Lecture Notes in Mathematics 1877

Editors:
J.-M. Morel, Cachan
F. Takens, Groningen
B. Teissier, Paris
Jrn Steuding

Value-Distribution
of L-Functions

ABC
Author
Jrn Steuding
Department of Mathematics
Chair of Complex Analysis
University of Wrzburg
Am Hubland
97074 Wrzburg
Germany
e-mail: [email protected]

Library of Congress Control Number: 2007921875

Mathematics Subject Classification (2000): 11M06, 11M26, 11M41, 30D35, 30E10,

60B05, 60F05

ISSN print edition: 0075-8434

ISSN electronic edition: 1617-9692
ISBN-10 3-540-26526-0 Springer Berlin Heidelberg New York
ISBN-13 978-3-540-26526-9 Springer Berlin Heidelberg New York

DOI 10.1007/978-3-540-44822-8

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,
reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,
1965, in its current version, and permission for use must always be obtained from Springer. Violations are
liable for prosecution under the German Copyright Law.
Springer is a part of Springer Science+Business Media
springer.com
c Springer-Verlag Berlin Heidelberg 2007

The use of general descriptive names, registered names, trademarks, 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.
Typesetting by the author and SPi using a Springer LATEX macro package
Cover design: WMXDesign GmbH, Heidelberg
Printed on acid-free paper SPIN: 11504160 VA41/3100/SPi 543210
Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IX

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 The Riemann Zeta-Function and the Distribution
of Prime Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Bohrs Probabilistic Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Voronins Universality Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Dirichlet L-Functions and Joint Universality . . . . . . . . . . . . . . . . 19
1.5 L-Functions Associated with Newforms . . . . . . . . . . . . . . . . . . . . . 24
1.6 The LinnikIbragimov Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . 28

2 Dirichlet Series and Polynomial Euler Products . . . . . . . . . . . . 35

2.1 General Theory of Dirichlet Series . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2 A Class of Dirichlet Series: The Main Actors . . . . . . . . . . . . . . . . 37
2.3 Estimates for the Dirichlet Series Coecients . . . . . . . . . . . . . . . 40
2.4 The Mean-Square on Vertical Lines . . . . . . . . . . . . . . . . . . . . . . . . 43

3 Interlude: Results from Probability Theory . . . . . . . . . . . . . . . . 49

3.1 Weak Convergence of Probability Measures . . . . . . . . . . . . . . . . . 49
3.2 Random Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3 Denjoys Probabilistic Argument for Riemanns Hypothesis . . . 54
3.4 Characteristic Functions and Fourier Transforms . . . . . . . . . . . . 56
3.5 Haar Measure and Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.6 Random Processes and Ergodic Theory . . . . . . . . . . . . . . . . . . . . 58
3.7 The Space of Analytic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 59

4 Limit Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.1 Associated Random Elements and the Main Limit Theorem . . . 63
4.2 Limit Theorems for Dirichlet Polynomials . . . . . . . . . . . . . . . . . . 67
4.3 An Ergodic Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.4 Approximation in the Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
VI Contents

4.5 A Limit Theorem for Absolutely Convergent Series . . . . . . . . . . 76

4.6 Proof of the Main Limit Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.7 Generalizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.8 A Discrete Limit Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

5 Universality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.1 Dense Sets in Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2 Application to the Space of Analytic Functions . . . . . . . . . . . . . . 94
5.3 Entire Functions of Exponential Type . . . . . . . . . . . . . . . . . . . . . . 96
5.4 The Positive-Density Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.5 The Support of the Limit Measure . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.6 The Universality Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.7 Discrete Universality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

6 The Selberg Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

6.1 Denition and First Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.2 Primitive Functions and the Selberg Conjectures . . . . . . . . . . . . 116
6.3 Non-Vanishing and Prime Number Theorems . . . . . . . . . . . . . . . 119
6.4 Pair Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.5 The PhragmnLindelf Principle . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.6 Universality in the Selberg Class . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.7 Lindelfs Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.8 Symmetric Power L-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

7 Value-Distribution in the Complex Plane . . . . . . . . . . . . . . . . . . 137

7.1 Sums Over c-Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.2 Riemannvon Mangoldt-Type Formulae . . . . . . . . . . . . . . . . . . . . 142
7.3 Nevanlinna Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
7.4 Uniqueness Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

8 The Riemann Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

8.1 Uniform Approximation and Zeros . . . . . . . . . . . . . . . . . . . . . . . . . 155
8.2 Bagchis Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.3 A Generalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
8.4 An Approach Towards Riemanns Hypothesis? . . . . . . . . . . . . . . 162
8.5 Further Equivalents of the Riemann Hypothesis . . . . . . . . . . . . . 163

9 Eective Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

9.1 The Problem of Eectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
9.2 Upper Bounds for the Density of Universality . . . . . . . . . . . . . . . 169
9.3 Value-Distribution on Arithmetic Progressions . . . . . . . . . . . . . . 174
9.4 Making Universality Visible . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
9.5 Almost Periodicity in the Half-Plane
of Absolute Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
9.6 Eective Inhomogeneous Diophantine Approximation . . . . . . . . 182
9.7 c-Values Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
 Contents VII

10 Consequences of Universality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

10.1 Dense Sets in the Complex Plane . . . . . . . . . . . . . . . . . . . . . . . . . . 193
10.2 Functional Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
10.3 Joint Functional Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
10.4 Anderssons Disproof of a Mean-Square Conjecture . . . . . . . . . . 200
10.5 Voronins Theorems and Physics . . . . . . . . . . . . . . . . . . . . . . . . . . 201
10.6 Shifts of Universal Dirichlet Series . . . . . . . . . . . . . . . . . . . . . . . . . 202

11 Dirichlet Series with Periodic Coecients . . . . . . . . . . . . . . . . . 209

11.1 Zero-Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
11.2 A Link to the Selberg Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
11.3 Strong Universality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
11.4 Hurwitz Zeta-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

12 Joint Universality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

12.1 A Joint Limit Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
12.2 A Transfer Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
12.3 Twisted L-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
12.4 First Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
12.5 A Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246

13 L-Functions of Number Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

13.1 Dedekind Zeta-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
13.2 Grssencharacters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
13.3 Hecke L-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
13.4 Universality for Hecke L-Functions . . . . . . . . . . . . . . . . . . . . . . . . 260
13.5 Artins Reciprocity Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
13.6 Artin L-Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
13.7 The Artin Conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
13.8 Joint Universality for Artin L-Functions . . . . . . . . . . . . . . . . . . . . 274
13.9 L-Functions to Automorphic Representations . . . . . . . . . . . . . . . 278

A Appendix: A Short History of Universality . . . . . . . . . . . . . . . . 285

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293

Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
Preface

L-functions are important objects in modern number theory. They are gener-
ating functions formed out of local data associated with either an arithmetic
object or with an automorphic form. They can be attached to smooth pro-
jective varieties dened over number elds, to irreducible (complex or p-adic)
representations of the Galois group of a number eld, to a cusp form or to
an irreducible cuspidal automorphic representation. All the L-functions have
in common that they can be described by an Euler product, i.e., a product
taken over prime numbers. In view of the unique prime factorization of integers
L-functions also have a Dirichlet series representation. The famous Riemann
zeta-function


1   1
1
(s) = = 1 s
n=1
ns p
p prime

may be regarded as the prototype. L-functions encode in their value-

distribution information on the underlying arithmetic or algebraic structure
that is often not obtainable by elementary or algebraic methods. For instance,
Dirichlets class number formula gives information on the deviation from
unique prime factorization in the ring of integers of quadratic number elds
by the values of certain Dirichlet L-functions L(s, ) at s = 1. In particu-
lar, the distribution of zeros of L-functions is of special interest with respect
to many problems in multiplicative number theory. A rst example is the
Riemann hypothesis on the non-vanishing of the Riemann zeta-function in
the right half of the critical strip and its impact on the distribution of prime
numbers. Another example are L-functions L(s, E) attached to elliptic curves
E dened over Q. The yet unproved conjecture of Birch and Swinnerton-Dyer
claims that L(s, E) has a zero at s = 1 whose order is equal to the rank of
the Mordell-Weil group of the elliptic curve E.
These notes present recent results in the value-distribution theory of such
L-functions with an emphasis on the phenomenon of universality. The starting
point of this theory is Bohrs achievement at the rst half of the twentieth
X Preface

century. He proved denseness results and rst limit theorems for the values of
the Riemann zeta-function. Maybe the most remarkable result concerning the
value-distribution of (s) is Voronins universality theorem from 1975, which
roughly states that any non-vanishing analytic function can be approximated
uniformly by certain shifts of the zeta-function in the critical strip. More
precisely: let 0 < r < 14 and suppose that g(s) is a non-vanishing continuous
function on the disc |s| r which is analytic in its interior. Then, for any
 > 0, there exists a real number such that
   
 3 
max  s + + i g(s) < ;
|s|r 4

moreover, the set of these has positive lower density:

     
1  3 
 
lim inf meas [0, T ] : max  s + + i g(s) <  > 0.
T T |s|r 4

This is a remarkable property! We say that (s) is universal since it allows

uniform approximation of a large class of functions. Voronins universality
theorem, in a spectacular way, indicates that Riemanns zeta-function is a
transcendental function; clearly, rational functions cannot be universal. In
some literature the validity of the Riemann hypothesis for abelian varieties
(proved by Hasse for elliptic curves and by Weil in the general case) is regarded
as evidence for the truth of Riemanns hypothesis for (s). However, the
zeta-function of an abelian variety is a rational function and so its value-
distribution is of a rather dierent type.
The LinnikIbragimov conjecture asserts that any Dirichlet series (which
has a suciently rich value-distribution) is universal. Meanwhile we know
quite many universal Dirichlet series; for instance, Dirichlet L-functions
(Voronin, 1975), Dedekind zeta-functions (Reich, 1980), Lerch zeta-functions
(Laurinikas, 1997), and L-functions associated with newforms (Laurinikas,
Matsumoto and Steuding, 2003). One aim of these notes is to prove an exten-
sion of Voronins universality theorem for a large class of L-functions which
covers (at least conjecturally) all known L-functions of number-theoretical
signicance.
These notes are organized as follows. In the introduction, we give an
overview on the value-distribution theory of the classical Riemann zeta-
function and Dirichlet L-functions; also we touch some allied zeta-functions
which we will not consider in detail in the following chapters. In Chap. 2,
we introduce a class S of Dirichlet series, satisfying certain analytic and
arithmetic axioms. The members of this class are the main actors in the
sequel. Roughly speaking, an L-function in S has a polynomial Euler product
and satises some hypothesis which may be regarded as some kind of prime
number theorem; besides, we require analytic continuation to the left of
the half-plane of absolute convergence for the associated Dirichlet series in
addition with some growth condition. The axioms dening S are kept quite
 Preface XI

general and therefore they may appear to be rather abstract and technical;
however, as we shall discuss later for many examples (in Chaps. 6, 12 and
13), they hold (or at least they are expected to hold) for all L-functions of
number theoretical interest. This abstract setting has the advantage that we
can derive a rather general universality theorem.
Our proof of universality, in the main part, relies on Bagchis probabilistic
approach from 1981. For the sake of completeness we briey present in Chap. 3
some basic facts from probability theory and measure theory. In Chap. 4, we
prove along the lines of Laurinikas extension of Bagchis method a limit
theorem (in the sense of weakly convergent probability measures) for functions
in the class S. In the following chapter we give the proof of the main result, a
universality theorem for L-functions in S. The proof depends on the limit theo-
rem of the previous chapter and the so-called positive density method, recently
introduced by Laurinikas and Matsumoto to tackle L-functions attached to
cusp forms. Furthermore, we discuss the phenomenon of discrete universality;
here the attribute discrete means that the shifts are taken from arithmetic
progressions. This concept of universality was introduced by Reich in 1980.
In Chap. 6, we introduce the Selberg class S consisting of Dirichlet
series with Euler product and a functional equation of Riemann-type (and
a bit more). It is a folklore conjecture that the Selberg class consists of all
automorphic L-functions. We study basic facts about S and discuss the main
conjectures, in particular, the far-reaching Selberg conjectures on primitive
elements. We shall see that the class S ts rather well into the setting of
the Selberg class S (especially with respect to Selbergs conjectures). Hence,
our general universality theorem extends to the Selberg class, unconditionally
for many of the classical L-function and conditionally to all elements of S
subject to some widely believed but rather deep conjectures. However, the
Selberg class is too small with respect to universality; for instance, a Dirichlet
L-function to an imprimitive character does not lie in the Selberg class (by
lack of an appropriate functional equation) but it is known to be universal.
Furthermore, some important L-functions are only conjectured to lie in the
Selberg class, and, in spite of this, for some of them we can derive universality
unconditionally.
In the following chapter, we consider the value-distribution of Dirichlet
series L(s) with functional equation in the complex plane. Following Levin-
sons approach from the 1970s, we shall prove asymptotic formulae for the
c-values of L, i.e., roots of the equation L(s) = c, and give applications in
Nevanlinna theory. In particular, we give an alternative proof of the Riemann
von Mangoldt formula for the elements in the Selberg class.
The main themes of Chap. 8 are almost periodicity and the Riemann
hypothesis. Universality has an interesting feedback to classical problems.
Bohr observed that the Riemann hypothesis for Dirichlet L-functions asso-
ciated with non-principal characters is equivalent to almost periodicity in
the right half of the critical strip. Applying Voronins universality theorem,
Bagchi was able to extend this result to the zeta-function in proving that if
XII Preface

the Riemann zeta-function can approximate itself uniformly in the sense of

Voronins theorem, then Riemanns hypothesis is true, and vice versa. We
sketch an extension of Bagchis theorem to other L-functions.
Chapter 9 deals with the problem of eectivity. The known proofs of uni-
versality are ineective, giving neither bounds for the rst approximating shift
nor for their density (with the exception of particular results due to Garunk-
tis, Good, and Laurinikas). We give explicit upper bounds for the density
of universality; more precisely, we prove upper bounds for the frequency with
which a certain class of target functions (analytic isomorphisms) can be uni-
formly approximated. Moreover, we apply eective results from the theory of
inhomogeneous diophantine approximation to prove several explicit estimates
for the value-distribution in the half-plane of absolute convergence.
In Chap. 10, we discuss further applications of universality, most of them
classical, e.g., an extension of Bohrs and Voronins results concerning the
value-distribution inside the critical strip, and the functional independence
which covers Ostrowskis solution of the Hilbert problem on the hyper-
transcendence of the zeta-function and some of its generalizations. Here a
function is called hyper-transcendental, if it does not satisfy any algebraic
dierential equation. Further, we study the value-distribution of linear com-
binations of (strongly) universal Dirichlet series. A subtle consequence of this
strong concept of universality, and a big contrast to L-functions, can be found
in the distribution of zeros o the critical line. Very likely a (universal) Dirich-
let series satisfying a functional equation of Riemann-type has either many
zeros to the right of the critical line (as a generic Dirichlet series with periodic
coecients) or none (as it is expected for L-functions). This seems to be the
heart of many secrets in the value-distribution theory of Dirichlet series.
Chapter 11 deals with Dirichlet series associated with periodic arithmeti-
cal functions. In general, these functions do not have an Euler product but
they are additively related to Dirichlet L-functions. Consequently, they share
certain properties with L-functions, e.g., a functional equation similar to the
one for Riemanns zeta-function. We prove universality for a large class of
these Dirichlet series; in contrast to L-functions they can approximate uni-
formly analytic functions having zeros (provided their Dirichlet coecients are
not multiplicative). Moreover, we study joint universality for Hurwitz zeta-
functions with rational parameters.
We conclude with joint universality; here joint stands for simultaneous
uniform approximation. In Chap. 12, we prove a theorem which reduces joint
universality for L-functions in S to a denseness property in a related function
space. Of course, we cannot have joint universality for any set of L-functions;
for example, (s) and (s)2 cannot approximate any given pair of admissi-
ble target functions simultaneously. However, we shall prove that in some
instances twists of L S with pairwise non-equivalent characters fulll this
condition (e.g., Dirichlet L functions). In the following chapter we present sev-
eral further applications. For instance, we prove joint universality for Artin
 Preface XIII

L-functions (which lie in the Selberg class if and only if the deep Artin con-
jecture is true). This universality theorem holds unconditionally despite the
fact that Artin L-functions might have innitely many poles in their strip of
universality; this was rst proved by Bauer in 2003 by a tricky argument.
At the end of these notes an appendix on the history of the general phe-
nomenon of universality in analysis is given. It is known that universality is a
quite regularly appearing phenomenon in limit processes, but among all these
universal objects only universal Dirichlet series are explicitly known. At the
end an index and a list of the notations and axioms which were used are given.
Value-distribution theory for L-functions with emphasis on aspects of uni-
versality was treated in the monographs of Karatsuba and Voronin [166] and
Laurinikas [186]. However, after the publication of these books, many new
results and applications were discovered; we refer the reader to the surveys of
Laurinikas [196] and of Matsumoto [242] for some of the progress made in the
meantime. The content of this book forms an extract of the authors habilita-
tion thesis written at Frankfurt University in 2003. We have added Chaps. 12
and 13 on joint universality and its applications as well as several remarks
and comments concerning the progress obtained in the meantime. Unfortu-
nately, we could not include the most current contributions as, for example,
the promising work [245] of Mauclaire which relates universality with almost
periodicity.
I am very grateful to Springer for publishing these notes; especially, I want
to thank Stefanie Zller and Catriona M. Byrne from Springer, the editors of
the series Lecture Notes in Mathematics, and, of course, the anonymous referees
for their excellent work, their valuable remarks and corrections. Furthermore,
I am grateful to my family, my friends and my colleagues for their interest
and support, in particular those from the Mathematics Departments at the
universities of Frankfurt, Madrid, and Wrzburg. Especially, I would like to
thank Ramunas Garunktis and Antanas Laurinikas for introducing me to
questions concerning universality, Ernesto Girondo, Aleksander Ivi, Roma
Kainskaite, Kohji Matsumoto, Georg Johann Rieger, Jrgen Sander, Wolf-
gang Schwarz, and Jrgen Wolfart for the fruitful discussions, helpful remarks
and their encouragement. Last but not least, I would like to thank my wife
Rasa.

Jrn Steuding
Wrzburg, December 2006
1
Introduction

The grandmother of all zeta-functions is the Riemann zeta-function.

David Ruelle

In this introduction we give some hints for the importance of the

Riemann zeta-function for analytic number theory and present rst classic
results on its amazing value-distribution due to Harald Bohr but also the re-
markable universality theorem of Voronin (including a sketch of his proof).
Moreover, we introduce Dirichlet L-functions and other generalizations of the
zeta-function, discuss their relevance in number theory and comment on their
value-distribution. For historical details we refer to Narkiewiczs monograph
[277] and Schwarzs surveys [317, 318].

1.1 The Riemann Zeta-Function and the Distribution

of Prime Numbers
The Riemann zeta-function is a function of a complex variable s = + it, for
> 1 given by


1  1
1
(s) = = 1 s ; (1.1)
n=1
ns p
p

here and in the sequel the letter p always denotes a prime number and the
product is taken over all primes. The Dirichlet series, and the Euler product,
converge absolutely in the half-plane > 1 and uniformly in each compact
subset of this half-plane. The identity between the Dirichlet series and the
Euler product was discovered by Euler [76] in 1737 and can be regarded as
an analytic version of the unique prime factorization of integers. The Euler
product gives a rst glance on the intimate connection between the zeta-
function and the distribution of prime numbers. A rst immediate consequence
is Eulers proof of the innitude of the primes. Assuming that there were only
2 1 Introduction

nitely many primes, the product in (1.1) is nite, and therefore convergent
for s = 1, contradicting the fact that the Dirichlet series dening (s) reduces
to the divergent harmonic series as s 1+. Hence, there exist innitely many
prime numbers. This fact is well known since Euclids elementary proof, but
the analytic access gives deeper knowledge on the distribution of the prime
numbers. It was the young Gauss [94] who conjectured in 1791 for the number
(x) of primes p x the asymptotic formula

(x) li(x), (1.2)

where the logarithmic integral is given by

 1
 x   x
du du
li(x) = lim + = 1.04 . . . ;

0+ 0 1+
log u 2 log u

this integral is a principal value in the sense of Cauchy. Gauss conjecture

states that, in rst approximation, the number of primes x is asymptotically
x
log x . By elementary means, Chebyshev [54, 55] proved around 1850 that for
suciently large x
log x
0.921 . . . (x) 1.055 . . . .
x
Furthermore, he showed that if the limit
log x
lim (x)
x x
exists, the limit is equal to one, which supports relation (1.2).
Riemann was the rst to investigate the Riemann zeta-function as a func-
tion of a complex variable. In his only one but outstanding paper [310] on
number theory from 1859 he outlined how Gauss conjecture could be proved
by using the function (s). However, at Riemanns time the theory of func-
tions was not developed suciently far, but the open questions concerning
the zeta-function pushed the research in this eld quickly forward. We shall
briey discuss Riemanns memoir. First of all, by partial summation

1 
N 1s [u] u
(s) = + +s du; (1.3)
ns s1 N us+1
nN

here and in the sequel [u] denotes the maximal integer less than or equal to u.
This gives an analytic continuation for (s) to the half-plane > 0 except for
a simple pole at s = 1 with residue 1. This process can be continued to the
left half-plane and shows that (s) is analytic throughout the whole complex
plane except for s = 1. Riemann gave the functional equation
s
 
1s
s/2 (s) = (1s)/2 (1 s), (1.4)
2 2
 1.1 The Riemann Zeta-Function and the Distribution of Prime Numbers 3

0.1

0.05

14 12 10 8 6 4 2

0.05

0.1

0.15

Fig. 1.1. (s) in the range s [14.5, 0.5]

where (s) denotes Eulers Gamma-function; it should be noted that

Euler [76] had partial results in this direction (namely, for integral s and
for half-integral s; see [7]). In view of the Euler product (1.1) it is easily seen
that (s) has no zeros in the half-plane > 1. It follows from the functional
equation and from basic properties of the Gamma-function that (s) vanishes
in < 0 exactly at the so-called trivial zeros s = 2n with n N (see Fig. 1.1
for the rst trivial zeros). All other zeros of (s) are said to be non-trivial,
and we denote them by  = + i. Obviously, they have to lie inside the
so-called critical strip 0 1, and it is easily seen that they are non-real.
The functional equation (1.4), in addition with the identity
(s) = (s),
shows some symmetries of (s). In particular, the non-trivial zeros of (s) are
distributed symmetrically with respect to the real axis and to the vertical line
= 12 . It was Riemanns ingenious contribution to number theory to point
out how the distribution of these non-trivial zeros is linked to the distribution
of prime numbers. Riemann conjectured that the number N (T ) of non-trivial
zeros  = + i with 0 < T (counted according multiplicities) satises
the asymptotic formula
T T
N (T ) log .
2 2e
This was proved in 1895 by von Mangoldt [235, 236] who found more precisely
T T
N (T ) = log + O(log T ). (1.5)
2 2e
Riemann worked with the function t  ( 12 + it) and wrote that very likely
all roots t are real, i.e., all non-trivial zeros lie on the so-called critical line
4 1 Introduction

0
10
0
10
20
30
40

Fig. 1.2. The reciprocal of the absolute value of (s) for [4, 4], t [10, 40].
The zeros of (s) appear as poles

= 12 . This is the famous, yet unproved Riemann hypothesis which we rewrite

equivalently as

Riemanns Hypothesis. (s) = 0 for > 12 .

In support of his conjecture, Riemann calculated some zeros; the rst one
with positive imaginary part is  = 12 + i14.134 . . . (see Fig. 1.2 and also
Fig. 8.1). Furthermore, Riemann conjectured that there exist constants A
and B such that
1 s
 s
  
s
s/2
s(s 1) (s) = exp(A + Bs) 1 exp .
2 2 
 

This was shown by Hadamard [113] in 1893 (recall the Hadamard product
theorem from the theory of functions). Finally, Riemann conjectured the so-
called explicit formula which states that

In 1932, Siegel [329] published an account of Riemanns work on the zeta-function

found in Riemanns private papers in the archive of the university library in
Gttingen. It became evident that behind Riemanns speculation there was
extensive analysis and computation.
 1.1 The Riemann Zeta-Function and the Distribution of Prime Numbers 5


(x1/n ) 
(x) + = li(x) li(x ) + li(x1 ) (1.6)
n=2
n
=+i
>0


du
+ log 2
x u(u2 1) log u

for any x 2 not being a prime power (otherwise a term 2k 1

has to be added on
k
the left-hand side, where x = p ); the appearing integral logarithm is dened
by
 (+i) log x
+i exp(z)
li(x )= dz,
(+i) log x z + i

where = +1 if > 0 and = 1 otherwise. The explicit formula was proved

by von Mangoldt [235] in 1895 as a consequence of both product representa-
tions for (s), the Euler product (1.1) on the one hand and the Hadamard
product on the other hand.
Riemanns ideas led to the rst proof of Gauss conjecture (1.2), the cel-
ebrated prime number theorem, by Hadamard [114] and de la Valle-Poussin
[357] (independently) in 1896. We give a very brief sketch (for the details we
refer to Ivi [141]). For technical reasons it is of advantage to work with the
logarithmic derivative of (s) which is for > 1 given by

 (n)
(s) = ,
n=1
ns

where the von Mangoldt -function is dened by

log p if n = pk with k N,
(n) = (1.7)
0 otherwise.

A lot of information concerning the prime counting function (x) can be

recovered from information about

(x) := (n) = log p + O x1/2 log x .

nx px

Partial summation gives

(x)
(x) .
log x
First of all, we shall express (x) in terms of the zeta-function. If c is a positive
constant, then
 c+i s
1 x 1 if x > 1,
ds = (1.8)
2i ci s 0 if 0 < x < 1.
6 1 Introduction

This yields the Perron formula: for x  Z and c > 1,

 c+i 
1 xs
(x) = (s) ds. (1.9)
2i ci s
Moving the path of integration to the left, we nd that the latter expression
is equal to the corresponding sum of residues, that are the residues of the
integrand at the pole of (s) at s = 1, at the zeros of (s), and at the pole of
the integrand at s = 0. The main term turns out to be
     s
xs 1 x
Res s=1 (s) = lim (s 1) + O(1) = x,
s s1 s1 s
whereas each non-trivial zero  gives the contribution
  
xs x
Res s= (s) = .
s 
By the same reasoning, the trivial zeros contribute

 
x2n 1 1
= log 1 2 .
n=1
2n 2 x

Incorporating the residue at s = 0, this leads to the the exact explicit formula

x  
1 1
(x) = x log 1 2 log(2),

 2 x

which is equivalent to Riemanns formula (1.6). Notice that the right-hand side
of this formula is not absolutely convergent. If (s) would have only nitely
many non-trivial zeros, the right-hand side would be a continuous function of
x, contradicting the jumps of (x) for prime powers x. However, going on it
is much more convenient to cut the integral in (1.9) at t = T which leads to
the truncated version

x x

(x) = x +O (log(xT ))2 , (1.10)

 T
||T

valid for all values of x. Next we need information on the distribution of the
non-trivial zeros. The largest known zero-free region for (s) was found by
Vinogradov [359] and Korobov [173] (independently) who proved
c
(s) = 0 in 1 ,
(log |t| + 3)1/3 (log log(|t| + 3))2/3
where c is some positive absolute constant; the rst complete proof due
to Richert appeared in Walsz [366]. In addition with the Riemannvon
Mangoldt formula (1.5) one can estimate the sum over the non-trivial zeros
in (1.10). Balancing out T and x, we obtain the prime number theorem with
the strongest existing remainder term:
 1.1 The Riemann Zeta-Function and the Distribution of Prime Numbers 7

Theorem 1.1. There exists an absolute positive constant C such that for suf-
ciently large x
  
(log x)3/5
(x) = li (x) + O x exp C .
(log log x)1/5
By the explicit formula (1.10) the impact of the Riemann hypothesis on the
prime number distribution becomes visible. Von Koch [172] showed that for
xed [ 12 , 1)
(x) li (x)  x+
(s) = 0 for > ; (1.11)
here and in the sequel  stands for an arbitrary small positive constant, not
necessarily the same at each appearance. With regard to known zeros of (s)
on the critical line it turns out that an error term with < 12 is impossible.
Thus, the Riemann hypothesis states that the prime numbers are as uniformly
distributed as possible!
Many computations were done to nd a counterexample to the Riemann
hypothesis. Van de Lune, te Riele and Winter [232] localized the rst
1 500 000 001 zeros, all lying without exception on the critical line; moreover
they all are simple! By observations like this it is conjectured, that all or
at least almost all zeros of the zeta-function are simple. This so-called essen-
tial simplicity hypothesis has arithmetical consequences. Cramr [63] showed,
assuming the Riemann hypothesis,
 X 2  
1 (x) x
 m() 2
dx  
log X 1 x    , (1.12)


where the sum is taken over distinct zeros and m() denotes their multiplicity.
The right-hand side is minimal if all the zeros are simple. Going further,
Goldston, Gonek and Montgomery [103] observed interesting relations
between the essential simplicity hypothesis, mean-values of the logarith-
mic derivative of (s), the error term in the prime number theorem, and
Montgomerys pair correlation conjecture.
A classical density theorem due to Bohr and Landau states that most
of the zeros lie arbitrarily close to the critical line. Denote by N (, T ) the
number of zeros  = + i of (s) for which > and 0 < T (counting
multiplicities). Bohr and Landau [34] (see also [224]) proved that
N (, T )  T = O(N (T )) (1.13)
1
for any xed > 2. Hence, almost all zeros of the zeta-function are clustered
around the critical line. A renement of their method led to bounds N (, T ) 
T with < 1 (see [35, 49]); for instance:
1
Theorem 1.2. For any xed with 2 < < 1,

N (, T )  T 4(1)+
.
For the various improvements of this density estimate we refer to Ivi [141].
8 1 Introduction

Hardy [116] showed that innitely many zeros lie on the critical line, and
Selberg [321] was the rst to prove that a positive proportion of all zeros lies
exactly on = 12 . Let N0 (T ) denote the number of zeros  of (s) on the
critical line with imaginary part 0 < T . The idea to use molliers to
dampen the oscillations of |( 12 + it)| led Selberg to
N0 (T + H) N0 (T )
lim inf > 0,
T N (T + H) N (T )
as long as H T 1/2+
. Karatsuba [164] improved this result to H T 27/82+

by some technical renements. The proportion is very small, about 106 as
2
Min calculated; a later renement by Zhuravlev gives after all 21 if H = T (cf.
[165, p. 36]). However, the localized zeros are not necessarily simple. By an
ingenious new method, working with molliers of nite length, Levinson [216]
localized more than one third of the non-trivial zeros of the zeta-function on
the critical line, and as Heath-Brown [122] and Selberg (unpublished) discov-
ered, they are all simple. By optimizing the technique, Levinson himself and
others improved the proportion 13 slightly, but more recognizable is Conreys
idea in introducing Kloosterman sums. So Conrey [57] was able to choose a
longer mollier to show that more than two-fths of the zeros are simple and
on the critical line. Bauer [17, 18] improved this proportion slightly. The use
of longer molliers leads to larger proportions. Farmer [77] observed that if it
is possible to take molliers of innite length, then almost all zeros lie on the
critical line and are simple. In [339], Steuding found a new approach (com-
bining ideas and methods of Atkinson, Jutila and Motohashi) to treat short
intervals [T, T + H], i.e., H = O(T ). It was proved that for H T 0.552 a
positive proportion of the zeros of the zeta-function with imaginary parts in
[T, T + H] lie on the critical line and are simple.
Recently, Garaev [81] showed that


|  ()|1  (log T )1/2 ;
0<<T

(
)=0


the divergence of the series  |  ()|1 was known before only subject to
the truth of the Riemann hypothesis (see [353, p. 374]). Garaevs result was
slightly improved by himself [82] and leeviiene and Steuding [333] (inde-
pendently) by a factor (log T )1/4 . Furthermore, using the results of [339] for
short intervals, in [319] it was proved that

  1 1

  + i   H(log T )1/4
 2 
T <<T +H

( 1 +i)=0
2

and

  
  1 
 + i   H(log T )9/4 ,
 2 
T <<T +H

( 1 +i)=0
2
 1.2 Bohrs Probabilistic Approach 9

both valid for T 0.552 H T . Such estimates measure how close the zeros
of the zeta-function are to being simple.

1.2 Bohrs Probabilistic Approach

In the second decade of the twentieth century, Harald Bohr rened former
studies on the value-distribution of the Riemann zeta-function by applying
diophantine, geometric, and probabilistic methods. Especially, for us, his prob-
abilistic approach is of interest. It seems to be rather dicult to locate concrete
values taken by the zeta-function, but it is much easier to study how often
the values lie in a given set. As Bohr found out these frequencies follow strict
mathematical laws (analogously to quantum mechanics, where it is impos-
sible to locate simultaneously space and time of a particle by Heisenbergs
uncertainty relation, but large numbers of particles obey statistical laws).
In the half-plane of absolute convergence > 1, we have

0 < |(s)| ().

Thus the values of (s) in half-planes 0 > 1 are lying in the disk of radius
(0 ) centered in the origin. It can be shown that (s) takes quite many of the
complex values inside this disk when t varies in R (very similar to the classic
theorem of Bloch in the theory of functions). On the other side () tends to
innity as 1+, and indeed Bohr [28] succeeded in proving

Theorem 1.3. In any strip 1 < < 1 + , (s) takes any non-zero value
innitely often.

Somehow it is more natural to study the logarithm of the zeta-function. To

dene log (s) for any s C, we choose the principal branch of the logarithm
on the intersection of the real axis with the half-plane of absolute convergence.
For other points s we dene log ( +it) to be the value obtained from log (2)
by continuous variation along the line segments [2, 2 + it] and [2 + it, + it],
provided that the path does not cross a zero or pole of (s); if it does, then we
take log ( + it) = lim
0+ log ( + i(t + )). Of course, for other L-functions
we may proceed analogously. For > 1, we have

 


1 1
log (s) = log 1 s = sk
. (1.14)
p
p p
kp
k=1

For xed prime p and xed > 1, the set of values taken by the inner sum
in the series representation on the right-hand side is a convex curve while
t runs through R. Adding up all these curves gives, using some facts from
the theory of diophantine approximation, information on the values taken
by log (s) itself. Actually, it follows that log (s) takes any complex value in
1 < < 1 +  which, by exponentiation, leads to the assertion of Theorem 1.3.
10 1 Introduction

2 1 1 2 3

Fig. 1.3. ( 35 + it) for t [0, 60]. The curve visits any neighbourhood of any given
point in the complex plane as t runs through the set of real numbers

The situation inside the critical strip is much more complicated. Here,
Bohr studied nite Euler products
  1
1
M (s) := 1 s . (1.15)
p
pM

As M tends to innity, these products do not converge any longer in the

critical strip but they approximate (s) in the mean (we will meet this
ingenious idea several times later). The value-distribution of nite Euler prod-
ucts is treatable by the theory of diophantine approximation, and by their
approximation property this leads to information on the values taken by the
zeta-function inside the critical strip. In a series of papers Bohr and his col-
laborators discovered that the asymptotic behaviour of (s) is ruled by prob-
ability laws on every vertical line to the right of = 12 . In particular, Bohr
and Courant [30] proved that for any xed ( 12 , 1] the set of values ( + it)
with t R lies dense in the complex plane (see Fig. 1.3). Later, Bohr rened
these results signicantly by applying probabilistic methods. Let R be an
arbitrary xed rectangle in the complex plane whose sides are parallel to the
real and the imaginary axes, and let G be the half-plane > 12 where all
points are removed which have the same imaginary part as, and smaller real
part than, one of the possible zeros of (s) in this region. Then a remarkable
limit theorem due to Bohr and Jessen [31, 32] states.
Theorem 1.4. For any > 12 , the limit
1
lim meas { [0, T ] : + i G, log ( + i ) R}
T T
exists.
 1.2 Bohrs Probabilistic Approach 11

Here and in the sequel meas A stands for the Lebesgue measure of a mea-
surable set A. The limit value of Theorem 1.4 may be regarded as the
probability how many values of log ( + it) belong to the rectangle R. Hattori
and Matsumoto [121] identied this limit distribution, which is too compli-
cated to be written down here. Next, for any complex number c, denote by
N c (1 , 2 , T ) the number of c-values of (s), i.e., the roots of the equation
(s) = c, inside the region 1 < < 2 , 0 < t T (counting multiplicities).
From the limit theorem mentioned above Bohr and Jessen [32] deduced

Theorem 1.5. Let c be a complex number = 0. Then, for any 1 and 2

satisfying 12 < 1 < 2 < 1, the limit

1 c
lim N (1 , 2 , T )
T T
exists and is positive.

In connection with the density Theorem 1.2 it follows that zeros are indeed
exceptional values of the zeta-function.
In 1935, Jessen and Wintner [149] proved limit theorems similar to the one
above by using more advanced methods from probability theory (innite con-
volutions of probability measures). We do not mention further developments
of Bohrs ideas by his successors Borchsenius, Jessen and Wintner but refer
for more details on Bohrs contribution and results of his collaborators to the
monograph of Laurinikas [186] and the survey of Matsumoto [242]. Bohrs
line of investigations appears to have been almost totally abandoned for some
decades. Only in 1972, Voronin [361] obtained some signicant generalizations
of Bohrs denseness result.
1
Theorem 1.6. For any xed numbers s1 , . . . , sn with 2 < Re sk < 1 for
1 k n and sk = s for k = , the set

{((s1 + it), . . . , (sn + it)) : t R}

is dense in Cn . Moreover, for any xed number s in 1

2 < < 1,

{((s + i ),  (s + i ), . . . , (n1) (s + i )) : R}

is dense in Cn .

This result (in addition with Voronins universality Theorem 1.7) started
a second period of intensive research in this eld. Comparable limit theo-
rems and related results were obtained, for example, by Laurinikas [180] for
Dirichlet L-functions, by Matsumoto [239] for more general L-functions, and
recently by leeviiene [331] for powers of the zeta-function, to give only
some examples.
12 1 Introduction

We conclude this section with a short view on the value-distribution of

the Riemann zeta-function on the critical line which appears to be rather
dierent. It is conjectured but yet unproved that also the set of values of (s)
taken on the critical line = 12 is dense in C. Selberg (unpublished) proved
that the values taken by an appropriate normalization of the Riemann zeta-
function on the critical line are normally distributed: let R be an arbitrary
xed rectangle in the complex plane whose sides are parallel to the real and
the imaginary axes, then

 
1 log 12 + it
lim meas t (0, T ] :  R
T T 1
log log T
2
  
1 x2 + y 2
= exp dx dy. (1.16)
2 R 2

The rst published proof is due to Joyner [150]. Laurinikas [184] obtained
results near the critical line, comparable to (1.16); more precisely, the weak
convergence of a suitably normalized measure to the normal distribution in
the region
1 1 1
+
2 2 log T
as T tends to innity. Selbergs limit theorem with remainder term includes
as a particular case the result
     
1  1  1

meas t (0, T ] :  
+ it  exp y log log T
T 2 2
 y  2  
1 x (log log log T )2
= exp dx + O .
2 2 log log T

As Hejhal [130] recently conrmed, this holds uniformly in y. This uniformity

is rather useful. Ivi [143] pointed out how to deduce results on small values
taken by the zeta-function and results on gaps between the ordinates of con-
secutive zeros on the critical line. Laurinikas and Steuding [208] investigated
applications to the distribution of large values (with respect to -results on
one hand and the Lindelf hypothesis on the other hand).

1.3 Voronins Universality Theorem

In 1975, Voronin [363, 364] proved a remarkable universality theorem for (s)
which states, roughly speaking, that any non-vanishing analytic function can
be approximated uniformly by certain purely imaginary shifts of the zeta-
function in the critical strip.
 1.3 Voronins Universality Theorem 13

Theorem 1.7. Let 0 < r < 14 and suppose that g(s) is a non-vanishing con-
tinuous function on the disk |s| r which is analytic in the interior. Then,
for any  > 0,
     
1  3 
lim inf meas [0, T ] : max  s + + i g(s) <  > 0.
T T |s|r 4

Thus the set of approximating shifts has positive lower density!

We say that (s) is universal since appropriate shifts approximate uni-
formly any element of a huge class of functions. What might have been
Voronins intention for his studies which had led him to the discovery of
this astonishing universality property? First of all, Voronins universality the-
orem can be seen as an innite dimensional analogue of the second part of
Theorem 1.6: the truncated Taylor series of the target function g(s) can be
approximated by the truncated Taylor series of a certain shift of zeta. This
becomes more clear in Sect. 10.1 when we will deduce the second assertion of
Theorem 1.6 from Theorem 1.7 as a particular case in a more general setting
(for more details we refer to the nice survey articles of Laurinikas [198] and
Matsumoto [244]). Moreover, in Sect. 10.6 we shall indicate how to deduce the
rst part from a rened version of Voronins universality theorem. Another
reason for Voronins investigations could have been Bohrs concept of almost
periodicity with which we will get in touch in Sect. 8.2.
We give a brief sketch of Voronins argument following the book of
Karatsuba and Voronin [166]. The Euler product (1.1) is the key to prove
the universality theorem in spite of the fact that it does not converge in the
region of universality. However, as Bohr [29] observed, an appropriate trun-
cated Euler product (1.15) approximates (s) in a certain mean-value sense
inside the critical strip; this is related to the use of modied truncated Euler
products in Voronins proof (see (1.19) and (1.20)).
It is more convenient to work with series than with products. Therefore,
we consider the logarithms of the functions in question. Since g(s) has no zeros
in |s| r, its logarithm exists and we may dene an analytic function f (s) by
g(s) = exp f (s). First we approximate f (s) by the logarithm of a truncated
Euler product. Let denote the set of all sequences of real numbers indexed
by the prime numbers in ascending order. Further, dene for every nite subset
M of the set of all primes, every = (2 , 3 , . . .) , and all complex s,

  exp(2ip )
1
M (s, ) = 1 .
ps
pM

Obviously, M (s, ) is a non-vanishing analytic function of s in the half-plane

> 0. Consequently, its logarithm exists and is equal to

 
exp(2ip )
log M (s, ) = log 1 ;
ps
pM
14 1 Introduction

in order to have a denite value we may choose the principal branch of the
logarithm. It might be unpleasant that f is only assumed to be continuous on
the boundary. However, since f (s) is uniformly continuous in the disk |s| r,
there exists some > 1 such that 2 r < 14 and
 s

  
max f f (s)< .
|s|r 2 2

The function f (s/2 ) is analytic and bounded on the disk |s| r =: R, and
thus belongs to the Hardy space HR 2
, i.e., the Hilbert space consisting of those
functions F (s) which are analytic for |s| < R with nite norm

F := lim |F (s)| d dt.
rR |s|r

Introducing the inner product


F, G = Re F (s)G(s) d dt,
|s|R

HR2
becomes a real Hilbert space. Denote by pk the kth prime number. We
consider the series


uk (s, ),
k=1

where  1
exp(2ipk )
uk (s, ) := log 1 s+(3/4)
.
pk
Here comes the rst main idea. Riemann proved that any conditionally
convergent series of real numbers can be rearranged such that its sum
converges to an arbitrary preassigned real number. Pechersky [288] gener-
alized Riemanns theorem to Hilbert spaces (Theorem 5.4). It follows, with
choice = 0 = ( 14 , 24 , 34 , . . .), that there exists a rearrangement of
the special
the series uk (s, 0 ) for which


s

ukj (s, 0 ) = f
j=1
2

(the rather dicult and lengthy verication of the conditions of Pecherskys

theorem uses classic results of Paley and Wiener and Plancherel from Fourier
analysis, a theorem on polynomial approximation due to Markov, and, most
importantly, the prime number Theorem 1.1). It might be interesting to notice
that in the older paper [362] Voronin had already developed the techniques
for proving universality only a rearrangement theorem for function spaces
was not at hand. Recently, Garunktis [87] found an argument to omit the
ineective rearrangement.
 1.3 Voronins Universality Theorem 15

The tail of the rearranged series can be made as small as we please, say
of modulus less than 2
. Thus, it turns out that for any  > 0 and any y > 0
there exists a nite set M of prime numbers, containing at least all primes
p y, such that
   
 3 
max log M s + , 0 f (s) < . (1.17)
|s|r 4

The next and main step in Voronins proof is to switch from log M (s)
to the logarithm of the zeta-function. Of course, log (s) has singularities at
the zeros of (s), but since the set of these possibly singularities has measure
zero by density Theorem 1.2, they are negligible. Note that for many higher
L-functions (i.e., functions which satisfy a Riemann-type functional equation
with many Gamma-factors), which we shall consider later, no density results
of this strength are known.
Now we choose > 1 and 1 (0, 1) such that r < 14 and
 s

 
max f f (s) < 1 . (1.18)
|s|r

Putting Q = {p : p z} and E = {s = + it : r < 2, |t| 1}, one can

show, using the approximate functional equation for (s) (i.e., a representation
as a Dirichlet polynomial related to (1.3)), that for any 2 > 0
       2
2T  1 3 3 
 s + + i, 0 s + + i 1 d dt d  42 T, (1.19)
 Q 4 4 
T E

provided that z and T are suciently large, depending on 2 ; here 0 :=

(0, 0, . . .); actually this is Hilfssatz 2 from Bohr [29]. Now dene
 
1
AT = [T, 2T ] : |Q (s + i, 0)(s + i ) 1|2 d dt < 22 .
E+ 34

Then it follows from (1.19) that, for suciently large z and T ,

meas AT > (1 2 )T, (1.20)

which is surprisingly large. It follows from Cauchys formula that, for su-
ciently small 2 ,
    
 3 3 
max log s + + i log Q s + + i, 0   2 , (1.21)
|s|r 4 4

provided AT , where the implicit constant depends only on . By (1.17)

there exists a sequence of nite sets of prime numbers M1 M2 . . . such
that k=1 Mk contains all primes and
16 1 Introduction
   s

 3

lim max log Mk s + , 0 f  = 0. (1.22)
k |s|r  4 
(0) (0)  
Let 0 = (2 , 3 , . . .). By the continuity of log M s + 34 , 0 , for any 1 >
0, there exists a positive such that, whenever the inequalities
p(0) p < for p Mk (1.23)
hold, where z denotes the minimal distance of z to an integer, then
    
 3 3 
max log Mk s + , 0 log Mk s + ,  < 1 .
 (1.24)
|s|r 4 4
Let   log p  
 
BT = [T, 2T ] :  p(0)  < .
2
Now we consider
      
1  3 3 2
 log Q s + + i, 0 log Mk s + + i, 0  d dt d,
T BT |s|r 4 4
respectively,
      
1  3 3 2
 log Q s + + i, 0 log Mk s + + i, 0  d d dt.
|s|r T BT 4 4
Putting  
log 2 log 3
( ) = , ,... , (1.25)
2 2
we may rewrite the inner integral as
     2
 
log Q s + 3 , ( ) log M s + 3 , ( )  d.
 4 k
4 
BT

Next we need Weyls renement of Kroneckers approximation theorem.

Let ( ) be any continuous function with domain of denition [0, ) and
range RN . Then the curve ( ) is said to be uniformly distributed mod 1 in
RN if, for every parallelepiped = [1 , 1 ] . . . [N , N ] with 0 j <
j 1 for 1 j N ,

1 N
lim meas { (0, T ) : ( ) mod 1} = (j j ).
T T
j=1

In some sense, a curve is uniformly distributed mod 1 if the proportion of

the fractional values which lie in an arbitrary parallelepiped of the unit
cube coincides with the volume of . In questions about uniform distribution
mod 1 one is interested in the fractional part only. For a curve ( ) in RN ,
we dene
{( )} = (1 ( ) [1 ( )], . . . , N ( ) [N ( )]),
where [x] denotes the integral part of x R.
 1.3 Voronins Universality Theorem 17

Lemma 1.8. (i) Let a1 , . . . , aN be real numbers, linearly independent over Q,

and let be a subregion of the N -dimensional unit cube with Jordan content
. Then
1
lim meas { (0, T ) : ( a1 , . . . , aN ) mod 1} = .
T T
(ii) Suppose that the curve ( ) is uniformly distributed mod 1 in RN . Let D
be a closed and Jordan measurable subregion of the unit cube in RN and let
be a family of complex-valued continuous functions dened on D. If is
uniformly bounded and equicontinuous, then
 
1 T
lim f ({( )})1D ( ) d = f (x1 , . . . , xN ) dx1 . . . dxN
T T 0 D

uniformly with respect to f , where 1D ( ) is equal to 1 if ( ) D mod 1,

and 0 otherwise.
Note that the notion of Jordan content is more restrictive than the notion of
Lebesgue measure. But, if the Jordan content exists, then it is also dened
in the sense of Lebesgue and equal to it. A proof of Weyls theorem can be
found in his paper [370] or in Karatsuba and Voronin [166].
The unique prime factorization of integers implies the linear independence
of the logarithms of the prime numbers over the eld of rational numbers.
Thus, in some sense, the logarithms of prime numbers behave like random
variables. This is the main reason why probabilistic methods can be applied
to the zeta-function or, more generally, to Euler products!
By Lemma 1.8(i), the curve ( ), dened by (1.25), is uniformly distrib-
uted mod 1. Application of Lemma 1.8(ii), to the curve ( ) yields
     2
1  
lim log Q s + 3 , ( ) log M s + 3 , ( )  d
T T B  4 k
4 
T
     
 2
= log Q s + 3 , log M s + 3 ,  d,
 4 k
4 
D

uniformly in s for |s| r, where D is the subregion of the unit cube in

RN given by the inequalities (1.23) with N = Mk , and d is the Lebesgue
measure. By the denition of M (s, ) it follows that for Mk Q

Q (s, ) = Mk (s, )Q\Mk (s, ),

and thus
     2
 
log Q s + 3 , log M s + 3 ,  d
 4 k
4 
D
   2
 
 meas D log Q\M s + 3 ,  d.
 k
4 
[0,1]N
18 1 Introduction
2r(1/2)
The latter integral is bounded above by yk provided that Mk contains
all primes yk . It follows that:
      
1 log Q s + 3 + i, 0 log M s + 3 + i, 0 2 d dt d
T BT |s|r 4 k 4
2r(1/2)
 yk meas D.
Applying once more Lemma 1.8(i) yields
1
lim meas BT = meas D,
T T
which implies, for suciently large yk ,
    
 3
meas BT : 
log Q s + 4 + i, 0
|s|r
 2 
3 
log Mk s + + i, 0  d dt < yk
r(1/4)
4
meas D
> T.
2
This gives via Lemma 1.8(ii)
   
1  3

meas BT : max log Q s + + i, 0
T |s|r 4
  
3 
log Mk s + + i, 0  < yk
(r1/4)/5
4
meas D
> . (1.26)
2
If we now take 0 < 2 < 1
2 meas D, then (1.20) implies
1 1
meas AT BT meas D 2 > 0.
T 2
Thus, in view of (1.18) and (1.22) we may approximate f (s) by
log Mk s + 34 , 0 (independent
 on ), with (1.24) and (1.26) the lat-
ter function
 by log Q s + 4 + i, 0 , and nally with regard to (1.21) by
3

log s + 34 + i on a set of with positive measure. Replacing T by 12 T , we

thus nd, for any  > 0,
     
1  3 
 
lim inf meas [0, T ] : max log s + + i f (s) <  > 0.
T T |s|r 4
Now it is obvious how to deduce Voronins theorem by taking the exponential.
Voronin called his universality theorem Teorema o kru
oqkah, the the-
orem about little disks. Reich [305] and Bagchi [9] improved Voronins result
signicantly in replacing the disk by an arbitrary compact set in the right half
of the critical strip with connected complement, and by giving a lucid proof in
the language of probability theory. The strongest version of Voronins theorem
has the form:
 1.4 Dirichlet L-Functions and Joint Universality 19

Theorem 1.9. Suppose that K is a compact subset of the strip 12 < < 1 with
connected complement, and let g(s) be a non-vanishing continuous function
on K which is analytic in the interior of K. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max |(s + i ) g(s)| <  > 0.
T T sK

The topological restriction on K is necessary. This follows from basic facts

in approximation theory (see the remark to Theorem 5.15). Also the restriction
on g(s) to be non-vanishing cannot be removed as we shall show in Sect. 8.1.
The domain in which the uniform approximation of admissible target functions
takes place is called the strip of universality. In the case of the zeta-function
this strip of universality is the open right half of the critical strip; later we
will meet examples where the strip of universality is more restricted.
It should be noticed that Voronins theorem implies that (s)1 is universal
(independent of the truth of the Riemann hypothesis). To see this we observe
that given a non-vanishing analytic function g(s) on some admissible set K
also its reciprocal is analytic and non-vanishing. Hence we can approximate
the function g(s)1 uniformly by shifts (s + i ). Since the set of such has
positive lower density and, by the density Theorem 1.2, the set of non-trivial
zeros of (s) is negligible, it follows that we can nd for which
(s + i )1 g(s) for s K.
Another consequence of the universality for the zeta-function is that its deriv-
atives and also the logarithmic derivative are universal; this was rst proved
by Bagchi [10] and Laurinikas [182] by a slight modication of the proof of
universality for (s).
We may interpret the absolute value of an analytic function as an analytic
landscape over the complex plane. Then the universality theorem states that
any nite analytic landscape can be found (up to an arbitrarily small error)
in the analytic landscape of the Riemann zeta-function (see Fig. 1.4 on the
next page). On the contrary, Steuding [338] showed that one cannot go to
innity along a rectiable path inside the critical strip avoiding large values,
roughly speaking, one cannot stay in valleys in the analytic landscape all the
time (the proof relies on a PhragmnLindelf type theorem for regions with
rectiable boundary and Ahlfors distortion theorem). Taking into account
explicit estimates for extreme values, so-called -results, as for example (9.2)
due to Montgomery [260], this can be formulated precisely.

1.4 Dirichlet L-Functions and Joint Universality

A special role in number theory is played by multiplicative arithmetical func-
tions and their associated generating functions. Multiplicative functions res-
pect the multiplicative structure of N: an arithmetic function f is called
multiplicative if f (1) = 0 and
20 1 Introduction

0.5
0.6
0.7

0.8

0.9

1
4

0
116
118
120
122

Fig. 1.4. Some dunes at the baltic sea shore or the analytic landscape of (s) for
[ 12 , 1], t [115, 122]

f (m n) = f (m) f (n)

for all coprime integers m, n; if the latter identity holds for all integers, f is
said to be completely multiplicative.
Let q be a positive integer. A Dirichlet character mod q is a non-
vanishing group homomorphism from the group (Z/qZ) of prime residue
classes modulo q to C. The character, which is identically one, is called prin-
cipal, and is denoted by 0 . By setting (a) = 0 on the non-prime residue
classes such a character extends via (n) = (a) for n a mod q to a com-
pletely multiplicative arithmetic function. For > 1, the Dirichlet L-function
L(s, ) attached to a character mod q is given by

 1
(n)  (p)
L(s, ) = = 1 s .
n=1
ns p
p

The zeta-function (s) may be regarded as the Dirichlet L-function to the

principal character 0 mod 1. It is possible that for values of n coprime
with q the character (n) may have a period less than q. If so, we say
that is imprimitive, and otherwise primitive; the principal character is not
regarded as a primitive character. Every non-principal imprimitive character
is induced by a primitive character. Two characters are non-equivalent if they
are not induced by the same character. The characters to a common modulus
 1.4 Dirichlet L-Functions and Joint Universality 21

are pairwise non-equivalent. If mod q is induced by a primitive character

mod q , then
 (p)

L(s, ) = L(s, ) 1 . (1.27)
ps
p|q

Being twists of the Riemann zeta-function with multiplicative characters,

Dirichlet L-functions share many properties with the zeta-function. For ins-
tance, there is an analytic continuation to the complex plane, only with the
dierence that L(s, ) is regular at s = 1 if and only if is non-principal. Fur-
thermore, L-functions to primitive characters satisfy a functional equation of
the Riemann-type; namely,

(s+)/2
()

(1+s)/2

q s+ q 1+s
L(s, ) = L(1 s, ),
2 i q 2

where := 12 (1 (1)) and

 
2ia
() := (a) exp
q
a mod q

is the Gauss sum attached to . One nds similar zero-free regions (with
the exception of possible Siegel zeros on the real line), density theorems, and
also for Dirichlet L-functions it is expected that the analogue of the Riemann
hypothesis holds; the so-called Generalized Riemann hypothesis states that
neither (s) nor any L(s, ) has a zero in the half-plane Re s > 12 .
Dirichlet L-functions were constructed by Dirichlet [69] to tackle the prob-
lem of the distribution of primes in arithmetic progressions. The main ingre-
dient in his approach are the orthogonality relations for characters linking
prime residue classes with character sums:


1 1 if = 0 ,
(a) = (1.28)
(q) 0 otherwise,
a mod q

and its dual variant


1 1 if a 1 mod q,
(a) =
(q) 0 otherwise,
mod q

valid for a coprime with q, where (q) is Eulers -function which counts
the number of prime residue classes mod q. By the latter relation a suitable
linear combination of characters can be used as indicator function of prime
residue classes modulo q. Using similar techniques as for (s), one can prove a
prime number theorem for arithmetic progressions. Let (x; a mod q) denote
the number of primes p x in the residue class a mod q, then, for a coprime
with q,
1
(x; a mod q) (x). (1.29)
(q)
22 1 Introduction

This shows that the primes are uniformly distributed in the prime residue
classes. Of course, one can prove also an asymptotic formula with error term
(the theorem of PageSiegelWalsz gives even an asymptotic formula which
is uniform in a small region of values q). Under assumption of the Generalized
Riemann hypothesis one has
1

(x; a mod q) = li (x) + O x1/2 log(qx)

(q)

for x 2, q 1, and a coprime with q, the implicit constant being absolute.

There are plenty of results which hold if Riemanns hypothesis is true. Often
one can replace this assumption by the celebrated theorem of Bombieri
Vinogradov due to Bombieri [37] and Vinogradov [360] (independently, with
a slightly weaker range for Q) which states that, for any A 1,
 

 1  x

max max (y; a mod q) li(y)  + Qx1/2 (log Qx)6 .
a mod q yx  (q) (log x)A
qQ (a,q)=1

This shows that the error term in the prime number theorem (1.29) is, on
average over q x1/2 (log x)A7 , of comparable size as predicted by the
Riemann hypothesis.
We return to the theme of universality. Voronin [364] proved that a col-
lection of Dirichlet L-functions to non-equivalent characters can uniformly
approximate simultaneously non-vanishing analytic functions; in slightly dif-
ferent form this was also established by Gonek [104] and Bagchi [9] (indepen-
dently; all these sources are unpublished doctoral theses). Again we state the
strongest version of this so-called joint universality:

Theorem 1.10. Let 1 mod q1 , . . . ,  mod q be pairwise non-equivalent

Dirichlet characters, K1 , . . . , K be compact subsets of the strip 12 < < 1
with connected complements. Further, for each 1 k , let gk (s) be a con-
tinuous non-vanishing function on Kk which is analytic in the interior of Kk .
Then, for any  > 0,
 
1
lim inf meas [0, T ] : max max |L(s + i, k ) gk (s)| < > 0.
T T 1k sKk

The proof of this joint universality theorem (in a slightly weaker form) can
be found in the monograph of Karatsuba and Voronin [166]. The proof uses the
orthogonality relation (1.28). This independence is essential for joint univer-
sality. Consider a character mod q induced by another character mod q .
It follows immediately from (1.27) that both L(s, ) and L(s, ) cannot
approximate uniformly a given function jointly.
Another type of universality was discovered by Bagchi. In [9], he proved
universality for Dirichlet L-functions with respect to the characters; more
precisely, if K is a compact subset of 12 < < 1 with connected complement
 1.4 Dirichlet L-Functions and Joint Universality 23

and g(s) is a non-vanishing continuous function on K, which is analytic in the

interior, then, for any suciently large prime number p and any > 0, there
exist a Dirichlet character mod p such that
   
 3 
max L s + , g(s) < ;
 (1.30)
sK 4
moreover, the latter inequality holds for more than cp characters mod p,
where c is a positive constant (recall that there are (p) = p 1 characters
mod p). Eminyan [75] showed that if K = {s : |s| r} with 0 < r < 14 , and
g(s) is a non-vanishing continuous function for s K, which is analytic in the
interior, then, for each prime number p and any > 0, there exist a positive
integer n and a Dirichlet character mod pn such that (1.30) holds.
We conclude with another interesting type of universality. Any integer
d = 0 with d 0, 1 mod 4 is called a discriminant; if d = 1 or d is the
discriminant of a quadratic number eld, d is said to be fundamental. To any
fundamental discriminant d we attach a real primitive character d mod |d|
dened by the Kronecker symbol ( d ). Recently, Mishou and Nagoshi [251]
investigated the functional distribution of the associated Dirichlet L-functions
L(s, d ) on 12 < < 1 as d varies and proved a universality theorem for
L(s, d ) in the d-aspect: let be a simply connected region in the open strip
D := {s : 12 < < 1} which is symmetric with respect to the real axis and
let g(s) be a non-vanishing analytic function on which takes positive real
values on D R; then for any compact subset K of and any positive ,
{1 d X : maxsK |L(s, d ) g(s)| < }
lim inf > 0. (1.31)
X {1 d X}

The condition on the target functions g(s) is natural with respect to the val-
ues taken by L(s, d ) on R. Mishou and Nagoshi [252] also obtained a
similar result for prime discriminants. Besides, they also studied the distribu-
tion of values L(1, d ). Denote by D+ the set of positive square-free integers
d > 1 with d 1 mod 8, and by D the set of negative square-free inte-
gers d 1 mod 8. In [251], they obtained the remarkable result that the sets
{L(1, d ) : d D } with {+, } are dense in R+ . They further deduced
from Dirichlets class number formula that also
   
h(d) log (d) h(d)
:dD +
and :dD
d d
are dense in R+, where h(d) is the class number of the associated quadratic

number eld Q( d) with d D , and (d) is the fundamental unit of Q( d)
with d D+ .
Nagoshi [275] extended this concept of universality to families of auto-
morphic L-functions, more precisely, to L-functions attached to newforms for
congruence subgroups of SL2 (Z) (which we will introduce in the following
section).
24 1 Introduction

1.5 L-Functions Associated with Newforms

First we recall some facts from the theory of modular forms; see Iwaniecs
book [144] on modular forms and associated L-functions for details.
Denote by H the upper half-plane {z := x + iy C : y > 0}, and let k and
N be positive integers, k being even. The subgroup
  
a b
0 (N ) := SL2 (Z) : c 0 mod N
c d
of the full modular group SL2 (Z) is called Hecke subgroup of level N or con-
gruence subgroup mod N . A holomorphic function f (z) on H is said to be a
cusp form of weight k and level N , if
 
az + b
f = (cz + d)k f (z)
cz + d
for all z H and all matrices
 
a b
0 (N ),
c d
and if f vanishes at all cusps. The vanishing of f at the cusps is equivalent
with
z := x + iy  y k |f (z)|2
is bounded on H. Then f has for z H a Fourier expansion



f (z) = c(n) exp(2inz). (1.32)
n=1

The cusp forms on 0 (N ) of weight k form a nite dimensional complex vector

space, denoted by Sk (0 (N )), with the Petersson inner product, dened by

dx dy
f, g = f (z)g(z)y k 2
H/0 (N ) y

for f, g Sk (0 (N ). Suppose that M |N . If f Sk (0 (M )) and dM |N , then

z  f (dz) is a cusp form on 0 (N ) of weight k too. The forms which may
be obtained in this way from divisors M of the level N with M = N span a
subspace Skold (0 (N )), called the space of oldforms. Its orthogonal complement
with respect to the Petersson inner product is denoted Sknew (0 (N )). For n N
we dene the Hecke operator T (n) by
 
1
k
az + b
T (n)f = a f
n d
ad=n 0b<d

for f Sk (0 (N )). The operators T (n) are multiplicative, i.e., T (mn) =

T (m)T (n) for coprime m, n, and they encode plenty of arithmetic information
 1.5 L-Functions Associated with Newforms 25

about modular forms. The theory of Hecke operators implies the existence of
an orthogonal basis of Sknew (0 (N )) made of eigenfunctions of the operators
T (n) for n coprime with N . By the multiplicity-one principle of Atkin and
Lehner [6], the elements f of this basis are in fact eigenfunctions of all T (n),
i.e., there exist complex numbers f (n) for which T (n)f = f (n)f and c(n) =
f (n)c(1) for all n N. Furthermore, it follows that the rst Fourier coecient
c(1) of such an f is non-zero. Such a simultaneous eigenfunction is said to be
an eigenform. A newform is dened to be an eigenform that does not come
from a space of lower level and is normalized to have c(1) = 1. The newforms
form a nite set which is an orthogonal basis of the space Sknew (0 (N )). For
instance, Ramanujans cusp form




(n) exp(2inz) := exp(2iz) (1 exp(2inz))24 (1.33)
n=1 n=1

is a normalized eigenform of weight 12 to the full modular group, and hence

a newform of level 1. Ramanujan [300] conjectured that the coecients (n)
are multiplicative and satisfy the estimate | (p)| 2p11/2 for every prime
number p. The multiplicativity was proved by Mordell [261], in particular by
the beautiful formula

mn

(m) (n) = d11 .

d2
d|(m,n)

The estimate was shown by Deligne. More precisely, Deligne [67] proved for
the coecients of any newform f of weight k the estimate
|c(n)| n(k1)/2 d(n), (1.34)

where d(n) := d|n 1 is the divisor function.
In the 1930s, Hecke [126] started investigations on modular forms and
Dirichlet series with a Riemann-type functional equation; his studies were
completed by Atkin and Lehner [6] (for newforms). Here we shall focus on
newforms. Given a newform f with Fourier expansion (1.32), we dene the
associated L-function by


c(n)
L(s, f ) = . (1.35)
n=1
ns

In view of the classic bound d(n)  n
it follows from (1.34) that the series
(1.35) converges absolutely for > k+1
2 . By the theory of Hecke operators, it
turns out that the Fourier coecients of newforms are multiplicative. Hence,
in the half-plane of absolute convergence, there is an Euler product represen-
tation and it is given by
 c(p)
1  
c(p) 1
1
L(s, f ) = 1 s 1 s + 2s+1k . (1.36)
p p p
p|N p
N
26 1 Introduction

Hecke [126], respectively, Atkin and Lehner [6], proved that L(s, f ) has an
analytic continuation to an entire function and satises the functional equation
N s/2 (2)s (s)L(s, f ) = (1)k/2 N (ks)/2 (2)sk (k s)L(k s, f ),

where = 1 is the eigenvalue of the AtkinLehner involution ( 01 N 0 ) on

Sk (0 (N )). Hecke proved a converse theorem which gives a characterization
of these L-functions by their functional equation; this beautiful result general-
izes Hamburgers theorem for the Riemann zeta-function [115] (see also [353,
Sect. 2.13]).
Laurinikas and Matsumoto [201] obtained a universality theorem for
L-functions attached to normalized eigenforms of the full modular group.
Laurinikas, Matsumoto and Steuding [204, 205] extended this result to
newforms:
Theorem 1.11. Suppose that f is a newform of weight k and level N . Let
K be a compact subset of the strip k2 < < k+1
2 with connected complement,
and let g(s) be a continuous non-vanishing function on K which is analytic in
the interior of K. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max |L(s + i, f ) g(s)| <  > 0.
T T sK

All known proofs of universality theorems rely on deep knowledge of the

coecients of the underlying Dirichlet series. Rankins celebrated asymptotic
formula [304] for the Fourier coecients c(n) of newforms f to the full modular
group states

x
c(p)2 p1k . (1.37)
log x
px

The main new ingredient in the proof of Theorem 1.11 is the use of an exten-
sion of Rankins formula to newforms to congruence subgroups 0 (N ). The
proof is inspired by Rankins approach; the non-vanishing of the associated
RankinSelberg L-functions (see Sect. 1.6) on the boundary of the critical
strip is essentially due to Moreno [262]. (See also Theorem 6.14.)
The universality of the logarithmic derivative and the rst derivative of
L-functions to normalized eigenforms for the full modular group were stud-
ied by Laurinikas [197]. Recently, Nagoshi [273, 274] proved universality
for L-functions to Maass forms (that are non-holomorphic eigenforms of the
LaplaceBeltrami operator on the upper half-plane). Delignes estimate (1.34)
played an essential part in the proofs of universality for L-functions attached
to holomorphic eigen- and newforms, respectively, but its analogue, the so-
called Ramanujan hypothesis, is not known for Maass forms in general. In
[273], Nagoshi used bounds of Kim and Sarnak for the local root numbers of
the Euler factors and obtained universality in a restricted range of the right
half of the critical strip; in [274], he proved unrestricted universality by use
of an asymptotic formula for the fourth moment of these local roots due to
 1.5 L-Functions Associated with Newforms 27

Murty. Remarkably, his argument also yields a proof for Theorem 1.11 with-
out using (1.34). Moreover, Nagoshi [275] proved universality for L-functions
to certain newforms and their derivatives in the level aspect.
We return to L-functions associated with holomorphic modular forms.
Laurinikas and Matsumoto [202] obtained a joint universality theorem for L-
functions associated with newforms twisted by characters. Let f Sk (0 (N ))
be a newform with Fourier expansion (1.32) and let be a Dirichlet character
mod q where q is coprime with N . The twisted L-function is dened by


c(n)
L (s, f ) = (n).
n=1
ns

As in the non-twisted case (1.35), this Dirichlet series has an Euler product
and extends to an entire function.

Theorem 1.12. Let q1 , . . . , qn be positive integers coprime with N and let

1 mod q1 , . . . , n mod qn be pairwise non-equivalent character. Further, for
1 j n, let gj be a continuous function on Kj which is non-vanishing in
the interior, where Kj is a compact subset of the strip {s C : k2 < < k+1
2 }
with connected complement. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max max |Lj (s + i, f ) gj (s)| < > 0.
T T 1jn sKj

The proof relies on a joint limit theorem due to Laurinikas [191] and some
kind of prime number theorem for the coecients of cusp forms with respect
to arithmetic progressions, namely

1 x
c(p)2 p1k , (1.38)
px
(q) log x
pa mod q

where a is coprime with q. The key for the proof of the latter result is again
the extension of Rankins asymptotic formula (1.37) to congruence subgroups
0 (N ); the transition to arithmetic progressions relies on analytic properties
of the twisted RankinSelberg L-function.
By Wiles celebrated proof of the ShimuraTaniyama conjecture for semi-
stable modular forms [371] (which was sucient to prove Fermats last the-
orem), and the proof by Breuil, Conrad, Diamond and Taylor of the general
case [44], every L-function attached to an elliptic curve over the rationals is
the L-function to some newform of weight 2 for some congruence subgroup.
Consequently, Theorem 1.11 proves the universality of L-functions associated
with elliptic curves. Laurinikas and Steuding [209] used Theorem 1.12 to give
an example of jointly universal L-functions associated with elliptic curves.
Here one may choose any nite family of elliptic curves of the form

Em : Y 2 = X 3 m2 X with squarefree m N;
28 1 Introduction

these curves were rst studied by Tunnell [356] with respect to the congruent
number problem. For this family one can avoid Wiles proof of the Shimura
TaniyamaWeil conjecture but show more or less directly that the L-function
associated with E1 corresponds to a newform f S2 (0 (32)) and  that the
L-function to Em is a twist of E1 with the Kronecker symbol m. (see also
[171]).

1.6 The LinnikIbragimov Conjecture

It is known that there exists a rich zoo of Dirichlet series having some univer-
sality property. We list now some more signicant examples.
The Dedekind zeta-function of a number eld K over Q is given by

1  1
1
K (s) = = 1 ,
a
N(a)s p
N(p)s

where the sum is taken over all non-zero integral ideals, the product is taken
over all prime ideals of the ring of integers of K, and N(a) is the norm of the
ideal a. The Riemann zeta-function may be regarded as the Dedekind zeta-
function for Q. Universality for the Dedekind zeta-function was rst obtained
by Voronin [364], Gonek [104], and, in full generality, by Reich [305, 306];
here the strip of universality is restricted to the strip max{ 12 , 1 d1 } < < 1,
where d is the degree of K over Q (we will later see the reason behind).
Reichs universality theorem [306] is discrete, i.e., under the same conditions
as in Theorem 1.9, for any real = 0 and any  > 0, the relation

max |K (s + in) g(s)| <  (1.39)

sK

holds for a set of positive integers n with positive lower density. (We shall
consider discrete universality more detailed in Sect. 5.7 and Dedekind zeta-
functions in Sect. 13.1.) Furthermore, in [305], Reich obtained universality for
Euler products formed with Beurling primes under the assumption of certain
extra conditions.
Dirichlet series attached to multiplicative functions appear naturally in
many aspects of number theory. They were intensively studied by Laurinikas
and leeviciene. Laurinikas [181] succeeded to prove universality for
Dirichlet series associated with a multiplicative function f , formally given by




f (n) 
f (pk )
= 1+ ,
n=1
ns p
pks
k=1
 1.6 The LinnikIbragimov Conjecture 29

subject to certain conditions on f as, for example, the existence of the mean
value
1

M (f ) = lim f (n),
x x
nx

the boundedness of f , and a technical condition on the local Euler factors.

In [207], Laurinikas and leeviiene introduced another class of Dirichlet
series attached to multiplicative functions and proved also universality for its
elements, and leeviiene [332] obtained joint universality for certain twists
by Dirichlet characters.
In [239] Matsumoto introduced the now called Matsumoto zeta-functions
and studied their value distribution; he obtained far-reaching generalizations
of the classical results of Bohr and his collaborators. For each positive integer
(j)
m let g(m) and f (j, m) with 1 j g(m) be positive integers, and let am
with 1 j g(m) be given complex numbers. Further, dene the polynomial


g(m)
Am (X) = (1 am
(j) f (j,m)
X ).
j=1

Denoting the mth prime number by pm , the associated Matsumoto zeta-

function is dened by


(s) = Am (ps
m )
1
.
m=1
(j)
Under the conditions g(m)  p
m and |am | pm the product dening (s)
converges for > 1++ and denes an analytic function in this region. The
Matsumoto zeta-function is an interesting generalization of many well-known
zeta-functions (e.g., Dedekind zeta-functions). Suppose now that (s) has an
analytic continuation to some half-plane > + + with 12 < 1 (except
for at most nitely many poles) with (s)  |t|c1 as |t| , where c1 is
some positive constant and bounded mean-square. Assuming
 
 g(m) 


 (j) 
 am  pm c2 > 0, (1.40)
 j=1 
 f (j,m)=1 

and, additionally, some technical condition, Laurinikas [190] obtained so-

called universality with weight, that is the relation
   
1 T
lim inf ( ) 1 ; : max |(s + i ) f (s)| <  d > 0
T U T sK
0

where T0 is a suciently large constant, is a positive function of bounded

variation, 1( ; A) is the indicator function of the set A,
 T
U := U (T, ) := ( ) d,
T0
30 1 Introduction

and K is a compact subset of the strip + + < < 1 + + with

connected complement.
A useful tool in the theory of modular forms are RankinSelberg L-
functions, introduced by Rankin [302, 303], and Selberg [320] (independently).
Given two cusp forms f and g of the full modular group with Fourier
series coecients a(n) and b(n), respectively, the associated RankinSelberg
L-function is dened by


a(n)b(n)
L(s, f g) = (2s) .
n=1
ns

For normalized eigenforms f to the full modular group, Matsumoto [241]

proved universality for L(s, f f ). The universality for such L-functions in
the level aspect was studied by Mishou and Nagoshi [254].
To mention only some more important examples of universal L-functions
(some of them we shall see in Chap. 13 again) this list can be continued with
Hecke L-functions to grssencharacters (Mishou [249, 250], and addition-
ally in the character aspect by Mishou and Koyama [174]);
Artin L-functions [16];
zeta-functions associated with nite abelian groups (Laurinikas [194,
196]).
There are other interesting examples which even have a stronger univer-
sality property: they can approximate functions with zeros on a set of posi-
tive lower density. The rst one is, of course, the logarithm of the Riemann
zeta-function (or the derivative). This follows immediately from the proof of
Voronins universality Theorem 1.7. Note that log (s) has a Dirichlet series
representation for > 1 (see (1.14)). The same argument applies to all uni-
versal Euler products.
Now we present a completely dierent example. For 0 < 1, R, the
Lerch zeta-function is given by


exp(2in)
L(, , s) = . (1.41)
n=0
(n + )s

This series converges absolutely for > 1. The analytic properties of L(, , s)
are quite dierent, if Z or not. If  Z, the series converges for > 0
and L(, , s) can be continued analytically to the whole complex plane. For
Z the Lerch zeta-function becomes the Hurwitz zeta-function


1
(s, ) = ; (1.42)
m=0
(m + )s

this function has an analytic continuation to C except for a simple pole at

s = 1 with residue 1. Denote by {} the fractional part of a real number .
Setting
 1.6 The LinnikIbragimov Conjecture 31

1 if Z,
+ = 1 {} and = (1.43)
{} otherwise,

one can prove the functional equation

(s) s

L(, , 1 s) = exp 2i L(, , s)

(2)s 4
s

+ exp 2i + + L(, + , s) . (1.44)

4
Twists with additive characters destroy the point symmetry of Riemann-type
functional equations. The function (s, ) was introduced by Hurwitz [137],
and L(, , s) by Lipschitz [222] and Lerch [215] (independently). Hurwitz
and Lerch also gave the rst proofs of the corresponding functional equations.
For more details on the Lerch zeta-function we refer to the monograph of
Garunktis and Laurinikas [89].
Gonek [104] and Bagchi [9] (independently) obtained strong universal-
ity for the Hurwitz zeta-function (s, ) if is transcendental or rational
= 12 , 1 (we will explain this restriction below and in Sects. 11.4). Laurinikas
[189] extended this result by proving that the Lerch zeta-function L(, , s)
is strongly universal if is not an integer and is transcendental. The joint
universality of Lerch zeta-functions L(j , j , s) was treated by Laurinikas
and Matsumoto [200, 203] and Nakamura [276]; here the parameters j have
to be algebraically independent over Q. This and the case of joint univer-
sality of Hurwitz zeta-functions with rational parameters will be studied
in Sect. 11.4.
The list of known zeta-function having the strong universality property
can be extended by, for example,

Estermanns zeta-function, dened by

 
 
k (n) k
E s; , = exp 2in ,
 n=1
ns 

where (n) is the generalized divisor function given by (n) = d|n d
for any complex (Garunktis et al. [90]).
General Dirichlet series, given by



an exp(sn ),
n=1

satisfying certain conditions on the coecients an and the exponents

n (Laurinikas, Schwarz and Steuding [206], respectively, Genys and
Laurinikas [98]). Unfortunately, these conditions are quite restrictive.
32 1 Introduction

For example, it is not known whether Selberg zeta-functions are universal

or not.

All examples of strongly universal Dirichlet series do not have an Euler

product and have many zeros in their region of universality; indeed, we shall
see (in Chaps. 8, 10 and 11) that the property of approximating analytic
functions with zeros is intimately related to the distribution of zeros of the
Dirichlet series in question. In particular, Euler products for which the ana-
logue of Riemanns hypothesis is expected to hold should not be capable of
approximating functions with zeros.
Roughly speaking, there are two methods to prove universality. First, one
can try to mimic Voronins proof or Bagchis probabilistic approach (here we
do not distinguish between these two since they have a lot in common). This
sounds more simple than it actually is, because one has to assure many ana-
lytic and arithmetic properties of the function in question. The second way is
to nd a representation as a linear combination or a product representation
of jointly universal functions (we will use this argument in Chaps. 1013). All
known proofs of universality of the rst type depend on a certain kind of
independence. For instance, the logarithms of the prime numbers are linearly
independent over Q (we used this property in the proof of Voronins univer-
sality Theorem 1.7 when we applied Weyls renement of Kroneckers approx-
imation theorem, Lemma 1.8). Another example are the numbers log(n + )
with non-negative integral n which are linearly independent over Q if is
transcendental. In order to prove universality for the Hurwitz zeta-function
(1.42), the rst type of proof yields the result aimed at for transcendental .
If is rational = 12 , 1, one can nd a representation of (s, ) as a linear com-
bination of non-equivalent Dirichlet L-functions for which we have the joint
universality Theorem 1.10; in the cases = 12 and = 1 the Hurwitz zeta-
function has an Euler product representation and is equal to the Riemann
zeta-function for = 1, resp., for = 12 ,
 
1
s, = 2s L(s, ),
2

where is the unique character mod2. In both cases the Hurwitz zeta-
function is universal but does not have the strong universality property. It
is an interesting open problem whether (s, ) is universal or even strongly
universal if is algebraic irrational. In this case neither of the two methods
seem to applicable directly. Laurinikas and Steuding [210, 211] obtained rst
limit theorems for Hurwitz zeta-functions with algebraic parameters, which
could be a rst step towards universality.
It was conjectured by Linnik and Ibragimov (cf. [207]) that all func-
tions given by Dirichlet series and analytically continuable to the left of the
half-plane of absolute convergence are universal. However, this has to be
 1.6 The LinnikIbragimov Conjecture 33

understood as a program and not as a conjecture. For example, put a(n) = 1

if n = 2k with k N0 := N {0}, and a(n) = 0 otherwise. Then

a(n)
1
s
= ks
= (1 2s )1 ,
n=1
n 2
k=0

and obviously, this function is far away from being universal. So we have to
ask for natural conditions needed to prove universality. Dirichlet series having
an innite Euler product seem to be good candidates.
2
Dirichlet Series and Polynomial Euler Products

What is man in nature? Nothing in relation to the innite, all in

relation to nothing, a mean between nothing and everything.
Blaise Pascal

In this chapter, we introduce a class of Dirichlet series satisfying several

quite natural analytic axioms in addition with two arithmetic conditions,
namely, a polynomial Euler product representation and some kind of prime
number theorem. The elements of this class will be the main actors in the
sequel; however, for some of the later results we do not need to assume all of
these axioms. Further, we shall prove mean-value estimates for the Dirichlet
series coecients of these L-functions as well as asymptotic mean-square
formulae on vertical lines in the critical strip. These estimates will turn out
to be rather useful in later chapters.

2.1 General Theory of Dirichlet Series

We start with a brief introduction to the general theory of ordinary Dirichlet
series. For more details and proofs we refer to Titchmarsh [352, Chap. IX].
An ordinary Dirichlet series is a series of the form


an
A(s) = , (2.1)
n=1
ns

where the coecients an are complex numbers and s = + it is a complex

variable. If such a Dirichlet series is convergent for s = s0 , then it is uniformly
convergent throughout the angular region given by

| arg(s s0 )| ,
2
where is any real number satisfying 0 < < 2 . Consequently, if a Dirichlet
series converges, it converges in some half-plane > c . Obviously, every
36 2 Dirichlet Series and Polynomial Euler Products

term of the Dirichlet series (2.1) is analytic and any point s with > c
is contained in a domain of uniform convergence. A well-known theorem of
Weierstrass states that the limit of a uniformly convergent sequence of analytic
functions is analytic. Hence, A(s) is an analytic function in its half-plane of
convergence > c . The abscissa of convergence is given by
 
 N   
log  n=1 an  log  n=N +1 an 
c = lim sup or c = lim sup
N log N N log N

according to whether n=1 an diverges or converges. Also the region of abso-
lute convergence of a Dirichlet series is a half-plane. The abscissa a of absolute
convergence is given by


N  
log |a |
n=1 n log n=N +1 |an |
a = lim sup or a = lim sup
N log N N log N

according to whether n=1 |an | is divergent or not. It is easily seen that
a c 1 (with equality, for example, for Dirichlet L-functions). If all
coecients an are non-negative real numbers, then the real point s = c is
a singularity of A(s) (e.g., the zeta-function). A function has at most one
Dirichlet series representation, i.e., given two Dirichlet series



an bn
A(s) = and B(s) = ,
n=1
ns n=1
ns

both absolutely convergent for > a , if A(s) = B(s) for each s in an innite
sequence {sk } such that Re sk as k , then an = bn for all n N.
This uniqueness implies the existence of some zero-free half-plane > 0 with
0 a .
Assume that A(s) is analytic in some strip 1 2 except for at most
a nite number of poles. Then A(s) is said to be of nite order in this strip if
there exists a positive constant c such that the estimate

A(s)  |t|c as |t| (2.2)

holds uniformly for 1 2 ; similarly, one denes the notion of nite

order for half-planes 1 . Clearly, a function given by a Dirichlet series is
of nite order in its half-plane of convergence. Given , dene () to be the
lower bound of all c for which (2.2) holds; this quantity is called the order of
A(s). One can show that the function () is convex downwards (in particular,
it is continuous). Moreover, () = 0 for > a and, conversely, the Dirichlet
series (2.1) is convergent in the half-plane where A(s) is regular and () = 0.
If
denotes the abscissa limiting the half-plane where A(s) is analytic with
() = 0, then c
a .
 2.2 A Class of Dirichlet Series: The Main Actors 37

Now we shall consider the mean-square of A(s) on vertical lines. In the

half-plane of absolute convergence for (2.1) it is easily shown that
 T

1 |a(n)|2
|A( + it)|2 dt . (2.3)
2T T n=1
n2
The convergence of the series on the right-hand side is a matter of course.
Carlson [50] proved that the mean-square of A(s) on vertical lines exists if
the left-hand side is bounded as T .
Theorem 2.1. Let A(s) be given by (2.1). Assume that A(s) is analytic except
for at most a nite number of poles, of nite order for 1 , and
 T
1
lim sup |A( + it)|2 dt < .
T 2T T

Then,


1 T
|an |2
lim |A( + it)|2 dt =
T 2T T n=1
n2
for > 1 , and uniformly in any strip 1 +  2 .
Carlsons theorem may be regarded as an analogue of Parsevals theorem
for Fourier series. We omit the proof but refer to Carlson [50] and
Titchmarsh [352].
In view of Carlsons theorem it makes sense to dene a so-called mean-
square half-plane by dening its abscissa m as the inmum over all 1 such
that (2.3) holds for all xed > 1 . One can show that
 
1
m max a , c (2.4)
2
and that () 12 for any > m . In our applications we will normalize the
Dirichlet series to be absolutely convergent for > a = 1. Inequality (2.4)
implies that (at least with present day methods) the strip of universality is
at most 12 < < 1; it seems reasonable that we cannot approximate analytic
functions uniformly on sets having a non-empty intersection with the half-
plane 12 .

2.2 A Class of Dirichlet Series: The Main Actors

Now we introduce several axioms for Dirichlet series; some of them might look
a bit technical on rst sight but they have been proved to be rather useful in
the analytic theory of Dirichlet series.
In the sequel we assume that a function L(s) has a representation as a
Dirichlet series in some half-plane:


a(n)
L(s) = . (2.5)
n=1
ns
38 2 Dirichlet Series and Polynomial Euler Products

First of all, we pose a hypothesis on the size of the Dirichlet series coecients.
(i) Ramanujan hypothesis. a(n)  n
for any  > 0, where the implicit con-
stant may depend on .
The Ramanujan hypothesis implies the absolute convergence of the Dirichlet
series (2.5) in the half-plane > 1, and uniform convergence in every compact
subset. Thus, L(s) is analytic for > 1 and it makes sense to ask for analytic
continuation.
(ii) Analytic continuation. There exists a real number L such that L(s) has
an analytic continuation to the half-plane > L with L < 1 except for
at most a pole at s = 1.
Actually, we could allow a nite number of poles on the line = 1, however,
we do not loose too much by the restriction above.
The next axiom excludes functions which have an extraordinary large order
of growth; Dirichlet series satisfying a functional equation are known to grow
moderately slowly (by the PhragmnLindelf principle; see Sect. 6.5).
(iii) Finite order. There exists a constant L 0 such that, for any xed
> L and any  > 0,
L( + it)  |t|L +
as |t| ;
again the implicit constant may depend on .
The nal two axioms are of purely arithmetical nature. Both are rather
restrictive; however, they t well to what we think an L-function should look
like. All known examples of L-functions of number theoretical interest are
automorphic or at least conjecturally automorphic L-functions, and for all of
them it turns out that the local Euler factors for prime p are the inverse of a
polynomial.
(iv) Polynomial Euler product. There exists a positive integer m and for every
prime p, there are complex numbers j (p), 1 j m, such that
 m  1
j (p)
L(s) = 1 .
p j=1
ps

The j (p) are called local roots at p. This axiom implies that the Dirichlet
series coecients a(n) are multiplicative (this will be proved in Lemma 2.2).
The Ramanujan hypothesis (i), in addition with the polynomial Euler product,
implies that the numbers j (p) have absolute value less than or equal to one
(see Lemma 2.2).
The last axiom may be regarded as some kind of prime number theorem
for the coecients of the Euler product:
(v) Prime mean-square. There exists a positive constant such that
1

lim |a(p)|2 = .
x (x)
px
 2.2 A Class of Dirichlet Series: The Main Actors 39

This asymptotic formula is intimately related to the deep Selberg conjectures

for L-functions in the Selberg class (see Chap. 6) and it is known to be satised
in many cases. This last axiom implies that there are innitely many primes
p for which not all numbers j (p) vanish. Consequently, the Euler product is
innite (and it makes sense to ask for universality).
In the sequel we shall not always use all of these axioms; however, we
denote the class of Dirichlet series (2.5) satisfying all axioms (i)(v) by S.
In Lemma 2.2 we will show that (iv) (i) and thus the assumption of the
Ramanujan hypothesis is superuous for the denition of the class S; however,
besides S we shall also consider other classes of Dirichlet series where we
assume (i) but not (iv). The class S is not empty. Obviously, Riemanns zeta-
function (s) is an element of S. Further examples are Dirichlet L-functions,
Dedekind zeta-functions, and many more. As a matter of fact, S is a subset of
the large class of Matsumoto zeta-functions, introduced by Matsumoto [239].
The restrictions going beyond (in particular, axioms (ii) and (v)) will turn out
to be rather important in the sequel. As already noted, they are motivated by
naturally appearing L-functions in number theory.
We give a simple example of an element in S. Assume that f is a newform
of weight k to some congruence subgroup 0 (N ) with Fourier expansion (1.32).
Writing
1k
a(n) = c(n)n 2
we nd via (1.36) the Euler product representation
   1
k1 a(p) (p)
L s+ ,f = 1 s + 2s ,
2 p
p p

where (p) = 0 if p | N , and (p) = 1 otherwise. In the latter case, i.e., p | N ,

then the corresponding Euler factor can be rewritten as
 1  1  1
a(p) 1 1 (p) 2 (p)
1 s + 2s = 1 1 ,
p p ps ps

where 1 (p), 2 (p) are complex numbers satisfying

1 (p) + 2 (p) = a(p) and 1 (p)2 (p) = 1.

Delignes estimate (1.34) translates into

|1 (p)| = |2 (p)| = 1, i.e., 1 (p) = 2 (p).

The transformation s  s + k1 2 maps the critical strip to 0 1

(independent of the weight). In the sequel we shall assume that L-functions
to modular forms are normalized in this way, and we denote them again by
L(s, f ). Note that the asymptotic formula (1.37) yields the prime number
theorem for the coecients of L(s, f ). It follows that L-functions to newforms
are in our class S.
40 2 Dirichlet Series and Polynomial Euler Products

2.3 Estimates for the Dirichlet Series Coecients

Mean-value estimates are very useful tools in analytic number theory. They
give information on the size of arithmetical objects. Now we shall prove some
elementary estimates for the coecients a(n) in the Dirichlet series expansion
(2.5). First of all, we study the relation between the Ramanujan hypothesis
and the size of the local roots of a polynomial Euler product.
Lemma 2.2. Suppose that L(s) is given by (2.5) and satises axiom (iv).
Then a(n) is multiplicative and


m
a(n) = j (p)kj , (2.6)
p|n k1 ,...,km 0 j=1
k1 +...+km =(n;p)

where (n; p) is the exponent of the prime p in the prime factorization of the
integer n. Moreover, if |j (p)| 1 for 1 j m and all primes p, then
a(n)  n
for any  > 0, and vice versa.
Thus, a polynomial Euler product of the form (iv) with xed degree m
together with the Ramanujan hypothesis (i) is equivalent to |j (p)| 1 for
1 j m and all primes p.
Proof. Taking into account the shape of the Euler product (iv) we have the
identity
m  1   


a(n)   j (p)
m
j (p)k
= 1 = 1+ ,
n=1
ns p j=1
ps p j=1
pks
k=1

valid for suciently large . This implies (2.6) and that a(n) is multiplicative.
Now we show that the estimate for the local roots implies the Ramanujan
hypothesis. If |j (p)| 1, then we get


|a(n)| 1 = dm (n), (2.7)
p|n k1 ,...,km 0
k1 +...+km =(n;p)

say. The generalized divisor function dm (n) appears as Dirichlet coecients

in the representation
 1
m

dm (n)
m
(s) = 1 s = ,
p
p n=1
ns

and is therefore multiplicative (by the same reasoning as for a(n)). For p prime
and N, we have
dm (p ) = {(k1 , . . . , km ) Nm
0 : k1 + . . . + km = }
 
m+1
= , (2.8)

 2.3 Estimates for the Dirichlet Series Coecients 41

  m + (n; p) 1 
and thus
dm (n) = .
p
(n; p)

The function dm (n) is the m-fold Dirichlet convolution of the function con-
stant 1: writing

(f g)(n) = f (b)g(n/b),
b|n

we have dk = dk1 1. Thus,

dm (n) = dm1 (b) dm1 (n) dm1 (n)d(n),
b|n b|n

where, d(n) = d2 (n) = d|n 1; here the rst inequality follows from (2.8).
Hence, for m 2,
dm (n) d(n)m1 . (2.9)
It remains to estimate the divisor function d(n). We start with
  
d(n) (1 + (n; p)) (1 + (n; p)) 2(n;p)
p px p>x
 x  log 2/ log n
log n
1+ p(n;p) .
log 2 p

Let n e3 . Taking x = log n(log log n)3 , we obtain

   
log n log log log n
d(n) exp log 2 1+O . (2.10)
log log n log log n
Hence, we obtain
 
log n
|a(n)| < exp (1 + )(m 1) log 2 (2.11)
log log n
for any  > 0 and suciently large n. This proves that the Ramanujan
hypothesis (i) holds provided |j (p)| 1.
Now assume that the Ramanujan hypothesis (i) is true. Consider the iden-
tity between the Dirichlet series and the Euler product representation, respec-
tively, the power series expansion



m
k
a(p )X = k
(1 j (p)X)1 ;
k=1 j=1

by the Ramanujan hypothesis, the series is analytic for |X| < 1, hence so is
the right-hand side which gives the desired bound. This concludes the proof
of the lemma. 

42 2 Dirichlet Series and Polynomial Euler Products

Estimate (2.10) is best possible. To see this note that

d(n) log log n

lim sup = log 2
n log n

(a proof can be found in [120, Sect. 18.1]). Estimate (2.11) is near to be best
possible since, if d(n) is replaced by dm (n), then the latter formula holds with
the right-hand side replaced by log m (for a proof, see [255]).
For our studies on the value-distribution of functions in S we need to prove
several mean-square formulae. It follows immediately from the Ramanujan
hypothesis (i) that

|a(n)|2  x1+
. (2.12)
nx

This rough bound is sucient for many applications. However, we can do a bit
better by a standard argument. Recall that m is the degree of the polynomial
in ps in the local Euler factor of L, independent of p.

Lemma 2.3. Suppose L(s) satises axioms (i) and (iv). Then, as x ,


|a(n)|2  x(log x)m 1 .
2

nx

Proof. In view of (2.7) it suces to nd a mean-square estimate for dm (n).

By the multiplicativity of dm (n),


dm (n)2 = g(d)
d|n

with some multiplicative function g. By (2.8), we nd

g(1) = dm (1)2 = 1, g(p) = dm (p)2 dm (1)2 = m2 1,

and by induction

m2
g(p ) = dm (p )2 dm (p1 )2 ,
!
as . Hence, we obtain






g(d) 
g(p )
2
dm (n) = g(d) x x 1+ .
d =1
p
nx dx nx dx px
n0 mod d

In view of the above estimates the right-hand side above equals asymptotically
 

m2 1
m2  m2 1

x 1+ + = x 1 + + O(x).
p =2
!p p
px px
 2.4 The Mean-Square on Vertical Lines 43

A classic estimate due to Mertens,

 1
  1
1
1+  1  log x
p p
px px

(see [120, Sect. 22.8]), in combination with

 m2 1
m2 1 1
1+ 1+ ,
p p
gives the estimate of the lemma. 


2.4 The Mean-Square on Vertical Lines

Our next aim is to derive an asymptotic mean-square formulae for L(s) on
some vertical lines to the left of the abscissa of absolute convergence = 1.
Theorem 2.4. Let L(s) satisfy the axioms (i)(iii). Then, for any satisfy-
ing  
1 1 L
> max ,1 , (2.13)
2 1 + 2L
we have  T

1 |a(n)|2
lim |L( + it)|2 dt = .
T 2T T
n=1
n2
In particular, the mean-square half-plane of L(s) has a non-empty intersection
with the strip 12 < < 1.
This result is essentially due to Carlson [50, 51]; we follow closely
Titchmarshs proof of the special case of the Riemann zeta-function in [353,
Sect. 7.9].
Proof. With regard to Carlsons Theorem 2.1 it suces to show that
 T
|L( + it)|2 dt  T (2.14)
T

for satisfying (2.13). For this aim we consider for > 0 the Dirichlet series


a(n)
exp(n).
n=1
ns

Since a(n)  n
by (i), this series converges absolutely in the whole complex
plane and uniformly in any compact subset. Let > 1 and c > be a constant.
By Mellins inversion formula,
 +i
1
exp() = (z)z dz, (2.15)
2i i
44 2 Dirichlet Series and Polynomial Euler Products

valid for positive real numbers and , we get

 c+i
a(n) a(n) 1
s
exp(n) = s 2i
(z)(n)z dz.
n=1
n n=1
n ci

Hence,

 c+i
a(n) 1
s
exp(n) = (z s)L(z) sz dz. (2.16)
n=1
n 2i ci

Here, interchanging summation and integration is allowed in view of the abso-

lute convergence of the Dirichlet series dening L(s) and Stirlings formula,
   
1 1 1
log (s) = s log s s + log 2 + O , (2.17)
2 2 |s|
valid for | arg s|  and |s| 1, uniformly in , as |s| . By (iii),

L(c + i )  | |L +
(2.18)

for > L as | | . This and Stirlings formula imply

 c+i
(z s)L(z) sz dz
ci
  
+
L +
+c 12 | |
 c
(1 + | | ) exp d.
2
Hence, the function on the right-hand side of (2.16) is an analytic function of
s in the half-plane > L for any c > L . Therefore, let > max{ 12 , L } and
choose 1 := Re z such that max{ 1, L } < 1 < 1. We move the path of
integration in (2.16) to the line Re z = 1 . We pass the simple pole of (z s)
at z = s with residue L(z). Moreover, by (ii), L(z) has at most a pole at z = 1
of order j, say. The residue coming from the hypothetical pole of L(z) is a
nite sum of terms of the form

c (u) (s 1) s1 (log )v ,

where c is a certain constant and u and v are non-negative integers with

u + v = j; by Stirlings formula (2.17), these terms are

 +
1 exp(BT ),

where B is a positive absolute constant. Now let > |T |B . Then, for >
|t|B we deduce from (2.16) by the calculus of residues

 1 +i
a(n) 1
L(s) = s
exp(n) (z s)L(z) sz dz
n=1
n 2i 1 i

+ O(exp(B|t|)). (2.19)
 2.4 The Mean-Square on Vertical Lines 45

Now we integrate the square of this expression with respect to t = Im s in the

range 12 T t T . Note that
  2  2

a(n)   1 +i 
   
|L( + it)|2   exp(n) +  (z s)L(z) sz
dz 
 ns   1 i 
n=1
+ O(exp(2B|t|)).

We nd
  2
T 

 a(n) 
 exp(n) dt
1  ns 
2T n=1





|a(n)|2 a(m)a(n) exp((m + n))
T exp(2n) + .
n=1
n2
m=1 n=1
(mn) | log m
n|
n=m

By the Ramanujan hypothesis (i), the rst series on the right-hand side is
convergent independent of > 0. The double series is bounded by



a(m)a(n) exp((m + n))
n .
n=1 m<n
(mn) log m

We split the summation and apply the Ramanujan hypothesis (i) with a suit-
ably small  > 0 to obtain the upper bound




exp((m + n))
+ n .
n n (mn)
log m
n=1 m< 2 2 m<n

n
In the rst sum, we have log m > log 2, and so it is bounded by

2  2

exp(n)
 u

exp(u) du  22
;
n=1
n
0

recall that  denotes an arbitrary small positive quantity, not necessarily the
same at each appearance. In the second sum we write m = n r with 1
r n2 . Then log m
n
= log(1 nr ) > nr , and we get in a similar manner the
upper bound



1 1
n1+2(
) exp(n)  22
log .
n=1 n r
r 2

Hence,



a(m)a(n) exp((m + n))
log n
 22
.
n=1 m<n
(mn) m
46 2 Dirichlet Series and Polynomial Euler Products

Thus,
  2
T 

 a(n) 
 exp(n) dt  T + 22
. (2.20)
1  ns 
2T n=1

Let z = 1 + i . Then
 1 +i 
(z s)L(z) sz
dz  1
| (z s)L(z)| d.
1 i

By the CauchySchwarz inequality this is bounded by

  1/2
1 | (z s)| d | (z s)L(z)2 | d .

By Stirlings formula (2.17) the rst integral is bounded. For the second one
we additionally apply (2.18) and obtain, for |t| T ,
  
2T
+ | (z s)L(z)2 | d
2T
    
2T
2L +
+1 (1/2) | |
 + | | exp d  exp(BT )
2T 4

with some absolute positive constant B (since |t | 12 | |). Consequently,

 T  1 +i 2

 
 (z s)L(z) sz dz  dt
1  1 i 
2T

 2T  T 
 221
| (z s)| dt |L(1 + i )|2 d + 221
2T 1
2T
 2T
 221 |L(1 + i )|2 d.
2T

By the trivial estimate

 2T
|L(1 + i )|2 d  T 1+2L +
,
2T

we get
  2
T  1 +i 
 
 (z s)L(z) sz dz  dt  221 T 1+2L +
.
1  1 i 
2T

This, (2.19), and (2.20) give

 T
|L( + it)|2 dt  T + 22
+ 221 T 1+2L +
. (2.21)
1
2T
 2.4 The Mean-Square on Vertical Lines 47

Using this with 21n T instead of T and summing up over all n N, we nd

that the same bound holds for the integral with limits 1 and T . The same
reasoning applies to the line segment [T, 1]. Thus, putting
1+2L +

=T 2(11 )
,

we nd, after a short computation, that all terms of the right-hand side of
(2.21) are  T if
 
1 1 1
> max , L , 1 .
2 1 + 2L + 

Sending  0 and since 1 > max{ 12 , L } can be chosen arbitrarily, we nd

that estimate (2.14) holds for the range (2.13). This proves the theorem.  

We give an example. Using (1.3) with N = t, we nd for the zeta-function

= 12 for = 12 which leads to the existence of the mean-square for > 34 .
This is far from being optimal. As a matter of fact, the mean-square exists for
> 12 but we may not forget that here we did not use any deeper property
of the zeta-function (only the rough estimate via (1.3)). In Sect. 6.5, we will
signicantly improve the range (2.13) of Theorem 2.4 for L-functions satisfying
a Riemann-type functional equation.
3
Interlude: Results from Probability Theory

Primes play a game of chance.

M. Kac

In this chapter, we briey present facts from probability theory which will
be used later. These results can be found in the monographs of Billingsley
[21, 22], Buldygin [45], Cramr and Leadbetter [64], Heyer [133], Laurinikas
[186], and Love [226]. However, there are two exceptions in this crash course
in probability theory. In Sect. 3.3 we present Denjoys heuristic probabilistic
argument for the truth of Riemanns hypothesis. Finally, in Sect. 3.7, we
introduce the universe for our later studies on universality, the space of ana-
lytic functions, and state some of its properties, following Conway [62] and
Laurinikas [186].

3.1 Weak Convergence of Probability Measures

The notion of weak convergence of probability measures is a useful tool in
investigations on the value-distribution of Dirichlet series. This powerful
theory was initiated by Kolmogorov, Erds and Kac and further developed
by Doob, Prokhorov, Skorokhod and others. To present the main properties
of weakly convergent probability measures we have to introduce the concept
of -eld and the axiomatic setting of probability measures.
Let be a non-empty set. By P() we denote the set of all subsets of .
A subset F of P() is called a eld (or algebra) if it satises the following
axioms:
, F;
Ac F for A F, where Ac denotes the complement of A;
F is closed under nite unions and nite intersections, i.e., if A1 , . . . , An
F, then

n 
n
Aj F and Aj F.
j=1 j=1
50 3 Interlude: Results from Probability Theory

F is called a -eld (or -algebra) if it satises the rst two axioms above in
addition with
F is closed under countable unions and countable intersections, i.e., if {Aj }
is a countable sequence of events in F, then



Aj F and Aj F.
j=1 j=1

For C P() we denote by (C) the smallest -eld containing C. This -eld
is said to be generated by C.
A non-negative function P dened on a -eld F with the properties:
P() = 0 and P() = 1;
For every countable sequence {Aj } of pairwise disjoint elements of F,





P Aj = P(Aj ),
j=1 j=1

is called a probability measure. The triple (, F, P) is said to be a probability

space. This setting of probability dates back to Kolmogorov who introduced
it in the early 1930s.
Let S be a topological space and let B(S) denote the class of Borel sets
of S, i.e., the -eld generated by the system of all open subsets of the space
S. Then each measure on B(S) is called Borel measure. Usually, we consider
probability measures dened on the Borel sets B(S) of some metric space S.
A class A of sets of S is said to be a determining class (also separating class)
in case the measures P and Q on (S, B(S)) coincide on the whole of S when
P(A) = Q(A) for all A A.
Given two probability measures P1 and P2 on (S1 , B(S1 )) and (S2 , B(S2 )),
respectively, there exists a unique measure P1 P2 such that

(P1 P2 )(A1 A2 ) = P1 (A1 )P2 (A2 )

for Aj B(Sj ). This measure is a probability measure on (S, B(S)), where

S = S1 S2 and B(S) = B(S1 ) B(S2 ), and is said to be the product measure
of the measures P1 and P2 .
In the sequel let Pn and P be probability measures on (S, B(S)). We say
that Pn converges weakly to P as n tends to innity, and write Pn P, if
for all bounded continuous functions f : S R
 
lim f dPn = f dP.
n S S

Since the integrals on the right-hand side completely determine P (which is

a consequence of Lebesgues dominated convergence theorem), the sequence
 3.1 Weak Convergence of Probability Measures 51

{Pn } cannot converge weakly to two dierent limits at the same time. Further,
note that weak convergence depends only on the topology of the underlying
space S, not on the metric that generates it.
A set A in S whose boundary A satises P(A) = 0 is called a continuity
set of P. The Portmanteau theorem provides useful conditions equivalent to
weak convergence.
Theorem 3.1. Let Pn and P be probability measures on (S, B(S)). Then the
following assertions are equivalent:
Pn P;
For all open sets G,
lim inf Pn (G) P(G);
n
For all continuity sets A of P,
lim Pn (A) = P(A).
n

This theorem is part of Theorem 2.1 in Billingsley [21]. Next we state a useful
criterion for weak convergence.
Lemma 3.2. We have Pn P if and only if every subsequence {Pnk }
contains a subsequence {Pnkj } such that Pnkj P.
This is Theorem 2.3 from Billingsley [21].
Now we consider continuous mappings between metric spaces S1 and S2 .
A function h : S1 S2 is said to be measurable if
h1 (B(S2 )) B(S1 ).
Let h : S1 S2 be a measurable function. Then every probability measure P
on (S1 , B(S1 )) induces a probability measure Ph1 on (S2 , B(S2 )) dened by
(Ph1 )(A) = P h1 (A) = P(h1 (A)),
where A B(S2 ). This measure is uniquely determined. A function h : S1
S2 is continuous if for every open set G2 S2 the set h1 (G2 ) is open in S1 .
Continuous mappings transport the property of weak convergence.
Theorem 3.3. Let h : S1 S2 be a continuous function. If Pn P, then
also Pn h1 Ph1 .
This theorem is a particular case of Theorem 5.1 from Billingsley [21].
A family {Pn } of probability measures on (S, B(S)) is said to be relatively
compact if every sequence of elements of {Pn } contains a weakly convergent
subsequence. A family {Pn } is called tight if for arbitrary > 0 there exists
a compact set K such that P(K) > 1 for all P from {Pn }. Prokhorovs
theorem is a powerful tool in the theory of weak convergence of probability
measures; it is given below as Theorem 3.4, the direct half, and as Theorem 3.5,
the converse half. These theorems connect relative compactness with the tight-
ness of a family of probability measures.
52 3 Interlude: Results from Probability Theory

Theorem 3.4. If a family of probability measures is tight, then it is relatively

compact.

Theorem 3.5. Let S be separable (i.e., S contains a countable dense subset)

and complete. If a family of probability measures on (S, B(S)) is relatively
compact, then it is tight.

These are Theorems 6.1 and 6.2 from Billingsley [21]. Note that a topological
space is said to be separable if it contains a countable dense subset.
In the theory of Dirichlet series we investigate the weak convergence of
probability measures PT P, where T is a continuous parameter which
tends to innity. As it is noted in [21], we have PT P, as T , if and
only if PTn P, as n , for every sequence {Tn } with limn Tn = .
All theorems on weak convergence analoguous to those stated above remain
valid in the case of continuous parameters.

3.2 Random Elements

The theory of weak convergence of probability measures can be paraphrased
as the theory of convergence of random elements in distribution.
Let (, F, P) be a probability space, (S, B(S)) be a metric space with its
class of Borel sets B(S), and X : S a mapping. If

X 1 (A) = { : X() A} F

for every A B(S), then X is called an S-valued random element dened on

; if S = R we say that X is a random variable. The distribution of an S-
valued random element X is the probability measure PX on (S, B(S)), given
by
PX (A) = P(X 1 (A)) = P{ : X() A}
for arbitrary A B(S) (in the sequel we will often write P in place of PX ).
We say that a sequence {Xn } of random elements converges in distribution
to a random element X if the distributions Pn of the elements Xn converge
weakly to the distribution of the element X, and in this case we write
D
Xn X;
n

D
if Pn PX , then we also write Xn PX .
n
Let S be a metric space with metric , and let Xn , Yn be S-valued random
elements dened on (, F, P). If Xn and Yn have a common domain, it makes
sense to speak of the distance (Xn (), Yn ()) for . If S is separable,
then (Xn , Yn ) is a random variable. In this case, convergence in distribution
of two sequences of random elements Xn and Yn is related to the distribution
of (Xn , Yn ) (convergence in probability).
 3.2 Random Elements 53

Theorem 3.6. Let S be separable and, for n N, let Yn , X1n , X2n , . . . be

S-valued random elements, all dened on (, F, P). Suppose that
D D
Xkn Xk for each k, and Xk X.
n k
If for any  > 0
lim lim sup P{(Xkn , Yn ) } = 0,
k n
D
then Yn X.
n
This is Theorem 4.2 of Billingsley [21].
The mean (expectation value) EX of a random element X is dened by

EX = X() dP

if the integral exists in the sense of Lebesgue. A simple but fundamental
result on the deviation of a random variable from its expectation value is
Chebyshevs (respectively, Markovs) inequality:
Lemma 3.7. Let X be a real-valued random variable, h : R [0, ) be a
non-negative function, and a > 0. Then
1
P{ : h(X) a} Eh(X).
a
A proof can be found, for example, in Billingsley [22]. Taking h(x) = x2 , we
deduce the classical Chebyshevs inequality:
1
P{|X| a} EX 2 .
a2
We say that some property is valid almost surely if there exists a set A F
with P(A) = 0 such that this property is valid for every \ A. Random
variables X and Y are said to be orthogonal if EXY = 0. An important
result on almost sure convergence of series of orthogonal random variables is
the following
Theorem 3.8. Assume that the random variables X1 , X2 , . . . are orthogonal
and that



E|Xn |2 (log n)2 < .
n=1

Then the series n=1 Xn converges almost surely.

Further, we need a similar result for independent random variables.

Random variables X and Y are said to be independent if for all A, B B(S)
P{ : X A, Y B} = P{ : X A} P{ : Y B}.
Usually, independence of random variables is dened via the -elds generated
by the related events; however, for the sake of simplicity, we introduced this
notion by the equivalent condition above.
54 3 Interlude: Results from Probability Theory

Theorem 3.9. Assume that the random variables X1 , X2 , . . . are indepen-

dent. If the series





EXn and E(Xn EXn )2
n=1 n=1

converge, then the series n=1 Xn converges almost surely.

Proofs of the last two theorems can be found in Love [226].

Now we present a criterion on almost sure convergence of series of inde-
pendent random variables for Hilbert spaces. Let H be a separable Hilbert
space with norm and dene for an H-valued random element X and a real
number c the truncated function

(c) X if X c,
X =
0 if X > c.

Then

Theorem 3.10. Let X1 , X2 , . . . be independent H-valued random elements. If

there is a constant c > 0 so that the series







E Xn(c) EXn(c) 2 , EXn(c) , and P{ Xn > c}
n=1 n=1 n=1

converge, then the series n=1 Xn converges in H almost surely.

A proof can be found in Buldygin [45].

3.3 Denjoys Probabilistic Argument

for Riemanns Hypothesis
At this point our survey of probability theory will be cut in order to give a
heuristic probabilistic argument for the truth of Riemanns hypothesis. The
Mbius -function is dened by (1) = 1, (n) = 0 if n has a quadratic divisor
= 1, and (n) = (1)r if n is the product of r distinct primes. It is easily seen
that (n) is multiplicative and appears as coecients of the Dirichlet series
representation of the reciprocal of the zeta-function:
 1


(n)
1
(s) = 1 s = ,
p
p n=1
ns

valid for > 1. Riemanns hypothesis is equivalent to the estimate

M (x) := (n)  x(1/2)+
.
nx
 3.3 Denjoys Probabilistic Argument for Riemanns Hypothesis 55

This is related to (1.11); for a proof see, for example, Titchmarsh [353,
Sect. 14.25].
Denjoy [68] argued as follows. Assume that {Xn } is a sequence of random
variables with distribution
1
P(Xn = +1) = P(Xn = 1) = .
2
Dene


n
S0 = 0 and Sn = Xj ,
j=1

then {Sn } is a symmetrical random walk in Z2 with starting point at 0. A

simple application of Chebyshevs inequality yields, for any positive c,
1 1
P{|Sn | cn 2 } ,
2c2
which shows that large values for Sn are rare events. By the theorem of
MoivreLaplace ([22, Theorem 27.1]) this can be made more precise. It follows
that:    c  2
1 1 x
lim P |Sn | < cn 2 = exp dx.
n 2 c 2
Since the right-hand side above tends to 1 as c , we obtain
 1

lim P |Sn |  n 2 +
= 1
n

for every  > 0. We observe that this might be regarded as a model for the
value-distribution of Mbius -function. To say it with the words of Edwards:
Thus these probabilistic assumptions about the values of (n) lead to the
conclusion, ludicrous as it seems, that M (x) = O(x1/2+
) with probability
one and hence that the Riemann hypothesis is true with probability one!
(cf. [74]). The law of the iterated logarithm [22, Theorem 9.5] would even
gives the stronger estimate
 
lim P |Sn |  (n log log n)1/2 = 1,
n

which suggests for M (x) the upper bound (x log log x)1/2 . This estimate is
pretty close to the so-called weak Mertens hypothesis which states
 X 2
M (x)
dx  log X.
1 x
Note that this bound implies the Riemann hypothesis and the essential sim-
plicity hypothesis. On the contrary, Odlyzko and te Riele [284] disproved the
original Mertens hypothesis [248],
1
|M (x)| < x 2 ,
56 3 Interlude: Results from Probability Theory

by showing that
M (x) M (x)
lim inf < 1.009 and lim sup > 1.06; (3.1)
x x1/2 x x1/2
for more details see also the notes to Sect. 14 in Titchmarsh [353].

3.4 Characteristic Functions and Fourier Transforms

There is an intimate relationship between weak convergence and characteristic
functions which makes characteristic functions very useful in studying limit
distributions.
The characteristic function ( ) of a probability measure P on (Rr , B(Rr ))
is dened by 
( ) = exp(i, x)P( dx),
Rr
where , x stands for the inner product of and x Rr . Notice that the
characteristic function uniquely determines the measure it comes from. Let
{Pn } be a sequence of probability measures on (Rr , B(Rr )) and let {n ( )}
be the sequence of the corresponding characteristic functions. Suppose that
lim n ( ) = ( )
n

for all , and that ( ) is continuous at the point 0 = (0, . . . , 0). Then
Lvys famous continuity theorem (see [22, Sect. 26]) yields the existence of
a probability measure P on (Rr , B(Rr )) such that Pn P, and ( ) is
the characteristic function of P. However, later we shall deal with Fourier
transforms instead of characteristic functions; their theory is quite similar to
the theory of characteristic functions (for details we refer to the rst chapter
from [186]).
Let = {s C : |s| = 1} and denote by m the cartesian product of m
copies of . Further, let P be a probability measure on ( m , B( m )), then the
Fourier transform g(k1 , . . . , km ) of the measure P is dened by

g(k1 , . . . , km ) = xk11 . . . xkmm dP,
m

where kj Z and xj for 1 j m. Similarly to the characteristic

function, the measure P is uniquely determined by its Fourier transform.
The role of Lvys continuity theorem in the theory of Fourier transforms is
played by
Theorem 3.11. Let {Pn } be a sequence of probability measures on
( m , B( m )) and let {gn (k1 , . . . , km )} be the sequence of the corresponding
Fourier transforms. Suppose that for every vector (k1 , . . . , km ) Zm the limit
g(k1 , . . . , km ) = lim gn (k1 , . . . , km )
n
 3.5 Haar Measure and Characters 57

exists. Then there is a probability measure P on ( m , B( m )) such that Pn

P, and g(k1 , . . . , km ) is the Fourier transform of P.

Theorem 3.11 is a special case of a more general continuity theorem for

probability measures on compact abelian groups. A proof of this result can
be found in Heyer [133, Theorem 1.4.2].

3.5 Haar Measure and Characters

Let G be a set equipped with the structures of a group and of a topological

space. If the function h : G G G, dened by h(x, y) = xy 1 , is continuous,
then G is called a topological group. A topological group is said to be compact
if its topology is compact. In what follows, G is assumed to be compact.
A Borel measure P on a compact topological group G is said to be invariant
if
P(A) = P(xA) = P(Ax)
for all A B(G) and all x G, where xA and Ax denote the sets {xy : y A}
and {yx : y A}, respectively. An invariant Borel measure on a compact
topological group is called Haar measure.

Theorem 3.12. On every compact topological group there exists a unique

probability Haar measure.

The uniqueness follows from m(G) = 1. For the proof see Hewitt and Ross
[132, Chap. IV] or Theorem 5.14 in Rudin [313].
In the sequel, we denote the Haar measure associated with a compact topo-
logical group G simply by m; there will be no confusion about the underlying
group.
Now assume further that G is a commutative group. A continuous homo-
morphism : G C is called a character of G. The character of G which
is identically 1 is called trivial or principal, and we denote it by 0 ; other
characters are said to be non-trivial or non-principal. The characters build up
a group G, the character group. The Fourier transform of a function f dened
on G is given by 
f() = (g)f (g) m( dg),
G

where is a character of G. Then f is a continuous map dened on G. The

orthogonality relation for characters states
 
1 if = 0 ,
(g) m( dg) = (3.2)
G 0 if = 0 .

This generalizes the concept of Dirichlet characters.

58 3 Interlude: Results from Probability Theory

3.6 Random Processes and Ergodic Theory

To be able to identify later the explicit form of the limit measure in limit
theorems, we recall some facts from ergodic theory.
Let (, F, P) be a probability space and let T denote a parameter set. A
nite real function X(, ) with T and is said to be a random (or
stochastic) process if  X(, ) is a random variable for each xed T .
For xed , the function  X(, ) is called a sample path of the ran-
dom process. Let 1 , . . . , n be an arbitrary set of values of T . Then the family
of all common distributions of random variables X(1 , ), . . . , X(n , ), i.e.,
P{X(1 , ) < x1 , . . . , X(n , ) < xn }
for all n N and all possible values of j with 1 j n, is called a
family of nite-dimensional distributions of the process X(, ). Part of the
structure of the random process is specied by its nite-dimensional distri-
butions. However, they do not determine the character of the sample paths
(see [22, Sect. 23], for a nice example). Kolmogorovs existence theorem (see
Theorem 36.1 in [22]) states that if a family of nite dimensional distributions
satises certain consistency conditions, then there exists on some probability
space a random process having exactly the same nite-dimensional distribu-
tions. For instance, a special application of Kolmogorovs existence theorem
yields a model for Brownian motion with continuous paths.
Let Y be the space of all nite real-valued functions y( ) with R.
In this case it is known that the family of nite-dimensional distributions of
each random process determines a probability measure P on (Y, B(Y )). Then,
on the probability space (Y, B(Y ), P), we dene for real u the translation gu
which maps each function y( ) Y to y( + u). It is easily seen that the
translations gu form a group. A random process X(, ) is said to be strongly
stationary if all its nite-dimensional distributions are invariant under the
translations by u. It is known that if a process X(, ) is strongly stationary,
then the translation gu is measure preserving, i.e., for any set A B(Y ) and
all u R the equality
P(A) = P(Au ), where Au := gu (A)
holds. A set A B(Y ) is called an invariant set of the process X(, ) if for
each u the sets A and Au dier from each other by a set of zero P-measure.
In other words, P(AAu ) = 0, where denotes the symmetric dierence of
two sets A and B:
AB := (A \ B) (B \ A).
It is easy to see that all invariant sets of Y form a -eld which is a sub--eld
of B(Y ). We say that a strongly stationary process X(, ) is ergodic if its
-eld of invariant sets consists only of sets having P-measure equal to 0 or 1.
For ergodic processes the BirkhoKhintchine theorem gives an expression
for the expectation of X(0, ) in terms of an integral taken over the sample
paths.
 3.7 The Space of Analytic Functions 59

Theorem 3.13. Let X(, ) be an ergodic process with E|X(, )| < and
almost surely Riemann-integrable sample paths over every nite interval. Then
 T
1
lim X(, ) d = EX(0, )
T T 0

almost surely.

A proof of this theorem can be found in Cramr and Leadbetter [64].

3.7 The Space of Analytic Functions

Let G be a simply connected region in the complex plane. We denote by H(G)

the space of analytic functions f dened on G equipped with the topology of
uniform convergence on compacta.
In order to introduce an appropriate metric on H(G) we note

Lemma 3.14. For any open set G in the complex plane there exists a sequence
of compact subsets Kj of G with the properties:
Kj Kj+1 for any j N;
If K is compact and K G, then K Kj for some j N;
such that


G= Kj
j=1

The proof is straightforward and can be found in Conways book [62,

Sect. VII.1].
Now, for f, g H(G) let

j (f, g) = max |f (s) g(s)|

sKj

and put


j (f, g)
(f, g) = 2j .
j=1
1 + j (f, g)

This denes a metric on H(G) which induces the desired topology; of course,
 depends on the family {Kj }. Note that the series above is dom-
the metric
inated by j 2j and therefore convergent.

Theorem 3.15. Let G be a simply connected region in the complex plane.

Then H(G) is a complete separable metric space.

In Conway [62, Sect. VII.1], it is shown that H(G) is a complete metric space;
the separability, i.e., the existence of a countable dense subset, follows from
60 3 Interlude: Results from Probability Theory

Runges approximation theorem (see [312, Sect. 13]) which states that the set
of polynomials is dense in H(G).
In our later studies, we deal with the supports of H(G)-valued random
elements. Let S be a separable metric space and let P be a probability
measure on (S, B(S)). The minimal closed set SP S with P(SP ) = 1 is
called the support of P. Note that SP consists of all x S such that for
every neighbourhood U of x the inequality P(U ) > 0 is satised. Let X be a
S-valued random element dened on the probability space (, F, P). Then
the support of the distribution P(X A) for A B(S) is called the support
of the random element X. We denote the support of X by SX .

Theorem 3.16. Let {Xn } be a sequenceof independent H(G)-valued random

elements, and suppose that the series n=1 Xn converges almost everywhere.
Then the support of the sum of this series is the closure of the set of all
f H(G) which may be written as a convergent series



f= fn , where fn SXn .
n=1

This is Theorem 1.7.10 of Laurinikas [186]. The proof follows the lines of an
analogous statement for independent real variables due to Lukacs [231]. We
only sketch the main ideas. Suppose that the random elements Xn are dened
on a probability space (, F, P ). Put



X := Xn = LN + RN ,
n=1

where


N


LN := Xn and RN := Xn .
n=1 n=N +1

Since the series n=1 Xn converges almost surely, we have

lim P { : (RN , 0) } = 0.
N

Let
PN (A) = P {LN A} and P(A) = P {X A}
for A B(H(G)). It follows that Pn P, which implies

SP lim SPN , (3.3)

where lim SPN denotes the set of all f H(G) such that any neighbourhood
of f contains at least one g which belongs to SPn for almost all n N.
To show the converse inclusion, put

QN (A) = P {RN A}
 3.7 The Space of Analytic Functions 61

for A B(H(G). The distribution of X = LN +RN is given by the convolution

PN QN , dened by

(PN QN )(A) = PN (A g)QN ( dg).
H(G)

The support of X = LN + RN is the closure of the set

{f H(G) : f = f1 + f2 , where f1 SLN , f2 SRN }. (3.4)

For g lim SPN let

A
:= {f H(G) : (f, g) < }.

It follows that PN (A
) = P (LN A
) > 0 and QN (A
) > 0 for N large
enough. This leads to

P(A2
) PN (A
)QN (A
) > 0.

This implies
SX = SP lim SPN .
By (3.3) it follows that the latter inclusion is an equality. In view of (3.4) the
support of LN is the set of all g H(G) which have a representation

N
g= fn ,
n=1

where fn SXn . From the denition of lim SPN we deduce that for any
f SX there exists a sequence of gN SLN which converges to f . This yields
the assertion of the theorem.
4
Limit Theorems

Measure what is measurable, and make measurable what is not so.

Galileo Galilei

In this chapter, we prove a limit theorem dealing with weakly convergent

probability measures for L-functions from the class S in the space of analytic
functions. Throughout this chapter, we assume that L S. We remark that we
will not make use of axiom (v), so the results hold in a more general context;
however, with respect to later applications, there is no need to introduce a
further class. We follow the presentation of Laurinikas [187, 188] (functions
in S form a subclass of Matsumoto zeta-functions considered herein). Besides,
we refer the interested reader to Laurinikas survey [185] and his monograph
[186].

4.1 Associated Random Elements and the Main Limit

Theorem
Assume that L Sand denote by D the intersection of the corresponding
mean-square half-plane > m with the critical strip (see Sect. 2.1):

D := {s C : m < < 1}. (4.1)

In view of Theorem 2.4 we know that D is not empty. Together with (2.4) we
have  
1 1 1 L
m max ,1 < 1, (4.2)
2 2 1 + 2L
where L and L are dened by axioms (ii) and (iii). In particular, as T ,
 T
1
|L( + it)|2 dt  1 for (m , 1). (4.3)
T 0
64 4 Limit Theorems

Now we attach to L(s) the probability measure PT by setting

1
PT (A) = meas { [0, T ] : L(s + i ) A} (4.4)
T
for A B(H(D)), where H(D) denotes the space of analytic functions on the
strip D (as introduced in Sect. 3.7). We shall show that this measure converges
weakly to some probability measure as T tends to innity. For later purposes
it is also important to identify the limit measure.
Denote by the compact unit circle {s C : |s| = 1} and put

= p ,
p

where p = for each prime p. With product topology and pointwise

multiplication this innite-dimensional torus is a compact topological
abelian group; the compactness of follows from Tikhonovs theorem (see
[169, Theorem 5.13]). Thus, by Theorem 3.12, the normalized Haar measure
m on (, B()) exists. This gives a probability space (, B(), m). Let (p)
stand for the projection of to the coordinate space p . The Haar
measure m on is the product of the Haar measures mp on the coordinate
spaces p , i.e., 
m{ : A} = mp { : (p) Ap },
p

where Ap is the projection of A B() onto p . Thus {(p) : p prime}

is a sequence of independent complex-valued random variables dened on
the probability space (, B(), m). Denote by G the set of all completely
multiplicative functions g : N . Further, let (1) = 1 and

(n) = (p)(n;p) , (4.5)
p

where (n; p) is the exponent of the prime p in the prime factorization of n.

This extends the function to the set of positive integers as a unimodular,
completely multiplicative function. On the other hand, the restriction of any
function g G to the set of primes is an element of . Consequently, we may
identify the elements of with the functions from G.
Next we are going to attach to


a(n)
L(s) =
n=1
ns

a random element. For s D and , we dene

a(n)(n)
L(s, ) = . (4.6)
n=1
ns

This is equal to L(s) for > 1 and (n) 1.

 4.1 Associated Random Elements and the Main Limit Theorem 65

Lemma 4.1. The function L(s, ) is an H(D)-valued random element on the

probability space (, B(), m).
Proof. Fix a real number 1 > m . Dene for n N
a(n)(n)
n = n () = ,
n1
then {n } is a sequence of complex-valued random variables on (, B(), m).
It is easy to compute that

a(m)a(n) |a(n)|2
Em n = if m = n,
(mn)1
(m)(n) dm = n21
0 otherwise.
This shows that the sequence {n } is pairwise orthogonal. Since m 12 by
(2.4), it follows that 21 > 1. This and the Ramanujan hypothesis (i) imply
that the series



|a(n)|2
E|n |2 (log n)2 = 21
(log n)2
n=1 n=1
n

is convergent. It follows from Theorem 3.8 that the series n=1 n converges
almost surely, respectively


a(n)(n)
n=1
n1
converges for almost all with respect to m. Since Dirichlet series
converge in half-planes, and uniformly in compact subsets of their half-plane
of convergence (see Sect. 2.1), it follows that, for almost all , the series
(4.6) converges uniformly on compact subsets of the half-plane > 1 . For
k N let Ak be the subset of for which the series (4.6) converges
uniformly on compact subsets of the half-plane m + k1 . Then m(Ak ) = 1
for all k N. Setting

A= Ak ,
k=1
we get m(A) = 1, and for A the series (4.6) converges uniformly on
compact subsets of D. Since each term of the series in question is an H(D)-
valued function, the assertion of the lemma follows. 

Taking into account the multiplicativity of the coecients a(n), (2.6), and
the decomposition (4.5) of , we may rewrite the series (4.6) (formally) as
 m  1
j (p)(p))
L(s, ) = 1
p j=1
ps



a(pk )(p)k
= 1+ (4.7)
p
pks
k=1

In the half-plane > 1 both the series (4.6) and the product (4.7) converge
absolutely.
66 4 Limit Theorems

Lemma 4.2. For almost all the product (4.7) converges uniformly on
compact subsets of D.

Proof. Dene for each prime p the function

a(pk )(p)k
xp = xp (s, ) = . (4.8)
pks
k=1

Then 
L(s, ) = (1 + xp (s, )).
p

By the uniform convergence of the xp -dening series on D the function xp (s, )

is an H(D)-valued random element for any p. Using the Ramanujan hypothesis
(i) we obtain, in a similar way as in the proof of Lemma 4.1, that


|xp (s, )|2
p

converges uniformly on compact subsets of D. To prove the almost sure con-

vergence of the product (4.7), it remains to show that the series


xp (s, )
p

converges almost surely. For this, we isolate the leading term in the series
expansion of xp . Put

a(p)(p)
yp (s) = yp (s, ) = ,
ps

then, by the Ramanujan hypothesis (i),

|a(pk )|
|xp (s, ) yp (s, )|  p
k  p
2 .
pk
k=2 k=2

Thus, for suciently small , the series

|xp (s, ) yp (s, )|
p

converges uniformly on compact

 subsets of D for all . Therefore, we
can focus on the series p y p (s, ). Clearly, the yp (s, ) dene a sequence
of independent H(D)-valued random elements. It follows from the denition
of the projections (p) that Eyp (s) = 0. Furthermore, by the Ramanujan
hypothesis (i),
|a(p)|2 1
E|yp (s)|2 =  2
.
p2 p
 4.2 Limit Theorems for Dirichlet Polynomials 67

This implies the convergence of

E|yp (s)|2 (log p)2
p

for all s D. Thus, by Theorem 3.9 the series p yp (s, ) converges almost
surely for each xed s D. It follows that also p xp (s, ) converges almost
surely for xed s D. In view of convergence properties of Dirichlet series
the convergence is uniform on compact subsets of D for almost all .
It is clear that the same assertion holds for the product too. This proves the
lemma. 

By analytic continuation the identity between the series (4.6) and the
product (4.7) holds whenever one of these representations converges uniformly.
Now let P denote the distribution of the random element L(s, ), i.e.,
P(A) = m{ : L(s, ) A} (4.9)
for A B(H(D)). The main result of this chapter is the limit theorem
Theorem 4.3. Let L S. The probability measure PT , dened by (4.4),
converges weakly to P as T .
The rst limit theorem of this type is due to Bagchi [9]. In the case of the
Riemann zeta-function this result might be regarded as an analogue of
the limit Theorem 1.4 of Bohr and Jessen on function spaces. To see this, note
that Laurinikas [183] translated their statement into a modern language by
dening a probability measure via
1
PT (A) = meas {t [0, T ] : ( + it) A} A B(C),
T
and proving that, for any xed > 12 , PT converges to a certain probability
measure as T . Meanwhile, functional limit theorems became a useful
tool in the value-distribution theory of Dirichlet series, not only with respect
to universality.

4.2 Limit Theorems for Dirichlet Polynomials

The rst step in the proof of Theorem 4.3 is to establish an analogous result
for Dirichlet polynomials. For N N let

N
(n)
fN (s) = ,
n=1
ns

where (n) C for 1 n N . Further, put

N
(n)g(n)
fN (s, g) = ,
n=1
ns
68 4 Limit Theorems

where g G is a unimodular, completely multiplicative arithmetic function;

in particular, fN (s, 1) = fN (s), where 1 : N C is the arithmetic func-
tion constant 1 (i.e., the principal character 0 mod 1). Later, we will put
(n) = a(n), where a(n) is the nth Dirichlet coecient of L S. We dene a
probability measure PgT,N (A), corresponding to PT , by setting
1
PgT,N (A) = meas { [0, T ] : fN (s + i, g) A} (4.10)
T
for A B(H(D)). First, we show that the measure PgT,N converges weakly to
some measure as T tends to innity.
Lemma 4.4. Let g G. Then there exists a probability measure PgN on
(H(D), B(H(D))) such that PgT,N converges weakly to PgN as T .
Proof. Let p1 , . . . , pr be the distinct prime divisors of the product

N
n. (4.11)
n=1
(n)=0

Moreover, set

r
r = pj .
j=1

Finally, dene the function hg : r H(D) by

N
(n)g(n)  (n;pj )
r
hg (x1 , . . . , xr ) = xj .
n=1
ns j=1

Obviously, hg is continuous on r . Since (n; p) is the exponent in the prime

factorization of n,
r
(n;pj )
n= pj ,
j=1

hg satises the identity

N
(n)g(n)  i (n;pj )
r
fN (s + i, g) = s
p = hg (pi i
1 , . . . , pr ). (4.12)
n=1
n j=1

Now we dene a probability measure HT on (r , B(r )) by

1
HT (A) = meas { [0, T ] : (pi
1 , . . . , pr ) A}
i
T
for A B(r ). For n1 , . . . , nr N, the Fourier transform hT of HT is given by
 
1 T  i nj
r
hT (n1 , . . . , nr ) = xn1 1 . . . xnr r dHT = p d.
r T 0 j=1 j
 4.2 Limit Theorems for Dirichlet Polynomials 69

Hence, hT (0, . . . , 0) = 1, and



r
exp iT j=1 nj log pj 1
hT (n1 , . . . , nr ) = r
iT j=1 nj log pj

if at least one nj = 0. Since the logarithm of the prime numbers are linearly
independent, we nd in the latter case

lim hT (n1 , . . . , nr ) = 0.
T

Thus

1 if (n1 , . . . , nr ) = (0, . . . , 0),
lim hT (n1 , . . . , nr ) = (4.13)
T 0 otherwise.

By Theorem 3.11 the measure HT converges weakly to a probability measure

on (r , B(r )) as T tends to innity; since this measure is uniquely deter-
mined by its Fourier transform (4.13), it has to be the Haar measure m on
(r , B(r )). In view of the continuity of the function hg , formula (4.12), and
Theorem 3.3, we deduce that the probability measure PgT,N converges weakly
to
m h1
g = mhg
1

as T tends to innity. This proves the lemma. 



Now we shall show that the limiting measure PgN = mh1

g is independent
of g.

Lemma 4.5. Let g be a unimodular completely multiplicative function. Then

both probability measures PgT,N and P1T,N converge weakly to the same measure
as T .

Proof. Let p1 , . . . , pr be, as in the last proof, the prime divisors of (4.11). We
dene the function h : r r by

h (x1 , . . . , xr ) = (x1 exp(i1 ), . . . , xr exp(ir )),

where j := arg g(pj ) for 1 j r. It follows from Lemma 4.1 that the
probability measures PgT,N and P1T,N converge weakly to the measures mh1
g
and mh1
1 , respectively. A simple computation shows

N
(n) 
r
(n;pj )
h1 (x1 , . . . , xr ) = s
exp(i(n; pj )j )xj
n=1
n j=1

N
(n)g(n)  (n;pj )
r
= x
n=1
ns j=1
= hg (h (x1 , . . . , xr )).
70 4 Limit Theorems

It follows that:
mh1 1
1 = m(hg (h )) = m(h )1 h1
g .

Since the Haar measure m is invariant with respect to translation of points in

, we deduce m(h )1 = m. In addition with the last but one identity we get
mh1 1
1 = mhg . The lemma is proved. 


4.3 An Ergodic Process

A one-parameter family { } of transformations on a compact topological

abelian group G is called a one-parameter group of transformations if

1 +2 (g) = 1 (2 (g)) and (g) = 1

(g)

for all real numbers , 1 and 2 and all g G. We consider one-parameter

groups of measurable transformations on . Dene

a : R ,  a = {pi : p prime}.

This map is continuous, and the image of every neighbourhood of the neutral
element of R is a neighbourhood of the neutral element in . Clearly, the
family {a : R} is a one-parameter group. Moreover, we dene the one-
parameter family { : R} of transformations on by setting

() = a (4.14)

for . Then { : R} denes a one-parameter group of measurable

and measure preserving transformations on .
Now we transport the notions from ergodic theory of random processes to
one-parameter groups. Let { : R} be an arbitrary one-parameter group
of measurable transformations on G. A set A B(G) is invariant with respect
to the group { : R} if for each the sets A and A := (A) dier one
from another by a set of zero m-measure, i.e.,

m(AA ) = 0,

where m is the Haar measure on G. All invariant sets form a sub--eld of

B(G). A one-parameter group is said to be ergodic if its -eld of invariant
sets consists only of sets having Haar measure equal to 0 or 1.

Lemma 4.6. The one-parameter group { : R} is ergodic.

Proof. For a positive rational number = ab with a, b coprime and we

put
(a)
() = .
(b)
 4.3 An Ergodic Process 71

This extends the function from N to the set of positive rational numbers.
Now dene : by () = (). Then, for any xed positive rational
number , is a character of . On the other side, since the dual group of
has a representation as a direct sum
$
Zp , where Zp = Z for any prime p
p

(see [132, Chap. 6]), for any character of there exists a positive rational
such that = .
Now suppose that A B() satises m(AA ) = 0 for any real . We
have to show that either m(A) = 0 or m(A) = 1. Denote by 1(; A) the
indicator function of A, i.e.,

1 if A,
1(; A) =
0 otherwise.
Since the set A is invariant, it follows that for arbitrary R we have
1(a ; A) = 1(; A) for almost all (4.15)
with respect to the Haar measure m.
First, assume that is non-trivial. By the same reasoning as above, there
exists a positive rational number such that = . Consequently,
(a ) = i ,
and so we can nd a 0 for which (a0 ) = 1. In view of (4.15) this implies
for the Fourier transform of the indicator function
 
1(; A) = ()1(; A) m( d) = (a0 ) ()1(; A) m( d)

= (a0 )1(; A).
Since we have chosen 0 such that (a0 ) = 1, it follows that 1(; A) = 0.
Obviously, this argument holds for any non-trivial character of .
Now suppose that 0 is the trivial character of , and let us assume that
1(0 ; A) = u. By the orthogonality relation for characters (3.2) we nd
 
u if = 0 ,
u() = u () m( d) =
0 otherwise.

Since 1(; A) = 0 for = 0 , we obtain, for any character ,

1(; A) = u().
The function 1(; A) is uniquely determined by its Fourier transform. Hence,
it follows that 1(; A) = u for almost all . Since 1(; A) is the indicator
function of A, we get u = 0 or u = 1, i.e., either 1(; A) = 0 for almost all ,
or 1(; A) = 1 for almost all . This implies that m(A) = 0 or m(A) = 1,
which proves the Lemma. 

72 4 Limit Theorems

Lemma 4.7. Let (m , 1). Then, for almost all ,

 T
|L( + it, )|2 dt  T.
0

Recall that m is the abscissa of the mean-square half-plane for L.

Proof. We start as in the proof of Lemma 4.1. For n N let

a(n)(n)
n (, ) = .
n
The n (, ) dene a sequence of pairwise orthogonal random variables with
second moment
|a(n)|2
E|n (, )|2 = .
n2
Thus the random variable
 2


 
(, ) :=  n (, ) = |L(, )|2
 
n=1

has the nite expectation

|a(n)|2
E(, ) = E|n (, )|2 = ; (4.16)
n=1 n=1
n2

the convergence follows from the Ramanujan hypothesis (i) (since m 1

2
and so 2 > 1). It is obvious that

(, ()) = |L(, a )|2 = |L( + it, )|2 (4.17)

for any transformation (4.14). Since the Haar measure m is invariant, the
equality m( (A)) = m(A) is valid for every A B() and every R.
Hence, |L( + i, )|2 is a strongly stationary process. We shall show that it
is even ergodic.
Let Q be the probability measure on (, B()) which is determined by
the random process |L( +i, )|2 . Further, let A be an invariant set of |L( +
i, )|2 , i.e.,
Q(AAu ) = 0,
where Au = u (A) with transformations u dened by (4.14). Now put

A = { : |L( + i, )|2 A}

and
Au = { : |L( + i, )|2 Au }.
 4.4 Approximation in the Mean 73

It follows that:
A = { : |L(, a )|2 A}
and

Au = { : |L( + i + iu, )|2 A}

= { : |L( + i, au )|2 A}.

Hence Au = u (A ). Since (AAu ) = A Au ,

m(A Au ) = m((AAu ) ) = Q(AAu ) = 0,

which shows that A is an invariant set with respect to { : R}. However,

by Lemma 4.6 the group { : R} is ergodic. Therefore, m(A ) = 0 or
m(A ) = 1. Hence it follows that Q(A) = 0 or Q(A) = 1, i.e., the process
|L( + i, )|2 is ergodic.
Clearly, (, ()) 0. Hence, from (4.16), (4.17) and the Birkho
Khintchine Theorem 3.13, we get
 T 
1 1 T
lim |L( + i, )| d = lim
2
(, ()) d
T T 0 T T 0

= E(, ) <

for almost all . This gives the assertion of the lemma. 



4.4 Approximation in the Mean

Let 1 > m , and dene for N N and 1 1

 
s s
N (s) = N s.
1 1

Further, for > m , put

 1 +i
1 dz
LN (s) = L(s + z)N (z) . (4.18)
2i 1 i z

Taking into account Stirlings formula (2.17) and axiom (iii), the existence of
the integral follows. Now we shall show that LN (s) approximates L(s) in an
appropriate mean.

Lemma 4.8. Let K be a compact subset of D. Then

 T
1
lim lim sup max |L(s + i ) LN (s + i )| d = 0.
N T T 0 sK
74 4 Limit Theorems

This result should be compared with formula (1.19) in the proof of Voronins
universality Theorem 1.7.
Proof. Let Re z = 1 . Since + 1 > 1 for s K, the function L(s + z) has
a representation as an absolutely convergent Dirichlet series:


a(n)
L(s + z) = s+z
.
n=1
n

Next we twist L(s) by an arithmetic function bN . For n, N N let

 1 +i
1 N (z) dz
bN (n) = .
2i 1 i nz z

By Stirlings formula (2.17),


1 1
bN (n)  |N (1 + it)| dt  .
n1 n1
Consequently, the series


a(n)bN (n)
,
n=1
ns
converges absolutely for > 1 1 . Hence,
 1 +i
 1 +i
1 dz a(n) 1 N (z) dz
L(s + z)N (z) = s 2i
2i 1 i z n=1
n 1 i nz z


a(n)bN (n)
= ;
n=1
ns

interchanging integration and summation is allowed by absolute convergence.

Thus (4.18) yields


a(n)bN (n)
LN (s) = .
n=1
ns
Using Mellins inversion formula (2.15), we obtain
 1 +i    z n
1

1 z N
bN (n) = dz = exp
2i1 1 i 1 n N
Thus we get


a(n) n
1

LN (s) = exp , (4.19)

n=1
ns N

where the series converges absolutely for > m . Now we move the path of
integration in (4.18) to the left. The integrand has a simple pole at z = 0, and
 4.4 Approximation in the Mean 75

a pole of order f at z = 1 s if L(s) has a pole at s = 1 of the same order.

Since K is compact in D, we may choose  > 0 such that [m + , 1 ]
for any s K. Putting 2 = m + 2
, we get by the calculus of residues
 2 +i
1 dz
LN (s) = L(s + z)N (z) (4.20)
2i 2 i z
 
N (z)
+ {Resz=0 + Resz=1s } L(s + z) .
z
On behalf of the functional equation for , for the residue at z = 0 we nd
 
N (z)
Resz=0 L(s + z) = lim L(s + z)N (z) = L(s).
z z0

If L(s) has a pole of order f at s = 1, then

   
N (z) N (w s)
Resz=1s L(s + z) = Resw=1 L(w)
z ws
 f 1 
1 d N (w s) 
= (w 1)f L(w)  = (s),
(f 1)! dw w s w=1
say.
Let C be a Jordan curve (i.e., a simple closed contour) in D of length |C|,
enclosing the set K, and denote by the distance from C to K. By Cauchys
formula,

1
max |L(s + i ) LN (s + i )| |L(z + i ) LN (z + i )| | dz|.
sK 2 C
Thus, for suciently large T ,

1 T
max |L(s + i ) LN (s + i )| d
T 0 sK
  T +Im z
1
 |L(Re z + i ) LN (Re z + i )| d | dz|
T C Im z
 2T
|C|
 max |L( + it) LN ( + it)| dt. (4.21)
T ; sC 0

We may choose the contour C such that for s C the inequalities m + 3
4

and 4
hold. In view of (4.20)

L( + it) LN ( + it)  |L(2 + it + i )||N (2 + i )| d

+ |(1 it)|.
76 4 Limit Theorems

Thus, for the same , we nd

 2T
1
|L( + it) LN ( + it)| dt
T 0
  | |+2T
1
 |N (2 + i )| |L(2 + it)| dt d
T | |
 2T
1
+ |(1 it)| dt.
T 0

By the CauchySchwarz inequality and the existence of the mean-square for

> m ,
  1/2
T T
|L( + it)| dt  T 1/2 |L( + it)|2 dt  T.
0 0

This leads in the last but one estimate to


1 2T
|L( + it) LN ( + it)| dt
T 0
  
| |
 max |N (2 + i )| 1 + d (4.22)
;sC T
 2T
1
+ max |(1 it)| dt.
T ;sK 0

Let I = [m 1 + 3
2 , 4 ]. Then, by Stirlings formula (2.17), the latter

expression is bounded by

max |N ( + i )|(1 + | |) d + o(1),
I

as T . By the denition of N (s), we obtain


lim max |N ( + i )|(1 + | |) d = 0.
N I

This in addition with (4.21) and (4.22) proves the lemma. 



4.5 A Limit Theorem for Absolutely Convergent Series

We dene for , s D and N N

a(n)(n) n
1

LN (s, ) = exp .
n=1
ns N
 4.5 A Limit Theorem for Absolutely Convergent Series 77

Due to the absolute convergence of the series LN (s), given by (4.19), LN (s, )
also converges absolutely for > m . Further, we dene the probability
measures PT,N and QT,N by setting
1
PT,N (A) = meas { [0, T ] : LN (s + i ) A}
T
and
1
QT,N (A) = meas { [0, T ] : LN (s + i, ) A}
T
for A B(H(D)). Now we prove that both measures converge to the same
limit as T tends to innity.
Lemma 4.9. There exists a probability measure PN on (H(D), B(H(D)))
such that the measures PT,N and QT,N both converge weakly to PN as T .
Proof. For , s D and M, N N let

M
a(n) n
1

LN,M (s) = exp

n=1
ns N

and

M
a(n)(n) n
1

LN,M (s, ) = exp .

n=1
ns N
Dene the probability measures PT,N,M and QT,N,M by putting
1
PT,N,M (A) = meas { [0, T ] : LN,M (s + i ) A}
T
and
1
QT,N,M (A) = meas { [0, T ] : LN,M (s + i, ) A}
T
for A B(H(D)). By Lemma 4.5 both measures converge weakly to the same
probability measure, say PN,M , as T tends to innity.
First, we show that the family {PN,M } is tight for xed N . Let be a
random variable uniformly distributed on [0, 1], dened on some probability
space (R, B(R), P ). Let
XT,N,M (s) = LN,M (s + iT ). (4.23)
Then, by Lemma 4.4,
D
XT,N,M XN,M , (4.24)
T

where XN,M = XN,M (s) is an H(D)-valued random element with distribu-

tion PN,M . Let {Kj } be a sequence of compact subsets of D satisfying the
conditions of Lemma 3.14 with G = D, i.e.,
Kj Kj+1 for any j N;
If K is compact and K D, then K Kj for some j N;
78 4 Limit Theorems

and


D= Kj .
j=1

Further, for j N let Mj be a positive parameter, which will be dened later.

By Chebyshevs inequality, Lemma 3.7,
   T
1
P max |XT,N,M (s)| > Mj max |LN,M (s + i )| d < .
sKj Mj T 0 sKj
Hence,
 

lim sup P max |XT,N,M (s)| > Mj
T sKj
 T
1 1
lim sup max |LN,M (s + i )| d.
Mj T T 0 sKj

Since the series for LN (s) converges absolutely in D,


1 T
sup lim sup max |LN,M (s + i )| d < Rj
N N T T 0 sKj

with some positive constant Rj depending on Kj . Setting

2j Rj
Mj = , (4.25)

we obtain
 

lim sup P max |XT,N,M (s)| > Mj < . (4.26)
T sKj 2j

By Theorem 3.3 and (4.24) it follows that

D
max |XT,N,M (s)| max |XN,M (s)|.
sKj T sKj

Thus, we obtain from (4.26)

 

P max |XN,M (s)| > Mj j . (4.27)
sKj 2

Now, for j N, dene a function kj : H(D) R by

kj (f ) = max |f (s)|.
sKj

The function kj is continuous. Furthermore, let

H
= {f H(D) : kj (f ) Mj for j N} .
 4.5 A Limit Theorem for Absolutely Convergent Series 79

A family of analytic functions is said to be compact on a region G if every

sequence of this family contains a subsequence which converges uniformly on
every compact subset of G. The theorem of Montel (see [62, Sect. VII.2] or
[313]) asserts that if a family of analytic functions on G is uniformly bounded
on every compact subset of G, then it is compact. Clearly, the family of func-
tions H
is uniformly bounded on every compact subset K D, and thus it
follows that H
is compact. Using (4.27), we obtain


 

P (XN,M (s) H
) 1 P max |XN,M (s)| > Mj
sKj
j=1


1
1 = 1 .
j=1
2j

Since PN,M is the distribution of XN,M , the inequality

PN,M (H
) 1 

holds for all M, N N. Therefore the family of probability measures {PN,M }

is tight with respect to M , and by Prokhorovs Theorem 3.4 it follows that it
is relatively compact.
It is clear that, for > m ,

lim LN,M (s) = LN (s),

M

and since the series dening LN (s) converges absolutely, the convergence is
uniform in any half-plane m + . Thus, for every  > 0,
1
lim lim sup meas { [0, T ] : (LN,M (s + i ), LN (s + i )) }
M T T
 T
1
lim lim sup (LN,M (s + i ), LN (s + i )) d = 0;
M T T 0

the denition of the metric  in the space of analytic function was given in
Sect. 3.7. Now let, similarly to (4.23),

YT,N (s) = LN (s + iT ).

Then the latter inequality can be rewritten as

lim lim sup P ((XT,N,M (s), YT,N (s)) ) = 0.

M T

Let {PN,Mk } be a subsequence {PN,M } which converges weakly to some mea-

sure PN . Then,
D
XN,Mk PN .
Mk
80 4 Limit Theorems

Since H(D) is separable (see Theorem 3.15), the conditions of Theorem 3.6
are fullled. Thus, we get by (4.24)
D
YT,N PN .
T

Hence, there exists a measure PN such that PT,N converges weakly to PN as
T tends to innity, and this limit is independent of the subsequence {PN,Mk }.
Since the family {PN,M } is relatively compact, by Theorem 3.2, we obtain
that PN,M converges weakly to PN as M tends to innity:
D
XN,M PN .
M

With the same argument applied to the random elements LN,M (s + iT, )
and LN (s + iT, ) we nd that QT,N converges weakly to PN as T tends to
innity. The assertion of the lemma follows. 


Let  be a subset of such that for  the series (4.6) for L(s, )
converges uniformly on compact subsets of D, and that for > m the esti-
mate  T
|L( + it, )|2 dt  T.
0

holds. It follows from Lemmas 4.1 and 4.7 that m(  ) = 1. By the same
argument as in the proof for Lemma 4.8 one can show

Lemma 4.10. Let K be a compact subset of D. Then, for  ,


1 T
lim lim sup max |L(s + i, ) LN (s + i, )| d = 0.
N T T 0 sK

4.6 Proof of the Main Limit Theorem

Dene a probability measure QT by setting
1
QT (A) = meas { [0, T ] : L(s + i, ) A}
T
for A B(H(D)).

Lemma 4.11. There exists a probability measure P on (H(D), B(H(D)))

such that the measures PT (given by (4.4)) and QT both converge weakly
to P as T .

Proof. Lemma 4.9 asserts that the measures PT,N and QT,N both converge
weakly with T to the same measure PN .
First, we prove that the family {PN } is tight. We follow the argument
given in the proof of Lemma 4.9 and use the notations introduced there.
 4.6 Proof of the Main Limit Theorem 81

Again, let be a random variable uniformly distributed on [0, 1], dened on

some probability space (R, B(R), P ). Dene

XT,N (s) = LN (s + iT ).

Then, by Lemma 4.4,

D
XT,N XN ,
T

where XN is an H(D)-valued random element with distribution PN . Now

let {Kj } be a sequence of compact subsets of D satisfying the conditions of
Lemma 3.14 with G = D. Since the series for LN (s) converges absolutely for
> m ,

1 T
sup lim sup max |LN (s + i )| d Rj < .
N N T T 0 sKj

By Chebyshevs inequality, Lemma 3.7,

   T
1
P max |XT,N (s)| > Mj max |LN (s + i )| d.
sKj Mj T 0 sKj

Thus,  

lim sup P max |XT,N (s)| > Mj < ,
T sKj 2j
where Mj and Rj are related by (4.25). We deduce

P (XN (s) H
) 1 ,

respectively,
PN (H
) 1 
for all N N. Since the set H
is compact, the family of probability measures
{PN } is tight, and by Theorem 3.4 it follows that it is relatively compact.
Applying the Chebyshev inequality once more, by Lemma 4.8, we obtain,
for every  > 0,
1
lim lim sup meas { [0, T ] : (L(s + i ), LN (s + i )) }
N T T
 T
1
lim lim sup (L(s + i ), LN (s + i )) d = 0.
N T T 0

Now let
YT (s) = L(s + iT ).
Then the latter inequality can be rewritten as

lim lim sup P ((XT,N (s), YT (s)) ) = 0.

N T
82 4 Limit Theorems

Let {PNj } be a subsequence {PN } which converges weakly to some measure

P . Then
D
XNj P .
j

This result and Theorem 3.6 imply

D
YT P ,
T

which is equivalent to the weak convergence of PT to P . Taking into account

that the family {PN } is relatively compact, by Lemma 3.2 we get
D
XN P .
N

Repeating this argument for the random elements LN (s + iT, ) and L(s +
iT, ), Lemma 4.10 shows that QT converges weakly to P as T tends to
innity. The lemma is proved. 

It remains to identify the limit of the probability measure PT as T .
This will nish the
Proof. By Lemma 4.11 the measures, PT and QT both converge weakly to
the same limit P as T tends to innity. It remains to show that P = P,
where P is dened by (4.9).
Let A B(H(D)) be a continuity set of P . By Theorem 3.1 and
Lemma 4.11 it follows that:
1
lim meas { [0, T ] : L(s + i, ) A} = P (A). (4.28)
T T

Fixing the set A, we dene the random variable on (, B(), m) by


1 if L(s, ) A,
() =
0 otherwise.
Then

E = dm = m{ : L(s, ) A} = P(A) < . (4.29)

By Lemma 4.6, and with the same argument as in the proof of Lemma 4.7,
we see that ( ()) is an ergodic process. Thus, in view of Theorem 3.13,

1 T
lim ( ()) d = E (4.30)
T T 0

for almost all . From the denition of and of the one-parameter group
{ : R} it follows that:

1 T 1
( ()) d = meas { [0, T ] : L(s + i, ) A}.
T 0 T
 4.7 Generalizations 83

This result, (4.29) and (4.30) imply

1
lim meas { [0, T ] : L(s + i, ) A} = P(A)
T T
for almost all . In view of (4.28) we deduce

P (A) = P(A)

for any continuity set A of the measure P . Since the continuity sets form a
determining class, we see that the latter equality holds for all A B(H(D)).
This proves the assertion of the theorem. 


4.7 Generalizations
The proof of limit Theorem 4.3 allows generalizations in various directions.
We conclude with the situation of Dirichlet series which do not necessarily
have multiplicative coecients.
Theorem 4.12. Assume that A(s) is a meromorphic function in the half-
plane > 0 , all poles of A(s) in this region are included in a compact set,
that A(s) satises the estimate

A( + it)  |t|

for 0 with some > 0, and that

 T
|A( + it + i )|2 d  T
0

for all > 0 . Further, suppose that A(s) has a representation as a Dirichlet
series


an
,
n=1
ns

the series being absolutely convergent for > 0 + 12 , and nx |an |2  x20 .
Then the probability measure, dened by
1
meas { [0, T ] : A(s + i ) A}
T
for A B(H(D0 )), where D0 := {s C : > 0 }, converges weakly to the
distribution of the random element


an (n)
 
n=1
ns

for s D0 , as T .
84 4 Limit Theorems

This limit theorem is due to Laurinikas [187, 192].

There are several further generalizations and extensions, most of them
belong to Laurinikas and his school. For instance, Laurinikas [188, 190]
proved a limit theorem for Matsumoto zeta-functions in the space of mero-
morphic functions. Other limit theorems for Dirichlet series (e.g., [185]) deal
with the distribution of values in the complex plane and may be regarded
as probabilistic analogues of Bohrs early results (see Sect. 1.2) and Selbergs
normal distribution result on the critical line. For a rst overview on the
various results we refer to [185]) and Matsumoto [244]; the best introduction
to limit theorems for Dirichlet series is Laurinikas monograph [186].
In the following section, we present another type of limit theorem.

4.8 A Discrete Limit Theorem

Whereas the preceding limit theorems were all of continuous character,
we shall now briey discuss the case of discrete limit theorems. Here the
attribute discrete refers to the value-distribution on arithmetic progressions.
We discuss this topic with respect to Matsumoto zeta-functions for which
Kainskaite [153, 154] obtained several discrete limit theorems.
We recall the denition of the Matsumoto zeta-function from Sect. 1.6. For
any positive integer m, we are given positive integers g(m) and further f (j, m)
(j)
as well as complex numbers am for 1 j g(m). Denoting the mth prime
number by pm , the associated Matsumoto zeta-function is given by


g(m)

1
(j) sf (j,m)
(s) = 1 am pm .
m=1 j=1

(j)
Under the conditions g(m)  p m and |am | pm the product converges for

> 1 + + . Suppose that (s) has an analytic continuation to some half-

plane > + + with 12 < 1 (except for at most nitely many poles)
with (s)  |t|c1 as |t| , where c1 is some positive constant and bounded
mean-square. We shall consider the value-distribution of (s) on arithmetic
progressions + in with n N in the strip

+ + < < 1 + + ,

where is a positive real number. For this aim we consider the probability
measure given by
1
PN (A) = {1 n N : (s + in) A}
N
for any Borel set A of the appropriate function space. Similarly as we did for
our L-functions, we consider the random variable
 4.8 A Discrete Limit Theorem 85

g(m)
 
(s, ) = (1 am
(j)
(pm )f (j,m) psf
m
(j,m) 1
) .
m=1 j=1

We denote by P the distribution of the random element (s, ). Kainskaite

[153, 154] proved

Theorem 4.13. Assume that is a real number such that

 
2k
exp (4.31)

is irrational for all k N. Then, for > + + 12 , the probability measure

PN converges weakly to P as N .

We omit the proof of this theorem but briey indicate where the assump-
tion on is needed. To obtain (4.13) in the proof of Lemma 4.4 the irrational-
ity of (4.31) for all k N was used. To overcome this unfortunate restriction,
Kainskaite and Laurinikas [155] recently investigated
 the case when there
exists a positive integer k for which exp 2k is rational. By considering a
certain subgroup of the innite torus , they succeeded in proving a discrete
limit theorem in the complex plane also for this case; their argument carries
over to spaces of analytic functions too.
The general idea and structure of proofs of discrete limit theorems are
rather similar to the continuous case, however, some arguments dier in detail.
5
Universality

Wer die Zetafunktion kennt, kennt die Welt!

Now we shall apply the limit theorem from Chap. 4 to derive information
on the value-distribution of L-functions. Our approach follows Bagchi [9],
respectively, the renements of Laurinikas [186]. Using the so-called positive
density method, introduced by Laurinikas and Matsumoto [200], we prove a
universality theorem for functions L S. Here, we shall make use of axiom (v).
This result is essentially due to Steuding [345] (under slightly more restrictive
conditions).

5.1 Dense Sets in Hilbert Spaces

In this section, we prove some preliminary results from the theory of Hilbert
spaces. For more details of the proofs below and the theory of Hilbert spaces
in general we refer to Bagchis thesis [10] and the monographs of Laurinikas
[186], Dunford and Schwartz [72], Duren [73] and Rudin [312].
Our rst aim is
Lemma 5.1. Let x1 , . . . , xn be linearly dependent vectors in a complex vector
space, and let a1 , . . . , an be complex numbers with |aj | 1 for 1 j n.
Then there exist complex numbers b1 , . . . , bn with |bj | 1 for 1 j n,
where at least one |bj | = 1, such that


n

n
aj xj = bj xj .
j=1 j=1

Proof. By assumption there are complex numbers c1 , . . . , cn , not all equal

zero, such that


n
cj xj = 0. (5.1)
j=1
88 5 Universality

Let
K = {(1 , . . . , n ) Cn : |j | 1 for 1 j n}
and
I = {t R : a + tc K},
where a = (a1 , . . . , an ) and c = (c1 , . . . , cn ). Since a K it follows that 0 I,
and so I is not empty. K is convex and thus I is convex, too. Hence, I is an
interval. Since K is compact and c is not the null vector, I is bounded. Denote
by t0 one of the endpoints of I. Let

b = (b1 , . . . , bn ) = a + t0 c.

Then b belongs to the boundary of K, that is |bj | 1 for 1 j n and

|bj | = 1 for at least one of the js. By (5.1),

n

n

n

n
bj xj = aj xj + t0 cj xj = aj xj ,
j=1 j=1 j=1 j=1

which proves the lemma. 



In what follows let H be a complex Hilbert space,

% and denote, as usual,
its inner product by x, y and its norm by x = x, x. Now we apply the
Lemma 5.1 to prove

Lemma 5.2. Let x1 , . . . , xn be points in a complex Hilbert space H and let

a1 , . . . , an be complex numbers with |aj | 1 for 1 j n. Then there
exist complex numbers b1 , . . . , bn with |bj | = 1 for 1 j n, satisfying the
inequality  2
 n 


n

n
 a x b x  4 xj 2 .
 j j j j 
 j=1 j=1  j=1

Proof. The proof is achieved by induction on n. The case n = 1 is trivial:

a1 x1 b1 x1 2 2x1 2 = 4 x1 2 .

Now assume that the assertion is true for n. Let x1 , . . . , xn+1 be points in H,
and let a1 , . . . , an+1 be complex numbers with |aj | 1 for 1 j n + 1.
Denote by yn+1 the orthogonal projection of xn+1 into the span of x1 , . . . , xn .
Then x1 , . . . , xn , yn+1 are linearly dependent. Thus, by Lemma 5.1, there exist
complex numbers c1 , . . . , cn+1 with |cj | 1 for 1 j n + 1, and |ck | = 1
for some 1 k n + 1, such that


n

n
aj xj + an+1 yn+1 = cj xj + cn+1 yn+1 .
j=1 j=1
 5.1 Dense Sets in Hilbert Spaces 89

First, suppose that k = n + 1. By the induction hypothesis there exist

complex numbers b1 , . . . , bn with |bj | = 1 for 1 j n such that
 2




 n n
 n
 c x b x  4 xj 2 .
 j j j j
j=1 j=1  j=1

Putting bn+1 = cn+1 we get

n+1

n+1
n

n
aj xj bj xj = cj xj bj xj + (an+1 cn+1 )zn+1 ,
j=1 j=1 j=1 j=1

where zn+1 := xn+1 yn+1 is orthogonal to x1 , . . . , xn (since yn+1 is the

orthogonal projection of xn+1 into the span of x1 , . . . , xn ). It follows that:
 2  2
n+1   n 



n+1
 

n

 aj xj   cj xj bj xj 
 + |an+1 cn+1 | zn+1
2 2
 bj xj  = 
 j=1 j=1   j=1 j=1 


n
4 xj 2 + 4 zn+1 2 .
j=1

Since zn+1 = xn+1 yn+1 2 xn+1 2 we are done in this case.

2 2

Next, suppose that 1 k n. Without loss of generality we may

assume that k = 1. By the induction hypothesis there are complex numbers
b2 , . . . , bn+1 with |bj | = 1 for 2 j n + 1 such that
 2




 n n
 n
 c x + c y b x b y  4 xj 2 + 4 yn+1 2 .
 j j n+1 n+1 j j n+1 n+1 
 j=2 j=2  j=2

Putting b1 = c1 we obtain


n+1

n+1

n

n
aj xj bj xj = cj xj + cn+1 yn+1 bj xj bn+1 yn+1
j=1 j=1 j=2 j=2
+ (an+1 bn+1 )zn+1 .
In view of the choice of yn+1 this leads to
 2  2
n+1   n 



n+1
 

n

 a x b x  =  c x + c y b x b y 
 j j j j  j j n+1 n+1 j j n+1 n+1 
 j=1 j=1   j=2 j=2 
+ |an+1 bn+1 |2 zn+1 2

n
4 xj 2 + 4 yn+1 2 + 4 zn+1 2 .
j=2

This gives the desired estimate for the second case and thus Lemma 5.2 is
proved. 

90 5 Universality

We have to recall a result from the theory of Hilbert spaces which can be
found, for example, in Trenogin [354]. A subset L of a complex Hilbert space
H is called a linear manifold (or subspace) if for all x, y L and all complex
numbers , the linear combination x + y is also an element of L. Denote
by L the orthogonal complement of L. If L is a linear manifold of H, every
element x H has a representation x = y + z, where y L and z L .

Lemma 5.3. A linear manifold L of a complex Hilbert space H is dense in H

if and only if L = {0}.

Now we are in the position to state the main result of this section.

Theorem 5.4. Let {xn } be a sequence in H satisfying


n=1 xn < ,
2

For 0 = x H, the series n=1 |xn , x| diverges to innity.
Then the set of all convergent series



an xn with |an | = 1
n=1

is dense in H.

This is essentially Pecherskys rearrangement theorem [288].

n : n N} with n = 1
Proof. First, we prove that there exists a sequence {
such that



n xn < . (5.2)
n=1

For this purpose, let {n : n N} be a sequence of independent random

variables dened on some probability space (R, B(R), P) such that
1
P(n = 1) = P(n = +1) = .
2
Then Xn := n xn is an H-valued random element with expectation EXn = 0.
By the rst assumption of the theorem, the sequence {Xn } is uniformly
bounded in norm. Moreover,





E Xn 2 = xn 2 < .
n=1 n=1

By the second assumption it follows that any non-zero element x H is not

orthogonal to all xn , and, in view of Lemma 5.3, the span of {xn } lies dense
in H. Consequently H is separable. Dene a set of linear combinations of the
{xn } by
 5.1 Dense Sets in Hilbert Spaces 91



m
S= rn xn : m N, rn Q(i) ,
n=1
where i denotes the imaginary unit. Next we show that S lies dense in H.
We observe that S is countable. Now let x0 be an arbitrary element of H.
By the above density result, for any positive , there exists an element from
the span of {xn }

M
y := cn xn
n=1

with cn C such that x0 y < . On the other side we can nd an element

M
z := rn xn
n=1

of S with coecients rn Q(i) for which y z < . This implies x0 z <

2, so S is dense in H with respect to the norm of H. 
Application of Theorem 3.10 implies that the series n=1 Xn converges
almost surely. Consequently, the desired sequence exists, and we denote it by
n }.
{
Next we have to show that for any x0 H andany positive  there exists
a sequence {an : |an | = 1} such that the series n=1 an xn converges and
 


 
x0 an xn  < .
 
n=1

By the rst assumption of the theorem we can nd a positive integer N

satisfying


2
xn 2 < , (5.3)
36
n=N

and, in view of (5.2),

 


  
 n xn  < (5.4)
  3
n=M

for any M N . Dene

M +N 


K= bn xn : |bn | 1, M N .
n=N

We show that K is dense in H.

Suppose that the contrary is true. Then there exists an element 0 = x0
H \ K. As in the proof of Lemma 5.1, K is convex, and so is K too. Thus,
{x0 } and K are disjoint closed convex subsets of H. The one-point set {x0 }
92 5 Universality

is compact. By the separation theorem for linear operators (see [72, Sect. V])
there exists a continuous linear functional f on H such that

Re f (x) > Re f (x0 )

for any x K. Clearly, f does not vanish identically. By the Riesz represen-
tation theorem (see [312, Theorem 6.19]) there exists an element 0 = z0 H
for which f (x) = x, z0 . Putting c = Re f (x0 ) we get

Re x, z0  = Re f (x) > c (5.5)

for any x K. Now choose bn so that |bn | = 1 and bn xn , z0  = |xn , z0 |.
The element


M +N
yM = bn xn ,
n=N

is in K. It follows from (5.5) that

M +N
Re yM , z0  = |xn , z0 | > c
n=N

for any M N. Consequently, the series n=N |xn , z0 | is convergent. Since
z0 = 0, this contradicts the second assumption of the theorem. So K is dense
in H.
By the density of K there exists a sequence {bn } with |bn | 1 such that
 


N 1

M +N 
  
x0 xn bn xn  < . (5.6)
  3
n=1 n=N

Lemma 5.2 in combination with (5.3) yields the existence of an element

M +N
u := an xn with |an | = 1
n=N

in H such that
M +N 2



M +N 

M +N
2
 
 bn xn an xn  4 xn 2 < .
  9
n=N n=N n=N

Thus,
 


M +N 
  
u bn xn  < . (5.7)
  3
n=N

Now dene



v= n xn .
n=M +N +1
 5.1 Dense Sets in Hilbert Spaces 93

In view of (5.4) we have


v < . (5.8)
3
Further, let


N 1


w= xn + u + v = an xn ,
n=1 n=1
where an = 1 for 1 n N 1 and an = n for n > M + N . Then (5.6),
(5.7), and (5.8) imply
   
 N
1 M
+N  M
+N

M +N 
   
x0 w x0 an xn bn xn  +  bn xn an xn 
   
n=1 n=N n=N n=N
 


 
+ an xn 
 
n=M +N +1
< ,
which proves the theorem. 

For our later purpose, we prove now a lemma which is not really related to
the topic of this section, however, its proof is very similar to a certain aspect
in the proof of the previous theorem.
Lemma 5.5. Let {zn } be a sequence of complex numbers such that the series

n=1 |zn | converges. If {n } is a sequence of independent random variables
2

on some probability space (R, B(R), P) such that

1
P(n = 1) = P(n = +1) =
2

for any n N, then the series n=1 n zn converges almost surely. In partic-
n } with n {1} such that
ular, there exists a sequence {



n zn
n=1
converges.
Proof. Let Xn = n zn , then {Xn } is a sequence of independent complex
random variables with expectation EXn = 0. By the assumption of the lemma







E(Re Xn E(Re Xn ))2 = E(Re Xn )2 |zn |2 < .
n=1 n=1 n=1

In view of Theorem 3.9the series n=1 Re Xn converges almost surely,

respectively, the series n=1 n Re z n converges for almost all sequences
{n =1}. By the same reasoning for the imaginary part of Xn , it follows

that n=1 n Im zn converges for almost all sequences {n } with n {1} as
well. This yields the assertion of the lemma. 

94 5 Universality

5.2 Application to the Space of Analytic Functions

Next we shall apply Theorem 5.4 to an appropriate function space. We already
dealt with Hardy spaces in the proof of Voronins universality theorem in
Sect. 1.3. However, the general situation is a bit more complicated; for details
we refer to Duren [73], Laurinikas [186, Sects. 6.2 and 6.3] and Rudin [312,
Chap. 17].
Let D be a simply connected domain in the complex plane with at least
two boundary points. The associated Hardy space H2 (D) consists of those
functions f (s) which are analytic in D such that the subharmonic function
|f (s)|2 has a harmonic majorant in D. If F (s) is the smallest harmonic
majorant for |f (s)|2 and s0 is a xed point in D, then the norm of f H2 (D)
is dened by
1
f = (F (s0 )) 2 .
With this norm H2 (D) becomes a complex Hilbert space.
Theorem 5.6. For any given f H2 (D) there exists a Borel measure f
with support concentrated in D such that if g H2 (D) has a continuous
extension to D, then the inner product of H2 (D) can be expressed by

f, g = g df .
D
Now we are in the position to apply Theorem 5.4 to the Hardy space
H2 (D) in order to derive a denseness result for the larger space of analytic
function H(D). Recall that denotes the unit circle in the complex plane.
Theorem 5.7. Let D be a simply connected domain in the complex plane.
Suppose that the sequence {fn } in H(D) satises the following assumptions:
If is a complex Borel measure on (C, B(C)) with compact support con-
tained in D such that



 
 fn d < ,
 
n=1 C
then 
sr d(s) = 0 for any r N0 .
C

The series n=1 fn converges in H(D).
For every compact K D,



max |fn (s)|2 < .
sK
n=1

Then the set of all convergent series

b(n)fn with b(n)
n=1

is dense in H(D).
 5.2 Application to the Space of Analytic Functions 95

This is Theorem 6.3.10 of Laurinikas [186], respectively, Lemma 5.29 of

Bagchi [9].
Proof. Let K be a given compact subset of D. We may choose a simply con-
nected domain G for which K G, the closure G of G is a compact subset of
D, and the boundary of G is a regular Jordan curve. In view of Theorem 5.6,
for any fn H2 (D) there exists a complex Borel measure fn on the Hilbert
space H2 (G) such that the norm of fn is given by

fn 2 = fn , fn  = fn dfn .
G

Consequently,

fn 2 max |fn (s)| | dfn |  max |fn (s)|2 .
sG G sG

Hence, by the third assumption,

fn 2 < . (5.9)
n=1

Now assume that g H2 (G) satises

|fn , g| < .
n=1

With the Borel measure g assigned to g (according Theorem 5.6) it follows

that:

 
 
 fn dg  < .

n=1 G

By the rst assumption of the theorem we get, for every r N0 ,


g(s), sr  = sr dg (s) = 0.
G

It follows that g is orthogonal to all polynomials. One can show that the set
of polynomials is dense in H2 (G) (see [73]), and thus, by Lemma 5.3, g(s) is
identically zero. Hence, for any g(s)  0,



|fn , g| = .
n=1

This and (5.9) show in view of Theorem 5.4 that the set of all convergent
series


n fn with |n | = 1
n=1

is dense in H2 (G). It remains to extend this result to the larger space H(D).
96 5 Universality

Now let f H(D) and  > 0. By Theorem 5.4 there exists a sequence
{n : |n | = 1} such that the series



n fn (s)
n=1

converges on D (in the topology of H2 (D)). This convergence is uniform on

every compact subset K of D and we may assume that
 


  
max  n fn (s) f (s) < .
sK   4
n=1

Hence, we can nd a number M for which

 


  
max  n fn (s) f (s) < ,
sK 
nM  2

and with regard to the second assumption of the theorem

 


  
max  fn (s) < .
sK   2
n>M

Now let 
n if n M,
b(n) =
1 otherwise.

Then the series n=1 b(n)fn (s) converges in H(D) and by the above inequal-
ities
 


 
max  b(n)fn (s) f (s)
sK  
n=1
   

 

   
max  (n)fn (s) f (s) + max  fn (s) < .
sK   sK  
nM n>M

This proves the theorem. 



5.3 Entire Functions of Exponential Type

Now we state some basic facts from the theory of entire functions.
A function f (s) which is analytic in the closed angular region | arg s| 0
with 0 < 0 is said to be of exponential type if
log |f (r exp(i))|
lim sup <
r r
for || 0 , uniformly in .
 5.3 Entire Functions of Exponential Type 97

Lemma 5.8. Let be a complex Borel measure on (C, B(C)) with compact
support contained in the half-plane > 0 . Moreover, for s C, dene the
function

f (s) = exp(sz) d(z).
C
Then f is an entire function of exponential type. If f (s) does not vanish
identically, then
log |f (r)|
lim sup > 0 .
r r
This is Lemma 6.4.10 of Laurinikas [186]. It is easily seen that the function
f (s), dened in the lemma, is indeed an entire function of exponential type.
The proof of the lemma relies on the Borel transform f of f which, for complex
s that are not contained in the support of , is given by

d(z)
f(s) = .
C sz
The Borel transform is analytic everywhere with the exception of a neigh-
bourhood of s = 0. This shows that the closed convex hull of the set of
singularities of the Borel transform, the so-called conjugate indicator diagram
of f , is contained in the convex hull of the support of . By the assumption
of the lemma, this is a subset of the half-plane > 0 . This leads via the
so-called PhragmnLindelf indicator function and a theorem of Boas to the
desired inequality; for the details we refer to Chap. 5 in Boas monograph [26]
and Sect. 6.4 in Laurinikas [186].
Plya investigated the rate of growth of functions of exponential type on
arithmetic progressions. This was extended by Bernstein [19] to more general
divergent sequences. We state a variant of Bernsteins theorem.
Theorem 5.9. Let f (s) be an entire function of exponential type, and let {m }
be a sequence of complex numbers. Moreover, assume that there are positive
real constants , and such that
log |f (iy)|
lim supy y ,
|m n | |m n|,
m
limm m = ,
< .
Then
log |f (m )| log |f (r)|
lim sup = lim sup .
m |m | r r
The case = 1 follows directly from Bernsteins theorem [19]. The general
case can be deduced by applying the case = 1 to the function F (s) = f (s)
and an appropriate scaling of the other appearing quantities. A proof can be
found in Laurinikas book [186, Sect. 6.4], or in Boas [26, Chap. 10]; notice
that the proof in [26] contains an error as was pointed out by Bagchi [10].
98 5 Universality

5.4 The Positive-Density Method

It follows from the prime number Theorem 1.1 that there exist constants c1
and c2 > 0 such that

1 %

= log log x + c1 + O exp c2 log x . (5.10)

p
px

This asymptotic formula plays an essential role in Bagchis probabilistic proof

of Voronins universality theorem (as a substitute of Voronins use of the prime
number theorem in his approach). However, for many L-functions, given by a
Dirichlet series


a(n)
,
n=1
ns
an analogous formula for

|a(p)|
,
p
px

is not known! Laurinikas and Matsumoto [200] overcame this diculty by

using sets of prime numbers of positive density within the set of all primes in
combination with (5.10). This is manifested in a mean-square formula for the
coecients a(p) (which are easier obtainable). This is the essential idea of the
so-called positive-density method which we shall use now. For more details
about the origin of this nice argument see Matsumoto [242].
In what follows, we assume that L S. Recall that D denotes the inter-
section of the mean-square half-plane > m for L(s) with the critical strip
(see (4.1)); we have already noticed that the strip D is not empty (see (4.2)).
For s D, b(p) = {z C : |z| = 1}, and any prime p, we dene the
function
m  1
j (p)b(p)
gp (s, b(p)) = log 1 (5.11)
j=1
ps

m  
j (p)b(p)
= log 1 .
j=1
ps

We observe the similarities between exp(gp (s, b(p))) and the corresponding
Euler factor in the Euler product representations of L(s) and of the associated
random element (4.6). Next we shall prove

Theorem 5.10. Let L S. The set of all convergent series

gp (s, b(p)) with b(p)
p

is dense in H(D), where is the unit circle in C.

 5.4 The Positive-Density Method 99

This denseness theorem plays a crucial role in the proof of the universality
theorem for S. We observe that the statement of the theorem is not true if D
is replaced, for example, by the half-plane of absolute convergence (since the
terms gp (s, b(p)) tend to zero as ).
Proof. Dene


m  
j (p)
gp (s) = gp (s, 1) = log 1 .
j=1
ps

For a parameter N N which will be chosen later, dene


gp (s) if p > N,
gp (s) =
0 if p N.

We claim that there exists a sequence {b(p) : b(p) } such that the series


b(p)gp (s) (5.12)
p

converges in H(D).
In order to prove this claim we consider the identity between the Dirichlet
series and the Euler product representation of L S,


m

j (p)k a(p)
gp (s) = = + rp (s), (5.13)
j=1 k=1
kpks ps

say, where



m


aj (p)k 1
rp (s) = 
kpks kpk
k=2 j=1 k=2

(since all j (p) are of modulus 1 by Lemma 2.2). Hence

1
rp (s)  < (2) (5.14)
p p
p2

and so the series p rp (s) converges uniformly on compact subsets of D. Now
let {j } be a strictly decreasing sequence of real numbers which tends to m
as j . Taking into account Lemma 5.5, for each j, there exists a sequence
{p = 1} of independent random variables such that the series

a(p)p

p
p j

converges almost surely. Thus, we can nd a sequence {b(p) : b(p) } for

which

a(p)b(p)

p
p j
100 5 Universality

converges for any j. Since Dirichlet series converge uniformly on compact

subsets in the half-plane of their convergence, it follows that the series

a(p)b(p)
(5.15)
p
ps

converges
 on any compact subset of D. Together with the convergence of
p rp (s), this proves our claim that the series (5.12) converges in H(D).
Now we shall show that the set of all convergent series


b(p)gp (s) with b(p) (5.16)
p

is dense in H(D). Obviously, it is sucient to show that the set of all conver-
gent series


b(p)fp (s) with b(p) (5.17)
p

is dense in H(D), where fp (s) := b(p)gp (s). To prove this we apply

Theorem 5.7. Taking into account (5.12)
 we have already veried the second
assumption, namely that the series p fp (s) converges in H(D). The third
assumption is easily shown by the estimate


1
|fp (s)|2  ,
p p
p2

since the series on the right converges on any compact subset of D (in a similar
manner as in (5.14).
In order to verify the rst assumption of Theorem 5.7, let be a complex
Borel measure on (C, B(C)) with compact support contained in D such that

 

 fp (s) d(s) < . (5.18)
 
p C

Dene
a(p)b(p)
hp (s) = .
ps
Then, by (5.13),


|a(p)|

|fp (s) hp (s)| = + |rp (s)|  1
p
p
pN p>N

uniformly on compact subsets of D. Hence, (5.18) implies

 

 hp (s) d(s) < . (5.19)
 
p C
 5.4 The Positive-Density Method 101

In view of the polynomial Euler product representation axiom (iv), respec-

tively, Lemma 2.2, for any prime p we have |a(p)| m (where m is the degree
of the polynomial dening the local Euler factors). Hence we may dene angles
p [0, 2 ] by setting
 


m 

|a(p)| =  j (p) = m cos p . (5.20)
 j=1 

Notice that cos p is non-negative. We rewrite (5.19) as

|(log p)| cos p < , (5.21)
p

where 
(z) := exp(sz) d(s).
C
We shall show that (z) vanishes identically; however, for this purpose it is of
advantage to consider an appropriate subseries of (5.21).
Fix a number with
 

0 < < min 1, ;
m

here is the quantity appearing in axiom (v) on the prime mean-square for
L S. Dene
P = {p : p prime and cos p > } .
Then (5.21) yields


|(log p)| < . (5.22)
pP

To deduce the vanishing of (z) we apply Theorem 5.9. We choose a suciently

large positive constant M such that the support of is contained in the region
{s C : m < < 1, |t| < M }. Then, by the denition of (z),

|(iy)| exp(M y) | d(s)|
C

for y > 0. Therefore,

log |(iy)|
lim sup M,
y y
and the rst condition of Theorem 5.9 is valid with = M . Now x a number

with 0 < < M , and dene
     & 
1 1
A = n N : r n , n + with |(r)| exp(r) .
4 4
102 5 Universality

We want to show that A has natural density 1 (i.e., (z) is small).

For n N, let
     
1 1
= (n) := exp n and = (n) := exp n+
4 4
(notice that and depend on n). We observe that for n  A we have
1
|(log p)| >
p
for all primes p (, ]. Thus, it follows that:





1
|(log p)| |(log p)| .
p
p P
nA p

P nA

P
p
<p <p <p

Hence, in view of (5.22),

1
< . (5.23)
p
nA
pP
<p

Let
(x) = {p x : p P }.
Then, for < u , we obtain




(cos p )2 1 + 2 1 (5.24)
<pu pP
P
p
<pu <pu

= (1 2 )( (u) ()) + 2 ((u) ()).

By axiom (v) on the mean-square of the coecients,

1

(cos p )2 = 2
|a(p)|2 2 (x). (5.25)
m m
px px

Let be a small positive constant. Substituting (5.25) in (5.24), gives

 
2
2
(u) () m + O(1) ((u) ())
1 2
for u (1 + ), as n . Thus, we get by partial summation

1  d (u)  2 2 
d(u)
= m
+ O(1)
p u 1 2
u
P
p

<p
 


2 1
m2
+ O(1) , (5.26)
1 2 p
(1+)<p
 5.4 The Positive-Density Method 103

as n . From (5.10) we deduce

1 log  
= log + O exp(c2 n)
p log((1 + ))
(1+)<p
   
1 log(1 + ) 1 1
= +O .
2 n n2

This gives in formula (5.26)

1 2    
2

1 log(1 + ) 1 1
m + O ,
p 1 2 2 n n2

pP
<p

as n . Hence, it follows from (5.23) that

1
< (5.27)
n
nA

and so A has natural density 1.

Let A = {ak : k N} with a1 < a2 < . . .. Then (5.27) implies
ak
lim = 1. (5.28)
k k
By the denition of the set A, there exists a sequence {k } such that
   
1 1
ak < k ak + and |(k )| exp(k ).
4 4

Hence, from (5.28) it follows that:

k log |(k )|
lim = and lim sup 1.
k k k k
Applying Theorem 5.9, we obtain
log |(r)|
lim sup 1. (5.29)
r r
However, by Lemma 5.8, if (z) does not vanish identically, then

log |(r)|
lim sup > 0,
r r
contradicting (5.29). Therefore (z) 0. Dierentiating this identity r-times
with respect to z, we obtain

0 = (1)r sr exp(sz) d(s).
C
104 5 Universality

Putting z = 0 we get 
sr d(s) = 0
C
for r N0 . Thus the rst assumption of Theorem 5.7 is also satised and we
obtain the denseness of all convergent series
 (5.16). It remains to show that
this does not change when we add terms pN gp (s).
Let f H(D), K be a compact subset of D, and  > 0. Further, choose
the parameter N such that




1 <  ,
max
(5.30)
sK
=2
p 4m
p>N

where m is still the degree of the local Euler factors of L(s). By the denseness
of all convergent series (5.16) in H(D) we see that there exists a sequence
{b(p) : b(p) } such that
 



  

max f (s) gp (s) b(p)gp (s) < . (5.31)
sK   2
pN p>N

Setting 
1 if p N,
b(p) =
b(p) otherwise,
then (5.30) and (5.31) imply
   
   

 


max f (s) 
gp (s, b(p)) = max f (s) gp (s) gp (s, b(p))
sK   sK  
p pN p>N

 
 




max f (s) gp (s) b(p)gp (s)
sK  
pN p>N
 
 



+ max  b(p)gp (s) gp (s, b(p))
sK  
p>N p>N




 1
< + 2m max < ;
2 sK
=2
p
p>N

for the last but one inequality we have used that |a(p)| m by Lemma 2.2.
Since f (s), K and  were arbitrary, the theorem is proved. 

 5.5 The Support of the Limit Measure 105

5.5 The Support of the Limit Measure

Now we are going to apply limit Theorem 4.3. We restrict our studies to the
space of functions analytic on bounded open rectangles. For this purpose, we
dene, for an arbitrary xed real number M > 0,

DM = {s = + it C : m < < 1, |t| < M }. (5.32)

Obviously, DM D by (4.1). Therefore, we obtain by the induced topology

that for s DM the function L(s, ), dened by (4.6), is an H(DM )-valued
random element on the probability space (, B(), m). Now denote by Q the
distribution of L(s, ) on (H(DM ), B(H(DM ))). Further, dene a probability
measure QT by setting
1
QT (A) = lim meas { [0, T ] : L(s + i ) A} (5.33)
T T
for A B(H(DM )). Then we deduce from Theorem 4.3

Corollary 5.11. Let L S. Then QT converges weakly to Q as T .

Next we examine the support of the measure QT :

Lemma 5.12. The support of the measure QT is the set

SM := { H(DM ) : (s) = 0 for s DM , or (s) 0}.

In order to prove this lemma we make use of Hurwitzs classical theorem

on uniformly convergent sequences of functions and their zeros.

Theorem 5.13. Let G be a region and {fn } be a sequence of functions analytic

on G which converges uniformly on G to some function f . Suppose that f (s) 
0, then an interior point s0 of G is a zero of f (s) if and only if there exists
a sequence {sn } in G which tends to s0 as n , and fn (sn ) = 0 for all
suciently large n.

The proof of Hurwitzs theorem relies in the main part on Rouchs the-
orem (see Theorem 8.1) and can be found in the monographs of Titchmarsh
[352, Sect. 3.4.5] or Conway [62, Sect. VII.2].
Now we can give the

Proof of Lemma 5.12. Recall the denition (5.11) of the functions gp (s, (p)).
Since {(p)} is a sequence of independent random variables on the probability
space (, B(), m), it follows that {gp (s, (p))} is a sequence of independent
H(DM )-valued random elements. The support of each (p) is the unit circle
, and therefore the support of the random elements gp (s, (p)) is the set

{ H(DM ) : (s) = gp (s, b) with b }.

106 5 Universality

Consequently, by Theorem 3.16, the support of the H(DM )-valued random

element

log L(s, ) = gp (s, (p))
p

is the closure of the set of all convergent series p gp (s, b(p)). By Theorem 5.10
the set of these series is dense in H(DM ). The map

exp : H(DM ) H(DM ), f  exp(f ),

is a continuous function sending log L(s, ) to L(s, ) and H(DM ) to SM \{0}.

Therefore, the support SL of L(s, ) contains SM \ {0}. On the other hand,
the support of L(s, ) is closed. By Hurwitzs Theorem 5.13 it follows that

SM \ {0} = SM .

Thus, SM SL . In view of Lemma 2.2 the functions

m  
j (p)(p)
exp(gp (s, (p))) = 1
j=1
ps

are non-zero for s DM and . Hence, L(s, ) is an almost surely con-

vergent product of non-vanishing factors. If we apply Hurwitzs Theorem 5.13
again, we conclude that L(s, ) SM almost surely. Therefore SL SM . The
proof is nished. 


5.6 The Universality Theorem

Now we are in the position to prove the main result of this chapter, a gener-
alization of Voronins universality theorem to the class S.

Theorem 5.14. Let L S, K be a compact subset of the strip D = {s C :

m < < 1} with connected complement, and let g(s) be a non-vanishing
continuous function on K which is analytic in the interior of K. Then, for
any  > 0
 
1
lim inf meas [0, T ] : max |L(s + i ) g(s)| <  > 0.
T T sK

So L-functions in S are universal and their strip of universality is at least the

intersection of their mean-square half-plane > m with the open right half
of the critical strip. Recall that, by (4.2), this strip of universality D contains
at least  
1 1 L
max ,1 < < 1,
2 1 + 2L
where L and L are dened by axioms (ii) and (iii).
 5.6 The Universality Theorem 107

Proof. Since K is a compact subset of D, there exists a number M such that

K DM , where DM is dened by (5.32).
First, we suppose that g(s) has a non-vanishing analytic continuation to
DM . By Lemma 5.12 the function g(s) is contained in the support SL of the
random element L(s, ). Denote by the set of functions H(DM ) such
that
max |(s) g(s)| < .
sK

Since by Corollary 5.11 the measure QT converges weakly to Q, as T ,

and the set is open, it follows from Theorem 3.1 and from properties of the
support that
 
1
lim inf meas [0, T ] : max |L(s + i ) g(s)| < 
T T sK

= lim inf QT () Q() > 0. (5.34)

T

This proves the theorem in the case of functions g(s) which have a non-
vanishing analytic continuation to DM . We remark that analytic functions on
D can be approximated uniformly on compact sets; here the restriction on K
to have connected complement is not necessary.
Now let g(s) be as in the statement of the theorem. In this case we have to
apply a complex analogue of Weierstrass approximation theorem, the theorem
of Mergelyan on the approximation of analytic functions by polynomials.
Theorem 5.15. Let K be a compact subset of C with connected complement.
Then any continuous function g(s) on K which is analytic in the interior of
K is uniformly approximable on K by polynomials in s.
This generalizes a classic result of Runge and gives the nal solution of the
problem of uniform approximation by polynomials on compacta. The inge-
nious proof relies on Greens formula and Riemanns mapping theorem and
can be found, for example, in Mergelyans article [246], Rudin [312] or Walshs
monograph [367, Sect. A.1]. The restriction to compact sets with connected
complement is natural. A necessary and sucient condition that every func-
tion analytic on a closed bounded point set C can be uniformly approximated
on C by polynomials is that C should not separate the plane, or equivalently,
C should be the complement of an innite region. For instance, the function
1
g(s) = s cannot be approximated uniformly on C by polynomials if C is
an annulus enclosing .
We continue with the proof of Theorem 5.14. By Mergelyans approxi-
mation Theorem 5.15 there exists a sequence of polynomials Gn (s) which
converges uniformly on K to g(s) as n . Since g(s) is non-vanishing on
K, we have Gm (s) = 0 on K for a suciently large m, and

max |g(s) Gm (s)| < . (5.35)
sK 4
108 5 Universality

Since the polynomial Gm (s) has only nitely many zeros, there exists a region
G whose complement is connected such that K G and Gm (s) = 0 on G. Hence
there exists a continuous branch log Gm (s) on G, and log Gm (s) is analytic in
the interior of G. Thus, by Mergelyans Theorem 5.15, there exists a sequence
of polynomials Fn (s) which converges uniformly on K to log Gm (s) as n .
Hence, for suciently large k,

max |Gm (s) exp(Fk (s))| < .
sK 4
From this and from (5.35) we obtain

max |g(s) exp(Fk (s))| < . (5.36)
sK 2
From (5.34) we deduce
 
1 
lim inf meas [0, T ] : max |L(s + i ) exp(Fk (s))| < > 0.
T T sK 2

In combination with (5.36) this proves Theorem 5.14. 



A close look on the proof of Theorem 5.14 shows that omitting a nite
number of Euler factors in the Euler product does not inuence the univer-
sality property. It is an interesting question to which extent innitely many
Euler factors can be omitted. A rst approach to answer this problem was
given by Schwarz, Steuding and Steuding [319] by using the notion of related
arithmetical functions.
We shall compare the universality Theorem 5.14 with the result of
Laurinikas [190] for Matsumoto zeta-functions (see also Sect. 1.6). Both
theorems apply to polynomial Euler products; however, the form of the Euler
product (iv) is more restrictive than the one of Matsumoto zeta-functions.
Furthermore, Theorem 5.14 relies on axiom (v) while Laurinikas assumes
(1.40); in our case his condition can be translated as follows: for all primes p,
|a(p)| is bounded from below by some positive constant. This is in general
a rather strong assumption. For example, Lehmer [212] conjectured that
Ramanujans -function (n), given implicitly as the Fourier coecients of the
modular discriminant (1.33), is non-zero for any n N. This is still unproved
(also in the form (p) = 0 for prime p). Laurinikas [190] also claims that
(1.40) can be replaced by the condition that, for  > 0,
1

1  x ,
(x) px
|a(p)|<

where 12 (again in our notation). Also this condition is rather

restrictive. For example, in the case of the L-function associated with the
modular discriminant one would need a quantitative version of the SatoTate
 5.7 Discrete Universality 109

conjecture (see (6.22)) in order to verify this assumption. Nevertheless, we

cannot exclude the possibility of a number-theoretical relevant Matsumoto
zeta-function which is not a polynomial Euler product of xed degree.
Reviewing our lengthy proof of Theorem 5.14 we may understand univer-
sality as a kind of ergodicity on function spaces. On the other side, all proofs
of universality results, known so far, depend on some arithmetical conditions;
but is universality really an arithmetic phenomenon or not? This question
was raised by Matsumoto [242]. It seems reasonable that the universality
of Dirichlet series is a common phenomenon in analysis (see the appendix),
that it is related to Julia rays in value-distribution theory and to ergodical
dynamical systems as well, but no link is known so far.

5.7 Discrete Universality

Now we consider the phenomenon of discrete universality, that means that the
shifts are restricted to arithmetic progressions. This concept of universality
was introduced by Reich [306] (see (1.39)). It is remarkable that this still leads
to universality results with positive lower density for the shifts . The precise
statement of Reichs discrete universality theorem for Dedekind zeta-functions
is
Theorem 5.16. Let K be an algebraic number eld of degree d over Q, let
K be a compact subset of the strip max{ 12 , 1 d1 } < < 1 with connected
complement, and assume that g(s) is a non-vanishing continuous function on
K which is analytic in its interior. Then, for any real = 0 and any  > 0,
 
1
lim inf  n N : max |K (s + in) g(s)| <  > 0.
N N sK

In Sect. 4.8 we presented a discrete limit theorem for Matsumoto zeta-

functions due to Kainskaite, where was assumed to satisfy a certain arith-
metical property. We also mentioned the work of Kainskaite and Laurinikas
[155] which removes this restriction. Interestingly, this problem was solved by
Reich in a dierent way. (For more details about Dedekind zeta-functions see
Sect. 13.1.)
Bagchi [9, 10] combined Reichs result with ideas from Gonek [104] in order
to obtain joint discrete universality for Dirichlet L-functions.
Theorem 5.17. Let 1 mod q1 , . . . ,  mod q be pairwise non-equivalent
Dirichlet characters, K1 , . . . , K be compact subsets of 12 < < 1 with con-
nected complements. Further, for each 1 j  let gj (s) be a continuous
non-vanishing function on Kj which is analytic in the interior of Kj . Then,
for any = 0 and any  > 0,
 
1
lim inf  n N : max max |L(s + in, j ) gj (s)| <  > 0.
N N 1j sKj
110 5 Universality

Bagchi considered only characters to a xed modulus q (in which case the
characters mod q are trivially non-equivalent) and he assumed the compact
sets Kj to be simply connected and locally path connected; however, without
big eort his result (Theorem 5.3.4 in [9]) can easily be extended to the form
given above.
It should be remarked that these results do not contain the continuous
analogues (as, for example, Theorem 5.14 in the latter case) nor that they
can be deduced from Theorem 5.14 (to our present knowledge). Nevertheless,
the methods of proof are rather similar. In fact, they depend on an appropriate
limit theorem as Theorem 4.13; the remaining arguments are essentially the
same.
Matsumoto [243] asked for discrete universality for L-functions asso-
ciated with cusp forms and Matsumoto zeta-functions. Garbaliauskiene and
Laurinikas [84] proved a discrete version of Theorem 1.11 for L-functions to
elliptic curves; Laurinikas, Matsumoto and Steuding [205] obtained discrete
universality for the more general case of L-functions to newforms.
6
The Selberg Class

What is a zeta-function (or an L-function)? We know one when we

see one. M.N. Huxley

In 1989, Selberg dened a rather general class S of Dirichlet series having

an Euler product, analytic continuation and a functional equation of Riemann-
type, and formulated some fundamental conjectures concerning them. His aim
was to study the value-distribution of linear combinations of L-functions. In
the meantime, this so-called Selberg class became an important object of
research, but still it is not understood very well. In this chapter, we shall
investigate universality for functions in the Selberg class. Therefore, we only
present results on this class which are related to our studies; for detailed
surveys on the Selberg class, we refer to Kaczorowski and Perelli [160],
Perelli [290], and M.R. Murty and V.K. Murty [270].

6.1 Denition and First Properties

Many authors, for example, Lekkerkerker [214], Perelli [289], and

Matsumoto [239] have introduced classes of Dirichlet series to nd common
patterns in their value-distribution. However, the most successful class seems
to be the class introduced by Selberg [323]. The Selberg class S consists of
Dirichlet series


a(n)
L(s) =
n=1
ns
satisfying the following hypotheses:
(1) Ramanujan hypothesis. a(n)  n
for any  > 0, where the implicit con-
stant may depend on .
(2) Analytic continuation. There exists a non-negative integer k such that
(s 1)k L(s) is an entire function of nite order.
112 6 The Selberg Class

(3) Functional equation. L(s) satises a functional equation of type

L (s) = L (1 s),

where

f
L (s) := L(s)Qs (j s + j )
j=1

with positive real numbers Q, j , and complex numbers j , with

Re j 0 and || = 1.
(4) Euler product. L(s) has a product representation

L(s) = Lp (s),
p

where 


b(pk )
Lp (s) = exp
pks
k=1

with suitable coecients b(pk ) satisfying b(pk )  pk for some < 12 .

The Euler product hypothesis implies that the coecients a(n) are multi-
plicative, and that each Euler factor has the Dirichlet series representation


a(pk )
Lp (s) = , (6.1)
pks
k=0

absolutely convergent for > 0. Dierentiation of the identity between the

two representations for the local Euler factors leads to the relation

 

ka(pk )
b(p )

log p = Lp (s) = Lp (s)
pks ps
k=1 =1


1

= log p ms
a(pk )b(p ).
m=1
p
m=k+

Comparing the coecients, we obtain a(p) = b(p) and

1

1
b(p ) = a(p ) ka(pk )b(pk ).

k=1

Moreover, it turns out that each Euler factor is absolutely convergent in the
half-plane > 0 and non-vanishing for > .
Axioms (1) and (2) imply that any element L(s) of the Selberg class is
analytic in the whole complex plane except for a possible pole at s = 1 (by
the same reasoning as for S in Sect. 2.2).
 6.1 Denition and First Properties 113

The structure of the Selberg class is of special interest. The degree of L S

is dened by

f
dL = 2 j . (6.2)
j=1

Although the data of the functional equation are not unique, the degree
is well-dened. If NL (T ) counts the number of zeros of L S in the
rectangle 0 1, |t| T (according to multiplicities), by standard contour
integration one can show
dL
NL (T ) T log T, (6.3)

in analogy to the Riemannvon Mangoldt formula (1.5) for Riemanns zeta-
function; we shall give a more precise asymptotic formula in Theorem 7.7. It
is conjectured that all L S have integral degree. Slightly stronger is the

Strong -conjecture. Let L S. All j appearing in the Gamma-factors

of the functional equation can be chosen to be equal to 12 .

The constant function 1 is the only element of S of degree zero. Recently,

Kaczorowski and Perelli [162] showed that all functions L S with degree
0 < dL < 53 have degree equal to one. For our purpose the following slightly
weaker result is sucient.

Theorem 6.1. If L S and 0 dL < 1, then L(s) 1.

For 0 < dL < 1, the statement is implicitly contained in the work of

Richert [308], Bochner [27] and Vignras [358]. We give a sketch of a simple
proof following Conrey and Ghosh [59] and Molteni [256].

Proof. Let B be a constant such that a(n)  nB . By Perrons formula,

 c+i  
1 xs xc+B
a(n) = L(s) ds + O ,
2i ci s T
nx

where c > 1 is a constant. Shifting the path of integration to the left, yields,
by the PhragmnLindelf principle (see Sect. 6.5), the asymptotic formula

 d 1

(1+B) dL +1 +

a(n) = xP (log x) + O x L ,
nx

where P (x) is a computable polynomial according to the principal part of the

Laurent expansion of L(s) at s = 1. By subtraction, this implies
dL 1
(1+B) dL +1 +

a(n)  n , (6.4)
114 6 The Selberg Class

where the implicit constant depends on B. For dL < 1 the exponent is

negative, and we may choose B arbitrarily large. Then L(s) is uniformly
bounded in every right half-plane. This is a contradiction for L S with
positive degree since the functional equation implies a certain order of growth.
This shows that S is free of elements having degree 0 < d < 1.
It remains to consider the case that dL = 0. Then the functional equation
takes the form:
Qs L(s) = Q1s L(1 s)
(there are no Gamma-factors). By (6.4) the a(n) are so small that the Dirichlet
series for L(s) converges in the whole complex plane. Thus, we may rewrite
the functional equation as

 2 s

Q a(n) s
a(n) = n . (6.5)
n=1
n n=1
n

We may regard this as an identity between absolutely convergent Dirichlet

series. Thus, if a(n) = 0, then Q2 /n is an integer. In particular, q := Q2 N.
Moreover, since q has only nitely many divisors, it follows that L(s) is a
Dirichlet polynomial. If q = 1, then L(s) 1 and we are done with the case
dL = 0. Hence, we may assume that q > 1.
Since the Dirichlet coecients a(n) are multiplicative, we have a(1) = 1
and via (6.5)
a(1)Q2s = Q1 a(Q2 )Q2s ;
thus, |a(q)| = Q. In particular, there exists a prime p such that the exponent
of p in the prime factorization of q is positive and, by the multiplicativity
of the a(n)s,
|a(p )| p/2 .
Now consider the logarithm of the corresponding Euler factor:
 


a(pm )
b(pk )
log 1 + ms
= .
m=1
p pks
k=1

Viewing this as a power series in X = ps , we write

log P (X) = Bk X k with Bk = b(pk ).
k=1

Since a(1) = 1, we nd




1
k

P (X) = 1 + a(pm )X m = (1 Cj X) with Bk = C .
m=1 j=1
k j=1 j

Now


|Cj | = |a(p )| p/2 ,
j=1
 6.1 Denition and First Properties 115

and thus the maximum of the values |Cj | is greater than or equal to p1/2 . We
have  1/k


 1


lim |b(p )|
k 1/k 
= lim  k
Cj  = max |Cj |;
k k  k  1j
j=1

by our foregoing observations the right-hand side is greater than or equal to

p1/2 . This is a contradiction to the condition b(pk )  pk with some < 12 in
the axiom on the Euler product. Hence, q = 1 and L(s) 1. This proves the
theorem. 


By the work of Kaczorowski and Perelli [161], it is known that the functions
of degree one in the Selberg class are the Riemann zeta-function and shifts
L(s + i, ) of Dirichlet L-functions attached to primitive characters with
R; parts of this classication are already contained in Bochners extension
of Hamburgers theorem [27] and in subsequent work of Grardin and Li [99]
and Piatetski-Shapiro and Rhagunathan [294] on converse theorems. How-
ever, for higher degree there is no complete classication so far. Examples
of degree two are L-functions associated with holomorphic newforms f ; here
the notion normalized means that a(n) = c(n)n(1k)/2 where the c(n) are
the Fourier coecients of f (as already indicated in Sect. 2.2). Normalized
L-functions attached to non-holomorphic newforms are expected to lie in
S but the Ramanujan hypothesis is not yet veried. The RankinSelberg
L-function of any normalized eigenform is an element of the Selberg class of
degree 4. Other examples are Dedekind zeta-functions to number elds K;
their degree is equal to the degree of the eld extension K/Q.
In view of the Euler product representation, it is clear that any element
L(s) of the Selberg class does not vanish in the half-plane of absolute con-
vergence > 1. This gives rise to the notions of critical strip and critical
line. The zeros of L(s) located at the poles of gamma-factors appearing in the
functional equation are called trivial. They all lie in 0, and it is easily
seen that they are located at
k + j
s= with k N0 and 1 j f. (6.6)
j

All other zeros are said to be non-trivial. In general, we cannot exclude the
possibility that L(s) has a trivial zero and a non-trivial one at the same point.
It is expected that for every function in the Selberg class the analogue of the
Riemann hypothesis holds.

Grand Riemann Hypothesis. If L S, then L(s) = 0 for > 12 .

The zero-distribution is essential for the Selberg class. This is also manifested
in the dening axioms. Following Conrey and Ghosh [59] and Kaczorowski
and Perelli [160], we motivate the axioms dening S. We have already seen
116 6 The Selberg Class

that the Ramanujan hypothesis implies the regularity of L(s) in > 1.

Further, we note:
The assumption that there be at most one pole, and that this one is located
at s = 1, is natural in the theory of L-functions. It seems that the point
s = 1 is the only possible pole for an automorphic L-function and that such
a pole is always related to the simple pole of the Riemann zeta-function in
the sense that the quotient with an appropriate power of (s) is another
L-function which is entire (examples for this scenario are Dedekind zeta-
functions).
The restriction Re j 0 in the functional equation comes from the
theory of Maass waveforms. If we assume the existence of an arithmetic
subgroup of SL2 (R) together with a Maass cusp form that corresponds
to an exceptional eigenvalue, and if we further suppose that all local
roots are suciently small (more precisely, that the RamanujanPetersson
conjecture holds), then the L-function associated with the Maass cusp
form has a functional equation where the j satisfy Re j < 0, but this
L-function violates Riemanns hypothesis.
Finally, consider the axiom concerning the Euler product. It is well-known
that the existence of an Euler product is a necessary (but not sucient)
condition for Riemanns hypothesis. On rst sight, the condition < 12
seems to be a little bit unnatural. However, for = 12 , there are examples
violating the Riemann hypothesis: the function


(1)n1
(1 21s )(s) = (6.7)
n=1
ns

has zeros o the critical line. Moreover, as we have seen in the proof of
Theorem 6.1, the bound for rules out non-trivial Dirichlet polynomials
from S as for example

(1 2as )(1 2bs ) with a + b = 1.

Some words of warning: It might turn out that the axioms of the Selberg class
are too restrictive; e.g., the form of the functional equation with the condition
on the j s to have positive real part might exclude relevant L-functions.

6.2 Primitive Functions and the Selberg Conjectures

The Selberg class is multiplicatively closed. A function L S is called prim-
itive if it cannot be factored as a product of two elements non-trivially, i.e.,
the equation
L = L1 L2 with L1 , L2 S,
implies L = L1 or L = L2 . The notion of a primitive function is fruitful for
studying the structure of S. Conrey and Ghosh [59] proved
 6.2 Primitive Functions and the Selberg Conjectures 117

Theorem 6.2. Every function in the Selberg class has a factorization into
primitive functions.

Proof. Suppose that L is not primitive, then there exist functions L1 and L2
in S \ {1} such that L = L1 L2 . Taking into account (6.3), from

NL (T ) = NL1 (T ) + NL2 (T )

we nd dL = dL1 + dL2 . In view of Theorem 6.1, both L1 and L2 have degree

at least 1. Thus, each of dL1 and dL2 is strictly less than dL . A continuation
of this process terminates since the number of factors is dL , which proves
the claim. 


In connection with Theorem 6.1 it follows that Riemanns zeta-function

and Dirichlet L-functions are primitive. A more advanced example of primitive
elements are L-functions associated with newforms is due to M.R. Murty [269].
On the contrary, Dedekind zeta-functions to cyclotomic elds = Q are not
primitive.
Denote by aL (n) the coecients of the Dirichlet series representation of
L S. The central claim concerning primitive functions is part of

Selbergs Conjectures.
(A) For all 1 = L S there exists a positive integer nL such that

|aL (p)|2
= nL log log x + O(1).
p
px

(B) For any primitive functions L1 and L2 S,

aL (p)aL (p)  log log x + O(1) if L1 = L2 ,
1 2
=
p O(1) otherwise.
px

In some sense, primitive functions are expected to form an orthonormal

system.
In view of the factorization into primitive functions, it is easily seen that
Conjecture B implies Conjecture A. In some particular cases, it is not too
dicult to verify Selbergs Conjecture A. For instance, (s) satises Selbergs
Conjecture A, which is basically due to Euler [76] who already wrote
1 1 1
+ + + = log log
2 3 5
(the rst complete proof was given by Mertens [247]). A stronger asymptotic
formula is (5.10) which plays an essential role in Bagchis proof of Voronins
universality theorem. By the prime number theorem for arithmetic progres-
sions (1.29), we get the same asymptotics for Dirichlet L-functions. Taking
into account the orthogonality relations for characters, one can also verify
118 6 The Selberg Class

Conjecture B for pairs of Dirichlet L-functions. The RankinSelberg convo-

lution method shows that L-functions associated with holomorphic modular
forms satisfy some kind of orthogonality (in terms of regularity at s = 1)
which is related to Selbergs conjectures. Recently, Liu, Wang and Ye [225]
proved Conjecture B for automorphic L-functions L(s, ) and L(s,  ), where
and  are automorphic irreducible cuspidal representations of GLm (Q) and
GLm
(Q), respectively; their result holds unconditionally for m, m 4 and in
other cases under the assumption of the convergence of

|a (pk )|2
(log p)2
p
pk

for k 2, where the a (n) denote the Dirichlet series coecients of L(s, ).
The latter hypothesis is an immediate consequence of the Ramanujan hypoth-
esis. (We shall return to these L-functions in Chap. 13.9.)
Selberg [323] wrote about his conjectures that these conjectures, which,
by the way, are not unrelated to several other conjectures like the SatoTate
conjecture, Langlands conjectures, etc., have been veried in a number of cases
for Dirichlet series with functional equation and Euler product that occur
in number theory, by assuming that the factorizations we can give are ac-
tually that a function is really primitive and cannot be factorized further.
For example, M.R. Murty [267] proved that Selbergs Conjecture B implies
Artins conjecture (we return to this topic in Chap. 13.7). Another important
consequence is due to Conrey and Ghosh [59].
Theorem 6.3. Selbergs conjecture B implies that every L S has a unique
factorization into primitive functions.
Proof. Suppose that L has two factorizations into primitive functions:

m 
n
L= Lj = Lk ,
j=1 k=1

and assume that no Lk is equal to L1 . Then it follows from

m

n
aLj (p) = aLk (p)
j=1 k=1

that

m

aLj (p)aL1 (p)

aLk (p)aL1 (p)
n
= .
j=1
p p
px k=1 px

By Selbergs conjecture B, the left-hand side tends to innity for x ,

whereas the right-hand side is bounded, giving the desired contradiction. 

The same argument gives a characterization of primitive functions in terms
of the quantity nL from Selbergs conjecture A:
 6.3 Non-Vanishing and Prime Number Theorems 119

Theorem 6.4. If the Selberg conjecture B is true, L S is primitive if and

only if nL = 1.

Selberg also considered twists with characters. If is a primitive Dirichlet

character, and L S is automorphic, then


a(n)
L (s) = (n)
n=1
ns

satises the axioms of S too. It can be shown that then L and L are either
both primitive or both not primitive (since they have the same nL ). Selbergs
twisting conjecture states that if L is not primitive, L and L will factor in the
same way into primitive factors, in the sense that one gets the factorization
of L by twisting all factors of L by .

6.3 Non-Vanishing and Prime Number Theorems

Furthermore, Selberg studied moments and distribution functions for L S
in order to prove an analogue of (1.16). For this aim, he had to assume some
unproved hypothesis on the distribution of zeros which hold, or are at least
expected to hold, for all known examples of elements in S. Let NL (, T ) count
the number of zeros  = + i of L(s) with > and || < T (counting
multiplicities). Kaczorowski and Perelli [163] proved

Theorem 6.5. For every L S,

NL (, T )  T 4( dL +3)(1)+
,

uniformly for 1
2 1.

Unfortunately, this estimate is only useful for close to 1. For smaller we

note the

Grand Density Hypothesis. For L S there is some positive constant c

such that for > 12
NL (, T )  T 1c(1/2)+
.

Selberg proved that if L S satises the Grand density hypothesis and Con-
jecture A, then the values of
 
log L 12 + it

nL log log t
are distributed in the complex plane according to the normal distribution.
Furthermore, he investigated the value-distribution of linear combinations of
independent elements of S. Selbergs argument was streamlined and extended
120 6 The Selberg Class

by Bombieri and Hejhal [39] to independent collections of L-functions having

polynomial Euler products with the emphasis just on probabilistic conver-
gence and the goal of applications to the zero-distribution. For this aim they
introduced a strong version of Selbergs conjecture B.
Strong Selbergs Conjecture B. For any primitive functions L1 and L2
in S there exist constants CL1 , CL1 ,L2 such that

aL (p)aL (p)  log log x + CL + R(x) if L1 = L2 ,

1 2 1
=
p CL1 ,L2 + R(x) otherwise,
px

where R(x)  1
log x .

Their theorem shows the statistical independence of any collection of indepen-

dent L-functions in any family of elements of S. Furthermore, Bombieri and
Hejhal applied their result to the zero-distribution of linear combinations of
independent L-functions. Assuming in addition the Grand Riemann hypoth-
esis and a weak conjecture on the well-spacing of the zeros, they proved that
almost all zeros of these linear combinations are simple and lie on the critical
line. Bombieri and Perelli [40] considered for the same class of functions the
distribution of distinct zeros. They proved that, for two dierent functions
L1 , L2 of the same degree,


max{mL1 () mL2 (), 0}  T log T, (6.8)
0T

where the sum is taken over the non-trivial zeros  = + i of L1 L2 (s) and
mLj () denotes the multiplicity of the zero  of Lj (s).
The Selberg conjectures refer to the analytic behaviour at the edge of the
critical strip. Conrey and Ghosh [59] proved the non-vanishing on the line
= 1 subject to the truth of Selbergs conjecture B:

Theorem 6.6. Let L S. If Selbergs conjecture B is true, then

L(s) = 0 for 1.

It is conjectured that the Selberg class consists only of automorphic L-

functions, and for those Jacquet and Shalika [148] obtained an unconditional
non-vanishing theorem.

Proof of Theorem 6.6. In view of the Euler product representation, in the

half-plane 1 zeros can only occur on the line = 1. By Theorem 6.2 it
suces to consider primitive functions L S. In case of (s), it is known that
there are no zeros on = 1. It is easily seen that if Selbergs Conjecture B is
true and if L S has a pole at s = 1 of order m, then the quotient L(s)/(s)m
is an entire function (this can be shown by the argument from the proof of
 6.3 Non-Vanishing and Prime Number Theorems 121

Theorem 6.3). Hence, we may assume that L(s) is entire. Then, for any real
, the function L(s + i) is a primitive element of S. Selbergs conjecture B
applied to L(s + i) and (s) yields

aL (p)
 1. (6.9)
p1+i
px

Now suppose that L(1 + i) = 0. Then

L(s) c(s (1 + i))k

as s = + i 1 + i for some complex c = 0 and some positive integer k.

It follows that:

log L( + i) k log( 1) (6.10)

as 1+. Since

aL (p)
log L(s) = + O(1)
p
ps

for > 1, we get by partial summation

aL (p) 
aL (p) dx
log L( + i) = ( 1) .
p
p+i 1 p1+i x
px

By (6.9) the right-hand side is bounded as 1+, which contradicts (6.10).

The theorem is proved. 


Recently, Kaczorowski and Perelli [163] investigated the analogue of the prime
number theorem for the Selberg class. It is well known that the non-vanishing
of (s) on the edge of the critical strip is equivalent to the (standard) prime
number theorem (without remainder term). For an arbitrary element L S,
the corresponding prime number theorem is given in form of the asymptotic
formula

L (x) := L (n) = kL x + O(x); (6.11)
nx

here kL = 0 if L(s) is regular at s = 1, otherwise kL is the order of the pole

of L(s) at s = 1, and L (n) is the von Mangoldt-function, dened by

L L (n)
(s) = ,
L n=1
ns

generalizing (1.7), the case of the Riemann zeta-function. Indeed, for poly-
nomial Euler products in the Selberg class one can prove this equivalence by
standard arguments involving an appropriate Tauberian theorem. In this case,
by Theorem 6.6, Selbergs conjecture B implies (6.11). However, this is not
122 6 The Selberg Class

satisfying with respect to S. Kaczorowski and Perelli [163] introduced a weak

form of Selbergs conjecture A:
Normality Conjecture. For all 1 = L S there exists a non-negative
integer kL such that

|aL (p)|2
= kL log log x + O(log log x).
p
px

Assuming this hypothesis, they proved the claim of Theorem 6.6, namely
the non-vanishing of any L(s) on the line = 1 and that this statement is
equivalent to the prime number theorem (6.11). It should be noted that their
proof of L(1 + iR) = 0 for a given L involves the assumption of the normality
conjecture for several elements in S.

6.4 Pair Correlation

Assuming the truth of the Riemann hypothesis, Montgomery [259] studied the
distribution of consecutive zeros 12 + i, 12 + i  of the Riemann zeta-function.
Montgomerys famous pair correlation conjecture states that, for xed ,
satisfying 0 < < ,
 
1 (  ) log T
lim  0 < ,  < T :
T N (T ) 2
   2 
sin u
= 1 du. (6.12)
u

Montgomery claims that (6.12) would follow from a suciently good estimate
for

(n)(n + h) c(h)x
nx

in a certain range of h, where c(h) is some quantity depending on h; however,

the HardyLittlewood twin prime conjecture [117] is too strong of an input
into this problem. Heath-Brown [123] obtained for the dicult related problem
of the distribution of dierences pn+1 pn of consecutive prime numbers under
assumption of the pair correlation conjecture the estimate


(pn+1 pn )2  x(log x)2 .
pn x

The pair correlation conjecture has many other important consequences. For
instance, (6.12) implies that almost all zeros of the zeta-function are simple
(actually, this is related to the asymptotic formula (1.12)).
Dyson remarked that the function on the right of (6.12) is the pair cor-
relation function of the eigenvalues of large random Hermitian matrices, or
 6.4 Pair Correlation 123

more specically of the Gaussian Unitary Ensemble. This supports an old idea
of Hilbert and Plya. Their approach towards Riemanns hypothesis was to
look for a self-adjoint linear operator of an appropriate Hilbert space whose
eigenvalues include the zeros of
1 s

(t) = s(s 1) s/2 (s)

2 2
with s = 12 + it. Since (t) is an entire function which vanishes exactly at the
non-trivial zeros  of (s), the property of self-adjointness would imply that
all zeros  lie on the line = 12 . In the last years, remarkable progress in
this direction was made. By the work of Odlyzko [283], it turned out that the
pair correlation and the nearest neighbour spacing for the zeros of (s) were
amazingly close to those for the Gaussian Unitary Ensemble. There is even
more evidence for the pair correlation conjecture than numerical data. In the
meantime many results from random matrix theory were found tting per-
fectly to certain results on the value-distribution of the Riemann zeta-function
(and even other L-functions; see Conreys survey article [58]). For example,
Keating and Snaith [168] showed that certain random matrix ensembles have
in some sense the same value-distribution as the zeta-function on the criti-
cal line predicted by Selbergs limit law (1.16). Further evidence for the pair
correlation conjecture was discovered by Rudnick and Sarnak. Normalize the
ordered non-trivial zeros n = 12 + in by setting
n
n = log |n |,
2
then it follows from the Riemannvon Mangoldt formula (1.5) that the num-
bers n have unit mean spacing. The pair correlation conjecture (6.12) can be
rewritten as follows: for any nice function f on (0, )


lim f (n+1 n ) = f (x)P (x) dx,
N 0
nN

where P is the distribution of consecutive spacings of the eigenvalues of a large

random Hermitean matrix. Rudnick and Sarnak [314] succeeded in showing
that the m-dimensional analogue of the latter formula, the m-level correlation,
holds for a large class of test functions. Finally, note that Katz and Sarnak
[167] proved a function eld analogue of Montgomerys pair correlation con-
jecture without assuming any unproved hypothesis.
Recently, Murty and Perelli [271] extended Montgomerys argument to
the Selberg class. For this purpose, they considered two primitive functions
L1 and L2 from S. To compare the zeros 12 + iL1 of L1 against the zeros
2 + iL2 of L2 dene
1

F(; L1 , L2 ) = T i dL1 (L1 L2 ) w(L1 L2 ),
dL1 T log T
T L1 ,L2 T
124 6 The Selberg Class

where w is a suitable weight function. The pair correlation conjecture for the
Selberg class takes then the form:
Pair Correlation Conjecture. Let L1 and L2 be primitive functions in S.
Under the assumption of the Grand Riemann hypothesis, uniformly in , as
T ,

L1 ,L2 || + dL1 T 2|| dL1 log T (1 + O(1)) if || < 1 ,
F(; L1 , L2 )
L1 ,L2 || otherwise.
Here 
1 if L1 = L2 ,
L1 ,L2 :=
0 otherwise,
is the Kronecker-symbol. This conjecture includes Montgomerys pair corre-
lation conjecture. It has plenty of important applications as M.R. Murty and
Perelli [271] worked out. For instance, the Artin conjecture follows from the
pair correlation conjecture. Further, the pair correlation conjecture implies
that almost all zeros of two dierent primitive functions L1 and L2 are sim-
ple and distinct. Moreover, if the pair correlation formula holds for at least
one value of , then S has unique factorization into primitive functions. This
shows what a powerful tool pair correlation is. Further, M.R. Murty and Perelli
proved that the Grand Riemann hypothesis and the pair correlation conjec-
ture imply the Selberg conjectures. The pair correlation conjecture plays a
complementary role to the Riemann hypothesis: vertical vs. horizontal distri-
bution of the non-trivial zeros of (s). Both together seem to be the key to
several unsolved problems in number theory!
In the sequel, we investigate universality for L-functions in the Selberg
class. We shall show how our general universality theorem for S extends to a
certain class of polynomial Euler products in S. We expect that it holds for
all functions L S \ {1}. Moreover, we can extend the strip of universality
signicantly on behalf of the functional equation.

6.5 The PhragmnLindelf Principle

The order of growth of a meromorphic function is of special interest. Recall
our observations on the order of growth of Dirichlet series from Sect. 2.1. If
the Dirichlet series under investigation satises a functional equation, we can
obtain more information about the mean-square half-plane. For L S we
dene
log |L( + it)|
L () = lim sup .
t log |t|
One can show that L () is a convex function of . Taking into account the
absolute convergence of the dening Dirichlet series we obtain immediately
L () = 0 for > 1. The order of growth in the half-plane to the left of the
critical strip is ruled by the functional equation which we may rewrite as
L(s) = L (s)L(1 s), (6.13)
 6.5 The PhragmnLindelf Principle 125

where
f
(j (1 s) + j )
L (s) := Q12s .
j=1
(j s + j )
Applying Stirlings formula (2.17), we get after a short computation
Lemma 6.7. Let L S. For t 1, uniformly in ,
   
 2 dL 1/2it i( dL ) 1
L ( + it) = Q t exp it dL + +O ,
4 t
where


f 
f
2
:= 2 (1 2j ) and := j j .
j=1 j=1

Using the so-called PhragmnLindelf principle, we can obtain upper

bounds for the order of growth inside the critical strip.
Theorem 6.8. Let L S. Uniformly in , as |t| ,
L( + it) |t|(1/2) dL |L(1 + it)|.
In particular,
0 if > 1,
L () dL (1 )
1
if 0 1,
1 2
( 2 ) dL if < 0.
Proof. The rst assertion follows immediately from the functional equation
and Lemma 6.7. This together with the trivial estimate L () = 0 for > 1
implies for < 0  
1
L () = dL .
2
The calculation of L () for 0 1 is more dicult. We have to apply a
kind of maximum principle for unbounded regions, the theorem of Phragmn
Lindelf.
Lemma 6.9. Let f (s) be analytic in the strip 1 2 with f (s) 
exp(|t|). If
f (1 + it)  |t|c1 and f (2 + it)  |t|c2 ,
then f (s)  |t|c() uniformly in 1 2 , where c() is linear with
c(1 ) = c1 and c(2 ) = c2 .
A proof can be found in the paper of Phragmn and Lindelf [221] or, for
example, in Titchmarsh [352]. Note that there are counterexamples if the
growth condition f (s)  exp(|t|) is not fullled.
We continue with the proof. In view of the axiom concerning the analytic
continuation L(s) is a function of nite order. Thus, Lemma 6.9 shows that
L () is non-increasing and convex downwards. By the estimates of L () for
outside of the critical strip the second assertion of the theorem follows. 

126 6 The Selberg Class

It should be noticed that we did not use the condition that the j appearing
in the gamma factors of the functional equation have positive real part. Thus,
if it will turn out that this condition is too restrictive, it does not inuence
the statement of Theorem 6.8 or what we will deduce from it.
In view of the functional equation, respectively, the convexity of L (),
the value for = 12 is essential. In particular, we obtain L ( 12 ) 14 dL , or
equivalently,
 
1
L + it  |t|(dL /4)+
(6.14)
2

for |t| 1; this bound is known as the convexity bound. The best known
upper bound for the Riemann zeta-function is ( 12 ) 20532
, due to Huxley
[138].
Next, we shall apply the following renement of Carlsons classic theorem
on the mean-square of Dirichlet series due to Potter [296].
Theorem 6.10. Suppose that the functions



an bn
A(s) = and B(s) =
n=1
ns n=1
ns

have a half-plane of convergence, are of nite order, and that all singularities
lie in a subset of the complex plane of nite area. Further, assume that



|an |2  xb+
and |bn |2  xb+
,
nx nx

as x , and that A(s) and B(s) satisfy

A(s) = h(s)B(1 s),

where h(s) |t|c(a/2) uniformly in for from a nite interval, as |t| ,

and a, b, c are some non-negative constants. Then
 T

1 |an |2
lim |A( + it)|2 dt =
T 2T T
n=1
n2

for > max{ a2 , 12 (b + 1) 1c }.

Taking into account (2.12) and Theorem 6.8, we may take a = b = 1 and
c = dL . We obtain
 
Corollary 6.11. Let L S. For > max 12 , 1 d1L ,


1 T
|a(n)|2
lim |L( + it)|2 dt = .
T 2T T n=1
n2
 6.5 The PhragmnLindelf Principle 127

Note that the series on the right-hand side converges on behalf of the
Ramanujan hypothesis (1). Further, we remark that the statement of the
corollary holds also for L-functions which do not fulll axiom (4) from the
denition of S, since we have used (2.12) and not Lemma 2.3. Corollary 6.11
should be compared with the mean-square estimates of Perelli [289] for his
class of L-functions; his argument is dierent.
Corollary 6.11 improves the range for the existence of the mean-square
given by Theorem 2.4, provided L(s) satises a Riemann-type functional equa-
tion: the mean-square half-plane of L then contains the region
 
1 1
> max ,1 .
2 dL

It would be desirable to extend this region further to the half-plane > 12

in general. However, this is hard. The diculties arise for large degrees dL .
For instance, Chandrasekharan and Narasimhan [52] obtained for Dedekind
zeta-functions the estimate
 T
|K ( + it)|2 dt  T d(1) (log T ) d
0

for 12 1 1d , where d is the degree of K (s). Potters theorem only yields

an asymptotic formula throughout > 12 if d 2. The diculties become
more obvious by noting that any result on the mean-square of an L-function
from the Selberg class of degree d is comparable to the corresponding result
for the 2dth moment of the Riemann zeta-function. The fourth moment of
(s) on the critical line is quite well understood (by the work of Hardy and
Littlewood [118], Ingham [139], Motohashi [263] and others), however, higher
moments are still unsettled; there are only estimates known. Similar problems
appear to the right of the critical line. For example, Ivi showed [141] proved
for 58 < < 1
 T
|( + it)|8 dt  T (118)/6+
. (6.15)
1

Recently some new insight was obtained by the analogies to random matrix
theory. Extending a conjecture of Conrey and Gonek [61], Keating and Snaith
[168] conjectured
   2k
1 T  
 1 + it  dt c(k)(log T )k2 ,
T  2 
0

where
 k 2
 2
G2 (k + 1)  1 (m + k)
c(k) := 1 2 pm
G(2k + 1) p p m=0
m! (k)
128 6 The Selberg Class

with the Barnes double Gamma-function G(z), dened by

   
1 z
n z2
G(z + 1) = (2) z/2
exp (z(z + 1) + z )
2
1+ exp z + ,
2 n=1
n n

and is the EulerMascheroni constant. For more details on mean-value

results and random matrix theory we refer to Ivis monograph [141],
Matsumotos survey [240], and Conreys paper [58].

6.6 Universality in the Selberg Class

Now we are going to apply the universality Theorem 5.14 to L-functions in
the Selberg class. First of all we compare the classes S and S. The axioms (i)
and (1) are identical. Concerning analytic continuation we note that axiom (2)
implies (ii) with any L < 1. Moreover, any L(s) satisfying (2) is a function
of nite order and together with the functional equation (3) we obtain (iii)
with L = L () (by Theorem 6.8). Only the arithmetic axioms (iv) and (v)
cannot be deduced from the axioms of the Selberg class; however, they are
expected to hold. As already mentioned, the elements in the Selberg class
are automorphic or at least conjecturally automorphic L-functions, and it is
conjectured that S consists of all automorphic L-functions. For all known
examples, the Euler product has the form of axiom (iv) in the denition of
S: there exists a positive integer m and for each prime p, there are complex
numbers j (p) such that
m 
 1
j (p)
L(s) = 1 .
p j=1
ps

The remaining axiom of S is (v) which states that

1

lim |a(p)|2 =
x (x)
px

for some positive constant . This axiom is closely related to Selbergs conjec-
tures. In fact, it implies a weak version of Selbergs conjecture A. By partial
summation,

|a(p)|2  x

1
du
= |a(p)| +
2
|a(p)|2 2
p x 2 pu u
px px
 x
du
log log x;
2 log u

this is the normality conjecture of Kaczorowski and Perelli [163] for the single
function L(s) (provided N which we do not require); indeed, the full
 6.6 Universality in the Selberg Class 129

normality conjecture is equivalent to the validity of axiom (v) for the whole
of S.1 Conversely, the strong Selberg Conjecture B with a remainder term
 
1
R(x) = O
log x
implies axiom (v) on the prime mean square with = nL . Moreover, from
this strong version it would follow that S S is multiplicatively closed.
Now we can formulate a rather general universality theorem for L-functions
in the Selberg class.
Theorem 6.12. Let L S S (i.e., L(s) satises the axioms (2), (3), (iv),
and (v)), let K be a compact subset of the strip m < < 1 (where m
max{ 12 , 1 d1L } < 1) with connected complement, and let g(s) be a non-
vanishing continuous function on K which is analytic in the interior of K.
Then, for any  > 0,
 
1
lim inf meas [0, T ] : max |L(s + i ) g(s)| <  > 0.
T T sK

Obvious examples for elements in S S are the Riemann zeta-function and

Dirichlet L-functions L(s, ) to primitive characters . Further examples of
elements are normalized L-functions associated with holomorphic newforms,
Dedekind zeta-functions, RankinSelberg L-functions. This can be shown by
using analogues of the prime number theorem (e.g., the prime ideal theorem).
If one is willing to accept some widely believed conjectures, then many L-
functions belong to this class. However, later we shall prove universality for
Dirichlet L-functions to imprimitive characters which are not elements of the
Selberg class (by lack of having the appropriate functional equation) and
further L-functions which are conjectured to lie in S. This marks an important
advantage of dealing with the class S rather than the Selberg class.
We shall briey discuss some renements of the universality Theorem 6.12
with respect to the strip of universality. In some particular cases of functions
L S with degree dL > 2 the existence of the mean-square

1 T
lim |L( + it)|2 dt
T T 1

is known for some 1 d1L . For instance, let L(s, ) be an arbitrary Dirichlet
L-function to a primitive character . Then (s)2 L(s, ) is an element of S
of degree 3, so Theorem 6.12 gives universality for 23 < < 1. Montgomery
[258] proved that


 T  1 4

L + it,  dt  (q)T (log(qT ))4 .
 2 
mod q T
primitive

1
This remark is due to Giuseppe Molteni; see MathReviews MR1981179
(2004c:11165). Selbergs conjecture A for a single L-function seems to be insu-
cient for a proof of universality (by lack of a suciently good error term).
130 6 The Selberg Class

It is easily seen that the same bound holds for the fourth moment on vertical
lines to the right as well. In combination with Ivis eigth-moment estimate
(6.15), the CauchySchwarz inequality yields
 T
|( + it)2 L( + it, )|2 dt
1
  1/2
T T
 |( + it)| dt
8
|L( + it, )| dt
4
1 1

 T (118)/12+
T 1/2+
 T

for any > 58 with suciently small . Thus (s)2 L(s, ) is universal in the
strip 58 < < 1. In this example, we can even do better by applying the joint
universality Theorem 1.10 which improves the range to 12 < < 1.
In Sect. 6.7, we show that the strip of universality for L S S can be
extended subject to a growth condition for L(s) on the critical line.

6.7 Lindelf s Hypothesis

For many applications in number theory, it is useful to assume Riemanns
hypothesis but quite often it suces to work with weaker conjectures.
Lindelf [220] conjectured that the order of growth of the zeta-function
is much smaller than the one the PhragmnLindelf principle gives. More
precisely, he expressed his belief that (s) is bounded if 12 +  for any
xed positive . In terms of the -function from Sect. 6.5 this would imply
( 12 ) = 0 or equivalently  
1
+ it  t

2
as t . The last statement is now known as Lindelfs hypothesis and it is
yet unproved. However, the boundedness conjecture is false as one can easily
deduce, for example, from Voronins universality theorem. It was proved by
Littlewood [223] that the Lindelf hypothesis follows from the truth of the
Riemann hypothesis. On the contrary, Backlund [8] proved that the Lindelf
hypothesis is equivalent to the much less drastic but yet unproved hypothesis
that for every > 12

N (, T + 1) N (, T ) = O(log T ). (6.16)

Furthermore, the Lindelf hypothesis implies the classic density hypothesis

which claims

N (, T )  T 2(1)+
. (6.17)

This should be compared with the density Theorems 1.2 and 6.5.
 6.7 Lindelfs Hypothesis 131

In the case of the Lerch zeta-function (1.41) the analogue of the density
hypothesis is not true in general.
 For instance, the function given by (6.7) can
be rewritten as L 12 , 1, s , and obviously it has innitely many zeros o the
critical line = 12 , at least the zeros s = 1 log 2 k with k Z\{0}, coming from
2i

1s
the factor 12 . This violates the analogue of the Riemann hypothesis and,
moreover, it violates the analogue of (6.17) as well. On the other side L( 12 , 1, s)
satises the analogue of the Lindelf hypothesis provided (s) satises the
Lindelf hypothesis. This example does not seem to be special. It is known
that a generic Lerch zeta-function has many zeros o the critical line. Denote
by  = + i the non-trivial zeros of L(, , s) (where the notion non-trivial
is dened quite similarly as for (s)). Garunktis and Steuding [91] (see also
[89]) proved, for 0 < , 1,

 1

T
= log + O(log T )
||T
2 2 +

(for the denition of the quantities + and see (1.43)). Consequently, any
L(, , s) with 2 > + has  T many zeros in the region > 12 , 0 < t
T . The proportion of the parameters , (0, 1] which satisfy 2 > + is
equal to  1%

1 (1 ) d = 1 .
0 8
Since most of these zeros o the critical line could lie arbitrarily close to = 12 ,
this alone does not violate the analogue of the density hypothesis for the Lerch
zeta-function. However, Garunktis [85] proved that for transcendental the
number of zeros  = + i of L(, , s) with > , || T is  T for
any xed ( 12 , 1]; the proof relies on the strong universality property of
L(, , s) (see also Garunktis and Laurinikas [89]). This indicates that if
L(, , s) has no Euler product representation, the density hypothesis does
not follow from the Lindelf hypothesis. On the other hand, Garunktis and
Steuding [92] showed that the analogue of the Lindelf hypothesis for Lerch
zeta-functions seems reasonable.
In [60] Conrey and Ghosh generalized the Lindelf hypothesis to the
Selberg class. They proved that if L S is an entire function which satis-
es the Riemann hypothesis, the Euler product is of the form (iv), and that
all j appearing in the functional equation are equal to 12 , then there exists a
constant C, depending only on  > 0 and the degree dL , such that


  
  f
L 1 + it  C Q(1 + |t|) dL /2 (1 + |j |) .
 2 
j=1

The proof follows the lines of Littlewoods proof that the Riemann hypo-
thesis implies the Lindelf hypothesis (in fact, it relies on a combination
of the PhragmnLindelf principle, the BorelCarathodory theorem and
132 6 The Selberg Class

Hadamards three circle theorem). In view of the PhragmnLindelf princi-

ple one can prove unconditionally
  
 
L 1 + it   (cQ2 (1 + |t|) dL )(1/4)+

 2 

for some positive constant c = c(, ) depending only on the data of the
functional equation. We already obtained with (6.14) a similar bound in the
t-aspect. The so-called subconvexity problem is to nd a > 0 such that
the exponent on the right-hand can be replaced by 14 . For certain L-
functions subconvexity bounds are known but in general, this seems to be a
dicult problem. However, solutions of the subconvexity problem (in the Q-
aspect) lead to several important applications in number theory, e.g., Hilberts
eleventh problem on the representation of integers by a given quadratic form;
we refer for details and further examples to the survey of Iwaniec and Sarnak
[145]. Here, we are only interested in the t-aspect.
Grand Lindelf Hypothesis. For L S and any  > 0,
 
1
L + it  |t|
f or |t| 1. (6.18)
2
The implicit constant may depend on  and L. This hypothetical estimate
coincides with Perellis Lindelf hypothesis for his class of L-functions [289].
Among other interesting results, Perelli showed for his L-functions that
Riemanns hypothesis implies the Lindelf hypothesis, and that Backlunds
reformulation (6.16) of the Lindelf hypothesis in terms of their zero-
distribution o the critical line holds. We observe that the Lindelf hypothesis
(6.18) for L(s) is equivalent to
 
1
L () = dL max 0, .
2
There are several further interesting reformulations of the Lindelf hypoth-
esis in the particular case of the Riemann zeta-function. One, given in terms
of moments on the critical line, was found by Hardy and Littlewood [119].
They proved that the Lindelf hypothesis is true if and only if for any k N
   2k
1 T  1 
 dt  T
.
 + it  (6.19)
T 1 2
Another equivalent form is due to Laurinikas [195]. He proved that the
Lindelf hypothesis is equivalent to the asymptotic formula
      
1  1  (T )
meas t [0, T ] :  + it  < xT
= 1 O
T 2 1 + xA
with positive , A and suciently large x, where (T ) is an arbitrary positive
function which tends to zero as T .
 6.8 Symmetric Power L-Functions 133

Now we consider the mean-square on vertical lines to the right of the

critical line subject to the Lindelf hypothesis. For this purpose we return
to Sect. 2.4. Assuming the Lindelf hypothesis, we may take L = 12 and
L = L ( 12 ) = 0 in Theorem 2.4. Thus, for L S S, the Lindelf hypothesis
implies


1 T |a(n)|2
lim |L( + it)|2 dt =
T T 1
n=1
n2
for any > 12 . In fact, there are no big diculties to consider also higher
moments, and also to derive an equivalent statement for Lindelfs hypothesis
analogous to (6.19).
If the Lindelf hypothesis is true, then the strip of universality can be
extended to the open right half of the critical strip.
Corollary 6.13. Assume that the Lindelf hypothesis is true for L S S.
Let K be a compact subset of the strip 12 < < 1 with connected complement,
and let g(s) be a non-vanishing continuous function on K which is analytic in
the interior of K. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max |L(s + i ) g(s)| <  > 0.
T T sK

In view of results on the frequency of the values taken in neighbourhood

of the critical line (see Theorem 7.6) it seems to be impossible to have univer-
sality (in the sense of Voronins theorem) in any region covering the critical
line.

6.8 Symmetric Power L-Functions

We conclude this chapter with an example. Symmetric power L-functions
became important by Serres reformulation of the SatoTate conjecture. For
the sake of simplicity, here we shall only consider the case where f is a
normalized eigenform of weight k to the full modular group. According to
the normalization from Sect. 2.2, we suppose that its Fourier expansion is
given by



f (z) = a(n)n(k1)/2 exp(2inz)
n=1
and the attached L-function has an Euler product representation of the form
 a(p) 1
1
L(s, f ) = 1 s + 2s .
p
p p

In view of Delignes estimate (1.34), for each prime p, we may dene an angle
p [0, ] by setting
a(p) = 2 cos p ; (6.20)
134 6 The Selberg Class

this should be compared with the angles dened in equation (5.20). Then, for
any non-negative integer m the symmetric mth power L-function attached to
f is for > 1 given by
m 
 1
exp(ip (m 2j))
Lm (s, f ) = 1 . (6.21)
p j=0
ps

Then
L0 (s, f ) = (s), L1 (s, f ) = L(s, f ),
and
(2s)
L2 (s, f ) = L(s, f f ),
(s)
where L(s, f f ) is the RankinSelberg L-function (introduced in Sect. 1.6).
Shimura [328] obtained the analytic continuation and a functional equation of
Riemann-type for the case m = 2. By the powerful methods of the Langlands
program, Shahidi [326] obtained the analytic continuation of Lm (s, f ) to 1
for m 4. In particular cases more is known; for example, Kim and Shahidi
[170] showed that certain third symmetric power L-function attached to non-
monomial cusp forms of GL2 over any number eld are entire, and Shahidi [327]
proved that the third and the fourth symmetric power L-function are cuspidal
unless the underlying cusp form is either of dihedral or tetrahedral type (in the
terminology of Kleins classication of nite subgroups of PGL2 (C)). However,
the case of m > 4 is open. Serre [324] conjectured that if f does not have
complex multiplication, the angles p are uniformly distributed with respect
to the SatoTate measure
2
(sin )2 d, (6.22)

if p ranges over the set of prime numbers (in analogy to a similar conjecture
for elliptic curves E due to Sato and Tate which was recently solved by
Taylor [351] assuming a mild condition on E). Furthermore, Serre proved
that the non-vanishing of Lm (s, f ) on the abscissa of convergence = 1 for
all m N implies the SatoTate conjecture for newforms, namely

1 2
lim {p x : < p < } = (sin )2 d. (6.23)
x (x)

Ogg [285] has shown that if for each m 2M , the function Lm (s, f ) has an
analytic continuation to a half-plane > 12 , then Lm (s, f ) does not vanish
on the line = 1. V.K. Murty [272] proved that it suces to have an analytic
continuation of any Lm (s, f ) to 1 for proving the SatoTate conjecture.
In this context, M.R. Murty [265] (see also [270]) proved asymptotic formulae
for the 2mth power moments of cos p .
 6.8 Symmetric Power L-Functions 135

Theorem 6.14. Let f be a normalized newform. If Lm (s, f ) has an analytic

continuation up to 12 for all m 2(M + 1), then, for m M + 1,
 
1
2m 1 2m
lim (2 cos p ) = ,
x (x) m+1 m
px

and, for m M ,
1

lim (2 cos p )2m+1 = 0.
x (x)
px

In view of Shimuras results, the asymptotic formulae of the theorem hold

unconditionally for M 1. This includes Rankins asymptotic formula (1.37).
If we expand the Euler product (6.21) into a Dirichlet series, the pth coecient
is given by


m
sin((m + 1)p )
ap := exp(ip (m 2j)) = = Um (cos p ),
j=0
sin p

where Um (x) is the Chebyshev polynomial of the second kind, dened by

U0 (x) = 1, U1 (x) = x and Um (x) = xUm1 (x) Um2 (x) for m 2. In view
of Theorem 6.14 it follows that:



|ap |2 = Um (cos p )2 (x)
px px

as x , where is a positive constant. Taking into account the functional

equation for Lm (s, f ) (see [219], resp. Cogdell and Michel [56]), it follows that
Lm (s, f ) is an element of degree m + 1 in the Selberg class S, unconditionally
for m 4, and conditionally for m > 4 (depending on the analytic continua-
tion which actually is a consequence of the Langlands conjectures). It is not
dicult to see that we also have Lm (s, f ) S under the same conditions.
Thus Theorem 5.14 yields
Corollary 6.15. Let f be a normalized newform, let K be a compact subset of
the strip max{ 12 , 1 m+1
1
} < < 1 with connected complement, and let g(s)
be a non-vanishing continuous function on K which is analytic in the interior
of K. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max |Lm (s + i, f ) g(s)| <  > 0,
T T sK

unconditionally if m 4, and for m 5 under the assumption of the holo-

morphy of Lm (s, f ) throughout C and the existence of a functional equation.
With slightly more eort, one can show that if f is a holomorphic newform
of some level N 1 and integer weight k 1 which is not of complex
multiplication type, then for m = 2, 3, 4 the mth symmetric power L-function
Lm (s, f ) is universal. Corollary 6.15 was independently obtained by Li and
Wu [219].
7
Value-Distribution in the Complex Plane

Une fonction entire, qui ne devient jamais ni a ni b est ncessaire-

ment une constante. Emile Picard

Many beautiful results on the value-distribution of L-functions follow from

the general theory of Dirichlet series like the Big Picard theorem (see Boas [26]
and Mandelbrojt [234]), but more advanced statements can only be proved
by exploiting the characterizing properties (the functional equation and the
Euler product). In this chapter, we study the distribution of values of Dirichlet
series satisfying a Riemann-type functional equation. These results are due to
Steuding [346, 347] and their proofs follow in the main part the methods of
Levinson [217], Levinson and Montgomery [218], and Nevanlinna theory.

7.1 Sums Over c-Values

Let c be any complex number. Levinson [217] proved that all but
 N (T )(log log T )1 of the roots of (s) = c in T < t < 2T lie in
 
  2
 1  < (log log T ) .
 2  log T

Thus, the c-values of the zeta-function are clustered around the critical line.
In particular, we see that the density estimate (1.13) alone does not indicate
the truth of the Riemann hypothesis. As we shall show in this chapter, this
distribution of c-values is typical for Dirichlet series satisfying a Riemann-type
functional equation.
Throughout this chapter, we shall assume that


a(n)
L(s) =
n=1
ns
138 7 Value-Distribution in the Complex Plane

satises the axioms (1)(3) from the denition of the Selberg class S, and so
we may dene the degree dL of L by (6.2); we shall not make use of axiom (4)
(neither do we use the condition on the real parts of the complex numbers j
in the Gamma-factors of the functional equation), however, for simplicity we
suppose that a(1) = 1. In some places we shall assume the Lindelf hypothesis
for L(s); by that we mean the estimate (6.18).
We give an example of a function satisfying these axioms which does not
have an Euler product. The DavenportHeilbronn zeta-function is given by

1 i 1 + i
L(s) = L(s, ) + L(s, ), (7.1)
2 2
where %
10 2 5 2
:=
51
and is the character mod 5 with (2) = i. It is easily seen that the
DavenportHeilbronn zeta-function satises the functional equation
 s/2    (1s)/2
5 s+1 5 s

L(s) = 1 L(1 s).

2 2

Davenport and Heilbronn [65] introduced this function as an example for a

Dirichlet series having zeros in the half-plane > 1 although L(s) satises a
Riemann-type functional equation; see Balanzario [14] for more examples of
a similar type.
The c-values of L(s) are the roots of the equation

L(s) = c, (7.2)

which we denote by c = c + ic . Our rst aim is to prove estimates for sums

taken over c-values, weighted with respect to their real parts.

Theorem 7.1. Assume that L(s) satises the axioms (1)(3) with a(1) = 1
and let c = 1. Then, for any b > max{ 12 , 1 d1L },


(c b)  T.
c >b
T <c 2T

Assuming the truth of Lindelf s hypothesis for L(s),

 
1
c = O(T log T ).
2
c > 1
2
T <c 2T
 7.1 Sums Over c-Values 139

The case c = 1 is exceptional since 1 = a(1) is the limit of L(s) as :

L(s) = 1 + O(2 ). (7.3)

We will briey discuss this case at the end of Sect. 7.2.

Proof. In view of (7.3) there exists a positive real number A depending on c

such that all real parts c of c-values satisfy c < A. Put
L(s) c
(s) = .
1c
Obviously, the zeros of (s) correspond exactly to the c-values of L(s). Next we
will apply Littlewoods lemma which relates the zeros of an analytic function
f (s) with a contour integral over log f (s).

Lemma 7.2 (Littlewood). Let b < a and let f (s) be analytic on R := {s

C : b a, |t| T }. Suppose that f (s) does not vanish on the right edge
= a of R. Let R be R minus the union of the horizontal cuts from the zeros
of f in R to the left edge of R, and choose a single-valued branch of log f (s)
in the interior of R . Denote by (, T ) the number of zeros  = + i of
f (s) inside the rectangle with > including zeros with = T but not those
with = T . Then
  a
log f (s) ds = 2i (, T ) d.
R b

This is an integrated version of the principle of the  argument. We give a

sketch of the simple proof. Cauchys theorem implies R
log f (s) ds = 0, and
so the left-hand side of the formula of the lemma, R , is minus the sum of
the integrals around the paths hugging the cuts. Since the function log f (s)
jumps by 2i across each cut (assuming for simplicity that the zeros of  f in
R are simple and have dierent height; the general case is no harder), R is
2i times the total length of the cuts, which is the right-hand side of the
formula in the lemma. For more details we refer to Titchmarsh [353, Sect. 9.9],
or Littlewoods original paper [224].
Let (, T ) denote the number of zeros c of (s) with c > and T < c
2T (counting multiplicities). Now let a be a parameter with a > max{A +
1, b}. Then Littlewoods Lemma 7.2, applied to the rectangle R with vertices
a + iT, a + 2iT, b + iT, b + 2iT , gives
  a
log (s) ds = 2i (, T ) d.
R b

Since
 a
 c

(, T ) d = d = (c b) (7.4)
b c >b b c >b
T <2T T <c 2T
140 7 Value-Distribution in the Complex Plane

and this quantity is real-valued, we get

 2T  2T
2 (c b) = log |(b + it)| dt log |(a + it)| dt +
c >b T T
T <c 2T
 a  a
arg ( + iT ) d + arg ( + 2iT ) d
b b


4
= Ij , (7.5)
j=1

say. To dene log (s) and log L(s) we may choose the principal branch of the
logarithm on the real axis, as ; for other points s the value of the
logarithm is obtained by continuous variation along line segments (this is in
agreement with Lemma 7.2).
We start with the vertical integrals. Obviously,
 2T
I1 (T, b) := I1 = log |L(b + it) c| dt T log |1 c|. (7.6)
T

By Jensens inequality the integral is

  
T 1 2T
log |L(b + it)| dt + O(T ).
2
2 T T

By Corollary 6.11 (which also applies to L-functions satisfying just axioms

(1)(3) as already remarked) this is  T for b > max{ 12 , 1 d1L }. Thus we
get I1 (T, b)  T unconditionally. An immediate consequence of Lindelfs
hypothesis is
 2T   2
 
L 1 + it  dt  T 1+

 2 
T
for any positive . Thus, assuming the truth of Lindelfs hypothesis we get
 
1
I1 T,  T log T.
2
Next we consider I2 . Since a > 1 we have

1
a(n)
(a + it) = 1 + , (7.7)
1 c n=2 na+it

and in view of (7.3) the absolute value of the series is less than 1 for suciently
large a. Therefore, we nd by the Taylor expansion of the logarithm



(1)k

a(n1 ) a(nk )
log |(a + it)| = Re .
k(1 c)k n =2 n =2 (n1 nk )a+it
k=1 1 k
 7.1 Sums Over c-Values 141

This leads to the estimate


(1)k

a(n1 ) a(nk ) 2T dt
I2 = Re
k(1 c)k n =2 n =2 (n1 nk )a T (n1 nk )it
k=1 1 k

 k

1
1
 1 (7.8)
k n=2 na

k=1

for suciently large a. It remains to estimate the horizontal integrals I3 , I4 .

Suppose that Re ( + iT ) has N zeros for b a. Divide the interval
[b, a] into at most N +1 subintervals in each of which Re (+iT ) is of constant
sign. Then
| arg ( + iT )| (N + 1). (7.9)
To estimate N let
1

g(z) = (z + iT ) + (z + iT ) .

2
Then we have g() = Re ( + iT ). Let R = a b and choose T so large that
T > 2R. Now, Im (z + iT ) > 0 for |z a| < T . Thus (z + iT ), and hence
g(z) is analytic for |z a| < T . Let n(r) denote the number of zeros of g(z)
in |z a| r. Obviously, we have
 2R  2R
n(r) dr
dr n(R) = n(R) log 2.
0 r R r
With Jensens formula (see for example, Titchmarsh [353, Sect. 3.61]),
 2R  2
n(r) 1   
dr = log g a + 2Rei  d log |g(a)|, (7.10)
0 r 2 0
we deduce

1 2    log |g(a)|
n(R) log g a + 2Rei  d .
2 log 2 0 log 2
By (7.7) it follows that: log |g(a)| is bounded. By Theorem 6.8, in any vertical
strip of bounded width,
L(s)  |t|B
as |t| with a certain positive constant B. Obviously, the same estimate
holds for g(z). Thus, the integral above is  log T , and n(R)  log T . Since
the interval (b, a) is contained in the disc |z a| R, the number N is less
than or equal to n(R). Therefore, with (7.9), we get
 a
|I4 | | arg ( + iT )| d  log T.
b

Obviously, I3 can be bounded in the same way.

Collecting all estimates, the assertions of the theorem follow. 

142 7 Value-Distribution in the Complex Plane

Now we want to include most of the c-values into our observations. In view
of Lemma 6.7 and Theorem 6.8 there exist positive constants C  , T  such that
there are no c-values in the region < C  , t T  . Therefore, assume that
b < C  1 and T T  + 1. By the functional equation in the form (6.13),
 
1
log |L(s) c| = log |L (s)| + log |L(1 s)| + O .
|L (s)L(1 s)|

In view of Lemma 6.7

   
1 1
log |L (s)| = ( dL log t + log(Q )) + O
2
.
2 t
Thus
 2T
log |L(b + it) c| dt
T
   2T
1
= b ( dL log t + log(Q2 )) dt
2 T
 2T
+ log |L(1 b it)| dt + O(log T ).
T

Now suppose that c = 1. The rst integral on the right-hand side is easily
calculated by elementary methods. The second integral is small if b is chosen
suciently large (see (7.8)). Together with (7.6) we get
  
1 4T
I1 = b dL T log + T log(Q ) T log |1 c| + O(log T ).
2
2 e

By (7.5) and with the estimates for the Ij s from the proof of Theorem 7.1,
we obtain

Theorem 7.3. Assume that L(s) satises the axioms (1)(3) with a(1) = 1
and let c = 1. Then, for suciently large negative b,

  
1 4T
2 (c b) = b dL T log + T log(Q2 )
2 e
T <c 2T

T log |1 c| + O(log T ).

7.2 Riemannvon Mangoldt-Type Formulae

We can rewrite the sum over c-values from Sect. 7.1 as follows:

 

 
1 1
(c b) = b 1+ c .
2 2
c c c
 7.2 Riemannvon Mangoldt-Type Formulae 143

The rst sum on the right counts the number of c-values and the second sum
measures the distances of the c-values from the critical line. Let N c (T ) count
the number of c-values of L(s) with T < c 2T . Then, subtracting the
formula of Theorem 7.3 with b + 1 instead of b from the formula with b, we
obtain
Corollary 7.4. Assume that L(s) satises the axioms (1)(3) with a(1) = 1.
Then, for c = 1,
dL 4T T
N c (T ) = T log + log(Q2 ) + O(log T ).
2 e 2
Furthermore,
Corollary 7.5. Assume that L(s) satises the axioms (1)(3) with a(1) = 1.
Then, for c = 1,

 1

T
c = log |1 c| + O(log T ).
2 2
T <c 2T

Thus, for c = 1 satisfying |1 c| = 1, the c-values, weighted with respect to

their distance to the critical line, lie asymmetrically distributed. Nevertheless,
our next aim is to show that most of the c-values lie close to the critical line.
Unfortunately, for this purpose we have to assume the Lindelf hypothesis.
Dene the counting functions (according multiplicities)
N+c (, T ) = {c : T < c 2T, c > },
and
Nc (, T ) = {c : T < c 2T, c < }.
Then
Theorem 7.6. Assume that L(s) satises the axioms (1)(3) with a(1) = 1
and let c = 1. Then, for any > max{ 12 , 1 d1L },
N+c (, T )  T, (7.11)
and assuming the truth of the Lindelf hypothesis, for any > 0,
   
1 1
Nc , T + N+c + , T  T log T.
2 2
Proof. First of all, let > max{ 12 , 1 d1L } and x 1 (max{ 12 , 1 dL }, ).
1

Then

1
N+c (, T ) (c 1 ).
1 c >
T <c 2T

The sum on the right hand-side is less than or equal to

 2T
(c 1 )  log |(1 + it)| dt + O(log T ),
c >1 T
T <c 2T
144 7 Value-Distribution in the Complex Plane

where we used Littlewoods Lemma 7.2 and the techniques from Sect. 7.1 for
the latter inequality. In view of the unconditional estimate for (7.6) in the
proof of Theorem 7.1 we obtain (7.11). Assuming the truth of the Lindelf
hypothesis, we get analogously
 
1 
N+c + , T  T log T (7.12)
2

for any positive .

Next we consider Nc . Let b be a suciently large constant. We have

 

 
1 1
(c b) b 1+ c .
1
2 1 1
2
c c c
2 2 2
T <c 2T T <c 2T T <c 2T

Hence


 

1 1
(c b) = b + c + (c b)
2 2
T <c 2T c < 1
2
c 1
2
T <c 2T T <c 2T
 
 
1 1
b N c (T ) + c
2 2
c < 1
2
T <c 2T

 
1
+ c .
2
c > 1
2
T <c 2T

By Theorem 7.1, the second sum on the right is bounded by T log T . Since
any term in the rst sum on the right is < , we obtain
 
 
1 1
Nc , T (c b) b N c (T ) + O(T log T ).
2 2
T <c 2T

In view of Theorem 7.3 and Corollary 7.4 we get

 
1 
Nc , T  T log T.
2

This is the same bound as for N+c in (7.12). Putting  = 2 we obtain the
assertion of the theorem. 


Thus, subject to the truth of the Lindelf hypothesis, we get by comparing

Corollary 7.4 and Theorem 7.6, for any positive ,
   
1 1
Nc , T + N+c + , T  N c (T ),
2 2
 7.2 Riemannvon Mangoldt-Type Formulae 145

so the c-values are clustered around the critical line for any c. This extra-
ordinary value distribution shows that if the Lindelf hypothesis for L(s)
is true, the critical line is a so-called Julia line from the classical theory of
functions. Julia [151] improved the Big Picard theorem by showing: if the
analytic function f has an essential singularity at a, then there exist a real
0 and at most one complex number z such that for every suciently small
>0
C \ {z} f ({a + r exp(i) : | 0 | < , 0 < r < });
the ray {a + r exp(i0 ) : r > 0} is called a Julia line. For more details on
Julias theorem we refer to Burckel [48, Sect. XII.4].
The distribution of the c-values close to the real axis is quite regular. It can
be shown that there is always a c-value in some neighbourhood of any trivial
zero of L(s) with suciently large negative real part, and with nitely many
exceptions there are no other in the left half-plane. The main ingredients for
the proof are Rouchs theorem (Theorem 8.1) and Stirlings formula (2.17).
With regard to (6.6), thus the number of these c-values having real part in
[R, 0] is asymptotically 12 dL R. On the other side, by (7.3) the behaviour
nearby the positive real axis is very regular. Note that all results from above
hold as well with respect to c-values from the lower half-plane.
Now let NLc (, T ) count the number of c-values c = c + ic of L(s)
satisfying c > , |c | T . Using Corollary 7.4 with 2n T for n N instead
of T and adding up, we get, for xed 0,



NLc (, T ) = 2 N c (, 2n T )
n=1
 

dL T T 1
= T log + log(Q2 ) n
e n=1
2

dL
log 4 n log 2
+ T + O(log T ).
n=1 2n

The appearing innite series are equal to 1 and 0, respectively. Hence, this
summation removes the factor 4 in the logarithmic term, and we have proved

Theorem 7.7. Assume that L(s) satises the axioms (1)(3) with a(1) = 1.
For any xed 0 and any complex c = 1,
dL T T
NLc (, T ) = T log + log(Q2 ) + O(log T ).
e
The case c = = 0 (the non-trivial zeros of L(s)) is a precise Riemannvon
Mangoldt formula (1.5). Similar results were obtained by Perelli [289] and
Lekkerkerker [214] for other classes of Dirichlet series. It should be noticed
that Tsang [355] investigated the number of c-values of (s) with respect to
short intervals for the imaginary parts. Let < 12 , T (1/2)+
H T , and
146 7 Value-Distribution in the Complex Plane

c be a complex number satisfying  |1 c| 1
with suciently small .

Assuming the truth of the Riemann hypothesis, Tsang proved
H T
Nc (, T + H) Nc (, T ) log
2
with an explicit error term depending on , H and T ; his result holds uncon-
ditionally provided 0.
An immediate consequence of Theorem 7.7 is that the multiplicity of non-
trivial zeros  of L(s) is bounded by 1 + log ||. More advanced results on
the multiplicities of the zeros were obtained by Ivi [142] in the case of the
Riemann zeta-function.
We conclude with another result from Selberg [323] for L-functions from S.
Assuming the truth of the Riemann hypothesis and of conjecture A, he
obtained for c = 1 the asymptotic formula

 1

nL % T |c|
c = T log log T + log +
1
2 4 3/2 4 1 |c|2
c >
2
0<c <T
 
(log log log T )3
+O T .
log log T
Furthermore, for

1 log log T dL
(T ) := and :=
2 log T 2 nL
with positive , he proved


(c (T ))
c >(T )
0<c <T
   
1 exp( 2 )
nL %
= + exp(x2 ) dx T log log T
2 2
  
T
+ log |c| exp(x ) dx log |1 c|
2
2
 
(log log log T )3
+O T .
log log T
From these results Selberg deduced that about half of the c-values lie to the
left of the critical line, statistically well distributed at distances of order

log log T
log T
o = 12 , and that

NLc ((T ), T ) NLc (T ) exp(x2 ) dx.

 7.3 Nevanlinna Theory 147

Most of the remaining c-values lie rather close to the critical line at distances
of order not exceeding
(log log log T )3
.
log T log log T
This improves some previous results of Selberg (unpublished) and Joyner [150]
and gives a much more detailed description of the clustering of the c-values
around the critical line.
In the exceptional case c = 1 one has to consider the function
qs
(s) = (L(s) 1),
a(q)
where q 1 is the least integer such that a(q) = 0. Then, by a similar
reasoning as in the proof of Theorem 7.7, one gets analogous results. For the
special case of the zeta-function this is carried out in Steuding [348, 349] where
Levinsons method is applied to Epstein zeta-functions. These methods also
allow to drop the condition a(1) = 1.

7.3 Nevanlinna Theory

Nevanlinna theory was created by Nevanlinna [281] in the 1920s to tackle
the value-distribution of meromorphic functions in general. We recall some
basic facts which, for example, can be found in Nevanlinnas monograph [281,
Chaps. VI and IX].
Let f be a meromorphic function and denote the number of poles of f (s)
in |s| r by n(f, , r) (counting multiplicities). The number of c-values of f
is given by  
1
n(f, c, r) = n , , r .
f c
The integrated counting function is
 r
d
N(f, c, r) = (n(f, c, ) n(f, c, 0)) + n(f, c, 0) log r.
0 
The proximity function is dened by
 2
1
m(f, r) = log+ |f (r exp(i))| d,
2 0
and, for c C, by  
1
m(f, c, r) = m ,r ,
f c
where log+ x := max{0, log x}. The function m(f, c, r) indicates how close
f (s) is to the value c on the circle |s| = r. The characteristic function of f is
dened by
T(f, r) = N(f, , r) + m(f, r).
148 7 Value-Distribution in the Complex Plane

Furthermore, let
T(f, c, r) = N(f, c, r) + m(f, c, r)
for c C. The rst main theorem in Nevanlinna theory states states that
T(f, c, r) diers from the characteristic function by a bounded quantity:

Theorem 7.8. Let f be a meromorphic function and let c be any complex

number. Then
T(f, c, r) = T(f, r) + O(1),
where the error term depends on f and c.

The proof relies on Jensens formula (7.10).

Thus, T(f, c, r) for dierent values of c is invariant up to additive terms
that are bounded. The invariant, the characteristic function T(f, r), encodes
information about the analytic behaviour of f .
The quantity
N(f, c, r)
(f, c) := 1 lim sup
r T(f, r)
is called the deciency of the value c of f . This deciency is positive only
if there are relatively few c-values. The second main theorem in Nevanlinna
theory implies the so-called deciency relation which states that


(f, c) 2 ;
cC{}

note that only for countably many values of c the deciency can dier from
zero. Another consequence is the Big Picard theorem.
Only recently Ye [372] computed the Nevanlinna functions for the Riemann
zeta-function. Without big eort we can extend his results to the class of
Dirichlet series under investigation. The Nevanlinna functions for those L(s)
are determined by the Gamma-factors in the functional equation.
First, let 0 > 1 be xed. We write s = r exp(i), so = r cos . It is easily
seen that 
1
log+ |L(r exp(i))| d  1.
2 {:r cos >0 }
Further, in view of Theorem 6.8,

1
log+ |L(r exp(i))| d  log r;
2 {:10 r cos 0 }

note that the Lebesgue measure of the set

{ [0, 2] : = r cos [1 0 , 0 ]}

is bounded by 1r . Finally, for 10 we deduce from the functional equation

in the form (6.13) that
 7.3 Nevanlinna Theory 149
f 


log |L(r exp(i))|
+
log+ | (j (1 r exp(i)) + j )|
j=1

+ log+ | (j r exp(i) + j )| + O(r).

Now we shall use Yes decomposition of the Gamma-function. For any z =

r exp(i), there is an integer n0 with n0 < r n0 + 1 such that
 
1

2n0
z
= F1 (z) F2 (z) with F1 (z) := z z ,
(z) n=1
n

where is the EulerMascheroni constant, and F2 (z) is an entire function with

m(F2 , r)  r. The order of growth of (z) is ruled by the order of growth of
F1 (z). Ye computed

log |F1 (z)| = r log r cos + O(r).

If is a positive real number and an arbitrary complex number, Yes

estimate leads to

1
log+ | ((1 r exp(i)) + )| d
2 {:r cos <10 }
 /2

r log r cos d + O(r) = r log r + O(r),
2 /2

and, similarly,

1
log+ | (r exp(i)) + )| d r log r + O(r).
2 {:r cos <10 }

Thus, we get

1 dL
log+ |L(r exp(i))| d r log r + O(r).
2 {:r cos <10 }

Adding the estimates for the other cases, we obtain for the proximity function
of L(s)
dL
m(L, r) r log r + O(r).

Since L(s) is regular except for at most a pole at s = 1,
 r
d
N(L, , r)  = log r. (7.13)
1 
Thus, we get
dL
T(L, r) r log r + O(r). (7.14)

150 7 Value-Distribution in the Complex Plane

It follows from Theorem 7.7 that:

dL
N(L, 0, r) = r log r + O(r). (7.15)

The rst main Theorem 7.8 implies
N(L, 0, r) T(L, 0, r) = T(L, r) + O(1).
In view of (7.14) and (7.15) we get an asymptotic formula for the characteristic
function:
Theorem 7.9. For L satisfying axioms (1)(3) with a(1) = 1,
dL
T(L, r) = r log r + O(r).

We deduce from this and (7.13) for the deciency value of innity:
N(L, , r)
(L, ) = 1 lim sup = 1.
r T(L, r)
In view of Theorem 7.7 the deciency values for c = 1, are equal to zero.
In combination with Theorem 7.7 the asymptotic formula of the theorem
shows that the counting function N(L, c, r) dominates the proximity function
m(L, c, r), at least for any complex value c = 1. In the exceptional case c = 1,
by the rst main Theorem 7.8, we may deduce from Theorem 7.9 that
dL
NL1 (T ) T log T + O(T ).

A more sophisticated analysis would show that this is actually an equality.
However, we do not go into the details. In Sect. 9.7 we return to the distribu-
tion of c-values in the half-plane of absolute convergence.
We conclude with a description of the analytic behaviour of the Dirichlet
series L under investigation in terms of the notion of nite order. A positive
function t(r) is said to be of nite order if
log t(r)
lim sup = ;
r log r
t(r) is of maximum, mean or minimum type of order if the upper limit
t(r)
lim sup
r r
is innite, nite and positive, or zero. A meromorphic function is dened to
be of the same order and the same type as its characteristic function T(r, f ).
Thus, by Theorem 7.9, we get
Corollary 7.10. Every L satisfying the axioms (1)(3) with a(1) = 1 is of
order one and of maximum type.
 7.4 Uniqueness Theorems 151

7.4 Uniqueness Theorems

We say that two meromorphic functions f and g share a value c C {} if
the sets of pre-images of the value c under f and under g are equal, for short
f 1 (c) := {s C : f (s) = c} = g 1 (c). (7.16)
We say that f and g share the value c counting multiplicities (CM) if (7.16)
holds and if the roots of the equations
f (s) = c and g(s) = c
have the same multiplicities; if there is no restriction on the multiplici-
ties, f and g are said to share the value c ignoring multiplicities (IM).
Nevanlinna [280] proved two fundamental results on shared values. His
remarkable ve-point theorem states that any two non-constant meromorphic
functions which share ve distinct values are equal. Since f (s) = exp(s) and
g(s) = exp(s) share the four values 0, 1, , the number 5 in Nevanlinnas
statement is best possible. If multiplicities are taken into account, Nevanlinna
proved that if two meromorphic functions f and g share four distinct values
c1 , . . . , c4 CM, then either f g or there exists a linear fractional transfor-
mation M such that g M f and
M (c1 ) = c1 , M (c2 ) = c2 , M (c3 ) = c4 , and M (c4 ) = c3 ;
in the latter case f and g do not assume the values c3 and c4 . Also the number
4 for the upper bound of shared values CM is best possible. The result can be
sharpened if two of the four values are allowed to be shared IM (see [111]).
In [347], Steuding, investigated how many values L-functions can share. In
this special case better estimates are possible than those which Nevanlinnas
theorems provide. It is expected that independent L-functions cannot share
any complex value which is actually taken.
First of all, we trivially note that two L-functions from the Selberg class
share the value CM if and only if both are entire or if they both have a pole
at s = 1 of the same order (other poles cannot occur), e.g., the Riemann zeta-
function (s) and a Dedekind zeta-function to a quadratic number eld. If
the orders of the poles dier, they share the value IM. Further, we observe
that two dierent L-functions in the Selberg class cannot share the value zero
CM. This follows immediately from a theorem of M.R. Murty and V.K. Murty
[268]. To see that denote the non-trivial zeros of L S by  and let mL ()
be the multiplicity of . Further, dene for L1 , L2 S the function


DL1 ,L2 (T ) = |mL1 () mL2 ()|,


where the summation is taken over all non-trivial zeros  of L1 and L2

(counting multiplicities). Then M.R. Murty and V.K. Murty proved that
L1 , L2 S are either equal or
1
lim inf DL1 ,L2 (T ) > 0
T T
152 7 Value-Distribution in the Complex Plane

(see also the related result of Bombieri and Perelli (6.8)). However, the trivial
example (s) and (s)2 shows that dierent elements of S can share the value
zero IM.
Concerning CM-shared values we shall prove that two dierent Dirichlet
series satisfying our axioms do not share any complex value CM. For sharing
values IM we shall only obtain an improvement of the ve-point theorem under
an additional assumption on the number of distinct c-values. For this purpose
let NLc (T ) count the number of distinct roots c of the equation L(s) = c lying
in the rectangle 0 1, |t| T .

Theorem 7.11. Assume that L1 , L2 satisfy the axioms (1)(3) with a(1) = 1.
(i) If L1 , L2 share a value c = CM, then L1 L2 .
(ii) If L1 , L2 satisfy the same functional equation and share two distinct values
c1 , c2 = IM such that

NLc1j (T ) + NLc2j (T ) 1
lim inf > + (7.17)
T NLc1j (T ) + NLc2j (T ) 2

for some positive  with either j = 1 or 2, then L1 L2 .

We briey discuss the second assertion of the theorem. Condition (7.17)

reects that more than 50% of the c1 - and c2 -values of Lj (s) are supposed to
be distinct. It should be noted that such conditions are very dicult to verify.
For instance, Farmer [78] proved that more than 63% of the zeros of (s) are
distinct; however, any extension to L-functions of larger degree seems to be
hard to realize.

Proof. We start with the rst assertion. In view of Theorem 7.7 two L-
functions satisfying the axioms (1)(3) can only share a value c = CM
if they have the same degree, d say. First of all assume that L1 , L2 are both
entire functions and share the value c = CM. Dene the function

L1 (s) c
(s) = .
L2 (s) c

Since L1 (s) assumes the value c if and only if L2 (s) = c and since for any
such root the multiplicities coincide, (s) is a non-vanishing entire function.
In view of the rst main Theorem 7.8 and Theorem 7.9
d
T(L2 , c, r) = T(L2 , r) + O(1) = r log r + O(r)

= T(L1 c, r) + O(r).

For a meromorphic function f denote its order by (f ). Then it follows that:

 
1
= (L2 c) = (L2 ) = 1.
L1 c
 7.4 Uniqueness Theorems 153

It is easily seen that the order of a nite product of functions of nite order is
less than or equal to the maximum of the order of the factors. Thus () 1.
By Hadamards factorization theorem (see [281, Sect. VIII.2]) this implies that
(s) is of the form
(s) = exp(P (s)),
where P is a polynomial of degree at most () 1. Since Lj (s) tends to one
as s for j = 1, 2, we have
1c
lim (s) = = 1.
s 1c
This implies that the polynomial P is vanishing identically, which implies
L1 L2 .
If L1 (s) or L2 (s) has a pole at s = 1 of order k, we may replace Lj (s)
by (s 1)k Lj (s) and repeat the argument from above. This proves the rst
assertion.
Now we shall prove the second statement. If L1 and L2 satisfy the same
functional equation, they both have the same degree, d say. Now consider the
function
(s) := L1 (s) L2 (s).
Obviously, also (s) satises the common functional equation for the Lj s.
Then the number N (T ) of zeros of (s) in the rectangle 0 1, |t| T
(counting multiplicities) is asymptotically given by
d
N (T ) T log T. (7.18)

Now suppose that L1 and L2 share two distinct complex values c1 , c2 IM.
Then (s) vanishes also for the pre-images of the ck s. Hence, we obtain a
lower bound for the number of zeros of (s) in terms of the c1 - and c2 -value
counting functions, namely

N (T ) NLc1j (T ) + NLc2j (T ), (7.19)

where we can take j = 1 or j = 2. Taking into account Theorem 7.7 and

(7.18) we can replace N (T ) by
1 c1

NLj (T ) + NLc2j (T ) + O(T ).

2
Thus we can rewrite (7.19) as

NLc1j (T ) + NLc2j (T ) 1
+ o(1).
NLc1j (T ) + NLc2j (T ) 2

This contradicts (7.17). Hence L1 and L2 can share at most one value c C.
Theorem 7.11 is proved. 

154 7 Value-Distribution in the Complex Plane

The functions L(s) and L(s)2 share the value zero IM. This is a special
example for two reasons. First, these functions are not independent in the sense
that they have the same primitive functions in their factorizations. Second,
they share the zeros. Bombieri and Hejhal [40] proved, assuming some widely
believed but yet unproved hypotheses, that almost all zeros of pairwise inde-
pendent L-functions are distinct. Of course, we expect the same to hold for
other c-values too. With respect to condition (7.17) this leads us to conjecture
that zero is the only possible shared value and that this happens only in cases
of dependent L-functions.
We conclude with a few words about the signicance of such studies.
Some problems in arithmetic (see Chap. 13.7) could be solved if one could
show that, given distinct primitive L-functions L1 (s), . . . , Lm (s) (in the sense
of the Selberg class), then Lj (k ) = 0 holds only for j = k, where the k
denote the non-trivial zeros of Lk (s). Clearly, this would also imply the unique
factorization into primitive elements.
8
The Riemann Hypothesis

...und es ist sehr wahrscheinlich, dass alle Wurzeln reell sind. Hiervon
wre allerdings ein strenger Beweis zu wnschen; ich habe indess die
Aufsuchung desselben nach einigen chtigen vergeblichen Versuchen
vorlug bei Seite gelassen... Bernhard Riemann

There is an interesting link between universality and the zero-distribution.

As we will show in this chapter, the question whether the zeta-function can
approximate itself in the right half of the critical strip turns out to be equiv-
alent to the Riemann hypothesis. This reformulation dates back to Bohr [30]
who proved its analogue for Dirichlet L-functions to non-principal characters.
Bagchi [9] extended this result to the Riemann zeta-function. We shall also
consider further generalizations.

8.1 Uniform Approximation and Zeros

In view of the phenomenon of universality a natural question arises: is the
condition on the non-vanishing of g(s) in the universality Theorem 5.14 nec-
essary or is it possible to approximate uniformly functions having zeros by a
universal L-function? The answer is negative. For the sake of simplicity we
consider only the case of the Riemann zeta-function. Assume that there is
an analytic function g(s) dened on an admissible set K with a zero in the
interior and
lim (s + ij ) = g(s)
j

on K for some innite sequence j which tends to innity as j . Then,

by Hurwitzs Theorem 5.13, the zero of g(s) is limit point of zeros of the
shifts (s + ij ) and, in particular, there would exist nontrivial zeros o the
critical line a violation of the Riemann hypothesis. However, the Riemann
hypothesis is still unproved and we shall sketch an unconditional argument
which can be xed with a bit more eort by the techniques of Sect. 8.2.
156 8 The Riemann Hypothesis

In order to see that (s) cannot approximate uniformly a function with a

zero, recall Rouchs theorem:

Theorem 8.1. Let f (s) and g(s) be analytic for |s| r. If

|f (s) g(s)| < |g(s)|

on |s| = r, then f (s) and g(s) have the same number of zeros in |s| < r.

This classical result follows from a simple application of the argument princi-
ple; for details see Burckel [48, Sect. VIII.3] or Titchmarsh [352, Sect. 3.42].
Now assume that g(s) is an analytic function on |s| r, where 0 < r < 14 ,
which has a zero with || < r, but which is non-vanishing on the boundary.
An application of Rouchs theorem shows that whenever the inequality
   
 3 
max  s + + i g(s) < min |g(s)|
 (8.1)
|s|=r 4 |s|=r

 
holds, s + 34 + i has to have a zero inside |s| < r. The zeros of an analytic
function lie either discretely distributed or the function vanishes identically,
and thus the inequality (8.1) holds if the left-hand side is suciently small. If
now for any  > 0
     
1  3 
 
lim inf meas [0, T ] : max  s + + i g(s) <  > 0,
T T |s|r 4

then we expect  T many in the interval [0, T ] each of which corresponds

via (8.1) to a complex zeros of (s) in the strip 34 r < < 34 + r up to T (for
a rigorous proof one has to consider the densities of values satisfying (8.1);
this can be done along the lines of the proof of Theorem 8.3). This contradicts
the density Theorem 1.2, which gives
 
3
N r, T = O(T ).
4

Thus, a given function having zeros cannot be approximated uniformly by

the zeta-function (in the sense of Voronins theorem)! (s) is not strongly
universal.
The above reasoning shows that the location of the complex zeros of Rie-
manns zeta-function is closely connected with the universality property.

8.2 Bagchis Theorem

Bohr introduced the fruitful notion of almost periodicity to analysis. An ana-
lytic function f (s), dened on some vertical strip a < < b, is called almost
periodic if, for any positive , and any , with a < < < b, there exists a
 8.2 Bagchis Theorem 157

1.5

0.5

1 1 2 3

0.5

1.5

Fig. 8.1. ( 12 + it) for t [0, 40]. In this range, there are six non-trivial zeros lying
on the critical line

length  = (f, , , ) > 0 such that every interval (t1 , t2 ) of length  contains
an almost period of f relatively to in the closed strip , i.e., there
exists a number (t1 , t2 ) such that

|f ( + it + i ) f ( + it)| < for any , t R. (8.2)

Bohr [30] proved.

Theorem 8.2. Every Dirichlet series is almost-periodic in its half-plane of

absolute convergence.

The most important open problem in the theory of the Riemann zeta-
function is the Riemann hypothesis on the location of the nontrivial zeros (see
Fig. 8.1 for the values taken by the zeta-function on the critical line). Bohr
discovered an interesting relation between the Riemann hypothesis and almost
periodicity; indeed, his aim in introducing the concept of almost periodicity
might have been Riemanns hypothesis. Bohr showed that if is a non-
principal character, then the Riemann hypothesis for the Dirichlet L-function
L(s, ) is equivalent to the almost periodicity of L(s, ) in the half-plane
> 12 . The condition on the character looks articial but it is necessary
for Bohrs reasoning. His argument relies in the main part on diophantine
approximation applied to the coecients of the Dirichlet series representation.
The Dirichlet series for L(s, ) with a non-principal character converges
throughout the critical strip, but the Dirichlet series for the zeta-function
does not.
More than half a century later Bagchi [9] proved that the Riemann
hypothesis is true if and only if for any compact subset K of the strip 12 < < 1
with connected complement and for any  > 0
158 8 The Riemann Hypothesis
 
1
lim inf meas [0, T ] : max |(s + i ) (s)| <  > 0.
T T sK

In [10], Bagchi generalized this result in various directions; in particular for

Dirichlet L-functions to arbitrary characters. One implication of his proof in
[10] relies essentially on Voronins universality theorem which, of course, was
unknown to Bohr. Later, Bagchi [11] gave another proof in the language of
topological dynamics, independent of universality, and therefore this property,
equivalent to Riemanns hypothesis, is called strong recurrence.
We extend Bagchis result slightly to
Theorem 8.3. Let 12 . Then (s) is non-vanishing in the half-plane >
if and only if, for any  > 0, any z with < Re z < 1, and for any 0 < r <
min{Re z , 1 Re z},
 
1
lim inf meas [0, T ] : max |(s + i ) (s)| <  > 0.
T T |sz|r

Proof. If Riemanns hypothesis is true we can apply Voronins universality

Theorem 1.7 with g(s) = (s), which implies the strong recurrence. More
generally, the non-vanishing of (s) for > would allow to approximate (s)
by shifts (s + i ) uniformly on appropriate subsets of the strip < < 1.
The idea for the proof of the other implication is that if there is at least one
zero to the right of the line = , then the strong recurrence property implies
the existence of many zeros, in fact too many with regard to the classic density
Theorem 1.2.
Suppose that there exists a zero of (s) with Re > . Without loss of
generality we may assume that Im > 0. We have to show that there exists a
disk with center z and radius r, satisfying the conditions of the theorem, and
a positive  such that
 
1
lim inf meas [0, T ] : max |(s + i ) (s)| <  = 0. (8.3)
T T |sz|r

Locally, the zeta-function has the expansion

 
(s) = c(s )m + O |s |m+1 | (8.4)

with some non-zero c C and m N. Now assume that for a neighbourhood

K := {s C : |s | } of the relation

max |(s + i ) (s)| <  min |(s)| (8.5)

sK |s|=

holds; the second inequality is fullled for suciently small  (by an argu-
ment already discussed in Sect. 8.1). Then Rouchs Theorem 8.1 implies the
existence of a zero  of (s) in

K + i := {s C : |s i | }.
 8.2 Bagchis Theorem 159

We say that the zero  of (s) is generated by the zero . With regard to (8.4)
and (8.5) the zeros and  = + i are intimately related; more precisely,

 > |() ( i )| = |( i )| |c| | i |m + O( m+1 ).

Hence,
 1/m


| i | + O 1+(1/m) .
|c|
In particular,
 1/m
1 
< Re 2 < < 1,
2 |c|
and  1/m

| ( + Im )| < 2 ,
|c|
for suciently small  and = O(m+1 ). Next we have to count the generated
zeros in terms of . Two dierent shifts 1 and 2 can lead to the same zero
, but their distance is bounded by
 1/m

|1 2 | < 4 .
|c|
If we now write
  
I(T ) := Ij (T ) := [0, T ] : max |(s + i ) (s)| <  ,
sK
j

where the Ij (T ) are disjoint intervals, it follows that there are

'   (  1/m
1/m
1 |c| 1 |c|
meas Ij (T ) + 1 > meas Ij (T )
4  4 

many distinct zeros according to Ij (T ), generated by . The number of

generated zeros is a lower bound for the number of all zeros. For the number
of all zeros having real part > Re 2( |c|

1/m
) up to level T it follows that
 1/m  1/m 
 
  = + i : > Re 2 , 0 < < T + Im + 2
|c| |c|
 1/m
|c|
1/4 meas I(T ).

This and the density Theorem 1.2 lead to

meas I(T ) = O(T ),

which implies (8.3). The theorem is proved. 


160 8 The Riemann Hypothesis

Using the same reasoning, Reichs discrete version of Voronins universality

theorem, Theorem 5.16, yields a discrete version of Theorem 8.3: Riemanns
hypothesis is true if and only if for any  > 0, any real number = 0, any z
with < Re z < 1, and any 0 < r < min{Re z , 1 Re z},
 
1
lim inf  1 n N : max |(s + in) (s)| <  > 0.
N N |sz|r

The expected strong recurrence of (s) may be regarded as a kind of

self-similarity. Assuming the truth of Riemanns hypothesis, this has a nice
interpretation. Consider the amplitude of light which is a physical bound for
the size of objects which human eyes can see, or the Planck length lP 1.616
1035 meter which is the smallest size of objects in quantum mechanics. If we
assume that  is less than one of these quantities, then we cannot physically
distinguish between (s) and (s + i ) for s from a compact subset K of the
right half of the critical strip, whenever

max |(s + i ) (s)| < .

sK

This shows that we cannot decide where in the analytic landscape of (s) we
actually are without moving to the boundary. The zeta-function is an amazing
maze!

8.3 A Generalization
It is obvious how Bagchis Theorem 8.3 can be generalized to other universal L-
functions, e.g., Dirichlet L-functions, for which appropriate density estimates
are known. Assume that L(s) is strongly recurrent in the strip m < < 1
in the sense of Theorem 8.3. In view of the positive lower density for the
set of shifts with which the target L(s) can be uniformly approximated by
L(s + i ) we need that the number of zeros with real part greater than m up
to level T is bounded by o(T ) as T to deduce the non-vanishing of L(s)
in the strip m < < 1. Thus, strong recurrence allows the conclusion that
if there are not too many zeros, then there are none!
For the set of L-functions in S S we can replace the density Theorem 1.2
by Theorem 6.5 from Kaczorowski and Perelli [163]. We observe for the number
of zeros in question that

NL (, T )  T 4( dL +3)(1)+
= o(T )

if
1
> = (L) := 1 ,
4( dL + 3)
which is larger than the left abscissa of the strip of universality in Theo-
rem 6.12. This yields the non-vanishing of L(s) in some strip inside the critical
 8.3 A Generalization 161

strip provided L(s) is strongly recurrent. For the line = 1, we notice that
axiom (v) on the mean-square for the Dirichlet series coecients of L S
on the primes implies the normality conjecture (as remarked in Sect. 6.6)
and Kaczorowski and Perelli [163] proved that the latter conjecture implies
L(1 + iR) = 0 (see Sect. 6.2). Hence,

Theorem 8.4. Let L S S. Then L(s) is non-vanishing in the half-plane

> if and only if, for any  > 0, any z with < Re z < 1 and any
0 < r < min{Re z , 1 Re z},
 
1
lim inf meas [0, T ] : max |L(s + i ) L(s)| <  > 0.
T T |sz|r

Of course, assuming the validity of the density hypothesis or the Lindelf

hypothesis, one can also prove a conditional result valid in the strip 12 < < 1.
In many cases one may obtain better results with a general density estimate
from Perelli [289]. For this purpose we introduce Perellis class of L-functions
[289] which has many similarities with the Selberg class.
Perellis class P of L-functions consists of Dirichlet series


a(n)
L(s) := ,
n=1
ns

which satisfy the following axioms.

Analytic continuation. L(s) can be analytically continued to a meromor-
phic function with at most one simple pole at s = 1.
Polynomial Euler product. There exists a positive integer m such that
for any prime p there exists an m m matrix Ap with complex entries
and eigenvalues j (p) with |j (p)| = 1 for all but nitely many p, and
|j (p)| < 1 otherwise, such that
m 
 1
j (p)
L(s) = 1 .
p j=1
ps

Finite order. There exist positive constants c and for which

L(s)  exp(exp(c|t|))

uniformly in 1 + .
Functional equation. There is a sequence of matrices Ap , all satisfying
the axiom on the polynomial Euler product, such that for L (s), dened
according to the axiom on the polynomial Euler product and satisfying all
previous axioms,
L (s) = L (1 s),
162 8 The Riemann Hypothesis

where

f
L (s) := L(s)Qs (j s + j ),
j=1

and where Q, j are real numbers and j and with || = 1 are complex
numbers.
For L P dene the quantities dL and , as for the elements of the Selberg
class S, by


f
f
dL = 2 j and = 2 (1 2j ).
j=1 j=1

Perellis class of L-functions is rather similar to S but it does not contain S

since, for example, (s)2  P by the axiom on the analytic continuation (there
is only a simple pole at s = 1 allowed). All axioms for P, apart from the one
on the analytic continuation, are covered by the related axioms for S. This
follows more or less directly from the axioms in addition with the Phragmn
Lindelf principle. The growth restriction in P follows via Theorem 6.8 from
the estimate
1
L(s)  tL ()+
 t dL ( 2 +)+
,
which is valid for 1 + with any positive . We do not know any
example of an element of P which does not lie in S.
Besides other interesting results (which we partially mentioned in
Sect. 6.7), Perelli proved zero-free regions subject to a natural condition on the
matrices Ap , a mean-square estimate, and an approximate functional equa-
tion. Moreover, using Montgomerys zero detection method [258], he obtained
a density theorem comparable with Theorem 1.2, respectively, (6.5). In many
instances this yields, for any > max{ 12 , 1 d1L },

NL (, T ) = O(T ).

Actually, Perelli proved stronger estimates for the whole right half of the
critical strip which are even uniform in Q, but this is not of interest for our
investigations.

8.4 An Approach Towards Riemanns Hypothesis?

Bagchis theorem oers an interesting but up to now insucient approach
toward the Riemann hypothesis.
In view of Voronins proof we may have the naive idea to start with a
truncated Euler product
  1
1
M (s) = 1 s ,
p
pM
 8.5 Further Equivalents of the Riemann Hypothesis 163

which converges throughout the half-plane > 0 (clearly, we are not allowed
to work with the logarithm of the zeta-function since we are interested in its
zeros). Obviously,

|(s + i ) (s)| |(s + i ) M (s)| + |M (s) (s)|.

Since M (s) is a non-vanishing analytic function in the half-plane > 0, by

Voronins universality theorem, the rst quantity on the right-hand side can
be made as small as we please for all s in some disk |s z| r for a set T of
values with positive lower density. For a proof of Riemanns hypothesis it
would be sucient to show that also the second quantity on the right-hand side
can be made arbitrarily small on a subset of T with positive lower density.
The second term is independent of but unfortunately, we cannot simply
approximate (s) by the truncated Euler product M (s). It is not dicult to
show that
 T
1
lim lim |M ( + it) ( + it)|2 dt = 0
M T T 0

for any > 12 (see [353, Sect. 7.11]; see also Lemma 4.8). This implies that
M (s) approximates (s) almost everywhere: given  > 0, there exist M0 , T0
such that for any M M0 , T T0 we have
 T
1
|M ( + it) ( + it)|2 dt < .
T 0

Hence the set

{t [0, T ] : |M ( + it) ( + it)| }

has Lebesgue measure o
(T ) and so M ( + it) ( + it) is small on average.

However, this is not sucient for |M (s) (s)| being small in general.

8.5 Further Equivalents of the Riemann Hypothesis

Mishou and Nagoshi [253] obtained necessary and sucient conditions for
the truth of the Riemann hypothesis in terms of the functional distribution
of quadratic L-functions L(s, d ) in the right half of the critical strip; here
L(s, d ) denotes the Dirichlet L-functions to the character d mod |d| given
by the Kronecker symbol ( d ) to a fundamental discriminant d (see Sect. 1.4).
They proved that the Riemann hypothesis for (s) is true if and only if, for
any compact subset K of the open strip 12 < < 1 and any positive ,

{1 d X : maxsK |L(s, d ) + (s)| < }

lim inf > 0;
X {1 d X}
164 8 The Riemann Hypothesis

an analogous result is stated and proved with respect to prime discrimi-

nants d. Their proof makes use of the universality theorems from Mishou and
Nagoshis paper [251, 252] (see (1.31)) and a zero density estimate for Dirichlet
L-functions of Jutila in combination with Rouchs theorem.
Another equivalent condition for the Riemann hypothesis was given by
leeviieneSteuding [334]. Remarkably, the Riemann zeta-function does not
appear in its formulation. In order to state this equivalent we rst recall
Beurlings generalized Euler products. In 1937, Beurling [20] investigated
prime number theorems for generalized multiplicative structures. For this aim
he introduced the so-called Beurling primes. We assume that we are given a
set P of positive real numbers pn which can be arranged as

p1 < p2 < < pn < pn+1 <

and satisfy limn pn = . Then the set G consisting of all possible products
of powers of elements pn including the positive integer 1 (the empty product)
forms a semigroup. The attached Euler product is dened by
  1
1
P (s) = 1 s .
pn
pn P

Assuming a certain distribution of the elements of G, Beurling [20] proved

that
x
P (x) := 1 ;
log x
P
pn x

later Nyman improved this result by giving an error term. It follows from
Reichs universality theorem [305] that Beurling zeta-functions P (s) are uni-
versal. In [110], Grosswald and Schnitzer suggested an interesting approach
toward the Riemann hypothesis. They considered a prime system Q = {qn }
formed with arbitrary but xed real numbers qn satisfying

pn qn pn+1 ,

where pn denotes the nth rational prime number. Then the associated Beurling
zeta-function Q (s) has an analytic continuation to the half-plane > 0 (but
in general the line = 0 is a border over which we cannot expect analytic
continuation) except for a simple pole at s = 1 and, most remarkably, for > 0
it has the same zeros with the same multiplicity as the Riemann zeta-function
(s). Mller [264] showed that the same statement holds if the condition of
Grosswald and Schnitzer is replaced by the weaker assumption

Q (x) = (x) + O(x
)

for any  > 0. Combining Reichs universality theorem, the zero-distribution of

the latter Beurling zeta-functions, and the ideas behind Bagchis Theorem 8.3,
leeviieneSteuding [334] showed that the Riemann hypothesis for (s) is
 8.5 Further Equivalents of the Riemann Hypothesis 165

true if and only if Q (s) can approximate itself uniformly, i.e., for certain disks
K and any  > 0,
1
lim inf meas { [0, T ] : max |Q (s + i ) Q (s)| < } > 0;
T T sK

here Q (s) is assumed to satisfy certain natural side-conditions. Furthermore,

in [334] with  
1
qn = pn exp ,
pn
an example of such a Beurling zeta-function Q (s) was given.
9
Eective Results

Kroneckers theorem is one of those mathematical theorems which

assert, roughly, that what is not impossible will happen sometimes
however improbable it may be. G.F. Hardy and E.M. Wright

In this chapter, we shall use ideas from the previous chapter in order to
obtain certain eective results on the value-distribution of L-functions. The
rst sections deal with the density of the approximating in universality
theorems. The derived upper bounds are due to Steuding [342, 350]. In the
following sections explicit estimates for c-values in the half-plane of absolute
convergence are obtained. These results are due to Girondo and Steuding [100]
and rely on a theorem of Rieger [310], resp. a quantied version of Steuding
[344], on eective inhomogeneous diophantine approximation.

9.1 The Problem of Eectivity

The known proofs of universality theorems are ineective, giving neither an
estimate for the rst approximating shift nor bounds for the positive lower
density with the exception of some attempts by Good, Laurinikas, and
Garunktis which we shall now discuss shortly.
If the Riemann hypothesis is true, then
    
 1  log t
log  + it  = O (9.1)
2 log log t

as t (see [353]). This is a signicant improvement of the bound for

(s) on the critical line predicted by the Lindelf hypothesis, but we may ask
whether it is the correct order? On the contrary, Montgomery [260] proved,
for xed 12 < < 1 and any real ,
 
(log t)1
Re {exp(i) log ( + it)} = , (9.2)
(log log t)
168 9 Eective Results

and the same estimate is valid for = 12 under assumption of the truth
of the Riemann hypothesis. By a dierent method, Balasubramanian and
Ramachandra [15] obtained the same estimate for = 12 unconditionally.
These -results were only slight improvements of earlier results and some
probabilistic heuristics suggest these estimates to be best possible, i.e., the
quantity in (9.2) describes the exact order of growth of (s). However, the
random matrix model predicts signicantly larger values: in analogy to large
deviations for characteristic polynomials one may expect that the estimate in
(9.1) gives the true order (see Hughes [136] for details). Steuding [338] showed
that Montgomerys estimate holds on any rectiable curve inside the right half
of the critical strip as well. In particular, assuming the truth of Riemanns
hypothesis, for any rectiable curve t  (t) + it, t R, in the right half of
the critical strip,
  
1 (log t)1(t)
((t) + it) = exp .
20 (log log t)(t)

Thus, the analytic landscape of the Riemann zeta-function over the critical
strip does not contain long valleys.
The proofs of Voronins theorem do not give any information about the
question how soon a given target function is approximated by (s + i ) within
a given range of accuracy, and Montgomerys approach does not give us
any idea of the shape of the set of values of (s) on vertical lines. Good
[105] combined Voronins universality theorem with the work of Montgomery
on extreme values of the zeta-function. This enabled him to complement
Voronins qualitative picture with Montgomerys quantitative estimates. For
suciently large T and with log T and r suciently small (but not
too small), he showed the existence of a certain constant c and more than
) positive numbers tn T with tn+1 > tn + 2r for n = 1, 2, . . .,
c
T exp( log
such that the annulus
    1+r 
c 1+r c
z C : exp |z| exp
log log

is contained in ( + itn + Br ), where Br := {z : |z| r}. This extends

Bohrs classic result on the denseness of the values (s) taken on vertical
lines (see Sect. 1.2). The other results of Good are too complicated and too
lengthy to be given here. He proved some kind of quantitative version of
Voronins universality Theorem 1.7 but his dissection of the target function
leads to certain function spaces which are not very well described yet. Recently,
Garunktis [86] proved another, more satisfying eective universality theorem
along the lines of Voronins proof in addition with some of Goods ideas. In
particular, his remarkable result shows that if f (s) is analytic in |s| 0.05
with max|s|0.05 |f (s)| 1, then for any 0 <  < 12 there exists a
  
0 exp exp 1013 (9.3)
 9.2 Upper Bounds for the Density of Universality 169

such that    
 
max log s + 3 + i f (s) < ,
|s|0.0001  4 

and further
     
1  3 
lim inf meas [0, T ] :  
max log s + + i f (s) < 
T T |s|0.0001  4
 13 
exp  . (9.4)

The original theorem is too complicated to be given here. Laurinikas [193]

(see also [89]) found another approach which gives conditional eective results
subject to certain assumptions on the speed of convergence of the related limit
distribution. However, the rate of convergence of weakly convergent probabil-
ity measures related to the space of analytic functions is not understood very
well.
All these attempts to quantify universality are remarkable but the given
quantitative universality results obtained so far have led to more open prob-
lems than they actually solved.

9.2 Upper Bounds for the Density of Universality

We shall prove eective upper bounds for the upper density of universality.
Denote by Br the closed disc of radius r > 0 with center in the origin.
For a meromorphic
 function L(s), an analytic function f : Br C with xed
r 0, 14 , and a positive , we dene the densities
     
1  3 
d(, f, L) = lim inf meas [0, T ] : max L s + + i f (s) <  ,
T T |s|r 4

and
     
1  3 
d(, f, L) = lim sup meas [0, T ] : max L s + + i f (s) <  .
T T |s|r 4

For continuity sets we do not have to distinguish between the lower and the
upper density of universality, since in this case, by the Portmanteau Theo-
rem 3.1, the limit exists.
For suciently large classes of functions L(s) and of functions f (s) we
shall prove eective upper bounds for the upper density d(, f, L) which tend
to zero as  0. For this purpose we consider analytic isomorphisms f : Br
B1 , i.e., the inverse f 1 exists and is analytic. Obviously, such a function f
has exactly one simple zero in the interior of Br . By the Schwarz lemma
(from the classical theory of functions, see [352, Sect. 5.2]) it turns out that
such a function f has the representation
170 9 Eective Results

s
f (s) = r exp(i) with R and || < r. (9.5)
r2 s

Denote by Ar the class of analytic isomorphisms from the closed disk Br (with
xed 0 < r < 14 ) to the unit disk. Further, let NL (1 , 2 , T ) count the number
of zeros of L(s) in 12 < 1 < < 2 < 1, 0 t < T (counting multiplicities).
The main result of this section is

Theorem 9.1. Let f Ar . Assume that L(s) is analytic in 34 r except

for at most O(T ) many singularities inside 34 r, 0 t  T , as T ,
and that d(, f, L) > 0 for all  > 0. Then, for any  0, 2r
1
4 + Re || ,
1

8r3 
d(, f, L) (9.6)
r2 ||2
 
1 3 3
lim sup NL + Re 2r, + Re + 2r, T .
T T 4 4

This theorem relates the density of universality to the value-distribution of L.

Note that one can obtain similar estimates for other c-values instead of c = 0
whenever c lies in the interior of Br . Since too many well distributed zeros of
L(s) in 12 < < 1 violate the universality property, very likely the limit in
(9.6) exists in general.

Proof. The idea of proof is that the zero of f is related to some zeros of L(s)
in 12 < < 1. Since f maps the boundary of Br onto the unit circle, Rouchs
Theorem 8.1 implies the existence of one simple zero of L(z) in
 
3
K := z = s + + i : s Br
4

whenever
   
 3 
max L s + + i f (s) <  < 1 = min |f (s)|.
 (9.7)
sBr 4 |s|=r

Recall that, in the language of Sect. 8.2, the zero of L(s) is generated by the
zero of f (s).
Universality is a phenomenon that happens in intervals. We prove an upper
bound for the distance of dierent shifts generating the same zero of L(s):

Lemma 9.2. Suppose that a zero of L(s), generated by , lies in two

dierent sets K1 and K2 . Then

8r3 
|1 2 | < .
r2 ||2
 9.2 Upper Bounds for the Density of Universality 171

Proof. Suppose that there exist complex numbers

3
sj = Re + itj Br ,
4
and j R for which
 
3
L sj + + ij = 0 for j = 1, 2,
4
such that
3 3
= s1 + + i1 = s2 + + i2 .
4 4
In view of (9.5),
r2 ||2
|f (s2 ) f (s1 )| = r |s2 s1 |.
|r2 s1 ||r2 s2 |
We deduce from (9.7) that |f (sj )| <  for j = 1, 2, and therefore

4r3 8r3 
|1 2 | = |t2 t1 | |f (s2 ) f (s1 )| < ,
r2 ||2 r2 ||2
which proves the lemma. 

We continue with the proof of the theorem. Denote by Ij (T ) the disjoint
intervals in [0, T ] such that (9.7) is valid exactly for

Ij (T ) =: I(T ).
j

Using Lemma 9.2, in every interval Ij (T ), there lie at least

) 2 &
r ||2 r2 ||2
1+ meas Ij (T ) meas Ij (T )
8r3  8r3 

zeros of L(s) in the strip 12 < < 1. Therefore, the number N (T ) of such
zeros satises the estimate
8r3 
N (T ) meas I(T ). (9.8)
r2 ||2
The next step is to replace N (T ) by the zero counting function appearing in
the theorem.
The value distribution of L(z) in K is ruled by that of f (s) in Br . This
gives a restriction on the real parts of the zeros .
Lemma 9.3. Let be a zero of L(s) generated by . Then
 
 
Re 3 Re  < 2r.
 4 
172 9 Eective Results

Proof. Let s Br . First of all, we consider the case |f (s)| . Then, in view
of (9.7),
     
   
L s + 3 + i  |f (s)| f (s) L s + 3 + i  > 0.
 4   4 

Since (9.5) implies

| s|
|f (s)| , (9.9)
2r
we obtain for s = 34 i by (9.7) that
     
   
 3 i  2r f 3 i 
 4   4 
  
 3 
2r|L()| + max f (s) L s + + i 

sBr 4
< 2r,

which yields the estimate of the lemma by taking the real parts.
If |f (s)| < , then we may deduce the desired estimate directly from (9.9).
The lemma is proved. 


Now we are in the position to nish the proof of the theorem. In view of
Lemma 9.3 we nd
 
3 3
N (T ) NL + Re 2r, + Re + 2r, T , (9.10)
4 4

where NL is the zero-counting function of L. On the other side, since

d(, f, L) > 0, for any > 0, there exists an increasing sequence {Tk } with
limk Tk = such that

meas (I(Tk )) ( d(, f, L) )Tk .

Consequently, this together with (9.10) leads in (9.8) to

 
8r3  3 3
NL + Re 2r, + Re + 2r, Tk ( d(, f, L) )Tk .
r2 ||2 4 4

Sending 0 yields the estimate (9.6) of the theorem. Since the set of
singularities of L(s) in 34 r has zero density but d(, f, L) > 0, the
singularities do not aect the above reasoning. The theorem is proved. 


In the case of the Riemann zeta-function we can get a slightly stronger

result. For this purpose we apply Theorem 9.1 to L(s) = log (s). In view of
the density Theorem 1.2 the set of singularities of log (s) has density zero.
As shown by Bohr and Jessen [32] (Hilfssatz 6 to Theorem 1.5), the limit
 9.2 Upper Bounds for the Density of Universality 173
 
1 3 3
lim Nlog + Re 2r, + Re + 2r, T (9.11)
T T 4 4

exists and tends to zero as  0. Now we proceed as above and obtain, for
f Ar ,
d(, exp f, (s)) = O().
Steuding [350] extended the above argument in order to obtain upper bounds
for the density of universality with respect to approximation of a rather general
class of functions g(s) = exp f (s); however, the implicit constants cannot be
given explicitly. Here we sketch the new idea in the case of the zeta-function.
Assume that g(s) is a non-constant, non-vanishing analytic function
dened on Br . Then there exists a complex number c in the interior of g(Br )
(which is not empty since g(s) is not constant) such that
 
g(s) = c + (s c ) + O |s c |2 (9.12)

for some c of modulus less than r and some = 0; this means that c is
a c-value of g(s) of multiplicity one. To see this suppose that for all c in the
interior of g(Br ) the local expansion is dierent than (9.12), i.e., g  (s) vanishes
identically in the interior. Then g is a constant function, a contradiction to
the assumption of the theorem.
Now suppose that
    
 3 
max  s + + i c {g(s) c} < min |g(s) c|.

|s|=r 4 |s|=r

Then, by Rouchs theorem, (z) has at least one c-value c in {z = s+ 34 +i :

|s| < r}. We rewrite the latter inequality as
   
 3 
max s + + i g(s) <  min |g(s) c|.
 (9.13)
|s|r  4 |s|=r

Since the zeta-function is universal, the rst inequality holds for a set of
with positive lower density. The second one follows for suciently small  from
the fact that c = g(c ) has positive distance to the boundary of g(Br ). Thus,
a c-value of g(s) generates many c-values of (z).
Assume that c = sj + 34 + ij with |sj | < r for j = 1, 2. It follows from
(9.13) that
|g(c ) g(sj )| = |c g(sj )| < . (9.14)
Since g  (c ) = = 0, there exists a neighborhood of c where the inverse
function g 1 exists and is a one-valued continuous function. By continuity,
(9.14) implies
|sj c | < = (), (9.15)
where () tends with  to zero; since g(s) behaves locally as a linear function
by (9.12), we have () . Now (9.15) implies
174 9 Eective Results

|2 1 | = |s1 s2 | |s1 c | + |s2 c | < 2. (9.16)

Now we can proceed as in the proof of Theorem 9.1 in addition with (9.11).
This yields:

Theorem 9.4. Suppose that g(s) is a non-constant, non-vanishing analytic

function dened on |s| r, where r (0, 14 ). Then, for any suciently small
 > 0,
d(, r, g, ) = O().

Thus, the decay of d(, g, ) with  0 is more than linear in  for any
non-constant, non-vanishing analytic function g.

9.3 Value-Distribution on Arithmetic Progressions

Now we consider the special case of discrete universality. The argument in
the proof of Theorem 9.4 which gave us a factor  for the upper bound does
not apply if we consider discrete shifts and so, in general, we do not get an
upper bound which tends with  to zero (as in Lemma 9.2 or (9.16)). Anyway,
for the zeta-function we obtain via Reichs discrete universality Theorem 5.16
and (9.11)
     
1  3 

lim sup  1 n N : max  s + + in g(s) <  = O(1) 
N N |s|r 4
(9.17)
for any real = 0, as  0. This is of interest with respect to an estimate of
Reich concerning small values of Dirichlet series on arithmetic progressions.
Let F (s) be a Dirichlet series, not identically zero, which has a half-plane of
absolute convergence > a , an analytic continuation to > m (m < a )
except for at most a nite number of poles on the line = a , such that
its mean square exists and F (s) is of nite order of growth in any closed
strip in m < < a . In [307], Reich proved under these assumptions (which
are rather similar to our axioms (ii) and (iii)) for any > m , = a , any
suciently small  > 0, and any real , neither being equal to zero nor of the
form 2 cos( qr ) with positive integers , q, r and q = r, that the relation

1
lim sup  {1 n N : |F ( + in)| < } < 1
N N

holds. In particular, it follows that F ( + in) cannot converge to zero as

n , and hence sn = + in cannot be a sequence of zeros of F (s). It
should be noticed that Reichs theorem also includes estimates for c-values on
arithmetic progressions (since with F (s) also F (s)c satises the conditions).
In the special case of the Riemann zeta-function we note the following
improvement of Reichs theorem:
 9.3 Value-Distribution on Arithmetic Progressions 175

Theorem 9.5. Let c be any constant and ( 12 , 1), and = 0 be real. Then

1
lim lim sup  {1 n N : |( + in) c| < } = 0.

0 N N
In particular, there does not exist an arithmetic progression sn = + in
(with and as in the theorem) on which (s) converges to any complex
number c.
We sketch the easy proof. Let g(s) be a non-constant, non-vanishing,
analytic function dened on a small disk centered at ( 12 , 1) such that
its closure lies inside the strip of universality for the zeta-function. Further
assume that
|g(s) c| < ;
this choice for g(s) is certainly possible for any given complex number c. By
the triangle inequality,

|( + in) c| |( + in) g(s)| + |g(s) c|

for any s. Hence, applying (9.17) yields

1
lim sup  {1 n N : |( + in) c| < 2}  .
N N
This is the assertion of the theorem.
An alternative proof can be given by using the deep Hauptsatz I of Bohr
and Jessen [32]; this approach does not depend on the universality property
of (s). As a matter of fact, this theorem may also be used to obtain the
estimate d(, r, g, )  2 for constant functions g  0.
There are remarkable results for a related problem. Putnam [297, 298]
showed that (s) does not have an innite vertical arithmetic progressions
of zeros (or even approximate zeros). Lapidus and van Frankenhuijsen [179,
Chap. 9] gave a dierent proof of Putnams theorem. Watkins (cf. [80]) was the
rst to give upper bounds for the length of such arithmetic progressions (valid
for any Dirichlet L-functions). Recently, van Frankenhuijsen [80] improved
these bounds by showing that

( + in) = 0 for 0 < |n| < N

with , > 0 and N 2 cannot hold for

 (1/)1

N 60 log
2

(his method also applies to Dirichlet L-functions). It is conjectured that there

are no arithmetic progressions at all; there are even no zeros known of the
form 12 + i and 12 + i2. It is conjectured that the ordinates of the nontrivial
zeros of (s) are linearly independent over Q. Ingham [140] showed that this
176 9 Eective Results

conjecture implies large values for the sum of the values of the Mbius -
function:
M (x) M (x)
lim inf 1/2 = and lim sup 1/2 = +,
x x x x
which would improve the unconditional bounds (3.1). The methods of
Putnam, Lapidus and van Frankenhuijsen do not apply to c-values.

9.4 Making Universality Visible

We return to the problem of eectivity in the universality theorem for (s).
Comparing the lower bound (9.4) of Garunktis [86] from Sect. 9.1 with the
upper bounds of Steuding [342, 350] from Theorem 9.4, it makes sense to ask
which estimate is more close to the truth. If a given function g(s) is suciently
nice, i.e., if its logarithm f (s) satises the condition of Garunktis theorem,
then  
exp 13  d(, g, ) d(, g, ) = O().
Given a positive  and a suciently small disk K located in the right half of
the critical strip, in principle, the estimate (9.3) allows to nd algorithmically
an approximating such that
max |L(s + i ) g(s)| < ;
sK

however, we cannot expect a reasonable running time for such an algorithm

when  is small. This idea was indeed considered in a project by Garunktis,
leeviieneSteuding and Steuding. For certain smooth functions g(s) and
rather large values for  approximating shifts were computed. Quite many
of these were found. However, it is impossible to deduce any information
about the density of universality as long as the running time of the underlying
algorithm cannot be signicantly improved. Nevertheless, we shall illustrate
this attempt toward eective universality by some data.
Our rst example (Fig. 9.1) is the exponential function on a small disk
centered at the origin. For example, we have
   
 3 
max  s + + 12 963 i exp(s) < 0.05.
|s|0.006 4
The shift is a positive integer since the discrete variant of universality was
used in order to simplify the algorithm.
In view of the assumptions on the set K in the universality Theorem 1.9,
we may choose K to be a line segment in the right half of the critical strip; in
this case the interior of K is empty and thus the target function needs only
to be continuous. Here is an example for such a function (Fig. 9.2):
  ) &
1  3 3 1 3 1
g(s) = + s  for s I = , + .
2 4 4 10 4 10
It was computed that
max |(s + 411 744i) g(s)| < 0.05.
sI
 9.4 Making Universality Visible 177

1.005

1.0025

1 2 3 4 5 6
0.9975

0.995

0.9925

0.99

0.005

1 2 3 4 5 6

0.005

0.01

Fig. 9.1. (s + 34 + 12 963 i) exp(s) for s = 0.006 exp(i) with 0 2. On the

left the real parts, on the right the imaginary parts are plotted; the zeta-function is
given in black, exp in grey

0.6

0.58

0.56

0.54

0.52

0.65 0.75 0.8 0.85

3 1
Fig. 9.2. (s + 4
+ i 411 744) g(s) := 2
+ |s 34 |; the zeta-function in black, g(s)
in grey
178 9 Eective Results

9.5 Almost Periodicity in the Half-Plane

of Absolute Convergence
Recall the notion of almost periodicity from Sect. 8.2. Bohr proved that every
Dirichlet series f (s), having a nite abscissa of absolute convergence a , is
almost periodic in the half-plane > a (Theorem 8.2); i.e., for any given
pair of positive numbers and , there exists a length  = (f, , ) such that
every interval of length  contains a number for which

|f ( + it + i ) f ( + it)| <

holds for any a + and all t. In general, it is a problem to nd explicitly

an admissible length (f, , ). Here we are interested in giving eective bounds
for these lengths in the case of polynomial Euler products; i.e., we consider
functions of the form

m  1
a(n)   j (p)
L(s) = = 1 , (9.18)
n=1
ns p j=1
ps

where the j (p) are complex numbers with |j (p)| 1; this is axiom (iv) in
addition with a bound for the local roots which is equivalent to the Ramanujan
hypothesis (i) by Lemma 2.2. Clearly, the Dirichlet series coecients a(n) are
multiplicative and satisfy


a(n) = j (p)kj  n

p|n k1 ,...,km 0
k1 ++km =p (n)

for any  > 0. Any such Euler product L(s) is regular and zero-free for > 1
and we may dene its logarithm log L(s) (by xing any of the single-valued
branches of the logarithm). This logarithm has also a Dirichlet series expan-
sion, for > 1 given by



m

j (pk )
log L(s) = . (9.19)
p j=1 k=1
kpks

The determination of an admissible length (log L, , ) is closely related to

simultaneous diophantine approximation. First of all, we shall prove the fol-
lowing transfer theorem:

Theorem 9.6. Let M (x, ) be a positive function, dened for 0 <  12 and
x 2, such that every interval (t1 , t2 ) R of length M (x, ) contains a
number for which
 
 log p 
 
 2  <  for all p x, (9.20)
 9.5 Almost Periodicity in the Half-Plane of Absolute Convergence 179

where z denotes the minimal distance of a real number z to the nearest

integer and p is prime. Suppose that L is given by (9.18) and let > 0. Then,
for suciently small ,
 1/ 

(log L, , ) = M , ( + O(1)) ,
5m 5m
  m 

(L, , ) =  log L, ,
1+
 m/ 1/  m 
1+
=M , ( + O(1)) .
5m 1+ 5m

This holds uniformly in 0 > 0.

We argue quite similar as Bohr [30] in his proof for almost periodicity of
Dirichlet series in their half-plane of absolute convergence.

Proof. It follows from (9.19) that

m
 
j (p)k 1
log L(s) log L(s + i ) = 1 .
p j=1 k=1
kpks pik

Since the j (p) all have absolute value less than or equal to one, we obtain

| log L(s) log L(s + i )| (9.21)

 
1  
m + + 1 1  ,
kpk  pik 
px ky px k>y p>x k=1

where x and y are parameters 2, which will be chosen later. First of all, we
shall bound the rst sum on the right-hand side for some values of .
In order to nd real numbers such that pik lies for p x and k y
suciently close to 1 we apply (9.20). It follows that, for any  (0, 12 ] and any
real number t1 , there exist integers xp and a real number (t1 , t1 + M (x, ))
such that
| log p 2xp | < 2 for all p x.
In particular,
(2k)2
cos(k log p) > cos(2k) 1 ,
2
and
sin(k log p) < sin(2k) 2k
for all primes p x, provided that k y < 1
4
. Then we obtain

5
|1 pik |2 = (1 cos(k log p))2 + sin2 (k log p) < (2k)2 (9.22)
4
180 9 Eective Results

for p x and k y. This yields

 


1  

1
1 1  < 5 . (9.23)
kp k  p i  pk
px ky px ky

In order to estimate the double sum on the right, note that by the uniqueness
of the prime factorization of the integers


1
1
.
pk y
n
px ky nx

Since

z
z  n  z
1 du du 1
< = < ,
n=2
n
n=2 n1
u
1 u 1
we may replace (9.23) by


 
1  
1 1  < 5 . (9.24)
kpk  p  1
i
px ky

Next we have to estimate the second and the third term on the right-hand
side of (9.21). For the sake of simplicity we shall not give the optimal and
rather complicated bounds. Later, it will turn out that these rough estimates
do not eect our result.
First of all,
   k


1  

1 2

1
1 1  2 .
kpk  pik  y pk y 2
px k>y px k>y px k>y

It is easily seen that the sum over k is less than

2k 2y 2 = 21y .
k>y =0

Thus, we have  


1  1  4(x)
1 < . (9.25)
kpk  pik  y2y
px k>y

It remains to bound
 



1  1 
1

1
1 ik  2 + . (9.26)
p>x
kpk  p p>x
p p>x kpk
k=1 k=2

We start with the second term on the right-hand side. Obviously,

 k
1 1

1 1
1
< = .
p>x
kp k 2 p>x p 2 p>x p 1
k=2 k=2
 9.5 Almost Periodicity in the Half-Plane of Absolute Convergence 181

Since
1 1 p 1 x 1 x 2
= < <
p 1
p p 1
p x 1
p x1 p
for p > x, it follows that




1
1
k
< .
p>x k=2
kp p>x
p

This gives in (9.26)

 
1
1  1 
k  1 ik  <4 . (9.27)
p>x k=1
kp p p>x
p

To estimate the sum on the right we make use of the prime number Theo-
rem 1.1. By partial integration we obtain

1    
u
= u d(u) = (1 + O(1)) u du.
p>x
p x x log u

The integral is equal to

 
log u 1 1 x1
u du < u
du = .
x (log u)2 log x x ( 1) log x

Substituting this estimate in (9.27) yields

1  
 1
1 1  < (4 + O(1))x . (9.28)
kpk  pik  ( 1) log x
px k=1

In view of (9.24), (9.25), and (9.28) we obtain in (9.21)

| log L(s) log L(s + i )|


5 4(x) (4 + O(1))x1
<m + +
1 y2y ( 1) log x

as x tends to innity. Taking the maximum over 1 + and incorporating

(1.1) we nd

max | log L(s) log L(s + i )|

1+

5 (4 + O(1))x (4 + O(1))
<m + + (9.29)
y2y log x x log x

for any which satises (9.20). For a suitable choice of our parameters x, y
we can make the right-hand side of (9.29) as small as we please. Let
182 9 Eective Results
 
1/ 1
x= and y = 1+ log ; (9.30)

then the condition 4y < 1 for (9.22) is fullled for suciently small . It
follows that:

5
max | log L(s) log L(s + i )| < m ( + O(1)) =: .
1+
This proves the assertion of the theorem for log L.
Since |j (p)| 1, we have for > 1
m 
 1  m
|j | 1
|L(s)| 1 1 = ()m .
j=1 p
p p
p

Hence
max |L(s)| (1 + )m .
1+

Furthermore,



1 1+
(1 + ) = 1 + <1+ u(1+) du = .
n=2
n1+ 1

Obviously,

L(s) L(s + i ) = L(s)(1 exp{log L(s + i ) log L(s)}).

It thus follows that

  m 

(L, , ) =  log L, , .
1+

This implies the assertion of the theorem for L. 



We are interested in giving eective estimates. Application of Weyls

approximation theorem, Lemma 1.8, would lead to an explicit lower bound
for the density of shifts L(s + i ) approximating L(s). However, this approach
does not give an upper bound for the rst approximating . We can solve
this problem by applying eective results from the theory of inhomogeneous
diophantine approximation which we derive in the following section.

9.6 Eective Inhomogeneous Diophantine

Approximation
Bohr and Landau [33, 36] (see also [353, Sect. 8.8]) proved for given N the
existence of a real number with 0 exp(N 6 ) such that
 9.6 Eective Inhomogeneous Diophantine Approximation 183

1
cos( log p ) < 1 + for 1 N,
N
where p denotes the th prime number. This can be regarded as a rst
eective version of Kroneckers approximation theorem, with a bound for
similar to the bound in Dirichlets approximation theorem. Using the idea of
Bohr and Landau in connection with Bakers estimate for linear forms, Rieger
[309] proved the following remarkable discrete approximation theorem:

Theorem 9.7. Let v, N N, b Z, 1 , U R. Let p1 < . . . < pN be

prime numbers (not necessarily consecutive) and

u Z, 0 < |u | U, R for 1 N.

Then there exist h Z, 0 N, and an eectively computable number

C = C(N, pN ) > 0, depending on N and pN only, so that
 u 
  1
h0 log p h  < for 1 N, (9.31)
v
and b h0 b + (2U v)C .

However, we are interested in an explicit bound for C. Therefore, we repeat

Riegers proof and add in the crucial step a result on an explicit lower bound
for linear forms in logarithms due to Waldschmidt.
Let K be a number eld of degree d over Q and denote by LK the set of
logarithms of the elements of K \ {0}, more precisely,

LK = { C : exp() K}.

If is an algebraic number with minimal polynomial P (X) over Z, then dene

the absolute logarithmic height of by

1 1
h() = log |P (exp(2i))| d;
d 0

note that h() = log || for every integer = 0, and h(0) = 0. Waldschmidt
[365] proved

Theorem 9.8. Let  LK and Q for 1 N , not all equal zero.

Dene a = exp( ) for 1 N , and

= 0 + 1 log a1 + + N log aN .

Let E, W and V , 1 N, be positive real numbers, satisfying

1
W max {h( )}, V1 . . . VN ,
1N d
184 9 Eective Results
 
| log a |
V max h(a ), for 1 N,
d
and   
4 dV
1 < E min exp(V1 ), min .
1N | log a |
Finally, dene V+ = max{V , 1} for = N and = N 1, with V1+ = 1 in
the case N = 1. If = 0, then
 

N
N 1
|| > exp c(N ) d N +2 + +
(W + log(E dVN )) log(E dVN 1 )(log E) V ,
=1

where c(N ) 28N +51 N 2N .

In conjunction with Riegers Theorem 9.7 this leads to

Theorem 9.9. With the notation and the assumptions of Theorem 9.7 there
exists an integer h0 such that (9.31) holds and
 

N
b h0 b + 2 + exp 28N +51 N 2N (1 + 2 log pN )(1 + log pN 1 ) log p
=2
((3U (N + 2) log pN )4 + 2)N +2 ;

in the special case that the primes p1 < p2 < . . . < pN are the rst N succes-
sive prime numbers, we obtain the stronger inequality

b h0 b + (U )(4+)N exp N (2+)N

for any  > 0 and suciently large N .

Proof. For t R dene

N u

f (t) = 1 + exp(2it) + exp 2i t log p .
=1
v

With 1 := 0, 1 := 0, 0 := 1, 0 := 0 and := uv log p , 1 N, we

have

N
f (t) = exp(2i(t )).
=1

By the multinomial theorem,

 

k!

N
f (t)k = exp 2i j (t ) .
j0
j1 ! jN ! =1
j1 ++jN =k
 9.6 Eective Inhomogeneous Diophantine Approximation 185

Hence, for 0 < B R and k N,

 b+B
J := |f (t)|2k dt
b

k!
k!
=  ! j !
j0
j1 ! jN ! j
0
j1 N
j1 +...+jN =k j
++j
=k
1 N

   
b+B

N

N
exp 2i (j j ) t (j j ) dt.
b =1 =1

Since the logarithms of prime numbers are linearly independent, the sum

N
(j j )
=1

vanishes if and only if j = j for 1 N . Thus, integration gives

 b+B   N 



N
 
exp 2i (j j ) t (j j ) dt = B
b =1 =1

if j = j for 1 N , and
   N  
 b+B


N 
   
 exp 2i (j j ) t (j j ) dt
 b 
=1 =1
 N 1
1 


 (j j ) 
 =1


if j = j for some {1, 0, . . . , N }. In the latter case, by Bakers esti-

mate [13] for linear forms, there exists an eectively computable constant A
such that  1

N 
 
 (j j )  < A.
 
=1

Setting 0 = j0 j0 ,
= uv (j j ), and a = p for 1 N , we nd,
with the notation of Theorem 9.8,

N
= (j j ) .
=1

We may take E = 1, W = log pN , V1 = 1 and V = log p for 2 N . If

N 2, Theorem 9.8 gives
186 9 Eective Results
 

N
|| > exp 28N +51 N 2N (1 + 2 log pN )(1 + log pN 1 ) log p .
=2

Thus we may take

 

N
8N +51 2N
A = exp 2 N (1 + 2 log pN )(1 + log pN 1 ) log p . (9.32)
=2

Hence we obtain

 2
k!
J B
j0
j1 ! jN !
j1 ++jN =k

A
k!
k!
 ! j ! . (9.33)
j0
j1 ! jN ! j
0
j1 N
j1 +...+jN =k j
++j
=k
1 N

Since

1 (k + 1)N +2 ,
j0
j1 ++jN =k

an application of the Cauchy Schwarz-inequality to the rst multiple sum and

of the multinomial theorem to the second multiple sum on the right-hand side
of (9.33) yields
2
 

B A k!
J
(k + 1)N +2 j0
j1 ! jN !
j1 ++jN =k
 
B A
(N + 2)2k .
(k + 1)N +2

Setting B = A(k + 1)N +2 and with [b, b + B] given by

|f ( )| = max |f (t)|,
t[b,b+B]

we obtain
B(N + 2)2k
J B|f ( )|2k .
2(k + 1)N +2
This gives

(N + 2)2 log k
|f ( )| > N + 2 2, where := ; (9.34)
3k
note that < 1 for k 11. By denition
 9.6 Eective Inhomogeneous Diophantine Approximation 187

N
f (t) = 1 + exp(2i(t )) + exp(2i(tm m )).
m=0
m=

Therefore, using the triangle inequality,

|f (t)| N + |1 + exp(2i( ))| for 0 N

and arbitrary t R. Thus in view of (9.34)

|1 + exp(2i( ))| > 2 2 for 0 N.

If h denotes the nearest integer to , then



| h | < for 0 N.
2

For = 0 this implies | h0 | < . Thus, replacing by h0 yields
 

|h0 h | < 1 + max | | for 1 N.
1N

Putting k = [(3wU (N + 2) log pN )4 ] + 1, we get

b 1 h0 b + 1 + B = b + 1 + A([(3U (N + 2) log pN )4 ] + 2)N +2 .

Substituting (9.32) and replacing b 1 by b, the rst assertion of Theorem 9.7

follows with the estimate of Theorem 9.9; the second estimate can be proved
by standard estimates involving the prime number Theorem 1.1. 


Now we are in the position to present our eective result on the almost
periodicity for the polynomial Euler products under consideration. For this
purpose we shall apply Theorem 9.9 with x = pN . Recall that m was the
degree of each Euler factor in the polynomial Euler product for L. For the
sake of simplicity we use the rough estimate N = (x) x which gives
 
M (x, ) = 4x exp x2x .

Hence, by Theorem 9.6,

Corollary 9.10. Let L(s) be given by (9.18) and > 0. For suciently small
 > 0,

C(m+1)/ 1/ 
(m + 1)/ 1/
(L, , ) = (C m+1 )C
exp C (m+1)/ 1/ ,

where C is an absolute positive constant, depending only on m.

188 9 Eective Results

9.7 c-Values Revisited

We return now to a topic already touched in Chap. 7. Almost periodicity

has a strong impact on the value distribution. In fact, if a value is taken by
a polynomial Euler product in the half-plane of absolute convergence, then
it is taken innitely often, with a certain regularity which reects almost
periodicity.

Theorem 9.11. Assume that L is given by (9.18). Let N (T ; a, b, c) count the

number of c-values of L(s) (i.e., the roots of the equation L(s) = c) in the
rectangle a < < b, 0 < t < T (counting multiplicities) and assume a 1. If
N (T ; a, b, c) 1, then, for suciently small positive and ,

1 1
lim inf N (T ; a, b, c) > 0.
T T (L, , )

Since the Euler products under observation have no zeros in their half-plane of
convergence, we have N (T ; 1, , 0) = 0; however, we expect N (T ; 1, , c) 1
for any c = 0 (as in the case of the zeta-function).
Again we shall use Rouchs theorem in order to localize c-values.

Proof. Suppose that N (T ; a, b, c) 1. Then there exists a complex number

c with a < Re c < b and 0 < Im c < T such that L(c ) = c. Locally, we
have
 
L(s) c = (s c )k + O |s c |k+1 , (9.35)

where is a non-zero complex number and k N. Now dene

K = {s C : |s c | }.

For suciently small > 0 the disk K is contained in the rectangle a < <
b, 0 < t < T . Now assume that there is some real number for which

max |L(s) c {L(s + i ) c}| < < min |L(s) c|; (9.36)
|sc |= |sc |=

the second inequality holds for suciently small and . By the maximum
principle the left-hand side equals

max |L(s) L(s + i )|.

sK

By almost periodicity, the latter quantity can be made arbitrarily small for a
set of with positive lower density. Now, for any satisfying (9.36), Rouchs
Theorem 8.1 implies the existence of a zero c of L(s) c in

K + i := {s C : |s i c | }.
 9.7 c-Values Revisited 189

In view of (9.35) the numbers c and c are intimately related. In fact,

> max |L(s) L(s + i )|

sK
|L(c i ) L(c )| = |L(c i ) c|
= || |c i c |k + O( k+1 ).

Hence,  1/k

|c i c | < + O 1+(1/k) .

||
In particular,
 1/k  1/k

Re c 2 < Re c < Re c + 2 ,
|| ||
and  1/k

|Im c ( + Im c )| < 2
||
for small = O(1/(k+1) ). Hence, for suciently small and , the root c of
the equation L(s) = c is contained in the open rectangle a < < b, 0 < t < T .
By the almost periodicity of L(s), we can nd a real number satisfying (9.36)
in any interval of length (L, , ). This implies the estimate of the theorem.


Taking into account Bohrs results on the value-distribution of the zeta-
function in the half-plane of absolute convergence (see Theorem 1.3), the
condition of Theorem 9.11 is fullled for any non-zero complex number c, i.e.,
N (T ; 1, , c) 1. Hence, for xed c = 0 and suciently small , > 0, we
nd for the number of c-values of (s) in the half-plane > 1 the estimate
1 1
lim inf  {c : Re c > 1, 0 < Im c < T, L(c ) = c} ,
T T (, , )
where

C2/ 1/ 
2 C 2/ 1/ 2/ 1/
(, , ) = (C ) exp C (9.37)

by Corollary 9.10.
We apply Theorem 9.6 and (9.37) to prove the existence of large values of
the zeta-function in some distance of the pole at s = 1. Recall that
1
(s) = + O(1)
s1
for s 1. Let c be a large positive real number; then there exists a real
number c with
1
1 < c 1 + for which c = (c ).
c
190 9 Eective Results

Put  := (, , ). By Theorem 9.6, for suciently small , we can nd a real

number (, 2) such that
1
|( + i )| |(c )| = c =:

for some > 1. In view of (9.37) we obtain after a short computation
1 log log log
.
log log log log
This leads to the estimate
|( + it)|
inf > 0.
>1,t> log log log t
It should be noted that this estimate for large values of the zeta-function
in the half-plane of absolute convergence is quite far from the best possible
estimate due to Bohr and Landau [34]; they proved the existence of innitely
many s = + it with 1+ and t + for which
|(s)|  log log t.
By a standard argument based on the PhragmnLindelf principle it follows
that
1
(1 + it) = (log log t) and = (log log t). (9.38)
(1 + it)
Under assumption of Riemanns hypothesis these estimates give the true order
of growth (see [353, Sect. 14.8]).

2.5

1.5

0.5

20 40 60 80 100

Fig. 9.3. |(1 + it)| for t (0, 100]. The limit superior of the peaks grows with t to
innity, the limit inferior of the valleys tends to zero
 9.7 c-Values Revisited 191

Theorem 9.9 is due to Steuding [344] and was used to prove -results for
Dirichlet L-functions in the half-plane of absolute convergence. It was shown
that for any real there are innitely many values of s = + it with 1+
and t + such that
log log log t
Re {exp(i) log L(s, )} log + O(1)
log log log log t
and
Re {exp(i) log L(1 + it, )} = (log log log log t).
It seems that also these estimates for L(s, ) are one iterated logarithm o
the true order. However, in contrast to -results, these large values are taken
frequently.
10
Consequences of Universality

Betrachten wir beispielsweise die Klasse derjenigen Funktionen, die

sich durch gewhnliche oder partielle Dierentialgleichungen charak-
terisieren lassen. In dieser Klasse von Funktionen kommen, wie wir
sofort bemerken, gerade solche Funktionen nicht vor, die aus der
Zahlentheorie stammen und deren Erforschung fr uns von hchster
Wichtigkeit ist. Beispielsweise gengt die schon frher erwhnte Funk-
tion (s) keiner algebraischen Dierentialgleichung, . . .
David Hilbert

The phenomenon of universality has many interesting consequences. There

are the classic results due to Bohr, mentioned in the introduction, as well as
extensions of Ostrowskis theorem on the non-existence of algebraic dierential
equations for Dirichlet series (which actually was part of the 18th Hilbert
problem). In this chapter, we prove generalizations for universal L-functions,
we discuss a disproof of a conjecture on certain mean-square estimates due to
Ramachandra by Andersson [1], and, nally, we study the value-distribution
of linear combinations of shifts of universal Dirichlet series.

10.1 Dense Sets in the Complex Plane

Let L S and let c be a complex number. A standard argument yields that

the number of c-values up to level T is  T provided c = 0. To see this, let
NLc (1 , 2 , T ) count the number of c-values of L(s) in the region 1 < < 2
and |t| T , i.e., NLc (1 , 2 , T ) = NLc (1 , T ) NLc (2 , T ) in the notation of
Sect. 7.2. We apply Theorem 5.14 to the function g(s) = c = 0 with a small
disk K inside the strip 1 < < 2 , where max{ 12 , 1 d1L } < 1 < 2 < 1.
For any  > 0, we nd
 
1
lim inf meas [0, T ] : max |L(s + i ) c| <  > 0.
T T sK
194 10 Consequences of Universality

By Rouchs Theorem 8.1, it follows that NLc (1 , 2 , T ) is bounded below

by a positive constant times T . In combination with the estimate (7.11) of
Theorem 7.6 we obtain
Theorem 10.1. Let L S and let c be a complex number = 0, 1. Then, for
any 1 , 2 satisfying max{ 12 , 1 d1L } < 1 < 2 < 1,

NLc (1 , 2 , T ) T.

This result should be compared with Bohrs limit Theorem 1.5.

In Sect. 8.1, we have seen that (s) cannot approximate functions having
zeros. However, the same argument as above can be used to count zeros of
the real part of (s). Garaev [83] used Voronins universality theorem in order
to detect zeros of Re (s). He showed that, for xed ( 12 , 1), the function
Re ( + it) has more than  T zeros t (0, T ); this follows directly from an
application of the universality theorem to the function g(s) = exp(Cs), where
C is a suciently large constant. Moreover, the function Re ( + it) has at
least cT sign changes in the interval (0, T ), where c is some positive constant.
Besides, Garaev proved also the upper bound T log T (the same statements
also hold for Im ( + it)).
Next we consider the multidimensional analogue and extend Voronins
Theorem 1.6 to
Theorem 10.2. Let L S and suppose that (max{ 12 , 1 dL }, 1)
1
is xed.
Then the set
{(L(s), L (s), . . . , L(n1) (s)) : t R}
lies everywhere dense in Cn .
Proof. First, for any vector (a0 , a1 , . . . , am ) Cm+1 with a0 = 0, we prove by
induction on m, that there exists another vector (b0 , b1 , . . . , bm ) Cm+1 for
which m 


m
ak k
exp bk sk
s mod sm+1 .
k!
k=0 k=0

For m = 0 one only has to choose b0 = log a0 . By the induction assumption

we may assume that with some
m 



m
ak k
exp bk sk
s + sm+1 mod sm+2 .
k!
k=0 k=0

For any ,
exp(sm+1 ) 1 + sm+1 mod s2m+2 .
Thus,
   


m

m
ak
exp k
bk s + sm+1
(1 + s m+1
) k m+1
s + s mod sm+2 .
k!
k=0 k=0
 10.2 Functional Independence 195

Let bm+1 = be the solution of the equation

am+1
a0 + = ,
(m + 1)!

 0. This shows
which exists since a0 =
m+1  m+1


ak
exp bk sk
sk mod sm+2 ,
k!
k=0 k=0

proving the claim.

Now let n1 



n1
ak k
g(s) = exp k
bk s s mod sn .
k!
k=0 k=0

Obviously, g (k) (0) = ak for 0 k n 1. Put

   
1 1 1
0 = max ,1 +1 .
2 2 dL

By the universality Theorem 5.14, for any  > 0, there exists a such that

rk
max |L(s + 0 + i ) g(s)| < 2
|s|r k!

for suciently small r. Applying Cauchys formula,

,
k! f (s)
f (k) (0) = ds
2i |s|=r sk+1

to the function
f (s) = L(s + 0 + i ) g(s)
with suciently small r, we obtain

|L(k) (0 + i ) ak | = |L(k) (0 + i ) g (k) (0)| < 

for 0 k n 1. This proves the theorem. 



10.2 Functional Independence

Hlder [135] proved in 1887 that the Gamma-function is hypertranscendental,

i.e., (s) does not satisfy any algebraic dierential equation: there exists no
polynomial P in n variables, not identically zero, such that

P z, (z),  (z), . . . , (n1) = 0

196 10 Consequences of Universality

for all z C. In Hilberts famous list of 23 challenging problems for the

twentieth century [134], originated from his lecture given at the International
Congress for Mathematicians in Paris 1900, he asked in problem 18 for a des-
cription of classes of function denable by dierential equations. In this con-
text Hilbert stated that the Riemann zeta-function is hypertranscendental;
the rst published proof for (s) was written down by Stadigh in his disser-
tation (cf. [286]). Hilbert also asked for a proof of the hypertranscendence for
the more general series


xn
(s, x) :=
n=1
ns
and proposed to use the functional equation

x (s, x) = (s 1, x).
x
This problem was solved by Ostrowski [286] as a particular case of a more
general theorem; his argument relies on a comparison of the dierential
independence with the linear independence of its frequencies. By a similar
reasoning Reich [307] proved that the Dedekind zeta-function does not satisfy
any dierence-dierential equation. Popken [295] introduced a measure for
the dierential-transcendence of (s) and other hypertranscendental Dirichlet
series (similar to transcendence measures in the theory of transcendental
numbers).
Following Voronin [362] we shall show now that universality implies the
more general concept of functional independence. We say that the functions
f1 (s), . . . , fm (s) are functionally independent if for any continuous functions
F0 , F1 , . . . , FN : Cm C, not all identically vanishing, the function


N
sk Fk (f1 (s), . . . , fm (s))
k=0

is non-zero for some values of s.

Theorem 10.3. Let L S, z = (z0 , z1 , . . . , zn1 ) Cn and suppose that
F0 (z), F1 (z), . . . , FN (z) are given continuous functions, not all identically
zero. Then


N

sk Fk L(s), L(s) , . . . , L(s)(n1) = 0

k=0
for some s C.
In particular, L S does not satisfy any algebraic dierential equation.
Proof. First, we show that if F (z) is a continuous function and if

F L(s), L(s) , . . . , L(s)(n1) = 0

identically in s C, then F vanishes identically.

 10.2 Functional Independence 197

Suppose the contrary, i.e., F (z)  0. Then there exists a Cn for which
F (a) = 0. Since F is continuous we can nd a neighbourhood U of a and a
positive  such that
|F (z)| >  for z U.
Choosing an arbitrary with max{ 12 , 1 d1L } < < 1, an application of
Theorem 10.2 yields the existence of some t for which

L( + it), L( + it) , . . . , L( + it)(n1) U,

which contradicts our assumption. This proves our claim which is the assertion
of the theorem with N = 0.
Without loss of generality we may assume that F0 (z) is not identically
zero. As above there exist an open bounded set U and a positive  such that
|F0 (z)| >  for z U.
Denote by M the maximum of all indices m for which
sup |Fm (z)| = 0.
zU

If M = 0, then the assertion of the theorem follows from the result proved
above. Otherwise, we may take a subset V U such that:
inf |FM (z)| > 
zV

for some positive . By Theorem 10.2, there exists a sequence tj , tending with
j to innity, such that

L( + itj ), L( + itj ) , . . . , L( + itj )(n1) V.

This implies
M 


  (n1) 
lim  ( + itj ) Fk (L( + itj ), L( + itj ) , . . . , L( + itj )
k
) = ,
j  
k=0

which proves the theorem. 


The zeta-function does not satisfy any algebraic dierential equation; how-
ever, another approximation property for zeta is related to a non-algebraic
dierential equation. Gauthier and Tarkhanov [95] proved that any mero-
morphic function on a compact subset K of C having at most simple poles
can be approximated by linear combinations of certain translates of the
Riemann zeta-function (not necessarily from the critical strip); if derivatives
of (s) are allowed, then arbitrary meromorphic functions can be approxi-
mated. The proof of this result does not rely on the universality property of
the zeta-function but the fact that (s) satises a certain (non-algebraic) inho-
mogeneous linear dierential equation of order one for the RiemannCauchy
operator.
198 10 Consequences of Universality

10.3 Joint Functional Independence

Now we extend the previous results by restricting on some kind of minimal
denition what universality might be. Assume that F1 (s), . . . , Fm (s) are mero-
morphic functions for > m with m < 1 satisfying the following property:
given arbitrary linearly independent continuous functions f1 (s), . . . , fm (s) on
a compact subset K of m < < 1 with connected complement, which are
analytic in the interior, for any  > 0 and all but nitely many points s0 K,
there exists a neighbourhood K of s0 with K K and a real number such
that
max max |Fj (s + i ) fj (s)| < . (10.1)
1jm sK

We then say that the family F1 , . . . , Fm is weakly jointly universal. This notion
was introduced by Sander and Steuding [315] in order to prove that families of
Hurwitz zeta-functions with rational parameters satisfying certain side condi-
tions are weakly jointly universal and functional independent.

Theorem 10.4. Suppose that F1 (s), . . . , Fm (s) are weakly jointly universal.
Then for almost all s0 with Re s0 (m , 1) (i.e., all but from a discrete set)
the image of the curve

(1) (1)
(t) := F1 (s0 + it), . . . , Fm (s0 + it), . . . , F1 (s0 + it), . . . , Fm (s0 + it)

lies dense in Cm .

We only give a sketch of the proof since the argument is in principal the
same as in the proof of Theorem 10.2. For j = 1, . . . , m and 0 k < , to
any given set of complex numbers akj with a0j = 0, there exists a polynomial
gj (s) such that

(k) 
akj = (exp(gj (s)))  . (10.2)
s=s0

Let K be the closed disk with center s0 and radius r > 0 such that K lies
inside the strip m < < 1. Now we make use of our assumption that the
functions Fj (s) are weakly jointly universal. We may assume that the functions
fj (s) := exp(gj (s)) are linearly independent (if not, then we may replace the
gj (s) by gj (s)+j (s) with suitable functions j (s) such that (10.2) still holds).
Further, we may assume that the Fj (s) are jointly universal with respect to
the functions fj (s) for a small neighbourhood K of s0 , a disk with center s0
of radius , say. In particular, for any  > 0, there exists a real number
such that (10.1) holds. Now dene j (s) = Fj (s + i ) gj (s). Using Cauchys
integration formula,
,
(k) k! j (s)
j (s0 ) = ds,
2i |ss0 |= (s s0 )k+1
 10.3 Joint Functional Independence 199

we obtain the estimate

 
 (k)  k!
j (s0 ) k max |j (s)|.
 sK

This leads to
 
 (k)  k!
max Fj (s0 + i ) akj  <  k
1jm 
for k = 0, 1, . . . ,  1. Taking  suciently small the assertion follows.
Now we may use the previous theorem to prove the functional indepen-
dence for jointly universal families and their derivatives:

Theorem 10.5. Suppose that F1 (s), . . . , Fm (s) are weakly jointly universal.
Let f0 , . . . , fn be continuous functions on Cm . If

n

(1) (1)
sk fk F1 (s), . . . , Fm (s), . . . , F1 (s), . . . , F1 (s) = 0
k=0

identically in m < < 1, then fk (s) 0 for k = 0, 1, . . . , n.

Again we give only a sketch of the proof. First, we show that if f is a continuous
function on Cm and

(1) (1)
f F1 (s), . . . , Fm (s), . . . , F1 (s), . . . , Fm (s) = 0

in m < < 1, then f vanishes identically.

Suppose the contrary, i.e., f (z0 ) = 0 for some z0 Cm . Since f is con-
tinuous, there exists a neighbourhood U of z0 and some positive  such that

|f (z)| >  for z U. (10.3)

By Theorem 10.4, for some xed s0 with Re s0 (m , 1), there exists a
sequence (tj ) tending to + for which

(1)
F1 (s0 + itj ), . . ., Fm (s0 + itj ), . . ., F1 (1)
(s0 + itj ), . . ., Fm (s0 + itj ) U.

This gives the desired contradiction. So we have proved the assertion of the
theorem in the case of n = 0.
To prove the full assertion we may now assume that f0 is not identically
zero. By the same reasoning as above, there exists an open bounded set U
and a positive  such that (10.3) holds with f0 in place of f . Now denote by
the maximum of all indices 1 k m such that the supremum of all values
|fk (z)| for z U is positive. For suciently small  > 0, we may nd a subset
V U for which
inf |f (z)| > .
zV
200 10 Consequences of Universality

By Theorem 10.4, for xed s0 , there exists a sequence (tj ) tending to +

such that

(1)
F1 (s0 + itj ), . . . , Fm (s0 + itj ), . . . , F1 (1)
(s0 + itj ), . . . , Fm (s0 + itj ) V.

This implies the divergence of


(1) 
sk fk F1 (s), . . . , Fm (s), . . . , F1 (1)
(s), . . . , Fm (s)  ,
s=s0 +itj
k=0

as j , giving the contradiction. Theorem 10.5 is proved.

We conclude with an interesting result of Kaczorowski and Perelli. If P is
an arbitrary entire function of minimum type of order one on Cn D, where
D is a region in C which contains the set
 
1
s C : , |t| T {s C : > 1, |t| T }
2
for some positive T , and if the functions L1 , . . . , Ln S are linearly indepen-
dent over Q, then Kaczorowski and Perelli [159] showed that the function

P (log L1 (s), . . . , log Ln (s), s)

has innitely many singularities in the half-plane 12 .

10.4 Anderssons Disproof of a Mean-Square Conjecture

There are further applications of universality. First, we briey discuss a dis-
proof of a conjecture concerning the mean-square of Dirichlet polynomials.
The theorem of MontgomeryVaughan states that, for arbitrary complex num-
bers a1 , . . . , aN ,
 N 2
T 


N
 it 
 an n  dt  |an |2 (T + n).
0  
n=1 n=1

In the other direction, Ramachandra [299] conjectured that for any N 0

and any complex numbers a1 , . . . , aN , there exists a positive constant C such
that
 T 

N
2


 it 
 an n  dt C |an |2 .
0  
n=1 nCT

Andersson [1] gave a disproof of this and two related conjectures on moments
of Dirichlet polynomials due to Ramachandra. Actually, Andersson proved the
following statement: for any H,  > 0 and 0 < < 12 there exists an N 2
 10.5 Voronins Theorems and Physics 201

and complex numbers a2 , . . . , aN , satisfying the estimate |an | n1/2 , such

that
 

N 
 
max 1 + an nit  < . (10.4)
0tH  
n=2

The proof relies on an approximation of (s) by a Dirichlet polynomial and

Voronins universality theorem. It follows from (1.3) that, for any (0, 12 )
and for suciently large T ,
 


 1it  
(1 + it) n < (10.5)
  3
n<2T

when T t T + H. Voronins universality Theorem 1.9, applied to g(s) =

3
and K = [1 , 1 + iH], yields
  
 
max (1 + it + iT )  <
tK 3 3
for some T > 0. Thus, if we replace t by t T , then
2
|(1 + it)| <
3
for T t T + H. By (10.5), this implies the inequality
 


 
 n1it  < ,
 
n<2T

valid for T t T + H, which leads to Anderssons claim (10.4).

It follows from (10.4) that, for each T 0 and any  > 0, there exist
complex numbers a1 , . . . , aN , for which
  2
T 

N 
 
1 + an nit  dt .
0  
n=2

This disproves Ramachandras conjecture. Andersson also gave a simpler

argument based on the SzaszMntz theorem.

10.5 Voronins Theorems and Physics

We should mention applications of Voronins results on the value distrib-
ution of the zeta-function to physics. Gutzwiller [112] discovered an inter-
esting link between quantum theory, hyperbolic geometry and the Riemann
zeta-function. On a closed surface of constant negative curvature the Selberg
202 10 Consequences of Universality

trace formula expresses the relation between classical and quantum states.
Gutzwiller constructed a solution of Schrdingers equation on the punctured
torus which may be regarded as the quotient of the upper half-plane by a free
subgroup of index six in SL2 (Z). This solution represents the scattering and
reection of a particle which enters the torus through the hole. The scattering
phase shift is essentially the argument of (s) on the line = 1. The univer-
sality theorem reects the chaotic behaviour of the zeta-function inside the
critical strip, and in particular the zero-distribution is expected to be closely
related to the quantum chaos in this model.
Bitar, Khuri and Ren [24] found an interesting application of Voronins
Theorems 1.6 and 1.7 to Feynmans path integral in quantum physics. They
obtained a formula for the partition function as a discrete sum over paths
with each path labeled by an integer and given by a zeta-function evaluated
at a xed set of points in the critical strip. These points are the image of the
spacetime lattice resulting from a linear mapping.

10.6 Shifts of Universal Dirichlet Series

Recently, Kaczorowski, Laurinikas and Steuding [157] studied shifts of uni-

versal Dirichlet series with respect to universality and their value-distribution.
Assume that F(s) is analytic in some open strip, D = {s C : 1 < < 2 }
say, and that K1 , . . . , Kn are disjoint compact subsets of D with connected
complements.
-n Let g(s) be any (non-vanishing) continuous function, dened
on j=1 Kj , which is analytic in the interior; here g(s) is allowed to have
zeros if and only if F(s) is strongly universal. If now for any  > 0, there
exists a real number such that

max |F(s + i ) g(s)| < ,

sn K
j=1 j

then also
max max |F(s + i ) gj (s)| < ,
1jn sKj

where the gj (s) are dened as restriction of g(s) on Kj ; of course, equivalently,

one can consider all gj (s) being dened on some compact subset K of D with
connected complement and study shifts of F(s).
Let 1 , . . . , n be complex numbers and put

:= min Re j and := max Re j . (10.6)

1jn 1jn

Further, let K be a compact set, and dene Kj := {s + j : s K}. Then the

shifts
Fj (s) := F(s + j )
 10.6 Shifts of Universal Dirichlet Series 203

of F(s) are said to be jointly universal with respect to 1 , . . . , n if, for every
compact K with connected complement and for which the sets Kj are dis-
joint subsets of D, every family of (non-vanishing) continuous functions gj (s)
dened on K which are analytic in the interior, and for any  > 0, we have
 
1
lim inf meas [0, T ] : max max : Fj (s + i ) gj (s)| <  > 0.
T T 1jn sK

Clearly, the assumption on the Kj to be disjoint is necessary. Moreover, to

have all shifts well dened on a non-empty subset of the complex plane we
need the following necessary condition:
< 2 1 (10.7)
with and dened in (10.6) and 1 , 2 being the abscissae bounding D.
The condition Kk K = for k =  implies
k =  for k = . (10.8)
According to the foregoing remarks one can formulate the following general
Shifts Universality Principle 1. Every (strongly) universal function in D
is jointly (strongly) universal with respect to every nite sequence of shifts
1 , . . . , n satisfying (10.7) and (10.8).
For instance, (s) and (s + i) are jointly universal for any real = 0.
Now suppose that F(s) is universal in D. By the same argument as in
Sect. 10.1 it follows that then for every complex numbers 1 , . . . , n satisfying
(10.7) and (10.8), for every s0 D(1 , . . . , n ) := {s C : 1 < <
2 }, the image of the curve

(1) (1)
F1 (s0 + it), . . . , Fn (s0 + it), . . . , F1 (s0 + it), . . . , Fn (s0 + it)

lies dense in Cn . It should be noticed that in the case of the zeta-function
this implies and generalizes both statements of Voronins Theorem 1.6. More-
over, as in Sect. 10.2, this statement can be used to deduce the functional
independence for these shifts, i.e., if f0 , . . . , fm are continuous functions on
Cn , and if


m

(1) (1)
sk fk F1 (s), . . . , Fn (s), . . . , F1 (s), . . . , Fn (s) = 0
k=0

identically for all s D(1 , . . . , n ), then fk 0 for k = 0, 1, . . . , m.

Further, Kaczorowski, Laurinikas and Steuding [157] considered linear
combinations of jointly universal shifts Fj (s). Let n 2 and a1 , . . . , an be
arbitrary, non-zero complex numbers and let


n
Z(s) := aj Fj (s).
j=1

This linear combination is strongly universal:

204 10 Consequences of Universality

Theorem 10.6. Suppose that n 2 and 1 , . . . , n are complex numbers

satisfying (10.7), (10.8), and let K be a non-empty compact subset with the
connected complement such that the sets Kj = {s + j : s K} are disjoint
subsets of D. Moreover, let F(s) be universal in D and let g(s) be a continuous
function on K which is analytic in the interior of K. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max |Z(s + i ) g(s)| <  > 0.
T T sK

We sketch the proof. Dene the functions

1 1
g1 (s) = (g(s) +  + 1) and g2 (s) = ( + 1),
a1 a2
where  := maxsK |g(s)|. Then g1 (s) does not vanish for s K. Further, for
j = 3, . . . , n, dene gj (s)  with some arbitrary positive . Then, for any ,

Z(s + i ) g(s)
= a1 {F1 (s + i ) g1 (s)} + a2 {F2 (s + i ) g2 (s)}

n
. /
n
+ aj Fj (s + i ) gj (s) + aj .
j=3 j=3

Now an application of the Shift Universality Principle yields the existence of a

set of positive real numbers having positive lower density for each of which

max |Z(s + i ) g(s)|

sK

n  

n
|aj | max max |Fj (s + i ) gj (s)| +  < 2 |aj |.
1jn sK
j=1 j=1

The right-hand side can be made as small as we please.

This theorem has interesting consequences on the zero-distribution. Under
the conditions of the previous theorem, the function Z(s) can uniformly
approximate functions g(s) having zeros, so Z(s) is strongly universal.
However, for a single shift Fj (s) this is not true in general (e.g., the Riemann
zeta-function is not strongly universal as shown in Sect. 8.1).
In the sequel we assume that F(s) is not vanishing identically, has a rep-
resentation as an ordinary Dirichlet series, more precisely,


f (n)
F(s) =
n=1
ns

for > 2 , and is regular and of nite order in the strip 1 < < 2 with
bounded second moment:

1 T
lim sup |F( + it)|2 dt < ; (10.9)
T T 0
 10.6 Shifts of Universal Dirichlet Series 205

notice that these assumptions hold for the Riemann zeta-function, Dirichlet L-
functions, Lerch zeta-functions, and many other naturally appearing Dirichlet
series with 1 = 12 and 2 = 1.
To obtain some information on the zero-distribution of Z(s) we denote by
NZ (, T ) the number of zeros  = + i of Z(s) satisfying > , 0 < < T .
Then

Theorem 10.7. Let 1 , . . . , n be in pairs dierent and satisfy (10.7). Then

for xed (1 , 2 ), there exist positive constants c1 , c2 , depending
only on and Z(s), such that
1 1
c1 lim inf NZ (, T ) lim sup NZ (, T ) c2 ;
T T T T

the constants c1 and c2 are explicitly given by (10.12) and (10.15) if condition
(10.13) holds.

The method of proof relies on Theorem 10.6 and techniques from Sects. 7.1
and 8.2 (respectively, Sect. 9.2).

Proof. We start with the lower bound. For  and r > 0 we shall consider the
function g(s) := s  dened on the disk of radius r and center , that is
K := { + s : |s| r}. We suppose that the disks Kj := {j + s : s K} are
disjoint and that 1 + r <  < 2 r. An application of Theorem 10.6
yields the existence of some positive real numbers such that

max |Z(s + i ) g(s)| < . (10.10)

sK

This implies

max |Z( + s + i ) g( + s)| <  r = min |g( + s)| (10.11)

|s|=r |s|=r

for suciently small . The function g( + s) = s has a zero inside the disk
|s| r and so, by Rouchs Theorem 8.1, there exists a zero of Z(s) inside any
of the sets { + s + i : |s| r} provided inequality (10.10) holds. The number
of these zeros will give a lower bound for the zero-counting function NZ in
the formulation of the theorem. As in the proof of Theorem 8.3 (respectively,
Theorem 9.1), we get
1
NZ (, T ) meas I(T ),
2r
where :=  r > 1 , and I(T ) is the set of [0, T ] such that (10.10)
is valid. In view of (10.11) we can take r = . By Theorem 10.6,
 
1
d() := lim inf meas [0, T ] : max |Z(s + i ) g(s)| <  > 0
T T sK
206 10 Consequences of Universality

for any positive  for which the sets Kj = {j + s : |s| } lie disjoint. This
yields the lower estimate of the theorem with the constant

1
c1 = d()
2
 
1 1
= lim inf meas [0, T ] : max |Z( +  + s + i ) s| <  (10.12)
2 T T |s|

For the upper bound we suppose that

n
f (1) = 1 and a := aj = 0. (10.13)
j=1

The general case can be treated analogously. We dene the function

1
Z(s) := Z(s).
a
Obviously, the zeros of Z(s) correspond one-to-one to the zeros of Z(s), and
Z(s) tends to 1 as . It follows that Z(s) has no zeros in some half-plane
3 . Littlewoods Lemma 7.2 yields
 2 
1
NZ (, T ) d = log Z(s) ds + O(1), (10.14)
1 2i R

where R is the rectangular contour with vertices 1 , 2 , 1 + iT, 2 + iT with

1 < 1 < 2 < 3 < 2 , and where the error term arises from possible
poles at s = 1 j . The right-hand side of (10.14) is equal to
 T
1
log |Z(1 + it)| dt + O(log T );
2 0

the error term estimate comes from a standard use of Jensens formula (as in
the proof of Theorem 7.1). By Jensens inequality, the appearing integral is
  
T
T 1
log |Z(1 + it)| dt .
2
4 T 0

Since F(s) is regular and of nite order with bounded second moment (10.9),
it follows from Carlsons Theorem 2.4 that, for > 1 ,


1 T
|f (n)|2
lim |F( + it)|2 dt = =: F (2).
T T 0 n=1
n2
 10.6 Shifts of Universal Dirichlet Series 207

This leads to
 
n

n
1 T 1 T
lim |Z(1 + it)| dt
2
|aj | lim
2
|Fj (1 + it)|2 dt
T T 0 |a|2 j=1 T T 0

n

n
= |aj |2 F (2(1 + Re j )).
|a|2 j=1

Thus, we may replace (10.14) by

T n

n
(Re  1 ) log |aj |2 F (2(1 + Re j )) + O(1) ;
Re
>
4 |a|2
j=1
1
0<Im
T

here the sum on the left-hand side is taken over all zeros  of Z(s) satisfying
the condition of summation. Since, for 1 < 1 < ,
1

NZ (, T ) (Re  1 ),
1 Re
>
1
0<Im
T

we obtain

T n
n
NZ (, T ) log |aj |2 F (2(1 + Re j )) + O(1) .
4( 1 ) |a|2 j=1

Thus, taking 1 = 12 (1 + ) the upper estimate of the theorem holds with

the constant

1 n
n
c2 = log 2 |aj |2 F (1 + + 2Re j ) . (10.15)
2( + 1 ) |a| j=1

The theorem is proved. 


We conclude with a few words concerning the zero-distribution of Dirichlet
series. It seems that universal Dirichlet series have in their region of univer-
sality either none or T many zeros with real part greater than for any
xed [ 12 , 1) up to level T . Selberg [323] investigated linear combinations
of linearly
independent elements of the Selberg class and conjectured that
 T log log T many of its zeros up to level T lie to the right of the critical
line, but that almost all zeros lie on the critical line. The latter conjecture
was proved by Bombieri and Hejhal [38] in the case of Epstein zeta-functions
attached to a rational quadratic form under the assumption of a well-spacing
of the zeros. Gritsenko [106] showed that linear combinations of primitive
degree 2-elements of the Selberg class have at least T (log log T ) many zeros
on the critical line up to level T , where is some constant. It is expected that
a positive proportion of these zeros is located on the critical line.
11
Dirichlet Series with Periodic Coecients

In Anschlu daran beweise ich u.a. eine Funktionalgleichung, der jede

Dirichletsche Reihe mit periodischen Koezienten gengt; diese ist
ganz berraschend einfach und scheint mir, wenn man sie auf die in
der Zahlentheorie eine so groe Rolle spielende Charakterreihen (L-
Reihen) anwendet, die Quelle zu sein, von der aus die Funktionalgle-
ichungen recht verstndlich werden. W. Schnee

In this chapter, we consider Dirichlet series associated with periodic arith-

metical functions f , sometimes also called periodic zeta-functions. This class
of Dirichlet series includes Dirichlet L-functions, but in general these func-
tions do not have an Euler product; anyway, we shall denote them by L(s, f ).
Such Dirichlet series are rather simple objects which have the advantage that
many computations can be done explicitly. We prove universality for Dirichlet
series attached to non-multiplicative periodic functions subject to some side
restrictions. This leads to an interesting zero-distribution which is rather dif-
ferent to the one of Dirichlet L-functions. The results of this chapter are due
to Steuding [340, 342, 343].

11.1 Zero-Distribution
Let q be a positive integer and let f : Z C be a q-periodic arithmetical
function, i.e.,
f (n + q) = f (n) for all n Z.
We dene the associated Dirichlet series by


f (n)
L(s, f ) = .
n=1
ns

Since the coecients are bounded, this series converges for > 1. In the
sequel, we assume that f does not vanish identically zero. However, we do
210 11 Dirichlet Series with Periodic Coecients

not require that q is the minimal period; consequently, certain quantities

appearing in asymptotic formulae below remain invariant if q is replaced by
an integer multiple.
By the periodicity of f we nd
 
1

q
a
L(s, f ) = s f (a) s, . (11.1)
q a=1 q

The analytic continuation of the Hurwitz zeta-function (1.42) leads immedi-

ately to an analytic continuation for L(s, f ). Thus, L(s, f ) is analytic through-
out the whole complex plane except for at mostone simple pole at s = 1;
q
actually, L(s, f ) is regular at s = 1 if and only if a=1 f (a) = 0 (respectively,
+
f (q) = 0 in the notation (11.2)). By the identity
 
(s) is
exp(2in)
(1 s, ) = exp
(2)s 2 n=1 ns
  
is
exp(2in)
+ exp ,
2 n=1 ns

valid for > 1 (this is a special case of the functional equation (1.44)), we
get from (11.1)
 
q
s (s) is
f (n)
L(1 s, f ) = exp
2 q 2 n=1 ns
  
is
f + (n)
+ exp
2 n=1 ns

for > 1, where

 
1

q
an
f (n) := f (a) exp 2i . (11.2)
q a=1 q

The functions f may be interpreted as the discrete Fourier transforms of the

functions f : Z C, n  f (n). Obviously, f+ = f , and f is q-periodic
as well. By analytic continuation we obtain the functional equation
q
s (s)  
is

L(1 s, f ) = exp L(s, f )
2 q 2
  
is
+ exp L(s, f ) , (11.3)
2
valid for all s, which was rst proved by Schnee [316].
Denote the zeros of L(s, f ) by  = + i. As in the case of Dirichlet
L-functions we have to distinguish between trivial and non-trivial zeros of
 11.1 Zero-Distribution 211

L(s, f ). Before we give a rigorous denition of trivial and non-trivial zeros we

establish a zero-free region on the right. Let

Cf = max{|f (a)| : 1 a q}

and
mf = min{a : 1 a q, f (a) = 0}.
Then, for > 1, we obtain

f (n)

f (n)
L(s, f ) = s
= (s, f ) + ,
n=m
n n=m +1
ns
f f

where
f (mf )
(s, f ) := .
msf
This leads to the estimate



1 dx Cf m1
f
|L(s, f ) (s, f )| Cf Cf .
n=mf +1
n mf x 1

Hence, as ,  
1
L(s, f ) = (s, f ) + O .
mf
This implies
L(s, f ) = 0 for > 1 + A(f ),
where
Cf
A(f ) := .
|(1, f )|
By the functional equation (11.3) and the non-vanishing of the Gamma-
function, L(1 s, f ) vanishes if and only if
   
is is
exp L(s, f ) = exp L(s, f + ).
2 2

Dene B(f ) = max{A(f )}. It follows that for < B(f ) the function
L(s, f ) can only have zeros close to the negative real axis if mf + = mf , and
close to the straight line, given by
t
=1+ mf ,
log mf +

if mf + = mf . We call zeros  = +i of L(s, f ) with < B(f ) trivial. One

can locate those zeros with the techniques of Spira [335] and Garunktis [88].
It is easily seen that the number of trivial zeros  with || R is cR, where
212 11 Dirichlet Series with Periodic Coecients

c is some positive, computable constant depending on f . We call other zeros

of L(s, f ) non-trivial. The non-trivial zeros lie in the vertical strip

B(f ) 1 + A(f ). (11.4)

If f is a Dirichlet character mod q, this denition of trivial and non-trivial

zeros does not correspond with the traditional one in the theory of Dirichlet L-
functions, where, for example, L(s, ) has a trivial zero at s = 1 if (1) =
1. But note that there are only nitely many trivial zeros  of L(s, f ) in any
vertical strip of bounded width (independent on this or any other reasonable
denition of trivial).
We start, similarly as in Chap. 7.1, with an asymptotic formula for a sum
taken over the zeros.

Theorem 11.1. Let f be a q-periodic arithmetic function and let b be a con-

stant satisfying b 3 + max{A(f ), B(f )}. Then,

 
1 T qT
( b) = b log
>b
2 2em f mf mf +
||T
  
 1 , f   1 , f + 
T  2 
+ log   1 2
2
 + O(log T ).
2  ,f 
2

Proof. Dene
L(s, f )
Z(s, f ) = .
(s, f )
Of course, the zeros of Z(s, f ) are exactly the zeros of L(s, f ). Let a = 2 +
max{A(f ), B(f )}. Then, by the condition on b, all non-trivial zeros of L(s, f )
and L(s, f ) have real parts in the interval (b, a).
First, suppose that L(s, f ) is regular at s = 1. Then Littlewoods
Lemma 7.2, applied to Z(s, f ) and the rectangle R with vertices a iT, b iT ,
yields

 T  T
2 ( b) = log |Z(b + it, f )| dt log |Z(a + it, f )| dt +
>b T T
||T
 a  a
arg Z( iT, f ) d + arg Z( + iT, f ) d
b b


4
= Ij , (11.5)
j=1

say. To dene log Z(s, f ) we choose the principal branch of the logarithm
on the real axis, as ; for other points s the value of the logarithm
is obtained by analytic continuation in a standard manner. If L(s, f ) is not
 11.1 Zero-Distribution 213

regular at s = 1, we have to replace Z(s, f ) by (s 1)Z(s, f ) at the expense

of an error O(1). To see this one applies Littlewoods lemma to (s 1)Z(s, f )
and the function s 1 separately and subtracts the resulting formulas. Since

i log(s 1) ds = 2(b + 1)  1,
R

we obtain (11.5) with an additional error term O(1).

To evaluate I1 note that
log |Z(b + it, f )| = log |(b, f )| + log |L(b + it, f )|.
By the functional equation (11.3) we get


 q 1bit (1 b it) 
log |L(b + it, f )| = log  
 (11.6)
2 q
  
 i(1 b) + t

+ log exp L(1 b it, f )
2
  
i(1 b) + t 
+ exp L(1 b it, f + ) .
2
For |t| > 1, Stirlings formula (2.17) implies

  
 q 1bit (1 b it) 

log   = 1 b log q|t| |t| + O(|t|1 ).
2 q  2 2 2
For t > 1, the second term on the right-hand side of (11.6) equals
  
 i(1 b) + t

log exp L(1 b it, f )
2
 
L(1 b it, f + ) 
1 + exp(i(1 b) t)
L(1 b it, f ) 
 
t  L(1 b it, f ) 
= 
+ log |(1 b, f )| + log   + O(exp(|t|)),
2 (1 b it, f ) 
and, for t < 1,
 
t  L(1 b it, f + ) 
+ log |(1 b, f + )| + log   + O(exp(|t|)).
2 (1 b it, f + ) 
Collecting, we obtain
 
1 qT 
I1 = 2 b T log + T log |(1 b, f )(1 b, f + )|
2 2e
 1  
 L(1 b it, f + ) 
2 log |(b, f )|) + log   dt
T (1 b it, f + ) 
 T  
 L(1 b it, f ) 
+ log   dt + O(1).
1 (1 b it, f ) 
214 11 Dirichlet Series with Periodic Coecients

The appearing integrals look similar to I2 . We estimate them all as follows.

For > 1,
  
f (n) mf
s


mf
Z(s, f ) = 1 + =1+O .
n=m +1
f (mf ) n mf + 1
f

This yields I2  1 by the same argument as in the proof of the estimate (7.8).
Hence, we get
 
1 T
I1 + I2 = 2 b T log
2 2emf mf mf +
  
 1 , f   1 , f + 
 2 
+ T log   2
2
 + O(1).
 1, f 
2

The horizontal integrals I3 , I4 can be estimated as in the proof of Theorem 7.1.

Thus the theorem is proved. 

Denote by N (T, f ) the number of non-trivial zeros  of L(s, f ) with || T
(according multiplicities). The formula of Theorem 11.1 implies
Corollary 11.2. Let f be a q-periodic arithmetic function. Then
T qT
N (T, f ) = log + O(log T ),
2emf mf mf +
and
  

   1 , f   1 , f + 
1 T  2 
= log   1 2
2
 + O(log T ).

non-trivial
2 2  ,f 
2
||T

The latter sum is O(N (T, f )). So most of the non-trivial zeros are either
approximately symmetrically distributed or lie close to the critical line. Indeed,
setting
N+ (, T, f ) = { : > , || T }
and
N (, T, f ) = { : < , || T },
one can prove by the techniques which we used for the proof of Theorem 7.6:
Theorem 11.3. Let f be a q-periodic arithmetic function. Then, uniformly
in > 0,
   
1 1 1
N+ + , T, f + N , T, f  T log log T
2 2
log log T
 N (T, f ).
log T
Moreover, for every xed > 12 ,
N+ (, T, f )  T.
 11.1 Zero-Distribution 215

Thus, the non-trivial zeros of any L(s, f ) are clustered around the critical line.
It seems that the clustering is a common pattern for Dirichlet series with a
Riemann-type functional equation.

 
Now dene
2 1
(f ) = lim .
T T 2

non-trivial
||T

It follows from Corollary 11.2 that L(s, f ) has innitely many zeros o the
critical line if   
 1 , f   1 , f + 
 2 
(f ) = log   1 2
2
 = 0.
 ,f 
2
Hence, a non-zero value of (f ) indicates an asymmetrical distribution of the
non-trivial zeros of L(s, f ) (with respect to the critical line), and in this case
the number of non-trivial zeros of L(s, f ) with = 12 and || T is
|(f )| T
.
1 + 2 max{A(f ), B(f )}
For example, the Ramanujan sum

 
an
cq (n) := exp 2i
a mod q
q
(a,q)=1

is q-periodic (in the n-aspect). Since cq (1) = (q) and cq (1) = q, where
is the Mbius -function, we have (cq ) = log q by Corollary 11.2 if q is
squarefree. It turns out that there are more than
log q T
1 + 2(q)
many zeros o the critical line up to level T . However, in this special example
we have Ramanujans identity [301]


L(s, cq ) = (s) (q/d)d1s , (11.7)
d|q

which gives precise information on the asymmetrically distributed zeros o

the critical line.
We conclude this section by a look on the zero distributions of L(s, f )
and L(s, f ) with respect to each other. Note that with f also the Fourier
transforms f are q-periodic. By the inversion formula for Fourier transforms,
(f + ) = f and (f ) = f ,
we deduce from (11.3) the functional equation
q
s (s)  
is
 
is
 
L(1 s, f ) = exp L(s, f ) + exp L(s, f ) .
2 q 2 2
Now we deduce from Corollary 11.2.
216 11 Dirichlet Series with Periodic Coecients

25

20

15

10

0
1 0.5 0 0.5 1 1.5 2

Fig. 11.1. The reciprocal of the absolute value of L(s, 11,5 ) for [1, 2], t [0, 25]
as a contour plot. In each white island there is a zero of (s, 1/5)

Corollary 11.4. For every q-periodic function f ,

(f+ f f + f ) = (f+ ) + (f ) + (f + ) + (f ) = 0.
For example, dene for positive integers a, q with 1 a q

1 if n a mod q,
1a,q (n) =
0 otherwise.
Obviously, 1a,q is q-periodic. One easily calculate
((1a,q )+ ) + ((1a,q ) ) + ((1a,q )+ ) + ((1a,q ) )
a qa q q
= log + log + log % + log %
q q (q a)a a(q a)
= 0.
Note that
 
1 a
L(s, 1a,q ) = s s, . (11.8)
q q
Hence, the zeros of L(s, 1a,q ) are exactly the zeros of (s, aq ) (see Fig. 11.1 for
an example). This special case should be compared with analogous results of
Garunktis and Steuding [91] for the Lerch zeta-function.

11.2 A Link to the Selberg Class

In this section, we are interested in the class S  consisting of all Dirichlet
series
 11.2 A Link to the Selberg Class 217

a(n)
L(s) := ,
n=1
ns

not identically zero, satisfying the axioms:

(ii) Analytic continuation. There exists a non-negative integer k such that
(s 1)k L(s) is an entire function of nite order.
(iii) Functional equation. L(s) satises a functional equation of type
L (s) = L (1 s),
where

f
L (s) := L(s)Qs (j s + j )
j=1

with positive real numbers Q, j , and complex numbers j , with

Re j 0 and || = 1.
These axioms are the analytic hypotheses of the Selberg class S. Therefore,
S S  . We refer to S  as the extended Selberg class. We dene the degree of
elements of S  in the same way as for S (see (6.2)) and denote by S d the set
of L S  of degree d.
Kaczorowski and Perelli [160] proved
Theorem 11.5. If L S0 , then L(s) is a Dirichlet polynomial and

a(n) n
L(s) = with a(n) = a(Q2 /n).
2
ns Q
n|Q

This follows more or less directly from the functional equation for L(s) which
might be viewed as an identity between absolutely convergent Dirichlet series
(see the proof of Theorem 6.1).
Our aim is to give a dierent classication for the degree zero-elements in
the Selberg class, namely as entire quotients of Dirichlet series with periodic
coecients and the Riemann zeta-function (s). One example we have already
met in (11.7). The general result is
Theorem 11.6. A function L lies in S0 if and only if there exists a periodic
arithmetical function f : N C which satises
f + = f with C, || = 1, (11.9)
and the quotient L(s) = L(s, f )/(s) denes an entire function. In this case,
L(s, f ) S1 .
This result should be compared with a related theorem of Kaczorowski and
Perelli [161] on a characterization of S1 -functions as S0 -linear combinations
of Dirichlet L-functions and with the construction of Dirichlet series with
periodic coecients having zeros o the critical line by Balanzario [14].
Before we give the proof of the theorem we note
218 11 Dirichlet Series with Periodic Coecients

Lemma 11.7. Let f be a q-periodic arithmetic function. If L(s, f )/(s) is an

entire function, then
2 q
s s

L(1 s, f ) = (s) cos L(s, f + ). (11.10)

q 2 2

Proof. To compensate the trivial zeros of (s) in s = 2m with m N, the

function L(s, f ) has to vanish there too. From the functional Equation (11.3)
we deduce
L(2m + 1, f + ) = L(2m + 1, f )
and


f + (n) f (n)
= 0 for all m N.
n=1
n2m+1
Since any convergent Dirichlet series, which does not vanish identically, has a
zero-free half-plane (see Sect. 2.1), it follows that f + = f , and, since f = f+ ,
we can replace (11.3) by (11.10). The lemma is proved. 


Now we are in the position to give the

Proof of Theorem 11.6. First, assume that L(s) = L(s, f )/(s) is an entire
function. By Lemma 11.7 we see that L(s, f ) satises the functional equation
(11.10). If f satises additionally (11.9), then we can rewrite (11.10) as
2 q
s s

L(1 s, f ) = (s) cos L(s, f ), (11.11)

q 2 2

or, using well-known identities for the Gamma-function,

q
(1s)/2  1 s  q
s/2 s

L(1 s, f ) = L(s, f ).
2 2

We note that L(s, f ) S1 . Using (11.11) also with f constant equal to 1 (the
functional equation for (s)), we obtain

q s/2 L(s) = q (1s)/2 L(1 s).

Therefore, L(s), being the entire quotient of two Dirichlet series in S1 , lies in
S0 . This proves the suciency.
Now assume that L S0 . By Theorem 11.5 there exist a positive integer
2
Q and complex numbers a(n) such that the coecients a(n) in the Dirichlet
series
 expansion of L(s) vanish if n does not divide Q2 . Now dene f (n) =
d|n a(d), then



f (n + Q2 ) = a(d) = a(d) = f (n),
d|(n+Q2 ) d|n
(d|Q2 ) (d|Q2 )
 11.3 Strong Universality 219

and thus f is a Q2 -periodic function. Further, L(s, f ) = (s)L(s), and it

remains to show (11.9). Using the functional equation for (s), we see that
L(s, f ) satises the functional equation (11.11), after replacing q by Q2 . More-
over, since L(s, f )/(s) is entire, Lemma 11.7 implies that further (11.10)
holds with Q2 instead of q. Comparing both functional equations, we nd
L(s, f + ) = L(s, f ). In view of the uniqueness of Dirichlet series expansions
we obtain (11.9). Theorem 11.6 is proved. 


It is interesting to investigate the dierence between the Selberg class and the
extended Selberg class, or which functions lie in S  \ S? The descriptions of
these classes are complete for degree less than or equal to one. For degree two
Kaczorowski et al. [158] gave examples with Dirichlet series associated with
cusp forms of certain Hecke groups. Note that for a positive real number ,
the Hecke group G() is dened as the subgroup of PSL2 (R) given by
0   1
1 0 1
G() := , .
0 1 1 0

Kaczorowski et al. showed that the associated Dirichlet series are elements
of S  or a related class of Dirichlet series where the axiom on the functional
equation is appropriately adjusted. Moreover, they showed that the Dirichlet
series associated to newforms for G() have an Euler productrepresentation
if
and only if G() can be arithmetically dened, i.e., if {1, 2, 3, 2}. Their
result is based on Heckes famous theorem on the correspondence of Dirichlet
series with functional equation and modular forms, and a result of Wolfart
on the arithmetic nature of the Fourier coecients of the associated modular
forms. For 2, Molteni and Steuding [257] proved that all these Dirichlet
series are almost primitive
(i.e., primitive up to factors of degree zero) and
primitive if  { 2, 3, 2}; if the latter condition is not fullled, there are
examples of non-primitive functions.

11.3 Strong Universality

As a more or less immediate consequence of the joint universality of Dirichlet

L-functions we obtain a strong universality result for L(s, f ):

Theorem 11.8. Suppose that q > 2 and f is a q-periodic arithmetical func-

tion, not a multiple of a character mod q, satisfying f (n) = 0 for n with
(n, q) > 1. Let K be a compact subset of the strip 12 < < 1 with connected
complement and let g(s) be a continuous function on K which is analytic in
the interior. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max |L(s + i, f ) g(s)| <  > 0.
T T sK
220 11 Dirichlet Series with Periodic Coecients

For example, the DavenportHeilbronn zeta-function (7.1) satises the condi-

tions and is therefore strongly universal.
The assumptions on f from Theorem 11.8 imply that f is not multiplica-
tive; this follows from a characterization of q-periodic multiplicative functions
due to Leitmann and Wolke [213]. In particular, the associated Dirichlet series
L(s, f ) has no Euler product representation.

Proof. For the (q) pairwise non-equivalent characters j mod q dene the
matrix
= (j (n)) 1nq,(n,q)=1 .
1j(q)

By the properties of characters the matrix is invertible (since pairwise

non-equivalent characters are linearly independent over C, and det is a
van der Monde-determinant). Hence there exist uniquely determined complex
numbers c1 , . . . , c(q) such that

(q)
f (n) = cj j (n) for 1 n q, (n, q) = 1,
j=1

where at least two distinct coecients cj are non-vanishing (since f is not a

multiple of a character mod q and (q) 2). In view of f (n) = 0 for integers
n which are not coprime with q, we obtain the representation

(q)
L(s, f ) = cj L(s, j ). (11.12)
j=1

Put g = 1 + maxsK |g(s)|. Then g(s) + g = 0 for any s K. Without loss

 0. Now dene
of generality, we may assume that c1 , c2 =

g(s) + g g
g1 (s) = , g2 = ,
c1 c2

and
gj (s) = for 3 j (q),

where is a small positive parameter, chosen later. Using the joint universality
Theorem 1.10 for Dirichlet L-functions, we obtain, for any > 0,
 
1
lim inf meas [0, T ] : max max |L(s + i, j ) gj (s)| < > 0.
T T 1j(q) sK

Note that
g(s) = c1 g1 (s) + c2 g2 (s).
 11.3 Strong Universality 221

Consequently, (11.12) implies

max |L(s + i, f ) g(s)|
sK

(q)


(q)
|cj | max |L(s + i, j ) gj (s)| + |cj |
sK
j=1 j=3

(q)( max |cj | + ).

1j(q)

With suciently small and the assertion of the theorem follows. 


Theorem 11.8 is due to Bagchi [9]; however, his argument diers slightly from
ours. Generalizations were given by Sander and Steuding [315], which include
universality for arbitrary Dirichlet series with periodic coecients and ele-
ments of degree 1 of the extended Selberg class in particular. The latter case
was independently proved by Kaczorowski and Kulas [156] in a stronger form.
Next we shall give explicit upper bounds for the density of universality.
With regard to Theorem 9.1 we get, using the same notation as in Sect. 9.2,
Theorem 11.9. Suppose that q > 2 and f is a q-periodic arithmetical
function, not a multiple of a character mod q, satisfying f (1) = 1 and
f (n) = 0 for n with (n, q) > 1. Further, assume that g Ar . Then, for
any  0, 2r1
4 + Re ||
1
 (where is the zero of g(s) in |s| r) and any
1 12 , 34 + Re 2r ,

2r3 
d(, g, L(s, f )) 3 
(r2 ||2 ) 4 + Re 2r 1
  
1

q
a
log |f (a)|2 21 , .
q 21 a=1 q

The proof is quite similar to some parts of the proof for Theorem 10.7.
Proof. Denote by N (, T, f ) the number of zeros  = + i of L(s, f ) with
> and 0 < T . Then Littlewoods Lemma 7.2 yields
 2 
1
N (, T, f ) d = log L(s, f ) ds + O(1), (11.13)
1 2i R
where R is a rectangular contour with vertices 1 , 2 , 1 + iT, 2 + iT , where
1
2 < 1 < 1 < 2 . Here the error term arises from the possible pole of L(s, f )
at s = 1 and log L(s, f ) is dened in a standard manner (as in the proof of
Theorem 11.1). Since
L(s, f ) = 1 + O(1)
as , we choose 2 so large that L(s, f ) has no zeros in the half-plane
2 . A standard application of Jensens formula (as in the proof of
Theorem 7.1) shows that the right-hand side of (11.13) is equal to
222 11 Dirichlet Series with Periodic Coecients

 T
1
log |L(1 + it, f )| dt + O(log T )
2 0
  
T 1 T
log |L(1 + it, f )| dt + O(log T ).
2
4 T 0

Therefore, using the mean-square formula of Kaenas and Laurinikas [152],

  
1

T q
1 a
|L( + it, f )| dt
2
|f (a)| 2,
2
,
T 0 q 2 a=1
q

valid for > 12 , and the identity (7.4), we may replace (11.13) by
  

T 1

q
a
( 1 ) log |f (a)| 21 ,
2
(11.14)
>
4 q 21 a=1 q
1
0<T

+O(log T ).

Now let 3 > 1 . Since

1

N (3 , T, f ) ( 1 ),
3 1 >
1
0<T

we obtain the estimate

 
3
N + Re 2r, T, f
4
  
1

q
T a
  log |f (a)| 21 ,
2
4 34 + Re 2r 1 q 21 a=1
q
+O(log T ).

In view of (9.6) this implies the estimate in the theorem. 



We give an example. For an odd prime q, consider the function L(s, 11,q ),
given by (11.8). By Theorem 11.8, L(s, 11,q ) has the universality property, and,
using Theorem 11.9, we obtain after some lengthy calculations, for suciently
small  > 0,

2 98 + ||
2 r3 
d(, g, L(s, 11,q )) 9/8+||/2 .
q (r2 ||2 )
Since the right-hand side tends to zero as q , the problem to approximate
an analytic isomorphism g(s) by a function L(s, 11,q ) becomes, in some sense,
more and more dicult with increasing q.
 11.4 Hurwitz Zeta-Functions 223

Following the reasoning of Chap. 10, universality for Dirichlet series with
periodic coecients leads to results on the denseness of the set of values
taken on vertical lines, and functional independence. Here we only consider
the particular case of the distribution of zeros. Strong universality leads to
the existence of many zeros o the critical line. Via Rouchs Theorem 8.1
the universality property implies the existence of T zeros in each strip 12 <
1 < < 2 < 1 up to level T . This reasoning is very similar to the proof
of Theorem 10.1. In particular, it follows that the real parts of the zeros lie
dense in [ 12 , 1]. In conjunction with the estimates of Theorem 11.3 we get the
exact order of size for the zero-counting function N (, T, f ), introduced in the
proof of Theorem 11.9.
Corollary 11.10. Suppose that L(s, f ) satises the conditions of Theo-
rem 11.8. Then, for 12 < < 1,
N (, T, f ) T.
There is another interesting result in this context. Kaczorowski and
Kulas [156] proved that any element L of degree 1 of the extended Selberg
class has o(T ) many zeros in the same range as in Theorem 11.9 if and only
if L(S) has a factorization
L(s) = P (s)L(s, ),

where P (s) is a Dirichlet polynomial from S0 and is a primitive Dirichlet

character. Their reasoning is based on the structure theorem for S1 from
Kaczorowski and Perelli [161] and a variation of the joint universality the-
orem for Dirichlet L-functions 1.10. The result of Kaczorowski and Kulas
gives a partial answer to the question whether any Dirichlet series satisfying
Riemanns hypothesis or, at least, a density estimate has necessarily an Euler
product.

11.4 Hurwitz Zeta-Functions

We consider Hurwitz zeta-functions with rational parameters (for their den-
ition see (1.42)). In view of (11.8) and Theorem 11.8 we see that the function
 
1 a
s, = L(s, 1a,q ). (11.15)
qs q
is strongly universal for q > 2. For { 12 , 1}, we have
 
1
s, = 2s L(s, ) and (s, 1) = (s), (11.16)
2
where is the unique character mod 2. Thus, it follows from Voronins results,
respectively, the renements due to Bagchi and Gonek, Theorem 1.9, 1.10, and
their discrete variant Theorem 5.17) that the functions
224 11 Dirichlet Series with Periodic Coecients
 
1 1
s, and (s, 1)
2s 2
are universal but not strongly universal (by the results from Chap. 8). The
appearance of the factor 2s in the case = 12 is a bit unpleasant here but it
can be removed as we shall explain now.
The rst argument is by an application of a universality theorem for
Dirichlet L-functions due to Gonek [104]. Incorporating some ideas from
Good [105], Gonek proved

Theorem 11.11. Let q be a positive integer and let K be a simply connected

compact set in 12 < < 1. Suppose that for each prime p|q we have 0 p < 1
and that for each character mod q, the function g (s) is continuous on K
and analytic in the interior. Then, for every  > 0, there is a real number
such that
min | log p 2(z + p )| <  for all p|q, (11.17)
zZ

and
max max |L(s + i, ) g (s)| < .
mod q sK

Goneks theorem does not show that the set of these has positive lower
density as in Voronins universality theorem for the zeta-function and also
the set K on which these approximations are realized is more restricted than
usually in universality theorems (however, it seems to be possible to extend
Goneks approach in order to obtain these stronger concepts of universality).
But this result has the option that the diophantine condition (11.17) can be
used to remove the factor 2s in (11.16) as well as the factor q s in (11.15).
Bagchi [9] gave a dierent argument based on statistical independence. A third
possibility is to use the joint discrete universality Theorem 5.17 for Dirichlet
L-functions due to Bagchi [9].
Sander and Steuding [315] used the latter argument to obtain a joint uni-
versality theorem for Hurwitz zeta-functions with rational parameters. We
sketch the simple argument which mimics the proof of Theorem 11.8. Dene
vectors
1  a 
t
t
Z := s s, and L := L(s, j ) .
q q 1aq;gcd(a,q)=1 1j(q)

Then we deduce from

 

1 a 1
s
s, = (a)L(s, )
q q (q)
mod q

(this is the inverse of (11.1) for f = ) that

(a)

j
Z = ML with M := ,
(q) 1aq;gcd(a,q)=1
1j(q)
 11.4 Hurwitz Zeta-Functions 225

where the j denote the (q) pairwise non-equivalent characters modulo q.

Clearly, M is a square matrix. Imagine we want to approximate a family of
nice target functions fa : K C with 1 a q and a coprime with q. The
idea is to nd appropriate functions gj : K C, 1 j (q) such that gj (s)
is approximated by L(s, j ) and
1
t
t
f = Mg , where f := fa (s) , g := gj (s) .
qs 1aq;gcd(a,q)=1 1j(q)

The functions gj (s) are uniquely determined since M is invertible. We have

g = M1 f. To approximate the functions gj (s) by Dirichlet L-functions, we
need to assume that the gj (s) are non-vanishing continuous functions on K
which are analytic in the interior. Then, given  > 0, for any real = 0,
by Theorem 5.17 we can nd a set of positive integers n with positive lower
density such that

max max |L(s + in, j ) gj (s)| < .

1j(q) sK

Hence, we may deduce from the identity Z f = M(L g) that

   
 1 a 

max max  in s + in, fa (s) < .
1aq sK q q
gcd(a,q)=1

Now, choosing appropriately, we may get rid of the factor q in and arrive
at the following weak joint universality theorem:

Theorem 11.12. Let K be a compact subset of 12 < < 1 with connected

complement, let q be a positive integer, and for each 1 a q with gcd(a, q) =
1 let fa (s) be a continuous functions on K which is analytic in the interior. If
all components of M1 f are non-vanishing on K (in the notation from above),
then, for any  > 0,
    
1  a 
lim inf  n N : max max  s + in, 
fa (s) <  > 0,
N N 1aq sK q
gcd(a,q)=1

where = 2k
log q with any k N if q 2, and 0 = R otherwise.

Notice that q is allowed to be equal to 1 or 2; however, in these cases we

do not obtain strong universality since then (q) = 1 and a = 1 and so the
condition on the non-vanishing of the components of M1 f is nothing but the
restriction for f1 (s) to be free of zeros in K. This corresponds to our remarks
concerning (11.16).
How strong is Theorem 9.5? By (11.1) with f = ,
 
1

q
a
L(s, ) = s (a) s, ,
q a=1 q
226 11 Dirichlet Series with Periodic Coecients

it follows that joint universality for Dirichlet L-functions is equivalent to joint

universality for Hurwitz zeta-functions to rational parameters. If the non-
vanishing condition for the components of M1 f would be dropped, it would
follow that Dirichlet L-functions could approximate functions with zeros which
is impossible.
Sander and Steuding [315] also investigated to which extent the condition
on the parameters of the Hurwitz zeta-functions to be reduced fractions with
a common denominator can be skipped. However, this is not always possible.
Assume that we are given a set A of integers 1 a q. By relations as for
example  
1
a
(s) = s s,
q q
1aq

it is easily seen that there exist dependence relations among the Hurwitz
zeta-functions (s, aq ) with a A if A > (q). However, if A (q),
then there exists restricted joint universality whenever the target functions are
linearly independent; here the conditions are rather technical. Nevertheless,
using the notion of weak joint universality from Sect. 10.3, it follows that
many families of Hurwitz zeta-functions with rational parameters are jointly
functional independent.
There is an interesting related, recent result due to Nakamura [276] who
proved that generalizations of Lerch zeta-functions are not jointly universal.
This is a rather surprising result. For , , (0, 1] and Re [0, 12 ) the
generalized Lerch zeta-function L(, , , , s) is for > 1 dened by


exp(2in)
L(, , , , s) =
n=0
(n + )s (n + )

and by analytic continuation for > 0 except for at most a simple pole at
s = 1. We note that for = or = 0 this function is indeed a Lerch
zeta-function:

L(, , , , s) = L(, , s) and L(, , , 0, s) = L(, , s).

In fact, Nakamura [276] showed joint universality for any nite family
L(j , j , j , , s) provided the j are algebraically independent transcenden-
tal numbers; however, he also proved that any family L(j , , j , , s) with
transcendental is not jointly universal in the sense of Voronin if at least two
of the j are equal. The latter statement follows from the fact that the set

(L(, , j , , s + i ), L(, , k , , s + i ))

is not dense for any admissible pair j , k in the according function space as
varies in R. However, this does not exclude the possibility of joint universality
for a large class of functions.
 11.4 Hurwitz Zeta-Functions 227

We conclude with another open problem in universality theory. In [243],

Matsumoto asked whether multiple zeta-functions are universal; for example,
the Barnes multiple zeta-function dened by




1
 (s; , 1 , . . . ,  ) = ... ,
m =0 m
( + m1 1 + . . . + m  )s

or the double HurwitzLerch zeta-function given by

1 exp(2im2 )
2 (s1 , s2 ; , , ) = .
m1 =0
( + m1 ) m =0 ( + m1 + m2 )s2
s 1
2
12
Joint Universality

What we know is not much. What we do not know is immense.

P.S. Laplace

In this chapter, we shall prove a conditional joint universality theorem

for functions in S. Joint universality means that we are concerned with
simultaneous uniform approximation, a topic invented by Voronin [362, 364].
Of course, such a result cannot hold for an arbitrary family of L-functions:
e.g., (s) and (s)2 cannot be jointly universal. The L-functions need to
be suciently independent to possess this joint universality property. We
formulate sucient conditions for a family of L-functions in order to be
jointly universal and give examples when these conditions are fullled; for
instance, Dirichlet L-functions to pairwise non-equivalent characters (this is
an old result of Voronin) or twists of L-functions in the Selberg class subject
to some condition on uniform distribution.

12.1 A Joint Limit Theorem

Our rst aim is to prove a joint limit theorem for L-functions from S; this
generalizes the limit Theorem 4.3 to a multi-dimensional limit theorem for a
family of elements Lk S, 1 k , which we consider as given by the
vector
L(s) := (L1 (s), . . . , L (s)).
As far as possible we shall use the same notation as in the previous chapters.
As for the proof of the one-dimensional limit theorem we will not make use
of axiom (v). Recall that m (L) is the abscissa of the mean-square half-plane
for L S. We denote by D the intersection of all mean-square half-planes
> m (Lk ) for Lk , 1 k , and the critical strip:
 
D = s C : max m (Lk ) < < 1 . (12.1)
1k
230 12 Joint Universality

In principal, we could also deal with dierent strips for the dierent Lk (s),
however, to simplify the presentation we may restrict on one for all. By The-
orem 2.4 (respectively, (4.2)), the set D is not empty. We write H(D) for
the -dimensional cartesian product of the spaces of analytic functions H(D).
L
Further, we dene a probability measure PT by setting

L 1
PT (A) = meas { [0, T ] : L(s + i ) A}
T
for A B(H(D) ). For s D and , we put

L(s, ) = (L1 (s, ), . . . , L (s, )),

where Lk (s, ) is given by (4.6). By Lemma 4.1, any component Lk (s, ) of

the vector L(s, ) is an H(D)-valued random element. Hence, L(s, ) is an
H(D) -valued random element on the probability space (, B(), m) with
as in Sect. 4.1. Let PL denote the distribution of L(s, ). We shall prove
L
Theorem 12.1. Let L1 , . . . , L S. The probability measure PT converges
weakly to PL , as T .

The proof relies in the main part on the one-dimensional limit Theorem 4.3
which applies individually to any component Lk (s) of the vector L(s).
We start with
L
Lemma 12.2. The family of probability measures {PT : T > 0} is relatively
compact.

Proof. To any Lk (s) we attach a probability measure PL

T by putting
k

1
PL
T (A) =
k
meas { [0, T ] : Lk (s + i ) A}
T
for A B(H(D)). By Theorem 4.3, restricted to D, the probability measure
PLT converges weakly to the distribution of the corresponding random element
k

Lk (s, ), i.e, as T ,
PLT P ,
k Lk

say. Hence, the family of probability measures {PL T : T > 0} is relatively

k

compact for any 1 k . Since H(D) is a complete separable space, we

obtain by Prokhorovs Theorem 3.5 that this family is also tight, i.e., for
every T > 0 and every  > 0, there exists a compact set Kk H(D) such that

PL
T (H(D) \ Kk ) < .
k
(12.2)

2 F, Q)
Now let T be a random variable dened on some probability space (,
with the probability measure
 12.1 A Joint Limit Theorem 231
 T
1
Q(T A) = 1(t; A) dt,
T 0

for A B(R), where 1(t; A) is the indicator function of the set A. Consider
the H(D) -valued random element

LT (s) = (L1 (s + iT ), . . . , L (s + iT )).

Then, by (12.2),

Q(Lk (s + iT ) H(D) \ Kk ) <

for 1 k . Hence, for K := K1 K ,
L
PT (H(D) \ K) = Q(LT (s) H(D) \ K)
  
=Q (Lk (s + iT ) H(D) \ Kk )
k=1



Q(Lk (s + iT ) H(D) \ Kk ) <
k=1

L
for all T > 0. Consequently, the family {PT } is tight and so, by Prokhorovs
Theorem 3.4, it is relatively compact. The lemma is proved. 

Let s1 , . . . , sr be arbitrary points in D. We put
1
0 = min Re sm and 1 = 0 .
1mr 2
Clearly, 0 > 12 and 1 < 0. Now let D1 = {s C : > 1 }. For given
complex numbers ukm for 1 k  and 1 m r, dene a function
h : H(D) H(D1 ) for s D1 by



r
h(f1 (s), . . . , f (s)) = ukm fk (sm + s). (12.3)
k=1 m=1

Lemma 12.3. Let

W (s) = h(L(s)).
Then W (s + iT ) converges to h(L(s, )) in distribution, where T is the
random variable, dened in the proof of Lemma 12.2.
Proof. We denote the nth Dirichlet coecient of Lk S by aLk (n). We have



r



r

aLk (n)
W (s) = ukm Lk (sm + s) = ukm
ns+sm
k=1 m=1 k=1 m=1 n=1


r
an ukm aLk (n)
= with an = .
n=1
ns m=1
nsm
k=1
232 12 Joint Universality

Since an  1 by (i), the Dirichlet series an
n ns converges for > 1. Moreover,


|an |2  x.
nx

In view of this estimate, we may apply Theorem 4.12 (with any 0 > 12 ) and
obtain the weak convergence of the probability measure given by
1
meas { [0, T ] : W (s + i ) A}
T
for A B(H(D1 )) to the distribution of the associated random element

an (n)
n=1
ns

for and s D1 , as T . Since


r

an (n) aLj (n)(n)
s
= u km
n=1
n m=1 n=1
ns+sm
k=1




r
= ukm Lk (s, ) = h(L(s, )),
k=1 m=1

the assertion of the lemma follows. 



Proof of Theorem 12.1. By Lemma 12.2, there exists a sequence T1 such

L
that PT1 converges weakly to some probability measure P. Suppose that P is
the distribution of an H(D) -valued random element

2 := (L21 (s), . . . , L2 (s)).

L(s)

Then, obviously,
D 2
L(s + iT1 ) L(s). (12.4)
T1

By the continuity of the function h, dened by (12.3), we deduce

2D
h(LT1 (s)) h(L(s)). (12.5)
T1

By Lemma 12.3,
D
h(LT1 (s)) h(L(s, )).
T1

This and (12.5) yield

2 D
h(L(s)) = h(L(s, )). (12.6)
 12.1 A Joint Limit Theorem 233

For f H(D1 ), dene h1 : H(D1 ) C by h1 (f ) = f (0). Then it follows from

(12.6) that:
2 D
h1 (h(L(s))) = h1 (h(L(s, ))),
and in particular
2 D
h(L(0)) = h(L(0, )).
Taking into account the denition of h this yields




r
D




r
ukm Lk (sm , ) = ukm L2k (sm ). (12.7)
k=1 m=1 k=1 m=1

The hyperplanes in the space R2r form a determining class and therefore,
the same is true for the hyperplanes in Cr . Taking into account (12.7), we
obtain that, for 1 k  and 1 m r, the random elements Lk (sm , )
and L2k (sm ) have the same distribution.
Now let K be a compact subset of D, and let f1 , . . . , f H(D). For an
arbitrary  > 0 we set
 
U = (g1 , . . . , g ) H(D) : max max |gk (s) fk (s)|  .

1k sK

Now choose a sequence {sm }, which is dense in K and dene, for r N,

 
Ur = (g1 , . . . , g ) H(D) : max max |gk (sm ) fk (sm )|  .

1k 1mr

From the properties of the random elements Lk (sm , ) and L2k (sm ) it follows
that:
2 Ur ).
m{ : L(s, ) Ur } = PL (L(s) (12.8)
Since the sequence {sm } was chosen to be dense in K, we have U1 U2 . . . ,
and Ur U , as r . Thus, letting r in (12.8), we nd
2 U ).
m{ : L(s, ) U } = PL (L(s)
Since the class of all sets U form a determining class, we obtain
D 2
L(s, ) = L(s).
This and (12.4) gives
D
L(s + iT1 ) L(s, ).
T1
L
This means that the probability measure PT1 converges weakly to the dis-
tribution of the random element L(s, ) as T1 . By Lemma 12.2, the
L
family {PT } is relatively compact. Since the random element L(s, ) does
not depend on the choice of the sequence T1 , the assertion of the theorem
follows. 

234 12 Joint Universality

Let m denote the abscissa limiting the intersection of all mean-square half-
planes for the Lk (s). Then, for M > 0, dene

DM = {s C : m < < 1, |t| < M }. (12.9)

Since DM D (see (12.1)), we obtain, by the induced topology, that L(s, )

is an H(DM ) -valued random element on the probability space (, B(), m).
Now denote by QL the distribution of L(s, ) on (H(DM ) , B(H(DM ) )).
L
Further, dene the probability measure QT by

L 1
QT (A) = meas { [0, T ] : L(s + i ) A} (12.10)
T
for A B(H(DM ) )). Then, by Theorem 12.1, we deduce
L
Corollary 12.4. Let L1 , . . . , L S. The probability measure QT converges
weakly to QL , as T .

This is the multi-dimensional analogue of Corollary 5.11.

12.2 A Transfer Theorem

Now, we shall prove a conditional joint universality theorem for elements

Lk S. Here, we shall make use of axiom (v). We cannot prove that any
two L-functions L1 (s), L2 (s) can approximate uniformly a nice function g(s)
simultaneously, since L1 (s) and L2 (s) need not be independent (see the remark
after Theorem 1.10 in Sect. 1.4). Therefore, we have to assume a certain inde-
pendence.
Given L1 , . . . , L S, we write their Euler product in the sequel
 m  1
jk (p)
Lk (s) = 1
p j=1
ps

for 1 k . In principle, the functions L1 (s), . . . , L (s) may have polynomial

Euler products of dierent degree; however, without loss of generality we may
assume that they all have degree equal to m by adding factors (1 jkps )1
(p)

with local roots jk = 0 if necessary. Recall that D denotes the intersection of

all mean-square half-planes of Lk (s) for 1 k  with the critical strip (see
(12.1)). In view of Theorem 2.4 (respectively, (4.2)), the strip of universality
D contains at least the strip
 
1 1 Lk
max max ,1 < < 1,
1k 2 1 + 2Lk

where Lk and Lk are dened by axioms (ii) and (iii).

 12.2 A Transfer Theorem 235

Further, we dene


m  
jk (p)b(p)
gpk (s, b(p)) = log 1 (12.11)
j=1
ps

for 1 k , where b(p) := {s C : |s| = 1}. Finally, we write

g p (s, b(p)) = (gp1 (s, b(p)), . . . , gp (s, b(p))). (12.12)

Then our transfer theorem takes the form:

Theorem 12.5. Let L1 (s), . . . , L (s) be elements of S, let K1 , . . . , K be com-

pact subsets of D with connected complements (where D is dened by (12.1)),
and, for each 1 k , let gk (s) be a continuous non-vanishing function
on Kk which is analytic in the interior of Kk . Suppose that the set of all
convergent series

g p (s, b(p)) with b(p) (12.13)
p

is dense in H(D) (where g p is dened by (12.11) and (12.12)). Then, for any


 > 0,
 
1
lim inf meas [0, T ] : max max |Lk (s + i ) gk (s)| <  > 0.
T T 1k sKk

Condition (12.13) is the multi-dimensional analogue of the assertion of

Theorem 5.10. This condition turns out to be essential; however, it would
be nice to replace it by a more lucid statement as, for example, statistical
independence.
Proof. Since all Kk are compact subsets of D, there exists a number M such
that Kk DM for 1 k , where the rectangle DM is dened by (12.9).
L
Our rst aim is to determine the support of the measure QT , dened by
(12.10). This will be done in a similar way as in the one-dimensional case,
Lemma 5.12. Recall that the minimal closed set SQL H(DM ) with
T

L
QT (SQL ) = 1
T

L
is called the support of QT ; it consists of all f := (f1 , . . . , f ) H(DM ) such
L
that for every neighbourhood G of f the inequality QT (G) > 0 holds.
L
Lemma 12.6. The support of the measure QT is the set

SM := { H(DM ) : (s) = 0 for s DM , or (s) 0}.
236 12 Joint Universality

For the proof we make use of

Lemma 12.7. Let {X n } be a sequence of independent H(G) -valued random

elements, where G is a region in C, and suppose that the series n=1 X n
converges almost everywhere. Then the support of the sum of this series is the
closure of the set of all f H(G) which may be written as a convergent series



f= fn with f n SX n ,
n=1

where SX n is the support of the (distribution of the) random element X n .

This is a multi-dimensional version of Theorem 3.16; for its proof, we refer
to Laurinikas [186] (Theorem 1.7.10).
Proof of Lemma 12.6. Recall that (p) is the projection of on the
coordinate space p = . The sequence {(p)} is a sequence of independent
random variables on the probability space (, B(), m). Thus
g p (s, (p)) := {(gp1 (s, (p)), . . . , gp (s, (p)))}

is a sequence of independent H(DM ) -valued random elements. The support

of each (p) is the unit circle , and therefore the support of the random
elements g p (s, (p)) is the set

{f H(DM ) : f (s) = (gp1 (s, b), . . . , gp (s, b))},

where


m  
jk (p)b
gpk (s, b) = log 1 with b .
j=1
ps

Consequently, by Lemma 3.16, the support of the H(DM ) -valued random

element


log L(s, ) = g p (s, (p))
p

is the closure of the set of all convergent series p f p (s). By condition (12.13),
the set of these series is dense in H(DM ) . The mapping h : H(DM )


H(DM ) , dened by
f = (f1 , . . . , f )  exp(f ) = (exp(f1 ), . . . , exp(f )),

is a continuous function sending log L(s, ) to L(s, ) and H(DM ) to SM


\{0}.
Therefore, the support SL of L(s, ) contains SM \ {0}. On the other hand,


the support of L(s, ) is closed. By Hurwitzs Theorem 5.13, it follows that

SM \ {0} = S  . Thus, S  S . By axiom (iv) on the polynomial Euler
M M L
product for elements in S, the functions
 12.2 A Transfer Theorem 237

m 
 
jk (p)(p)
exp(gpk (s, (p))) = 1
j=1
ps

are non-zero for s DM and . Hence, the functions Lk (s, ), 1 k ,

are almost surely convergent products of non-vanishing factors. If we apply
the Hurwitz theorem again, we conclude that L(s, ) SM 
almost surely.
Therefore, SL SM 
. The lemma is proved. 


Now we have nished all preliminaries and can start with the main part
of the proof of Theorem 12.5.
First, we suppose that the functions g1 (s), . . . , g (s) all have a non-
vanishing analytic continuation to DM . By Lemma 12.6, the gk s are con-
tained in the support SL of the random element L(s, ). Denote by the set
of functions = (1 , . . . ,  ) H(DM ) such that

max max |k (s) gk (s)| < .

1k sKk

L
Since by Corollary 12.4 the measure QT converges weakly to QL , as T ,
and since the set is open, it follows from Theorem 3.1 and properties of the
support that:
 
1
lim inf meas [0, T ] : max max |Lk (s + i ) gk (s)| < 
T T 1k sKk

L
= lim inf PT () QL () > 0. (12.14)
T

This proves the theorem in the case of functions gk (s) which have a non-
vanishing analytic continuation to DM .
Now assume that the functions gk (s) are as in the statement of the the-
orem. By Mergelyans approximation Theorem 5.15, for 1 k , there
exists a sequence of polynomials Gkn (s) which converges uniformly on Kk to
gk (s) as n . Since the gk (s) are non-vanishing on Kk , it follows that, for
suciently large m, the functions Gkm (s) do not vanish on Kk and

max max |gk (s) Gkm (s)| < . (12.15)
1k sKk 4

Since any of the polynomials Gkm (s) has only nitely many zeros, there exists
a region G whose complement is connected such that Kk G for 1 k 
and Gkm (s) = 0 on G. Hence, there exist continuous branches log Gkm (s)
on G, and the logarithms log Gkm (s) are analytic in the interior of G. Thus,
by Mergelyans Theorem 5.15, for 1 k , there exists a sequence of
polynomials Fkn (s) which converges uniformly on Kk to log Gkm (s) as n .
Thus, for all 1 k  and all suciently large n,
238 12 Joint Universality

max |Gkm (s) exp(Fkn (s))| < .
sKk 4

This and (12.15) imply


max max |gk (s) exp(Fkn (s))| < . (12.16)
1k sKk 2

From (12.14) we deduce

 
1 
lim inf meas [0, T ] : max max |Lk (s + i ) exp(Fkn (s))| < > 0.
T T 1k sKk 2

In combination with (12.16) this proves Theorem 12.5. 



12.3 Twisted L-Functions

Next, we want to give an example for a family of L-functions which satises

Condition (12.13) of Theorem 12.5 and thus is jointly universal.
For L S and a Dirichlet character mod q dene


a(n)
L (s) = (n). (12.17)
n=1
ns

It is obvious that such a twist satises the Ramanujan hypothesis (i) (since
|(n)| 1). By the complete multiplicativity of Dirichlet characters, we obtain
the identity

 m  1
a(n) j (p)
L (s) = (n) = 1 (p) ;
n=1
ns p j=1
ps

hence, also the axiom on the polynomial Euler product (iv) holds for L (s).
Next we observe that




|a(p)(p)|2 = |a(p)|2 = |a(p)|2 + O(1), (12.18)
px px px
p
q

since there are only nitely many primes dividing q. Hence, L (s) satises
the axiom on the mean-square with the same constant > 0 as for L(s).
However, we cannot show that the analytic axioms (ii) and (iii) hold in this
general frame analytic continuation is a dicult problem in analytic number
theory. Besides the assumption of axioms (ii) and (iii), we shall need one more
condition on the distribution of the values of the Dirichlet coecients a(p) for
prime p in arithmetic progressions.
 12.3 Twisted L-Functions 239

Theorem 12.8. Let L S. Let q1 , . . . , q be positive integers, let 1 mod

q1 , . . . ,  mod q be pairwise non-equivalent characters. Further, for 1
k , let gk be a continuous non-vanishing function on Kk which is analytic
in the interior, where Kk is a compact subset of the strip D with connected
complement (where D is dened by (12.1)). For 1 k , assume that Lk (s)
satises the axioms (ii) and (iii), and that
1

lim |a(p)|2 = (12.19)
x (x) (q)
px
ph mod q

holds for any prime residue class h mod q, where q is the least common mul-
tiple of q1 , . . . , qk , and is the quantity from axiom (v). Then, for any  > 0,
 
1
lim inf meas [0, T ] : max max |Lk (s + i ) gk (s)| < > 0.
T T 1k sKk

Notice that the additional condition (12.19) implies axiom (v).

Proof. Because of Theorem 12.5 it suces to prove that the set of all conver-
gent series

g p (s, b(p)) with b(p)
p

is dense in H(D) , where, according to (12.11),



m  
j k (p)b(p)
gpk (s, b(p)) = log 1 , (12.20)
j=1
ps

and g p (s, b(p)) = (gp1 (s, b(p)), . . . , gp (s, b(p))). We follow the proof of The-
orem 5.10. Almost all arguments can be easily extended to our multi-
dimensional case; however, in one instance the linear independence of our
characters plays a crucial role. The main tool for the proof will be a multi-
dimensional analogue of Theorem 5.7:
Theorem 12.9. Let D be a simply connected domain in the complex plain.
Suppose that the sequence of functions {f n = (fn1 , . . . , fn )} be a sequence in
H(D) satises:
If 1 , . . . ,  are complex Borel measures on (C, B(C)) with compact sup-
ports contained in D such that
  





 
 fnk dk  < ,
 C 
n=1 k=1

then

sr dk (s) = 0 for r N {0}, k = 1, . . . , .
C
240 12 Joint Universality

The series n=1 f n converges on H(D) .
For every compact K D



max |f n (s)|2 < .
sK
n=1

Then the set of all convergent series n=1 b(n)f n with b(n) is dense in
H(D) .

A proof can be found in Bagchi [10]; it follows along the lines of Theorem 5.7.
We continue with the proof of Theorem 12.8. For 1 k , dene

m  
j (p)k (p)
gpk (s) = gpk (s, 1) = log 1 ,
j=1
ps

and g p (s) = (gp1 (s), . . . , gp (s)). Moreover, for 1 k  and a parameter
N N which will be chosen later, dene

gpk (s) if p > N,
gpk (s) =
0 if p N,

and let g p (s) = (gp1 (s), . . . , gp (s)). In the proof of Theorem 5.10, we have
shown that there exists a sequence {b(p) : b(p) } such that the series


b(p)gpk (s)
p

converges in H(D) (see (5.12). Similarly as in the one-dimensional case, one

can show the existence of a sequence {b(p) : b(p) } such that the series


b(p)g p (s) (12.21)
p

converges in H(D). Next, we prove that the set of all convergent series


b(p)g p (s) with b(p) (12.22)
p

is dense in H(D) . It suces to show that the set of all convergent series


b(p)f p (s) with b(p) (12.23)
p

is dense, where fpk (s) := b(p)gpk (s) and f p (s) := (fp1 (s), . . . , fp (s)). For this
purpose we apply Theorem  12.9. In view of the convergence of (12.21), it
follows that the series p f p (s) converges in H(D) ; this implies the second
 12.3 Twisted L-Functions 241

assumption of Theorem 12.9. The third assumption follows immediately from

the fact that for any component


1
|fpk (s)|2  ,
p p
p2

as in the one-dimensional case.

To verify the rst assumption, let 1 , . . . ,  be complex Borel measures
on (C, B(C)) with compact support contained in D such that
   




 fpk (s) dk (s) < . (12.24)
 C 
p k=1

On account of the Euler product representation, we nd

m

j (p)k (p) a(p)k (p)
gpk (s) = = + rpk (s)
j=1 =1
ps ps

for 1 k , where rpk (s)  p2 . Then,

 

 a(p)k (p)b(p) 
|a(p)|

fpk (s) = + |rp (s)|  1
p
 ps  pN p
p>N
p
qk

uniformly on compact subsets of D. Hence, (12.24) implies

 


  
s 
a(p) k (p)p dk (s) < . (12.25)
p
 C 
k=1

Recall that q denotes the least common multiple of the moduli q1 , . . . , q and,
for 1 k , denote by k the character mod q induced by k mod qk . Now
let h be an integer satisfying 1 h q and coprime with q. If p h mod q,
then
k (p) = k (p) = k (h).
Hence, we may rewrite (12.25) as
  


 

s
a(p) k (h) p dk (s) <
 C 
ph mod q k=1

or, equivalently,


|a(p)|h (log p)| < , (12.26)
ph mod q
242 12 Joint Universality

where



h (z) := exp(sz) dh (s) and h (s) := k (h)k (s) (12.27)
C k=1

for 1 h q coprime with q. Recall our argument in the one-dimensional

case. For any prime p we have dened angles p [0, 2 ] by
 
m 



|a(p)| =  j (p) = m cos p .
 j=1 

Hence, we can replace (12.26) by

cos p |h (log p)| < , (12.28)
ph mod q

Our next aim is to show that h (z) vanishes identically. We apply Theo-
rem 5.9. Now, we choose a suciently large positive constant M such that the
support of all k is contained in the region {s C : m < < 1, |t| < M }.
Then, by the denition of h (s),
 


|h (iy)| exp(M y) | dk (s)|
k=1 C

for y > 0. Therefore,

log |h (iy)|
lim sup M,
y y
and the rst condition
 ofTheorem 5.9 is valid with = M . Fix a number
with 0 < < min 1, m , where m is the degree of the polynomial dening
the local Euler factors, and is the quantity appearing in the axiom on the
mean-square for L S. Next, denote by P(h) the set of all prime numbers
p h mod q and dene

P (h) = {p : p h mod q prime and cos p > } .

Then (12.28) yields

|h (log p)| < (12.29)
pP (h)

for any h with 1 h q, coprime with q. Further, x a number with

0<< M , and dene
     & 
1 1
A = n N : r n , n + with |h (r)| exp(r) .
4 4
 12.3 Twisted L-Functions 243

Now let
     
1 1
:= (n) := exp n and := (n) := exp n+ .
4 4
Then





1
|h (log p)| |h (log p)|
p
pP (h) nA p

P(h) nA pP
(h)
<p <p

and, in view of (12.29),

1
< . (12.30)
p
nA p P
(h)
<p

Let
(x; h mod q) = {p x : p P (h)}.
Then we obtain, for < u ,




(cos p )2 1 + 2 1
P
p (h) p

P
(h) P
p (h)\

P
(h)
<pu <pu <pu

= (1 2 )( (u; h mod q) (; h mod q))

+ 2 ((u; h mod q) (; h mod q). (12.31)

By condition (12.19),

1

(cos p )2 = |a(p)|2 (x; h mod q).
px
m2 px
m2
ph mod q ph mod q

Let be a small positive constant. Substituting the latter formula in (12.31)

gives

(u; h mod q) (; h mod q)

 
2
2
m + O(1) ((u; h mod q) (; h mod q))
1 2

for u (1 + ), as n . Thus, we get by partial summation

1  d (u; h mod q)  2 2 
d(u; h mod q)
= m + O(1)
p u 1 2
u
pP
(h)
<p
 


2 1
m2
+ O(1) , (12.32)
1 2 p
p (h)P
(1+)<p
244 12 Joint Universality

as n . By the prime number theorem for arithmetic progressions (that

is (1.29) with a suitable remainder term), we nd for the prime residue class
h mod q

1 1
= log log x + Ch + O((log x)2 )
px
p (q)
ph mod q

with some constant Ch , depending only on h mod q. Thus

   
1 1 1 log(1 + ) 1 1
= +O .
p (q) 2 n n2
p (h)P
(1+)<p

This gives in (12.32)

   
1 1 m2 2 1 log(1 + ) 1 1
+O ,
p (q) 1 2 2 n n
pP
(h)
<p

as n . Hence, it follows from (12.30) that:

1
< .
n
nA

Let A = {aj : j N} with a1 < a2 < . This implies

aj
lim = 1. (12.33)
j j

By the denition of the set A, there exists a sequence {j } such that

   
1 1
aj < j aj + and |h (j )| exp(j ).
4 4

Hence, from (12.33) it follows that:

j log |h (j )|
lim = and lim sup 1.
j j j j

Applying Theorem 5.9, we obtain

log |h (r)|

lim sup 1. (12.34)
r r

However, by Lemma 5.8, if h (z) does not vanish identically, then

log |h (r)|

lim sup > 0,
r r
 12.4 First Applications 245

contradicting (12.34). Therefore, h (z) vanishes identically. By (12.27),

dierentiation yields 
sr dh (s) 0,
C
respectively


 
k (h) sr k (s) 0,
k=1 C

for r = 0, 1, 2, . . .. The latter identity holds for any h coprime with q. Multi-
plying with i (h) for some arbitrary i with 1 i  and summing up over
all prime residue classes h mod q, we get
 



0 sr dk (s) i (h)k (h)
k=1 C 1hq
(h,q)=1

By the orthogonality relation for characters (1.28), this gives


0 sr i (s),
C

and so i (s) 0. Since i was arbitrary, all 1 (s), . . . ,  (s) vanish identically.
Thus, the rst assumption of Theorem 12.9 is also satised and we deduce
the denseness of all convergent series (12.22). It  remains to show that this
does not change when we add terms of the form pN g p (s). However, this
can be done in a similar way as in the one-dimensional case. The theorem is
proved. 


12.4 First Applications

As a rst application of Theorem 12.8 we extend Voronins joint universality

Theorem 1.10 for Dirichlet L-functions to pairwise non-equivalent characters.
Dirichlet L-functions to primitive characters lie in the Selberg class (see
Sect. 6.1) and so they are universal by Theorem 6.12; however, this is not
true for imprimitive characters since the axiom on the functional equation
is violated. What about Dirichlet L-functions to imprimitive characters with
respect to universality? Assume that is an imprimitive character mod q,
induced by a primitive character mod q . Then, by (1.27),
 (p)


L(s, ) = L(s, )f (s), where f (s) := 1 .
ps
p|q

Obviously, f (s) is analytic in the whole complex plane, all its zeros are located
on the line = 0, and it satises |f ( + it)| 1 for 0. Thus, L(s, ) is an
246 12 Joint Universality

entire function. This yields axiom (ii) and axiom (iii). In case of a primitive
character, the mean-square half-plane of L(s, ) is > 12 by Corollary 6.11.
For Dirichlet L-functions to imprimitive characters, we can do better than
applying Theorem 2.4. Taking into account the properties of the function
f (s) outlined above, we nd
 T  T
|L( + it, )| dt
2
|L( + it, )|2 dt
1 1

for 0. Thus, by Corollary 6.11 and Carlsons Theorem 2.1, the asymp-
totic mean-square of L(s, ) exists in the same region as the mean-square of
L(s, ). It remains to verify condition (12.19) which we may rewrite as
1

lim 1= ,
x (x) (q)
px
ph mod q

where h is coprime with q. This is the prime number theorem in arithmetic

progressions (1.29). The same reasoning holds for principal characters. Thus,
we have proved Theorem 1.10: Dirichlet L-functions to pairwise non-equivalent
characters are jointly universal.
Another example for an application of Theorem 12.8 is provided by L-
functions associated with newforms (see Sect. 1.5). The joint universality for
twists of these L-functions, Theorem 1.12, was obtained by Laurinikas and
Matsumoto [202]. Here, condition (12.19) is satised because of the asymptotic
formula (1.38).

12.5 A Conjecture

We conclude this chapter with a conjecture. So far, all known explicit fam-
ilies of jointly universal Dirichlet series were constructed in a very special
way, namely as twists of a single universal Dirichlet series twisted by pairwise
non-equivalent characters. Examples are Dirichlet L-functions (see Theorem
1.10 for the continuous and Theorem 5.17 for discrete joint universality) and
L-functions to newforms twisted by Dirichlet characters (Theorem 1.12); The-
orem 12.8 may be considered as a conditional attempt to generalize this type
of joint universality. The knowledge of these jointly universal L-functions can
be used to prove joint universality for other Dirichlet series as, for example,
Dirichlet series with periodic coecients (see Theorem 11.8) or Hurwitz zeta-
functions with rational parameters (Theorem 11.12), Dedekind zeta-functions
(see Sect. 13.1) or further L-functions to number elds (Sect. 13.8). In all these
examples, the result is obtained by suitable representations of the functions in
question as linear combinations or products of jointly universal L-functions.
Here, the necessary independence for these joint universality theorems is
deduced from the linear independence of Dirichlet characters (or ray class
 12.5 A Conjecture 247

group characters in Sect. 13.8) over C. We should mention that there exist
also analogous results for additive characters: Matsumoto and Laurinikas
[200, 203] obtained joint universality for certain classes of Lerch zeta-functions,
which may be regarded as twists of strongly universal Hurwitz zeta-functions
by additive characters exp(2i aq ) with rational aq (in addition with some
algebraic side restrictions).
To replace the orthogonality relation of (multiplicative or additive) char-
acters by a more general statement, Laurinikas [191], Laurinikas and
Matsumoto [199], respectively, introduced an interesting matrix condition. To
explain this matrix condition in a few words, given a collection of Matsumoto
zeta-functions, it is assumed that there are nitely many distinct subclasses
Pj of the set of prime numbers each of which of positive density such that
the Dirichlet coecients ak (p) for prime p multiplied with some weight are
constant Bkj on the Pj . If the rank of the matrix built from the values (Bkj )
is as large as possible, the coecients ak (p) are linearly independent which
allows to deduce the joint universality of the Matsumoto zeta-functions. Since
the subsets Pj are not necessarily related to the multiplicative structure of
prime residue classes, this approach allows generalizations of joint universal-
ity theorems for Dirichlet series twisted by multiplicative characters.
We may ask whether this is all? The methods sketched above fail to
prove, for example, joint universality for (s) and an L-function to some new-
form. More generally, we ask for a necessary and sucient condition that a
given nite family of L-functions is jointly universal? In the context of the
Selberg class, we expect that Selbergs Conjecture B (or a suitable quantita-
tive extension) could be used to answer this question. By Selbergs Conjecture
B, primitive functions are expected to form an orthonormal system. Recall
that Bombieri and Hejhal [39] proved, assuming a stronger version of Selbergs
conjecture B, the statistical independence of any collection of independent L-
functions in any family of independent elements of S. With regard to this
statistical independence, predicted by Selbergs Conjecture B, it seems rea-
sonable to conjecture that any nite collection of distinct primitive functions
in the Selberg class is jointly universal. Moreover, we expect that any two
functions L1 (s), L2 (s) S are jointly universal, if and only if

a1 (p)a2 (p)
= O(1), (12.35)
p
px

where the Dirichlet series coecients of Lj (s) are denoted by aj (n) and the
summation is over the prime numbers. We shall briey illustrate how this ts
to some special cases of pairs of jointly universal functions.
Assume that we are given functions Lj (s) which are twists of an element of
S by distinct primitive Dirichlet characters j mod q, i.e., aj (n) = a(n)j (n)
for j = 1, 2. Then
248 12 Joint Universality

a1 (p)a2 (p)
|a(p)|2

= 1 (p)2 (p)
p p
px px


|a(p)|2
= 1 (h)2 (h) .
p
h mod q px
ph mod q

Now suppose a certain equidistribution: for any prime residue class h mod q,
the inner sum on the right is asymptotically equal to c log log x + O(1), where
c is some positive constant, independent on h mod q. By the orthogonality
relation for characters, it follows that:

a1 (p)a2 (p)

=c 1 (h)2 (h) log log x + O(1) = O(1),
p
px h mod q

which is (12.35). On the other hand, with regard to Theorem 12.8 we expect
L1 (s) and L(s) to be universal (which is known in particular cases).
Finally, let us consider an example where we do not have joint universality,
(s) and (s)2 , say. Here, since the Dirichlet coecients of (s)2 are given by
the divisor function d(n), the left-hand side of (12.35) is obviously unbounded.
13
L-Functions of Number Fields

The zeta-function of a eld is like the atom of physics.

H.M. Stark

In this chapter, we shall obtain universality for many classical L-functions,

including Dedekind zeta-functions as well as Hecke and Artin L-functions.
Further, we shall briey discuss the arithmetic axioms in the denition of S
with respect to the Langlands program. We give only a sketch of the analytic
theory of all these L-functions and refer to Bump et al. [47] for further details.
For details from algebraic number theory we refer to Heilbronns survey [129],
the monographs of Murty and Murty [270], of Neukirch [279], and, last but
not least, Starks article [337].

13.1 Dedekind Zeta-Functions

Let K be an algebraic number eld (i.e., a nite extension of Q). The associ-
ated Dedekind zeta-function is for > 1 dened by

1  1
1
K (s) = = 1 ;
a
N(a)s p
N(p)s

here the sum is taken over all non-zero integral ideals, the product is taken
over all prime ideals of the ring of integers of K and N(a) denotes the norm of
the ideal a. This generalization of the Riemann zeta-function was introduced
by Dedekind [66] in order to obtain information about the multiplicative struc-
ture of number elds. The identity between series and product is an analytic
version of the unique factorization of integral ideals into prime ideals. In an
algebraic number eld K over Q of degree d any rational prime number p has
a unique factorization into a product of prime ideals

r
e


r
(p) = pj j with N(pj ) = pfj and ej fj = d; (13.1)
j=1 j=1
250 13 L-Functions of Number Fields

of course, the non-negative integers ej , fj , and r depend on p (which is not

indicated here for simplicity). Hence we can write
 1
1   r 
1
1
K (s) = 1 = 1 . (13.2)
p
N(p)s p j=1
psfj
pj |(p)

Assume that K = Q() with an algebraic integer and F (X) is the minimal
polynomial of . Then for all but nitely many primes p, the factorization of
F (X) modulo p determines the arithmetic of the local Euler factor: if

r
F (X) Fj (X) mod p (13.3)
j=1

is the factorization into irreducible factors, then fj = deg Fj in (13.2). Of

course, we have r [K : Q] = d. Thus, the Dedekind zeta-function has a
representation as a polynomial Euler product. Since the norm of an integral
ideal is a positive rational integer, the series can be rewritten as an ordinary
Dirichlet series

1

fK (n)
s
= ,
a
N(a) n=1
ns
where fK (n) counts the number of integral ideals a with norm N(a) = n.
We see that the Riemann zeta-function is the Dedekind zeta-function of Q
and, as a matter of fact, Dedekind zeta-functions share many properties with
Riemanns zeta-function. The Dirichlet series dening the Dedekind zeta-
function K (s) converges for > 1, independent of the eld K. To see this
note that, for > 1,
 1
 1   1
d
|K (s)| 
1 N(p)s  1
p
= ()d ,
p p

since there are at most d = [K : Q] many primes p lying above each rational
prime p and N(p) is smallest if (p) splits completely.
Landau [176] proved the analogue of the prime number theorem for number
elds, that is the prime ideal theorem
x
K (x) := {p OK prime : N(p) x} . (13.4)
log x
Hecke [124] obtained the rst deeper results concerning the analytic behaviour
of Dedekind zeta-functions. He showed that K (s) has an analytic continuation
to C except for a simple pole at s = 1 and satises a Riemann-type functional
equation. Especially the residue of the simple pole contains important infor-
mation about the underlying number eld. Hecke proved that
2r1 (2)r2 hR
lim (s 1)K (s) = % , (13.5)
s1+ |dK |
 13.1 Dedekind Zeta-Functions 251

where r1 is the number of real conjugate elds, 2r2 is the number of complex
conjugate elds, h is the class number, R is the regulator, is the number of
roots of unity in the group of units, and dK is the discriminant of K. The class
number is the number of equivalence classes of fractional ideals of K modulo
principal ideals, and so it measures the deviation of OK from having unique
prime factorization. Gauss conjectured that the class numbers h = h(d) of
imaginary quadratic number elds K = Q( D) with discriminant d < 0
tend with d to innity; note that d = D if D 1 mod 4, and d = 4D if
D 2, 3 mod 4. This was rst proved by Heilbronn [128] and in rened form
by Siegel [330]. The problem of nding an eective algorithm to determine
all imaginary quadratic elds with a given class number h is known as the
Gauss class number h problem. This problem is of interest with respect to the
non-existence of exceptional real zeros of Dirichlet L-functions o the critical
line. The general Gauss class number problem was solved by Goldfeld, Gross
& Zagier [102, 107]. A complete determination of the imaginary quadratic
elds with class number one was rst given by Heegner [127] (but his solution
was not completely accepted due to a number of gaps), Baker [12], and Stark
[336] (independently):

h=1 d {3, 4, 7, 8, 11, 19, 43, 67, 163}.

Note that class number one is equivalent to unique prime factorization in the
corresponding ring of algebraic integers.
Voronin [362] proved functional independence for any nite collection of
Dedekind zeta-functions to distinct quadratic number elds and
Dedekind zeta-functions to distinct cyclotomic number elds Q(r ), where
r is a primitive rth root of unity and r is assumed to be odd and square-
free.1
Gonek [104] showed that a single Dedekind zeta-function to an abelian number
eld is universal in the strip 1/2 < < 1 and Reich [305, 306] succeeded in
proving discrete universality for Dedekind zeta-functions to arbitrary number
elds (Theorem 5.16); however, here the strip of universality is restricted to
1 [K : Q]1 < < 1. We shall explain the reason for this dierence.
If we are dealing with a Dedekind zeta-function K (s) to a non-abelian eld,
we may apply our universality Theorem 6.12 (this is easily deduced from the
properties listed above). Since K (s) is an element of degree d = [K : Q] in the

1
Voronins paper is published in the same year, 1975, as his universality theorem
for (s) and his reasoning is based on the very same ideas as in his approach
to universality, however, while writing [362], the step to approximating func-
tions rather than numbers was not done (actually, the rearrangement theorem for
series in Hilbert spaces was then not available; cf. [198]). Later Voronin included
the joint universality for Dedekind zeta-functions for such number elds in his
thesis [364] (and also the monograph [166] of Karatsuba & Voronin contains these
results).
252 13 L-Functions of Number Fields

Selberg class, this yields universality in the same strip as obtained by Reich.
Now assume that we are given an abelian number eld K (i.e., a nite Galois
extension with abelian Galois group). Then


K (s) = L(s, k ), (13.6)
k=1

where k denotes the primitive character which induces k , resp. the principal
character mod 1 which induces the trivial character of the Galois group,
and  is the number of primitive characters. The simplest example provide
quadratic number elds K = Q( D) with
K (s) = (s)L(s, d ),
where d is the Jacobi symbol and d is the discriminant. (For these facts we
refer to Washington [368].)
Now it is clear how to apply the joint universality for Dirichlet L-functions
in order to obtain universality for K (s). The product (13.6) consists of
Dirichlet L-functions to pairwise non-equivalent characters and we may apply
the joint universality Theorem 1.10. For all but one of the factors L(s, k ) we
dene the target function to be gk (s) 1, whereas the last one may be chosen
freely, g(s) say. Then
 
g(s) = gk (s).
k=1
If g(s) is assumed to be non-vanishing and continuous on some compact subset
K of 12 < < 1 with connected complement, and analytic in the interior, then
Theorem 1.10 implies
 
1
lim inf meas [0, T ] : max max |L(s + i, k ) gk (s)| < > 0.
T T 1k sK

Hence, we nd for the product (13.6)

 
1
lim inf meas [0, T ] : max |K (s + i ) g(s)| < > 0.
T T sK

However, we can do slightly more. In the sequel we shall sketch how to gen-
eralize this idea in order to obtain joint universality.
Dependence relations between Dedekind zeta-functions are of special
interest with respect to so-called class number relations (see Brauer [43]).
Besides functional independence we shall consider the concept of multiplica-
tive independence. The functions f1 (s), . . . , fm (s) are said to be multiplica-
tively independent if the identity

m
fj (s)aj = 1 with aj Z
j=1

holds only trivially, i.e., a1 = = am = 0.

 13.1 Dedekind Zeta-Functions 253

Recently, Marszaek [238] showed that the multiplicative independence

of Dedekind zeta-functions to abelian elds is equivalent to their functional
independence; he also gave a sucient condition for their independence as
well as all possible multiplicative dependence relations. His argument relies
essentially on the representation of Dedekind zeta-functions as products of
Dirichlet L-functions.
Assume that we are given abelian number elds K1 , . . . , Km . Then, by
the theorem of KroneckerWeber, their compositum K is a subeld of some
cyclotomic eld Q(r ), where r is an rth root of unity. We denote by
G = {1 , . . . ,  } the group of Dirichlet characters associated with K and
by G1 , . . . , G the subgroups of G corresponding to K1 , . . . , Km , respectively.
We put 
1 if k Gj ,
jk = (13.7)
0 otherwise.
Then, for 1 j m, by class eld theory we have the multiplicative repre-
sentation
 
Kj (s) = L(s, k )jk , (13.8)
k=1

where kdenotes the primitive character which induces k . (For these facts
we once more refer to Washington [368].)
Sander and Steuding [315] showed that both concepts of independence,
functional independence and multiplicative independence, are equivalent to
the discrete joint universality for the Dedekind zeta-functions in question.
Theorem 13.1. Suppose that K1 , . . . , Km are abelian number elds. Then the
following statements are equivalent:
(i) The associated Dedekind zeta-functions are jointly universal: Let K be a
compact subset of 12 < < 1 with connected complement, let q be a positive
integer, and for each 1 j m let fj (s) be a continuous non-vanishing
function on K which is analytic in the interior. Then, for any real = 0
and any  > 0,
 
1
lim inf  n N : max max |Kj (s + in) fj (s)| <  > 0.
N N 1jm sK

(ii) The matrix M := (jk ) 1jm with jk given by (13.7) has rank m .
1k
(iii) The associated Dedekind zeta-functions are multiplicatively independent.
(iv) The associated Dedekind zeta-functions are functionally independent.
The proof is quite simple. The equivalence of (ii)(iv) is from Marszaek [238].
We sketch the argument. Suppose that the matrix M has rank m  and that

m
Kj (s)aj = 1
j=1
254 13 L-Functions of Number Fields

for some integers aj . In view of (13.8) this implies


m 
 
 m
L(s, k )aj jk = L(s, k ) j=1 j jk ,
a
1=
j=1 k=1 k=1

resp.


m
aj jk = 0 for k = 1, . . . , 
j=1

by the functional independence of Dirichlet L-functions. However, since the

rank of M is equal to m, the latter system of linear equations has a unique
solution: a1 = = am = 0. Conversely, if we would have m >  or the rank
would be less than m, the system would have non-trivial solutions. Hence,
(ii) is equivalent to the multiplicative independence (iii). Clearly, functional
independence implies multiplicative independence. Next we show that the
converse is also true.
Assume that the Dedekind zeta-functions are multiplicatively independent
and that, for continuous functions F0 , F1 , . . . , FN : Cm C, not all identi-
cally vanishing, and any s,


N
sk Fk (K1 (s), . . . , Km (s)) = 0. (13.9)
k=0

For each k dene Gk = Fk (P (s)), where P := (P1 , . . . , Pm ) : C Cm is

given by



Pj (s1 , . . . , s ) = skjk .
k=1
Now it follows from (13.9) that


N
sk Gk (L(s, 1 ), . . . , L(s,  )) = 0.
k=0

The functional independence of Dirichlet L-functions implies that the Gk are

vanishing identically and it is easy to deduce the same for the Fk . This shows
that multiplicative and functional independence are equivalent. In particular,
we have shown (ii) (iv).
In Sect. 10.3, we have shown that joint universality implies functional
independence: (i) (iv). It remains to prove that (ii) implies (i). Given the
condition on the rank of the matrix M, the system of linear equations

t
t
F = MG , where F := Fj (s) , G := Gk (s)
1jm 1k

with Fj (s) := log fj (s) is solvable in G; note that the logarithms Fj (s) =
log fj (s) for s K exist by the condition of the theorem. Now we switch from
the identities
 13.1 Dedekind Zeta-Functions 255


Fj (s) = jk Gk (s)
k=1

by exponentiation to


fj (s) = gk (s)jk ,
k=1

where gk (s) := exp(Gk (s)); here the functions gk (s) are continuous non-
vanishing functions on K which are analytic in the interior. By Theorem 5.17
we may approximate any gk (s) by certain shifts L(s + in, k ), i.e.,

max max |L(s + in, k ) gk (s)| < . (13.10)

1k sK

In view of (13.8) we have



Kj (s + in) = L(s + in, k )jk
k=1


= (gk (s) + O())jk = fj (s) + O()
k=1

for any s K and any j. Since (13.10) holds for a set of positive integers n
with positive lower density, the assertion follows.
Of course, in view of Theorem 1.10 we can also replace statement (i) about
discrete universality by joint continuous universality: for any  > 0,
 
1
lim inf meas [0, T ] : max max |Kj (s + i ) fj (s)| <  > 0.
T T 1jm sK

Theorem 13.1 is best possible. We cannot have joint universality for

any family of Dedekind zeta-functions to dierent number elds. If m > ,
then there exist restrictive dependencies for the Dedekind zeta-functions. For
instance,
Q(3) Q(21/3 ) = Q Q(21/3 ,exp(2i/3)) (13.11)
(see Sect. 13.6). Indeed, if the Galois group has more normal subgroups
than conjugacy classes, then such algebraic relations for the corresponding
Dedekind zeta-functions exist and they cannot be jointly universal. Recently,
Bauer [16] proved that if there is no algebraic relation for a collection of
Dedekind zeta-function to nite normal extensions of the rationals, then they
are functionally independent; his proof relies on the joint universality for
Artin L-functions to non-equivalent characters of the Galois group of a Galois
extension K/Q (see Sect. 13.8).
It is an interesting question to which extent the Dedekind zeta-function
determines the eld. One can show that the Dedekind zeta-function K (s)
determines the minimal normal extension L of Q containing K and thus we
256 13 L-Functions of Number Fields

have to ask whether there exist non-conjugate subgroups of Gal(L/Q) giving

the same induced trivial character. This is indeed possible! Two number elds
K1 and K2 are said to be arithmetically equivalent if their Dedekind zeta-
functions are the same. The rst non-trivial example was given by Gassmann
[93]. Perlis [291] proved that arithmetically equivalent non-isomorphic elds
have at least degree 7 and that this bound cannot be improved. An explicit
example of degree 8 is for instance Q((3)1/8 ) and Q((48)1/8 ) which is due
to Perlis and Schinzel [292]. By the rank condition from Marszaek [238] it
follows that two distinct abelian elds K1 and K2 cannot be arithmetically
equivalent since otherwise their Dedekind zeta-functions have the same factor-
ization into Dirichlet L-functions and K1 = K2 if and only if their character
groups are equal: G1 = G2 (see Washington [368]). A stronger statement was
shown by Nicolae [282] who proved that Dedekind zeta-functions to distinct
nite Galois extensions are linearly independent over C. This also shows that
linear independence of Dedekind zeta-functions does not imply their joint
universality.

13.2 Grssencharacters
In 1920, Hecke [125] introduced a new class of L-functions which generalize
the concepts of Dedekind zeta-functions and Dirichlet L-functions. However,
rst of all we have to introduce Hecke grssencharacters which represent the
most general extension of Dirichlet characters to number elds.
Given a number eld K of degree n over Q, there are exactly n embeddings
K(j) of K into C, for 1 j n given by

K   (j) K(j) ,

where the (j) denote the conjugates of ; we assume that among these there
are r1 real and 2r2 complex embeddings (that makes n = r1 + 2r2 ). We denote
the real embeddings by
K(1) , . . . , K(r1 )
and the complex embeddings which are pairwise complex conjugate by

K(r1 +1) , . . . , K(r1 +r2 +1) = K(r1 +1) , . . . , K(n) = K(r1 +r2 ) .

Let f be a non-zero integral ideal of K. The unit group modulo f is dened

to be the set of all units  1 mod f which are totally positive (i.e., all its
real conjugates are positive) and we denote it by U(f). It is easily seen that
U(f) is indeed a group. By Dirichlets unit theorem there exist r = r1 + r2 1
units 1 , . . . , r and a root of unity in K such that any  U(f) has a unique
representation
 = m 1n1 rnr
 13.2 Grssencharacters 257

with integers m, nk . The units 1 , . . . , r are said to be fundamental units of

U(f) although they are not uniquely determined. Dene the matrix

(j) 1 if 1 j r1 ,
(ej log |k |)1j,kr , where ej :=
2 if r1 < j r1 + r2 .

Then the regulator R(f) is dened to be the absolute value of the determinant
of this matrix:
(j)
R(f) = | det(ej log |k |)|;
it should be noted that the regulator does not depend on the choice of the
fundamental units k .
We further denote by I(f) the multiplicative group generated by all ideals
coprime with f. The principal ray class P(f) is the subgroup of I(f) consisting
of all principal ideals of the form (/) satisfying
0 = , OK (the ring of integers);
mod f;
/ is totally positive.
The factor group
G(f) := I(f)/P(f)
is called the ray class group mod f, and its elements, the ray classes, might be
regarded as the analogues of the residue classes in the rational number eld
case. One can show that G(f) is a nite abelian group and we denote its order
by h(f).
We shall give a brief
example. For the sake of simplicity we shall consider
the number eld Q( 5) and choose f = (1) in which case G = G((1)) is the
class group. (In general, the narrow divisor class group G((1)) is not equal to
G, however, in complex quadratic elds there are no sign conditions.) There is
no unique prime factorization, and so the class number h is greater than one.
One can deduce from Minkowskis theorem on linear forms that every class of
G contains an integral ideal a with norm
%
N(a) | dK |,

where dK is the discriminant of the number eld, that is in our case dK = 20.
Obviously, this observation proves the niteness of the class number. However,
we can also use it to get information about the structure of the class group.
For this purpose we observe that the only prime ideals p with N(p) 4 can
be among the prime ideal divisors of (2) and (3). By the splitting of primes
in quadratic number elds, we nd

(2) = p21 with p1 = (2, 1 + 5) = (2, 1 5),

(3) = p2 p2 with p2 = (3, 1 + 5) = p2 = (3, 1 5).

Hence, the ideals with norms less than or equal to 4 are p1 , p2 , p2 , and (2) = p21 .
It is easy to see that p1 is not principal and represents a class of order two.
258 13 L-Functions of Number Fields

Furthermore, it turns out that

all other ideals lie in this class or are principal.
Hence the class number of Q( 5) equals two and we have a description of
the associated class group.
Now we are in the position to dene Hecke characters. Suppose we are
given numbers aj and k satisfying
aj {0, 1} for 1 j r1 and aj Z for r1 < j r1 + r2 ;
k R for 1 k r1 + r2 such that 1 + + r1 +r2 = 0.
Then we dene a function : K C by
r1
+r2 r1
+r2  aj
(j)
() = |(k) |ik .
k=1 j=1
|(j) |

Obviously, is unimodular. Since the sum of the k s vanishes, it follows

that is trivial on Q . We suppose that the kernel of contains the unit
group modulo f, i.e., () = 1 for any  U(f). Then induces a character
on P(f).
If a non-trivial homomorphism : I(f) C is identied with on
P(f), that is
(a) = () for a = () P(f),
then is said to be a grssencharacter modulo f (resp. Hecke character in
some literature). If all numbers aj , k are equal to zero, then is said to be a
ray class character, and if additionally f = (1), then is an ideal class group
character (that is one of the nitely many characters of the class group of K).
However, there are innitely many grssencharacters since, for example, the
function  4m

a = ()  m (a) = = exp(4im arg()) (13.12)
||
for a = 0 and any integer m is a primitive grssencharacter. If there exists an
ideal f f and a grssencharacter mod f such that = on I(f), then
is said to be induced by ; otherwise is called primitive and f is said to
be the conductor of .

13.3 Hecke L-Functions

Let K be a number eld, f be an ideal of K, and let modulo f be a grssen-

character. We extend to the group I of all fractional ideals of K by setting
(a) = 0 if a is not coprime with f. Then the Hecke L-function associated to
is (formally) given by

(a)  (p)
1
L(s, ) = = 1 , (13.13)
a
N(a)s p
N(p)s
 13.3 Hecke L-Functions 259

where the sum is taken over all non-zero integral ideals a of K, the product is
taken over all prime ideals p, and N(a) denotes the norm of the ideal a (as in
the case of the Dedekind zeta-function).
Hecke L-functions to grssencharacters are the analogues of Dirichlet L-
functions: if K = Q, f = (q) with q Z, and 1, then the construction
above leads without the totally positive condition to

G(f) = (Z/qZ) /{1}.

If is the trivial (principal) character, then the Hecke L-function for a number
eld K is nothing else but the Dedekind zeta-function. Note that what we
call Hecke L-functions are in some literature called (generalized) Dirichlet
L-functions.
Both the series and the product (13.13) dening L(s, ) are absolutely
convergent for > 1 and uniformly convergent in any compact subset. To
see this we proceed as we did for Dedekind zeta-functions via the splitting of
primes (13.1). Indeed, we can rewrite (13.13) as an ordinary Euler product
 (p)
1   r 
(pj )
1
L(s, ) = 1 = 1 sfj .
p
N(p)s p j=1
p
pj |(p)

Thus, L(s, ) has a representation as a polynomial Euler product. Hence we

may also rewrite this as an ordinary Dirichlet series:


a(n)
L(s, ) = ,
n=1
ns

where


r
a(n) = (pj )kj .
p|n k1 ,...,kr 0 j=1
k1 f1 ++kr fr =(n;p)

Since the degree of the local Euler factors is bounded by the degree of the eld
extension K/Q, it follows (as in the case of the Dedekind zeta-function) that
the Dirichlet series and so the Euler product converge for > 1 absolutely
and that the Ramanujan hypothesis holds.
Hecke [125] proved that L(s, ) extends to an entire function and satises
a functional equation of Riemann-type provided is primitive. Let dK denote
the discriminant of K. We dene
r1
+r2  1/2
| dK |N(f)
() = 2ik /2
, A(f) = 2r2 ,
n
k=r1 +1

and   r1  

r1
s + aj ij
+r2
|aj | ij
(s, ) = s+ .
j=1
2 j=r1 +1
2
260 13 L-Functions of Number Fields

Then
(1 s, ) = ()(s, ),
where () is a complex number with |()| = 1, depending only on , and

(s, ) := ()A(f)s (s, )L(s, ).

In view of all the properties mentioned, it follows that Hecke L-functions

L(s, ) to primitive grssencharacters are elements of the Selberg class of
degree n = [K : Q]. The Hecke L-function to the trivial character is, as
already mentioned, identical with the Dedekind zeta-function and so it is an
element of the Selberg class too.
We sketch some arithmetic consequences of the analytic properties of Hecke
L-functions. First of all, we note that L(s, ) does not vanish on the edge of
the critical strip; the proof follows the proof for Dirichlet L-functions (see
Heilbronn [129]). We have already mentioned that L(s, ) is entire and so it
is regular at s = 1 unless is trivial. If is not trivial, then

 
x
(p) = o
log x
N(p)x

as x ; for trivial we obtain the prime ideal theorem (13.4).

13.4 Universality for Hecke L-Functions

Universality for Hecke L-functions was rst proved by Mishou. In [249], he
obtained universality for Hecke L-functions to ray class characters. The proof
proceeds along the lines of Voronins original proof; besides, arithmetical prop-
erties of those rational primes which split completely play a central role.
In Mishou [250], Bagchis approach and, in particular, the positive density
method are applied to tackle the general case of grssencharacters.

Theorem 13.2. Let K be a nite extension of Q of degree n and be a

grssencharacter modulo f. Let K be a compact subset of the strip 1 n1 < < 1
with connected complement and g(s) be a non-vanishing and continuous on K
which is analytic in the interior of K. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max |L(s + i, ) g(s)| <  > 0.
T T sK

The main part of the proof relies on a discussion of the various ways how
rational primes can split. This is used to show that there exist innitely many
rational primes p for which the character sum


a(p) = (p),
p|(p)
 13.5 Artins Reciprocity Law 261

where the sum is taken over all prime ideals of degree one which divide p, is
bounded away from zero. In the case of ray class characters one can deduce
this quite easily by applying class eld theory; in fact, the set of all rational
primes which split completely in the class eld for the ideal class group has
innitely many elements. The general case is more delicate (for the details see
[249, 250]).
For the special type of grssencharacters m of the form (13.12) for Q(i),
Koyama and Mishou [174] succeeded to prove a mixed version of universality
for Hecke L-functions L(s + i, m ) in the - and in the -aspect:

Theorem 13.3. Let K be a compact set in the open right half of the critical
strip. For any function g(s) which is non-vanishing and continuous on K and
which is analytic in the interior of K, and for any  > 0,
 
1
lim inf 2 meas
(, m) T : max |L(s + i, m ) g(s)| <  > 0,
T T sK

where T := [0, T ] {0, . . . , T } and the measure meas

is dened by

T
meas
(X) = meas { [0, T ] : (, m) X}
m=0

for X T .

As Koyama and Mishou mention, it is unfortunate that the range of and

m has to be the same. They remark that a conjecture of Duke [71] on mean-
values for certain Dirichlet series over Z(i) would imply the universality in the
character aspect alone.

13.5 Artins Reciprocity Law

Now we want to study a further class of L-functions which play a central

role in algebraic number theory ever since Artin introduced them in order to
nd higher reciprocity laws. However, rst of all, we briey motivate their
denition.
In algebraic number theory, a fundamental problem is to describe how a
rational prime factors into primes in the ring of integers OK of a given number
eld K. Now assume that K is a Galois extension over Q with Galois group
G := Gal(K/Q) (i.e., Q is xed with respect to automorphisms from G). Then
K is the splitting eld of some monic polynomial with rational coecients, and
G is the group of eld automorphisms of K xing Q pointwise. The splitting
type of a prime p in OK is completely determined by the size of the subgroup
of G which xes any pj above p. For simplicity, assume that the rational
prime p is unramied in K, i.e., the primes pj in (13.1) are all distinct, then
262 13 L-Functions of Number Fields

these subgroups are all cyclic. Information about the factorization of such p
is encoded in the so-called Frobenius automorphism pj of G, the canonical
generator of the subgroup of G which maps any pj into itself. The Frobenius is
determined only up to conjugacy in G; nevertheless, the resulting conjugacy
class, which we denote by p , completely determines the splitting type of
(13.1).
If, for example, K = Q(i) = {a + bi : a, b Q}, then OK = Z[i], the set
of Gaussian integers m + ni with m, n Z. In this case, p is the identity
if 1 is a quadratic residue mod p, and the complex conjugation otherwise.
Hence, we may identify G with the subgroup {1} of C := C \ {0} via the
homomorphism  : G {1}:
 
1
(p ) = .
p

By a part of the quadratic reciprocity law, the Legendre symbol can be

expressed in terms of a congruence condition on p which states for unram-
ied (odd) primes p
  
1 p1 +1 if p 1 mod 4,
= (1) 4 =
p 1 otherwise.

Thus, the factorization of p in Z[i] depends only on its residue mod 4.

One goal of class eld theory is to nd a similar description of p for
arbitrary Galois extensions K. In general, one cannot expect that there exists
a modulus q such that p is the identity if and only if p lies in some arithmetic
progression mod q. However, if K is abelian, i.e., G = Gal(K/Q) is abelian,
and  : G C is a homomorphism, then it is known that there exists a
Dirichlet character mod q such that

: (Z/qZ) C with (p ) = (p) (13.14)

for all primes p, unramied in K. This is the theorem of KroneckerWeber.

It follows that the splitting properties of p in K depend only on its residue
modulo some xed number q depending on K. In particular, this implies the
general quadratic reciprocity law of Gauss. As a matter of fact, the factoriza-
tion of Dedekind zeta-functions
%
K (s) = (s)L(s, ), where K = Q( (1)q),

for all quadratic elds K is equivalent with quadratic reciprocity.

What can be said for non-abelian Galois extensions? Recognizing the util-
ity of studying groups in terms of their matrix representations, Artin focused
attention on homomorphisms

 : G = Gal(K/Q) GLm (C),

 13.6 Artin L-Functions 263

i.e., m-dimensional representations of the Galois group G; note that one-

dimensional representations are simply characters. Artin transferred the prob-
lem of analysing conjugacy classes in G to the analogous problem in GLm (C),
where the corresponding classes are completely determined by their charac-
teristic polynomials  
(p )
det 1 ,
ps
where 1 denotes the unitary matrix. Introducing the so-called Artin L-function
  1
(p )
L(s, ) = det 1
p
ps

(we give a precise denition in Sect. 13.6), Artin was able to reduce the prob-
lem to one involving these analytic objects: is it possible to dene L(s, ) in
terms of the arithmetic of Q alone? It was in this context that Artin proved
his reciprocity law. Indeed, for abelian elds K and one-dimensional , Artin
showed that L(s, ) is identical to a Dirichlet L-function L(s, ) with an
appropriate character mod q. Since an identity between two Euler prod-
ucts implies an identity between the local Euler factors, this implies Artins
reciprocity law (13.14). Actually, Artin proved a stronger result for abelian
extensions L/K (see Theorems 13.4 and (13.16)).

13.6 Artin L-Functions

Let L/K be a Galois extension of number elds with Galois group G. Further,
let  : G GLm (V ) be a representation (group homomorphism) of G on a
nite dimensional complex vector space V . In order to give the denition of
the Artin L-function attached to these data, we recall some facts on prime
ideals in number elds and their ramication in Galois extensions.
For each prime p of K, and a prime P of L with P|p, we dene the
decomposition group by

DP = { G : P = P} = Gal(LP /Kp ),

where LP and Kp are the completions of L at P and K at p, respectively.

Denote by kP /kp the residue eld extension. By Hensels lemma, we have a
surjective map from DP to Gal(kP /kp ); its kernel IP is the inertia group at
P, dened by

IP = { G : () mod P for all OL }.

We thus have an exact sequence

1 IP DP Gal(kP /kp ) 1.
264 13 L-Functions of Number Fields

Hence, there is an isomorphism

DP /IP ! Gal(kP /kp ).

Now kP /kp is a Galois extension of nite elds, and hence the group
Gal(kP /kp ) is cyclic, generated by the map  N(p) , where N(p), the
absolute norm of p, is the cardinality of kp . We can choose an element P DP
whose image in Gal(kP /kp ) is this generator; this P is called Frobenius ele-
ment at P, i.e.,
P () N(p) mod P
for all OL . Note that the Frobenius element is only dened mod IP . The
Frobenius describes how prime ideals split when lifted from a smaller ring
of integers to a larger ring of integers. For unramied p (and in particular,
these are all but nitely many p), the Frobenius is well dened since IP =
{1}. The action of the Galois group on the set of primes in L above p is
transitive, and thus for any pair of primes P1 and P2 lying above p, there
exists an automorphism in G which simultaneously conjugates DP1 into DP2 ,
IP1 into IP2 , and P1 into P2 . This implies an identity for the characteristic
polynomials of Pj on the subspace VPj of V on which IPj acts trivially:
   
(P1 )  (P2 ) 
det 1 VP1 = det 1 V P 2 .
N(p)s N(p)s

Thus, these characteristic polynomials are independent of the choice of P .

Denote by p the conjugacy class of Frobenius elements at primes P above
p; in case of unramied p the inertia group is trivial, and p is called Artin
symbol.
Following Artin [4], we dene the Artin L-function attached to  by
 1
 (p ) 
L(s, , L/K) = det 1 V P , (13.15)
p
N(p)s

where p runs through the prime ideals of the ring of integers in K; this Euler
product converges for > 1 by the same reasoning as for previous L-functions
of number elds.
We shall illustrate this by an example (see also Heilbronn [129] and Stark
[337]). Assume that L/Q is normal with Galois group equal to the symmetric
group S3 on three letters:

G := {1, (), (), (), (), ()}

say. For instance, one may consider the cubic eld K = Q(21/3 ) and its normal
closure L = Q(, e2i/3 ) = Q(, , ), where

= 21/3 , = e2i/3 21/3 , = e4i/3 21/3 .

 13.6 Artin L-Functions 265

Since automorphisms of L are determined by their action on , and , we

nd G = Gal(L/Q) for the Galois group of L. The splitting of primes from
Q to K, and likewise from K to L, is ruled by the Frobenius automorphisms.
Suppose that P is an unramied prime of L which lies above p of K which in
turn lies above the rational prime p of Q. Then the Frobenius automorphism
of P relative to Q is given by one of the following conjugacy classes:
P = 1. Since the Frobenius has order one, by (13.1), there are six primes
in L above p. Obviously, P Gal(L/K) = {1, ()}. In this case p splits
in K into three dierent primes pj (1 j 3) each of which splits into
two prime ideals Pk (1 k 6) of L.
P is in the conjugacy class {(), (), ()} of elements of order two.
We may choose P such that P = () Gal(L/K). Then f = 2 in (13.1)
and so there are three second degree primes Pk (1 k 3) above p; we
may assume that P = P1 . We observe that the Frobenius automorphism
of P relative to K is equal to P . Hence, we nd N(P) = N(p)2 and
N(p) = p for some prime p = p1 of K. For the other two primes P2 and P3
the Frobenius P is equal to () and (), respectively. In these cases
we nd P 2
= 1 Gal(L/K) and N(P) = N(p) and N(p) = p2 for some
prime p = p2 of K. Thus, the primes P2 and P3 have relative degree one
over a single prime p2 of K (which is of degree two).
P is in the conjugacy class {(), ()} of elements of order three. In
this case we have f = 3 in (13.1) and there are two third degree primes
P1 and P2 of L above p, for one of them P = () and for the other
P = (). In both cases neither P nor P 2
lie in Gal(L/K) = {1, ()}
and so both P1 and P2 lie above a single prime p of K (which must be of
degree three).
Before we continue we remark that the splitting of primes can be computed
by use of the following statement (analogous to (13.3)): suppose that F (X) is
the minimal polynomial of K over Q and that it splits factors mod p into
irreducible pieces as

F (X) F1 (X)e1 . . . Fr (X)er mod p.

If the power of p in the polynomial discriminant of F (X) is the same as the

power of p in the relative discriminant DL/K of L/K, then p splits in L as

p = Pe11 . . . Perr ,

where Pj = (p, Fj ()) is of relative degree deg Fj . This together with

Eisensteins irreducibility criterion gives the basic tools to do arithmetic com-
putations in number elds.
We may represent the Galois group G by matrices as follows. For g G
we write
266 13 L-Functions of Number Fields


g = M(g) ,

where M(g) is the permutation matrix corresponding to g. Thus, in our
example, we can represent the six elements of G by

100 010 001
1  0 1 0 , ()  0 0 1 , ()  1 0 0 ,
001 100 010

010 001 100
()  1 0 0 , ()  0 1 0 , ()  0 0 1 .
001 100 010

The map  : g  M(g) denes a homomorphism: M(gh) = M(g)M(h); it is

an example of a three-dimensional permutation representation of the group
G. The conjugacy classes of the symmetric group on , , are precisely the
conjugacy classes of Frobenius automorphisms arising from prime numbers
which split in the indicated form; that are

C1 : {1}, C2 : {(), ()}, C3 : {(), (), ()}.

For each of them we observe that the associated Euler factors are of the form
as predicted by (13.15). To see this we have a look on every individual Euler
factor. Since the eld extension K/Q has degree 3, there are the following
possibilities to consider.
The prime p splits completely into three dierent prime divisors; e.g.,
(31) = p1 p2 p3 with

p1 = (31, 4), p2 = (31, 7), p3 = (31, 20).

In this case the local Euler factor at p is of the form

1
 3 100
1 1
1 s = det 1 0 1 0 s .
p p
001
Obviously, the appearing matrix has the eigenvalue +1 with multiplicity
three. This Euler factor corresponds to the class C1 .
The prime p can be factored into a product of two factors, one of rst
degree and one of second degree; for example, (5) = p1 p2 with

p1 = (5, 3), p2 = (5, 2 + 3 + 9).

Here we have
 1  1  1
1 1 1
1 s 1 2s = det 1 M s ,
p p p
 13.6 Artin L-Functions 267

for any of the matrices

01 0 001 100
M = 1 0 0, 0 1 0, and 0 0 1,
00 1 100 010
corresponding to C3 . The eigenvalues of the (similar) matrices are 1 and
+1 with multiplicities one and two, respectively.
The prime p is a prime ideal of third degree; e.g., (7) = p. In this case we
have  1  1
1 1
1 3s = det 1 M s
p p
for the matrices associated with C2 :

010 001
M = 0 0 1 and 1 0 0 .
100 010
Here the eigenvalues of the (similar) matrices are the third roots of unity.
Now we introduce a more convenient notation of Artin L-functions. To
any representation  of G, we attach a character of G by setting
(g) = trace((g))
for g G. The degree of a character is dened by deg = (1). If h is another
element of G, then
(h1 gh) = (h)1 (g)(h),
so that (h1 gh) and (g) are similar matrices and thus have the same trace.
This shows that characters of G are constant on the conjugacy classes. Two
representations are said to be equivalent if they have the same character. If
1 and 2 are representations of G with characters 1 and 2 , then
 
1 (g) 0
(g) =
0 2 (g)
also denes a representation of G with character 1 + 2 , and in this case
 is said to be reducible; any representation which is not reducible is called
irreducible. We shall use the same attributes for the associated character.
It turns out that any conjugacy class of G corresponds to an irreducible
representation and one can show that there are not more; of course, distinct
irreducible representations are non-equivalent (these observations are analo-
gous to the case of Dirichlet characters and the group of residue classes of Z).
In our example we nd for the three conjugacy classes of G:
C1 C2 C3
1 +1 +1 +1
2 +1 +1 1
3 +2 1 0
268 13 L-Functions of Number Fields

Hence, for G = S3 , there are three irreducible characters (in some literature
simple characters).
It is easily seen that our characters satisfy the orthogonality relations, that
are 
1
1/C if C = D,
(C)(D) =
G 0 otherwise,
G

where C and D are conjugacy classes. Since the Euler factors in (13.15) depend
only on the conjugacy class p , in the sequel we will talk sometimes in terms
of characters and denote the Artin L-function (13.15) by L(s, , L/K) (and
sometimes we shall even write L(s, ) for short).
For further illustration we continue with our example. We can construct
more characters from the irreducible characters listed above, for example,
a third degree character related to the permutation representation ().
Taking the character relations into account we nd = 1 + 3 . For the
related Artin L-functions we note that

L(s, , L/K) = L(s, 1 + 3 , L/K) = L(s, 1 , L/K)L(s, 3 , L/K).

This additivity property holds in general.

For the eld L = Q(, , ) there are four subelds up to conjugacy:
rstly, the eld Q itself, xed by all of G, secondly, Q( 3) xed by G2 :=
{1, (), ()} (corresponding to the conjugacy class C2 ), thirdly, K =
Q(21/3 ) xed by G3 := {1, ()} (corresponding to the conjugacy class C3 ),
and nally L xed just by {1}.

L {1}
tt 444 t 99
ttt 44 () tt t 99
tt tt 99
t tt 44 t tt 99
t 44 t
44 tt 9(),()
99
K = Q(21/3 ) G3 77 99
88 44 77
88 44 77 99
88 44 77 99
88 77 9
8
3 88 Q( 3) 77 G2
88 w 77 rrr
88 ww 77 r r
88 w
ww 2
w 77 rrr
ww rrr
Q Gal(L/Q) = S3

We obtain the following factorizations of the associated Dedekind zeta-

functions into products of Artin L-functions to L/Q:

(s) = Q (s) = L(s, 1 ),

Q(3) (s) = L(s, 1 ) L(s, 2 ),
Q(21/3 ) (s) = L(s, 1 ) L(s, 3 ),
L (s) = L(s, 1 ) L(s, 2 ) L(s, 3 ).
 13.7 The Artin Conjecture 269

We observe that any of the Dedekind zeta-functions on the left-hand side

is divisible by the Riemann zeta-function. It follows from these factoriza-
tions and the analytic behaviour of Dedekind zeta-functions that each of the
involved Artin L-functions with = 0 possesses a meromorphic continua-
tion to the whole complex plane; the only possible poles can occur at zeros
of other Artin L-functions. Furthermore, one can deduce functional equations
of the Riemann-type. This is a remarkable way to deduce analytic properties
for L-functions!

13.7 The Artin Conjecture

One of the most fundamental conjectures in algebraic number theory is
Artins Conjecture. Let L/K be a nite Galois extension with Galois group
G. For any irreducible character = 1 of G the Artin L-function L(s, , L/K)
extends to an entire function.
We discuss briey one of its important consequences. Dedekinds conjecture
claims that the quotient L (s)/K (s) is entire provided L/K is an extension of
number elds, not necessarily Galois. Here one may recall the factorizations of
the Dedekind zeta-functions in our example from Sect.13.6. If L/K is a Galois
extension, then the so-called ArtinTakagi factorization gives a factorization
of the Dedekind zeta-function of a number eld relative to a subeld (see
Heilbronns survey article [129]); more precisely,
L(s, 1, L/K) = K (s) and L(s, RG , L/K) = L (s),

where RG is the regular character of G (the character dened by (1)),
and 
L (s) = L(s, , L/K)(1) ,
G

where G denotes the set of irreducible characters of G. In the case of Galois

extensions L/K, a theorem of Aramata [2] yields the truth of Dedekinds
conjecture; its proof relies mainly on the ArtinTakagi factorization (an easier
proof was given by Brauer [42] by his induction theorem; see also Heilbronn
[129] or Murty and Murty [270], Sect. 2.3). In the general case, if L/K is a
nite (not necessarily Galois) extension, then Dedekinds conjecture follows
from Artins conjecture by studying the normal closure of L/K.
As indicated in a previous section, Artin [5] proved his conjecture if
is one-dimensional and L/K is abelian. In this case, the related Artin L-
function coincides with a Hecke L-function attached to some ray class group.
The Frobenius p induces a map from the fractional ideals of K coprime with
f to the Galois group:
(f) : I(f) Gal(L/K).
By class eld theory it is surjective and if the nite primes dividing f are
suciently large, then its kernel is a congruence subgroup for f. This is Artins
reciprocity law:
270 13 L-Functions of Number Fields

Theorem 13.4. Let L/K be abelian and let be a one-dimensional character

of G = Gal(L/K). Then there exists a modulus f = f(L/K) divisible by all
ramied primes (nite and innite) such that

I(f)/ker((f)) ! G.

Using the isomorphism (f), we dene a character of the class group by

setting (p) = (p ) for p coprime with f. Then

L(s, , L/K) = L(s, ). (13.16)

Artin proved this theorem by means of class eld theory and, in particular,
using Chebotarevs density theorem. We shall briey explain the latter result.
Let L/K be a nite Galois extension with Galois group G and let C be a
subset of G, closed under conjugation. Further, denote by C (x) the number
of prime ideals p of K, unramied in L, for which p C and which have
norm N(p) x in K. Then, Chebotarevs density theorem [53] states
C
C (x) (x). (13.17)
G
This rather deep theorem can be seen as a higher analogue of the prime
number theorem in arithmetic progressions. A modern proof can be found,
for example, in Narkiewicz [277]. The Chebotarev density theorem can be
used to determine the Galois group of a given irreducible polynomial P (X)
of degree n by counting the number of unramied primes up to a certain
bound for which P factors in a certain way and comparing the results with
the fractions of elements of each of the transitive subgroups of the symmetric
group Sn with the same cyclic structure.
We illustrate this by returning to our example from Sect.13.6. By means
of class eld theory one can show that K = Q(21/3 ) is the class eld for a
ray class group G(f) over Gal(L/K) which is cyclic of order three. Besides
the trivial character, there are two cubic characters 2 and 3 each of which
is the square of the other one. Their common conductor is f = (6) and, by
Chebotarevs density theorem it follows that:
1/6 of the rational primes split completely in K (and they have a repre-
sentation as p = x2 + 27y 2 with x, y Z),
1/2 of the rational primes split as the product of rst and second degree
prime factors (namely p 2 mod 3), and
1/3 of the rational primes generate third degree primes of K.
Brauer [42] obtained a signicant extension of Artins result. Let H be a
subgroup of G. Then any character on H induces a character IndGH () on G
and
L(s, , L/LH ) = L(s, IndGH (), L/K), (13.18)
where LH is the corresponding xed eld. Brauer [42] proved that any char-
acter of a nite group G can be written as a linear combination
 13.7 The Artin Conjecture 271


= nj IndGHj (j ),
j

where the Hj are nilpotent subgroups of G, the j are one-dimensional, and

the nj s are integers. Hence, it follows that

L(s, , L/K) = L(s, IndGHj (j ), L/K)nj .
j

However, by (13.18),

L(s, , L/K) = L(s, j , L/LHj )nj .
j

Now applying Artins reciprocity law to the one-dimensional characters j ,

we obtain 
L(s, , L/K) = L(s, j )nj (13.19)
j

for certain ray class characters j related to j . Using the analytic proper-
ties of Hecke L-functions, Brauer [42] deduced a functional equation of the
Riemann-type for Artin L-functions which gives a meromorphic continuation
throughout the complex plane (see also Neukirch [279]), Sect. VII.12).
Brauers theorem implies the Artin conjecture if G is nilpotent or super-
solvable since then every character of G is monomial (see Serre [325]). Murty
[267] observed that

Theorem 13.5. Selbergs Conjecture B implies Artins conjecture.

Murty and Perelli [271] replaced Selbergs conjecture by the pair correlation
conjecture (as already mentioned above).
The proof uses some easy properties of Artin L-functions which we did not
prove or even did not mention above. The reader may have a look into the
literature, e.g., Heilbronn [129], and may consult the examples from Sect.13.6.

Proof. Let L be the normal closure of L over Q. Then, L/K and L/Q are
Galois. Thus, can be considered as a character of Gal(L/K), and by the
properties of Artin L-functions it turns out that

L(s, , L/K) = L(s, , L/K).

By the invariance induction of from Gal(L/K) to Gal(L/Q), it follows that


L(s, , L/K) = L(s, , L/Q)m() ,

where the product is taken over all irreducible characters of Gal(L/Q) and
m() are non-negative integers. To prove Artins conjecture, it suces to
272 13 L-Functions of Number Fields

show that all appearing L(s, , L/Q) are entire. By Brauers induction theo-
rem (13.18) and Artins reciprocity law, Theorem 13.4, respectively, formula
(13.19), we have
L(s, 1 )
L(s, , L/Q) = ,
L(s, 2 )
where 1 , 2 are characters of Gal(L/Q) and L(s, 1 ), L(s, 2 ) are products
of Hecke L-functions (13.13). Since Hecke L-functions belong to the Selberg
class S, and S is multiplicatively closed, the functions L(s, 1 ) and L(s, 2 )
belong to S too. Now, by Theorem 6.2, there exist primitive functions Lj S
such that


f
L(s, , L/Q) = Lj (s)ej , (13.20)
j=1

where ej Z. By comparing the pth coecient in the Dirichlet series expan-

sions of both sides, we get

f
(p) = ej aLj (p).
j=1

Thus,
 2

|(p)|2
1 
f 

=  e a (p) . (13.21)
p 
p  j=1
j Lj 
px px 

Selbergs Conjecture B yields the asymptotic formula

 2

1 
f 

f
 ej aLj (p) = e2j log log x + O(1). (13.22)
p  j=1  j=1
px

Next, we decompose the sum on the left-hand side of (13.21) according to

the conjugacy classes C of G := Gal(L/Q) to which the Frobenius element p
belongs. If gC denotes any element of C, this leads to

|(p)|2

1
= |(gC )|2 .
p p
px C px
p C

By partial summation, we deduce from Chebotarevs density theorem (13.17)

1 C
= log log x + O(1).
px
p G
p C
 13.7 The Artin Conjecture 273

This gives

|(p)|2
C
= |(gC )|2 log log x + O(1).
p G
px C

Since is irreducible, we have

C 1

|(gC )|2 = C = 1,
G G
C C

which implies via (13.22) and (13.21)

f
e2j = 1.
j=1

Thus, f = 1 and e1 = 1. The case e1 = 1 implies

1
L(s, , L/Q) = ,
L1 (s)

which is impossible since L(s, , L/Q) has trivial zeros (their existence fol-
lows from the functional equation). Hence, e1 = +1, and we conclude that
L(s, , L/Q) = L1 (s) is entire. 


The proof shows that if is an irreducible non-trivial character of Gal(K/Q),

then the Artin L-function L(s, , K/Q) is a primitive element in the Selberg
class S if Selbergs Conjecture B is true; the case of Artin L-functions to
reducible characters follows from the fact that S is multiplicatively closed.
Furthermore, the splitting of primes implies axiom (iv) on the polynomial
Euler product from the denition of S. Decomposing as in Murtys proof, we
nd
1
1


|(p)|2 = |(gC )|2 1.
(x) (x)
px C px
p C

By Chebotarevs density theorem (13.17), this is asymptotically equal to

1

C = 1,
G
C

which implies axiom (v) on the prime mean-square. So, L(s, , K/Q) S pro-
vided that Artins conjecture is true; note that in the case of one-dimensional
characters this is unconditional since these L-functions are known to be entire.
Thus, on behalf of Theorem 5.14 we obtain universality for Artin L-functions
L(s, , K/Q) to Galois extensions K/Q subject to Selbergs Conjecture B.
However, following the lines of Voronins proof, Bauer [16] proved universality
for any Artin L-function without assumption of any unproved hypothesis. In
Sect.13.8 we shall prove and extend this result by use of Bagchis method.
274 13 L-Functions of Number Fields

13.8 Joint Universality for Artin L-Functions

We shall use joint universality for Artin L-functions to one-dimensional rep-

resentations in combination with Brauers induction theorem (13.19) in order
to deduce joint universality for arbitrary Artin L-functions of K/Q.

Theorem 13.6. Let K be a nite Galois extension of Q and let 1 , . . . ,  be

C-linearly independent characters of Gal(K/Q). Let K be a compact subset of
max{ 12 , 1 [K : Q]1 } < < 1 with connected complement. Further, for each
1 k  let gk (s) be a continuous non-vanishing function on K which is
analytic in the interior of K. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max max |L(s + i, k , K/Q) gk (s)| < > 0.
T T 1k sK

Proof. Without loss of generality we may assume that the characters

1 , . . . ,  form a basis of the set of class functions on Gal(K/Q). In view
of Brauers induction theorem (13.19) any of the Artin L-functions has a rep-
resentation 
L(s, k , K/Q) = L(s, jk )njk , (13.23)
j

where the L(s, jk ) are Hecke L-functions to number elds Kjk contained in
K, the jk are ray class characters of Kjk , and the numbers njk are non-zero
integers.
It suces to prove that the family of Artin L-functions to all pairwise non-
equivalent one-dimensional characters k to subgroups of Gal(K/Q) is jointly
universal. Then the assertion of the theorem follows from (13.23) in just the
same way as in the nal step of the proof of Theorem 13.1.
Now let {L(s, k )} be such a family of distinct Artin L-functions to pair-
wise non-equivalent one-dimensional characters k to subgroups of Gal(K/Q).
We want to prove their joint universality. In view of Artins reciprocity law,
Theorem 13.4, each L(s, k ) is a Hecke L-function to a ray class group char-
acter for some subeld of K; in particular, they belong to S (as we have seen
in Sect. 13.7). According to (13.1),


rp
e


rp
p= pj j with N(pj ) = pfj , ej fj = [K : Q],
j=1 j=1

we write
  1   1
k (p ) k (p)
L(s, k ) = det 1 = 1
p
ps p
ps
rp  1

 k (pj ) ak (n)
= 1 fj s =
p j=1
p n=1
ns
 13.8 Joint Universality for Artin L-Functions 275

with


rp
ak (p) = k (pj ).
j=1
fj =1

Clearly, m := max rp [K : Q]; this denes the local roots jk . In order to

prove joint universality for {L(s, k )} we shall apply Theorem 12.5, i.e., we
have to show that the set of all convergent series


g p (s, b(p)) with b(p)
p

is dense in H(D) , where the g p are dened by (12.11) and (12.12), that are

m  
jk (p)b(p)
gpk (s, b(p)) = log 1 ,
j=1
ps
g p (s, b(p)) = (gp1 (s, b(p)), . . . , gp (s, b(p))),

and  is the cardinality of {L(s, k )}. Most of the reasoning is similar as in

the proof of Theorem 12.8; however, the verication of the rst assumption
of Theorem 12.9 is slightly dierent. We sketch the main idea.
Let 1 , . . . ,  be complex Borel measures on (C, B(C)) with compact sup-
port contained in D such that
 


  

 b(p)gpk (s) dk (s) < ;
 C 
p k=1

here, as in Sect.13.7,

gpk (s) if p > N,
gpk (s) =
0 if p N,
g p (s) = (gp1 (s), . . . , gp (s)),

and the b(p) form a sequence such that p b(p)gpk (s) converges in H(D).
Following the proof of Theorem 12.8, we nd
ak (p)
gpk (s) = + rpk (s)
ps
with a suciently small remainder rpk (s). It is easily seen that we can replace
(12.25) by  


  

 ak (p)ps dk (s) < . (13.24)
p
 C 
k=1

Denote by C1 , . . . , CJ the dierent conjugacy classes of G = Gal(K/Q). Then J

is the dimension of the vector space of class functions on G, and so J is greater
than or equal to the number of C-linear independent characters. Dene
276 13 L-Functions of Number Fields

Pj = {p P : p Cj } for 1 j J;

recall that the Frobenius p completely determines the splitting of the primes.
Clearly, P = 1jJ Pj , and

ak (p) = trace(k (p )) = trace(k (Cj )) for p Pj .

Hence, we can replace (13.24) by

|j (log p)| < ,
pPj

where



j (z) := exp(sz) dj (s) and j (s) := trace(k (Cj ))k (s).
C k=1

Next we need Chebotarevs prime number theorem (13.17) with remainder

term due to Artin [4]

Cj
(x; Cj ) := 1= li(x) + O(x exp(c(log x)1/2 )
px
G
P
p j

(see also [129]). This implies

1  x
1 du
= (x; Cj ) + (u; Cj ) 2
px
p x 2 u
P
p j

Cj
= log log x + cj + O((log x)2 ),
G
where the cj certain constants depending only on Cj . Now we can proceed
as in the proof of Theorem 12.8 and obtain that j (z) vanishes identically,
respectively, 
sr dj (s) 0 for r = 0, 1, 2, . . .
C
That means


 
trace(k (Cj )) sr k (s) 0
k=1 C

for r = 0, 1, 2, . . . and all 1 j J. Hence

 
k (gj ) sr k (s) 0
k=1 C
 13.8 Joint Universality for Artin L-Functions 277

for any gj in the conjugacy class Cj . Since the characters k are linear inde-
pendent over C, it follows that all

(q) sr k (s) 0.
C

Thus, any k (s) vanishes identically and so we are done. Now we continue in
the standard way. This yields the assertion of the theorem. 


We conclude our studies on Artin L-functions by some interesting con-

sequences of Theorem 13.6 due to Bauer [16]. First of all, any single Artin
L-function to an arbitrary normal extension K/L is universal.

Corollary 13.7. Let L/K be a normal extension and an arbitrary character

of Gal(L/K). Further, let K be a compact subset of max{ 12 , 1 d1 } < < 1
with connected complement, where d is the degree of a normal extension of Q
containing L, and let g(s) be a continuous non-vanishing function on K which
is analytic in the interior. Then, for any  > 0,
 
1
lim inf meas [0, T ] : max |L(s + i, , L/K) g(s)| < > 0.
T T sK

The proof follows from Theorem 13.6 by observing that a normal extension
L of Q containing L is also a normal extension of K and that L(s, , L/K) =
L(s, , L/K) by well-known properties of Artin L-functions. Moreover,
Gal(L/K) is a subgroup of Gal(L/K) and so L(s, , L/K) = L(s, , L/Q)
for the induced character . Application of Theorem 13.6 to L(s, , L/Q)
yields the assertion.
As a second application we note that there is no non-trivial continuous
relation between Artin L-functions to irreducible characters of any xed Galois
extension K/Q. This remarkable result follows in a similar way as the theorem
on the functional independence; for a precise formulation and the details we
refer to Bauer [16]. Similarly, it follows that there is no non-trivial continuous
relation between Dedekind zeta-functions to nite normal extensions Kj of Q
satisfying Ki Kj = Q for i = j. The latter condition cannot be removed
as the example (13.11) shows. Bauer also showed that the only non-trivial
relations are monomial algebraic.
Davenport and Heilbronn [65] proved that the zeta-function of an ideal
class of an imaginary quadratic eld has an innity of zeros in the half-plane
of convergence > 1 provided that the class number is greater than one.
Voronin [364] extended this result to the right half of the critical strip and gave
lower bounds for the number of these zeros. Following Voronins reasoning in
addition with some basics from class eld theory, Bauer [16] deduced from the
joint universality of Artin L-functions that any partial zeta-function (s, A)
associated with an arbitrary class A of the ray class of any algebraic number
eld has more than cT many zeros in 12 < < 1, |t| < T for T suciently large,
where c is a positive constant, provided the ray class group has cardinality
278 13 L-Functions of Number Fields

greater than one (in particular for class groups of number elds with class
number greater than one).
Finally, we shall remark that Hecke L-functions and Artin L-functions
form overlapping classes of Euler products, however, neither is contained in
the other. For example, a Hecke L-functions associated with a grssenchar-
acter which is not a ray class group character is certainly not an Artin L-
function. Conversely, if L/K is a non-abelian Galois extension of Q and  is
an irreducible representation of Gal(L/K), then the attached Artin L-function
is a Hecke L-function if and only if  is one-dimensional. Weil [369] gave gen-
eralizations of both Hecke and Artin L-functions which can be unied in a
single theory; the corresponding objects are so-called ArtinWeil L-functions.
Their universality is still open.

13.9 L-Functions to Automorphic Representations

The Langlands program has emerged in the late 1960s of the last century in
a series of far-reaching conjectures tying together seemingly unrelated objects
in number theory, algebraic geometry, and the theory of automorphic forms.
These disciplines are linked by Langlands L-functions associated with auto-
morphic representations, and by the relations between the analytic properties
and the underlying algebraic structures. There are two kinds of L-functions:
motivic L-functions which generalize Artin L-functions and are dened purely
arithmetically, and automorphic L-functions, dened by transcendental data.
In its comprehensive form, an identity between a motivic L-function and an
automorphic L-function is called a reciprocity law. Langlands reciprocity
conjecture claims, roughly, that every L-function, motivic or automorphic,
is equal to a product of L-functions attached to automorphic representations.
For an introduction to the Langlands program we refer to the excellent surveys
of Gelbart [97], Murty [266] and Langlands lecture [178] at the International
Congress in Helsinki. We shall see that the axioms dening S, resp. S are
intimately related to fundamental conjectures concerning general L-functions;
the paper [145] by Iwaniec and Sarnak is a nice survey on these conjectures
and the progress made in the last years. Besides, we give some applications
of our universality Theorem 5.14.
At the heart of Langlandss program is the notion of an automorphic rep-
resentation and its L-function L(s, ). These objects, both dened via group
theory and the theory of harmonic analysis on adle groups, will be briey
explained below, omitting the technical details; for a more detailed description
we refer to the monographs of Bump [46], Gelbart [96], Langlands [177], and
the paper [41] of Borel.
Let K be a number eld (one loses not too much by restricting to Q).
For each absolute value on K, there is a completion K of K which is R,
C, or a p-adic eld, where p is a prime ideal in K. Denote by O the ring
of integers in K . In discussing localglobal problems it is often necessary to
 13.9 L-Functions to Automorphic Representations 279

consider several places simultaneously. At rst sight it seems natural to form

the product of all the K which is a topological ring, but it does not have
satisfactory compactness properties. Since any K is a p-adic integer for
almost all p, we restrict to elements

= ,

where lies in O for all but nitely many places ; such elements are called
adles. The adles form a set-theoretic (restricted) product. This product is
a topological ring, the adle ring AK of K. One can think of K as embedded
in AK via the map  (, , . . .).
For m 1 let GLm (AK ) be the group of m m matrices over AK whose
determinant is a unit in AK . By the product topology of the adle ring,
GLm (AK ) becomes a locally compact group in which GLm (K), embedded di-
agonally, is a discrete subgroup of GLm (AK ). A character of K \ GL1 (AK )
is called grssencharacter, where K := K \ {0}. For a xed grssencharacter
we consider the Hilbert space

L2 := L2 (GLm (K) \ GLm (AK ), )

of measurable functions f on GLm (K) \ GLm (AK ) satisfying the conditions

f (zg) = (z)f (g) for any z Z, g GLm (K) \ GLm (AK );
the integral 
|f (g)|2 dg
ZGLm (K)\GLm (AK )

is bounded.
Elements f L2 generalize the concept of twisted modular forms to discrete
subgroups of the full modular group. In order to introduce a subspace of cusp
forms we have to consider appropriate subgroups. Any parabolic subgroup P
of GLm (R), where R is a commutative ring with identity, has a decomposition,
called the Levi decomposition, of the form P = M N , where N is the unipotent
radical of P ; M is called the Levi component of P . We denote the unipotent
radical of P in the Levi decomposition of a parabolic subgroup P in GLm (R)
by NP (R).
The subspace of cusp forms

L20 := L2 (GLm (K) \ GLm (AK ), )

of L2 is dened by the additional vanishing condition

for all parabolic subgroups P of GLm (AK ) and every g GLm (AK ),

f (ng) dn = 0.
NP (K)\NP (AK )
280 13 L-Functions of Number Fields

The right regular representation R of GLm (K) on L2 is given by

(R(g)f )() = f (g)
for each f L2 and any , g GLm (AK ). An automorphic representation is
a subquotient of the right regular representation of GLm (AK ) on L2 , and a
cuspidal automorphic representation is a subrepresentation of the right regular
representation of GLm (AK ) on L20 .
A representation of GLm (AK ) is called admissible if its restriction to the
maximal subgroup
  
K := Um (C) Om (R) GLm (O )
complex real finite

contains each irreducible representation of K with nite multiplicity; here

Um and Om denote the groups of unitary and orthogonal m m matrices,
respectively.
Now let be an irreducible, admissible, cuspidal automorphic representa-
tion of GLm (K). Then can be factored into a direct product
= ,
where ranges over all (nite and innite) places of K, and each is an
irreducible representation of GLm (K ); see Flath [79]. For all but a nite num-
ber of places the representation is unramied (that means the quotient
obtained by inducing a quasi-character from the Borel subgroup of GLm (K )
to GLm (K ) is unique).
In order to dene the L-function attached to an automorphic representa-
tion we dene the local Euler factors for non-archimedean (nite) unramied
places by
 1
A
L (s, ) = det 1 ,
N(p)s
where A is the semi-simple conjugacy class corresponding to and p is
the prime ideal of K belonging to the place . We do not explain here the
rather technical denition of the Euler factors L (s, ) for ramied places .
However, any Euler factor L (s, ) for a non-archimedean place associated
with the prime ideal p, unramied or not, can be rewritten as
m  1
j (p)
L (s, ) = 1 , (13.25)
j=1
N(p)s

where the numbers j (p) for 1 j m are so-called Satake, resp. Langlands
parameters, determined from the local representations . At the archimedean
(innite) places we put for certain numbers j ()

m
L (s, ) = (s j ())
j=1
 13.9 L-Functions to Automorphic Representations 281

with

2 (s/2) if
s
K ! R,
(s) := (13.26)
(2)s (s) if K ! C,

where, again, the appearing numbers j () for 1 j m are determined

from the local representations . Then the global L-function associated with
is given by 
L(s, ) = L (s, ),
non-archimedean

and the completed L-function is dened by


(s, ) = L(s, ) L (s, ).
archimedean

By the work of Hecke [126], Jacquet and Langlands [147], and Godement and
Jacquet [101] we have

Theorem 13.8. Let K be a number eld and be an irreducible, admissible,

cuspidal automorphic representation of GLm (AK ). Then (s, ) has a mero-
morphic continuation throughout the complex plane and satises the functional
equation
(s, ) =  Ns1/2 (1 s, ),
where is the contragradient representation of , N N is the conductor
of and  is the root number (these quantities are completely determined by
the local representations). (s, ) is entire unless m = 1 and is trivial, in
which case it has a pole at s = 1.

For m = 1 one simply obtains the Riemann zeta-function, Dirichlet L-

functions and Hecke L-functions attached to grssencharacters, whereas for
m = 2 one gets L-functions associated with newforms; see Jacquets paper
[146] for the details. The similarities between these general L-functions and
those of the Selberg class are obvious. On one hand we have the Selberg
class dened by axioms which are known to be the most common pattern of
many L-functions in number theory, on the other hand we have Langlands
construction of general L-functions out of group representations.
Langlands visionary program can be regarded as a continuation of the
famous Artin conjecture. One of his central conjectures claims that all zeta-
functions arising in number theory are special realizations of L-functions to
automorphic representations constructed above.
Langlands Reciprocity Conjecture. Suppose L is a nite Galois exten-
sion of a number eld K with Galois group G, and  : G V is an irreducible
representation of G, where V is an m-dimensional vector space. Then there
exists an automorphic cuspidal representation of GLm (AK ) such that

L(s, , L/K) = L(s, ).

282 13 L-Functions of Number Fields

This means that there are identities between certain L-functions, which are a
priori of dierent type! Since Hecke grssencharacters are automorphic repre-
sentations of GL1 (A), Artins conjecture is a special case of the Langlands reci-
procity conjecture. By Artins work, if m = 1 and L/K is abelian, Langlands
reciprocity law is settled by means of class eld theory. In the case of function
elds, the Langlands conjecture has been proved by Drinfeld [70] in dimension
2, and recently by Laorgue [175] for arbitrary dimension (for which both of
them were awarded with a Fields medal).
Now we consider the local Euler factors of L-functions attached to auto-
morphic representations. Petersson [293] extended Ramanujans conjecture
on the values of the -function to modular forms. Delignes estimate (1.34)
proved the desired bound for newforms but it is expected that an analogue
should hold for all L-functions of arithmetical nature.
RamanujanPetersson Conjecture. Let be a cuspidal automorphic
representation of GLm (AK ) which is unramied at a place . If is non-
archimedean, then

|j (p)| = 1 for 1 j m,

where p is the prime ideal associated with the place . If is archimedean,

then Re j () = 0 for 1 j m.
Note that, for K = Q and Q = R, this conjecture includes Selbergs eigen-
value conjecture [322] on the smallest eigenvalue of the Laplacian as a special
case. The RamanujanPetersson conjecture might look very restrictive on the
rst view, but it is nothing else than the local analogue of the Grand Riemann
hypothesis. We refer to Iwaniec and Sarnak [145] for details and the current
knowledge concerning this conjecture.
We shall speculate a little bit about all these widely believed conjectures
and the axioms dening the Selberg class and the class S, in particular. It is
expected that all functions in the Selberg class are automorphic L-functions. If
L S is primitive and automorphic, then it is also attached to an irreducible
automorphic representation. Conversely, every irreducible automorphic repre-
sentation should give a primitive function in S. This is not known in general,
but it has been proved by Murty [267, 269] for GL1 and GL2 . The axioms
on the analytic continuation and on the functional equation follow immedi-
ately from Theorem 13.8. The polynomial Euler product in the denition of
S ts (by the splitting of primes in K) perfectly to the Euler product of Lang-
lands L-functions attached to automorphic representations (13.25) and the
RamanujanPetersson conjecture. The axiom on the mean-square was already
discussed in the context of the Selberg conjectures. Finally, let us note that
the Euler factor at the innite places (13.26) is of the form, predicted by
the strong -conjecture (from Sect. 6.1). Of course, all these axioms and the
hypotheses too, are guessed from known examples of L-functions in number
theory, and so they have to share certain patterns. Anyway, we are led to see
 13.9 L-Functions to Automorphic Representations 283

a close connection between Langlands general L-functions and the elements

of the Selberg class.
Assuming Selbergs Conjecture B, Murty [267] proved that
if is an irreducible cuspidal automorphic representation of GLm (AQ )
which satises the RamanujanPetersson conjecture, then L(s, ) is a
primitive function in S,
if K is a Galois extension of Q with solvable Galois group G, and if is
an irreducible character of G of degree m, then there exists an irreducible
cuspidal automorphic representation of GLm (AQ ) such that

L(s, ) = L(s, ).

The rst assertion identies certain L-functions to automorphic represen-

tations as primitive functions in the Selberg class subject to the truth
of Selbergs Conjecture B and the RamanujanPetersson conjecture; if we
assume additionally axiom (v) on the mean-square, L(s, ) is an element of
S too.
The second assertion is Langlands reciprocity conjecture if K/Q is solv-
able. The constraint of solvability arises from a theorem of Arthur and Clozel
[3], who showed that the maps of base change and automorphic inductions for
automorphic representations exist if the extension is cyclic of prime degree.
We do not explain here what that means, but these maps are conjectured to
exist in general, which would yield Langlands reciprocity conjecture in its full
generality. However, Murtys proof shows that if the Dedekind zeta-function
of K is the L-function of an automorphic representation over Q, then Selbergs
Conjecture B implies Langlands reciprocity conjecture.
Concerning universality, Theorem 5.14, in addition to the work of Liu,
Wang, and Ye [225], implies that if n is a positive integer, a cuspidal auto-
morphic representation of GLn (Q) satisfying the RamanujanPetersson con-
jecture (including the innite place), then the GodementJacquet L-function
L(s, ) is universal.
A
Appendix: A Short History of Universality

Thus it appears that universality is a generic phenomenon in analysis.

K.-G. Grosse-Erdmann

We conclude with a brief historical overview on the phenomenon of universal-

ity. We refer to Grosse-Erdmann [109] for a detailed and interesting survey on
more or less all known types of universalities and an approach to unify them
all.

The rst universal object appearing in the mathematical literature was dis-
covered by Fekete
 in 1914/15 (see [287]); he proved that there exists a real
power series n=1 an xn with the property that for any continuous function
f (x) on the interval [1, 1] satisfying f (0) = 0 there is a sequence of positive
integers mk such that


mk
an xn f (x) uniformly on [1, 1].
k
n=1

The proof relies essentially on Weierstrass approximation theorem which

states that every continuous function on a compact interval is the limit of
a uniformly convergent sequence of polynomials (see [367]).
We illustrate Feketes theorem and its proof by a p-adic version. The p-adic
numbers were introduced by Hensel [131] in 1897. Given a prime number p,
we dene the p-adic absolute value | . |p on Z by
 (;p)
p if = 0,
||p =
0 otherwise;

here (; p) is the exponent of p in the prime factorization of Z. It is an

easy task to extend | |p from Z to Q. The p-adic absolute values satises the
strong triangle inequality: for any , Q,

| + |p max{||p , ||p }
286 Appendix

in contrast to the standard absolute value. The p-adic absolute value is non-
archimedean (i.e., sup{|n|p : n N} is bounded). The integers that are
p-adically close to zero are precisely the ones that are highly divisible by p.
The set of p-adic numbers is the completion of Q with respect to the p-adic
absolute value, and is denoted by Qp . The p-adic absolute value extends to a
non-archimedean absolute value on Qp , and Qp becomes a complete, locally
compact, totally disconnected Hausdor space. Similarly to the completion R
of Q with respect to the standard (archimedean) absolute value, Qp is a eld,
and its elements have a p-adically convergent series representation: for any
Qp , there exist integers and ak such that


= ak pk with 0 ak < p.
k(;p)

The set of p-adic numbers with ||p 1 forms a ring, the ring Zp of p-adic
integers. Another construction of p-adic numbers uses the representation of
Zp as a projective limit of the ring of residue classes mod pk . It is customary
to write | | for the standard absolute value on Q, Q for R, and then
call R the completion of Q at the innite prime p = . Two absolute values
are called equivalent if they induce the same topology. Ostrowski (see [279,
Sect. II.3]) showed that every non-trivial absolute value on Q is equivalent to
one of the absolute values | |p , where p is a prime number or p = . Thus,
completion of Q with respect to its non-equivalent absolute values yields a
family of complete, locally compact topological elds Qp which contain Q,
one for each place p :

Q " Q2 , Q3 , Q5 , . . . , and Q = R.

p-adic analysis is quite dierent from real analysis. A non-archimedean

absolute value induces a curious (ultrametric) topology. In p-adic analysis,
the role of the intervals in R are played by the balls

a + p Zp := { Qp : | a|p p },

where a Qp and Z. These balls are clopen sets, i.e., they are closed and
open. It follows that each point inside a ball is center of the ball, and that any
two balls are either disjoint or contained one in another.
The algebraic closure Qp of Qp has innite degree over Qp , but is not
complete, and so it is not the right eld for doing analysis. However, the
completion Cp of Qp is algebraically closed. If

n = 0 + 1 p + 2 p2 + and k = 0 + 1 p + 2 p2 +

are the p-adic expansions of the integers n and k, then one can show that
n
   
0 1
mod p.
k 0 1
 A Short History of Universality 287

In view of the binomic inversion formula this leads for any continuous functions
f : Zp Cp to the representation

n n

f (n) = ak (f ) ,
k
k=0

where n N and

k  
kj k
ak (f ) := (1) f (j).
j=0
j

It is remarkable that this yields a series representation, the so-called Mahler

series of f . For k N0 , we dene the nite dierence operator " by

k  
k
"k f (x) = (1)kj f (x + j).
j=0
j

Then ak = "k f (0) for k N. Mahler [233] proved

Theorem A.1. Let f : Zp Cp be a continuous function. Then ak 0 as

k , and

n x

ak f (x) uniformly on Zp .
k n
k=0

A proof can be found in Robert [311, Sect. 4.2.4]. Notice that this theorem
does not hold over R.
In particular, Mahlers Theorem A.1 implies a p-adic version of
Weierstrass approximation theorem: for any continuous function f : Zp Cp ,
there exists a sequence of polynomials fn with coecients in Cp that converges
uniformly to f . Further applications of Mahlers theorem are the construc-
tion of p-adic analogues of classical functions like the exponential function,
the logarithm or the Gamma-function; for details we refer again to Roberts
monograph [311].
Now, it is not too surprising that an analogue of Feketes universality
theorem holds in the p-adic case too. Steuding [341] obtained

Theorem A.2. There exists a p-adic power series n=1 an xn with the prop-
erty that for any ball a + p Zp , where a, Z, 0 < a < p, and any continuous
function f , dened on a+p Zp and satisfying f (a) = 0, there exists a sequence
of positive integers mk such that

mk
an xn f (x) uniformly on a + p Zp .
k
n=1

We follow Luh [227] in his proof of a version of Feketes theorem.

288 Appendix

Proof. Let {Qn } be a sequence of all polynomials with integral coecients.

We construct a sequence of polynomials {Pn } as follows: let P0 = Q0 and
assume that P0 , . . . , Pn1 are known. Denote by dn the degree of Pn1 . Now
let n be a continuous function on Zp such that
 


n1
n (x) = Qn (x) Pk (x) xdn 1
k=0

for x a + pn Zp . In view of Mahlers Theorem A.1 there exists a polynomial

fn so that
max |fn (x) n (x)|p pdn .
xa+pn Zp

Setting Pn (x) = fn (x)xdn +1 , the sequence {Pn } is constructed. We note

 n 


 
max  Pk (x) Qn (x) = max |(fn (x) n (x))xdn +1 |p
xa+pn Zp   xa+pn Zp
k=0 p

pdn .

By construction, distinct Pn have no powers in common. Thus we can

rearrange formally the polynomial series into a power series:





an xn := Pn (x).
n=1 n=0

Again by Mahlers Theorem A.1, for any continuous function f on a + p Zp

there exists a sequence of positive integers nk , tending to innity with k, such
that
max |f (x) Qnk (x)|p pnk .
xa+pnk Zp

For suciently large k the ball a + p Zp is contained in a + pnk Zp . In view

of the above estimates we obtain
   
 dnk +1   

 
nk

max  f (x) a xn
= max  f (x) P (x)

xa+pnk Zp 
n  xa+pnk Zp 
n

n=1  n=0 p
p
 

nk 
 
max |f (x) Qnk (x)|p , Qnk (x) Pn (x)
xa+pnk Zp  
n=0 p
nk dnk
max{p ,p },

which tends to zero as k . Thus, putting mk = dnk + 1, the assertion of

the theorem follows. 

 A Short History of Universality 289

In the years after Feketes discovery many universal objects were found.
For instance, Birkho [23] proved in 1929 the existence of an entire function
f (z) with the property that to any given entire function g(z) there exists a
sequence of complex numbers an such that
f (z + an ) g(z) uniformly on compacta in C.
n

The proof relies in the main part on Runges approximation theorem. This
type of universality is very similar to the one of the Riemann zeta-function
and other Dirichlet series. Birkhos theorem states the existence of an entire
function with wild behaviour near innity. Luh [228] constructed holomorphic
monsters, that are holomorphic functions with an extraordinary wild bound-
ary behaviour in arbitrary simply connected open sets. More precisely: let G
be a proper open subset of C with simply connected components. Then there
exists a function f holomorphic on G such that for every boundary point z of
G, every compact subset K with connected complement and every continuous
function g on K which is holomorphic in the interior of K, there exist linear
transformations n (z) = an z + bn with n (K) G and dist(n (K), z) 0 as
n for which
f (n (z)) g(z) uniformly on K;
n

in addition, each derivative of f and each antiderivative of f of arbitrary order

has the boundary behaviour described above. We shall explain the notion
antiderivative of a holomorphic function f dened on a simply connected open
subset of C; in fact, this notion is not unique. Here, for a negative integer j,
the jth antiderivative f (j) of f with order |j| is dened by

d|j| (j)
f (z) = f (z).
d|j| z
Any other antiderivative of f with the same order |j| diers from f (j) on each
component of G by some polynomial of degree less than |j|. Moreover, in [230],
Luh proved the existence of multiply universal functions, that are holomorphic
functions that satisfy, along with their derivatives and antiderivatives, six
universal properties at the same time.
Marcinkiewicz [237] was in 1935 the rst to use the notion universal-
ity when he proved the existence of a continuous function whose dierence
quotients approximate any measurable function in the sense of convergence
almost everywhere. This should be compared with the result of Blair and
Rubel [25] who proved that there exists an entire function f such that the
set {f (n) : n N0 } of all derivatives of f is dense in the space of all entire
functions in the topology of uniform convergence on compact subsets of the
complex plane. Other universal objects are, for example, conformal mappings
composed with universal functions, discovered by Luh [229]. However, for a
long time no explicit example of a universal object was found until Voronin
discovered in 1975 that the Riemann zeta-function is universal!
290 Appendix

In all these examples of universality there are two characteristic aspects

of universality, namely the existence of a single object which
is maximal divergent
(via a countable process) allows to approximate a maximal class of objects.
This observation led to understand universality as a phenomenon which occurs
quite naturally in certain limiting processes. Meanwhile it turned out that the
phenomenon of universality is anything but a rare event in analysis! Many
analytical processes which diverge or behave irregularly in some cases produce
universal objects.
Grosse-Erdmann gave a rather general description of universality as fol-
lows. There is a topological space X of objects, a topological space Y of ele-
ments to be approximated, and a family of continuous mappings Tj : X Y
for j J. Then an object x X is called universal if every element y Y can
be approximated by certain Tj (x), i.e., the set {Tj (x) : j J} lies dense in Y.
In the special case when the mappings Tj form a group of homeomorphisms,
the concept of universality is well-known in topological dynamics under the
name of topological transitivity (this reminds us of Bagchis reformulation
of Riemanns hypothesis in Sect. 8.2). In operator theory, where iterates T j
of an operator T are studied, universal elements are said to be hypercyclic.
The general setting of Grosse-Erdmann covers quite many universality results.
In [108, 109] he proved the following universality criterion.
Theorem A.3. Suppose that X is a Baire space and Y is of the second cate-
gory. Then the following assertions are equivalent:
The set U of universal elements is residual in X .
The set U of universal elements is dense in X .
To every pair of non-empty sets V X and W Y there exists some
j J with
Tj (V ) W = .
If one of these conditions holds, then U is a dense G -subset of X .
We briey explain the topological notions before we discuss a special case
of this theorem; for details we refer to Grosse-Erdmann [108], Kelleys mono-
graph [169] and Rudins monograph [312]. In 1899, Baire introduced the notion
of category to measure the size of subsets of topological spaces. A subset E
of a topological space X is called nowhere dense if its closure E contains no
non-empty open subset of X . Any countable union of nowhere dense sets is
called a set of the rst category (meager); all other subsets of X are said to
be of the second category (non-meager). The complement of a set of the rst
category is called residual (co-meager). A topological space X is said to be
a Baire space if the intersection of any countable family of open and dense
subsets of X is dense in X . Countable intersections of open sets are called G -
sets. The theorem of Baire states that any complete metric space is a Baire
space.
 A Short History of Universality 291

In many applications of Theorem A.3, both X and Y are metric spaces.

Then the rst two assumptions of the theorem are fullled if X is complete and
Y is separable. Furthermore, the third assertion can be rewritten as follows:
For every x X and y Y there exist sequences {xn } in X and {jn } in
J such that Tjn (xn ) tends to y as xn x.
For a verication of this statement one needs a suitable approximation
theorem; for example, Weierstrass approximation theorem for Feketes uni-
versality theorem, respectively, Runges approximation theorem for Birk-
hos universality result. With regard to universality for Dirichlet series,
this ts to the denseness Theorem 5.10 for the space of analytic functions
(which is a complete separable metric space by Theorem 3.15) and the use of
Mergelyans approximation Theorem 5.15, in the proof of the general univer-
sality Theorem 5.14, for S. There is another remarkable aspect of Theorem
A.3: if we observe the phenomenon of universality in some space X , then
the set of universal elements is dense. This observation supports the Linnik
Ibragimov conjecture (see Sect. 1.6)! This can be stated in a more explicit way:
Nestoridis
 and Papadimitropoulos [278] proved the existence of a Dirichlet se-
ries n=1 anns , absolutely convergent for any > 0, with the property that for
every admissible set K {s C : 0} and every entire function g, there
exists a sequence of positive integers {j }j such that for any  = 0, 1, 2, . . .
()


j
an
g(s)() uniformly on K,
ns
n=1

as j . Furthermore, the set of such Dirichlet series is dense and G (in the
Baire sense) in the space of absolutely convergent Dirichlet series in the right
half-plane. Here a set K is called admissible if K is compact with connected
complement, and K is the nite union of sets Kr each of which contained in
a vertical strip of width less than 12 . Their approach does not produce any
explicit example of a universal Dirichlet series.
Universality is far away from being completely understood. In particular,
the discovery of explicit examples of universal objects (zeta- and L-functions)
has led to many new and interesting questions. It seems that universality of
general Dirichlet series is not an arithmetic phenomenon at all, but it is much
easier to nd universal Dirichlet series explicitly among those associated with
arithmetic objects.
References

1. J. Andersson, Disproof of some conjectures of K. Ramachandra,

HardyRamanujan J. 22 (1999), 27
2. H. Aramata, ber die Teilbarkeit der Dedekindschen Zetafunktionen, Proc.
Imp. Acad. Jpn. 9 (1933), 3134
3. J. Arthur, L. Clozel, Simple Algebras, Base Change and the Advanced Theory
of the Trace Formula, Ann. Math. Stud., vol. 120, Princeton University Press
(1990)
4. E. Artin, ber eine neue Art von L-Reihen, Abh. Math. Sem. Univ. Hamburg
3 (1923), 89108
5. E. Artin, Beweis des allgemeinen Reziprozittsgesetzes, Abh. Math. Sem. Univ.
Hamburg 5 (1927), 353363
6. A.O.L. Atkin, J. Lehner, Hecke operators on 0 (m), Math. Ann. 185 (1970),
134160
7. R. Ayoub, Euler and the zeta-function, Am. Math. Mon. 81 (1974), 10671086
8. R. Backlund, ber die Beziehung zwischen Anwachsen und Nullstellen der
Zetafunktion, fversigt Finska Vetensk. Soc. 61(9) (19181919)
9. B. Bagchi, The statistical Behaviour and Universality Properties of the
Riemann Zeta-Function and Other Allied Dirichlet Series, Ph.D. Thesis,
Calcutta, Indian Statistical Institute, (1981)
10. B. Bagchi, A joint universality theorem for Dirichlet L-functions, Math. Z. 181
(1982), 319334
11. B. Bagchi, Recurrence in topological dynamics and the Riemann hypothesis,
Acta Math. Hung. 50 (1987), 227240
12. A. Baker, Linear Forms in the Logarithms of Algebraic Numbers. I, Mathe-
matika 13 (1966), 204216
13. A. Baker, Transcendental Number Theory, Cambridge University Press (1975)
14. E.P. Balanzario, Remark on Dirichlet series satisfying functional equations,
Divulg. Mat. 8 (2000), 169175
15. R. Balasubramanian, K. Ramachandra, On the frequency of Titchmarshs phe-
nomenon for (s). III. Proc. Indian Acad. Sci. 86 (1977), 341351
16. H. Bauer, The value distribution of Artin L-series and zeros of zeta-functions,
J. Number Theory 98 (2003), 254279
17. P. Bauer, ber den Anteil der Nullstellen der Riemannschen Zeta-Funktion auf
der Kritischen Geraden, Diploma Thesis, Frankfurt University (1992), available
at www.math.uni-frankfurt.de/pbauer/diplom.ps
294 References

18. P. Bauer, Zeros of Dirichlet L-series on the critical line, Acta Arith. 93 (2000),
3752
19. S.N. Bernstein, The extension of properties of trigonometric polynomials
to entire functions of nite degree, Izvestiya Akad. Nauk SSSR 12 (1948),
421444 (Russian)
20. A. Beurling, Analyse de la loi asymptotique de la distribution des nombres
premiers gnraliss, I, Acta. Math. 68 (1937), 255291
21. P. Billingsley, Convergence of Probability Measures, Wiley, New York (1968)
22. P. Billingsley, Probability and Measure 3rd Edition, Wiley, New York (1995)
23. G.D. Birkho, Dmonstration dun thorme lmentaire sur les fonctions
entires, C. R. Acad. Sci. Paris 189 (1929), 473475
24. K.M. Bitar, N.N. Khuri, H.C. Ren, Path integrals and Voronins theorem on
the universality of the Riemann zeta-function, Ann. Phys. 211 (1991), 172196
25. C. Blair, L.A. Rubel, A universal entire function, Am. Math. Mon. 90 (1983),
331332
26. R.P. Boas, Entire Functions, Academic, New York (1954)
27. S. Bochner, On Riemanns functional equation with multiple gamma factors,
Ann. Math. 67 (1958), 2941
28. H. Bohr, ber das Verhalten von (s) in der Halbebene > 1, Nachr. Akad.
Wiss. Gttingen II Math. Phys. Kl. (1911), 409428
29. H. Bohr, Zur Theorie der Riemannschen Zetafunktion im kritischen Streifen,
Acta Math. 40 (1915), 67100
30. H. Bohr, ber eine quasi-periodische Eigenschaft Dirichletscher Reihen mit
Anwendung auf die Dirichletschen L-Funktionen, Math. Ann. 85 (1922),
115122
31. H. Bohr, B. Jessen, ber die Werteverteilung der Riemannschen Zetafunktion,
erste Mitteilung, Acta Math. 54 (1930), 135
32. H. Bohr, B. Jessen, ber die Werteverteilung der Riemannschen Zetafunktion,
zweite Mitteilung, Acta Math. 58 (1932), 155
33. H. Bohr, E. Landau, ber das Verhalten von (s) und (k) (s) in der Nhe der
Geraden = 1, Nachr. Ges. Wiss. Gttingen Math. Phys. Kl. (1910), 303330
34. H. Bohr, E. Landau, Ein Satz ber Dirichletsche Reihen mit Anwendung auf
die -Funktion und die L-Funktionen, Rend. di Palermo 37 (1914), 269272
35. H. Bohr, E. Landau, Sur les zros de la fonction (s) de Riemann, Comptes
Rendus Acad. Sci. Paris 158 (1914), 106110
36. H. Bohr, E. Landau, Nachtrag zu unseren Abhandlungen aus den Jahren 1910
und Nachr. Ges. Wiss. Gttingen Math. Phys. Kl. (1924), 168172
37. E. Bombieri, On the large sieve, Mathematika 12 (1965), 201225
38. E. Bombieri, D.A. Hejhal, Sur les zros des fonctions zta dEpstein, Comptes
Rendus Acad. Sci. Paris 304 (1987), 213217
39. E. Bombieri, D.A. Hejhal, On the distribution of zeros of linear combinations
of Euler products, Duke Math. J. 80 (1995), 821862
40. E. Bombieri, A. Perelli, Distinct zeros of L-functions, Acta Arith. 83 (1998),
271281
41. A. Borel, Automorphic L-functions, in: Automorphic Forms, Representations
and L-Functions, Proceedings of the Symposium in Pure Mathematics, vol. 33,
part 2, 2761, Am. Math. Soc., Providence (1979)
42. R. Brauer, On Artins L-series with general group characters, Ann. Math. 48
(1947), 502514
 References 295

43. R. Brauer, Beziehungen zwischen Klassenzahlen von Teilkrpern eines galoiss-

chen Krpers, Math. Nachr. 4 (1951), 158174
44. C. Breuil, B. Conrad, F. Diamond, R. Taylor, On the modularity of elliptic
curves over Q: Wild 3-adic exercises, J. Am. Math. Soc. 14 (2001), 843939
45. V.V. Buldygin, Convergence of Random Series in Topological Spaces, Naukova
Dumka, Kiev (1985) (in Russian)
46. D. Bump, Automorphic Forms and Representations, Cambridge University
Press (1997)
47. D. Bump, J.W. Cogdell, D. Gaitsgory, E. de Shallit, S.S. Kudla, An Introduc-
tion to the Langlands Program, J. Bernstein, S. Gelbart (Eds.), Birkhuser,
Boston (2003)
48. R.B. Burckel, Introduction to Classical Complex Analysis, vol. I, Birkhuser,
Boston (1979)
49. F. Carlson, ber die Nullstellen der Dirichletschen Reihen und der
Riemannschen -Funktion, Arkiv fr Mat. Astr. och Fysik 15(20) (1920)
50. F. Carlson, Contributions a la theorie des series de Dirichlet. I, Arkiv fr Mat.
Astr. och Fysik 16(18) (1922)
51. F. Carlson, Contributions a la theorie des series de Dirichlet. II, Arkiv fr Mat.
Astr. och Fysik 19(26) (1926)
52. K. Chandrasekharan, R. Narasimhan, The approximate functional equation for
a class of zeta-functions, Math. Ann. 152 (1963), 3064
53. N. Chebotarev, Determination of the density of the set of prime numbers,
belonging to a given substitution class, Izv. Ross. Akad. Nauk 17 (1924),
205250 (in Russian)
54. P.L. Chebyshev, Sur la fonction qui dtermine la totalit des nombres premiers
infrieurs une limite donne, Mmoires des savants trangers de lAcad. Sci.
St. Ptersbourg 5 (1848), 119
55. P.L. Chebyshev, Mmoire sur nombres premiers, Mmoires des savants
trangers de lAcad. Sci. St. Ptersbourg 7 (1850), 1733
56. J. Cogdell, P. Michel, On the complex moments of symmetric power L-functions
at s = 1, Int. Math. Res. Not. 31 (2004), 15611617
57. J.B. Conrey, More than two fths of the zeros of the Riemann zeta-function
are on the critical line, J. Reine Angew. Math. 399 (1989), 126
58. J.B. Conrey, The Riemann hypothesis, Notices Am. Math. Soc. 50 (2003),
341353
59. J.B. Conrey, A. Ghosh, On the Selberg class of Dirichlet series: Small degrees,
Duke Math. J. 72 (1993), 673693
60. J.B. Conrey, A. Ghosh, Remarks on the generalized Lindelf hypothesis,
preprint, available at www.math.okstate.edu/conrey/papers.html
61. J.B. Conrey, S.M. Gonek, High moments of the Riemann zeta-function, Duke
Math. J. 107 (2001), 577604
62. J.B. Conway, Functions of One Complex Variable I 2nd Edition, Springer,
Berlin Heidelberg New York (1978)
63. H. Cramr, Ein Mittelwertsatz in der Primzahltheorie, Math. Z. 12 (1922),
147153
64. H. Cramr, M. Leadbetter, Stationary and Related Stochastic Processes, Wiley,
New York (1967)
65. H. Davenport, H. Heilbronn, On the zeros of certain Dirichlet series I, II, J.
Lond. Math. Soc. 11 (1936), 181185; 307312
296 References

66. R. Dedekind, ber die Anzahl der Ideal-Klassen in den verschiedenen Ord-
nungen eines endlichen Krpers, in: Festschrift der Technischen Hochschule in
Braunschweig zur Skularfeier des Geburtstages von C.F. Gau, Braunschweig
(1877), 155; Gesammelte Werke, vol. 1, Vieweg, Braunschweig (1930), 104158
67. P. Deligne, La Conjecture de Weil I, II, Publ. I.H.E.S. 43 (1974), 273307; 52
(1981), 313428
68. A. Denjoy, LHypothse de Riemann sur la distribution des zros de (s), relie
la thorie des probabilits, Comptes Rendus Acad. Sci. Paris 192 (1931),
656658
69. P.G.L. Dirichlet, Beweis des Satzes, dass jede unbegrenzte arithmetische
Progression, deren erstes Glied und Dierenz ganze Zahlen ohne gemein-
schaftlichen Factor sind, unendlich viele Primzahlen enthlt, Abhdl. Knigl.
Preuss. Akad. Wiss. (1837), 4581
70. V.G. Drinfeld, Langlands Conjecture for GL(2) Over Function Fields, Proceed-
ings of the International Congress of Mathematicians, Helsinki 1978, 565574,
Acad. Sci. Fennica, Helsinki (1980)
71. W. Duke, Some problems in multidimensional analytic number theory, Acta
Arith. 52 (1989), 203228
72. N. Dunford, J.T. Schwartz, Linear Operators, vol. 1, Interscience, New York
(1958)
73. P.L. Duren, Theory of H p -Spaces, Academic, New York (1970)
74. H.M. Edwards, Riemanns Zeta-Function, Academic, New York (1974)
75. K.M. Eminyan, -universality of the Dirichlet L-function, Mat. Zametki 47
(1990), 132137 (Russian); Math. Notes 47 (1990), 618622
76. L. Euler, Variae observationes circa series innitas, Comment. Acad. Sci.
Petropol 9 (1737), 160188
77. D.W. Farmer, Long molliers of the Riemann zeta-function, Matematika 40
(1993), 7187
78. D.W. Farmer, Counting distinct zeros of the Riemann zeta-function, Electron.
J. Combin. 2 (1995), 5 (electronic)
79. D. Flath, Decomposition of representations into tensor products, in: Auto-
morphic Forms, Representations and L-Functions, vol. 33, Am. Math. Soc.,
Providence (1979), 179184
80. M. van Frankenhuijsen, Arithmetic progressions of zeros of the Riemann zeta-
function, J. Number Theor. 115 (2005), 360370
81. M.Z. Garaev, On a series with simple zeros of (s), Math. Notes 73 (2003),
585587
82. M.Z. Garaev, One inequality involving simple zeros of (s), HardyRamanujan
J. 26 (2003), 1822
83. M.Z. Garaev, On vertical zeros of R(s) and I(s), Acta Arith. 108 (2003),
245251
84. V. Garbaliauskiene, A. Laurinikas, Discrete value-distribution of L-functions
of elliptic curves, Publ. Inst. Math. (Beograd) 76 (2004), 6571
85. R. Garunktis, On zeros of the Lerch zeta-function II, in: Probability Theory
and Mathematical Statistics, Proceedings of the Seventh Vilnius Conference,
Grigelionis et al. (Eds.), TEV Vilnius (1998), 267276
86. R. Garunktis, The eective universality theorem for the Riemann zeta-
function, in: Special Activity in Analytic Number Theory and Diophantine
Equations, Proceedings of a Workshop at the Max Planck-Institut Bonn 2002,
R.B. Heath-Brown and B. Moroz (Eds.), Bonner Math. Schriften 360 (2003)
 References 297

87. R. Garunktis, On Voronins universality theorem for the Riemann zeta-

function, Fiz. Mat. Fak. Moksl. Semin. Darb. 6 (2003), 2933
88. R. Garunktis, A. Laurinikas, On zeros of the Lerch zeta-function, in: Num-
ber Theory and its Applications, S. Kanemitsu and K. Gyry (Eds.), Kluwer
Academicdelete Publishers", Dordrecht (1999), 129143
89. R. Garunktis, A. Laurinikas, The Lerch Zeta-Function, Kluwer, Dordrecht
(2002)
90. R. Garunktis, A. Laurinikas, R. leeviiene, J. Steuding, On the universality
of the Estermann zeta-function, Analysis 22 (2002), 285296
91. R. Garunktis, J. Steuding, On the zero distribution of Lerch zeta-functions,
Analysis 22 (2002), 112
92. R. Garunktis, J. Steuding, Does the Lerch zeta-function satisfy the Lindelf
hypothesis?, in: Analytic and Probabilistic Methods in Number Theory, Pro-
ceedings of the Third International Conference in Honour of J. Kubilius,
Palanga 2001, A. Dubickas et al. (Eds.), TEV, Vilnius (2002), 6174
93. F. Gamann, ber Beziehungen zwischen den Primidealen eines algebraischen
Krpers und den Substitutionen seiner Gruppen, Math. Z. 25 (1926), 661675
94. C.F. Gauss, Asymptotische Gesetze der Zahlentheorie, Werke, vol. 10.1 (1791),
1116, Teubner 1917
95. P.M. Gauthier, N. Tarkhanov, Approximation by the Riemann zeta-function,
Complex Variables 50 (2005), 211215
96. S.S. Gelbart, Automorphic Forms on Adele Groups, Princeton University Press
and University of Tokyo Press, Princeton (1975)
97. S.S. Gelbart, An elementary introduction to the Langlands program, Bull. Am.
Math. Soc. 10 (1984), 177219
98. J. Genys, A. Laurinikas, Value distribution of general Dirichlet series. V, Lith.
Math. J. 44 (2004), 145156
99. P. Grardin, W. Li, Functional equations and periodic sequences, in: Thorie
des nombres, Quebec 1987, J.-M. De Koninck and C. Levesque (Eds.), de
Gruyter, Berlin (1989), 267279
100. E. Girondo, J. Steuding, Eective estimates for the distribution of values of
Euler products, Monatshefte Math. 145 (2005), 97106
101. R. Godement, H. Jacquet, Zeta-Functions of Simple Algebras, Lecture Notes
260, Springer, Berlin Heidelberg New York (1972)
102. D. Goldfeld, Gauss class number problem for imaginary quadratic number
elds, Bull. Am. Math. Soc. 13 (1985), 2337
103. D.A. Goldston, S.M. Gonek, H.L. Montgomery, Mean values of the logarithmic
derivative of the Riemann zeta-function with aplications to primes in short
intervals, J. Reine Angew. Math. 537 (2001), 105126
104. S.M. Gonek, Analytic Properties of Zeta and L-Functions, Ph.D. Thesis, Uni-
versity of Michigan, (1979)
105. A. Good, On the distribution of the values of Riemanns zeta-function, Acta
Arith. 38 (1981), 347388
106. S.A. Gritsenko, On zeros of linear combinations of analogues of the Riemann
function, Tr. Mat. Inst. Steklova 218 (1997), 134150 (Russian); translation
in Proc. Steklov Inst. Math. 218 (1997), 129145
107. B. Gross, D.B. Zagier, Heegner points and derivatives of L-series, Invent. Math.
84 (1986), 225320
108. K.-G. Grosse-Erdmann, Holomorphe Monster und universelle Funktionen,
Mitt. Math. Sem. Giessen 176 (1987), 181
298 References

109. K.-G. Grosse-Erdmann, Universal families and hypercyclic operators, Bull.

Am. Math. Soc. 36 (1999), 345381
110. E. Grosswald, F.J. Schnitzer, A class of modied and L-functions, Pacific J.
Math. 74(2) (1978), 357364
111. G.G. Gundersen, Meromorphic functions that share four values, Trans. Am.
Math. Soc. 277 (1983), 545567
112. M.C. Gutzwiller, Stochastic behavior in quantum scattering, Physica D 7
(1983), 341355
113. J. Hadamard, tude sur le propits des fonctions entires et en partic-
ulier dune fonction considre par Riemann, J. Math. Pures Appl. 9 (1893),
171215
114. J. Hadamard, Sur les zros de la fonction (s) de Riemann, Comptes Rendus
Acad. Sci. Paris 122 (1896), 14701473
115. H. Hamburger, ber die Riemannsche Funktionalgleichung der -Funktion,
Math. Z. 10 (1921), 240254
116. G.H. Hardy, Sur les zros de la fonction (s) de Riemann, Comptes Rendus
Acad. Sci. Paris 158 (1914), 10121014
117. G.H. Hardy, J.E. Littlewood, Some problems of partitio numerorum; III: On
the expression of a number as a sum of primes, Acta Math. 44 (1922), 170
118. G.H. Hardy, J.E. Littlewood, The approximate functional equation in the the-
ory of the zeta-function, with applications to the divisor problems of Dirichlet
and Piltz, Proc. Lond. Math. Soc. 21 (1922), 3974
119. G.H. Hardy, J.E. Littlewood, On Lindelfs hypothesis concerning the Riemann
zeta-function, Proc. Royal Soc. 103 (1923), 403412
120. G.H. Hardy, E.M. Wright, An Introduction to the Theory of Numbers, 5th
Edition, Oxford Science, Oxford (1979)
121. T. Hattori, K. Matsumoto, A limit theorem for BohrJessens probability
measures for the Riemann zeta-function, J. Reine Angew. Math. 507 (1999),
219232
122. D.R. Heath-Brown, Simple zeros of the Riemann zeta-function on the critical
line, Bull. Lond. Math. Soc. 11 (1979), 1718
123. D.R. Heath-Brown, Gaps between primes and the pair correlation between
zeros of the Riemann zeta-function, Acta Math. 41 (1982), 8599
124. E. Hecke, ber die Zetafunktion beliebiger algebraischer Zahlkrper, Nachr.
Ges. Wiss. Gttingen (1917), 7789
125. E. Hecke, ber eine neue Art von Zetafunktionen, Math. Z. 6 (1920), 1151
126. E. Hecke, ber die Bestimmung Dirichletscher L-Reihen durch ihre Funktion-
algleichung, Math. Ann. 112 (1936), 664699
127. K. Heegner, Diophantische Analysis und Modulfunktionen, Math. Z. 56 (1952),
227253
128. H. Heilbronn, On the class number in imaginary quadratic elds, Quarterly J.
Math. 5 (1934), 150160
129. H. Heilbronn, Zeta-functions and L-functions, in: Algebraic Number Theory,
J.W.S. Cassels and A. Frhlich (Eds.), Academic, New York, (1967), 204230
130. D.A. Hejhal, On a result of Selberg concerning zeros of linear combinations of
L-functions, Internat. Math. Res. Notices 11 (2000), 551577
131. K. Hensel, ber eine neue Begrndung der Theorie der algebraischen Zahlen,
Jahresberichte DMV 6 (1897), 8388
132. E. Hewitt, K.A. Ross, Abstract Harmonic Analysis, vol. I, Springer, Berlin
Heidelberg New York (1979)
 References 299

133. H. Heyer, Probability Measures on Locally Compact Groups, Springer, Berlin

Heidelberg New York (1977)
134. D. Hilbert, Mathematische Probleme, Archiv Math. Physik 1 (1901), 4463,
213317
135. O. Hlder, ber die Eigenschaft der Gammafunktion keiner algebraischen Dif-
ferentialgleichung zu gengen, Math. Ann. 28 (1887), 113
136. C.P. Hughes, On the Characteristic Polynomial of a Random Unitary Matrix
and the Riemann Zeta-Function, Ph.D. Thesis, University of Bristol (2001)
137. 
A. Hurwitz,
 1 Einige Eigenschaften der Dirichletschen Funktionen F (s) =
D
n ns , die bei der Bestimmung der Klassenzahlen binrer quadratischer
Formen auftreten, Zeitschrift Math. Physik 27 (1887), 86101
138. M.N. Huxley, Exponential sums and the Riemann zeta-function, V, Proc. Lond.
Math. Soc. 90 (2005), 141
139. A.E. Ingham, Mean-value theorems in the theory of the Riemann zeta-function,
Proc. Lond. Math. Soc. 27 (1926), 273300
140. A.E. Ingham, On two conjectures in the theory of numbers, Am. J. Math. 64
(1942), 313319
141. A. Ivi, The Theory of the Riemann Zeta-Function with Applications, Wiley,
New York (1985)
142. A. Ivi, On the multiplicity of zeros of the zeta-function, Bull. Acad. Serbe Sci.
Arts, Classe Sci. Math. 24 (1999), 119132
143. A. Ivi, On small values of the Riemann zeta-function on the critical line and
gaps between zeros, Lith. Math. J. 42 (2002), 2536
144. H. Iwaniec, Topics in Classical Automorphic Forms, Am. Math. Soc., Provi-
dence (1997)
145. H. Iwaniec, P. Sarnak, Perspectives on the analytic theory of L-functions,
Geom. Funct. Anal., special volume - GAFA (2000), 705741
146. H. Jacquet, Principal L-functions for GL(n), Proc. Symp. Pure Math. 61 (1997),
321329
147. H. Jacquet, R.P. Langlands, Automorphic Forms on GL(2), Lecture Notes 114,
Springer, Berlin Heidelberg New York (1970)
148. H. Jacquet, J.A. Shalika, A non-vanishing theorem for zeta-functions on GLn ,
Invent. Math. 38 (1976), 116
149. B. Jessen, A. Wintner, Distribution functions and the Riemann zeta-function,
Trans. Am. Math. Soc. 38 (1935), 4888
150. D. Joyner, Distribution Theorems of L-Functions, Pitman Research Notes in
Mathematics, (1986)
151. G. Julia, Lecons sur les fonctions uniformes point singulier essentiel isol,
Gauthier-Villars, Paris (1924)
152. A. Kaenas, A. Laurinikas, On the periodic zeta-function, Liet. Mat. Rink.
38 (1998), 8297 (Russian); Lith. Math. J. 38 (1998), 6476
153. R. Kainskaite, A discrete limit theorem for the Matsumoto zeta-function in the
space of analytic functions, Liet. Mat. Rink. 41 (2001), 441448 (in Russian);
Lith. Math. J. 41 (2001), 344350
154. R. Kainskaite, Discrete Limit Theorems for the Matsumoto Zeta-Function,
Ph.D. Thesis, Vilnius University (2002)
155. R. Kainskaite, A. Laurinikas, On the value distribution of the Matsumoto
zeta-function, Acta Math. Hung. 105 (2004), 339359
300 References

156. J. Kaczorowski, M. Kulas, On the non-trivial zeros o the critical line for
L-functions from the extended Selberg class, Monatshefte Math. 150 (2007),
217232
157. J. Kaczorowski, A. Laurinikas, J. Steuding, On the value distribution of shifts
of universal Dirichlet series, Monatshefte Math. 147 (2006), 309317
158. J. Kaczorowski, G. Molteni, A. Perelli, J. Steuding, J. Wolfart, Heckes theory
and the Selberg class, Funct. Approx. Comment. Math. 35 (2006), 183194
159. J. Kaczorowski, A. Perelli, Functional independence of the singularities of a
class of Dirichlet series, Am. J. Math. 120 (1998), 289303
160. J. Kaczorowski, A. Perelli, The Selberg class: A survey, in: Number Theory
in Progress. Proceedings of the International Conference in Honor of the 60th
Birthday of Andrej Schinzel, Zakopane 1997. vol. 2: Elementary and analytic
number theory. De Gruyter (1999), 953992
161. J. Kaczorowski, A. Perelli, On the structure of the Selberg class, I: 0 d 1,
Acta Math. 182 (1999), 207241
162. J. Kaczorowski, A. Perelli, On the structure of the Selberg class, V: 1 < d <
5/3, Invent. Math. 150 (2002), 485516
163. J. Kaczorowski, A. Perelli, On the prime number theorem in the Selberg class,
Arch. Math. 80 (2003), 255263
164. A.A. Karatsuba, On the zeros of the function (s) on short intervals of the
critical line, Izv. Akad. Nauk. SSSR Ser. Mat. 48 (1984) (Russian); Math.
USSR-Izv. 24 (1985), 523537
165. A.A. Karatsuba, Complex Analysis in Number Theory, CRC, Boca Raton, FL
(1995)
166. A.A. Karatsuba, S.M. Voronin, The Riemann Zeta-Function, de Gruyter, New
York (1992)
167. N.M. Katz, P. Sarnak, Random Matrices, Frobenius Eigenvalues, and
Monodromy, Am. Math. Soc., Providence (1999)
168. J.P. Keating, N.C. Snaith, Random matrix theory and ( 12 + it), Comm. Math.
Phys. 214 (2000), 5789
169. J.L. Kelley, General Topology, Van Nostrand, New Jersey (1955)
170. H.H. Kim, F. Shahidi, Symmetric cube L-functions for GL2 are entire, Ann.
Math. 150 (1999), 645662
171. N. Koblitz, Introduction to Elliptic Curves and Modular Forms, Springer,
Berlin Heidelberg New York (1984)
172. H. von Koch, Sur la distribution des nombres premiers, Comptes Rendus Acad.
Sci. Paris 130 (1900), 12431246
173. N.M. Korobov, Estimates of trigonometric sums and their applications, Uspehi
Mat. Nauk 13 (1958), 185192 (Russian)
174. S. Koyama, H. Mishou, Universality of Hecke L-functions in the
grossencharacter-aspect, Proc. Jpn. Acad. 78 (2002), 6367
175. L. Laorgue, Chtoucas de Drinfeld et correspondance des Langlands, Invent.
Math. 147 (2002), 1241
176. E. Landau, Neuer Beweis des Primzahlsatzes und Beweis des Primidealsatzes,
Math. Ann. 56 (1903), 645670
177. R.P. Langlands, Problems in the Theory of Automorphic Forms, Lecture Notes
170, Springer, Berlin Heidelberg New York (1970), 1886
178. R.P. Langlands, L-functions and automorphic representations, Proceedings of
the International Congress of Mathematicians, Helsinki 1978, 165175, Acad.
Sci. Fennica, Helsinki (1980)
 References 301

179. M.L. Lapidus, M. van Frankenhuijsen, Fractal Geometry and Number Theory,
Birkhuser, Boston (2000)
180. A. Laurinikas, A limit theorem for Dirichlet L-functions, Mat. Zametki 25
(1979), 481485
181. A. Laurinikas, Distributions des valeurs de certaines sries de Dirichlet,
Comptes Rendus Acad. Sci. Paris 289 (1979), 4345
182. A. Laurinikas, Zeros of the derivative of the Riemann zeta-function, Litovsk.
Mat. Sb. 25 (1985), 111118 (Russian)
183. A. Laurinikas, Limit theorems for the Riemann zeta-function in the complex
space, in: Probability Theory and Mathematical Statistics, Proceedings Sixth
Vilnius Conference, B. Grigelionis et al. (Eds.), VSP/Mokslas, Vilnius 1990,
5969
184. A. Laurinikas, A limit theorem for the Riemann zeta-function near the critical
line in the complex space, Acta Arith. 59 (1991), 19
185. A. Laurinikas, On limit theorems for the Riemann zeta-function in some
spaces, in: Probability Theory and Mathematical Statistics, Proceedings of the
Sixth Vilnius Conference 1993, Grigelionis et al. (Eds.) VSP Utrecht/Vilnius
(1994), 457483
186. A. Laurinikas, Limit Theorems for the Riemann Zeta-Function, Kluwer,
Dordrecht 1996
187. A. Laurinikas, On the limit distribution of the Matsumoto zeta-function II,
Liet. Mat. Rink. 36 (1996), 464485 (Russian); Lith. Math. J. 36 (1996),
371387
188. A. Laurinikas, On the limit distribution of the Matsumoto zeta-function I,
Acta Arith. 79 (1997), 3139
189. A. Laurinikas, The universality of the Lerch zeta-functions, Liet. Mat. Rink.
37 (1997), 367375 (Russian); Lith. Math. J. 37 (1997), 275280
190. A. Laurinikas, On the Matsumoto zeta-function, Acta Arith. 84 (1998), 116
191. A. Lauriniaks, On the zeros of linear combinations of Matsumoto zeta-
functions, Liet. Mat. Rink. 38 (1998) (Russian), 185204; Lith. Math. J. 38
(1998), 144159
192. A. Laurinikas, A limit theorem in the theory of nite Abelian groups, Publ.
Math. Debrecen 52 (1998), 517533
193. A. Laurinikas, On the eectivization of the universality theorem for the Lerch
zeta-function, Liet. Mat. Rink. 40 (2000), 172178 (Russian); Lith. Math. J.
40 (2000), 135139
194. A. Laurinikas, The universality of Dirichlet series attached to nite abelian
groups, in: Number Theory, Proceedings of the Turku Symposium on Number
theory in Memory of Kustaa Inkeri (1999), M. Jutila, T. Metsnkyl (Eds.),
Walter de Gruyter, Berlin, New York (2001), 179192
195. A. Laurinikas, A probabilistic equivalent of the Lindelf hypothesis, in: An-
alytic and Probabilistic Methods in Number Theory, Proceedings of the Third
International Conference in Honour of J. Kubilius, Palanga 2001, A. Dubickas
et al. (Eds.), TEV, Vilnius (2002), 157161
196. A. Laurinikas, The universality of zeta-functions, Proceedings of the Eighth
Vilnius Conference on Probability Theory and Mathematical Statistics, Part I
(2002), Acta Appl. Math. 78(13) (2003), 251271
197. A. Laurinikas, On the derivatives of zeta-functions of certain cusp forms,
Glasgow Math. J. 47 (2005), 8796
302 References

198. A. Laurinikas, Prehistory of the Voronin universality theorem, iauliai Math.

J. 1 (9) (2006), 4153
199. A. Laurinikas, K. Matsumoto, On estimation of the number of zeros of linear
combinations of certain zeta-functions, LMD Mokslo Darbai (1998), 4348
200. A. Laurinikas, K. Matsumoto, The joint universality and the functional inde-
pendence for Lerch zeta-functions, Nagoya Math. J. 157 (2000), 211227
201. A. Laurinikas, K. Matsumoto, The universality of zeta-functions attached to
certain cusp forms, Acta Arith. 98 (2001), 345359
202. A. Laurinikas, K. Matsumoto, The joint universality of twisted automorphic
L-functions, J. Math. Soc. Jpn. 56 (2004), 923939
203. A. Laurinikas, K. Matsumoto, Joint value-distribution theorems on Lerch
zeta-functions II, Liet. Mat. Rink. 46 (2006), 332350 (Russian); Lith. Math.
J. 46 (2006), 271286
204. A. Laurinikas, K. Matsumoto, J. Steuding, The universality of L-functions
associated with newforms, Izvestija Math. 67 (2003), 7790; Izvestija Ross.
Akad. Nauk Ser. Mat. 67 (2003), 8396 (Russian)
205. A. Laurinikas, K. Matsumoto, J. Steuding, Discrete universality of L-functions
for newforms, Math. Notes 78 (2005), 551558; Mat. Zametki 78 (2005),
595603 (Russian)
206. A. Laurinikas, W. Schwarz, J. Steuding, The universality of general Dirichlet
series, Analysis 23 (2003), 1326
207. A. Laurinikas, R. leeviciene, The universality of zeta-functions with mul-
tiplicative coecients, Integral Transforms and Special Functions 13 (2002),
243257
208. A. Laurinikas, J. Steuding, A short note on the Lindelf hypothesis, Lith.
Math. J. 43 (2003), 5155; Liet. Mat. Rink. 43 (2003), 5964 (Russian)
209. A. Laurinikas, J. Steuding, Joint universality for L-functions attached to a
family of elliptic curves, Proceedings of the ELAZ 2004, Conference on ele-
mentary and analytic number theory, Mainz 2004, W. Schwarz, J. Steuding
(Eds.), Steiner, Stuttgart (2006), 155163
210. A. Laurinikas, J. Steuding, A limit theorem for the Hurwitz zeta-function
with an algebraic irrational parameter, Arch. Math. 85 (2005), 419432
211. A. Laurinikas, J. Steuding, Limit theorems in the space of analytic functions
for the Hurwitz zeta-function with an algebraic irrational parameter, (submit-
ted)
212. D.H. Lehmer, The Vanishing of Ramanujans Function, Duke Math. J. 14
(1947), 429433
213. D. Leitmann, D. Wolke, Periodische und multiplikative zahlentheoretische
Funktionen, Monatsh. Math. 81 (1976), 279289
214. C.G. Lekkerkerker, On the Zeros of a Class of Dirichlet Series, Proefschrift,
van Gorcum, NV (1955) 
215. M. Lerch, Note sur la fonction K(w, x, s) = k=0 e2kix (w + k)s , Acta Math.
11 (1887), 1924
216. N. Levinson, More than one third of Riemanns zeta-function are on = 12 ,
Adv. Math. 13 (1974), 383436
217. N. Levinson, Almost all roots of (s) = a are arbitrarily close to = 1/2, Proc.
Nat. Acad. Sci. USA 72 (1975), 13221324
218. N. Levinson, H.L. Montgomery, Zeros of the derivative of the Riemann zeta-
function, Acta Math. 133 (1974), 4965
 References 303

219. H.-Z. Li, J. Wu, The universality of symmetric power L-functions and their
RankinSelberg L-functions, J. Math. Soc. Jpn. (in print)
220. E. Lindelf, Quelques remarques sur la croissance de la fonction (s), Bull. Sci.
Math. 32 (1908), 341356
221. E. Lindelf, E. Phragmn, Sur une extension dun principe classique danalyse
et sur quelques proprits de fonctions monognes dans le voisinage dun point
singulier, Acta Math. 31 (1908), 381406
222. R. Lipschitz, Untersuchung einer aus vier Elementen gebildeten Reihe, J. Reine
und Angew. Math. 54 (1857), 313328
223. J.E. Littlewood, Quelques consquences de lhypothse que la fonction (s) de
Riemann na pas de zros dans de demi-plan R(s) > 12 , Compt. Rend. Acad.
Sci. Paris 154 (1912), 263266
224. J.E. Littlewood, On the zeros of the Riemann zeta-function, Proc. Camb.
Philol. Soc. 22 (1924), 295318
225. J. Liu, Y. Wang, Y. Ye, A proof of Selbergs orthogonality for automorphic
L-functions, Manuscripta Math. 118 (2005), 135149
226. M. Love, Probability Theory, Van Nostrand, Toronto (1955)
227. W. Luh, Universalfunktionen in einfach zusammenhngenden Gebieten,
Aequationes Mathematicae 19 (1979), 183193
228. W. Luh, Holomorphic monsters, J. Approx. Theory 53 (1988), 128144
229. W. Luh, Universal functions and conformal mappings, Serdica 19 (1993),
161166
230. W. Luh, Multiply universal holomorphic functions, J. Approx. Theory 89
(1997), 135155
231. E. Lukacs, Characteristic Functions, 2nd Edition, Grin, London (1970)
232. J. van de Lune, H.J.J. te Riele, D.T. Winter, On the zeros of the Riemann
zeta-function in the critical strip, IV, Math. Comp. 46 (1986), 667681
233. K. Mahler, p-adic Numbers and their Functions 2nd Edition, Cambridge Uni-
versity Press (1981)
234. S. Mandelbrojt, Dirichlet Series, Reidel, Dordrecht (1972)
235. H. von Mangoldt, Zu Riemanns Abhandlung ber die Anzahl der Primzahlen
unter einer gegebenen Grsse, J. Reine Angew. Math. 114 (1895), 255305
236. H. von Mangoldt, Zur Verteilung der Nullstellen der Riemannschen Funktion
(t), Math. Ann. 60 (1905), 119
237. J. Marcinkiewicz, Sur les nombres drivs, Fund. Math. 24 (1935), 305308
238. R. Marszaek, The multiplicative and functional independence of Dedekind
zeta-functions of abelian elds, Coll. Math. 103 (2005), 1116
239. K. Matsumoto, Value-Distribution of Zeta-Functions, in Analytic Number
Theory, Proceedings of the Japanese-French Symposium, Tokyo 1988,
K. Nagasaka and . Fouvry (Eds.), Lecture Notes in Maths 1434, Springer,
Berlin Heidelberg New York (1990), 178187
240. K. Matsumoto, Recent developments in the mean square theory of the Riemann
zeta and other zeta-functions, in: Number Theory, Trends Math., Birkhuser,
Boston (2000), 241286
241. K. Matsumoto, The mean values and the universality of Rankin-Selberg L-
functions, in: Number Theory, Proceedings of the Turku Symposium on Num-
ber Theory in Memory of Kustaa Inkeri (1999), M. Jutila, T. Metsnkyl
(Eds.), Walter de Gruyter, Berlin, New York (2001), 201221
304 References

242. K. Matsumoto, Probabilistic value-distribution theory of zeta-functions,

Sugaku 53 (2001), 279296 (in Japanese); English translation in Sugaku Expo-
sitions 17 (2004), 5171
243. K. Matsumoto, Some problems on mean values and the universality of zeta
and multiple zeta-functions, in: Analytic and Probabilistic Methods in Num-
ber Theory, Proceedings of the Third International Conference in Honour of
J. Kubilius, Palanga 2001, A. Dubickas et al. (Eds.), TEV, Vilnius (2002),
195199
244. K. Matsumoto, An introduction to the value-distribution theory of zeta-
functions, iauliai Math. J. 1 (9) (2006), 6183
245. J.-L. Mauclaire, Almost periodicity and Dirichlet series, preprint
246. S.N. Mergelyan, Uniform approximations to functions of a complex variable,
Usp. Mat. Nauk 7 (1952), 31122 (Russian); Am. Math. Soc. (Translated) 101
(1954), 99
247. F. Mertens, Ein Beitrag zur analytischen Zahlentheorie, J. Reine Angew. Math.
78 (1874), 4662
248. F. Mertens, ber eine zahlentheoretische Funktion, Sem. Ber. Kais. Akad.
Wiss. Wien 106 (1897), 761830
249. H. Mishou, The universality theorem for L-functions associated with ideal class
characters, Acta Arith. 98 (2001), 395410
250. H. Mishou, The universality theorem for Hecke L-functions, Acta Arith. 110
(2003), 4571
251. H. Mishou, H. Nagoshi, Functional distribution of L(s, d ) with real characters
and denseness of Quadratic Class Numbers, Trans. Am. Math. Soc. 358 (2006),
43434366
252. H. Mishou, H. Nagoshi, The universality of quadratic L-series for prime dis-
criminants, Acta Arith. 123 (2006), 143161
253. H. Mishou, H. Nagoshi, Equivalents of the Riemann hypothesis, Arch. Math.
86 (2006), 419424
254. H. Mishou, H. Nagoshi, The universality for Rankin-Selberg L-functions in the
level aspect, preprint
255. D.S. Mitrinovic, J. Sndor, B. Crstici, Handbook of Number Theory, Kluwer,
Dordrecht 1995
256. G. Molteni, A note on a result of Bochner and Conrey-Ghosh about the Selberg
class, Arch. Math. 72 (1999), 219222
257. G. Molteni, J. Steuding, (Almost) primitivity of Hecke L-functions, Monat-
shefte Math. (in print)
258. H.L. Montgomery, Topics in Multiplicative Number Theory, Lecture Notes 227,
Springer, Berlin Heidelberg New York (1971)
259. H.L. Montgomery, The pair correlation of zeros of the zeta-function, Proc.
Symp. Pure Math. 24 (1973), 181193
260. H.L. Montgomery, Extreme values of the Riemann zeta-function, Comment.
Math. Helv. 52 (1977), 511518
261. L.J. Mordell, On Ramanujans empirical expansions of modular functions,
Proc. Camb. Philol. Soc. 19 (1920), 117124
262. C.J. Moreno, Explicit formulas in the theory of automorphic forms, in: Number
Theory Day, Proceedings of the Conference in Rockefeller University, New
York 1976, Lecture Notes in Math. 626, Springer, Berlin Heidelberg New York
(1977), 73216
 References 305

263. Y. Motohashi, Spectral Theory of the Riemann Zeta-Function, Cambridge Uni-

versity Press (1997)
264. H. Mller, ber eine Klasse modizierter - und L-Funktionen, Arch. Math.
36 (1981), 157161
265. M.R. Murty, Oscillations of Fourier coecients of modular forms, Math. Ann.
262 (1983), 431446
266. M.R. Murty, A motivated introduction to the Langlands program, in: Advances
in Number theory, F. Gouvea and N. Yui, (Eds.), Clarendon Press Oxford
(1993), 3766
267. M.R. Murty, Selbergs conjectures and Artin L-functions, Bull. Am. Math. Soc.
31 (1994), 114
268. M.R. Murty, V.K. Murty, Strong multiplicity one for Selbergs class, C.R.
Acad. Sci. Paris Ser. I Math. 319 (1994), 315320
269. M.R. Murty, Selbergs conjectures and Artin L-functions, II, in: Current Trends
in Mathematics and Physics, S.D. Adhikari, (Ed.), Narosa, New Delhi (1995),
154168
270. M.R. Murty, V.K. Murty, Non-Vanishing of L-Functions and Applications,
Birkhuser, Boston (1997)
271. M.R. Murty, A. Perelli, The pair correlation of zeros of functions in the Selberg
class, Int. Math. Res. Notices 10 (1999), 531545
272. V.K. Murty, On the Sato-Tate conjecture, in: Number Theory Related to
Fermats Last Theorem, N. Koblitz, (Ed.), Birkhuser, Boston (1982), 195205
273. H. Nagoshi, On the universality for L-functions attached to Maass forms,
Analysis 25 (2005), 122
274. H. Nagoshi, The universality of L-functions attached to Maass forms, Advanced
Studies in Pure Mathematics, Proceedings of the International Conference on
Probability and Number Theory, Kanazawa (2005) (in print)
275. H. Nagoshi, The universality of families of automorphic L-functions, preprint
276. T. Nakamura, The existence and the non-existence of joint t-universality for
Lerch zeta-functions, J. Number Theory (in print)
277. W. Narkiewicz, The Development of Prime Number Theory, Springer, Berlin
Heidelberg New York (2000)
278. V. Nestoridis, C. Papadimitropoulos, Abstract theory of universal series and
an application to Dirichlet series, C.R. Acad. Sci. Paris 341 (2005), 539543
279. J. Neukirch, Algebraische Zahlentheorie, Springer 1992; English Translation:
Algebraic Number Theory, Springer, Berlin Heidelberg New York (1999)
280. R. Nevanlinna, Einige Eindeutigkeitsstze in der Theorie der meromorphen
Funktionen, Acta Math. 48 (1926), 367391
281. R. Nevanlinna, Analytic functions, Springer, Berlin Heidelberg New York
(1970), translated from the 2nd German edition (1953)
282. F. Nicolae, ber die lineare Unabhngigkeit der Dedekindschen Zetafunktionen
galoisscher Zahlkrper, Math. Nachr. 220 (2000), 111113
283. A.M. Odlyzko, The 1020 th zero of the Riemann zeta-function and 70 million
of its neighbors, in: Dynamical, spectral, and arithmetic zeta functions. San
Antonio, TX, (1999), 139144, Contemp. Math. 290, Am. Math. Soc., Provi-
dence (2001)
284. A.M. Odlyzko, H.J.J. te Riele, Disproof of Mertens conjecture, J. Reine Angew.
Math. 367 (1985), 138160
285. A.P. Ogg, A remark on the Sato-Tate conjecture, Invent. Math. 9 (1970),
198200
306 References

286. A. Ostrowski, ber Dirichletsche Reihen und algebraische Dierentialgleichun-

gen, Math. Z. 8 (1920), 241298
287. J. Pl, Zwei kleine Bemerkungen, Tohoku Math. J. 6 (1914/15), 4243
288. D.V. Pechersky, On the permutation of the terms of functional series, Dokl.
Akad. Nauk SSSR 209 (1973), 12851287 (Russian)
289. A. Perelli, General L-functions, Ann. Mat. Pura Appl. 130 (1982), 287306
290. A. Perelli, A survey of the Selberg class of L-functions, Part I, Milan J. Math.
73 (2005), 1952
291. R. Perlis, On the equation K (s) = K
(s), J. Number Theory 9 (1977), 342360
292. R. Perlis, A. Schinzel, Zeta functions and the equivalence of integral forms,
J. Reine Angew. Math. 309 (1979), 176182
293. H. Petersson, Konstruktion der smtlichen Lsungen einer Riemannschen
Funktionalgleichung durch Dirichletreihen mit Eulerscher Produktentwicklung
II, Math. Ann. 117 (1940/41), 3964
294. I. Piatetski-Shapiro, R. Rhagunathan, On Hamburgers theorem, Am. Math.
Soc. Transl. 169 (1995), 109120
295. J. Popken, A measure for the dierential-transcendence of the zeta-function
of Riemann, in: Number Theory and Analysis, (papers in honor of Edmund
Landau), Plenum, New York 1969, 245255
296. H.S.A. Potter, The mean values of certain Dirichlet series I, Proc. Lond. Math.
Soc. 46 (1940), 467468
297. C.R. Putnam, On the non-periodicity of the zeros of the Riemann zeta-
function, Am. J. Math. 76 (1954), 9799
298. C.R. Putnam, Remarks on periodic sequences and the Riemann zeta-function,
Am. J. Math. 76 (1954), 828830
299. K. Ramachandra, On Riemann zeta-function and allied questions, Astrisque
209 (1992), 5772 Journes Arithmtiques, Geneva (1991)
300. S. Ramanujan, On certain arithmetical functions, Trans. Camb. Phil. Soc. 22
(1916), 159184
301. S. Ramanujan, On certain trigonometrical sums and their applications in the
theory of numbers, Trans. Camb. Philol. Soc. 22 (1918), 259276
302. R.A. Rankin, Contributions to the theory of Ramanujans  -function and sim-
(n)
ilar arithmetical functions. I. The zeros of the function n=1 ns
on the line
Rs = 13 2
, Proc. Camb. Philol. Soc. 35 (1939), 351356
303. R.A. Rankin, Contributions to the theory of Ramanujans -function and sim-
ilar arithmetical functions. II. The order of the Fourier coecients of integral
modular forms, Proc. Camb. Philol. Soc. 35 (1939), 357372
304. R.A. Rankin, An -result for the coecients of cusp forms, Math. Ann. 203
(1973), 239250
305. A. Reich, Universelle Wertverteilung von Eulerprodukten, Nach. Akad. Wiss.
Gttingen, Math.-Phys. Kl. (1977), 117
306. A. Reich, Wertverteilung von Zetafunktionen, Arch. Math. 34 (1980), 440451
307. A. Reich, Zetafunktionen und Dierenzen-Dierentialgleichungen, Arch. Math.
38 (1982), 226235
308. H.-E. Richert, ber Dirichletreihen mit Funktionalgleichungen, Publ. Inst.
Math. Acad. Serbe Sci. 11 (1957), 73124
309. G.J. Rieger, Eective simultaneous approximation of complex numbers by
conjugate algebraic integers, Acta Arith. 63 (1993), 325334
310. B. Riemann, ber die Anzahl der Primzahlen unterhalb einer gegebenen
Grsse, Monatsber. Preuss. Akad. Wiss. Berlin (1859), 671680
 References 307

311. A.M. Robert, A course in p-adic Analysis, Springer, Berlin Heidelberg New
York (2000)
312. W. Rudin, Real and Complex Analysis, Mc Graw-Hill, New York (1966)
313. W. Rudin, Functional Analysis, Mc Graw-Hill, New York (1973)
314. Z. Rudnick, P. Sarnak, Zeros of principal L-functions and Random Matrix
Theory, Duke Math. J. 81 (1996), 269322
315. J. Sander, J. Steuding, Joint universality for sums and products of Dirichlet
L-functions, Analysis 26 (2006), 295312
316. W. Schnee, Die Funktionalgleichung der Zetafunktion und der Dirichletschen
Reihen mit periodischen Koezienten, Math. Z. 31 (1930), 378390
317. W. Schwarz, Geschichte der analytischen Zahlentheorie seit 1890, in: Ein
Jahrhundert Mathematik 18901990, DMV, Vieweg (1990), 741780
318. W. Schwarz, Some remarks on the history of the prime number theorem from
1896 to 1960, in: Development of Mathematics 19001950, J.-P. Pier, (Ed.)
Birkhuser, Boston 1994 (Symposium Luxembourg 1992), 565615
319. W. Schwarz, R. Steuding, J. Steuding, Universality for Euler products and
related arithmetical functions, (submitted)
320. A. Selberg, Bemerkungen ber eine Dirichletsche Reihe, die mit der Theorie
der Modulformen nahe verbunden ist, Arch. Math. Naturvid. 43 (1940), 4750
321. A. Selberg, On the zeros of the Riemann zeta-function, Skr. Norske Vid. Akad.
Oslo 10 (1942), 159
322. A. Selberg, On the estimation of Fourier coecients of modular forms, Proc.
Sym. Pure Math. VIII (1965), 115
323. A. Selberg, Old and new conjectures and results about a class of Dirichlet series,
Proceedings of the Amal Conference on Analytic Number Theory, Maiori
1989, E. Bombieri et al. (Eds.), Universit di Salerno (1992), 367385
324. J.-P. Serre, Abelian l-adic representations and elliptic curves, Research Notes
in Mathematics, AK Peters (1968)
325. J.-P. Serre, Linear representations of finite groups, Springer, Berlin Heidelberg
New York (1977)
326. F. Shahidi, Symmetric power L-functions for GL(2), in: Elliptic curves and
related objects, H. Kisilevsky and M.R. Murty,(Ed.), CRM Proceedings and
Lecture Notes, vol. 4 (1994), 159182
327. F. Shahidi, Automorphic L-functions and functoriality, Proceedings of the
International Congress of Mathematicians, Bejing (2002), vol. III, 655666
328. G. Shimura, On the holomorphy of certain Dirichlet series, Proc. Lond. Math.
Soc. 31 (1975), 7998
329. C.L. Siegel, ber Riemanns Nachlass zur analytischen Zahlentheorie, Quellen
u. Studien zur Geschichte der Math. Astr. Phys. 2 (1932), 4580
330. C.L. Siegel, ber die Classenzahl quadratischer Zahlkrper, Acta Arith. 1
(1935), 8386
331. R. leeviiene, Joint limit theorems and universality for the Riemann and
allied zeta-functions, Ph.D. Thesis, Vilnius University (2002)
332. R. leeviiene, The joint universality for twists of Dirichlet series with multi-
plicative coecients, in: Analytic and probabilistic methods in Number Theory,
Proceedings of the Third International Conference in Honour of J. Kubilius,
Palanga 2001, A. Dubickas et al. (Eds.), TEV, Vilnius (2002), 303319
333. R. leeviiene, J. Steuding, Short series over simple zeros of the Riemann
zeta-function, Indagationes Math. 15 (2004), 129132
308 References

334. R. leeviiene-Steuding, Universality for generalized Euler products, Analysis

26 (2006), 337345
335. R. Spira, Zeros of Hurwitz zeta-functions, Math. Comp. 30 (1976), 863866
336. H.M. Stark, A Complete Determination of the Complex Quadratic Fields of
Class Number One, Michigan Math. J. 14 (1967), 127
337. H.M. Stark, Galois theory, algebraic numbers and zeta functions, in: From
Number Theory to Physics, Waldschmitd et al. (Eds.), Springer, Berlin
Heidelberg New York (1989), 313393
338. J. Steuding, Grosse Werte der Riemannschen Zetafunktion auf rektizierbaren
Kurven im kritischen Streifen, Arch. Math. 70 (1998), 371376
339. J. Steuding, On simple zeros of the Riemann zeta-function in short intervals
on the critical line, Acta Math. Hung. 96 (2002), 259308
340. J. Steuding, On Dirichlet series with periodic coecients, Ramanujan J. 6
(2002), 295306
341. J. Steuding, The world of p-adic numbers and p-adic functions, Fiz. Mat. Fak.
Moksl. Semin. Darb. 5 (2002), 90107
342. J. Steuding, Upper bounds for the density of universality, Acta Arith. 107
(2003), 195202
343. J. Steuding, A note on the extended Selberg class (degrees 0 and 1), Arch.
Math. 81 (2003), 2225
344. J. Steuding, Extremal values of Dirichlet L-functions in the half-plane of ab-
solute convergence, J. Tho. Nombres Bordeaux 16 (2004), 221232
345. J. Steuding, Universality in the Selberg class, in: Special Activity in Analytic
Number Theory and Diophantine Equations, Proceedings of a Workshop at
the Max Planck-Institut Bonn 2002, R.B. Heath-Brown and B. Moroz (Eds.),
Bonner math. Schriften 360 (2003)
346. J. Steuding, On the value-distribution of L-functions, Fiz. Mat. Fak. Moksl.
Semin. Darb. 6 (2003), 87119
347. J. Steuding, How many values can L-functions share? Fiz. Mat. Fak. Moksl.
Semin. Darb. 7 (2004), 7081
348. J. Steuding, On the zero-distribution of Epstein zeta-functions, Math. Annalen
333 (2005), 689697
349. J. Steuding, On the value-distribution of Epstein zeta-functions, Publicacions
Mat., Proceedings of Primeras Jornadas de Teoria de Nmeros, Vilanova i la
Geltr (Barcelona) 2005 (in print)
350. J. Steuding, Upper bounds for the density of universality II, Acta Mathematica
Universitatis Ostraviensis 13 (2005), 7382
351. R. Taylor, Automorphy for some l-adic lifts of automorphic mod l representa-
tions II. (preprint) available at http://www.math.harvard.edu/rtaylor/
352. E.C. Titchmarsh, The Theory of Functions, 2nd Edition, Oxford University
Press (1939)
353. E.C. Titchmarsh, The Theory of the Riemann Zeta-Function, 2nd Edition,
revised by D.R. Heath-Brown, Oxford University Press (1986)
354. V.A. Trenogin, Functional analysis, Nauka, Moscow (1980) (in Russian)
355. K.M. Tsang, Distribution of values of the Riemann zeta-function, Ph.D. Thesis,
Princeton (1984)
356. J. Tunnell, A classical diophantine problem and modular forms of weight 3/2,
Inventiones Math. 72 (1983), 323334
357. C.J. de la Valle-Poussin, Recherches analytiques sur la thorie des nombres
premiers, I-III, Ann. Soc. Sci. Bruxelles 20 (1896), 183256, 281362, 363397
 References 309

358. M.-F. Vignras, Facteurs gamma et quations fonctionnelles, in: Modular Func-
tions of one Variable I, Proceedings of the International Conference, University
of Bonn, 1976, A. Dold and B. Eckmann (Eds.), Lecture Notes in Math. 627,
Springer, Berlin Heidelberg New York (1977), 79103
359. I.M. Vinogradov, A new estimate for the function (1 + it), Izv. Akad. Nauk
SSSR, Ser. Mat. 22 (1958), 161164 (Russian)
360. A.I. Vinogradov, On the density hypothesis for Dirichlet L-functions, Izv. Akad.
Nauk SSSR 29 (1965), 903934 (Russian)
361. S.M. Voronin, The distribution of the non-zero values of the Riemann zeta-
function, Izv. Akad. Nauk Inst. Steklov 128 (1972), 131150 (Russian)
362. S.M. Voronin, On the functional independence of Dirichlet L-functions, Acta
Arith. 27 (1975), 493503 (Russian)
363. S.M. Voronin, Theorem on the universality of the Riemann zeta-function, Izv.
Akad. Nauk SSSR, Ser. Matem., 39 (1975), 475486 (Russian); Math. USSR
Izv. 9 (1975), 443445
364. S.M. Voronin, Analytic Properties of Dirichlet Generating Functions of Arith-
metic Objects, Ph.D. Thesis, Moscow, Steklov Math. Institute, (1977) (Russian)
365. M. Waldschmidt, A lower bound for linear forms in logarithms, Acta Arith. 37
(1980), 257283
366. A. Walsz, Weylsche Exponentialsummen in der neueren Zahlentheorie, VEB
Deutscher Verlag der Wissenschaften (1963)
367. J.L. Walsh, Interpolation and approximation by rational functions in the com-
plex domain, Am. Math. Soc. Coll. Publ. 20, (1960)
368. L.C. Washington, Introduction to Cyclotomic Fields, Springer, Berlin
Heidelberg New York (1980)
369. A. Weil, Sur la thorie du corps de classes, J. Math. Soc. Jpn. 3 (1951), 135
370. H. Weyl, ber ein Problem aus dem Gebiete der diophantischen Approxima-
tion, Gttinger Nachrichten (1914), 234244
371. A. Wiles, Modular elliptic curves and Fermats last theorem, Ann. Math. 141
(1995), 443551
372. Z. Ye, The Nevanlinna functions of the Riemann zeta-function, J. Math. Analy-
sis Appl. 233 (1999), 425435
Notations

We indicate here some of the notations and conventions used in these notes;
most of them are standard. However, this list is not complete; we omit notions
which only appear in one chapter (where they are dened in situ) or which
are covered by the index or which are standard.
As usual, we denote by N = {1, 2, 3, . . .} the set of positive integers. The
sets of integers, rational numbers, real numbers, and complex numbers are
denoted by Z, Q, R, and C, respectively. Following an old tradition in the
theory of the zeta-function, the complex variable is given by a mixture of
greek and latinletters: we write s = + it, where , t R, and i is the
imaginary unit 1 in the upper half-plane.
The letter p, with and without a subscript, denotes a prime number. We
write d | n (respectively, d
n) if the integer d divides (respectively, does not
divide) the integer n. The symbol stands either for some congruence or it
denotes that a function is constant. The number of elements of a nite set
A is denoted by A. The function (x) counts the number of primes p x.
Besides, we use many other arithmetical functions; in our notation we follow
the classic [120].
The logarithm is, as usual in number theory, always taken to the basis e =
exp(1). The integer part and fractional part of a real number x are indicated
by [x] and {x}, respectively. Very convenient is the use of the Landau- and
Vinogradov-symbols. Given two functions f (x) and g(x), both dened for
x X, where g(x) is positive for all x X, we write:
f (x) = O(g(x)) and f (x)  g(x), respectively, if there exists a constant
C 0 such that

|f (x)| Cg(x) for all x X;

f (x) g(x) if f (x)  g(x)  |f (x)|;

here X is specied either explicitly or implicitly. Usually, the set X is an

interval [, ) for some real number ; in this case we also write:
312 Notations

f (x) g(x) if the limit

|f (x)|
lim
x g(x)
exists and is equal to 1.
f (x) = o(g(x)) if the latter limit exists and is equal to zero.
f (x) = (g(x)) if
|f (x)|
lim inf >0
x g(x)

(this is the negation of f (x) = O(g(x))).

Sometimes the limit x is replaced by another limit x x0 , where

x0 is some complex number; in this case the limit x0 is explicitly stated. In
estimates,  always denotes a small positive number, not necessarily the same
at each appearance.
The letters P and Q always denote probability measures, often with a
subscript. meas (A) is the Lebesgue measure and m(A) stands for the Haar
measure of a measurable set A. By NL ( ) and NLc ( . ) we denote the numbers
of zeros and of c-values of the function L(s) in some region, often specied by
some data in the brackets or by subscripts.
Finally, we list all axioms which were used in these notes:

(i), (1) Ramanujan hypothesis. a(n)  n
for any  > 0, where the implicit
constant may depend on .
(ii) Analytic continuation. there exists a real number L such that L(s)
has an analytic continuation to the half-plane > L with L < 1
except for at most a pole at s = 1.
(2) Analytic continuation. There exists a non-negative integer k such that
(s 1)k L(s) is an entire function of nite order.
(iii) Finite order. There exists a constant L 0 such that, for any xed
> L and any  > 0,

L( + it)  |t|L +
as |t| ;

the implicit constant may depend on .

(3) Functional equation. L(s) satises a functional equation of type

L (s) = L (1 s),

where

f
L (s) := L(s)Q s
(j s + j )
j=1

with positive real numbers Q, j , and complex numbers j , with

Re j 0 and || = 1.
 Notations 313

(iv) Polynomial Euler product. There exists a positive integer m and for
every prime p, there are complex numbers j (p), 1 j m, such that
m 
 1
j (p)
L(s) = 1 .
p j=1
ps

(4) Euler product. L(s) satises




b(pk )
L(s) = Lp (s), where Lp (s) = exp
p
pks
k=1

with suitable coecients b(pk ) satisfying b(pk )  pk for some < 12 .

(v) Prime mean-square. There exists a positive constant such that
1

lim |a(p)|2 = .
x (x)
px

We denote by S the class of Dirichlet series satisfying the axioms (i)(v) and
by S  the so-called extended Selberg class of Dirichlet series satisfying (2) and
(3); the subclass of the latter class of all elements satisfying additionally the
axioms (1) and (4) is the Selberg class S.
Index

algebraic dierential equations 195 Dedekind zeta-function 28, 249

almost sure convergence 53 degree 113
almost periodic 156, 178 Delignes estimate 25, 282
analytic continuation 38, 111 Denjoys heuristics 54
analytic isomorphism 169 denseness 90, 98, 193, 235
approximation in the mean 10, 15, 73 density hypothesis 119, 131
arithmetical equivalence 256 theorem 7, 119, 160
Artin conjecture 269 Dirichlet character 20
Artin L-funtion 264 Dirichlet L-function 21, 245
Artins reciprocity law 262, 270 Dirichlet polynomial 67, 217
asymmetric zero-distribution 215 Dirichlet series 35
automorphic representation 280 with periodic coecients 209
discrete limit theorem 84
Bagchis theorem 158 discrete universality 28, 109
Beurling zeta-function 164 distribution 52
Birkho-Khintchine theorem 59
Bohr & Jessen limit theorem 11, 67, 194 eectivity 168
Bohrs probabilistic method 9 elliptic curves 27
Borel set 50 entire function of exponential type 96
Brauer induction theorem 270 ergodic 58, 70
Estermann zeta-function 31
Carlsons theorem 37, 43 Euler product 1, 21, 25, 38, 112
character 20, 57, 256 extended Selberg class 216
characteristic function 56
Chebotarevs density theorem 270, 276 Feketes universality theorem 285
Chebyshevs inequality 53 nite order 38, 150
clustering of zeros 137, 144, 214 Fourier transform 56, 57
convergence in distribution 52 functional distribution 23, 163
cusp form 24 functional equation 2, 21, 26, 112
c-values 138, 188, 193 functional independence 196, 253, 277

Davenport-Heilbronn zeta-function 138, general Dirichlet series 31

220 generated zero 159, 171, 188
316 Index

Grand Lindelf hypothesis 132 order 36, 38, 124

Grand Riemann hypothesis 115 orthogonality relation 21, 57, 245
grssencharacter 256
pair correlation conjecture 122
Haar measure 57 Pecherskys rearrangement theorem 14,
Hardy space 14 90
Hecke L-function 259 Perellis class 161
holomorphic monster 289 Perrons formula 5
Hurwitzs theorem 105 Phragmn-Lindelf principle 124
Hurwitz zeta-function 30, 223 polynomial Euler product 38, 40
hypertranscendence 196 Portmanteau theorem 51
positive density method 98, 243, 276
Jensens formula 141 Potters theorem 126
joint limit theorem 230 prime ideal theorem 250
joint universality 22, 239, 253, 274 prime mean-square 38, 102, 238
prime number theorem 7
Kroneckers approximation theorem 17, for arithmetic progressions 22
183 for the Selberg class 121
primitive 116, 219
Langlands reciprocity conjecture 281 probability measure 50
Laurinikas general limit theorem 83 Prokhorovs theorem 52
Lerch zeta-function 30, 131, 216, 226
Levinsons method 138 Ramanujans cusp form 25
L-function IX Ramanujan hypothesis 26, 38, 40, 111
to automorphic representations 281 Ramanujan-Petersson conjecture 282
to newforms/elliptic curves 25, 27 random element 52, 64, 230
limit theorem 67, 83 random matrix theory 123
Lindelf hypothesis 130, 133 random process 58
linear independence of log p 17, 69 ray class character 258
Linnik-Ibragimov conjecture 32, 291 relative compact 51
Littlewoods lemma 139 Riegers eective approximation
local roots 38, 40 theorem 183
Riemann hypothesis 4, 7, 54, 115, 155
Matsumoto zeta-function 29, 108 Riemann zeta-function IX, 1, 155
mean 53 Riemann-von Mangoldt formula 3, 113,
mean-square half-plane 37, 43, 63 145, 214
Mellins inversion formula 43 Rouchs theorem 156
Mergelyans theorem 107 Runges approximation theorem 60
Mertens hypothesis 55, 176
modular form 24 Selberg class 111, 216, 219
multiplicative 20 Selberg conjectures 120, 247, 271
multiplicative independence 252 shift universality principle 203
space of analytic functions 59, 94
Nevanlinna theory 147 splitting of primes 249, 265
newform 25 Stirlings formula 44
normality conjecture 122 strip of universality 19
strongly stationary process 58, 72
Omega-results 167, 190 strong reccurrent 158
one-parameter group 70 strong universality 30, 204, 219
 Index 317

support 60, 105, 235 universality 13, 285

symmetric power L-functions 133 in the character aspect 23
universality theorem 13, 19, 22, 26, 106
topological group 57 upper bound for the density of
topology of uniform convergence on universality 169, 221
compacta 59
torus 64
Voronins universality theorem 13, 19
twist 119, 238

uniform distribution 16 weak convergence 50

uniqueness theorems 151 weak joint universality 198, 225
Lecture Notes in Mathematics
For information about earlier volumes
please contact your bookseller or Springer
LNM Online archive: springerlink.com

Vol. 1711: W. Ricker, Operator Algebras Generated by Vol. 1733: K. Ritter, Average-Case Analysis of Numerical
Commuting Projections: A Vector Measure Approach Problems (2000)
(1999) Vol. 1734: M. Espedal, A. Fasano, A. Mikelic, Filtration in
Vol. 1712: N. Schwartz, J. J. Madden, Semi-algebraic Porous Media and Industrial Applications. Cetraro 1998.
Function Rings and Reflectors of Partially Ordered Editor: A. Fasano. 2000.
Rings (1999) Vol. 1735: D. Yafaev, Scattering Theory: Some Old and
Vol. 1713: F. Bethuel, G. Huisken, S. Mller, K. Stef- New Problems (2000)
fen, Calculus of Variations and Geometric Evolution Prob- Vol. 1736: B. O. Turesson, Nonlinear Potential Theory and
lems. Cetraro, 1996. Editors: S. Hildebrandt, M. Struwe Weighted Sobolev Spaces (2000)
(1999) Vol. 1737: S. Wakabayashi, Classical Microlocal Analysis
Vol. 1714: O. Diekmann, R. Durrett, K. P. Hadeler, in the Space of Hyperfunctions (2000)
P. K. Maini, H. L. Smith, Mathematics Inspired by Biol- Vol. 1738: M. mery, A. Nemirovski, D. Voiculescu,
ogy. Martina Franca, 1997. Editors: V. Capasso, O. Diek- Lectures on Probability Theory and Statistics (2000)
mann (1999) Vol. 1739: R. Burkard, P. Deuflhard, A. Jameson, J.-L.
Vol. 1715: N. V. Krylov, M. Rckner, J. Zabczyk, Stochas- Lions, G. Strang, Computational Mathematics Driven
tic PDEs and Kolmogorov Equations in Infinite Dimen- by Industrial Problems. Martina Franca, 1999. Editors:
sions. Cetraro, 1998. Editor: G. Da Prato (1999) V. Capasso, H. Engl, J. Periaux (2000)
Vol. 1716: J. Coates, R. Greenberg, K. A. Ribet, K. Ru- Vol. 1740: B. Kawohl, O. Pironneau, L. Tartar, J.-P. Zole-
bin, Arithmetic Theory of Elliptic Curves. Cetraro, 1997. sio, Optimal Shape Design. Tria, Portugal 1999. Editors:
Editor: C. Viola (1999) A. Cellina, A. Ornelas (2000)
Vol. 1717: J. Bertoin, F. Martinelli, Y. Peres, Lectures Vol. 1741: E. Lombardi, Oscillatory Integrals and Phe-
on Probability Theory and Statistics. Saint-Flour, 1997. nomena Beyond all Algebraic Orders (2000)
Editor: P. Bernard (1999) Vol. 1742: A. Unterberger, Quantization and Non-
Vol. 1718: A. Eberle, Uniqueness and Non-Uniqueness holomorphic Modular Forms (2000)
of Semigroups Generated by Singular Diffusion Opera- Vol. 1743: L. Habermann, Riemannian Metrics of Con-
tors (1999) stant Mass and Moduli Spaces of Conformal Structures
Vol. 1719: K. R. Meyer, Periodic Solutions of the N-Body (2000)
Problem (1999) Vol. 1744: M. Kunze, Non-Smooth Dynamical Systems
Vol. 1720: D. Elworthy, Y. Le Jan, X-M. Li, On the Geo- (2000)
metry of Diffusion Operators and Stochastic Flows (1999) Vol. 1745: V. D. Milman, G. Schechtman (Eds.), Geomet-
Vol. 1721: A. Iarrobino, V. Kanev, Power Sums, Goren- ric Aspects of Functional Analysis. Israel Seminar 1999-
stein Algebras, and Determinantal Loci (1999) 2000 (2000)
Vol. 1746: A. Degtyarev, I. Itenberg, V. Kharlamov, Real
Vol. 1722: R. McCutcheon, Elemental Methods in Ergodic
Enriques Surfaces (2000)
Ramsey Theory (1999)
Vol. 1747: L. W. Christensen, Gorenstein Dimensions
Vol. 1723: J. P. Croisille, C. Lebeau, Diffraction by an
(2000)
Immersed Elastic Wedge (1999)
Vol. 1748: M. Ruzicka, Electrorheological Fluids: Model-
Vol. 1724: V. N. Kolokoltsov, Semiclassical Analysis for ing and Mathematical Theory (2001)
Diffusions and Stochastic Processes (2000)
Vol. 1749: M. Fuchs, G. Seregin, Variational Methods
Vol. 1725: D. A. Wolf-Gladrow, Lattice-Gas Cellular for Problems from Plasticity Theory and for Generalized
Automata and Lattice Boltzmann Models (2000) Newtonian Fluids (2001)
Vol. 1726: V. Maric, Regular Variation and Differential Vol. 1750: B. Conrad, Grothendieck Duality and Base
Equations (2000) Change (2001)
Vol. 1727: P. Kravanja M. Van Barel, Computing the Zeros Vol. 1751: N. J. Cutland, Loeb Measures in Practice:
of Analytic Functions (2000) Recent Advances (2001)
Vol. 1728: K. Gatermann Computer Algebra Methods for Vol. 1752: Y. V. Nesterenko, P. Philippon, Introduction to
Equivariant Dynamical Systems (2000) Algebraic Independence Theory (2001)
Vol. 1729: J. Azma, M. mery, M. Ledoux, M. Yor (Eds.) Vol. 1753: A. I. Bobenko, U. Eitner, Painlev Equations in
Sminaire de Probabilits XXXIV (2000) the Differential Geometry of Surfaces (2001)
Vol. 1730: S. Graf, H. Luschgy, Foundations of Quantiza- Vol. 1754: W. Bertram, The Geometry of Jordan and Lie
tion for Probability Distributions (2000) Structures (2001)
Vol. 1731: T. Hsu, Quilts: Central Extensions, Braid Vol. 1755: J. Azma, M. mery, M. Ledoux, M. Yor
Actions, and Finite Groups (2000) (Eds.), Sminaire de Probabilits XXXV (2001)
Vol. 1732: K. Keller, Invariant Factors, Julia Equivalences Vol. 1756: P. E. Zhidkov, Korteweg de Vries and Nonlin-
and the (Abstract) Mandelbrot Set (2000) ear Schrdinger Equations: Qualitative Theory (2001)
Vol. 1757: R. R. Phelps, Lectures on Choquets Theorem Vol. 1786: S. D. Cutkosky, Monomialization of Mor-
(2001) phisms from 3-Folds to Surfaces (2002)
Vol. 1758: N. Monod, Continuous Bounded Cohomology Vol. 1787: S. Caenepeel, G. Militaru, S. Zhu, Frobenius
of Locally Compact Groups (2001) and Separable Functors for Generalized Module Cate-
Vol. 1759: Y. Abe, K. Kopfermann, Toroidal Groups gories and Nonlinear Equations (2002)
(2001) Vol. 1788: A. Vasilev, Moduli of Families of Curves for
Vol. 1760: D. Filipovic, Consistency Problems for Heath- Conformal and Quasiconformal Mappings (2002)
Jarrow-Morton Interest Rate Models (2001) Vol. 1789: Y. Sommerhuser, Yetter-Drinfeld Hopf alge-
Vol. 1761: C. Adelmann, The Decomposition of Primes in bras over groups of prime order (2002)
Torsion Point Fields (2001) Vol. 1790: X. Zhan, Matrix Inequalities (2002)
Vol. 1762: S. Cerrai, Second Order PDEs in Finite and Vol. 1791: M. Knebusch, D. Zhang, Manis Valuations
Infinite Dimension (2001) and Prfer Extensions I: A new Chapter in Commutative
Vol. 1763: J.-L. Loday, A. Frabetti, F. Chapoton, F. Goi- Algebra (2002)
chot, Dialgebras and Related Operads (2001) Vol. 1792: D. D. Ang, R. Gorenflo, V. K. Le, D. D. Trong,
Vol. 1764: A. Cannas da Silva, Lectures on Symplectic Moment Theory and Some Inverse Problems in Potential
Geometry (2001) Theory and Heat Conduction (2002)
Vol. 1765: T. Kerler, V. V. Lyubashenko, Non-Semisimple Vol. 1793: J. Corts Monforte, Geometric, Control and
Topological Quantum Field Theories for 3-Manifolds with Numerical Aspects of Nonholonomic Systems (2002)
Corners (2001) Vol. 1794: N. Pytheas Fogg, Substitution in Dynamics,
Vol. 1766: H. Hennion, L. Herv, Limit Theorems for Arithmetics and Combinatorics. Editors: V. Berth, S. Fer-
Markov Chains and Stochastic Properties of Dynamical enczi, C. Mauduit, A. Siegel (2002)
Systems by Quasi-Compactness (2001) Vol. 1795: H. Li, Filtered-Graded Transfer in Using Non-
Vol. 1767: J. Xiao, Holomorphic Q Classes (2001) commutative Grbner Bases (2002)
Vol. 1768: M. J. Pflaum, Analytic and Geometric Study of Vol. 1796: J.M. Melenk, hp-Finite Element Methods for
Stratified Spaces (2001) Singular Perturbations (2002)
Vol. 1769: M. Alberich-Carramiana, Geometry of the Vol. 1797: B. Schmidt, Characters and Cyclotomic Fields
Plane Cremona Maps (2002) in Finite Geometry (2002)
Vol. 1770: H. Gluesing-Luerssen, Linear Delay-
Vol. 1798: W.M. Oliva, Geometric Mechanics (2002)
Differential Systems with Commensurate Delays: An
Vol. 1799: H. Pajot, Analytic Capacity, Rectifiability,
Algebraic Approach (2002)
Menger Curvature and the Cauchy Integral (2002)
Vol. 1771: M. mery, M. Yor (Eds.), Sminaire de Prob-
Vol. 1800: O. Gabber, L. Ramero, Almost Ring Theory
abilits 1967-1980. A Selection in Martingale Theory
(2003)
(2002)
Vol. 1801: J. Azma, M. mery, M. Ledoux, M. Yor
Vol. 1772: F. Burstall, D. Ferus, K. Leschke, F. Pedit,
(Eds.), Sminaire de Probabilits XXXVI (2003)
U. Pinkall, Conformal Geometry of Surfaces in S4 (2002)
Vol. 1773: Z. Arad, M. Muzychuk, Standard Integral Vol. 1802: V. Capasso, E. Merzbach, B. G. Ivanoff,
Table Algebras Generated by a Non-real Element of Small M. Dozzi, R. Dalang, T. Mountford, Topics in Spatial
Degree (2002) Stochastic Processes. Martina Franca, Italy 2001. Editor:
Vol. 1774: V. Runde, Lectures on Amenability (2002) E. Merzbach (2003)
Vol. 1775: W. H. Meeks, A. Ros, H. Rosenberg, The Vol. 1803: G. Dolzmann, Variational Methods for Crys-
Global Theory of Minimal Surfaces in Flat Spaces. talline Microstructure Analysis and Computation (2003)
Martina Franca 1999. Editor: G. P. Pirola (2002) Vol. 1804: I. Cherednik, Ya. Markov, R. Howe, G. Lusztig,
Vol. 1776: K. Behrend, C. Gomez, V. Tarasov, G. Tian, Iwahori-Hecke Algebras and their Representation Theory.
Quantum Comohology. Cetraro 1997. Editors: P. de Bar- Martina Franca, Italy 1999. Editors: V. Baldoni, D. Bar-
tolomeis, B. Dubrovin, C. Reina (2002) basch (2003)
Vol. 1777: E. Garca-Ro, D. N. Kupeli, R. Vzquez- Vol. 1805: F. Cao, Geometric Curve Evolution and Image
Lorenzo, Osserman Manifolds in Semi-Riemannian Processing (2003)
Geometry (2002) Vol. 1806: H. Broer, I. Hoveijn. G. Lunther, G. Vegter,
Vol. 1778: H. Kiechle, Theory of K-Loops (2002) Bifurcations in Hamiltonian Systems. Computing Singu-
Vol. 1779: I. Chueshov, Monotone Random Systems larities by Grbner Bases (2003)
(2002) Vol. 1807: V. D. Milman, G. Schechtman (Eds.), Geomet-
Vol. 1780: J. H. Bruinier, Borcherds Products on O(2,1) ric Aspects of Functional Analysis. Israel Seminar 2000-
and Chern Classes of Heegner Divisors (2002) 2002 (2003)
Vol. 1781: E. Bolthausen, E. Perkins, A. van der Vaart, Vol. 1808: W. Schindler, Measures with Symmetry Prop-
Lectures on Probability Theory and Statistics. Ecole d erties (2003)
Et de Probabilits de Saint-Flour XXIX-1999. Editor: Vol. 1809: O. Steinbach, Stability Estimates for Hybrid
P. Bernard (2002) Coupled Domain Decomposition Methods (2003)
Vol. 1782: C.-H. Chu, A. T.-M. Lau, Harmonic Functions Vol. 1810: J. Wengenroth, Derived Functors in Functional
on Groups and Fourier Algebras (2002) Analysis (2003)
Vol. 1783: L. Grne, Asymptotic Behavior of Dynamical Vol. 1811: J. Stevens, Deformations of Singularities
and Control Systems under Perturbation and Discretiza- (2003)
tion (2002) Vol. 1812: L. Ambrosio, K. Deckelnick, G. Dziuk,
Vol. 1784: L. H. Eliasson, S. B. Kuksin, S. Marmi, J.-C. M. Mimura, V. A. Solonnikov, H. M. Soner, Mathematical
Yoccoz, Dynamical Systems and Small Divisors. Cetraro, Aspects of Evolving Interfaces. Madeira, Funchal, Portu-
Italy 1998. Editors: S. Marmi, J.-C. Yoccoz (2002) gal 2000. Editors: P. Colli, J. F. Rodrigues (2003)
Vol. 1785: J. Arias de Reyna, Pointwise Convergence of Vol. 1813: L. Ambrosio, L. A. Caffarelli, Y. Brenier,
Fourier Series (2002) G. Buttazzo, C. Villani, Optimal Transportation and its
Applications. Martina Franca, Italy 2001. Editors: L. A. Vol. 1837: S. Tavar, O. Zeitouni, Lectures on Probabil-
Caffarelli, S. Salsa (2003) ity Theory and Statistics. Ecole dEt de Probabilits de
Vol. 1814: P. Bank, F. Baudoin, H. Fllmer, L.C.G. Saint-Flour XXXI-2001. Editor: J. Picard (2004)
Rogers, M. Soner, N. Touzi, Paris-Princeton Lectures on Vol. 1838: A.J. Ganesh, N.W. OConnell, D.J. Wischik,
Mathematical Finance 2002 (2003) Big Queues. XII, 254 p, 2004.
Vol. 1815: A. M. Vershik (Ed.), Asymptotic Com- Vol. 1839: R. Gohm, Noncommutative Stationary
binatorics with Applications to Mathematical Physics. Processes. VIII, 170 p, 2004.
St. Petersburg, Russia 2001 (2003) Vol. 1840: B. Tsirelson, W. Werner, Lectures on Probabil-
Vol. 1816: S. Albeverio, W. Schachermayer, M. Tala- ity Theory and Statistics. Ecole dEt de Probabilits de
grand, Lectures on Probability Theory and Statistics. Saint-Flour XXXII-2002. Editor: J. Picard (2004)
Ecole dEt de Probabilits de Saint-Flour XXX-2000. Vol. 1841: W. Reichel, Uniqueness Theorems for Vari-
Editor: P. Bernard (2003) ational Problems by the Method of Transformation
Vol. 1817: E. Koelink, W. Van Assche (Eds.), Orthogonal Groups (2004)
Polynomials and Special Functions. Leuven 2002 (2003) Vol. 1842: T. Johnsen, A. L. Knutsen, K3 Projective Mod-
Vol. 1818: M. Bildhauer, Convex Variational Problems els in Scrolls (2004)
with Linear, nearly Linear and/or Anisotropic Growth Vol. 1843: B. Jefferies, Spectral Properties of Noncom-
Conditions (2003) muting Operators (2004)
Vol. 1819: D. Masser, Yu. V. Nesterenko, H. P. Schlick- Vol. 1844: K.F. Siburg, The Principle of Least Action in
ewei, W. M. Schmidt, M. Waldschmidt, Diophantine Geometry and Dynamics (2004)
Approximation. Cetraro, Italy 2000. Editors: F. Amoroso, Vol. 1845: Min Ho Lee, Mixed Automorphic Forms, Torus
U. Zannier (2003) Bundles, and Jacobi Forms (2004)
Vol. 1820: F. Hiai, H. Kosaki, Means of Hilbert Space Vol. 1846: H. Ammari, H. Kang, Reconstruction of Small
Operators (2003) Inhomogeneities from Boundary Measurements (2004)
Vol. 1821: S. Teufel, Adiabatic Perturbation Theory in Vol. 1847: T.R. Bielecki, T. Bjrk, M. Jeanblanc, M.
Quantum Dynamics (2003) Rutkowski, J.A. Scheinkman, W. Xiong, Paris-Princeton
Vol. 1822: S.-N. Chow, R. Conti, R. Johnson, J. Mallet- Lectures on Mathematical Finance 2003 (2004)
Paret, R. Nussbaum, Dynamical Systems. Cetraro, Italy Vol. 1848: M. Abate, J. E. Fornaess, X. Huang, J. P. Rosay,
2000. Editors: J. W. Macki, P. Zecca (2003) A. Tumanov, Real Methods in Complex and CR Geom-
Vol. 1823: A. M. Anile, W. Allegretto, C. Ring- etry, Martina Franca, Italy 2002. Editors: D. Zaitsev, G.
hofer, Mathematical Problems in Semiconductor Physics. Zampieri (2004)
Cetraro, Italy 1998. Editor: A. M. Anile (2003) Vol. 1849: Martin L. Brown, Heegner Modules and Ellip-
tic Curves (2004)
Vol. 1824: J. A. Navarro Gonzlez, J. B. Sancho de Salas,
C Differentiable Spaces (2003) Vol. 1850: V. D. Milman, G. Schechtman (Eds.), Geomet-
ric Aspects of Functional Analysis. Israel Seminar 2002-
Vol. 1825: J. H. Bramble, A. Cohen, W. Dahmen, Mul-
2003 (2004)
tiscale Problems and Methods in Numerical Simulations,
Vol. 1851: O. Catoni, Statistical Learning Theory and
Martina Franca, Italy 2001. Editor: C. Canuto (2003)
Stochastic Optimization (2004)
Vol. 1826: K. Dohmen, Improved Bonferroni Inequal-
Vol. 1852: A.S. Kechris, B.D. Miller, Topics in Orbit
ities via Abstract Tubes. Inequalities and Identities of
Equivalence (2004)
Inclusion-Exclusion Type. VIII, 113 p, 2003.
Vol. 1853: Ch. Favre, M. Jonsson, The Valuative Tree
Vol. 1827: K. M. Pilgrim, Combinations of Complex (2004)
Dynamical Systems. IX, 118 p, 2003.
Vol. 1854: O. Saeki, Topology of Singular Fibers of Dif-
Vol. 1828: D. J. Green, Grbner Bases and the Computa- ferential Maps (2004)
tion of Group Cohomology. XII, 138 p, 2003. Vol. 1855: G. Da Prato, P.C. Kunstmann, I. Lasiecka,
Vol. 1829: E. Altman, B. Gaujal, A. Hordijk, Discrete- A. Lunardi, R. Schnaubelt, L. Weis, Functional Analytic
Event Control of Stochastic Networks: Multimodularity Methods for Evolution Equations. Editors: M. Iannelli,
and Regularity. XIV, 313 p, 2003. R. Nagel, S. Piazzera (2004)
Vol. 1830: M. I. Gil, Operator Functions and Localization Vol. 1856: K. Back, T.R. Bielecki, C. Hipp, S. Peng,
of Spectra. XIV, 256 p, 2003. W. Schachermayer, Stochastic Methods in Finance, Bres-
Vol. 1831: A. Connes, J. Cuntz, E. Guentner, N. Hig- sanone/Brixen, Italy, 2003. Editors: M. Fritelli, W. Rung-
son, J. E. Kaminker, Noncommutative Geometry, Martina galdier (2004)
Franca, Italy 2002. Editors: S. Doplicher, L. Longo (2004) Vol. 1857: M. mery, M. Ledoux, M. Yor (Eds.), Smi-
Vol. 1832: J. Azma, M. mery, M. Ledoux, M. Yor naire de Probabilits XXXVIII (2005)
(Eds.), Sminaire de Probabilits XXXVII (2003) Vol. 1858: A.S. Cherny, H.-J. Engelbert, Singular Stochas-
Vol. 1833: D.-Q. Jiang, M. Qian, M.-P. Qian, Mathemati- tic Differential Equations (2005)
cal Theory of Nonequilibrium Steady States. On the Fron- Vol. 1859: E. Letellier, Fourier Transforms of Invariant
tier of Probability and Dynamical Systems. IX, 280 p, Functions on Finite Reductive Lie Algebras (2005)
2004. Vol. 1860: A. Borisyuk, G.B. Ermentrout, A. Friedman,
Vol. 1834: Yo. Yomdin, G. Comte, Tame Geometry with D. Terman, Tutorials in Mathematical Biosciences I.
Application in Smooth Analysis. VIII, 186 p, 2004. Mathematical Neurosciences (2005)
Vol. 1835: O.T. Izhboldin, B. Kahn, N.A. Karpenko, Vol. 1861: G. Benettin, J. Henrard, S. Kuksin, Hamil-
A. Vishik, Geometric Methods in the Algebraic Theory tonian Dynamics Theory and Applications, Cetraro,
of Quadratic Forms. Summer School, Lens, 2000. Editor: Italy, 1999. Editor: A. Giorgilli (2005)
J.-P. Tignol (2004) Vol. 1862: B. Helffer, F. Nier, Hypoelliptic Estimates and
Vol. 1836: C. Nastasescu, F. Van Oystaeyen, Methods of Spectral Theory for Fokker-Planck Operators and Witten
Graded Rings. XIII, 304 p, 2004. Laplacians (2005)
Vol. 1863: H. Fhr, Abstract Harmonic Analysis of Con- Vol. 1890: M. Bunge, J. Funk, Singular Coverings of
tinuous Wavelet Transforms (2005) Toposes (2006)
Vol. 1864: K. Efstathiou, Metamorphoses of Hamiltonian Vol. 1891: J.B. Friedlander, D.R. Heath-Brown,
Systems with Symmetries (2005) H. Iwaniec, J. Kaczorowski, Analytic Number Theory,
Vol. 1865: D. Applebaum, B.V. R. Bhat, J. Kustermans, Cetraro, Italy, 2002. Editors: A. Perelli, C. Viola (2006)
J. M. Lindsay, Quantum Independent Increment Processes Vol. 1892: A. Baddeley, I. Brny, R. Schneider, W. Weil,
I. From Classical Probability to Quantum Stochastic Cal- Stochastic Geometry, Martina Franca, Italy, 2004. Editor:
culus. Editors: M. Schrmann, U. Franz (2005) W. Weil (2007)
Vol. 1866: O.E. Barndorff-Nielsen, U. Franz, R. Gohm, Vol. 1893: H. Hanmann, Local and Semi-Local Bifur-
B. Kmmerer, S. Thorbjnsen, Quantum Independent cations in Hamiltonian Dynamical Systems, Results and
Increment Processes II. Structure of Quantum Lvy Examples (2007)
Processes, Classical Probability, and Physics. Editors: M. Vol. 1894: C.W. Groetsch, Stable Approximate Evaluation
Schrmann, U. Franz, (2005) of Unbounded Operators (2007)
Vol. 1867: J. Sneyd (Ed.), Tutorials in Mathematical Bio- Vol. 1895: L. Molnr, Selected Preserver Problems on
sciences II. Mathematical Modeling of Calcium Dynamics Algebraic Structures of Linear Operators and on Function
and Signal Transduction. (2005) Spaces (2007)
Vol. 1868: J. Jorgenson, S. Lang, Posn (R) and Eisenstein Vol. 1896: P. Massart, Concentration Inequalities and
Series. (2005) Model Selection, Ecole dEt de Probabilits de Saint-
Vol. 1869: A. Dembo, T. Funaki, Lectures on Probabil- Flour XXXIII-2003. Editor: J. Picard (2007)
ity Theory and Statistics. Ecole dEt de Probabilits de Vol. 1897: R. Doney, Fluctuation Theory for Lvy
Saint-Flour XXXIII-2003. Editor: J. Picard (2005) Processes, Ecole dEt de Probabilits de Saint-Flour
Vol. 1870: V.I. Gurariy, W. Lusky, Geometry of Mntz XXXV-2005. Editor: J. Picard (2007)
Spaces and Related Questions. (2005) Vol. 1898: H.R. Beyer, Beyond Partial Differential Equa-
Vol. 1871: P. Constantin, G. Gallavotti, A.V. Kazhikhov, tions, On linear and Quasi-Linear Abstract Hyperbolic
Y. Meyer, S. Ukai, Mathematical Foundation of Turbu- Evolution Equations (2007)
lent Viscous Flows, Martina Franca, Italy, 2003. Editors: Vol. 1899: Sminaire de Probabilits XL. Editors:
M. Cannone, T. Miyakawa (2006) C. Donati-Martin, M. mery, A. Rouault, C. Stricker
Vol. 1872: A. Friedman (Ed.), Tutorials in Mathemati- (2007)
cal Biosciences III. Cell Cycle, Proliferation, and Cancer Vol. 1900: E. Bolthausen, A. Bovier (Eds.), Spin Glasses
(2006) (2007)
Vol. 1873: R. Mansuy, M. Yor, Random Times and En- Vol. 1901: O. Wittenberg, Intersections de deux
largements of Filtrations in a Brownian Setting (2006) quadriques et pinceaux de courbes de genre 1, Inter-
Vol. 1874: M. Yor, M. mery (Eds.), In Memoriam Paul- sections of Two Quadrics and Pencils of Curves of Genus
Andr Meyer - Sminaire de probabilits XXXIX (2006) 1 (2007)
Vol. 1875: J. Pitman, Combinatorial Stochastic Processes. Vol. 1902: A. Isaev, Lectures on the Automorphism
Ecole dEt de Probabilits de Saint-Flour XXXII-2002. Groups of Kobayashi-Hyperbolic Manifolds (2007)
Editor: J. Picard (2006) Vol. 1903: G. Kresin, V. Mazya, Sharp Real-Part Theo-
Vol. 1876: H. Herrlich, Axiom of Choice (2006) rems (2007)
Vol. 1877: J. Steuding, Value Distributions of L-Functions
(2007)
Vol. 1878: R. Cerf, The Wulff Crystal in Ising and Percol-
ation Models, Ecole dEt de Probabilits de Saint-Flour
XXXIV-2004. Editor: Jean Picard (2006)
Vol. 1879: G. Slade, The Lace Expansion and its Applica-
tions, Ecole dEt de Probabilits de Saint-Flour XXXIV-
2004. Editor: Jean Picard (2006)
Vol. 1880: S. Attal, A. Joye, C.-A. Pillet, Open Quantum
Systems I, The Hamiltonian Approach (2006)
Vol. 1881: S. Attal, A. Joye, C.-A. Pillet, Open Quantum
Systems II, The Markovian Approach (2006)
Vol. 1882: S. Attal, A. Joye, C.-A. Pillet, Open Quantum
Systems III, Recent Developments (2006) Recent Reprints and New Editions
Vol. 1883: W. Van Assche, F. Marcelln (Eds.), Orthogo-
nal Polynomials and Special Functions, Computation and Vol. 1618: G. Pisier, Similarity Problems and Completely
Application (2006) Bounded Maps. 1995 2nd exp. edition (2001)
Vol. 1884: N. Hayashi, E.I. Kaikina, P.I. Naumkin, Vol. 1629: J.D. Moore, Lectures on Seiberg-Witten
I.A. Shishmarev, Asymptotics for Dissipative Nonlinear Invariants. 1997 2nd edition (2001)
Equations (2006) Vol. 1638: P. Vanhaecke, Integrable Systems in the realm
Vol. 1885: A. Telcs, The Art of Random Walks (2006) of Algebraic Geometry. 1996 2nd edition (2001)
Vol. 1886: S. Takamura, Splitting Deformations of Dege- Vol. 1702: J. Ma, J. Yong, Forward-Backward Stochas-
nerations of Complex Curves (2006) tic Differential Equations and their Applications. 1999
Vol. 1887: K. Habermann, L. Habermann, Introduction to Corr. 3rd printing (2005)
Symplectic Dirac Operators (2006) Vol. 830: J.A. Green, Polynomial Representations of
Vol. 1888: J. van der Hoeven, Transseries and Real Differ- GLn , with an Appendix on Schensted Correspondence
ential Algebra (2006) and Littelmann Paths by K. Erdmann, J.A. Green and
Vol. 1889: G. Osipenko, Dynamical Systems, Graphs, and M. Schocker 1980 2nd corr. and augmented edition
Algorithms (2006) (2007)