Theorem 14.20.
Exercise 14.29. Prove the congruence stated in Remark 14.24.
Appendix
Curious and Exotic Identities for Bernoulli
Numbers

Don Zagier

Bernoulli numbers, which are ubiquitous in mathematics, typically appear either as

the Taylor coefficients of x= tan x or else, very closely related to this, as special
values of the Riemann zeta function. But they also sometimes appear in other guises
and in other combinations. In this appendix we want to describe some of the less
standard properties of these fascinating numbers.
In Sect. A.1, which is the foundation for most of the rest, we show that, as well
as the familiar (and convergent) exponential generating series1
1
X Bn
x x x2 x4 x6
D x n
D 1  C  C   (A.1)
ex  1 nD0
nŠ 2 12 720 30240

defining the Bernoulli numbers, the less familiar (and divergent) ordinary generat-
ing series
1
X x x2 x4 x6
ˇ.x/ D Bn x n D 1  C  C  (A.2)
nD0
2 6 30 42

also has many virtues and is often just as useful as, or even more useful than, its
better-known counterpart (A.1). As a first application, in Sect. A.2 we discuss the
“modified Bernoulli numbers”
!
X n
nCr Br  

Bn D n1 : (A.3)
rD0
2r nCr

1
Here, and throughout this appendix, we use the convention B1 D 1=2, rather than the
convention B1 D 1=2 used in the main text of the book.

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 239
in Mathematics, DOI 10.1007/978-4-431-54919-2, © Springer Japan 2014
240 Appendix: Curious and Exotic Identities for Bernoulli Numbers

These numbers, which arose in connection with the trace formula for the Hecke
operators acting on modular forms on SL.2; Z/, have several unexpected properties,
including the surprising periodicity

BnC12 D Bn (n odd) (A.4)

and a modified form of the classical von Staudt–Clausen formula for the value
of Bn modulo 1. The following section is devoted to an identity discovered by
Miki [A10] (and a generalization due to Gessel P [A4]) which has Pthe
n striking
property of involving Bernoulli sums both of type Br Bnr and r
Br Bnr ,
i.e., sums related to both the generating functions (A.1) and (A.2). In Sect. A.4 we
look at products of Bernoulli numbers and Bernoulli polynomials in more detail.
In particular, we prove the result (discovered by Nielsen) that when a product
of two Bernoulli polynomials is expressed as a linear combination of Bernoulli
polynomials, then the coefficients are themselves multiples of Bernoulli numbers.
This generalizes to a formula for the product of two Bernoulli polynomials in two
different arguments, and leads to a further proof, due to I. Artamkin, of the Miki–
Gessel identities. Finally, in Sect. A.5 we discuss the continued fraction expansions
of various power series related to both (A.1) and (A.2) and, as an extra titbit,
describe an unexpected appearance of one of these continued fraction expansions
in connection with some recent and amazing discoveries of Yu. Matiyasevich
concerning the non-trivial zeros of the Riemann zeta function.
This appendix can be read independently of the main text and we will recall
all facts and notations needed. We should also add a warning: if you don’t like
generating functions, don’t read this appendix!

A.1 The “Other” Generating Function(s)

for the Bernoulli Numbers

Given a sequence of interesting numbers fan gn0 , one often tries to understand
them by using the properties of the corresponding generating P1functions. The two
n
most popular choices for theseP1 generating functions are nD0 na x (“ordinary
n
generating function”) and nD0 an x =nŠ (“exponential generating function”).
Usually, of course, at most one of these turns out to have useful properties. For the
Bernoulli numbers the standard choice is the exponential generating function (A.1)
because it has an expression “in closed form.” What is not so well known is that the
ordinary generating function of the Bernoulli numbers, i.e., the power series (A.2),
even though it is divergent for all non-zero complex values of x, also has extremely
attractive properties and many nice applications. The key property that makes it
useful, despite its being divergent and not being expressible as an elementary
function, is the following functional equation:
A.1 The “Other” Generating Function(s) for the Bernoulli Numbers 241

Proposition A.1. The power series (A.2) is the unique solution in QŒŒx of the
equation

1  x 
ˇ  ˇ.x/ D x : (A.5)
1x 1x

Proof. Let fBn g be unspecified numbers and define ˇ.x/ by the first equality
in (A.2). Then comparing the coefficients of x m in both sides of (A.5) gives
! 
X
m1
m 1 if m D 1;
Bn D (A.6)
nD0
n 0 if m > 1:

This is the same as the standard recursion for the Bernoulli numbers obtained by
multiplying both sides of (A.1) by e x  1 and comparing the coefficients of x m =mŠ
on both sides. t
u
The functional equation (A.5) can be rewritten in a slightly prettier form by
setting
1
X
ˇ1 .x/ D x ˇ.x/ D Bn x nC1 ;
nD0

in which case it becomes simply

 x 
ˇ1  ˇ1 .x/ D x 2 : (A.7)
1x
A generalization of this is given by the following proposition.
Proposition A.2. For each integer r  1, the power series
1
!
X nCr 1
ˇr .x/ D Bn x nCr (A.8)
nD0
n

satisfies the functional equation

 x 
ˇr  ˇr .x/ D r x rC1 (A.9)
1x
and is the unique power series having this property.
Proof. Equation (A.9) for any fixed value of r  1 is equivalent to the recur-
sion (A.6), by the calculation
242 Appendix: Curious and Exotic Identities for Bernoulli Numbers

! !
 x  X1
nCr 1
1
X `
ˇr  ˇr .x/ D Bn x `C1
1x nD0
n nCr 1
`DnCr

1
! ! !
X ` X
`r
`r C1
D x `C1
Bn D r x rC1 :
r 1 nD0
n
`Dr

Alternatively, we can deduce (A.9) from (A.7) by induction on r by using the easily
checked identity

x 2 ˇr0 .x/ D r ˇrC1 .x/ .r  1/ (A.10)

and the fact that

d  x   x 2 0  x 
x2 F D F (A.11)
dx 1x 1x 1x

for any power series F .x/. t

u
We observe next that the definition (A.8) makes sense for any r in Z,2 and that
the properties (A.9) and (A.10) still hold. But this extension is not particularly
interesting since ˇk .x/ for k 2 Z0 is just a known polynomial in 1=x :

1
! !
X nk1 Xk
n k Bn
ˇk .x/ D Bn x nk
D .1/
nD0
n nD0
n x kn
1 k 1   1
D Bk C D Bk C 1 D .1/k Bk  ;
x x k1 x x

where Bk .X / is the kth Bernoulli polynomial. (One can also prove these identities
by induction on k, using either (A.10) or else (A.9) together with the uniqueness
statement in Proposition A.2 and the corresponding well-known functional equation
for the Bernoulli polynomials.) However, there is a different and more interesting
way to extend the definition of ˇr to non-positive integral values of r. For k 2 Z,
define
X .n  1/Š
k .x/ D BnCk x n 2 x QŒŒx :
.n C k/Š
n  max.1;k/

Then one easily checks that r .x/ D .r  1/Š ˇr .x/ for r > 0, so that the
negative-index power series k are just renormalized versions of the positive-index
power series ˇr . But now we do get interesting power series (rather than merely
polynomials) when k  0, e.g.

2
Or even in C if we work formally in x r QŒŒx.
A.1 The “Other” Generating Function(s) for the Bernoulli Numbers 243

1
X X1 X1
Bn x n BnC1 x n BnC2 x n
0 .x/ D ; 1 .x/ D ; 2 .x/ D :
nD1
n nD1
n.n C 1/ nD1
n.n C 1/.n C 2/
(A.12)
The properties of these new functions corresponding to (A.10) and (A.9) are
given by:
Proposition A.3. The power series k .x/ satisfy the differential recursion
Bk
x 2 k0 .x/ D k1 .x/  x .k  0/ (A.13)
kŠ
.with 1 .x/ D ˇ1 .x// as well as the functional equations
 x 
0  0 .x/ D log.1  x/ C x ; (A.14)
1x
 x   1 1
1  1 .x/ D   log.1  x/  1 ;
1x x 2

and more generally for k  1

 x      1 
.1/k 1
k  k .x/ D Bk log.1  x/ C Pk1 ; (A.15)
1x kŠ x x
where Pk1 .X / is a polynomial of degree k  1, the first few values of which are
P0 .X / D 1, P1 .X / D X  12 , P2 .X / D X 2  X C 12 1
, P3 .X / D X 3  32 X 2 C
1
3
X C 12 and P4 .X / D X 4  2X 3 C 4 X 2 C 4 X  360 .
1 3 1 13

Proof. Equation (A.13) follows directly from the definitions, and then Eqs. (A.14)
and (A.15) (by induction over k) follow successively from (A.7) using the general
identity (A.11). t
u
We end this section with the observation that, although ˇ.x/ and the related
power series ˇr .x/ and k .x/ that we have discussed are divergent and do not give
the Taylor or Laurent expansion of any elementary functions, they are related to
the asymptotic expansions of very familiar, “nearly elementary” functions. Indeed,
Stirling’s formula in its logarithmic form says that the logarithm of Euler’s Gamma
function has the asymptotic expansion
 1 1 X1
Bn
log .X /
X  log X  X C log.2/ C X nC1
2 2 nD2
n.n  1/

as X ! 1, and hence that its derivative .X / (“digamma function”) has the

expansion

0 .X / 1 X Bn 1  1
.X / WD
log X   X n D log X  0 
.X / 2X nD2 n X
244 Appendix: Curious and Exotic Identities for Bernoulli Numbers

as X ! 1, with 0 .x/ defined as in Eq. (A.12), and the functions ˇr .x/ correspond
similarly to the derivatives of .x/ (“polygamma functions”). The transformation
x 7! x=.1x/ occurring in the functional equations (A.5), (A.9), (A.14) and (A.15)
corresponds under the substitution X D 1=x to the translation X 7! X C1, and the
compatibility equation (A.11) simply to the fact that this translation commutes with
the differential operator d=dX , while the functional equations themselves reflect
the defining functional equation .X C 1/ D X .X / of the Gamma function.

A.2 An Application: Periodicity of Modified Bernoulli

Numbers

The “modified Bernoulli numbers” defined by (A.3) were introduced in [A14].

These numbers, as already mentioned in the introduction, occurred naturally in a
certain elementary derivation of the formula for the traces of Hecke operators acting
on modular forms for the full modular group [A15]. They have two surprising
properties which are parallel to the two following well-known properties of the
ordinary Bernoulli numbers:

n > 1 odd ) Bn D 0 ; (A.16)

X 1
n > 0 even ) Bn   .mod 1/ (A.17)
p prime
p
.p1/jn

(von Staudt–Clausen theorem). These properties are given by:

Proposition A.4. Let Bn .n > 0/ be the numbers defined by (A.3). Then for n odd
we have

 ˙3=4 if n ˙ 1 .mod 12/,
Bn D (A.18)
1=4 if n ˙ 3 or ˙5 .mod 12/,

and for n even we have the modified von Staudt–Clausen formula

X 1
2nBn  Bn  .mod 1/ : (A.19)
p prime
p
.pC1/jn

Remark. The modulo 12 periodicity in (A.18) is related, via the above-mentioned

connection with modular forms on the full modular group SL.2; Z/, with the well-
known fact that the space of these modular forms of even weight k > 2 is the sum
of k=12 and a number that depends only on k .mod 12/.
A.2 An Application: Periodicity of Modified Bernoulli Numbers 245

Proof. The second assertion is an easy consequence of the corresponding

property (A.17) of the ordinary Bernoulli numbers and we omit the proof. (It is
given in [A15].) To prove the first, we use the generating functions for Bernoulli
numbers introduced in Sect. A.1. Specifically, for 2 Q we define a power series
g .t/ 2 QŒŒt by the formula
 t 
g .t/ D 0  log.1  t C t 2 / ;
1  t C t 2
P
where 0 .x/ D n>0 Bn x
n
=n is the power series defined in (A.12). For D 2 this
specializes to
1
X X1
Br tr
g2 .t/ D  2 log.1  t/ D 2 Bn t n : (A.20)
rD1
r .1  t/2r nD1

with Bn as in (A.3). On the other hand, the functional equation (A.14) applied to
x D t=.1  t C t 2 /, together with the parity property 0 .x/ C x D 0 .x/, which
is a restatement of (A.16), implies the two functional equations

t
g C1 .t/ D g .t/ C D g .t/
1  t C t 2

for the power series g . From this we deduce

     
g2 .t/  g2 .t/ D g2 .t/  g1 .t/ C g1 .t/  g0 .t/ C g0 .t/  g1 .t/
t t t 3t  t 3  t 5 C t 7 C t 9  3t 11
D C C D ;
1  t C t2 1 C t2 1 C t C t2 1  t 12

and comparing this with (A.20) immediately gives the desired formula (A.18) for
Bn , n odd. u
t
We mention one further result about the modified Bernoulli numbers from [A15].
The ordinary Bernoulli numbers satisfy the asymptotic formula

2 nŠ
Bn
.1/.n2/=2 (n ! 1, n even). (A.21)
.2/n

As one might expect, the modified ones have asymptotics given by a very similar
formula:

.n  1/Š
Bn
.1/.n2/=2 (n ! 1, n even). (A.22)
.2/n

The (small) surprise is that, while the asymptotic formula (A.21) holds to all orders
in 1=n (because the ratio of the two sides equals .n/ D 1 C O.2n /), this is not
246 Appendix: Curious and Exotic Identities for Bernoulli Numbers

true of the new formula (A.22), which only acquires this property if the right-hand
side is replaced by .1/n=2  Yn .4/, where Yn .x/ is the nth Bessel function of the
second kind.
Here is a small table of the numbers Bn and BQn D 2nBn  Bn for n even:

n 2 4 6 8 10 12 14 16 18 20 22
Bn 1
24
 27
80
 1260
29 451
1120
 264
65
 12012
6571 571
312
 181613
38080
23663513
1220940
 10188203
83600
564133
552
BQn 0  83  10
3 136
21
5  330
4249 651
13
 3056
21
109269
170
 247700
57
38775

A.3 Miki’s Identity

The surprising identity described in this section was found and proved by
Miki [A10] in an indirect and non-elementary way, using p-adic methods. In
this section we describe two direct proofs of it, or rather, of it and of a very similar
identity discovered by Faber and Pandharipande in connection with Chern numbers
of moduli spaces of curves. The first, which is short but not very enlightening,
is a variant of a proof I gave of the latter identity [A2] (but which with a slight
modification works for Miki’s original identity as well). The second one, which
is more natural, is a slight reworking of the proof given by Gessel [A4] based on
properties of Stirling numbers of the second kind. In fact, Gessel gives a more
general one-parameter family of identities, provable by the same methods, of which
both the Miki and the Faber–Pandharipande identities are special cases. In Sect. A.4
we will give yet a third proof of these identities, following I. Artamkin [A1].
Proposition A.5 (Miki). Write Bn D .1/n Bn =n for n > 0. Then for all n > 2 we
have
!
X
n2 X
n2
n
Bi Bni D Bi Bni C2Hn Bn ; (A.23)
i D2 i D2
i

where Hn D 1 C 1
2 CC 1
n denotes the nth harmonic number.
B2g
(Faber–Pandharipande). Write bg D .2  22g / for g  0. Then for all
.2g/Š
g > 0 we have

X .2g1  1/Š .2g2  1/Š X 22n B2n

g
bg 1 bg 2 D bgn CH2g1 bg : (A.24)
g1 Cg2 Dg
2 .2g  1/Š nD1
2n .2n/Š
g1 ; g2 >0
A.3 Miki’s Identity 247

First proof. We prove (A.24), following [A2]. Write the identity as a.g/ D b.g/ C
P
1 P
1
c.g/ in the obvious way, and let A.x/ D a.g/ x 2g1 , B.x/ D b.g/ x 2g1
gD1 gD1
P
1
and C.x/ D c.g/ x 2g1 be the corresponding odd generating functions. Using
gD1
P
1 1
the identity bg x 2g1 D , we obtain
gD0 sinh x

X Z x
1
A.x/ D bg 1 bg 2 t 2g1 1 .x  t/2g2 1 dt (by Euler’s beta integral)
2 g1 ; g2 >0 0
Z 1
1 x
1  1 1 
D   dt ;
2 0 t sinh t x  t sinh.x  t/
1
1 X 22n B2n 2n 1  sinh x 
B.x/ D x D log ;
sinh x nD1 2n .2n/Š sinh x x
Z Z 
1  1 1 
1
X x
x 2g1  t 2g1 x
1
C.x/ D bg dt D  C dt ;
gD1 0 xt 0 x  t sinh x sinh t xt

and hence, symmetrizing the integral giving C.x/ with respect to t ! x  t,

Z

x
1  1
1 1 
2 A.x/  2 C.x/ D  
0 t sinh t x  t sinh.x  t/
 1 1  1 1 
 C C
xt t sinh x x

1 1 1 1
C C dt
x  t sinh t t sinh.x  t/
Z x 
1 x 1
D  dt
0 sinh.t/ sinh.x  t/ sinh x t .x  t/
1  sinh t x  t ˇˇt Dx
D log  ˇ D 2 B.x/ :
sinh x t sinh.x  t/ t D0

A similar proof can be given for Miki’s original identity (A.23), with “sinh” replaced
by “tanh”. t
u
Second proof. Now we prove (A.23), following the method in [A4]. Recall that the
Stirling number of the second kind S.k; m/ is defined as the number of partitions
of a set of k elements into m non-empty subsets or, equivalently, as 1=mŠ times the
number of surjective maps from the set f1; 2; : : : ; kg to the set f1; 2; : : : ; mg. It can
be given either by the closed formula
248 Appendix: Curious and Exotic Identities for Bernoulli Numbers

!
1 X
m
m` m
S.k; m/ D .1/ `k (A.25)
mŠ `
`D0

(this follows immediately from the second definition and the inclusion-exclusion
principle, since `k is the number of maps from f1; 2; : : : ; kg to a given set of `
elements) or else by either of the two generating functions
1
X xm
S.k; m/ x k D ;
.1  x/.1  2x/    .1  mx/
kD0
(A.26)
1
X xk .e x  1/m
S.k; m/ D ;
kŠ mŠ
kD0

both of which can be deduced easily from (A.25). (Of course all of these formulas
are standard and can be found in many books,  including Chap. 2 of this one, where
S.k; m/ is denoted using Knuth’s notation mk .) From either generating function one
finds easily that S.k; m/ vanishes for k < m, S.m; m/ D 1, S.m C 1; m/ D m 2Cm ,
2

and more generally that S.m C n; m/ for a fixed value of n is a polynomial in m

(of degree 2n, and without constant term if n > 0). Gessel’s beautiful and very
natural idea was to compute the first few coefficients of this polynomial using each
of the generating functions in (A.26) and to equate the two expressions obtained. It
turned out that this gives nothing for the coefficients of m0 and m1 (which are found
from either point of view to be 0 and Bn , respectively), but that the equality of the
coefficients of m2 obtained from the two generating functions coincides precisely
with the identity that Miki had discovered!
More precisely, from the first formula in (A.26) we obtain
X  X  1  X 1r C 2r C    mr r
1 m 1
log S.m C n; m/ x n D log D x
nD0 j D1
1  jx rD1
r
1 
X 
Br .1/r1 Br1 2
D mC m C    xr
rD1
r 2

(the last line by the Bernoulli–Seki formula) and hence, exponentiating,

 X
n2  m2
S.m C n; m/ D Bn mC nBn1 C Bi Bni C .n  3/ ; (A.27)
i D2
2
 
while
P from the second formula in (A.26) and the expansion log .e x  1/=x D
Bn x n =nŠ we get
n>0
A.4 Products and Scalar Products of Bernoulli Polynomials 249

S.m C n; m/
 m  m  m x n  e x  1 m
D 1C 1C  1C  Coefficient of in
1 2 n nŠ x
  X n1
!  2 
n m
D 1 C Hn m C    Bn m C Bi Bni C
i D1
i 2
!
 X
n1
n  m2
D Bn m C 2Hn Bn C Bi Bni C    .n  1/ : (A.28)
i D1
i 2

Comparing the coefficients of m2 =2 in (A.27) and (A.28) gives Eq. (A.23). t

u
Finally, we state the one-parameter generalization of (A.23) and (A.24) given
in [A4]. For n > 0 denote by Bn .x/ the polynomial Bn .x/=n.
Proposition A.6 (Gessel). For all n > 0 one has
  X !
n X
n1 n
n
Bn1 .x/ C Bi .x/Bni .x/ D Bi Bni .x/CHn1 Bn .x/ : (A.29)
2 i D1 i D1
i

Gessel does not actually write out the proof of this identity, saying only that it can
be obtained in the same way as his proof of (A.23) and pointing out that, because
Bn .1/ D Bn and 22g B2g .1=2/ D .2g  1/Š bg , it implies (A.23) and (A.24) by
specializing to x D 1 and x D 1=2, respectively.

A.4 Products and Scalar Products of Bernoulli Polynomials

If A is any algebra over Q and e0 ; e1 ; : : : is an additive basis of A, P then each

product ei ej can be written uniquely as a (finite) linear combination k cijk ek for
certain numbers cijk 2 Q and the algebra structure on A is completely determined
by specifying the “structure constants” cijk . If we apply this to the algebra A D QŒx
and the standard basis ei D x i , then the structure constants are completely trivial,
being simply 1 if i C j D k and 0 otherwise. But the Bernoulli polynomials also
form a basis of QŒx, since there is one of every P degree, and we can ask what the
structure constants defined by Bi .x/Bj .x/ D k cijk Bk .x/ are. It is easy to see that
cijk can only be non-zero if the difference rWDi C j  k is non-negative (because
Bi .x/Bj .x/ is a polynomial of degree i C j ) and even (because the nth Bernoulli
polynomial is .1/n -symmetric with respect to x 7! 1  x). The surprise is that,
up to an elementary factor, cijk is equal simply to the kth Bernoulli number, except
when k D 0. This fact, which was discovered long ago by Nielsen [A11, p. 75]
(although I was not aware of this reference at the time when Igor Artamkin and I
had the discussions that led to the formulas and proofs described below), is stated in
250 Appendix: Curious and Exotic Identities for Bernoulli Numbers

a precise form in the following proposition. The formula turns out to be somewhat
simpler if we use the renormalized Bernoulli polynomials Bn .x/ D Bnn.x/ rather
than the Bn .x/ themselves when n > 0. (For n D 0 there is nothing to be calculated
since the product of any Bi .x/ with B0 .x/ D 1 is just Bi .x/.)
Proposition A.7. Let i and j be strictly positive integers. Then
! !
X 1 i 1 j
Bi .x/ Bj .x/ D C B2` Bi Cj 2` .x/
i Cj
i 2` j 2`
0`< 2 (A.30)
i 1
.1/ .i  1/Š .j  1/Š
C Bi Cj :
.i C j /Š

Note that, despite appearances, the (constant) second term in this formula is
symmetric in i and j , because if Bi Cj ¤ 0 then i and j have the same parity.
Proof. Write Bi;j .x/ for the right-hand side of (A.30). We first show that the
difference between Bi;j .x/ and Bi .x/Bj .x/ is constant. This can be done in two
different ways. First of all, using Bn .x C 1/  Bn .x/ D x n1 we find
! !
X 1 i 1 j
Bi;j .x C 1/  Bi;j .x/ D C B2` x i Cj 2`1
i Cj
i 2` j 2`
0`< 2
 1   1 
D x j 1 Bi .x/ C x i 1 C x i 1 Bj .x/ C x j 1
2 2
D Bi .x C 1/Bj .x C 1/  Bi .x/Bj .x/ :

It follows that the Bi;j .x/  Bi .x/Bj .x/ is periodic and hence, since it is also
polynomial, constant. Alternatively, we can use that Bn0 .x/ equals 1 for n D 1 and
.n1/Bn1 .x/ for n > 1 to show by induction on i Cj that Bi;j .x/ and Bi .x/Bj .x/
have the same derivative (we omit the easy computation) and hence again that their
difference is constant. To show that this constant vanishes, it suffices to show that the
integrals of the two sides of (A.30) over the interval [0,1] agree. Since the integral
of Bn .x/ over this interval vanishes for any n >p0, this reduces to the following
statement, in which to avoid confusion with i D 1 we have changed i and j to
r and s. t
u
Proposition A.8. Let r and s be positive integers. Then
Z 1
rŠ sŠ
Br .x/ Bs .x/ dx D .1/r1 BrCs : (A.31)
0 .r C s/Š

Proof. Here again we give two proofs. The first uses the Fourier development

kŠ X e 2 i nx
Bk .x/ D  .0 < x < 1; k  1/ (A.32)
.2 i /k n2Z nk
n¤0
A.4 Products and Scalar Products of Bernoulli Polynomials 251

discussed in Chap. 4, Theorem 4.11 of this book. (For k D 1 the sum converges
R1
only conditionally and one has to be a little careful.) Since the integral 0 e 2 i kx dx
equals ık;0 , this gives
Z 1
rŠ sŠ X 1 rŠ sŠ
Br .x/ Bs .x/ dx D .1/r D .1/r1 BrCs
0 .2 i / rCs
n2Z
nrCs .r C s/Š
n¤0

as desired. (The second equality, giving the well-known connection between

Bernoulli numbers and the values at positive even integers of the Riemann zeta
function, is just the case k D r C s, x ! 0 of (A.32).) The second proof, using
generating functions, is just as short. Denote the left-hand side of (A.31), also for r
or s equal to 0, by Ir;s . Then we have

X Z
t r1 us1 1
e xt e xu 1 1 e t Cu  1
Ir;s D dx D
r; s0
rŠ sŠ 0 et  1 eu  1 et  1 eu  1 t C u
 X1
1 1 1 Bk t k1  .u/k1
D  u
D
t Cu e 1 e 1
t kŠ t Cu
kD0

1 X Bk X
D C t r1 .u/s1 ;
tu kŠ r; s1
k2
rCsDk

and Eq. (A.31) follows by equating the coefficients of t r1 us1 .

Before continuing, we show that Proposition A.7 immediately yields another
proof of the identities of Miki and Gessel discussed in the preceding section. This
method is due to I. Artamkin [A1] (whose proof, up to a few small modifications,
we have followed here). Indeed, summing (A.30) over all i; j  1 with i C j D n,
and using the easy identities
! !
X
n1
1 i 1 n1
D .r > 0/
i D1
i r r r

and

X X n1 Z
.i  1/Š .j  1/Š 1
.1/i 1 D .x/i 1 .1  x/ni 1 dx
i; j 1
.n  1/Š i D1 0
i Cj Dn
Z 1
 1 C .1/n
D .1  x/n1  .x/n1 dx D
0 n
252 Appendix: Curious and Exotic Identities for Bernoulli Numbers

(where the first equation is the beta integral again), we obtain

!
1 X X
n1
n1 Bn .0/
Bi .x/Bj .x/ D Hn1 Bn .x/ C Br .0/ Bnr .x/ C ;
2 i; j 1 rD2
r n
i Cj Dn
(A.33)
which is equivalent to Gessel’s identity (A.29).
Proposition A.8 describes the scalar products among the Bernoulli polynomials
R1
with respect to the scalar product .f; g/ D 0 f .x/g.x/dx. It is more natural to
replace the Bernoulli polynomials Bk .x/ by their periodic versions B k .x/ (defined
for x … Z as Bk .x  Œx/ or by the right-hand side of (A.32), and for x 2 Z by
continuity if k ¤ 1 and as zero if k D 1), since then the scalar product is simply the
integral of B r .x/B s .x/ over the whole domain of definition R=Z. The first proof
just given then carries over almost unchanged to give the following more general
result:
Proposition A.9. Let r and s be integers  1 and ˛, ˇ two real numbers. Then
Z 1
rŠ sŠ
B r .x C ˛/ B s .x C ˇ/ dx D .1/r1 B rCs .˛  ˇ/ : (A.34)
0 .r C s/Š

Using this, one finds, with almost the same proof as before, the following
generalization of Proposition A.7:
Proposition A.10. Let i and j be positive integers. Then for any two variables x
and y we have

max.i;j /
! !
X 1 i .1/m j
Bi .x/ Bj .y/ D Bi Cj m.y/ C Bi Cj m.x/ BmC .x  y/
mD0
i m j m

.i  1/Š .j  1/Š C
C .1/j 1 Bi Cj .x  y/ ; (A.35)
.i C j /Š

where BmC .x/ denotes the symmetrized Bernoulli polynomial

Bm .x/ C .1/m Bm .x/ Bm .x C 1/ C Bm .x/ m

BmC .x/ D D D Bm .x/ C x m1 :
2 2 2
The same calculation as was used above to deduce (A.33) from (A.30), but now
applied to (A.35) instead of (A.30), gives the following generalization of Gessel’s
identity (A.29):
A.4 Products and Scalar Products of Bernoulli Polynomials 253

X  
Bi .x/Bj .y/  Hn1 Bn .x/ C Bn .y/
i; j 1
i Cj Dn
!
X n1   B C .x  y/
n1
D Bnm .y/ C .1/m Bnm .x/ m

mD1
m m

1 C .1/n C
C Bn .x  y/ : (A.36)
n2
We observe that Eq. (A.36) was also found by Hao Pan and Zhi-Wei Sun [A12] in a
slightly different form, the right-hand side in their formula being

!
X n1  Bm .y  x/ 
n
Bm .x  y/
Bnm .y/ C Bnm .x/
mD1
m1 m2 m2

1 Bn .x/  Bn .y/
C ; (A.37)
n xy

which is easily checked to be equal to the right-hand side of (A.36); their formula
has the advantage of being more visibly symmetric in x and y and of using only the
Bernoulli polynomials Bm .x/ rather than the symmetrized Bernoulli polynomials
BmC .x/, but the disadvantage of having a denominator x  y (which of course
disappears after division into the numerator Bn .x/Bn .y/) rather than being written
in an explicitly polynomial form.
We end this section by giving a beautifully symmetric version of the multiplica-
tion law for Bernoulli polynomials given by the same authors in [A13].
Proposition
 A.11 (Sun–Pan). For each integer n  0 define a polynomial
r s
in four variables r, s, x and y by
xy n

 ! !
X
r s i r s
D .1/ Bj .x/ Bi .y/ : (A.38)
xy n i j
i; j 0
i Cj Dn

Then for any six variables r, s, t, x, y and z satisfying r CsCt D n and xCyCz D 1
we have
  
r s s t t r
t Cr Cs D0: (A.39)
xy n yz n zx n

First proof (sketch). We can prove (A.39) in the same way as (A.36) was proved
above, replacing the product Bj .x/Bi .y/ in (A.38) for i and j positive using
formula (A.35) (with x and y replaced by 1  y and x) and then using elementary
254 Appendix: Curious and Exotic Identities for Bernoulli Numbers

binomial coefficient identities to simplify the result. We do not give the full
calculation, which is straightforward but tedious. t
u
Second proof. An alternative, and easier, approach is to notice that, since the left-
hand side of (A.39) is a polynomial in the variables x, y and z D 1  x  y, it is
enough to prove the identity for x; y; z > 0 with x C y C z D 1. But for x and y
between 0 and 1 we have from (A.32)
 X
r s
.2 i / n
D Cn .r; sI a; b/ e 2 i.bxay/
xy n
a; b2Z

with
⎧P
⎪
⎪
i j
i; j 1; i Cj Dn .r/i .s/j a b if a ¤ 0, b ¤ 0
⎪
⎪
⎨  .r/n an if a ¤ 0, b D 0
Cn .r; sI a; b/ D
⎪
⎪  .s/n b n if a D 0, b ¤ 0
⎪
⎪
⎩
0 if a D 0, b D 0

where .x/m D x.x  1/    .x  m C 1/ is the descending Pochhammer symbol.

Equation (A.39) then follows from the identity

t Cn .r; sI a; b/Cr Cn .s; tI b; c/Cs Cn .t; rI c; a/ D 0 .aCbCc D 0; rCsCt D n/ :

whose elementary proof (using partial fractions if abc ¤ 0) we omit. t

u
We end by remarking on a certain formal similarity between the cyclic iden-
tity (A.39) and a reciprocity law for generalized Dedekind sums proved in [A5].
The classical Dedekind sums, introduced by Dedekind while posthumously editing
some unpublished calculations of Riemann’s, are defined by
X h  bh 
s.b; c/ D B1 B1 (b; c 2 N coprime),
c c
h .mod c/

where B 1 .x/ as usual is the periodic version of the first Bernoulli polynomial (equal
to x  12 if 0 < x < 1, to 0 if x D 0, and periodic with period 1), and satisfy the
famous Dedekind reciprocity relation

b2 C c2 C 1 1
s.b; c/ C s.c; b/ D  :
12bc 4

This was generalized by Rademacher, who discovered that if a, b and c are pairwise
coprime integers then the sum
X  ah   bh 
s.a; bI c/ D B1 B1 (A.40)
c c
h .mod c/
A.4 Products and Scalar Products of Bernoulli Polynomials 255

which equals s.a0 ; c/ for any a0 with aa0 b .mod c/ or ba0 a .mod c/, satisfies
the identity

a2 C b 2 C c 2 1
s.a; bI c/ C s.b; cI a/ C s.c; aI b/ D  : (A.41)
12abc 4

A number of further generalizations, in which the functions B 1 in (A.40) are

replaced by periodic Bernoulli polynomials with other indices and/or the arguments
of these polynomials are shifted by suitable rational numbers, were discovered later.
The one given in [A5] concerns the sums
  X  hCz   hCz 
abc
Sm;n D Bm a  x Bn b y ; (A.42)
xy z c c
h .mod c/

where m and n are non-negative integers, a, b and c natural numbers with no

common factor, and x, y and z elements of TWDR=Z. (The hth summand in (A.42)
depends on z modulo c, not just modulo 1, but the whole sum has period 1 in z.)
For fixed m and n these sums do not satisfy any relation similar to the 3-term
relation (A.41) for the case m D n D 1, but if we assemble all of the functions
Sm;n (m; n  0) into a single generating function
0 1
 
a b c  X m1  Y n1
a b c X 1
@
S x y z DA Sm;n ; (A.43)
mŠ nŠ xy z a b
X Y Z m; n0

in which X , Y and Z (which does not appear explicitly on the right) are formal
variables satisfying X C Y C Z D 0, then we have the following relation:
Proposition A.12 ([A5]). Let a; b; c be three natural numbers with no common
factor, x; y; z three elements of T, and X; Y; Z three formal variables satisfying
X C Y C Z D 0. Then
0 1 0 1 0 1 
a b c b c a c a b
1=4 if .x; y; z/ 2 .a; b; c/T ;
S@ x y z A C S@ y z x A C S@ z x y A D
X Y Z Y ZX ZX Y 0 otherwise.

We do not give the proof of this relation, since three different proofs (all similar
in spirit to various of the proofs that have been given in this appendix) are given
in [A5], but we wanted to at least mention this generalized Dedekind–Rademacher
reciprocity law because of its formal resemblance, and perhaps actual relationship,
to the Sun–Pan reciprocity law (A.39).
256 Appendix: Curious and Exotic Identities for Bernoulli Numbers

A.5 Continued Fraction Expansions for Generating

Functions of Bernoulli Numbers

There are several classical formulas expressing various versions of the standard
(exponential) generating functions of the Bernoulli numbers as continued fractions.
A simple example is
 X 2n .2n  1/Bn 
x
tanh x D x n1 D ; (A.44)
nŠ x2
n2 1C
x2
3C
x2
5C
::
:

whose proof is recalled below, and a somewhat more complicated one, whose proof
we omit, is
 X B2n 
x=2 1
D x 2n
D (A.45)
tanh x=2 .2n/Š a1 x 2
n0 1C
a2 x 2
1C
::
:

with an defined by
⎧
⎪
⎪ 1
⎪
⎪  if n D 1 ,
⎪
⎪ 12
⎪
⎪
⎪
⎨ .n C 1/.n C 2/
an D if n is even ,
⎪
⎪ .2n  2/.2n  1/.2n/.2n C 1/
⎪
⎪
⎪
⎪ .n  2/.n  1/
⎪
⎪
⎪
⎩ .2n  1/.2n/.2n C 1/.2n C 2/ if n > 1 is odd .

It was discovered by M. Kaneko that the convergents Pn .x/=Qn .x/ of the continued
fraction (A.45) could be given in a simple closed form, namely
! !1
X
n=2
n 2n C 1 xi
Pn .x/ D
i D0
2i 2i .2i C 1/Š
! !1
X
n=2
nC1 2n C 2 xi
Qn .x/ D
i D0
2i 2i .2i /Š
A.5 Continued Fraction Expansions for Generating Functions of Bernoulli. . . 257

if n is even and a similar but slightly more complicated expression if n is odd.

(It was in connection with this discovery that he found the short recursion formula
for Bernoulli numbers discussed in Sect. 1.2 of the book.) Again we omit the proof,
which is given in [A6].
What is perhaps more surprising is that there are also nice continued fraction
expansions for certain non-standard (ordinary) generating functions of Bernoulli
numbers of the type considered in Sect. A.1, and these are in some sense of even
more interest because the continued fractions, unlike the power series themselves,
converge for positive real values of the argument (and give the appropriate deriva-
tives of .X / as discussed in the last paragraph of Sect. A.1). For instance, on the
cover of the Russian original of Lando’s beautiful book on generating functions [A7]
one finds the pair of formulas3

x3 x5 x7
1xC2 C 16  C 272  C    D tan x
3Š 5Š 7Š
x
1  x C 2  x 3 C 16  x 5 C 272  x 7 C    D
1  2 x2
1
2  3 x2
1
3  4 x2
1
1 
The numbers 1, 2, 16, 272, . . . defined by the first of these two formulas are just
the numbers .4n  2n /jBn j=n, so the second formula gives a continued fraction
expansion for the non-exponential generating function for essentially the Bernoulli
numbers. Again we omit the proof, referring for this to the book cited, mentioning
only the following alternative and in some ways prettier form of the formula:

1 2 16 272 1
 3 C 5  7 C D (A.46)
X X X X 1
XC
X 1
C
2 X
C
3
  X C4   
in which the continued fraction is convergent and equal to 1 X2 4  X C2
4
for all X > 0.
Other continued fraction expansions for non-exponential Bernoulli number
generating functions that can be found in the literature include the formulas

3
In the English translation [A8] (which we highly recommend to the reader) this formula has been
relegated to the exercises: Chapter 5, Problem 5.6, page 85.
258 Appendix: Curious and Exotic Identities for Bernoulli Numbers

1
X x
B2n .4x/n D ;
1 x
nD1 1C C
2 1 1 x
C C
2 3 1 1 x
C C
3 4 ::
:

or the equivalent but less appealing identity

1
X 1
Bn x n D ;
x
nD0 1C
2 x

1 2x
3C
2 2x

2 3x
5C
2 3x

3 4x
7C
2 4x

4 5x
9C
::
:

and
1
X x
.2n C 1/ B2n x n D
x
nD1 1C1C
1 x
1C C
2 1 1 x
C C
2 2 1 1 x
C C
2 3 1 1 x
C C
3 3 ::
:

all given by J. Frame [A3] in connection with a statistical problem on curve fitting.
For good conscience’s sake we give the proofs of one continued fraction of each
of the two above types, choosing for this purpose the two simplest ones (A.44)
and (A.46). We look at (A.44) first. Define functions I0 ; I1 ; : : : on .0; 1/ by
Z a
t n .1  t=a/n t  
In .a/ D e dt n 2 Z0 ; a 2 R>0 :
0 nŠ
A.5 Continued Fraction Expansions for Generating Functions of Bernoulli. . . 259

Integrating by parts twice, we find that

Z 
a
d 2 t nC1 .1  t=a/nC1
InC1 .a/ D et dt
0 dt 2 .n C 1/Š
Z a  n1
t .1  t=a/n1 4n C 2 t n .1  t=a/n
D et  dt
0 .n  1/Š a nŠ
4n C 2
D In1 .a/  In .a/
a

In1 .a/ 4n C 2 InC1 .a/

for n > 0. Rewriting this as D C and noting that
In .a/ a In .a/
 2  2
I0 .a/ D e a  1 ; I1 .a/ D e a 1  C 1C
a a
by direct calculation, we obtain

1 e 2x C 1 1 I1 .2x/ 1 1
D 2x D C D C ;
tanh x e 1 x I0 .2x/ x 3 1
C
x 5 1
C
x ::
:

which is equivalent to (A.44). Similarly, for (A.46), we define functions J0 ; J1 ; : : :

on .0; 1/ by
Z 1
 n  
Jn .X / D tanh .t=X / e t dt n 2 Z0 ; X 2 R>0 :
0

This time J0 .X / is simply the constant function 1, while J1 .X / has the exact
evaluation
X X  X X 1
J1 .X / D 1  C1 C C ; (A.47)
2 4 2 4 2
as is easily deduced from Euler’s integral representation
Z 1
1  t x1
.x/ D  C dt ;
0 1t
as well as the asymptotic expansion
Z 1 
1 2 t3 16 t 5 272 t 7
J1 .X /
t 3 C 5  7 C    e t dt
0 X X 3Š X 5Š X 7Š
1 2 16 272

 3 C 5  7 C
X X X X
260 Appendix: Curious and Exotic Identities for Bernoulli Numbers

as X ! 1. (This last expression can be written as 1  X0 .2=X / C X0 .4=X /

with 0 as in (A.12), in accordance with (A.47) and the relationship between 0 .X /
and .X / given at the end of Sect. A.1.) On the other hand, integrating by parts and
using tanh.x/0 D 1  tanh.x/2 , we find
Z 1
d  n 
Jn .X / D e ttanh.t=X / dt
0 dt
Z 1
n  n1   2 
D e t tanh.t=X / 1  tanh.t=X / dt
X 0
n  
D Jn1 .X /  JnC1 .X /
X

Jn1 .X / X JnC1 .X /
for n > 0, and rewriting this as D C we obtain that J1 .X / D
Jn .X / n Jn .X /
J1 .X /
has the continued fraction expansion given by the right-hand side of (A.46),
J0 .X /
as claimed. t
u
We end this appendix by describing an appearance of the continued frac-
tion (A.46) in connection with the fantastic discovery of Yuri Matiyasevich that
“the zeros of the Riemann zeta function know about each other.” Denote the zeros
of .s/ on the critical line <.s/ D 12 by n and n with 0 < =.1 /  =.2 /    
and for M  1 consider the finite Dirichlet series M .s/ defined as the N  N
determinant4
ˇ ˇ
ˇ 1 1  1 1 ˇˇ
ˇ 1
ˇ : :: : : : :: :: ˇˇ
ˇ :: : ::
ˇ : : : ˇ
ˇ 1 1 ˇ
ˇn n    nM nM ns ˇ
M .s/ D ˇ : :: ˇˇ ;
ˇ : :: : : :: ::
ˇ : : : : : : ˇ
ˇ ˇ
ˇ ˇ
ˇ  ˇ
ˇN 1 N 1    N M N M N s ˇ

where N D 2M C 1. This function clearly vanishes when s D n or n for 1 

n  M , but Matiyasevich’s discovery (for which we refer to [A9] and the other
papers and talks listed on his website) was that its subsequent zeros are incredibly
close to the following zeros of the Riemann zeta function, e.g., the first zero of
50 on 12 C R>0 following 50 differs in absolute value from 51 by less than 4 
1015 , the first zero of 1500 after 1500 differs in absolute value from 1501 by less
than 5101113 , and even the 300th zero of 1500 after 1500 differs in absolute value
766
from P1801 by less than 5  10 ! Moreover, if we write the Dirichlet series M .s/
N s
as cM nD1 aM;n n with the normalizing constant cM chosen to make aM;1 D 1,

4
We have changed Matiyasevich’s notations slightly for convenience of exposition.
References 261

1
then it turns out that the function cM M .s/ not only has almost the same zeros,
but is itself a very close approximation to .1  21s /.s/ over a long interval of the
critical line.
In studying this latter function, Matiyasevich
P was led to consider the real
numbers M defined by M D 4M 2M nD1 M;n =n, where M;n denotes the
coefficient of ns in the Dirichlet series cM 1
M .s/=.s/. Since by the nature
of his investigation he was working to very high precision, he obtained very
precise decimal expansions of these numbers, and in an attempt to recognize
them, he computed the beginning of their continued fraction expansions. (Recall
that rational numbers and real quadratic irrationalities can be recognized
numerically by the fact that they have terminating or periodic continued
fraction expansions.) To his surprise, when M was highly composite these
numbers had very exceptional continued fraction expansions. For instance, for
2M D l.c.m.f1; 2; : : : ; 10g D 2520, the number M has a decimal expansion
beginning 0:9998015873172093    and a continued fraction expansion beginning
Œ0; 1; 5039; 2520; 1680; 1260; 1008; 840; 720; 630; 560; 504. In view of the
fact that nearly all real numbers (in a very precise metrical sense) have continued
fraction expansions with almost all partial quotients very small, this is certainly
not a coincidence, and it is even more obviously not one when we notice that
the numbers 5040, 2520, . . . 504 are 5040=n for n D 1; 2; : : : ; 10. This leads
one immediately to the continued fraction (A.46) with X D 4M and hence, in
view of the evaluation of that continued
 fraction
 given
 above, to the (conjectural)
approximation M  12 M C 1  12 M C 12 , which turns out indeed to be a
very good one for M large, the two numbers differing only by one part in 10108 in
the above-named case 2M D 2520. We take this somewhat unusual story as a fitting
place to end our survey of curious and exotic identities connected with Bernoulli
numbers.

References

[A1] I.V. Artamkin : lementarnoe dokazatelstvo todestva Miki–

Zagira–Gessel , Uspehi mat. nauk , 62 (2007), 165–166. [I. Artamkin : An
elementary proof of the Miki–Zagier–Gessel identity, Russian Mathematical Surveys, 62
(2007), 1195–1196.]
[A2] C. Faber and R. Pandharipande : Logarithmic series and Hodge integrals in the tautological
ring (with an appendix by D. Zagier, Polynomials arising from the tautological ring), Mich.
Math. J., 48 (2000), 215–252 (appendix: pp. 240–252).
[A3] J. Frame : The Hankel power sum matrix inverse and the Bernoulli continued fraction,
Math. Comp., 33 (1979), 15–826.
[A4] I. Gessel : On Miki’s identity for Bernoulli numbers, J. Number Theory, 110 (2005),
75–82.
[A5] R. Hall, J. Wilson and D. Zagier : Reciprocity formulae for general Dedekind–Rademacher
sums, Acta Arithmetica, 73 (1995), 389–396.
[A6] M. Kaneko : A recurrence formula for the Bernoulli numbers, Proc. Japan Acad., 71
(1995), 192–193.
262 References

[A7] S.K. Lando : Lekcii o proizvodw ih funkcih , Moskva, MCNMO (2002),

144 pages.
[A8] S. Lando : Lectures on Generating Functions, Student Mathematical Library, Amer. Math.
Soc., 27 (2003), 150 pages. [Dtranslation of [A7]]
[A9] Yu. Matiyasevich : Calculation of Riemann’s zeta function via interpolating determinants
(talk given at the MPI in March 2013). On the author’s website at http://logic.pdmi.ras.
ru/~yumat/personaljournal/artlessmethod.
[A10] H. Miki : A relation between Bernoulli numbers, J. Number Theory, 10 (1978), 297–302.
[A11] N. Nielsen : Traité élémentaire des nombres de Bernoulli, Gauthiers-Villars, Paris (1923),
398 pages.
[A12] H. Pan and Z.-W. Sun : New identities involving Bernoulli and Euler polynomials,
J. Comb. Theory, 113 (2006), 156–175.
[A13] Z.-W. Sun and H. Pan : Identities concerning Bernoulli and Euler polynomials, Acta
Arithmetica, 125 (2006), 21–39.
[A14] D. Zagier : Hecke operators and periods of modular forms, Israel Math. Conf. Proc., 3
(1990), 321–336.
[A15] D. Zagier : A modified Bernoulli number, Nieuw Archief voor Wiskunde, 16 (1998), 63–72.
References

1. Ahlfors , L.: Complex Analysis, 3rdedn. McGraw Hill (1979)

2. Akiyama, S., Tanigawa, Y.: Multiple zeta values at non-positive integers. Ramanujan J. 5(4),
327–351 (2001)
3. Andrianov, A.N.: Quadratic Forms and Hecke Operators. Grundlehren der mathematischen
Wissenshaften, vol. 286. Springer (1987)
4. Ankeny, N.C., Artin, E., Chowla, S.: The class-number of real quadratic number fields. Ann.
Math. 56(2), 479–493 (1952)
5. Apéry, R.: Irrationalité de .2/ et .3/. Astérisque 61, 11–13 (1979)
6. Arakawa, T.: Generalized eta-functions and certain ray class invariants of real quadratic fields.
Math. Ann. 260, 475–494 (1982) P1
7. Arakawa, T.: Dirichlet series nD1 .cot  n˛/=ns , Dedekind sums, and Hecke L-functions
for real quadratic fields. Comment. Math. Univ. St. Pauli 37, 209–235 (1988)
8. Arakawa, T.: A noteon the Hirzebruch sum. Comment. Math. Univ. St. Pauli 42, 81–92 (1993)
9. Arakawa, T., Kaneko, M.: Multiple zeta values, poly-Bernoulli numbers, and related zeta
functions. Nagoya Math. J. 153, 189–209 (1999)
10. Arakawa, T., Kaneko, M.: On poly-Bernoulli numbers. Comment. Math. Univ. St. Pauli 48,
159–167 (1999)
11. Barnes, E.: The genesis of the double gamma functions. Proc. Lond. Math. Soc. 31, 358–381
(1899)
12. Barnes, E.: The theory of the double gamma functions. Philos. Trans. Roy. Soc. (A) 196,
265–388 (1901)
13. Berggren, L., Borwein, J., Borwein, P.: Pi, A Source Book. Springer, New York (1997)
14. Berndt, B.C.: Dedekind sums and a paper of G. H. Hardy. J. Lond. Math. Soc. 13, 129–137
(1976)
15. Berndt, B.C., Evans, R.J., Williams, K.S.: Gauss and Jacobi Sums. Canadian Math. Soc.
Series of Monographs and Advanced Texts, vol. 21. Wiley-Interscience (1998)
16. Bernoulli, J.: Ars Conjectandi, in Werke, vol. 3, pp. 107–286. Birkhäuser (1975)
17. Biermann, K.-R.: Thomas Clausen, Mathematiker und Astronom. J. Reine Angew. Math. 216,
159–198 (1964)
18. Biermann, K.-R.: Kummer, Ernst Eduard, in Dictionary of Scientific Biography, vols. 7 & 8.
Charles Scribner’s Sons, New York (1981)
19. Bjerknes, C.A.: Niels-Henrik Abel, Tableau de sa vie et de son Action Scientifique.
Cambridge University Press, Cambridge (2012)
20. Brewbaker, C.: Lonesum .0; 1/-matrices and poly-Bernoulli numbers of negative index.
Master’s thesis, Iowa State University (2005)
21. Buchmann, J., Vollmer, U.: Binary quadratic forms. An algorithmic approach. Algorithms
and Computation in Mathematics, vol. 20. Springer, Berlin (2007)

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 263
in Mathematics, DOI 10.1007/978-4-431-54919-2, © Springer Japan 2014
264 References

22. Carlitz, L.: Arithmetic properties of generalized Bernoulli numbers. J. Reine Angew. Math.
202, 174–182 (1959)
ˇ ˇ:ˇC1
23. Clausen, T.: Ueber die Fälle, wenn die Reihe von der Form y D 1C ˛1   xC ˛:˛C1
1:2
 : C1 x 2 C
0 ˇ0 0 0 0 ˇ 0 :ˇ 0 C1 0 0
etc: ein Quadrat von der Form z D 1 C ˛1   0 ı"0 x C ˛ :˛1:2C1   0 : 0 C1  ı"0 :ı:"0 C1
C1 2
x C etc: hat.
J. Reine Angew. Math. 3, 89–91 (1828)
24. Clausen, T.: Über die Function sin ' C 212 sin 2' C 312 sin 3' C etc:. J. Reine Angew. Math.
8, 298–300 (1832)
25. Clausen, T.: Beweis, daß die algebraischen Gleichungen Wurzeln von der Form a Cbi haben.
Astron. Nachr. 17, 325–330 (1840)
26. Clausen, T.: Lehrsatz aus einer Abhandlung über die Bernoullischen Zahlen. Astron. Nachr.
17, 351–352 (1840)
27. Coates, J., Wiles, A.: Kummer’s criterion for Hurwitz numbers, in Algebraic number theory
(Kyoto Internat. Sympos., RIMS, Kyoto, 1976), pp. 9–23. Japan Soc. Promotion Sci., Tokyo
(1977)
28. Dilcher, K.: A Bibliography of Bernoulli Numbers, available online, http://www.mscs.dal.
ca/~dilcher/bernoulli.html
29. Dirichlet, P.G.L.: Lectures on Number Theory, with Supplements by R. Dedekind. History of
Mathematics Sources, vol. 16. Amer. Math. Soc. (1999)
30. Edwards, A.W.F.: A quick route to sums of powers. Am. Math. Monthly 93, 451–455 (1986)
31. Edwards, H.W.: Fermat’s Last Theorem. A Genetic Introduction to Algebraic Number Theory.
Springer, New York-Berlin (1977)
32. Euler, L.: Remarques sur un beau rapport entre les séries des puissances tant directes que
réciproques. Opera Omnia, series prima XV, 70–90 (1749)
33. Euler, L.: De numero memorabili in summatione progressionis harmonicae naturalis occur-
rente. Opera Omnia, series prima XV, 567–603 (1785)
34. Freudenthal, H.: Hurwitz, Adolf, in Dictionary of Scientific Biography, vols. 5 & 6. Charles
Scribner’s Sons, New York (1981)
35. Gauss, C.F.: DisquisitionesArithmeticae (1801)
36. Gould, H.W.: Explicit formulas for Bernoulli numbers. Am. Math. Monthly 79, 44–51 (1972)
37. Gouvéa, F.Q.: p-adic Numbers, an Introduction. Springer
38. Graham, R., Knuth, D., Patashnik, O.: Concrete Mathematics. Addison-Wesley (1989)
39. Hashimoto, K.: Representation of the finite symplectic group Sp.2; Fp / in the space of Siegel
modular forms. Contemporary Math. 53, 253–276 (1986)
40. Hensel, K.: Festschrift zur Feier des 100. Geburtstages Eduard Kummers, B.G. Teubner, 1910.
(Kummer’s collected papers, 33–69)
41. Hilbert, D.: Adolf Hurwitz, Gedächtnisrede, Nachrichten von der k. Gesellshaft der Wis-
senschaften zu Göttingen (1920), pp. 75–83 (Hurwitz’s Mathematische Werke, I xiii–xx)
42. Hungerford, T.W.: Algebra, Graduate Texts in Mathematics, vol. 73. SpringerP(1974) D 1
43. Hurwitz, A.: Einige Eigenschaften der Dirichlet’schen Funktionen F .s/ D n
 ns , die
bei der Bestimmung der Klassenanzahlen binärer quadratischer Formen auftreten. Zeitschrift
für Math. und Physik 27, 86–101 (1882). (Mathematische Werke I, 72–88)
44. Hurwitz, A.: Über die Entwicklungskoeffizienten der lemniskatischen Funktionen. Math.
Ann. 51, 196–226 (1899). (Mathematische Werke II, 342–373)
45. Ibukiyama, T.: On some elementary character sums. Comment. Math. Univ. St. Pauli 47, 7–13
(1998)
46. Ibukiyama, T., Saito, H.: On zeta functions associated to symmetric matrices and an explicit
conjecture on dimensions of Siegel modular forms of general degree. Int. Math. Res. Notices
8, 161–169 (1992)
47. Ibukiyama, T., Saito, H.: On a L-functions of ternary zero forms and exponential sums of Lee
and Weintraub. J. Number Theory 48-2, 252–257 (1994)
48. Ibukiyama, T., Saito, H.: On zeta functions associated to symmetric matrices I. Am. J. Math.
117-5, 1097–1155 (1995); II: Nagoya Math. J. 208, 265–316 (2012); III: Nagoya Math. J.
146, 149–183 (1997)
References 265

49. Ibukiyama, T., Saito, H.: On “easy” zeta functions (trans. by Don Zagier). Sugaku Exposition,
14(2), 191–204 (2001). Originally in Sugaku, 50–1, 1–11 (1998)
50. Ireland, K., Rosen, M.: A Classical Introduction to Modern Number Theory, 2nd edn.
Graduate Texts in Mathematics, vol. 84. Springer (1990)
51. Iwasawa, K.: Lectures on p-adic L-functions. Annals of Math. Studies, vol. 74. Princeton
University Press, Princeton (1972)
52. Jacobi, C.G.J.: De usu legitimo formulae summatoriae Maclaurinianae. J. Reine Angew.
Math. 12, 263–272 (1834). (Mathematische Werke VI, 64–75)
53. Jordan, Ch.: Calculus of Finite Differences, Chelsea Publication, New York (1965). (First
edition, Budapest, 1939)
54. Kaneko, M.: A generalization of the Chowla-Selberg formula and the zeta functions of
quadratic orders. Proc. Jpn. Acad. 66(A)-7, 201–203 (1990)
55. Kaneko, M.: A recurrence formula for the Bernoulli numbers. Proc. Jpn. Acad. 71(A)-8,
192–193 (1995)
56. Kaneko, M.: Poly-Bernoulli numbers. J. Th. Nombre Bordeaux 9, 199–206 (1997)
57. Kaneko, M.: Multiple zeta values. Sugaku Expositions 18(2), 221–232 (2005)
58. Katz, N.: The congruences of Clausen-von Staudt and Kummer for Bernoulli-Hurwitz
numbers. Math. Ann. 216, 1–4 (1975)
59. Knuth, D.: Two notes on notation. Am. Math. Monthly 99, 403–422 (1992)
60. Knuth, D.: Johann Faulhaber and sum of powers. Math. Comp. 61(203), 277–294 (1993)
61. Knuth, D., Buckholtz, T.J.: Computation of tangent Euler and Bernoulli numbers. Math.
Comp. 21, 663–688 (1967)
62. Kronecker, L.: Bemerkungen zur Abhandlung des Herrn Worpitzky. J. Reine Angew. Math.
94, 268–270 (1883). (Mathematische Werke II, 405–407)
˛ˇ ˛.˛C1/ˇ.ˇC1/
63. Kummer, E.E.: Über die hypergeometrische Reihe 1 C 1 x C 12. C1/ x 2 C
˛.˛C1/.˛C2/ˇ.ˇC1/.ˇC2/ 3
123. C1/. C2/
x C : : : . J. Reine Angew. Math. 15, 39–83, 127–172 (1836).
(Collected papers II, 75–166)
64. Kummer, E.E.: Über eine allgemeine Eigenschaft der rationalen Entwicklungscoefficienten
einer bestimmten Gattung analytischer Functionen. J. Reine Angew. Math. 41, 368–372
(1851). (Collected papers I, 358–362)
65. Lampe, E.: Nachruf für Ernst Eduard Kummer. Jahresbericht der Deutschen Mathematiker-
vereinigung 3, 13–21 (1894). (Kummer’s collected papers I, 15–30)
66. Lang, S.: Cyclotomic Fields, Graduate Texts in Mathematics, vol. 59. Springer (1980)
67. Lang, S.: Elliptic Functions, 2nd edn. Graduate Texts in Mathematics, vol. 112. Springer
(1987). (Original edition, Addison-Wesley Publishing, 1973)
68. Launois, S.: Combinatorics of H-primes in quantum matrices. J. Algebra 309(1), 139–167
(2007)
69. Lee, R., Weintraub, S.H.: On a generalization of a theorem of Erich Hecke. Proc. Natl. Acad.
Sci. USA 79, 7955–7957 (1982)
70. Leopoldt, H.-W.: Eine Verallgemeinerung der Bernoullischen Zahlen. Abh. Math. Sem. Univ.
Hamburg 22, 131–140 (1958)
71. Lindemann, F.: Ueber die Zahl . Math. Ann. 20, 213–225 (1882)
72. Meissner, E.: Gedächtnisrede auf Adolf Hurwitz. Vierteljahrsschrift der Naturforschenden
Gesellschaft in Zürich. 64, 855–857 (1919)
73. Nielsen, N.: Handbuch der Theorie der Gammafunktion. Chelsea Publication, New York
(1965). (First edition, Leipzig, 1906)
74. Noether, M.: Zur Erinnerung an Karl Georg Christian von Staudt. Jahresbericht d. Deutschen
Mathem.-Vereinigung 32, 97–119 (1923)
75. Nörlund, N.E.: Vorlesungen über Differenzenrechnung. Chelsea Publication, New York
(1954). (First edition, Springer, Berlin, 1924)
76. Pólya, G.: Some mathematicians I have known. Am. Math. Monthly 76, 746–753 (1969)
77. Rademacher, H.: Lectures on Elementary Number Theory, Die Grundlehren der mathematis-
chen Wissenschaften, vol. 169. Springer (1773)
266 References

78. Reid, C.: Hilbert. Springer, Berlin-Heidelberg-New York (1970)

79. Ribenboim, P.: 13 Lectures on Fermat’s Last Theorem. Springer, New York-Heidelberg-
Berlin (1979)
80. Rivoal, T.: La fonction zêta de Riemann prend une infinité de valeurs irrationnelles aux entiers
impairs. C. R. Acad. Sci. Paris, Sér. l. Math. 331, 267–270 (2000)
81. Scherk, H.F.: Über einen allgemeinen, die Bernoullischen Zahlen und die Coëfficienten der
Secantenreihe zugleich darstellenden Ausdruck. J. Reine Angew. Math. 4, 299–304 (1829)
82. Serre, J.-P.: Formes modulaires et fonctions zêta p-adiques. Lect. Notes in Math., vol. 350,
pp. 191–268. Springer (1973). (Collected papers III, 95–172)
83. Serre, J.-P.: Cours d’arithmétique, Presses Universitaires de France, 1970. English translation:
A course in arithmetic, Graduate Text in Mathematics, vol. 7. Springer (1973)
84. Seki, T.: CollectedWorks of Takakazu Seki. In: Hirayama, A., Shimodaira, K., Hirose, H.
(eds.) Osaka Kyouiku Tosho (1974)
85. Shintani, T.: On a Kronecker limit formula for real quadratic fields. J. Fac. Sci. Univ. Tokyo,
Sec. IA 24, 167–199 (1977)
86. Shintani, T.: On special values of L-functions of number fields. Sugaku 29–3, 204–216 (1977)
87. Shintani, T.: On certain ray class invariants of real quadratic fields. J. Math. Soc. Jpn. 30,
139–167 (1978)
88. Slavutskii, I.Sh.: Staudt and arithmetical properties of Bernoulli numbers. Historia Scien-
tiarum 5–1, 69–74 (1995)
89. Stark, H.M.: L-functions at s D 1, III. Totally real fields and Hilbert’s twelfth problem.
Advances in Math. 22, 64–84 (1976)
90. Titchmarsh, E.C.: TheTheory of the Riemann Zeta-function, 2nd edn. (revised by D.R. Heath-
Brown). Oxford, (1986)
91. Tsushima, R.: The spaces of Siegel cusp forms of degree two and the representation of
Sp.2; Fp /. Proc. Jpn. Acad. 60, 209–211 (1984)
92. Tsushima, R.: Dimension formula for the spaces of Siegel cusp forms and a certain
exponential sum. Mem. Inst. Sci. Tech. Meiji Univ. 36, 1–56 (1997)
93. Vandiver, H.S.: On developments in an arithmetic theory of the Bernoulli and allied numbers.
Scripta Math. 25, 273–303 (1961)
94. Volkenborn, A.: Ein p-adisches Integral und seine Anwendungen. I. Manuscripta Math. 7,
341–373 (1972)
95. Volkenborn, A.: Ein p-adisches Integral und seine Anwendungen. II. Manuscripta Math. 12,
17–46 (1974)
96. von Staudt, K.G.C.: Beweis eines Lehrsatzes, die Bernoullischen Zahlen betreffend. J. Für
Reine u. Angew. Math. 21, 372–374 (1840)
97. von Staudt, K.G.C.: Beweis des Satzes, daß jede algebraische rationale ganze Function von
einer Veränderlichen in Faktoren vom ersten Grade aufgelöst werden kann. J. Für Reine u.
Angew. Math. 29, 97–102 (1845)
98. Waldschmidt, M.: Valeurs zêta multiples. Une introduction. Colloque International de Théorie
des Nombres (Talence, 1999). J. Théor. Nombres Bordeaux 12(2), 581–595 (2000)
99. Waldshcmidt, M.: Open diophantine problems. Moscow Math. J. 4–1, 245–300 (2004)
100. Washington, L.C.: Introduction to Cyclotomic Fields. Graduate Text in Mathematics, vol. 83.
Springer (1982)
101. Weber, H.: Lehrbuchder Algebra, vol. III, Chelsea Publication, New York. (First edition,
Friedrich Vieweg und Sohn, Braunschweig, 1908)
102. Weil, A.: Basic Number Theory. Springer, New York (1973)
103. Weil, A.: Number Theory: An Approach Through History; From Hammurapi to Legendre.
Birkhäuser, Boston (1983)
104. Whittaker, E.T., Watson, G.N.: A Course of Modern Analysis. Cambridge University Press,
Cambridge (1927)
105. Zagier, D.: A Kronecker limit formula for real quadratic fields. Math. Ann. 213, 153–184
(1975)
References 267

106. Zagier, D.: Modular forms whose Fourier coefficients involve zeta functions of quadratic
fields, in Modular functions of one variable VI. Lect. Notes in Math., vol. 627, pp. 105–169.
Springer (1977)
107. Zagier, D.: Zetafunktionen und Quadratische Körper. Springer (1981)
108. Zagier, D.: A one-sentence proof that every prime p  1 .mod 4/ is a sum of two squares.
Am. Math. Monthly 97–2, 144 (1990)
109. Zagier, D.: Values of zeta functions and their applications, in ECM volume. Progress Math.
120, 497–512 (1994)
Index

Page numbers shown in bold indicate biographical note.

Symbols B
L-function Bachmann, 49
p-adic —, 43, 183 Barnes, 210
— of prehomogeneous vector space, 174, — double zeta function, 212, 214
175 contour integral representation of —,
Dirichlet —, 148, 176 214
p-adic transformation formula of —, 214, 216
— L-function, 43, 183 — multiple zeta function, 209–211
— integral, 44 contour integral representation of —,
— modular form, 41, 96 210
— number field, 183, 184 Berndt, 217
measure on — number field, 184 Bernoulli measure, 196, 199
ring of — integers, 183 Bernoulli number
p-adic integral expression of —, 198
— and class number, 95
A — and class number of imaginary quadratic
Abel, 46, 203 field, 90
Academia Algebrae, 3 — of higher order, 224
Akiyama, 36 definition of —, 3
Akiyama–Tanigawa algorithm, 36 denominator of —, 41
algebraic irrational number, 212 different definition of —, 6
ambig, 92, 178 formula of — involving Stirling number, 35
Ankeny, 96 generalized —, 53, 58, 107, 110, 111, 127
Ankeny–Artin–Chowla congruence, 96 generating function of —s, 53
Ars Conjectandi, 1 generating function of —s, 13, 20, 23
Artin, 96 poly- —, 223
asteroid recurrence formula of —, 3, 12, 22
Ceres, 107 sign of —, 23
Pallas, 47, 107 table of —, 6
Vesta, 107 Bernoulli polynomial, 10, 54, 55, 65, 141, 144,
Astronomische Nachrichten, 46 209, 217
automorphic factor, 220 definition of —, 55
automorphic form, 209 Fourier expansion of —, 59
dimension formula of —s, 124, 174, 177 generating function of —s, 56

T. Arakawa et al., Bernoulli Numbers and Zeta Functions, Springer Monographs 269
in Mathematics, DOI 10.1007/978-4-431-54919-2, © Springer Japan 2014
270 Index

Bernoulli, Daniel, 1 — representation of Hurwitz zeta function,

Bernoulli, Jakob, 1–3, 6 143
Bernoulli, Johann, 1 convolution, 188, 195
Bessel, 46 cycle, 27
Bessel function, 46 cyclotomic field, 49, 105
Biermann, 46 arithmetic of —, 43
binomial coefficient, 2 class number of —, 43
binomial expansion, 190
binomial polynomial, 185
Bleuler, 207 D
bounded linear functional, 191, 192 de Moivre, 2
— and measure, 192 de Moivre’s formula, 2
Buckholtz, 24 Dedekind domain, 78
Dedekind zeta function, 156
analytic continuation of —, 166
C functional equation of —, 166
Cantor, 49 di-Bernoulli number, 227, 233
Cauchy, 95 denominator of —, 230
Cauchy’s integral theorem, 143 difference, 55
character difference calculus, 32
conductor of —, 52 Dilcher, 3
conjugate —, 53 dilogarithm, 47
Dirichlet —, 51 dimension formula, 124, 174, 177
even —, 53 Diophantine approximation, 213
odd —, 53 Dirichlet, 48, 55, 95
primitive —, 52 — L-function, 148, 176
trivial —, 51 functional equation of —, 150, 151
character sum, 103, 104 special value of —, 148, 152
characteristic function, 191 — character, 51
Chinese remainder theorem, 110, 169 — series, 212
Chowla, 96 Dirichlet’s Theorem, 55
class number, 82, 95 discriminant
— and Bernoulli number, 90 — of order of quadratic field, 79
— formula, 55, 89, 95, 155, 162, 173 — of quadratic field, 78
— in the narrow sense, 82 — of quadratic form, 76
— in the wide sense, 82 Disquisitiones Arithmeticae, 91, 92
— of cyclotomic field, 43 distribution property, 184
— of order of quadratic field, 82, 159 Dorpat observatory, 46
— of quadratic forms, 77 double gamma function, 209
finiteness of —, 88 du Bois–Reymond, 49
Clausen, 41, 46, 47, 48
— function, 47
— identity, 47 E
theorem of — and von Staudt, 25, 41, 95 Eisenstein, 41
commutative integral domain, 13 Eisenstein series, 41
conductor Euler, 22, 47, 65, 73, 102, 205
— of Dirichlet character, 52 — factor, 170, 171
— of order of quadratic field, 79 — function, 52, 87, 109, 187
congruence zeta function, 132 — number, 102
conjugate character, 53 — product, 55, 156
continued fraction, 218 —’s constant, 69
contour integral, 141 —’s formula, 67
— representation of Barnes multiple zeta —’s theorem, 52
function, 210 criterion of —, 99
Index 271

formula of —, 95 generating function of —s, 53

summation formula of — and Maclaurin, genus theory, 92
65, 73 Geometrie der Lage, 47
even character, 53 Gordan, 49
expansion of cotangent, 23 Gould, 34
expansion of tangent, 23 group of totally positive units, 216, 219, 221
exponential sum, 104, 107, 110, 135

H
F Hashimoto, 174
Faulhaber, 3 Hensel, 46
Faulhaber’s theorem, 11 Hilbert, 207
Fermat, 42, 47, 205 Holst, 46
— number, 47 Hurwitz, 95, 139, 203, 207
—’s little theorem, 42, 237 — integral series, 97
formal Laurent series, 20 — number, 203
formal power series, 13, 14 — zeta function, 139, 209, 210
derivative of —, 19 contour integral representation of —,
integral of —, 20 143
inverse of —, 18 functional equation of —, 147
product of —, 14 values at non-positive integers of —,
substitution of —, 16 144
sum of —, 14 hyperbolic function, 23
Fourier, 163 hypergeometric series, 47, 48
Fourier transform, 163, 193
Freudenthal, 207
Frobenius, 207 I
Fuchs, 49 ideal
fundamental discriminant — and quadratic form, 82
— of quadratic field, 78 — class group, 159
fundamental domain, 221 — group, 159
fundamental period of continued fraction — of order of quadratic field, 79
expansion, 218 equivalence of — (in the narrow, wide
Fundamental theorem of finitely generated sense), 82
abelian groups, 78 fractional —, 157
fundamental theorem on algebra, 48 inverse —, 158
fundamental unit, 90, 91, 219 norm of —, 79
primitive —, 79
proper —, 80
G proper fractional —, 158
Galois theory, 105 standard basis of —, 79
gamma function, 25, 139, 140, 147, 165 imaginary quadratic field, 77
duplication formula of —, 147 indeterminate, 14, 105
reflection formula of —, 147 inverse Fourier transform, 193
Gauss, 47, 47–49, Ireland, 230, 232
75, 91, 92, 105, 107 irregular prime, 43
—ian sum, 104, 107, 108, 111 Iwasawa, 183
— of quadratic form, 108 Iwasawa isomorphism, 186, 188
formula of —, 110
sign of —, 107
conjecture of —, 91 J
Gauss symbol, 59 Jacobi, 3, 46, 48, 203
generalized Bernoulli number, 53, 58, 107, Joachimsthal, 48
110, 111, 127 Joseph von Utzschneider Optical Institute, 46
272 Index

K multiple zeta function, 210

Katsuyou Sanpou, 2 multiple zeta value, 36, 224
Klein, 207
Knuth, 24, 25, 28
Kronecker, 34, 48, 207 N
Kronecker’s delta, 29, 114 Nielsen, 25
Kummer, 43, 44, 48, 48, 49, 207 Noether, 46, 47
— congruence, 43, 48, 183, 198 norm
— surface, 49 — of element of quadratic field, 78
— of ideal, 79

L
O
l’Hôpital, 106
Ochiai, 234
Lampe, 46
odd character, 53
Lang, 155, 168, 183, 188, 189
Olbers, 107
lattice, 157, 216
order
— of quadratic field, 157
— of lattice, 216
double zeta function of —, 220
— of quadratic field, 78
order of —, 216
product of —, 157
Law of Large Numbers, 1
P
Lebesgue, 210
Pólya, 207, 208
— dominant convergence theorem, 210
period of continued fraction expansion, 218
Lebesgue integral, 210
Peters, 47
Leibniz, 115
Piazzi, 107
Leibniz rule, 115
Poisson, 162
Lerch, 209
Poisson summation formula, 163
Lerch type zeta function, 209
poly-Bernoulli number, 223
Lindemann, 73, 207
denominator of —, 227
formula of — in terms of Stirling number,
226
M generating function of —s, 223
Möbius, 170 recurrence formula of —, 224, 225
— function, 170 table of —, 224, 234
— inversion formula, 171 polylogarithm, 223
Maclaurin, 65 prehomogeneous vector space
Mahler, 189 zeta function of —, 135, 155, 177
theorem of —, 189 primitive
Maple, 6 — character, 52
Mathematica, 6 — ideal, 79
maximal order, 77 principal ideal domain, 78
measure projective geometry, 47
— and formal power series, 185, 186, 188, proper ideal, 80
190 purely periodic continued fraction expansion,
— on p-adic number field, 184 218
convolution of —, 188, 195
Mendelssohn, 48
Minkowski, 207 Q
modular form quadratic field, 77
p-adic —, 96 (fundamental) discriminant of —, 78
modular group, 76, 213 class number of order of —, 82
modulus of Dirichlet character, 51 discriminant of order of —, 79
multiple gamma function, 209 ideal group of —, 155
Index 273

imaginary —, 77 Seki, 2, 3, 6
integer of —, 77 Serre, 55, 183
maximal order of —, 77 Shintani, 209
norm of element of —, 78 simplicial cone, 221
order of —, 78 standard basis, 79
class number formula of —, 89 Stark, 209
conductor of —, 79 Stark–Shintani conjecture, 209
ideal of —, 79 Stirling, 25
real —, 77 Stirling number, 25
class field construction over —, 209 — of the first kind, 27, 28, 112, 113
ring of integers of —, 77 recurrence formula of —, 28
trace of element of —, 78 table of —, 29
quadratic form, 75 — of the second kind, 25, 26, 116
— and ideal, 82 recurrence formula of —, 26
— with square discriminant, 86 table of —, 27
class number formula of —s, 87 alternative definition of —, 33
class number of —s, 77 various formulas of —, 28
discriminant of —, 76 structure theorem for finitely generated abelian
equivalence of —, 76 groups, 157
primitive —, 75 sum of powers, 1, 10
quadratic residue proof of formula of —, 6, 58, 73, 116
— symbol, 88 symmetric group, 27
reciprocity law of —, 89 symmetric matrix
quartic surface, 49 equivalence of —, 76
half-integral —, 76, 173
prehomogeneous vector space of symmetric
R matrices, 177
real quadratic field, 77 primitive —, 76
real quadratic irrational number, 216
continued fraction expansion of —, 218
reduced —, 219 T
reduced real quadratic irrational number, 219 tangent number, 24, 98
regular prime, 43 Tanigawa, 36
residue theorem, 60, 141 theorem of Clausen and von Staudt, 25, 41, 95
Riemann, 71 theorem of l’Hôpital, 106
Riemann zeta function, 22, 67, 71, 139, 174 theta function, 108, 162, 164
analytic continuation of —, 71 transformation formula of —, 162, 165
functional equation of —, 148 totally real algebraic number field, 209
special values of —, 6 trace formula, 127
values at negative integers of —, 66, 72, 96 trace of element of quadratic field, 78
values at positive even integers of —, 22, transcendental number, 73, 207
61, 73 trivial character, 51
ring of all integers, 77
Rosen, 230, 232 U
Roth, 213 unique factorization domain, 78
Roth’s theorem, 213 uniqueness of prime ideal decomposition, 156
upper half plane, 162
Utzschneider, 46
S
Scherk, 48
Schönflies, 49 V
Schubert, 207 Vandiver, 232, 237, 238
Schumacher, 46 von Staudt, 34, 41, 47
Schwarz, 49, 207 theorem of Clausen and —, 25, 41, 95
274 Index

W Barnes double —, 212, 214

Washington, 43, 183 contour integral representation of —,
Watson, 139 214
Weber, 155, 168 Barnes multiple —, 209, 210
Weierstrass, 207 special values of —, 211
Weil, 73 congruence — of algebraic curve, 132
Whittaker, 139 Dedekind —, 156
Wilson, 229 double — of lattice, 220
theorem of —, 229 Hurwitz —, 139, 143, 209, 210
functional equation of —, 147
values at non-positive integers of —,
Z 144
Zagier, 62, 168, 218, 221 Lerch type —, 209
zeta function Riemann —, 22, 67, 71, 96, 139
— of ideal class, 164 analytic continuation of —, 71
functional equation of —, 166 functional equation of —, 148
analytic continuation of —, 166 special values of —, 6
— of order of quadratic field, 168, 170 values at negative integers of —, 66, 72
— of prehomogeneous vector space, 135, values at positive even integers of —,
155, 177 22, 61, 73