EQUIVALENTS OF THE RIEMANN HYPOTHESIS

Volume One: Arithmetic Equivalents

The Riemann hypothesis (RH) is perhaps the most important outstanding problem
in mathematics. This two-volume text presents the main known equivalents to RH
using analytic and computational methods. The books are gentle on the reader with
definitions repeated, proofs split into logical sections, and graphical descriptions of
the relations between different results. They also include extensive tables,
supplementary computational tools, and open problems suitable for research.
Accompanying software is free to download.
These books will interest mathematicians who wish to update their knowledge,
graduate and senior undergraduate students seeking accessible research problems in
number theory, and others who want to explore and extend results computationally.
Each volume can be read independently.
Volume 1 presents classical and modern arithmetic equivalents to RH, with some
analytic methods. Volume 2 covers equivalences with a strong analytic orientation,
supported by an extensive set of appendices containing fully developed proofs.

Encyclopedia of Mathematics and Its Applications

This series is devoted to significant topics or themes that have wide application in
mathematics or mathematical science and for which a detailed development of the
abstract theory is less important than a thorough and concrete exploration of the
implications and applications.

Books in the Encyclopedia of Mathematics and Its Applications cover their

subjects comprehensively. Less important results may be summarized as exercises
at the ends of chapters. For technicalities, readers can be referred to the
bibliography, which is expected to be comprehensive. As a result, volumes are
encyclopedic references or manageable guides to major subjects.
 Encyclopedia of Mathematics and Its Applications

All the titles listed below can be obtained from good booksellers or from Cambridge
University Press. For a complete series listing visit
www.cambridge.org/mathematics.
119 M. Deza and M. Dutour Sikirić Geometry of Chemical Graphs
120 T. Nishiura Absolute Measurable Spaces
121 M. Prest Purity, Spectra and Localisation
122 S. Khrushchev Orthogonal Polynomials and Continued Fractions
123 H. Nagamochi and T. Ibaraki Algorithmic Aspects of Graph Connectivity
124 F. W. King Hilbert Transforms I
125 F. W. King Hilbert Transforms II
126 O. Calin and D.-C. Chang Sub-Riemannian Geometry
127 M. Grabisch et al. Aggregation Functions
128 L. W. Beineke and R. J. Wilson (eds.) with J. L. Gross and T. W. Tucker Topics in Topological
Graph Theory
129 J. Berstel, D. Perrin and C. Reutenauer Codes and Automata
130 T. G. Faticoni Modules over Endomorphism Rings
131 H. Morimoto Stochastic Control and Mathematical Modeling
132 G. Schmidt Relational Mathematics
133 P. Kornerup and D. W. Matula Finite Precision Number Systems and Arithmetic
134 Y. Crama and P. L. Hammer (eds.) Boolean Models and Methods in Mathematics, Computer
Science, and Engineering
135 V. Berthé and M. Rigo (eds.) Combinatorics, Automata and Number Theory
136 A. Kristály, V. D. Rădulescu and C. Varga Variational Principles in Mathematical Physics,
Geometry, and Economics
137 J. Berstel and C. Reutenauer Noncommutative Rational Series with Applications
138 B. Courcelle and J. Engelfriet Graph Structure and Monadic Second-Order Logic
139 M. Fiedler Matrices and Graphs in Geometry
140 N. Vakil Real Analysis through Modern Infinitesimals
141 R. B. Paris Hadamard Expansions and Hyperasymptotic Evaluation
142 Y. Crama and P. L. Hammer Boolean Functions
143 A. Arapostathis, V. S. Borkar and M. K. Ghosh Ergodic Control of Diffusion Processes
144 N. Caspard, B. Leclerc and B. Monjardet Finite Ordered Sets
145 D. Z. Arov and H. Dym Bitangential Direct and Inverse Problems for Systems of Integral and
Differential Equations
146 G. Dassios Ellipsoidal Harmonics
147 L. W. Beineke and R. J. Wilson (eds.) with O. R. Oellermann Topics in Structural Graph Theory
148 L. Berlyand, A. G. Kolpakov and A. Novikov Introduction to the Network Approximation Method
for Materials Modeling
149 M. Baake and U. Grimm Aperiodic Order I: A Mathematical Invitation
150 J. Borwein et al. Lattice Sums Then and Now
151 R. Schneider Convex Bodies: The Brunn–Minkowski Theory (Second Edition)
152 G. Da Prato and J. Zabczyk Stochastic Equations in Infinite Dimensions (Second Edition)
153 D. Hofmann, G. J. Seal and W. Tholen (eds.) Monoidal Topology
154 M. Cabrera Garcı́a and Á. Rodrı́guez Palacios Non-Associative Normed Algebras I: The
Vidav–Palmer and Gelfand–Naimark Theorems
155 C. F. Dunkl and Y. Xu Orthogonal Polynomials of Several Variables (Second Edition)
156 L. W. Beineke and R. J. Wilson (eds.) with B. Toft Topics in Chromatic Graph Theory
157 T. Mora Solving Polynomial Equation Systems III: Algebraic Solving
158 T. Mora Solving Polynomial Equation Systems IV: Buchberger Theory and Beyond
159 V. Berthé and M. Rigo (eds.) Combinatorics, Words and Symbolic Dynamics
160 B. Rubin Introduction to Radon Transforms: With Elements of Fractional Calculus and Harmonic
Analysis
161 M. Ghergu and S. D. Taliaferro Isolated Singularities in Partial Differential Inequalities
162 G. Molica Bisci, V. Radulescu and R. Servadei Variational Methods for Nonlocal Fractional
Problems
163 S. Wagon The Banach–Tarski Paradox (Second Edition)
164 K. Broughan Equivalents of the Riemann Hypothesis I: Arithmetic Equivalents
165 K. Broughan Equivalents of the Riemann Hypothesis II: Analytic Equivalents
166 M. Baake and U. Grimm Aperiodic Order II: Representation Theory and the Zelmanov Approach
 Encyclopedia of Mathematics and Its Applications

Equivalents of the Riemann Hypothesis

Volume One: Arithmetic Equivalents

KEVIN BROUGHAN
University of Waikato, New Zealand
 University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
4843/24, 2nd Floor, Ansari Road, Daryaganj, Delhi – 110002, India
79 Anson Road, #06-04/06, Singapore 079906

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781107197046
DOI: 10.1017/9781108178228

c Kevin Broughan 2017
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2017
Printed in the United Kingdom by Clays, St Ives plc
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Broughan, Kevin A. (Kevin Alfred), 1943– author.
Title: Equivalents of the Riemann hypothesis / Kevin Broughan,
University of Waikato, New Zealand.
Description: Cambridge : Cambridge University Press, 2017– |
Series: Encyclopedia of mathematics and its applications ; 164 |
Includes bibliographical references and index. Contents: volume 1. Arithmetic equivalents
Identifiers: LCCN 2017034308 | ISBN 9781107197046 (hardback : alk. paper : v. 1)
Subjects: LCSH: Riemann hypothesis.
Classification: LCC QA246 .B745 2017 |
DDC 512.7/3–dc23 LC record available at https://lccn.loc.gov/2017034308
ISBN – 2 Volume Set 978-1-108-29078-4 Hardback
ISBN – Volume 1 978-1-107-19704-6 Hardback
ISBN – Volume 2 978-1-107-19712-1 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Dedicated to Jackie, Jude and Beck
RH is a precise statement, and in one sense what it means is clear, but what
it is connected with, what it implies, where it comes from, can be very un-
obvious.
Martin Huxley
 Contents for Volume One

Contents for Volume Two page x

List of Illustrations xiv
List of Tables xvi
Preface for Volume One xvii
List of Acknowledgements xxi

1 Introduction 1
1.1 Chapter Summary 1
1.2 Early History 1
1.3 Volume One Summary 8
1.4 Notation 12
1.5 Background Reading 13
1.6 Unsolved Problems 14

2 The Riemann Zeta Function 15

2.1 Introduction 15
2.2 Basic Properties 16
2.3 Zero-Free Regions 21
2.4 Landau’s Zero-Free Region 25
2.5 Zero-Free Regions Summary 29
2.6 The Product Over Zeta Zeros 30
2.7 Unsolved Problems 39

3 Estimates 40
3.1 Introduction 40
3.2 Constructing Tables of Bounds for ψ(x) 41
3.3 Exact Verification Using Computation 51
3.4 Estimates for θ(x) 54
3.5 More Estimates 65
3.6 Unsolved Problems 67

vii
viii Contents for Volume One

4 Classical Equivalences 68
4.1 Introduction 68
4.2 The Prime Number Theorem and Its RH Equivalences 69
4.3 Oscillation Theorems 81
4.4 Errors in Arithmetic Sums 88
4.5 Unsolved Problems 93

5 Euler’s Totient Function 94

5.1 Introduction 94
5.2 Estimates for Euler’s Function ϕ(n) 98
5.3 Preliminary Results With RH True 110
5.4 Further Results With RH True 123
5.5 Preliminary Results With RH False 130
5.6 Nicolas’ First Theorem 135
5.7 Nicolas’ Second Theorem 137
5.8 Unsolved Problems 142

6 A Variety of Abundant Numbers 144

6.1 Introduction 144
6.2 Superabundant Numbers 147
6.3 Colossally Abundant Numbers 153
6.4 Estimates for x2 () 161
6.5 Unsolved Problems 163

7 Robin’s Theorem 165

7.1 Introduction 165
7.2 Ramanujan’s Theorem Assuming RH 169
7.3 Preliminary Lemmas With RH True 174

7.4 Bounding p≤x (1 − p−2 ) From Above With RH True 180
7.5 Bounding loglog N From Below With RH True 184
7.6 Proof of Robin’s Theorem With RH True 186
7.7 An Unconditional Bound for σ(n)/n 188
7.8 Bounding loglog N From Above Without RH 190
7.9 A Lower Bound for σ(n)/n With RH False 191
7.10 Lagarias’ Formulation of Robin’s Criterion 193
7.11 Unconditional Results for Lagarias’ Formulation 196
7.12 Unitary Divisor Sums 197
7.13 Unsolved Problems 198

8 Numbers That Do Not Satisfy Robin’s Inequality 200

8.1 Introduction 200
8.2 Hardy–Ramanujan Numbers 202
8.3 Integers Not Divisible by the Fifth Power of Any Prime 208
 Contents for Volume One ix

8.4 Integers Not Divisible by the Seventh Power of Any Prime 211
8.5 Integers Not Divisible by the 11th Power of Any Prime 214
8.6 Unsolved Problems 217

9 Left, Right and Extremely Abundant Numbers 218

9.1 Introduction 218
9.2 Grönwall’s Theorem 220
9.3 Further Preliminary Results 223
9.4 Riemann Hypothesis Equivalences 225
9.5 Comparing Colossally and Left Abundant Numbers 232
9.6 Extremely Abundant Numbers 235
9.7 Unsolved Problems 235

10 Other Equivalents to the Riemann Hypothesis 236

10.1 Introduction 236
10.2 Shapiro’s Criterion 239
10.3 Farey Fractions 241
10.4 Redheffer Matrix 247
10.5 Divisibility Graph 250
10.6 Dirichlet Eta Function 252
10.7 The Derivative of ζ(s) 253
10.8 A Zeta-Related Inequality 256
10.9 The Real Part of the Logarithmic Derivative of ξ(s) 259
10.10 The Order of Elements of the Symmetric Group 271
10.11 Hilbert–Pólya Conjecture 282
10.12 Epilogue 285
10.13 Unsolved Problems 286

Appendix A Tables 287

A.1 Extremely Abundant Numbers 287
A.2 Small Numbers Not Satisfying Robin’s Inequality 288
A.3 Superabundant Numbers 289
A.4 Colossally Abundant Numbers 290
A.5 Primes to Make Colossally Abundant Numbers 291
A.6 Small Numbers Satisfying Nicolas’ Reversed Inequality 292
A.7 Heights of Integers 293
A.8 Maximum Order of an Element of the Symmetric Group 293

Appendix B RHpack Mini-Manual 294

B.1 Introduction 294
B.2 RHpack Functions 296

References 313
Index 321
 Contents for Volume Two

Contents for Volume One page xi

List of Illustrations xiv
List of Tables xvi
Preface for Volume Two xvii
List of Acknowledgements xxi

1 Introduction 1
1.1 Why This Study? 1
1.2 Summary of Volume Two 2
1.3 How to Read This Book 7

2 Series Equivalents 8
2.1 Introduction 8
2.2 The Riesz Function 10
2.3 Additional Properties of the Riesz Function 14
2.4 The Series of Hardy and Littlewood 15
2.5 A General Theorem for a Class of Entire Functions 16
2.6 Further Work 22

3 Banach and Hilbert Space Methods 23

3.1 Introduction 23
3.2 Preliminary Definitions and Results 25
3.3 Beurling’s Theorem 29
3.4 Recent Developments 35

4 The Riemann Xi Function 37

4.1 Introduction 37
4.2 Preliminary Results 40
4.3 Monotonicity of |ξ(s)| 49

x
 Contents for Volume Two xi

4.4 Positive Even Derivatives 51

4.5 Li’s Equivalence 54
4.6 More Recent Results 59

5 The De Bruijn–Newman Constant 62

5.1 Introduction 62
5.2 Preliminary Definitions and Results 66
5.3 A Region for Ξλ (z) With Only Real Zeros 69
5.4 The Existence of Λ 77
5.5 Improved Lower Bounds for Λ 77
5.5.1 Lehmer’s Phenomenon 78
5.5.2 The Differential Equation Satisfied by H(t, z) 81
5.5.3 Finding a Lower Bound for ΛC Using Lehmer Pairs 87
5.6 Further Work 92

6 Orthogonal Polynomials 93
6.1 Introduction 93
6.2 Definitions 94
6.3 Orthogonal Polynomial Properties 96
6.4 Moments 99
6.5 Quasi-Analytic Functions 104
6.6 Carleman’s Inequality 106
6.7 Riemann Zeta Function Application 113
6.8 Recent Work 116

7 Cyclotomic Polynomials 117

7.1 Introduction 117
7.2 Definitions 118
7.3 Preliminary Results 119
7.4 Riemann Hypothesis Equivalences 124
7.5 Further Work 126

8 Integral Equations 127

8.1 Introduction 127
8.2 Preliminary Results 129
8.3 The Method of Sekatskii, Beltraminelli and Merlini 133
8.4 Salem’s Equation 139
8.5 Levinson’s Equivalence 142

9 Weil’s Explicit Formula, Inequality and Conjectures 150

9.1 Introduction 150
9.2 Definitions 152
9.3 Preliminary Results 152
9.4 Weil’s Explicit Formula 154
xii Contents for Volume Two

9.5 Weil’s Inequality 159

9.6 Bombieri’s Variational Approach to RH 166
9.7 Introduction to the Weil Conjectures 173
9.8 History of the Weil Conjectures 174
9.9 Finite Fields 176
9.10 The Weil Conjectures for Varieties 178
9.11 Elliptic Curves 178
9.12 Weil Conjectures for Elliptic Curves – Preliminary Results 182
9.13 Proof of the Weil Conjectures for Elliptic Curves 186
9.14 General Curves Over Fq and Applications 188
9.15 Return to the Explicit Formula 190
9.16 Weil’s Commentary on his 1952 and 1972 Papers 192

10 Discrete Measures 193

10.1 Introduction 193
10.2 Definitions 194
10.3 Preliminary Results 195
10.4 A Mellin-Style Transform 197
10.5 Verjovsky’s Theorems 200
10.6 Historical Development of Non-Euclidean Geometry 206
10.7 The Hyperbolic Upper Half Plane H 208
10.8 The Groups PSL(2, R) and PSL(2, Z) 209
10.9 Eisenstein Series 211
10.10 Zagier’s Horocycle Equivalence 216
10.11 Additional Results 219

11 Hermitian Forms 221

11.1 Introduction 221
11.2 Definitions 223
11.3 Distributions 226
11.4 Positive Definite 228
11.5 The Restriction to C(a) for All a > 0 231
11.6 
Properties of K(a) and K(a) 236
11.7 Matrix Elements 242
√
11.8 An Explicit Example With a = log 2 247
11.9 Lemmas for Yoshida’s Main Theorem 258
11.10 Hermitian Forms Lemma 260
11.11 Yoshida’s Main Theorem 269
11.12 The Restriction to K(a) for All a > 0 270

12 Dirichlet L-Functions 274

12.1 Introduction 274
12.2 Definitions 277
 Contents for Volume Two xiii

12.3 Properties of L(s, χ) 283

12.4 The Non-Vanishing of L(1, χ) 284
12.5 Zero-Free Regions and Siegel Zeros 288
12.6 Preliminary Results for Titchmarsh’s Criterion 295
12.7 Titchmarsh’s GRH Equivalence 296
12.8 Preliminary Results for Gallagher’s Theorem 298
12.9 Gallagher’s Theorems 302
12.10 Applications of Gallagher’s Theorems 307
12.11 The Bombieri–Vinogradov Theorem 311
12.12 Applications of Bombieri–Vinogradov’s Theorem 323
12.13 Generalizations and Developments for Bombieri–Vinogradov 326
12.14 Conjectures 327

13 Smooth Numbers 332

13.1 Introduction 332
13.2 The Dickman Function 335
13.3 Preliminary Lemmas for Hildebrand’s Equivalence 346
13.4 Riemann Hypothesis Equivalence 349
13.5 Further Work 357

14 Epilogue 359

Appendix A Convergence of Series 361

Appendix B Complex Function Theory 363
Appendix C The Riemann–Stieltjes Integral 377
Appendix D The Lebesgue Integral on R 381
Appendix E The Fourier Transform 388
Appendix F The Laplace Transform 405
Appendix G The Mellin Transform 409
Appendix H The Gamma Function 418
Appendix I The Riemann Zeta Function 425
Appendix J Banach and Hilbert Spaces 442
Appendix K Miscellaneous Background Results 451
Appendix L GRHpack Mini-Manual 459
L.1 Introduction 459
L.1.1 Installation 459
L.1.2 About This Mini-Manual 460
L.2 GRHpack Functions 461

References 473
Index 485
 Illustrations

1.1 Euclid (about 325–265 BCE). page 2

1.2 L. Euler, 1707–1783. 3
1.3 J. C. F. Gauss, 1777–1855. 4
1.4 J. L. Dirichlet, 1805–1859. 4
1.5 B. Riemann, 1826–1866. 5
1.6 David Hilbert, 1862–1943. 7
1.7 Some relationships between chapters. 12
2.1 The contours of |ζ(s)|. 17
2.2 The flow ṡ = ζ(s). 18
2.3 The flow ṡ = ξ(s). 19
2.4 The function −S (t) for 1000 ≤ t ≤ 1040. 24
2.5 The function T (t) on [950, 1060] with a phantom zero at t = 1010. 25
3.1 The values of b for 10 ≤ b ≤ 100 and 1 ≤ m ≤ 5. 52
3.2 The values of b for 100 ≤ b ≤ 1000 and 3 ≤ m ≤ 7. 52
3.3 The values of b for 103 ≤ b ≤ 104 and 3 ≤ m ≤ 7. 53
3.4 A plot of b, where x = eb , vs log (x) for 103 ≤ b ≤ 104 . 60
3.5 A plot of b, where x = e , vs log (x) for 10 ≤ b ≤ 10 .
b 5 6
60
4.1 J. E. Littlewood (1885–1977).
√ 87
4.2 The function M(x)/ x for 1 ≤ x ≤ 105 . 91
5.1 Jean-Louis Nicolas. 95
5.2 Some relationships between results in Chapter 5. 98
5.3 Some relationships between results in Chapter 5. 99
5.4 The function f (x). 113
5.5 The function H(x). 116
5.6 The function F1/2 (x). 120
5.7 Some relationships between results in Chapter 5. 126
5.8 The relation c(n) ≤ c(Nk ) for k = 3. 129
5.9 Proportion of solutions to eγ loglog n < n/ϕ(n) for 1 ≤ n ≤ 105 . 137
5.10 The function Δk for 1 ≤ k ≤ 4000. 137

xiv
D
 List of Illustrations xv

5.11 The sequence c(n). 138

5.12 The sequence c(Nk ) for 120 568 ≤ k ≤ 106 . 141
6.1 Paul Erdős, 1913–1996. 145
6.2 The functions F(3, α) and F(2, α) showing F(3, α) < F(2, α). 154
6.3 The function n → σ(n)/n1+4 . 157
6.4 The function n → σ(n)/n1+0.1 . 158
7.1 The values of σ(n)/(neγ loglog n) for 5041 ≤ n ≤ 6041. 168
7.2 Timothy Trudgian. 169
7.3 Relationships between some of the results in Chapter 7. 170
7.4 Further relationships between some of the results in Chapter 7. 171
7.5 Additional relationships between some of the results in Chapter 7. 171
7.6 Further relationships between some of the results in Chapter 7. 172
7.7 Jeffrey Lagarias. 194
7.8 The values of σ(n)/(Hn + exp(Hn ) log(Hn )) for 2 ≤ n ≤ 1000. 196
8.1 YoungJu Choie. 201
9.1 Dependences between some results in this chapter. 220
9.2 Further dependences between some results in this chapter. 221
9.3 More dependences between
√ some results in this chapter. 221
10.1 The Farey sums over n with 1 ≤ n ≤ 100. 248
10.2 Individual terms in the Farey sum for n = 101. 248
10.3 The divisibility graph for n = 6. 250
10.4 The first 300 imaginary coordinate gaps between zeros of ζ(s). 267
10.5 The function S (T ) on the domain [400, 450]. 268
10.6 Values of the function h(n) excluding
 prime powers. 274
10.7 Values of the function log g(n)/ n log n for 3 ≤ n ≤ 1000. 280
 Tables

2.1 Some improvements for Backlund’s zero counting estimate. page 20

2.2 Zeros on the critical line and corresponding heights. 22
2.3 Examples for finding values of R. 29
3.1 Upper bounds used for km (0), 1 ≤ m ≤ 28. 50
3.2 Values of b, m and b . 50
3.3 More values of b, m and b . 51
3.4 Values of b = log x, and rounded log (x) for R = 8 and R = 18. 59
3.5 Larger values of b and log (x). 59
6.1 Indexing for the first eight colossally abundant numbers. 158
7.1 Evaluation of η(α, a, b), given α, a and b. 185
8.1 Values of n1 (t) and Nn1 (t) . 217
10.1The first 20 Riemann zeros and gaps. 266
A.1 The first seven extremely abundant numbers. 287
A.2 Numbers n ≤ 5041 with σ(n)/n ≥ eγ loglog n. 288
A.3 The first 31 superabundant numbers. 289
A.4 The first 30 colossally abundant numbers. 290
A.5 Initial values of h(n); compare with item A008475 in
OEIS. 293
A.6 Initial values of g(n); compare with item A000793 in
OEIS. 293

xvi
 Preface

Why have these two volumes on equivalences to the Riemann hypothesis

been written? Many would say that the Riemann hypothesis (RH) is the most
noteworthy problem in all of mathematics. This is not only because of its
relationship to the distribution of prime numbers, the fundamental building
blocks of arithmetic, but also because there exist a host of related conjectures
that will be resolved if RH is proved to be true and which will be proved
to be false if the converse is demonstrated. These are the RH equivalences.
The lists of equivalent conjectures have continued to grow ever since the
hypothesis was first enunciated, over 150 years ago.
The many attacks on RH that have been reported, the numerous failed
attempts, and the efforts of the many whose work has remained obscure,
have underlined the problem’s singular nature. So too has its mythology.
The great English number theorist, Godfrey Hardy, wrote a postcard to
Harald Bohr while returning to Cambridge from Denmark in rough weather
that read: “Have proof of RH. Postcard too short for proof.” He didn’t
believe in a God, but was certain he would not be allowed to drown with
his name associated with an infamous missing proof. David Hilbert, the
renowned German mathematician, was once asked, “If you were to die and be
revived after five hundred years, what would you then do?” Hilbert replied
that he would ask “Has someone proved the Riemann hypothesis?” More
recently, towards the end of the twentieth century, Enrico Bombieri, an Italian
mathematician at the Institute for Advanced Study, Princeton, issued a joke
email announcing the solution of RH by a young physicist, on 1 April of
course!
There are several ways in which the truth of the hypothesis has been
supported but not proved. These have included increasing the finite range
of values T > 0 such that the imaginary part of all complex zeros of ζ(s) up
to T all have real part 12 [68], increasing the lower bound for the proportion

xvii
xviii Preface

of zeros that are on the critical line s = 12 [40], and increasing the size of the
region in the complex plane where ζ(s) can be proved to be non-zero [63].
This volume includes a detailed account of some recent work that
takes a different approach. It is based on inequalities involving some
simple accessible arithmetic functions. Broadly outlined RH implies that
an inequality is true for all integers, or all integers sufficiently large, or all
integers of a particular type. If RH fails, there is an integer in the given
range, or of the given type, for which the inequality fails, so the truth of
the inequality is equivalent to the truth of RH. Progress under this approach
is made whenever the nature of any counterexample is shown to be more
restricted than previously demonstrated. The reader may also wish to consult
the introductions to Chapters 3 to 7 for further details.
The relatively recent work depends critically on a range of explicit
estimates for arithmetic functions. These have been derived using greater
computing power than was available in the 1940s and 1960s when these
sorts of estimates were first published. In many cases the details are included,
along with simplified presentations.
Also included are a range of other equivalences to RH, some by now
classical. The more recent work depends on these classical equivalences for
both the results and techniques, so it is useful to set both out explicitly. It also
shows how some equivalences are more fundamental than others. This is not
to suggest new equivalences are easy consequences of older established ones,
even though this may be true in some cases.
The aim of these volumes is to give graduate students and number theory
researchers easy access to these methods and results in order that they might
build on them. To this end, complete proofs have been included wherever
possible, so readers might judge for themselves their depth and crucial steps.
To provide context, a range of additional equivalences has been included in
this volume, some of which are arithmetical and some more analytic. An
intuitive background for some of the functions employed is also included
in the form of graphical representations. Numerical calculations have been
reworked, and values different from those found in the literature have often
been arrived at.
To aid the reader, definitions are often repeated and major steps in proofs
are numbered to give a clear indication of the main parts and allow for easy
proof internal referencing. When possible, errors in the literature have been
corrected. Where a proof has not been verified, either because this author
was not able to fill gaps in the argument, or because it was incorrect, it
has not been included. There is a website for errata and corrigenda, and
readers are encouraged to communicate with the author in this regard at
[email protected]. The website is linked to the author’s homepage:
www.math.waikato.ac.nz/∼kab.
 Preface xix

Also linked to this website is a suite of MathematicaTM programs, called

RHpack, related to the material in this volume, which is available for
free download. Instructions on how to download the software are given in
Appendix B.
The two volumes are distinct, with a small amount of overlap. This volume,
Volume One, has an arithmetic orientation, with some analytic methods,
especially those relying on the manipulation of inequalities. The equivalences
found here are the Möbius mu estimate of Littlewood, the explicit ψ(x)
function estimate of Schoenfeld, the Liouville λ(n) limit criterion of Landau,
two Euler totient function criteria of Nicolas, the sum of divisors inequality
of Ramanujan and Robin and its reformulation by Lagarias, the criterion of
Caveney, Nicolas and Sondow based on so-called “extraordinary numbers”,
the criterion of Nazardonyavi and Yakubovich based on extremely abundant
numbers, the estimate of Shapiro that uses the integral of ψ(x), the Franel–
Landau Farey fraction criterion, the divisibility matrix criterion of Redheffer,
the Levinson–Montgomery criterion that uses counts of the zeros of the
derivative ζ  (s), the inequality of Spira relating values of zeta at s and 1 − s,
the self-adjoint operator criteria of Hilbert and Pólya, and the criterion of
Lagarias and Garunkstis based on the real part of the logarithmic derivative
of ξ(s). In addition, Volume One has criteria based on the divisibility matrix
of Redheffer and a closely related graph, the Dirichlet eta function, and an
estimate for the size of the maximum order of an element of the symmetric
group.
The Appendix includes tables of the numbers that appear in some of the
equivalences and a mini-manual for RHpack.
Volume Two [32] contains equivalences with a strong analytic orientation.
To support these, there is an extensive set of appendices containing
fully developed proofs. The equivalences set out are named Amoroso,
Hardy–Littlewood, Báez-Duarte, Beurling, Bombieri, Bombieri–Lagarias,
de Bruijn–Newman, Cardon–Roberts, Hildebrand, Levinson, Li, Riesz,
Sekatskii–Beltraminelli–Merlini, Salem, Sondow–Dumitrescu, Verjovsky,
Weil, Yoshida and Zagier. For summary details, see the Preface for Volume
Two and Chapter 1 of Volume Two. In addition, Bombieri’s proof of Weil’s
explicit formula, a discussion of the Weil conjectures and a proof of the
conjectures for elliptic curves are included.
In the case of the general Riemann hypothesis (GRH) for Dirichlet L-
functions, in Volume Two the Titchmarsh criterion is given, as well as proofs
of the Bombieri–Vinogradov and Gallagher theorems and a range of their
applications. There is a small supporting Mathematica package, GRHpack.
The set of appendices for Volume Two gives comprehensive statements
and proofs of the special results that are needed to derive the equivalences in
that volume. In addition, there is a GRHpack mini-manual.
xx Preface

A note concerning the cover figure. This represents integral paths for the
flow ṡ = ζ(s) in a small region of the upper complex plane, rotated and
reflected in σ = 0. It was produced using an interactive program written by
the author and Francis Kuo in Java . It includes three critical zeros and three
trivial zeros.
Many people have assisted with the development and production of
these volumes. Without their help and support, the work would not have
been possible, and certainly not completed in a reasonable period of time.
They include Sir Michael Berry, Enrico Bombieri, Jude Broughan, George
Csordas, Daniel Delbourgo, Tomás Garcia Ferrari, Pat Gallagher, Adolf
Hildebrand, Geoff Holmes, Stephen Joe, Jeff Lagarias, Wayne Smith, Tim
Trudgian, John Turner and Michael Wilson. The support of the University of
Waikato and especially its Faculty of Computing and Mathematical Sciences
and Department of Mathematics and Statistics has been absolutely essential.
Cambridge University Press has also provided much encouragement and
support, especially Roger Astley and Clare Dennison. Last, but not least, I
am grateful for my family’s belief in me and support of my work.
Kevin Broughan
December 2016
 Acknowledgements

The author gratefully acknowledges the following sources and/or permissions for the non-
exclusive use of copyrighted material.

Euclid: Figure 1.1, being an excerpt from Raphael’s fresco “School of Athens”, Visions of
America/Superstock/Getty Images.
L. Euler: Figure 1.2, Apic/Hulton Archive/Getty Images.
C. F. Gauss: Figure 1.3, Bettmann/Getty Images.
J. L. Dirichlet: Figure 1.4, Stringer/Hulton Archive/Getty Images.
B. Riemann: Figure 1.5, Author: Konrad Jacobs. Source: Archives of the Mathematisches
Forschungsinstitut Oberwolfach.
D. Hilbert: Figure 1.6, Ullstein bild/Getty Images.
Zeta flow: Figure 2.2, being figure 2, p. 990, of K. A. Broughan and A. R. Barnett, The
holomorphic flow of the Riemann zeta function, Mathematics of Computation 73 (2004),
287–1004. Permission of the American Mathematical Society.
Xi flow: Figure 2.3, being figure 1, p. 1274, of K. A. Broughan, The holomorphic flow of
Riemann’s function ξ(s), Nonlinearity 18 (2005), 1289–1294. Copyright c IOP Publishing
and London Mathematical Society. Permission of the American Mathematical Society. All
rights reserved.
J. E. Littlewood: Figure 4.1, used by permission of the Master and Fellows of Trinity College
Cambridge.
J.-L. Nicolas: Figure 5.1, used by permission of J.-L. Nicolas.
P. Erdős: Figure 6.1, Author: Kay Piene. Source: Ragni Piene and Archives of the
Mathematisches Forschungsinstitut Oberwolfach.
T. S. Trudgian: Figure 7.2, used by permission of T. S. Trudgian.
J. C. Lagarias: Figure 7.7, used by permission of J. C. Lagarias.
Y. Choie: Figure 8.1, used by permission of Y. Choie.

Section 8.5 was derived from a previously published article: K. A. Broughan and T. Trudgian,
Robin’s inequality for 11-free integers, Integers 15 (2015), A12 (5pp.).

xxi
 1
Introduction

1.1 Chapter Summary

This chapter is discursive. It begins with an overview of the early history
of the Riemann hypothesis (RH) and the evolution of ideas relating to the
Ramanujan–Robin inequality. Then in Section 1.3 there is a summary of
the contents of the entire volume, first in brief and then in more detail. The
section also describes the tables in Appendix A and the software RHpack in
Appendix B.
There is a section on notational conventions and special notations, most of
which are quite standard. A guide to the reader and two problems complete
the chapter.

1.2 Early History

Here the main players in the evolution of the Riemann hypothesis are noted:
Euclid, Euler, Gauss, Dirichlet and, last but not least, Riemann himself. The
first is Euclid of Alexandria (Figure 1.1) who lived around 300 BCE. His
Elements includes a proof that there are an infinite number of primes, and
that they are the fundamental building blocks of numbers, through the unique
factorization of integers. How are the primes distributed? On the face of it,
the only pattern appears to be that all are odd, except 2. Do they appear
completely at random, or are they uniformly distributed in some sense?
Don Zagier gives a good description of this random/uniform dichotomy:
“The first fact is the prime numbers belong to the most arbitrary and ornery
objects studied by mathematicians: they grow like weeds among the natural
numbers, seeming to obey no other law than that of chance, and nobody can
predict where the next one will sprout.”
“The second fact is even more astonishing, for it states just the opposite:
that the prime numbers exhibit stunning regularity, that there are laws
governing their behaviour, and that they obey these laws with almost military
precision.”

1
2 Introduction

Figure 1.1 Euclid (about 325–265 BCE).

Euler (1707–1783; Figure 1.2) was an amazing mathematician, who

wrote a text on calculus, including the first treatment of trigonometric
functions, and invented many parts of mathematics that are important today,
including the calculus of variations, graph theory and divergent series. In
number theory, quadratic reciprocity and Euler products are two of his many
contributions, as well as his extensive work on the zeta function. In fact, Euler
gave birth to the Riemann zeta function, writing down its definition for the
first time. He studied the following sums and product:

1 1 1 π2
ζ(2) = 1 + 2
+ 2 + 2 + ··· = ,
2 3 4 6
1 1 1 π4
ζ(4) = 1 + 4 + 4 + 4 + · · · = ,
2 3 4 90
1 1 1
ζ(s) = 1 + s + s + s + · · · ,
2 3 4
 1
−1
ζ(s) = 1− s .
p
p

For Euler, the variable s was an integer, and the product over all primes. He
tried in vain to find a closed expression for ζ(3) having the same style as that
for ζ(2) and ζ(4).
 1.2 Early History 3

Figure 1.2 L. Euler, 1707–1783.

The next main player was Gauss (1777–1855; Figure 1.3), although
he does not appear to have used the zeta function. He was the giant of
nineteenth-century mathematics and physics. For example, he used calculus
to compute correctly the position of Ceres after it had passed behind the
Sun. Although he spent most of his life in his observatory in Göttingen in
Germany, he contributed to statistics, non-Euclidean geometry, curvature,
geodesy, electromagnetism and complex numbers, as well as to number
theory, for which he had an abiding passion.
Gauss was given a book of log tables, at around the age of 14, that included
a table of primes. He extended the table, counting the number of primes up to
a real positive variable x, now called π(x). He considered the average number
of primes in each interval of numbers [1, 2, 3, . . . , N], and in this way arrived
at his prime number conjecture, an inspired guess that was proved about 100
years later:
x π(x)
π(x) ∼ which means → 1, x → ∞.
log x x/log x
It was this conjecture that inspired many mathematicians during the
nineteenth century.
Dirichlet (1805–1859; Figure 1.4) was a fine teacher and had a great
influence on Bernhard Riemann, who attended his number theory lectures
in Berlin during 1847–1849. According to Jacobi, Dirichlet was not only
creative, but knew how to make a robust proof: “Only Dirichlet, not I, nor
4 Introduction

Figure 1.3 J. C. F. Gauss, 1777–1855.

Figure 1.4 J. L. Dirichlet, 1805–1859.

 1.2 Early History 5

Cauchy, not Gauss, knows what a perfectly rigorous proof is, but we learn it
only from him. When Gauss says he has proved something it is very likely.
When Cauchy says it, it’s a fifty-fifty bet. When Dirichlet says it, it is certain.”
Dirichlet showed that there were an infinite number of primes in arithmetic
progressions whenever the step and initial value are coprime. He used groups,
characters and zeta functions. He was a wonderfully inspiring lecturer, and
joined Riemann in Göttingen in 1855.
According to Felix Klein: “Riemann was bound to Dirichlet by the
strong inner sympathy of a like mode of thought. Dirichlet loved to
make things clear to himself with an intuitive underpinning. Along with
this he would give acute, logical analyses of foundational questions
and avoid long computations as much as possible.” His manner suited
Riemann, who adopted it and worked in many ways according to Dirichlet’s
methods.
So we come to the central character in this account, Bernhard Riemann
(1826–1866; Figure 1.5). Riemann was impoverished for most of his life. He
was shy and withdrawn, sickly and a hypochondriac. Other than the two years
in Berlin, he spent most of his working life in the still very beautiful walled
German town of Göttingen, but he changed mathematics forever. As well as
numbers, formulae and concepts, the idea of mathematical spaces and of the

Figure 1.5 B. Riemann, 1826–1866.

6 Introduction

relationships between them can be traced back to Riemann. He went beyond

Dirichlet by supporting his profound ideas with extensive calculations and
manipulations.
Riemann contributed to real analysis (the Riemann integral), complex
analysis, (Cauchy–Riemann equations), potential theory and geometry
(Riemannian manifolds). He richly deserves to be called a genius and a great
mathematician.
An eight-page paper Riemann published in the Notices of the Berlin
Academy in 1859, titled On the number of primes less than a given magnitude
[140], contained an outline of a possible proof of Gauss’s prime number
conjecture. It started from ideas of Cauchy and Dirichlet. For example, he
extended the domain of ζ(s) to the whole of the complex plane other than the
point s = 1. The paper did not contain proofs, but radically changed analytic
number theory. It took 30 years for mathematicians to begin to appreciate
what Riemann’s ideas really meant, and that his assertions were provable.
Over 70 years later the analytic and computational underpinnings for the
paper, through Siegel and his exploration of Riemann’s hand-written notes,
became clear. Riemann states in the paper what is now known as the Riemann
hypothesis, or RH. Here is a translation of what he wrote:
One finds in fact about this many zeros of the zeta function within these
bounds on the critical line, and it is very likely that all of the zeros are on
the critical line. One would of course like to have a rigorous proof of this, but
I have put aside the search for such a proof after some fleeting vain attempts,
because it is not necessary for the immediate objective of my investigation.
(For information about the term “critical line” see Section 2.2 (7).)
Indeed it was shown later that to complete the proof of the prime number
theorem, the main objective of the paper, one did not need to prove the RH,
only to show that the zero-free region for the zeta function includes all of the
line s = 1.
At the 1900 International Congress of Mathematicians, ICM1900, held in
Paris, the great mathematician David Hilbert (Figure 1.6) gave an address in
which he listed the 23 most outstanding problems of the day he considered
worthy of the efforts of mathematicians, and fundamental for the further
advancement of the subject. Problem number eight was the “Riemann
hypothesis”, which conjectured that the complex zeros of ζ(s) all had real
part 12 .
The final event in this brief history, and significant for the work reported
in this volume, is one of the consequences of assuming the hypothesis is
true. In 1914, soon after the commencement of World War I when resources
where scarce, the great Indian mathematician Srinivasa Ramanujan published
an article in the Proceedings of the London Mathematical Society titled
Highly composite numbers. At over 50 pages long it must have represented a
considerable publishing challenge. As it turned out, there was a large amount
 1.2 Early History 7

Figure 1.6 David Hilbert, 1862–1943.

of additional material on related topics written by Ramanujan which was not

included in the published work. This was discovered in more recent times,
and a typeset version with notes was eventually published in 1997. It includes
an inequality, derived by Ramanujan, who assumed the Riemann hypothesis
in this case, relating to the sum-of-divisors of an integer arithmetic function
σ(n). This can be stated as follows: for all n ∈ N sufficiently large we have

σ(n)
< eγ loglog n.
n

Here γ is Euler’s constant,

⎛⎛ n ⎞ ⎞
⎜⎜⎜⎜⎜⎜ 1 ⎟⎟⎟ ⎟
γ = lim ⎜⎜⎜⎝⎜⎜⎜⎝ ⎟⎟⎟ − log(1 + n)⎟⎟⎟⎟⎟ .
n→∞ j⎠ ⎠
j=1

The nice surprise is that this inequality, with an explicit lower bound for n
replacing “sufficiently large”, and other related arithmetic inequalities and
equalities, is equivalent to the Riemann hypothesis. The original inequality
involving σ(n) is the most famous, and the equivalence is known as “Robin’s
theorem”. A detailed history of its evolution is given in [128] with quotes
from Erdős, Hardy, Rankin, Berndt and Nicolas.
8 Introduction

1.3 Volume One Summary

Except for Chapter 10, the text contains a mostly linear progression of ideas.
In Chapter 2, key properties of the Riemann zeta function, which will be
needed, are outlined and two key parameters developed. This is then followed
in Chapter 3 by explicit estimates for functions of primes, given in such a way
that their derivations can potentially be improved. Then in Chapter 4 some
classical equivalences of the Riemann hypothesis are proved in full. These
are used in the work that follows. In Chapter 5 a set of equivalences, for
the most part in the form of inequalities, involving the Euler totient function
ϕ(n), are derived. Chapter 6 provides preparatory material for Chapter 7 by
developing the properties of two types of so-called “abundant” numbers,
wherein it is the number of divisors that are abundant. These are the numbers
that appear as possible counterexamples to inequalities. In Chapter 7 an
inequality based on values of σ(n), the sum-of-divisors function, is shown
to be equivalent to the hypothesis. In Chapter 8 the focus shifts to numbers
not satisfying the inequality, with several results showing the numbers are
very constrained, so sit in a narrow class. It is expected that this class will
be further constrained. Chapter 7 already contained an alternative inequality,
equivalent to the Riemann hypothesis, and Chapter 9 continues in this mode,
expressing equivalent formulations for the hypothesis in terms of so-called
“extraordinary” numbers and “extremely abundant” numbers.
The final chapter (Chapter 10) breaks the sequence of ideas, by giving ten
other equivalent statements for the Riemann hypothesis, mostly proved in
full. In one form or another, inequalities play a role in these formulations.
The idea of this chapter is to reveal the ubiquitous nature of the hypothesis,
and be the source of new ideas.
Further summary details are given in this section in the paragraphs below.
This first chapter, in Section 1.2, gives a sketch of the genesis of the
Riemann hypothesis. Interestingly, it was not posed as a problem, conjecture
or even hypothesis by its principal author, Bernhard Riemann, but we know
that the issues therein were first publicly discussed in 1859. This date also
marks the origin of what we now call the Riemann zeta function as a function
of a complex variable, ζ(s).
Chapter 2 summarizes some basic properties of the Riemann zeta function,
and gives an intuitive idea of its behaviour. Appendix H “The gamma
function” and Appendix I “Riemann zeta function” in Volume Two [32]
give more background and proofs. The fundamental parameters H and R
are introduced and used throughout the text. The symbol H represents a
y-value up to which all zeros of ζ(s) with positive imaginary part have been
demonstrated to have their real part equal to 1/2. Comments are made, when
needed, on which value of H has been chosen in a particular circumstance.
The value of H that could be used is expected to increase in time, and
theorems and algorithms are presented so new values can be adopted easily.
 1.3 Volume One Summary 9

The value of R describes a simple form for a zero-free region of ζ(s),

namely for ζ(β + iγ) = 0 we have

1
β < 1− .
R log γ

It is expected that the value of R will decrease in time. We give the derivation
of Landau, and quote more recent improvements.
Chapter 3 is devoted to numerical estimates, especially of the arithmetical
functions θ(x) and ψ(x). Note that there are additional estimates of products
of functions of primes in Section 5.2. Estimates in this chapter are primarily
derived without using RH. The material depends not only on the work of
Rosser and Schoenfeld in their separate and joint papers of 1941, 1962, 1963
and 1975, but also on von Mangold’s theorems of 1905.
Chapter 4 gives derivations of some well-known explicit classical equiv-
alences to RH. It may be skipped by anyone familiar with introductory
material. The proof of an important theorem of Landau, used in many cases in
the text where RH is assumed to be false, is included. The chapter concludes
with overview material on zero-free regions and a summary of heights up
to which the hypothesis has been shown to hold, with the corresponding
numbers of critical zeros.
The work of Rosser and Schoenfeld is remarkable in that it contains sharp
explicit results produced before computer-based methods became ubiquitous.
In this chapter we take some advantage of the much greater processing
speed now available to extend the ranges for which computer verification
of inequalities is practicable, and the greatly improved height below which
all the zeros of ζ(s), with positive imaginary part, have real part σ = 12 .
Improving these estimates is the subject of current research, since they have
a strong influence on the equivalences to RH.
Moving on to Chapter 5, Nicolas proved in 1983 that RH is true if and only
if for every non-prime primorial q we have
q
> eγ .
ϕ(q) loglog(q)

Here a primorial is the product of all primes up to a given prime, so the kth
primorial, say q, is q = Nk := p1 · · · pk , where pi is the ith prime starting with
p1 = 2.
Nicolas improved this result in 2012, and found four statements equivalent
to the Riemann hypothesis. Let Nk = 2 · 3 · · · pk be the kth primorial and let
 
n γ

c(n) := − e loglog(n) log n
ϕ(n)
10 Introduction

and define
 1
β := = 2 + γ − log π − 2 log 2 = 0.046191 . . . ,
ρ
ρ(1 − ρ)

where ρ ranges through the non-trivial zeros of ζ(s) with increasing absolute
value of the imaginary part. Then RH is equivalent to each of the following:
(1) lim supn→∞ c(n) = eγ (2 + β) = 3.644415 . . . .
(2) For all n ≥ N120569 = 2 · 3 · · · 1591 883 we have c(n) < eγ (2 + β).
(3) For all n ≥ 2, c(n) ≤ c(N66 ) = c(2 · 3 · · · 317) = 4.0628356921.
(4) For all k ≥ 1 we have c(Nk ) ≥ c(N1 ) = c(2) = 2.208589 . . . .
From this one shows that the Riemann hypothesis is equivalent to the
following inequality holding for all n ≥ N120569 :
n eγ (2 + β)
< eγ loglog n +  .
ϕ(n) log n
The chapter includes the derivation of useful estimates for treating the Euler
phi function, namely some products and sums of functions of primes. These
depend on the relationship between primes and zeta zeros, and form an
essential basis, along with other estimates, for developing the equivalences.
In Chapter 6, fundamental results to do with two types of number with
many divisors are developed. These are called superabundant and colossally
abundant. The work includes that of Alaoglu, Erdős and Nicolas. These
numbers appear as counterexamples to RH, so are used in the chapters that
follow. We include a corrected proof of a fundamental theorem of Alaoglu
and Erdős.
Chapter 7 is also a fundamental chapter. Grönwall proved in 1913 that
σ(n) σ(n)
G(n) := =⇒ lim sup = eγ = 1.78107 . . . .
n loglog(n) n→∞ n loglog n
Ramanujan showed, probably in 1915, that if RH is true then for n sufficiently
large we must have G(n) < eγ . Then Robin showed in 1983 that RH is true if
and only if n > 5040 implies G(n) < eγ . Also included is the equivalence of
Lagarias, which is dependent on the result of Robin, namely if Hn = 1 + 1/2 +
· · · + 1/n is the n harmonic number, then RH is equivalent to the inequality
σ(n) ≤ Hn + exp(Hn ) log(Hn ), n ≥ 1.
In Chapter 8, properties of numbers that do not satisfy Robin’s inequality
are explored. The work of Choie, Lichiardopol, Moree and Solé is developed
and extended. In particular it is shown that any integer greater than 5040
not satisfying Robin’s inequality must be even, fail to be squarefree and
not squarefull. The smallest such number must be a so-called Hardy–
Ramanujan number and superabundant. Then it is shown, successively, that
 1.3 Volume One Summary 11

any counterexample must be divisible by the fifth power of at least one prime,
the seventh power and finally the 11th power of at least one prime.
We now move on to Chapter 9. Define a composite integer n > 1 to be left
abundant or GA1 if it satisfies G(n/p) ≤ G(n) for all primes p | n. Call it right
abundant or GA2 if G(n) ≥ G(an) for all natural numbers a > 1. Then in 2011
Caveney, Nicolas and Sondow showed that the Riemann hypothesis is true
if and only if 4 is the only number that is both left and right abundant. They
called such numbers extraordinary. They also showed that a right abundant
number n > 5040 exists if and only if the Riemann hypothesis is false, and,
if so, such an integer n is even with n > 108576 . The chapter concludes with
the equivalence of Nazardonyavi and Yakubovich based on “champions” for
G(n), which they call extremely abundant.
Although Chapter 9 represents the final section of the main body of this
text, a range of other equivalences that have caught the attention of the author,
and are not in Volume Two [32], have been included in Chapter 10. Proofs
are given, as in the earlier chapters, to show depth and illustrate technique:
Farey fractions, the Redheffer matrix, the divisibility graph, the Dirichlet eta
function, the derivative of the Riemann zeta function, a simple zeta-related
inequality, and an inequality for the real part of the logarithmic derivative
of the Riemann xi function, ξ(s). The chosen set of equivalences is far from
comprehensive. For detailed descriptions of these equivalences, the reader is
invited to consult Chapter 10. An improvement of a result of Lagarias, which
gives, for every t ≥ 168π, the imaginary part of a zeta zero γ with |t − γ| ≤ 1.5,
is included.
In the penultimate section of Chapter 10, an equivalence regarding the
maximum multiplicative order, ord n, of any element of the symmetric group
S n is outlined. Not all of the details are included – these are spread over
many papers. The development is based on the work of Landau and Shah,
and includes the theorem of Massias, Nicolas and Robin, which says that RH
is equivalent to the inequality

log ord(n) < li−1 (n), n sufficiently large,

where li(x) is the logarithmic integral. A new “straight line” proof of the
theorem of Shah which is fundamental to the work is included, as is an
introduction to the related work of Erdős and Turan on the statistics of S n .
In the final section of Chapter 10, a simple equivalence, but very influential
conjecture, going under the name of the “Hilbert–Pólya conjecture”, is
introduced.
In Figure 1.7 we give some of the dependences between the chapters in
this volume. An arrow from Chapter A to Chapter B means some results in
B depend on material in A. Not all dependences have been included, but the
network flow diagram provides a general overview.
12 Introduction

Ch 9:Extreme

Ch 5:Totient Ch 10:Other

Ch 7:Robin Ch 4:Classical

Ch 6:Abundant Ch 2:Zeta

Ch 3:Estimates

Ch 8:NotRobin

Figure 1.7 Some relationships between chapters.

Appendix A consists of a set of tables giving examples of numbers that

appear in the text: numbers less than 5041 that do not satisfy Robin’s
inequality, superabundant numbers, colossally abundant numbers, the values
of the maximum order of an element of the symmetric group, solutions of
what is called Nicolas’ reversed inequality, a list of primes in order which can
be used to make colossally abundant numbers, and some extremely abundant
numbers, all for small initial values.
Appendix B is a mini-manual for the Mathematica software RHpack, a
package that is tailored for this book. As well as a description of each RHpack
function, instructions are included on how to locate, download and install the
package.

1.4 Notation
We use the following notation, sometimes without (further) explanation:

n = pα1 1 · · · pαmm with α1 > 0 is the standard prime factorization,


d(n) := 1, the number of divisors of n,
d|n

σ(n) := d, the divisor sum,
d|n

μ(n) := (−1)m if α j = 1 for all j, else 0,

ω(n) := m, the number of distinct prime divisors of n,
 1.5 Background Reading 13

m
Ω(n) := αi , the total number of prime power divisors of n,
i=1
m
λ(n) := (−1) i=1 αi , Liouville’s function,
Λ(n) := log p if n is a power of p, and 0 otherwise,
ϕ(n) := #{ j : 1 ≤ j ≤ n and (n, j) = 1}, Euler’s phi function,
 n
1
Hn := , the harmonic number,
j=1
j
ν p (n) := k where pk | n and pk+1  n,
π(x) := #{p : p ≤ x, p is prime},

θ(x) := log p,
p≤x

ψ(x) := log p,
pn ≤x

M(x) := μ(n),
n≤x
 1−  x 
dt dt
li(x) := lim + , the logarithmic integral,
→0 log t 1+ log t
 ∞ 0
dt
Li(x) := , also called the logarithmic integral,
2 log t
ord(σ), the multiplicative order of the element σ of a group.

A note on the use of symbols: Many authors have built explicit rational
constants into their explicit results. We have attempted to avoid this practice,
since an improvement in a constant will invalidate a result. Constants are
called a1 , a2 , . . . , c1 , c2 , . . ., etc. and given explicit definitions. There is
also a problem in this part of the literature with overloaded symbols. Two
examples are the function names φ(·) and K(·), both of which have a variety
of meanings, sometimes in the same context. We have attempted to avoid
this practice.
The symbols p and q invariably represent rational prime numbers. The
Landau–Vinogradov symbols O, o, , and , with their usual meanings,
are also used. Additional notation is introduced as needed.
In addition, e, π and i have their usual meanings, γ is Euler’s constant, B is
Mertens’ constant, β is 2 + γ − log π − 2 log 2 and H is a known height up to
which all complex zeros of ζ(s) lie on the critical line.

1.5 Background Reading

There are several books about the Riemann hypothesis that set it in its
historical and cultural context – see for example Sabbagh [148] and du
14 Introduction

Sautoy [151]. Books by Edwards [51], Patterson [131] and Borwein et al.
[13] give accessible mathematical introductions, whilst Titchmarsh [167],
Ivić [82] and Karatsuba and Voronin [88] are more advanced monographs.

1.6 Unsolved Problems

(1) Investigate why Dirichlet’s famous theorem on primes in arithmetic
progressions was not included in his original set of lecture notes.
(2) There are Riemann hypotheses in more general or analogous contexts:
extended Riemann hypotheses, the generalized Riemann hypothesis and
the grand Riemann hypothesis for example. Conjecture and prove an
equivalent form in one or more of these settings.
 2
The Riemann Zeta Function

2.1 Introduction
The purpose of this chapter is to give some insight into the nature of
the Riemann zeta function, and describe some of its properties, which are
essential for the development of arithmetic equivalences to RH.
Following this introductory section, the second section of this chapter
discusses some of the main properties of the Riemann zeta function, such as
how it is defined, the functional equation and the function’s zeros, which are
all important since they are the subject of the Riemann hypothesis. An idea
of how ζ(s) behaves is given with a presentation of plots of its modulus and
argument in small parts of the complex plane, including some of the complex
zeros. There is no attempt to be comprehensive. It is interesting to note that
few of the properties of ζ(s) are used in deriving equivalent formulations to
the hypothesis.
In Section 2.2, statements of the formulae of von Mangoldt are included,
which give explicit representations of errors in a form of the prime number
theorem in terms of sums of quantities dependent on the zeros of ζ(s). This
material is presented with references but without proofs, since there are a
number of excellent texts that give comprehensive expositions.
In Section 2.3 information is given on heights up to which all zeros of
ζ(s) have been proved, numerically, to be on the critical line. The zero-free
region of Landau is derived in Section 2.4. The zeta function properties and
zero-free regions are used in this volume in essential ways.
There are quite a few very important results in Section 2.6, which has
complete proofs. These include the product formula for ξ(s) given by
 s

ξ(s) = ξ(0) 1− ,
ρ
ρ
and the explicit expression for the convergent sum
1 γ 1
= 1 + − log(4π).
ρ
ρ 2 2
15
16 The Riemann Zeta Function

Both the product and sum are over the non-trivial zeros of ζ(s). When each ρ
is associated with the zero 1 − ρ, the product and sum converge absolutely.
The following RHpack functions (see Appendix B) relate to the material
in this chapter: Psi0, PlotVonMangoldtPsi, BacklundB, ZetaZeroHeight,
LargestZetaZero and NextZetaZero.

2.2 Basic Properties

In this section a summary of some of the properties of the Riemann zeta
function is given. These are well set out and derived in many texts such as
the classics [6, 51, 82, 167], and the derivations will not be repeated here. As
usual s = σ + it is a complex variable and ζ(s) is the complex value of the
Riemann zeta function at s ∈ C.
(1) ζ(s) is holomorphic on C \ {1}. It has a simple pole at s = 1 with
residue 1.
(2) On the half plane σ > 1 it can be represented by the absolutely
convergent series and Euler product
  −1
1 
∞
1
ζ(s) = = 1− s .
n=1
n s p∈P p
(3) On the right half plane σ > 0 it can be represented by the conditionally
convergent series
∞
(−1)n
ζ(s) = − .
n=1
ns
(4) Also on the half plane σ > 0 it can be represented by the integral
 ∞
s {x}
ζ(s) = −s dx
s−1 1 x s+1
where {x} is the fractional part of x ∈ R.
(5) It satisfies a functional equation. This can be written in the form ξ(s) =
ξ(1 − s) where ξ(s) is the so-called “completed” zeta function, the Riemann
xi function, defined by
s(s − 1) −s/2  s 
ξ(s) := π Γ ζ(s).
2 2
The function ξ(s) is entire.
(6) The complex zeros of ζ(s) and ξ(s) are the same. In addition ζ(s) has
zeros at each negative even integer s ∈ {−2, −4, −6, . . .}. It is traditional to call
these the “trivial zeros” of ζ(s). Other than at the trivial zeros, the function is
non-zero outside of the vertical open strip 0 < σ < 1. If ρ = β + iγ is a complex
zero of ζ(s) with β  12 then we have a set of four associated zeros:
{ρ = β + iγ, ρ̄ = β − iγ, 1 − ρ = 1 − β − iγ, 1 − ρ̄ = 1 − β + iγ}.
 2.2 Basic Properties 17

(7) There are an infinite number of simple zeros of ζ(s) on the so-called
“critical line”, σ = 12 [76]. The Riemann hypothesis is the conjecture that all
of the non-trivial zeros of ζ(s) lie on this line. This is a long story, beginning
with Hardy and Littlewood [74]. Conrey [40] has shown that more than two-
fifths of the critical zeros are on this line.
To give the reader some intuitive understanding of how ζ(s) behaves
for small values of |s|, three graphical descriptions are given. The first is
Figure 2.1, which includes some of the contours for |ζ(s)|. We see the first
three zeros on the critical line s = 12 surrounded by closed curves.
The graphic of Figure 2.2 gives the direction modulo π of the complex
number ζ(s) at each s in the domain. It is a phase portrait of the flow ṡ = ζ(s).
The first three real zeros at s = −2, −4 and −6, where the flow is alternately a
source or a sink, and the first three complex zeros, where the flow is a source
focus, are shown in the figure.
The passage of ζ(s) → 1 as s → ∞ is also clear. For further information
see the papers [17, 19, 20, 23, 25, 26].
The graphic of Figure 2.3 is a depiction of part of the flow ṡ = ξ(s). It shows
the symmetry of the flow stemming from the functional equation, the first
three complex zeros, each of which is a centre for the flow, and the trending
of the separatrices to the positive and negative x-axis. For further information
see the papers [21, 22, 26].

30 0.75
2

2
1
25 0.1 0.25 0.5
0.75
1
1
0.5
0.1 0.25 0.75
20 1

1
15 0.75
0.1 0.25 0.5
0.5
0.75
1

10
0.0 0.2 0.4 0.6 0.8 1.0

Figure 2.1 The contours of |ζ(s)|.

18 The Riemann Zeta Function
30

25

20

15

10

0
−10 −5 0 5 10

Figure 2.2 The flow ṡ = ζ(s).

(8) Bounds on the modulus of zeta, or its multiplicative inverse, are crucial
in estimating integrals using Cauchy’s theorem. One we use is the following:
assuming the Riemann hypothesis, for every  > 0 and δ > 0 there is a T 0 > 0
such that for a positive constant K, dependent on the chosen constants, such
that for every s = σ + it with σ ≥ 12 +  and t ≥ T 0 we have the bound |1/ζ(s)| ≤
Ktδ . Detailed derivations of bounds of this type are given in Volume Two
[32, Appendix I].
(9) We also need bounds on zeta and its derivative that apply on vertical
strips a ≤ σ ≤ b where a and b are any given real numbers. The bound is
unconditional and the implied constant and power A depend on a and b. There
is a constant A > 0 such that for every s = σ + it with a ≤ σ ≤ b and |t| ≥ δ > 0,
we have |ζ(s)| t A and |ζ  (s)| t A [167].
 2.2 Basic Properties 19
40

35

30

25

20

15

10

–15 –10 –5 0 5 10 15 20

Figure 2.3 The flow ṡ = ξ(s).

At times, more precise bounds for the modulus of ζ(s) on the critical line
are needed. These have been the subject of close study over a considerable
period. For example, there is the inequality of Cheng and Graham [37]:
|ζ( 12 + it)| ≤ min(6t1/4 + 57, 3t1/6 log t), t ≥ 3. (2.1)
We make particular use of the inequality of Platt and Trudgian [134], which
is better than (2.1):
|ζ( 12 + it)| ≤ 0.732t1/6 log t, t ≥ 2. (2.2)
Some use is also made of a bound of Richert [139] as made completely
explicit by Ford [60], and which gives a bound for the entire right half of the
critical strip. It is most useful for σ near 1:
3/2
|ζ(σ + it)| ≤ 76.2|t|4.45(1−σ) log2/3 |t|, |t| ≥ 3, 1
2
≤σ<1 (2.3)
(see also Cheng [36]).
We also need to use explicit zero-free regions for ζ(s), beyond heights
at which all zeros are known to be on the critical line. These are studied in
some detail in Section 2.3. We use the result of Korobov [96] and Vinogradov
20 The Riemann Zeta Function
Table 2.1 Some improvements for Backlund’s zero counting estimate.

Year Author a1 a2 a3 T0

1919 Backlund [7] 0.137 0.443 4.35 200

1938 Rosser [143] 1.12 0 9.5 1450
1941 Rosser [144] 0.137 0.443 1.588 1467
2014 Trudgian [169] 0.122 0.278 2.510 e

[176]: there is an absolute constant c > 0 such that if s = σ + it satisfies

c
1−σ ≤ (2.4)
log |t| loglog1/3 |t|
3/2

and |t| is sufficiently large, then ζ(s)  0.

(10) Having a value of the height y, denoted by H, such that every complex
zero ρ = β + iγ of ζ(s) with 0 < γ ≤ y has β = 12 , is useful in proving explicit
statements. A summary history of known results is given in Section 2.3. An
approximation to N(T ) (see (11)) is derived in Volume Two [32, Appendix I,
Theorem I.1].
(11) It is also important in applications to be able to count the number
of zeta zeros with 0 < γ ≤ T with specified accuracy. First we need some
definitions. Let N(T ) be the number of zeros ρ = β + iγ of ζ(s) for which
0 < β < 1 and 0 < γ ≤ T . Let
T T T 7
F(T ) := log − +
2π 2π 2π 8
and
B(T ) := a1 log T + a2 loglog T + a3 , T ≥ T0,
where a1 , a2 and a3 are real numbers – see Table 2.1. We use the result of
Backlund [7] as refined by Trudgian:
|N(T ) − F(T )| ≤ B(T ), T ≥ e.
For a longer list of progressive improvements, see [169, table I]. For a
description of Backlund’s approach, see [51, section 6.7].
(12) Essential tools in this study are the two explicit formulae of von
Mangoldt. They express ψ(x), or its integral, as a sum of terms involving x
and the complex zeros of ζ(s), together with some additional terms of minor
significance. Recall the definition:
   log(x) 
ψ(x) := log p = log p.
pn ≤x p≤x
log p
 2.3 Zero-Free Regions 21

In spite of their central importance, I do not give their derivation as this is set
out fully in [51, sections 3.2 and 3.5]. Let ρ range over the complex zeros of
ζ(s) with increasing absolute value of the imaginary part, and let p represent
a rational prime and n a natural number. Then we have:
Theorem 2.1 (von Mangoldt) For x > 1 which is not a power of a prime
number we have
 xρ ∞  xρ 1  
1 ζ  (0) 1
ψ(x) = x − + − = x− − log 1 − 2 − log 2π.
ρ
ρ n=1 2nx2n ζ(0) ρ
ρ 2 x

If x is a prime power the expression given on the right-hand side is equal to

the average (ψ(x−) + ψ(x+))/2, so differs from ψ(x) by O(log x).
Theorem 2.2 (von Mangoldt) For x > 1 we have
 x
x2  xρ+1 ∞
x−2n+1 ζ  (0) ζ  (−1)
ψ(t) dt = − − − x+ .
0 2 ρ
ρ(ρ + 1) n=1 2n(2n − 1) ζ(0) ζ(−1)

2.3 Zero-Free Regions

A principal tool in this work is a mixture of the classical style of zero-free
region for ζ(s), asymptotically tending to the boundary of the critical strip
along the lines of Theorem 2.4, and a height, denoted by H, below which
all critical zeros are on the critical line. Intensive use is made of the explicit
formulae of von Mangoldt, Theorems 2.1 and 2.2, using these properties to
bound infinite sums. We do not always use the best forms that are available,
but write results as far as possible in terms of symbolic parameters, so it
should be straightforward to introduce improvements.
Although the zero-free region with explicit constants of Kevin Ford [59] is
better, asymptotically, than that developed by Landau [105] described below
in Section 2.4, it is only when the height is of the order of 1070 that it is
better. Since checking the Riemann hypothesis to this height is currently
unattainable, we will present the method of Landau, which gives a zero-free
region that is used in explicit applications, especially in Chapter 3, to derive
explicit estimates. See also the notes at the end of Section 2.5 where there is
a summary of recent results, including those of Ford, Kadiri, Jang and Kwon
[83], and Platt and Trudgian.
Table 2.2 of heights H up to which all imaginary zeros are on the critical
line has been compiled using data from Borwein et al. [13, p. 39] and
published articles or direct computations for the heights up to 35 × 105 . A
variety of computational methods were used to establish these zeros and
heights. Good examples include [14, 15, 68, 178]. For references consult
[13], or for the most recent result, which is that of David Platt, see [132] and
[133, section 7]. Below we discuss the seminal 1953 value of Alan Turing.
22 The Riemann Zeta Function
Table 2.2 Zeros on the critical line and corresponding heights.

Year Number of zeros Height H Author(s)

1856? 3 25.02 B. Riemann

1903 15 65.12 J. P. Gram
1914 79 198.02 R. J. Backlaund
1925 138 299.85 J. I. Hutchinson
1935 195 388.85 E. C. Titchmarsh
1936 1041 1466.51 L. J. Comrie, E. C. Titchmarsh
1953 1104 1539.31 A. M. Turing
1956 15 000 14 040.46 D. H. Lehmer
1956 25 000 21 942.60 D. H. Lehmer
1958 35 337 29 750.17 N. A. Meller
1966 25 × 104 170 570.23 R. S. Lehman
1968 35 × 105 1459 433.86 J. B. Rosser, J. M. Yohe,
L. Schoenfeld
1977 4 × 107 — R. P. Brent
1979 75 × 106 32 585 736.4 R. P. Brent
1982 2 × 108 + 1 81 702 130.19 R. P Brent, J. van de Lune,
H. J. J. te Riele, D. T. Winter
1983 3 × 108 + 1 119 590 809.282 J. van de Lune, H. J. J. te Riele
1986 15 × 108 + 1 545 439 823.215 J. van de Lune, H. J. J. te Riele,
D. T. Winter
2004 9 × 1011 — S. Wedeniwski
2004 1013 — X. Gourdon
2011 103 800 788 359 30 610 046 000 D. J. Platt

Because the final three items in Table 2.2 are “unpublished” at the time of
writing, we have used the 1986 value of van de Lune, te Riele and Winter
[111] in our own derivations in Chapter 3. However Platt’s value of H was
used, as reported in [133], to show that

π(1024 ) = 18 435 599 767 349 200 867 866,

which is his theorem 7.1. This result corresponds to that of Büthe, Franke,
Jost and Kleinjung of 2010 in [33], which was derived using RH. Thus there
is considerable confidence that Platt’s value of H is correct.
The most influential method to date for verifying RH, up to some height
H, has been Turing’s method, which was published in 1953, a year before
his most unfortunate death [174, 175]. Almost all methods following Turing
share important components with it. There are details in [51, section 8.2].
For an introduction see Booker’s article [12] or that by Hejhal and Odlyzko
[78]; see also Trudgian [172]. Here we give a brief outline, with many details
skipped.
We saw in Section 2.2 (11) that if we define E(t) := N(t) − F(t) we can
write |E(t)| ≤ B(t). Now −F(t) is a smooth decreasing function and N(t) a step
 2.3 Zero-Free Regions 23

function that increases by 1 at each simple critical line zero, and by at least 2
if t is the ordinate of a pair of zeros off the critical line. We will see below that
detecting (and hence counting) zeros on the critical line is straightforward.
We have more analytic information on S (t), a good approximation to E(t),
defined below, which is contradicted if a pair of off-critical-line zeros occurs.
The great strength of Turing’s method is that it enables RH to be checked
by considering only the values of ζ(s), and related functions, with s on the
critical line.
Some further details: let γ(s) := π−s/2 Γ(s/2) and Λ(s) := γ(s)ζ(s). Then the
functional equation can be written Λ(s) = Λ(1 − s), and this can be used to
show that Λ( 12 + it) is real for all real t. The zeros of Λ( 12 + it) are precisely the
zeros of ζ( 12 + it). Now let ϑ(t) be the argument of γ( 12 + it), with ϑ(0) := 0, or
in other words
γ( 12 + it) = |γ( 12 + it)|eiϑ(t) .
Then it can be shown that
 
ϑ(t) 1
+ 1 = F(t) + O .
π t
The real-valued function of a real variable S (T ) is the argument of ζ(s) on
the critical line at the point 12 + iT which is not the ordinate of a zero. The
argument is fixed by taking the value found by continuous variation from 0
along the horizontal line from (∞, iT ) to ( 12 , iT ). At a zero the value is defined
to be limt→0+ S (t). See Figure 10.4 to get a feeling for how S (T ) behaves. It
is well known that we can write
 T  7  
T 1
N(T ) = log + + S (T ) + O .
2π 2πe 8 T
See for example Edwards [51, chapters 8 and 9]. It follows that
  
ϑ(t) 1
S (t) = N(t) − +1+O .
π t
Now |Λ(s)| decays exponentially on the critical line, so, for detecting zeros,
it is easier to work with the function
Λ( 1 + it)
Z(t) := 12 ,
|γ( 2 + it)|
which also has the same zeros for real t as ζ( 12 + it). Each simple zero on the
critical line for ζ(s) is accompanied by a sign change in Z(t). This function
also has the advantage in computations of requiring fewer terms in a summa-
tion than competing methods, via the famous Riemann–Siegel formula:
√
 t/2π  
cos(ϑ(t) − t log n) 1
Z(t) = 2 √ + O 1/4 .
n=1
n t
24 The Riemann Zeta Function

A key ingredient, exploited by Turing, is the theorem of Littlewood, which

states that the average value of S (t) is zero:

1 T
lim S (t) dt = 0.
T →∞ T 0

Turing developed an explicit form for this, which was recently improved by
Trudgian. For h > 0 and T > 168π we have
 T +h
S (t) dt ≤ 2.067 + 0.059 log(T + h). (2.5)
T

See Lemma 10.29 below.

So, with these preliminaries in hand, we can describe the essence of the
method: the function S (t) oscillates about zero in a reasonably controlled
manner, with average value zero. Proceed up the critical line counting zeros
and “storing” −S (t) using those values and checking (2.5) for h a small
multiple of log T . Should a pair of zeros be missed (because they are off the
critical line), or a zero on the line be missed, then this would register as an
immediate increase in the “central” value of −S (t), and soon become evident
as a change in the moving average, by violating (2.5). The region in C where
this occurred could be investigated further to determine whether a zero or
multiple zero on the critical line had been missed, or indeed a counterexample
to RH had been discovered. This is illustrated in Figures 2.4 and 2.5 for which
−S (t) is contrasted with T (t), a function that is the same as −S (t), but has a
“fake” missing double zero at t = 1010, registered by S (t), but not by N(t).

–S(t)

1.0

0.5

t
1010 1020 1030 1040

–0.5

–1.0

Figure 2.4 The function −S (t) for 1000 ≤ t ≤ 1040.

 2.4 Landau’s Zero-Free Region 25
T(t)

t
980 1000 1020 1040 1060

–1

Figure 2.5 The function T (t) on [950, 1060] with a phantom zero at t = 1010.

2.4 Landau’s Zero-Free Region

Zero-free regions play a vital role in this development. Although Landau’s
derivation of a zero-free region for ζ(s) is by now quite classical and has
undergone successive improvements (see Section 2.5), it has the advantage
of relative simplicity and clear ideas. The progress that has been made has
been with considerable effort, but will point to the way forward if RH is
shown to be false.

Lemma 2.3 [105, section 79] If 1 < σ ≤ 2 and t ≥ 2 there is a real constant
b1 such that
ζ  (σ + it)  log p cos(mt log p)
− =
ζ(σ + it) p, m ≥ 1 pmσ
1  σ−β β

< log t + b1 − + ,
2 ρ
(σ − β)2 + (t − γ)2 β2 + γ2

where the sum is over the imaginary zeros ρ = β + it of ζ(s).

Proof First note that when σ > 1 we can write

 1
−1
ζ(s) = 1− s ,
p
p
26 The Riemann Zeta Function

with the product converging absolutely when σ is bounded away from 1.

Taking the negative of the logarithmic derivative gives [6, p. 239]
ζ  (s)  log p cos(mt log p)
− = .
ζ(s) p,m
pmσ
Now consider the Weierstrass product for ξ(s), Theorem 2.11, together with
its definition in terms of ζ(s):
 s
 s 
−s/2
ξ(s) = ξ(0) 1 − = (s − 1)π Γ + 1 ζ(s),
ρ
ρ 2
where the product is over the imaginary zeros of ζ(s) taken in pairs ρ and
1 − ρ. Take the logarithmic derivative of this identity at points that are not
zeta zeros, and use Lemma 2.10, to get
 
ζ  (s) 1 1 Γ (s/2 + 1)  1 1
= b− − + + ,
ζ(s) s − 1 2 Γ(s/2 + 1) ρ
s−ρ ρ
where the constant b = log 2π − 1 − γ/2. Then take the negative of the real
part of this equation, with s = σ + it and ρ = β + iγ, to get
 log p cos(mt log p) σ−1 1 Γ (σ/2 + 1 + ti/2)
= − b + + 
p,m
p mσ (σ − 1)2 + t2 2 Γ(σ/2 + 1 + ti/2)
  
σ−β β
− + .
ρ
(σ − β)2 + (t − γ)2 β2 + γ2
Now use the bound |Γ (s)/Γ(s)| ≤ log t + O(1), which can be derived from the
equation following (H.4) in Volume Two [32], and which is valid for 0 ≤ σ ≤ 2
and t ≥ 2 to get
 log p cos(mt log p) 1  σ−β β

< log t + b 1 − + ,
p,m
pmσ 2 ρ
(σ − β)2 + (t − γ)2 β2 + γ2
which completes the proof of the lemma. 
Theorem 2.4 [105, section 78] If for all θ we have
a0 + a1 cos(θ) + a2 cos 2θ) + · · · + an cos nθ ≥ 0
with each a j ≥ 0 and 0 < a0 < a1 , and ρ = β + iγ is a zero of ζ(s) with γ > 0,
then
1
β < 1− ,
R log γ
where R is any real number greater than
a 1 + a2 + · · · + a n
√ √ .
2( a1 − a0 )2
 2.4 Landau’s Zero-Free Region 27

Proof From Lemma 2.3, if we apply the stricter bounds 1 < σ ≤ 2 and t ≥ 2
we have
 log p cos(mt log p) 1  σ−β β

< log t + b 1− + ,
p,m
pmσ 2 ρ
(σ − β)2 + (t − γ)2 β2 + γ2

so in particular we have for some constant b1 the bound

  
ζ (σ + it) 1
− < log t + b1 .
ζ(σ + it) 2
It also follows that for t = γ ≥ 2, if γ is the ordinate of a zero ρ of ζ(s) and
σ ∈ (1, 2], then
 log p cos(mγ log p) 1 1
< log γ + b1 − .
p,m
p mσ 2 σ −β

Therefore, since for all θ we have

a 0 a2 an
−cos θ ≤ + cos 2θ + · · · + cos nθ,
a1 a1 a1
we can write
1 1  log p cos(mγ log p)
< log γ + b1 −
σ−β 2 p,m
pmσ
1   a 0 a2
≤ log γ + b1 + + cos(2mγ log p) + · · ·
2 p,m
a1 a1

an log p
· · · + cos(nmγ log p) mσ
a1 p

  
1 a0 ζ (σ) a2 ζ (σ + 2γi)
= log γ + b1 − −  − ···
2 a1 ζ(σ) a1 ζ(σ + 2γi)
  
an ζ (σ + nγi)
··· −  .
a1 ζ(σ + nγi)
Also, since n is fixed and 1 < σ ≤ 2, using the Laurent series for ζ(s) from
Volume Two [32, Theorem I.2] to bound ζ  (σ)/ζ(σ) and the bound derived
above, there is a constant b2 such that we have
 
1 1 a2 an a0 1
< b2 + 1 + + · · · + log γ +
σ−β 2 a1 a1 a1 σ − 1
a0 1 a 1 + · · · + an
= b2 + + log γ.
a1 σ − 1 2a1
If χ := a0 /a1 and a real number λ satisfies
a 1 + · · · + an
λ> ,
2a1
28 The Riemann Zeta Function

it follows that for all sufficiently large γ and 1 < σ ≤ 2 we have

1 χ
< + λ log γ.
σ−β σ−1
If we set σ = 1 + g/log γ, with σ fixed in (1, 2], this will define a positive real
number g. Applying the inequality we have derived, when g is sufficiently
large:
 
1 χ
< + λ log γ
1 − β + g/log γ g
 
g 1 1
1−β+ >
log γ χ/g + λ log λ
   
1 1 1 − χ − λg 1
1−β > −g = .
χ/g + λ log γ χ/g + λ log γ
Now choose g so that the coefficient
1 − χ − λg 1
= −g
χ/g + λ χ/g + λ
is as large as possible: we have
 
d 1
0= −g
dg χ/g + λ
χ
= − 1,
(χ + gλ)2
and therefore
√
χ−χ
g= .
λ
Because 0 < a0 < a1 , the number g is real and positive. The maximum value
of the coefficient is
√ √ √
1 1− χ χ − χ (1 − χ )2
−g = − =
χ/g + λ λ λ λ
so that for all sufficiently large γ we have
√
(1 − χ )2 1 1
1−β > =
λ log γ R log γ
√ √ 2
with R > (a1 + a2 + · · · + an )/[2( a1 − a0 ) ]. This completes the proof. 
Landau used the choices a0 = 5, a1 = 8, a2 = 4 and a3 = 3, with Theorem 2.4
to derive R = 18.52.
 2.5 Zero-Free Regions Summary 29
Table 2.3 Examples for finding values of R.

No. Year Author Form R

(1) 1898 Hadamard 3 + 4 cos(θ) + cos(2θ) 34.82

(2) 1909 Landau 5 + 8 cos(θ) + 4 cos(2θ)
+ 3 cos(3θ) 18.52
(3) 1941 Rosser 18 + 30 cos(θ) + 17 cos(2θ)
+ 6 cos(3θ) + cos(4θ) 17.72
(4) 1962 Rosser/Schoenfeld 2(1 + cos(θ))2 (3 + 10 cos(θ))2 17.51
(5) 2000 Ford (cos(θ) + 0.225)2 (cos(θ) + 0.9)2 8.463
(6) 2014 Kadiri 8(cos(θ) + 0.265)2 (cos(θ) + 0.91)2 5.69693
(7) 2014 Jang/Kwon 8(cos(θ) + a)2 (cos(θ) + b)2 5.68371
16
(8) 2014 Platt/Trudgian k=0 ak cos(kθ) 5.573412

2.5 Zero-Free Regions Summary

In this section we present a summary of some of the developments regarding
zero-free regions of ζ(s). Example (1) was the original used by Hadamard
and de la Vallée Poussin in their proof of the prime number theorem, (2)
appears in Landau’s book [105], (3) was employed by Rosser in his 1941
paper, (4) after probably some serious searching by Rosser and Schoenfeld
in their 1962 paper, with only a modest improvement in R. Example (5) was
found using numerical optimization by Ford [59].
The value from (4) is
515
R= √ √ = 17.51631055012728537 . . . .
( 546 − 322 )2
Using Ford’s explicit bound from 2002 [60], namely for ρ = β + iγ a zeta zero
with γ > 0,
1
β < 1− ,
57.54 log γ loglog1/3 γ
2/3

one could make R arbitrarily small, provided H is sufficiently large. At the

time of writing, the known value of H is well out of range.
In their 1975 paper [146], Rosser and Schoenfeld used a different method,
which is more analytic than Landau’s, to find a better, but quite convenient,
error-free region for ζ(s). It takes a form that is just slightly better than
1
β < 1− ,
R1 log(γ/17)
where ρ = β + iγ is a zero of zeta with
√ √ √
R1 := (1 − 1/ 5 )515/( 546 − 322 )2 = 9.6827783291110 . . .
30 The Riemann Zeta Function

and γ ≥ 21. This form for the zero-free region, together with a better
approximation for the integrals
 ∞  t 
φm (t) log dt,
h 2π
than that which has been used here in Chapter 3, gives, on the face of it,
better results than those we computed and use when calculating estimates.
However, since we do not need sharper estimates than those obtained by
using larger values of H, and in the interests of simplicity of exposition, we
have not used R1 .
Finally, note some more recent results: Kadiri [84] showed that we could
choose R = 5.69693. He used Weil’s explicit formula with the method and
one of the test functions of Heath-Brown. For background definitions, proofs
and applications of the explicit formula, see Volume Two [32, Chapter 9].
The value was slightly improved by Jang and Kwon, who used an alternative
test function in the explicit formula and Platt’s height H = 3.0610046 × 1010
of 2011 to derive R = 5.68371; see [83]. This has been further improved
to R = 5.573412 by Mossinghoff and Trudgian [121] by finding a well-
adapted trigonometric polynomial (of degree 16) using simulated annealing,
and carefully analysing the error term of Kadiri. They used Platt’s value of H.
Ford used the value for H established by van de Lune, te Riele and Winter
[111], and Kadiri used the 2004 height of Wedeniwski. We have chosen to
use a value R = 8.0 in the computations reported in Chapter 3.
The details of these more recent results are outside the scope of this book.
The interested reader could consult the fundamental papers [77, 179], or the
references given above.

2.6 The Product Over Zeta Zeros

In this section we show that the completed zeta function ξ(s) has a product
representation
 s

ξ(s) = ξ(0) 1− .
ρ
ρ

This is given in Theorem 2.11. First we have two lemmas, which take the
form of evaluations, and then a call-out to Hadamard’s product expansion,
which is proved in Volume Two [32, Appendix B].
Lemma 2.5 We have the special values:
Γ (1)
(1) = −γ,
Γ(1)
Γ (3/2)
(2) = 2 − γ − log 4,
Γ(3/2)
 2.6 The Product Over Zeta Zeros 31
ζ  (0)
(3) = log(2π).
ζ(0)
Proof (1) By Lemma H.1 of Volume Two [32], for all z which is not a
negative integer, we have the following representation for the logarithmic
derivative of Γ(z):
∞  
Γ (z) 1  1 1
= −γ − + − ,
Γ(z) z n=1 n n + z

and the evaluation follows on setting z = 1.

(2) By the formula in Step (1) we also have
∞  
Γ (3/2) 2  1 1
= −γ − + −
Γ(3/2) 3 n=1 n n + 3/2
2 8 − 3 log 4
= −γ − +
3 3
= 2 − γ − log 4.
Alternatively, the value in part (2) of the lemma follows from Gauss’s
digamma theorem [92]: for p, q ∈ N
Γ (p/q) 1
= − γ − log(2q) − π cot(πp/q)
Γ(p/q) 2

+2 cos((2πpn)/q) log(sin(πn/q)).
0<n<q/2

(3) A proof using Euler–Maclaurin summation is as follows. By Abel’s

theorem [6, theorem 4.2] we have for σ > 1
∞  ∞
1 1 1 {x}
= + − s dx,
n=N
n s (s − 1)N s−1 N s
N x
s+1

and
 ∞
dx 1
= .
N x s (s − 1)N s−1
Hence
∞  ∞  ∞
1 dx 1 {x}
s
− s
= s
− s s+1
dx.
n=N
n N x N N x

For s = σ > 1 and positive integer N, evaluating the integral gives

∞  ∞
1 1 dx
ζ(s) − = −
s − 1 n=1 n s
1 xs
32 The Riemann Zeta Function
  N  ∞
1 1
N−1 ∞
dx dx
= s
+ s
− s
−
n=1
n n=N
n 1 x N xs
  N  ∞
dx  1
N−1 ∞
1 dx
= s
− s
+ s
−
n=1
n 1 x n=N
n N xs

N−1  N  ∞
1 dx 1 {x}
= s
− s
+ s −s s+1
dx.
n=1
n 1 x N N x

The right-hand side converges to a holomorphic function on the half plane

σ > 0 and gives an analytic continuation of the left-hand side in this half
plane. At s = 1 the right-hand side converges, as N → ∞, to Euler’s constant
γ, because we have the representation
 
1 1
γ = lim 1 + + · · · + − log N .
N→∞ 2 N
Therefore the Taylor expansion of g(s) := (s − 1)ζ(s) about s = 1 has the form
g (1) γ
g(s) = (s − 1)ζ(s) = 1 + γ(s − 1) + O(|s − 1|2 ) =⇒ = = γ.
g(1) 1
Next write the functional equation of ζ(s) in the form [6, theorem 12.7]
ζ(s) = Γ(1 − s)(2π) s−1 2 sin(sπ/2)ζ(1 − s).
This implies (s − 1)ζ(s) = −Γ(2 − s)(2π) s−1 2 sin(sπ/2)ζ(1 − s). Take the
logarithmic derivative of this equation at s = 1 and use part (1) of the lemma
to derive
g (1) Γ (1) π cos(π/2) ζ  (0)
=γ=− + log(2π) + −
g(1) Γ(1) 2 sin(π/2) ζ(0)
ζ  (0)
=⇒ = log(2π),
ζ(0)
which completes the derivation. 
Lemma 2.6 If γ is Euler’s constant then
  
ζ (s) 1
lim + = γ.
s→1 ζ(s) s−1
Proof This follows from Step (3) of the proof of Lemma 2.5. 
We also need, in several places below, the well-known product expansion
for ξ(s) in terms of the non-trivial zeta zeros. This can be proved directly from
properties of the constituent functions (see for example [51, chapter 2]), but
we have chosen to give a proof based on Hadamard’s expansion, which is
proved in Volume Two [32, Theorem B.10]; see also [43, section 12].
 2.6 The Product Over Zeta Zeros 33

Recall the definition: the order of an entire function f (z) is the infimum of
real values b such that as z → ∞, f (z) exp(|z|b ). Next we give Hadamard’s
theorem statement.
Theorem 2.7 (Hadamard factorization) Let f (z) be an entire function of
finite order ρ, having f (0)  0 and zeros z1 , z2 , . . . , with multiple zeros
appearing multiple times. Then f (z) has the absolutely and uniformly
converging on compact subsets product representation
∞   
z Q(z/zn )
f (z) = e P(z)
1− e , (2.6)
n=1
zn
where P(z) is a polynomial of degree less than or equal to ρ, and Q(z) is a
polynomial of the form
z2 zm
Q(z) = z + + ··· + ,
2 m
where m ≥ 0 is the smallest integer such that
∞
1
< ∞.
|z
n=1 n
|m+1
Recall the definition:
ξ(s) := 12 s(s − 1)π−s/2 Γ(s/2)ζ(s).
Lemma 2.8 For all s ∈ C we have constants A and B such that
   
s s/ρ
ξ(s) = e A+Bs
1− e ,
ρ
ρ
where, if we combine the factors associated with (1 − s/ρ) and (1 − s/(1 − ρ)),
the product converges absolutely and uniformly on compact subsets of C.
Proof (1) It follows from properties of the zeros and poles of ζ(s) and Γ(s)
that ξ(s) is an entire function. We first show that ξ(s) has order at most 1. By
the functional equation we need only consider σ ≥ 12 . In this half plane, for
|s| sufficiently large and some constant c1 > 0 we have
| 21 s(s − 1)π−s/2 | ≤ exp(c1 |s|).
By Stirling’s approximation (see (H.5) in Theorem H.5 in Volume Two [32]),
there is a constant c2 > 0 such that, for s → ∞ in σ ≥ 12 ,
Γ(s/2) ≤ exp(c2 |s| log |s|).
Then using Abel’s theorem [6, theorem 4.2] for σ > 1, and extending the
domain of validity of the expression by analytic continuation to σ > 0,
 ∞
s {x}
ζ(s) = −s dx,
s−1 1 x s+1
34 The Riemann Zeta Function

which gives |ζ(s)| ≤ c3 |s| as s → ∞ in σ ≥ 12 . Combining these estimates with

the definition of ξ(s), and the functional equation, gives for each  > 0, as
|s| → ∞,
|ξ(s)| ≤ exp(|s|1+ ).
Hence the order is at most 1. We cannot do better than this in terms of the
value of the exponent on the right-hand side, since when σ → ∞, log Γ(σ) is
asymptotic to σ log σ. Hence the order of ξ(s) is 1.
(2) Next we show that the parameter m inHadamard’s theorem
(Theorem
 2.7) has the value 1. This is because the ρ 1/|ρ|2 converges and
ρ 1/|ρ| diverges. To verify the former write, since the number of Riemann
zeta zeros up to T , by Lemma 2.9, satisfies N(T ) T log T ,
 1  1  1  ∞
d(t log t)
=2 ≤2 < ∞.
ρ
|ρ| 2
γ>0
β +γ
2 2
γ>0
γ 2
1 t2

To verify the latter statement, by Theorem 10.32 proved below, for each n ∈ N
there is a zeta zero γ jn such that |4n − γ jn | ≤ 1.5. Therefore if n  n , we have
γ jn  γ jn and we can write
 1  1  1  ∞
1 
∞
1
≥2  ≥ ≥ ≥ = ∞.
ρ
|ρ| γ>0 1/4 + γ2 γ>0 |γ| n=1 |γ jn | n=1 |4n − 2|
(3) It follows from Steps (1) and (2) and Hadamard’s theorem that there
are constants A, B such that for s ∈ C
 
s s/ρ
ξ(s) = e A+Bs
1− e . (2.7)
ρ
ρ

This completes the proof. 

In Section 2.2 (11), we commented on available good estimates for zeta
zeros in the critical strip. Sometimes in what follows all we need to have is
an asymptotic upper bound for the number of zeros up to T as a function of
T . This is given in Lemma 2.9, which is a standard application of Jensen’s
formula. Indeed, we used the bound N(t) t log t in Step (3) of the previous
lemma, and will use it again in Lemma 2.10. Note the argument here is not
circular!
Lemma 2.9 The function ξ(s) has at most O(R log R) zeros in any disc
B(0, R].
Proof Let 0 < R, let n(R) be the number of zeros of ξ(z) in B(0, R], and let
M(R) = sup{|ξ(z)| : |z| ≤ R}. By the proof of Lemma 2.8, Step (1), we have
M(R) ≤ exp(c1 R log R).
 2.6 The Product Over Zeta Zeros 35

Label the zeros (an ) of ξ(z) in B(0, R] such that

|a1 | ≤ |a2 | ≤ · · · .
Next we use Jensen’s formula, from Volume Two [32, Theorem B.5], to write
  2π  n(2R)
 2R  n(R)
1 2R
M(2R) ≥ exp log |ξ(2Re )| dθ =
iθ
≥ ≥ 2n(R) .
2π 0 n=1
|a n | n=1
|a n |

Therefore n(R) ≤ log M(2R)/log 2. Since we have M(R) ≤ ec1 R log R so n(R) ≤
c3 R log R, and therefore n(R) = O(R log R). This completes the proof. 
We also need, frequently in the developments that follow in later chapters,
several expressions and estimates for sums over critical zeta zeros. Statement
(a) in Lemma 2.10 was, apparently, known to Riemann. We saw in the proof
of Lemma 2.8 that the convergence was conditional. In the second sum, as
usual 1/ρ is associated with 1/(1 − ρ), in which form the series is absolutely
convergent.
Lemma 2.10 We have
1 γ 1
(a) = 1 + − log 4π,
ρ
ρ 2 2
 1
(b) = 2 + γ − log 4π,
ρ
ρ(1 − ρ)
 1
(c) ≤ 0.046191441,
ρ
|ρ| 2

where the sums are over all of the non-trivial zeros of ζ(s).
Proof By the definition of ξ(s) and Lemma 2.8, for some real U, V we can
write
1  s  
s s/ρ
−s/2
ξ(s) = s(s − 1)π Γ ζ(s) = eU−V s
1 − e = ξ(1 − s), (2.8)
2 2 ρ
ρ

where the product is over the non-trivial zeros of ζ(s).

Logarithmic differentiation gives
ξ (s) ζ  (s) 1 1 1 Γ (s/2 + 1)
= + − log π +
ξ(s) ζ(s) s − 1 2 2 Γ(s/2 + 1)
 1 1

= −V + + . (2.9)
ρ
s − ρ ρ

By Lemmas 2.5 (2) and 2.6, respectively, we know that

  
1 Γ (3/2) γ ζ (s) 1
− = − 1 + log 2 and lim + = γ,
2 Γ(3/2) 2 s→1 ζ(s) s−1
36 The Riemann Zeta Function

so using (2.9) we can write

  
ζ (s) 1 1 1 Γ (3/2)
−V = lim + − log π +
s→1 ζ(s) s−1 2 2 Γ(3/2)
γ 1
= 1 + − log(4π).
2 2
But by (2.8) we have
ξ (0) ξ (1)
V =− = ,
ξ(0) ξ(1)
so using (2.9) we get
 1 1  1 1

−V + + =V− + .
ρ
−ρ ρ ρ
1−ρ ρ

Because ζ(ρ) = 0 if and only if ζ(1 −ρ) = 0, there are matching terms in these
sums that cancel, leading to 2V = 2 ρ 1/ρ and therefore
1 γ 1
V= = 1 + − log(4π). (2.10)
ρ
ρ 2 2

This completes the proof of part (a). Part (b) is immediate since
 1 1 1
 1
= + =2 .
ρ
ρ(1 − ρ) ρ
ρ 1−ρ ρ
ρ

Now choose a height H up to which all of the zeta zeros are on the critical
line and define
 1  1
SH = and S ∞ = .
|γ|≤H
1/4 + γ2 |γ|>H
β2 + γ 2

We choose in this derivation H = 600 269.6771, giving 106 zeros with positive
imaginary part. Then
 1
≤ S H + S ∞.
ρ
|ρ|2

Note also that

 
β 1 β
|γ| > H =⇒ < 1+ 2 2
γ 2 H β + γ2
and
 
β−1 1 β−1
< 1+ 2 ,
γ 2 H (β − 1)2 + γ2
 2.6 The Product Over Zeta Zeros 37

and therefore
 1   β 1 − β
S∞ < = + 2
|γ|>H
γ2 |γ|>H γ2 γ
  
1 β 1−β
< 1+ 2 + .
H |γ|>H β2 + γ2 (1 − β)2 + γ2

But
1  
1  2β 1
2V = + = =⇒ S ∞ < 1 + 2 (2V − S H ),
ρ
ρ ρ̄ ρ
|ρ|2 H

and therefore
 1  
1
< S H + 1 + 2 (2V − S H ) < 0.046191441,
ρ
|ρ|2 H

which completes the proof of part (c). 

Theorem 2.11 For all s ∈ C we have
 s

ξ(s) = ξ(0) 1− ,
ρ
ρ

where if we combine the factors (1 − s/ρ) and (1 − s/(1 − ρ)) the product
converges absolutely and uniformly on compact subsets of C. In addition
ξ(0) = 12 .
Proof (1) By Lemma 2.8 there are constants A, B such that for s ∈ C
 
s s/ρ
ξ(s) = e A+Bs
1− e . (2.11)
ρ
ρ

Use the definition of ξ and take the logarithmic derivative of both sides of
 
s s/ρ 1
e A+Bs
1 − e = s(s − 1)π−s/2 Γ(s/2)ζ(s)
ρ
ρ 2

to get
 
ζ  (s) 1 1 1 Γ (s/2 + 1)  1 1
= B + log π − − + + .
ζ(s) 2 s − 1 2 Γ(s/2 + 1) ρ∈S s − ρ ρ

By (2.10) of Lemma 2.10 we get the value of B:

1 1 γ
B = −V = − = log(4π) − 1 − .
ρ
ρ 2 2
38 The Riemann Zeta Function

(2) From the definition of ξ(s) and the functional equation ξ(s) = ξ(1 − s)
we can write
1 1
ξ(0) = ξ(1) = π−1/2 Γ(1/2) lim(s − 1)ζ(s) = .
2 s→1 2
Thus, by (2.11), we have eA = ξ(0) = 12 .
(3) Next we consider the convergence of the product. With ρ = β + iγ let
 s
 
s(1 − s)

1− = 1− .
ρ
ρ γ>0
ρ(1 − ρ)

This converges absolutely provided γ>0 1/|ρ(1 − ρ)| < ∞. To see this,
complete the square
   1  ∞
1 1 d(t log t)
=   < < ∞.
|ρ(1 − ρ)|
γ>0
(ρ − ) + 
γ>0
1 2
2
1
4 γ>0
|ρ − |
1 2
2 1 t2

(4) Because V = −B we are now able to write

⎛ ⎛ ⎞⎞
⎜⎜⎜ ⎜⎜⎜  1 ⎟⎟⎟⎟⎟⎟   s
  s

⎜ ⎜
ξ(s) = ξ(0) exp ⎜⎜⎝ s ⎜⎜⎝ B + ⎟ ⎟
⎟⎟⎟⎟ 1 − = ξ(0) 1− .
ρ
ρ ⎠⎠ ρ ρ ρ
ρ

This completes the proof. 

Finally, as a corollary to Theorem 2.11, we can derive a useful expression
for the logarithmic derivative of ζ(s).
Lemma 2.12 [51, section 3.2] Let S be the zeros of ζ(s) with 0 < s < 1.
Then for all s  1 and not in S in C we have
 
ζ  (s) γ 1 1 Γ (s/2 + 1)  1 1
= log(2π) − 1 − − − + + .
ζ(s) 2 s − 1 2 Γ(s/2 + 1) ρ∈S s − ρ ρ

Proof From the definition of ξ(s) and Theorem 2.11 we have

 s

−s/2
Γ(s/2 + 1)π (s − 1)ζ(s) = ξ(s) = ξ(0) 1− ,
ρ
ρ

where the product is over the non-trivial zeros of ζ(s). If s is not a zeta zero
we can take the logarithmic derivative to get
ζ  (s) 1 1 1 Γ (s/2 + 1)  1
− = − log π + + − .
ζ(s) 2 s − 1 2 Γ (s/2 + 1) ρ
s−ρ

Evaluating this at s = 0 and using Lemma 2.5 (1) gives

ζ  (0) 1 γ 1
= log π + 1 + − .
ζ(0) 2 2 ρ
ρ
 2.7 Unsolved Problems 39

Adding these two equations gives

 
ζ  (s) ζ  (0) γ 1 1 Γ (s/2 + 1)  1 1
− + = 1+ + + − + .
ζ(s) ζ(0) 2 s − 1 2 Γ (s/2 + 1) ρ
s−ρ ρ

Finally substitute for ζ  (0)/ζ(0) from Lemma 2.5 (3) to complete the proof
of the lemma. 

2.7 Unsolved Problems

(1) Prove that ζ(3) is a rational multiple of π3 .
(2) Further improve the result of Backlund which gives an estimate for the
number of zeros of ζ(s) in the positive critical strip up to height T .
(3) Find a sum of cosines giving  a smaller value of R, searching through
functions of the form f (θ) = 1≤i≤n (ai + bi cos θ)2 or otherwise.
(4) Determine, using Dirichlet’s pigeonhole principle or otherwise, whether
the value of R can be made arbitrarily close to 12 , i.e. for all  > 0 there is
a positive sum of cosines such that 12 < R < 12 + .
(5) Improve the proof of Rosser and Schoenfeld of [146, theorem 1] to find a
smaller value of the denominator R1 log(t/17) with |t| ≥ t0 := 21, by either
reducing the value of R1 or replacing 17 with a larger value, at possibly
the cost of using a much larger t0 .
(6) Study the works given in the references at the end of Section 2.3, and use
the methods therein, with improved test functions, to find a smaller value
of R.
 3
Estimates

3.1 Introduction
This chapter includes the derivation of estimates that are used throughout the
volume. The aim here is not to be comprehensive and derive all estimates of a
given type, or to present the best that are, in some sense, currently available.
Almost all of the results that are needed for deriving equivalences included
in the volume are deduced from first principles, in a manner which could be
followed step by step, should the reader so wish.
The main focus is on numerical estimates, especially of the arithmetical
functions ψ(x) and θ(x). Note that there are additional estimates of products
of functions of primes in Section 5.2. In the main, the estimates in this
chapter are derived without using RH, and so are significantly weaker than
Theorems 4.5 or 4.6, which give good estimates for ψ(x) and θ(x), which
could be regarded as best possible. The material is based on the work of
Rosser and Schoenfeld, scattered through their separate or joint papers of
1941, 1962 and 1975, but also depends on von Mangoldt’s theorems of 1905.
The work of Rosser and Schoenfeld is remarkable in that they produced
sharp explicit results before computer-based methods became easy to use
and ubiquitous. In this chapter we take some advantage of the much greater
processing speed available in the early twenty-first century to extend the
ranges for which computer verification of inequalities is practicable. We also
are able to take advantage of the much greater height, denoted by H, below
which all the zeros of ζ(s), with positive imaginary part, have real part σ = 12 .
Results are expressed in terms of key parameters, regarded as constants, but
which are likely to change
 in time with further
 improvements.
Recall that θ(x) := n≤x log p and ψ(x) := pm ≤x log p. In Section 3.2 tables
of values of b and b > 0 are derived such that
x ≥ eb =⇒ |ψ(x) − x| < b x,
where e is the exponential number. This represents a recalculation of the table
developed by Rosser and Schoenfeld in 1962 [145, table I], with improved

40
 3.2 Constructing Tables of Bounds for ψ(x) 41

parameters and some more minor improvements. The range is increased from
b = 5000 to b = 10 000, but the vital tool is an algorithm to compute b given
b, rather than the table.
In Section 3.3 some details are given of the simple strategy used to derive
exact results from inequalities using floating-point arithmetic. In Section 3.4
estimates are given for θ(x) of the form
x
|θ(x) − x| <  ,
log x
with x in a range depending on , and also the useful bounds 0.8x ≤ θ(x) ≤
1.02x valid for x ≥ 121. The inequality θ(x) < 1.000081x, used by some
authors, was able to be proved, but only by assuming RH. This proof is
included in the final section, as are some useful bounds for ψ(x) − θ(x).
Many authors simply assume these estimates, especially those from [145].
However, the work presented here, especially that in Chapters 5 and 7, is
totally dependent on their validity, so sufficient detail of derivations is given
in this chapter for readers to reproduce results, or to set about improving
them.
The RHpack (see Appendix B) functions ComputeKm, PsiSmallXEp-
silon, PsiLargeXEpsilon, PsiLargeXEpsilon2 and VerifyThetaX relate to the
material in this chapter.

3.2 Constructing Tables of Bounds for ψ(x)

The first lemma, Lemma 3.1, gives two bounds for integrals that are used in
the derivation of the b , the first bound rather complex and the second simpler.
We used the more complex form in computations, but the simpler form could
find theoretical uses. Note that in their 1975 paper, Rosser and Schoenfeld use
modified Bessel functions to find better upper bounds for the integrals, and
their method was implemented by Dusart as reported in [49]. Those results
are not needed for the RH equivalences developed in later chapters.
The parameter R is described in Section 2.3: if ρ = β + iγ is an imaginary
zero of ζ(s), with γ > 0, then
1
β < 1− .
R log γ

Lemma 3.1 [144, lemma 19] If m ≥ 1 is a natural number with real x

satisfying
0 < 2 log x < Rm log2 h,
then for h > 2πe, if we define
φm (t) := t−m−1 e−(log x)/(R log t) ,
42 Estimates

we have
 
h
 ∞  t  1 + m log
2π
φm (t) log dt <  
h 2π (1 + m log(h/2π)) log x
2
mh xm 1/(R log h) 1−
log(h/2π)Rm2 log2 h
=: B1 (m, x, h).

If however we have 0 < 4 log x < Rm log2 h, again with h ≥ 2eπ, then the bound
takes a simpler form:
 
he
 ∞  t  2 log
2π
φm (t) log dt < .
h 2π mh x log h)
m 1/(R

∞
Proof First integrate I(h) := h φm (t) log(t/2π) dt by parts, noting that
   t 
 log t 1 + m log
2π dt = − 2π ,
tm+1 m2 tm
to get
    
h
1 + m log  ∞ 1 + m log t log x  
2π 2π   log t φ (t) dt.
I(h) = 2 m 1/(R log h) + t m
mh x h Rm2 log2 t log 2π
2π
Therefore
  ⎛   ⎞
h ⎜⎜⎜ h ⎟⎟⎟
1 + m log ⎜⎜⎜ 1 + m log log x ⎟⎟⎟
2π ⎜ 2π ⎟
I(h) < 2 m 1/(R log h) + ⎜⎜⎜⎜   ⎟⎟⎟⎟ I(h).
mh x ⎜⎜⎜ h ⎟⎟⎟
⎝ Rm2 log2 h log ⎠
2π
Solving for I(h) gives the stated upper bound. Note that the condition on log x
ensures the denominator of the upper bound is positive. The more stringent
condition ensures the bracketed term in the denominator is greater than 12
and the numerator less than m log(he/2π). This completes the proof of the
lemma. 

In Chapter 4 a simplified version of the following development is given,

sufficient for what is needed to prove one of the equivalences to RH presented
there. Here, in order that Tables 3.2 and 3.3 might be derived, the additional
integer parameter m, used in Lemma 3.1, is needed. In addition, we replace
 3.2 Constructing Tables of Bounds for ψ(x) 43

ψ(x) − x with an approximating function η(x) which simplifies the estimation

problem. Let
 
1 1
η(x) := ψ(x) − x + log(2π) + log 1 − 2 .
2 x
The basic definition needed here is an extension of E(x, h), which will be
defined in the proof of Lemma 4.3, to all integers m ≥ 1:
 h  h
Em (x, h) := ··· η(x + y1 + · · · + ym ) dy1 · · · dym .
0 0

Lemma 3.2 [146, lemma 8, p. 256] If x > 1, m is a positive integer and

0 < δ < (1 − 1/x)/m, we have
−1 x ≤ η(x) ≤ 2 x,
where
Em (x, −xδ) mδ Em (x, xδ) mδ
1 := + and 2 := + .
(−1) x δ
m+1 m+1 m 2 xm+1 δm 2
Proof (1) First we need two definitions:
Em (x, h) nhα
fm,n,α (x, h, z) := + − zhα−1 ,
hn 2
Em (x, h) mh
fm (x, h, z) := fm,m,1 (x, h, z) = + − z.
hm 2
Also assume x > 1 and x+mh > 1. The variable h could be negative; indeed we
will later choose h = ±δx. In the properties below, (2) and (3) are immediate
and (4) can be deduced by induction on n ≥ 1 using (3) and (4) applied to the
integrand
 on the right.
h
(2) fm,1,α (x, h, z) dz = Em (x, h).
0 h
(3) fm,n,α (x, h, y1 + · · · + yn ) dyn = fm,n−1,α+1 (x, h, y1 + · · · + yn−1 ).
0  h  h
(4) Em (x, h) = ··· fm,n,α (x, h, y1 + · · · + yn ) dyn · · · dy1 .
0 0
(5) If 0 < h there is a z with 0 ≤ z ≤ mh such that η(x + z) ≤ fm (x, h, z); if not
we would have by (4)
 h  h
Em (x, h) = ··· η(x + y1 + · · · + ym ) dy1 · · · dym
0 0
 h  h
> ··· fm (x, h, y1 + · · · + ym ) dy1 · · · dym = Em (x, h),
0 0

which is false.
44 Estimates

(6) If 0 > h there is a z with 0 ≥ z ≥ mh such that η(x + z) ≥ fm (x, h, z); if not
we would have
 0  0
(−1) Em (x, h) =
m
··· η(x + y1 + · · · + ym ) dy1 · · · dym
h h
 0  0
< ··· fm (x, h, y1 + · · · + ym ) dy1 · · · dym
h h
= (−1) Em (x, h).
m

(7) If δ and x satisfy 0 < δ < (1 − 1/x)/m, then −1 x ≤ η(x) ≤ 2 x. To see
this let h = δx with δ > 0. By (5) and the definition of 2 , there is a z ∈ [0, h]
such that
Em (x, δx) mδx
η(x + z) ≤ + − z ≤ x(1 + 2 ).
δm xm 2
Therefore, since z ≥ 0 and − 12 log(1 − 1/(x + z)2 ) ≤ − 12 log(1 − 1/x2 ) we get
 
1 1
ψ(x) ≤ ψ(x + z) ≤ x(1 + 2 ) − log(2π) − log 1 − 2 .
2 x
In other words, η(x) ≤ 2 x. The proof of the lower bound follows in a similar
manner. 
Now define for m ≥ 1
 xβ−1
V(m, x) := ,
ρ
|γ|m+1

where the sum is over all complex zeros ρ = β + iγ of ζ(s).

Lemma 3.3 [144, theorems 15 and 16] Let m ≥ 1, x > 1, 0 < δ < (1 − 1/x)/m,
and suppose also that
δ ≥ 2V(m, x)1/(m+1) .
Then if the real number θ satisfies
 
δ bm (δ)
θ≥ +m ,
2 2m
m m
where bm (δ) := j=0 j
(1 + jδ)m+1 , we have max(1 , 2 ) < θ.
Proof (1) To begin we use Theorem 2.2 which is for x > 1
 x
x2  xρ+1 
∞
x−2n+1 ζ  (0) ζ  (−1)
ψ(t) dt = − − − x+ .
0 2 ρ
ρ(ρ + 1) n=1 2n(2n − 1) ζ(0) ζ(−1)
 3.2 Constructing Tables of Bounds for ψ(x) 45

First integrate the definition of η(x) using ζ  (0)/ζ(0) = log(2π) from

Lemma 2.5:
 h  h  h
η(x + z) dz = ψ(x + z) dz − (x + z) dz
0 0

0
 
1 h 1
+ log 1 − dz + h log(2π)
2 0 (x + z)2
 x+h   
1 2 1 x+h 1
= ψ(u) du − hx − h + log 1 − 2 du
x 2 2 x u
+ h log(2π)
 xρ+1 − (x + h)ρ+1
= .
ρ
ρ(ρ + 1)

Thus
 h  xρ+1 − (x + h)ρ+1
η(x + z) dz = .
0 ρ
ρ(ρ + 1)

We claim further integrations lead to

⎛ m   ⎞
 xρ+m ⎜⎜⎜ m ⎟⎟⎟
Em (x, ±xδ) = ⎜⎜⎜ (−1) j+m+1
(1 ± jδ) ⎟⎟⎟⎠ .
ρ+m
ρ(ρ + 1) · · · (ρ + m) j=0⎝ j
ρ

To see this, use the absolute convergence of ρ 1/|ρ|2 , and integrate twice to
get
 h h
η(x + y1 + y2 ) dy1 dy2
0 0
 h
(x + y2 )ρ+1 − (x + y2 + h)ρ+1
= dy2
0 ρ
ρ(ρ + 1)
  h 
1
= (x + y2 )ρ+1 − (x + y2 + h)ρ+1 dy2
ρ
ρ(ρ + 1) 0
 1  h
= (x + y2 )ρ+2 − (x + y2 + h)ρ+2
ρ
ρ(ρ + 1)(ρ + 2) 0

 xρ+2 − 2(x + h)ρ+2 + (x + 2h)ρ+2

= − .
ρ
ρ(ρ + 1)(ρ + 2)

Continuing in this manner and using induction, after m integrations followed

by a substitution h = ±δx, we obtain the given formula for Em (x, ±δx).
(2) Now we derive the bound
|Em (x, ±xδ)| < xm+1 ((1 + δ)m+1 + 1)V(m, x).
46 Estimates

Take absolute values of the result of Step (1) using the upper bound
 
 xρ+m  < x
β+m

 ρ(ρ + 1) · · · (ρ + m)  |γ|m+1

and the bound |(1 ± jδ)ρ+m | < (1 + jδ)m+1 , to get

 ⎛ m   ⎞
 xρ+m ⎜⎜⎜ ⎟⎟
ρ+m ⎟
|Em (x, ±xδ)| =  ⎜⎜⎜ (−1) j+m+1 m
(1 ± jδ) ⎟⎟⎟
 ρ ρ(ρ + 1) · · · (ρ + m) ⎝ j=0 j ⎠
m    xβ−1
m
<x m+1
(1 + jδ)m+1 .
j=0
j ρ
|γ|m+1

Hence
 xβ−1
|Em (x, ±xδ)| ≤ xm+1 bm (δ)
ρ
|γ|m+1
= xm+1 bm (δ)V(m, x).
(3) Since we are given δ ≥ 2V(m, x)1/(m+1) we have
V(m, x) 1
≤ m+1 ,
δm+1 2
and therefore
Em (x, xδ) mδ
2 = +
xm+1 δm 2
xm+1 bm (δ)  xβ−1 mδ
< m+1 m +
x δ ρ
|γ|m+1 2
 
δ V(m, x)
= 2bm (δ) m+1 + m
2 δ
 
δ bm (δ)
≤ +m .
2 2m
The proof for 1 is similar. 
Recall the definition φm (γ) = exp(−log x/(R log γ))/|γ|m+1 . Also note that
we are designating H as a fixed height below which all complex zeros of ζ(s)
are on the critical line, and
 xβ−1
V(m, x) := ,
ρ
|γ|m+1

where the sum is over all non-trivial zeros ρ = β + iγ of ζ(s).

The next lemma shows how the value of H and an explicit zero-free region
using the parameter R can be used to obtain an upper bound for V(m, x).
 3.2 Constructing Tables of Bounds for ψ(x) 47

Lemma 3.4 For x > 1 and m ≥ 1 we have

1  1 
V(m, x) ≤ √ + φm (γ).
x ρ |γ| m+1
H<γ

Proof Splitting the sum in the definition of V(m, x) we get

 xβ−1  xβ−1
V(m, x) = + .
1
|γ|m+1 1
|γ|m+1
β≤ 2 β> 2

For the first sum

 xβ−1 1  1 1  1
≤ √ ≤ √ ,
1
|γ|m+1 x 1 |γ|m+1 x ρ |γ|m+1
β≤ 2 β≤ 2

and for the second

 xβ−1  xβ−1
= 2 .
1
|γ|m+1 1
|γ|m+1
β> 2 β> 2 , γ>0

In addition, by Theorem 2.4, for γ > H we have

 
log x
exp −
xβ−1 R log γ
< = φm (γ),
γ m+1 γ m+1

and so
 xβ−1  
< 2 φ m (γ) ≤ φm (γ).
1
|γ|m+1 1 γ>H
β> 2 β> 2 , γ>0

This completes the proof. 

Next we obtain an upper bound for the sums of the φm (γ). Recall the
improvement to Backlund’s result, given in Section 2.2, for the difference
between the number N(T ) of zeta zeros with positive imaginary part up to
height T and its approximation F(T ) = (T/2π) log(T/2πe) + 7/8, namely
B(T ) := a1 log T + a2 loglog T + a3 ,
where, using the best available values at the time of writing, a1 = 0.122, a2 =
0.278 and a3 = 2.510 for T ≥ e.
Lemma 3.5 [145, theorem 27] If 2 ≤ h then
  ∞  t 
−(log x)/(R log h)
φm (γ) < P(h)e + Q(h) φm (t) log dt,
h<γ h 2π
48 Estimates

where
a1 log h + a2 loglog h + a3
P(h) := 2
hm+1
and
1 a1 log h + a2
Q(h) := + .
2π h log h log(h/2π)
Proof By Lemma 4.2 (which relies simply on Lemma 4.1, a generalized
form of Abel summation) and Backlund’s bound B(T ), we can write
  ∞  t    
1 a2 1
φm (γ) < log + a1 + φm (t) dt + 2B(h)φm (h).
h<γ h 2π 2π log t t

But for h ≤ t we have

 t 
h log 1 1
≤  2π  and  t ≤  .
t h t log h
log 2π h log
2π 2π
Therefore
   ∞  t 
1 a1 log h + a2
φm (γ) < + φm (t) log dt + 2B(h)φm (h)
h<γ
2π h log h log(h/2π) h 2π
 ∞  t 
= P(h)e−(log x)/(R log h) + Q(h) φm (t) log dt.
h 2π
This completes the proof. 

For m ≥ 1 let km (h) := ρ,γ>h 1/γ . Now we can combine Lemmas 3.4
m+1

and 3.5 to obtain an upper bound for V(m, x). The reader might compare this
with [145, lemma 6]. Recall that we have defined B1 (m, x, H) in the proof of
Lemma 3.1:
 
h
1 + m log
2π
B1 (m, x, H) :=    .
1 + m log (h/2π) log x
2 m
mh x 1/(R log h) 1−
log (h/2π) Rm2 log2 h
Lemma 3.6 Let 0 < 2 log x < Rm log2 H. Then
 
km (0) P(H)
V(m, x) < √ + 1/(R log H) + Q(H)B1 (m, x, H) =: B2 (m, x, H).
x x
Next recall the definition:
 
1 1
η(x) := ψ(x) − x + log(2π) + log 1 − 2 , x > 1.
2 x
 3.2 Constructing Tables of Bounds for ψ(x) 49

Finally, putting all of these derivations together with Lemma 3.3, we

are able to prove the following theorem, which will enable us to derive
Tables 3.2 and 3.3.
Theorem 3.7 [144, theorem 21] If positive real numbers δ, , a and x and
integer m ≥ 1 are such that
 
δ [(1 + δ)m+1 + 1]m
(δ) := +m ,
2 2m
a > 1 + mδa,
2 log a < Rm log2 H,
δ ≥ 2B2 (m, a, H)1/(m+1) ,
then for all x ≥ a we have − x < η(x) <  x.
Proof Note that for x ≥ a we have V(m, x) ≤ V(m, a) so
 δ m+1
V(m, x) ≤ V(m, a) < B2 (m, a, H) ≤
2
and we can use Lemma 3.2 with i = (δ) and the rather gross estimate bm (δ) <
[(1 + δ)m+1 + 1]m . 
In deriving Tables 3.2 and 3.3, we used Rosser’s 1941 value of R and a
reasonably large value of H for 2014, namely H = 5.4 × 108 ; see the 1986
entry of Table 2.2 from van de Lune, te Riele and Winter. The values of km (0),
which occur in Table 3.1, were obtained in a manner similar to those in [144,
p. 225], which were based on [143, lemma 8]. The bound for k1 (0) comes
from Lemma 2.10. Those with m > 1 are computed by evaluating the first
100 terms 1/γm+1 numerically and bounding the remainder using the bound
for km−1 (0), which had been computed previously.
Finally, we present only a subset of the values of b that are required.
A copy of the Mathematica program used to compute the successive km (0)
values, ComputeKm, is part of RHpack – see Appendix B.
Of course, Theorem 3.7 gives an  that works for the function η and table I
of [145] is in terms of ψ(x), so a small adjustment needs to be made. Indeed,
we have shown that
log 2π 1 ψ(x) log 2π 1
− x ≤ η(x) ≤  x =⇒ 1 −  − + 2≤ ≤ 1+ − + 2.
x 2x x x x
Hence if log x ≥ log a = b we need to replace  with
 
log 2π 1 log 2π 1
  := max  + − ,  − + .
eb 2e2b eb e2b
Thus, with this revised  denoted b we obtain Tables 3.2 and 3.3 where the
row b gives the value of b, the value of m giving the smallest , and the
50 Estimates
Table 3.1 Upper bounds used for km (0), 1 ≤ m ≤ 28.

m km (0) bound m km (0) bound

1 0.046300 2 0.0016700
3 0.000074400 4 4.46301 × 10−6
5 2.88354 × 10−7 6 1.93507 × 10−8
7 1.32606 × 10−9 8 9.19838 × 10−11
9 6.42733 × 10−12 10 4.5113 × 10−13
11 3.17539 × 10−14 12 2.23908 × 10−15
13 1.58067 × 10−16 14 1.11669 × 10−17
15 7.89294 × 10−19 16 5.58061 × 10−20
17 3.94653 × 10−21 18 2.79132 × 10−22
19 1.97443 × 10−23 20 1.3967 × 10−24
21 9.88051 × 10−26 22 6.98986 × 10−27
23 4.94499 × 10−28 24 3.49838 × 10−29
25 2.47499 × 10−30 26 1.75098 × 10−31
27 1.23877 × 10−32 28 8.76397 × 10−34

Table 3.2 Values of b, m and b .

b m b

10 1 0.03604387616
11 1 0.02792600226
12 1 0.02166958397
13 1 0.01683159959
14 1 0.013082730087
15 1 0.010173897079
16 1 0.007914814691
17 1 0.006159239110
18 1 0.004794353324
19 1 0.003732913461
20 1 0.002907340379
25 1 0.0008427406872
30 1 0.0002748595135
35 1 0.0001515932079
40 1 0.0001348313906
45 2 0.0001315791411
50 2 0.00008556326873
55 2 0.00003720592016
60 2 0.000016278904663
65 2 7.594348788 × 10−6
75 2 4.457990402 × 10−6
80 2 4.377481479 × 10−6
85 2 4.330519430 × 10−6
90 2 4.286798206 × 10−6
95 3 2.5990044920 × 10−6
100 3 1.4581963574 × 10−6
 3.3 Exact Verification Using Computation 51
Table 3.3 More values of b, m and b .

b m b

200 5 2.3461827847 × 10−7

300 8 1.1237935640 × 10−7
400 11 8.578404834 × 10−8
500 14 7.743154524 × 10−8
600 15 7.3869596744 × 10−8
700 14 7.1000004141 × 10−8
800 14 6.8129091697 × 10−8
900 13 6.5258532748 × 10−8
1 000 13 6.2436274496 × 10−8
2 000 7 3.4825165615 × 10−8
3 000 3 1.1625210647 × 10−8
4 000 3 2.563076921 × 10−9
5 000 4 1.0662477419 × 10−9
6 000 4 3.162913973 × 10−10
7 000 5 2.252110357 × 10−10
8 000 5 8.153215009 × 10−11
9 000 6 7.617681319 × 10−11
10 000 7 7.415233861 × 10−11

corresponding value of b so that

|ψ(x) − x| ≤ b x, x ≥ eb .
We also present plots of the value of b as a function of b in three ranges,
corresponding to the tables going up to b = 104 . Figures 3.1, 3.2 and 3.3 show
how the optimal value of m changes in the mid-range.
The program used to compute the b is PsiSmallXEpsilon, which is part of
RHpack – see Appendix B.
A recent article by Faber and Kadiri [57, corollary 1.2] uses the Platt value
H = 3.061 × 1010 to derive, for example,
|ψ(x) − x| ≤ 5.3688 × 10−4 x, x ≥ e20 .
The estimates for ψ(x) derived here are sufficient for our purposes.

3.3 Exact Verification Using Computation

In verifying estimates over a finite range, both in this chapter and in what
follows, we often use a computational method, not always giving details.
These computations are very useful since they sometimes simplify proofs,
and sometimes make covering the infinite part of variable ranges easier.
We give here a description of the approach that has been taken. It might be
regarded as a very simple form of interval arithmetic.
52 Estimates

0.04
m=1

0.03

0.02

0.01

b
0 20 40 60 80 100

Figure 3.1 The values of b for 10 ≤ b ≤ 100 and 1 ≤ m ≤ 5.

7. × 10–7

6. × 10–7

5. × 10–7

4. × 10–7

3. × 10–7

—m=3
2. × 10–7

1. × 10–7

b
0 200 400 600 800 100

Figure 3.2 The values of b for 100 ≤ b ≤ 1000 and 3 ≤ m ≤ 7.

(1) Suppose we wish to test the validity of the bound


n
ai < f (n), 1 ≤ n ≤ N0 , (3.1)
i=1

where (ai ) is a given real sequence and f (n) is a real-valued function.

The values of each element of the sequence and of f (n) are only known
 3.3 Exact Verification Using Computation 53

5. × 10–8

4. × 10–8

3. × 10–8

2. × 10–8
m=7

1. × 10–8

b
0 2000 4000 6000 8000 10 000

Figure 3.3 The values of b for 103 ≤ b ≤ 104 and 3 ≤ m ≤ 7.

approximately. We set a target for success


n
ai + 0 < f (n), 1 ≤ n ≤ N0 .
i=1

To achieve that target we need to approximate f (n) and the terms of (ai ) with
a “local accuracy”  > 0 so we can calculate rationals (bi ) and g(n) such that
|ai − bi | ≤  and | f (n) − g(n)| ≤  for all 1 ≤ i, n ≤ N0 . Then we need to have,
for each 1 ≤ n ≤ N0 ,

n 
n
0 ≤ f (n) − ai ≤ g(n) +  − bi + n,
i=1 i=1

so we should test 1 := 0 − (n + 1) ≤ g(n) − ni=1 bi , and that should be
sufficient to verify (3.1) exactly.
For example, if we wanted to show that θ(x) < x up to x = 1011 with
accuracy 10−10 we could set the local accuracy to  = 10−23 and define
an = log n for prime n and zero otherwise.
(2) Continuing with this example, the equation θ(n) < n will have been
verified at integer values n of x in the given range. Since θ(x) is constant
between integers and f (x) = x is increasing, evaluation at integers is sufficient
to show it is true for all real numbers x up to N0 .
(3) In case the inequality is similar, for example, to 0.8x < θ(x) for 0 < x ≤
106 say, then the process adopted is similar, except we verify the equation,
using the approach of Step (1), 0.8(n + 1) < θ(n) at integers n. Then, since
54 Estimates

both sides are increasing, if n < x < n + 1 and n is an integer, we have

0.8x < 0.8(n + 1) < θ(n) ≤ θ(x),
and the equation has been verified for all real x in the given range.

3.4 Estimates for θ(x)

The next theorem provides estimates for θ(x) and ψ(x) in an unbounded
range. It depends on a function (x), defined in the statement, and depends
on x being sufficiently large.
The proof, given by Rosser and Schoenfeld in 1962 in [145] as theorem 11
(with a different conclusion from that stated here), but based in part on the
proofs from 1938 in [143], is by dividing the range into sections and using a
variety of lemmas. In problem (1) at the end of this chapter we suggest that
a different and simpler proof should be possible, with potentially a different
and smaller form for the upper bound.
Here we have used an improved value for the lower bound for the range of
values of R, from R ≥ 18 to R ≥ 8, at the cost of a small increase in the leading
coefficient. This enables us to use R = 8 in calculations; see Section 2.5.
Recall the definitions of km (h), N(T ) and F(T ):
 1
km (h) := ,
ρ, γ>h
γm+1
N(T ) := #{ρ : 0 < ρ ≤ T },
T  T  7
F(T ) := log + ,
2π 2πe 8
and |N(T ) − F(T )| < 1.12 log T + 9.5 for T ≥ 1450, which is Rosser’s
improvement of Backlund’s bound from 1938 [143, lemma 5]. For other
bounds and ranges see part (11) of Section 2.2.
Theorem 3.8 Let
⎛  ⎞
 ⎜⎜⎜ log x ⎟⎟⎟
(x) := 2 log x exp ⎜⎜⎝− ⎟⎟ ,
R ⎠
where 8 ≤ R ≤ 18. Then for all x ≥ e140 we have
|ψ(x) − x| < (x)x and |θ(x) − x| < (x)x.
Proof (1) First we have some definitions. Let
 1
S 8 := < 3.32 × 10−4 < (0.274)8e−8 ,
γ 2
1469≤γ≤e8
 1
S n := , n ≥ 9,
γ2
n−1
e <γ≤e
n
 3.4 Estimates for θ(x) 55

k1 (0) 
∞
1.9 3
1 (x) := √ + S n x−1/Rn + + 2,
x n=9
x x
∞  
log x
2 (x) := S n exp − ,
n=9
Rn
k1 (0) 1.9 3
3 (x) := √ + + 2 so 1 = 2 + 3 .
x x x
(2) Next we claim that for S r , when r ≥ 9 we have S r < (0.274)r exp(−r).
To see this, define

Δ := F(er ) − F(er−1 ),

so Δ = er−1 (e − 1)(r − a)/(2π) where a := log 2π + 1 − 1/(e − 1) > 2.25. We also

define for r fixed g(x) := F(xer−1 ), so g(x) is increasing and concave upwards
on [1, e], and so
 
e−1
g 1+ n ≤ g(1) + n, 1 ≤ n ≤ Δ,
g(e) − g(1)
which translates to
 
2πn
F e r−1
+ ≤ F(er−1 ) + n.
r−a
Therefore, since N(er ) ≤ F(er ) + 9.5 + 1.12r, if we let M be n plus that upper
bound for N(er ) we must have
2πn
er−1 + < γM
r−a
for 0 ≤ n ≤ Δ. But, considering the lower bound for N(er−1 ),

m < F(er−1 ) − 9.5 − 1.12(r − 1) =⇒ γm < er−1 .

Hence
 1
Sr =
γ2
er−1 <γ≤e
r

19 + 2.24r 
Δ
1
< +  2
e2r−2 2πn
n=1
+ er−1
r−a
1 
Δ
2.24(r + 9) 1
< +  2
e2r−2 e2r−2 n=1 (e − 1)n
1+
Δ
56 Estimates
 Δ
2.24(r + 9) 1 dt
< 2r−2
+ 2r−2  2
e e 0 e−1
1+ t
Δ
2.24(r + 9) Δ
= 2r−2
+ 2r−1
e e
2.24(r + 9) (r − a)(e − 1)
= +
e2r−2 2πer
< 16.6(r + 9)e−2r + 0.274(r − 2.25)e−r < 0.274re−r ,
where the last inequality is true for r ≥ 9. This completes the proof of the
claim.
(3) In this step we focus on the term that depends on the zeta zeros ρ where
ρ = β + iγ, namely 2 (x). Using the result of Step (2) we have
∞   ∞
log x
2 (x) = S n exp − < (0.274) ne−n x−1/Rn .
n=9
Rn n=9

Next we claim that for all y > 0 and n ≥ 1 we have

 
n √ 1 1 √
n+y/n
< y + + √ e−2 y . (3.2)
e 2 8 y
To see this, let
n  y √ 
f (y, n) := √ exp −n − + 2 y
y n
√ √
and regard y and n as real variables. Note that f (y, y ) = 1 and let n1 = y.
We get
√ 2
∂ f (y, n) e−(n− y) /n
= √ (n − n2 + y),
∂n n y

which for n ≥ 1 is zero at the unique value n2 := (1 + 4y + 1 )/2. Also, at
n = n2 we get
⎛ √  ⎞
  ⎜⎜⎜ (−2 y + 4y + 1 + 1)2 ⎟⎟⎟
4[2y( 4y + 1 + 2) + 4y + 1 + 1] exp ⎜⎝−⎜  ⎟⎟⎠
∂2 f (y, n) 2( 4y + 1 + 1)
=− √  ,
∂n2 y ( 4y + 1 + 1)3
which is negative. Thus for fixed y, n2 gives the maximum value for f (y, n).
Then, setting h := n2 − n1 and using convexity and a linear approximation
we get
√
∂ f (y, y)
f (y, n1 + h) ≤ f (y, n1 ) + h
∂n
 3.4 Estimates for θ(x) 57
⎛  ⎞  1/2
1 ⎜⎜⎜⎜ 1 + 4y + 1 √ ⎟⎟⎟⎟ 1 1
= 1 + √ ⎜⎝ − y⎟⎠ = √ + 1 +
y 2 2 y 4y
1 1
< 1+ √ + .
2 y 8y
Therefore
 
n √ √
−2 y √ 1 1 √
= f (y, n) y e < y + + √ e−2 y ,
en+y/n 2 8 y
which completes the proof of the claim.
Setting y := (log x)/R, it follows that for n ≥ 1, 8 ≤ R ≤ 18 and x ≥ e140 , we
have
⎛ ⎞
⎜
⎜ ⎟⎟⎟ −2 √(log x)/R
ne−n x−1/Rn ≤ ⎜⎜⎜⎝
log x 1 1 ⎟⎟ e
+ + 
R 2 8 (log x)/R ⎠

log x −2 √(log x)/R
< 1.2 e . (3.3)
R
Returning to the sum for 2 (x) given in Step (1), we can now write
 
2 (x) = (0.274) ne−n x−1/Rn + (0.274) ne−n x−1/Rn
√ √
9≤n≤2 (log x)/R n>2 (log x)/R
  ⎛  ⎞
log x log x ⎜⎜⎜ log x ⎟⎟⎟
< (0.274)2 (1.2) exp ⎜⎜⎝−2 ⎟⎟
R R R ⎠
 
log x ∞
+ (0.274) te−t dt
R 2 √(log x)/R
 ⎛ ⎞ ⎛  ⎞
log x ⎜⎜⎜⎜ 1 1 ⎟⎟⎟ ⎜⎜⎜
⎟⎟⎠ exp ⎜⎜⎝−2 log x ⎟⎟⎟
⎟⎟
< (2×0.274) ⎜1.2 + 
⎝ +
R (log x)/R 2(log x)/R R ⎠
⎛  ⎞
log x ⎜⎜⎜ log x ⎟⎟⎟
< (0.8895) exp ⎜⎜⎝−2 ⎟⎟ ,
R R ⎠
∞
where we have used the integral b t exp(−t) dt = (b + 1)e−b , and the final
inequality follows for x ≥ exp(140). Finally in this step, using the bound we
have derived for 2 (x) and that for 3 (x) from Step (1), for x ≥ e140 we get,
since in addition R ≥ 8,
⎛  ⎞
1 ⎜⎜⎜ log x ⎟⎟⎟⎟
1 (x) = 2 (x) + 3 (x) < log x exp ⎜⎝−2⎜ ⎟. (3.4)
8 R ⎠
(4) By Theorem 2.2 we have for x > 1
 x
x2  xρ+1 
∞
x−2n+1 ζ  (0) ζ  (−1)
ψ1 (x) := ψ(t) dt = − − − x+ .
0 2 ρ
ρ(ρ + 1) n=1 2n(2n − 1) ζ(0) ζ(−1)
58 Estimates

Therefore
 
 x2  x2 x2
ψ1 (x) −  ≤ 2x 1 (x) =⇒ − 2x2 1 (x) ≤ ψ1 (x) ≤ + 2x2 1 (x).
2
2 2 2
Therefore, if h > 0, because 1 (x) > 1 (x + h) we have
ψ1 (x + h) − ψ1 (x) < 12 (x + h)2 − 12 x2 + 2(x + h)2 1 (x + h) + 2x2 1 (x)
≤ xh + 12 h2 + 21 (x)(2x2 + 2xh + h2 ).
Discarding the positive last term in the first lower bound below
ψ1 (x) − ψ1 (x − h) ≥ 12 x2 − 2x2 1 (x) − 12 (x − h)2 + 2(x − h)2 1 (x − h)
> xh − 12 h2 − 2x2 1 (x)
> xh − 12 h2 − 21 (x)(2x2 + 2xh + h2 ).
Thus, because ψ(x) is non-decreasing we have
1 21 (x) 2 ψ1 (x) − ψ1 (x − h)
x− h− (2x + 2xh + h2 ) ≤ ≤ ψ(x)
2 h h
ψ1 (x + h) − ψ1 (x) 1 21 (x) 2
≤ ≤ x+ h+ (2x + 2xh + h2 ).
h 2 h
√
Finally, set h := 4x 1 (x) and use (3.4) to derive
√   
|ψ(x) − x| ≤ x 1 3 + 4 1 (x) + 81 (x)
⎛  ⎞
 ⎜⎜⎜ log x ⎟⎟⎟⎟
< 4x (log x)/8 exp ⎜⎝− ⎜ ⎟
R ⎠
⎛  ⎞
 ⎜⎜⎜ log x ⎟⎟⎟
= x 2 log x exp ⎜⎜⎝− ⎟⎟ ,
R ⎠
for x ≥ e140 , completing the proof of the estimate for ψ(x).
(5) Bringing the estimates together,
 
θ(x) − x ≤ ψ(x) − x ≤ 3.983x (log x)/8 exp(− (log x)/R) < (x)x.
√ √
Then, using Lemma 3.10 which gives ψ(x) − θ(x) < 1.02 x + 3 3 x (x > 0),
we get
√ √
θ(x) − x ≥ ψ(x) − x − 1.02 x − 3 3 x
  √ √
≥ −3.983x (log x)/8 exp(− (log x)/R) − 1.02 x − 3 3 x
 
≥ −4x (log x)/8 exp(− (log x)/R) = −x(x),
with the last inequality holding for x ≥ e81 . Therefore
|θ(x) − x| < (x)x, x ≥ e140 .
This completes the proof. 
 3.4 Estimates for θ(x) 59
Table 3.4 Values of b = log x, and rounded log (x) for R = 8 and R = 18.

b log (x), R = 8 log (x), R = 18 b log (x), R = 8 log (x), R = 18

1 000 −7 −3 10 000 − 30 −18

2 000 −11 −6 20 000 − 44 −28
3 000 −15 −8 30 000 − 55 −35
4 000 −17 −10 40 000 − 65 −41
5 000 −20 −12 50 000 − 73 −47
6 000 −22 −13 60 000 − 80 −52
7 000 −24 −15 70 000 − 87 −56
8 000 −26 −16 80 000 − 94 −61
9 000 −28 −17 90 000 −100 −65
10 000 −30 −18 100 000 −105 −68

Table 3.5 Larger values of b and log (x).

b = log x log (x), R = 8 log (x), R = 18

100 000 −105 − 68

200 000 −151 − 99
300 000 −186 −122
400 000 −216 −142
500 000 −243 −160
600 000 −266 −175
700 000 −288 −190
800 000 −309 −204
900 000 −328 −216
1000 000 −346 −228

In Tables 3.4 and 3.5 we give values of log (x) for a range of values of b =
log x. The second and fifth columns of Table 3.4 are computed rounding the
result of Theorem 3.8, as are the R = 8 plots in Figures 3.4 and 3.5. The third
and sixth columns
 are computed
 using the Rosser–Schoenfeld expression for
(x), namely log x exp(− (log x)/R), as are the R = 18 plots in Figures 3.4
and 3.5.
As an application of these values, first consider Table 3.5 at line b = 106 .
Since exp(−346) < 5.43 × 10−151 , we get
6
|ψ(x) − x| ≤ x × 5.43 × 10−151 , x ≥ e10 .
At the other extreme, from Table 3.4, line b = 1000,
3
|ψ(x) − x| ≤ x × (0.00092), x ≥ e10 .
60 Estimates
x

b
2000 4000 6000 8000 10 000

–5

–10

–15

—R=18
–20

–25

–30 —R=8

Figure 3.4 A plot of b, where x = eb , vs log (x) for 103 ≤ b ≤ 104 .

x
b
200 000 400 000 600 000 800 000 1 × 106
–50

–100

–150

–200
—R=18
–250

–300

–350 —R=8

Figure 3.5 A plot of b, where x = eb , vs log (x) for 105 ≤ b ≤ 106 .

This compares badly with the value obtained in Section 3.2; see Table 3.3,
which gives
3
|ψ(x) − x| ≤ x × (4.45 × 10−9 ), x ≥ e10 .

There is one more preliminary lemma in this chapter, viz. Landau’s lemma
(Lemma 3.9). We have made some small changes in his proof and result,
but the interesting parts of a fascinating proof are his. Other than this, the
remaining material consists of essential estimates, with applications spread
 3.4 Estimates for θ(x) 61

throughout the rest of the book. For example, Lemma 3.14 below is used in
Lemmas 3.10, 5.6, 7.7, 8.13 and 9.16, and in Theorems 5.25, 7.11 and 7.13.
Lemma 3.9 [105, pp. 89–90] For all x ≥ 1 we have
ψ(x) < αx + (log x + 2)(3 log x)
where
  
6 log 2 log 3 log 5 log 30
α := + + − < 1.105551.
5 2 3 5 30
Proof (1) For x ≥ 1 define

T (x) := log n = log (x!) .
1≤n≤x

Then

x−1 x−1 
 n+1
T (x) = log n + logx ≤ log t dt + logx
n
 x
n=1 n=1

≤ log t dt + log x = x log x − x + 1 + log x,

1

x x  n
T (x) ≥ log n ≥ log t dt
n=2 n=2 n−1
 x  x  x
= log t dt = log t dt − log t dt
x

1 1
x
≥ log t dt − log x > x log x − x + 1 − log x.
1

Hence there is a θ0 , which depends on x, with |θ0 | ≤ 1 such that we have the
estimate
T (x) = x log x − x + 1 + θ0 log x.
Define U(x) := T (x) − T (x/2) − T (x/3) − T (x/5) + T (x/30) and evaluate U(x)
using the estimate for T (x) to obtain the expression, for some θ1 with |θ1 | ≤ 1
and x ≥ 30,
 
log 2 log 3 log 5 log 30
U(x) = x + + − + θ1 (5 log x)
2 3 5 30
= βx + θ1 (5 log x), (3.5)
where β = 0.921292 . . . . Equation (3.5) was proved for x ≥ 30. A simple
computation shows it is also true for 1 ≤ x ≤ 30. Note that the combination of
T evaluations has been chosen to make this bound possible.
62 Estimates
x
(2) Next we claim we can write T (x) = n=1 ψ(x/n). To see this derive
     
x x
T (x) = log (x!) = log p + 2 + ···
p≤x
p p
 ∞    ∞   
x x
= log p m
= log p m ,
p≤x m=1
p m=1 m√ p
p≤ x

and so

∞ 
x  ∞  x  
x
T (x) = log p = θ m
√
m=1 n=1 p≤ m x m=1 n=1
n
x  ∞    x  x
x
= θ m = ψ .
n=1 m=1
n n=1
n

Then substitute this expression for T (x) into the definition of U(x) to get
∞     x  x  x  x   ∞  x
x
U(x) = ψ −ψ −ψ −ψ +ψ = cn ψ ,
n=1
n 2n 3n 5n 30 n=1
n

where the coefficients cn are determined by the remainder when n is divided

by 30 according to the following rules:

cn = −1 for remainders 6, 10, 12, 15, 18, 20, 24, 0,

cn = 0 for remainders 2, 3, 4, 5, 8, 9, 14, 16, 21, 22, 25, 26, 27, 28,
cn = 1 for remainders 1, 7, 11, 13, 17, 19, 23, 29.

(3) Now expand U(x) in increasing order of denominators of x to give

 x  x  x  x  x
U(x) = ψ(x) − ψ +ψ −ψ +ψ −ψ + ··· ,
6 7 10 11 12
and we note that the signs of the successive terms alternate and their sizes
decrease. A minor miracle! Hence
 x
ψ(x) − ψ ≤ U(x),
6
and therefore, since by what we have shown above in (3.5)
 x
ψ(x) − ψ < βx + 5 log x.
6
Hence, for n ≥ 0 with x ≥ 6n and x ≥ 1 we have
x  x  x x
ψ n − ψ n+1 < β n + 5 log n .
6 6 6 6
 3.4 Estimates for θ(x) 63

Now sum these expressions over n:

x/log 6 
log x  x 
ψ(x) = ψ − ψ
n=0
6n 6n+1
∞  
1 log x  
≤ βx n
+ + 1 5 log x
n=0
6 log 6
6
< βx + 3 log x (log(x) + 2).
5
This completes the proof. 
We need the following Lemma 3.10 in Chapter 4 where it is used in the
proof of Theorem 4.5. It plays an essential role in the development of the
ideas in the book, being used more than 10 times.
Lemma 3.10 [145, theorem 18]
(a) For all x with 0 < x ≤ 6.66 × 1011 we have θ(x) < x.
(b) For 6.66 × 1011 < x we have θ(x) ≤ 1.0009x.
√ √
(c) For all x > 0 we have ψ(x) − θ(x) < θ( x ) + 3 3 x.
(d) For x ≥ 121 we have 0.8x ≤ θ(x).
Proof (a) This is proved by computation; see Section 3.3.
(b) This follows directly from the b = 25 line of Table 3.2.
(c) A direct computation shows that for integers x in the range 1 ≤ x ≤ 105 ,
and thus for all real x in that range, we have
√ √
ψ(x) − θ(x) < θ( x ) + 3 3 x. (3.6)
We now extend the range for (3.6) to 105 ≤ x < 1015 . We claim that for x ≥ 1
we have θ(x) < 1.106207x. This is immediate for 1 ≤ x < 106 by the right-hand
member of (b), and for x ≥ 106 we get, using Lemma 3.9,
 
(3 log x)(log x + 2)
θ(x) ≤ ψ(x) ≤ x 1.105551 + < 1.106207x.
x
Using this result, θ(x) < 1.106207x, x > 0, and the identity

∞
√
ψ(x) = θ( n x ), (3.7)
n=1

we can write
log
√
x/log 2

ψ(x) − θ(x) − θ( x ) ≤ 1.2x 1/3

x1/n−1/3 < 3x1/3
n=3

when x ∈ [10 , 10 ].
5 18
64 Estimates

We can now complete the proof of (3.6) in the range x ≥ 1018 . Using (3.7)
and (a) and (b) again, we can write for x ≥ 1018
√ √  √
ψ(x) − θ(x) − θ( x ) = θ( 3 x ) + θ( n x )
n≥4
√ √
≤ θ( 3 x ) + θ( 4 x )(log x/log 2 − 3)
 
√3 log x/log 2 − 3 √
< 1.0009 x 1 + √
12
< 3 3 x.
x
This completes the proof of (c).
(d) A numerical check shows that for integers x in the range 0 < x ≤ 1010
we have
√
x − 2.05282 x < θ(x) < x. (3.8)
√
Thus, for 0 < x ≤ 1010 , by (3.8) we have x − 2.05282 x < θ(x), and so for
121 ≤ x ≤ 1010
 
2.05282
0.813337x ≤ x 1 − √ < θ(x).
x
√ √
By (c) we have ψ(x) − 1.0009 x − 3 3 x < θ(x) for x > 0, and by Table 3.2,
x(1 − 0.0361) < ψ(x) for x > 1010 . Therefore, for x ≥ 1010
 
1.0009 3
0.8x < x 1 − 0.0361 − √ − √6 < θ(x)
x x
which completes the proof. 
A note regarding part (a) is needed here. It is known, see Platt and Trudgian
[135, theorem 1], that θ(x) < x for 2 < x < 1.39 × 1017 , improving on a result
by Schoenfeld [153, p. 360] (see also [142, (4), p. 194]). In addition Platt
and Trudgian showed that for all x > 0 we have θ(x) < (1 + 7.5 × 10−7 )x.
The paper [135] also contains the result (corollary 2), which is then an easy
consequence,
√ √
ψ(x) − θ(x) < (1 + 7.5 × 10−7 ) x + 3 3 x, x > 0.
The note [153, p. 360] states that for all 0 < x we have θ(x) < 1.000081x. This
was not proved by Schoenfeld in that paper, but is the subject of a final note,
with details to follow. However, the details have not been found. We have a
simple derivation of the estimate, using RH, which is all that is needed.
Lemma 3.11 If we assume RH, then for 0 < x we have θ(x) < 1.000081x.
Proof Since by Lemma 3.10, θ(x) < x for 0 < x ≤ 1011 , the upper bound
for θ(x) holds in that range. By Theorem 4.6, which depends on RH,
 3.5 More Estimates 65

we have
√  
x log2 x log2 x
θ(x) − x ≤ =⇒ θ(x) ≤ x 1 + √ < 1.000081x
8π 8π x
for x ≥ 1011 . 
Using the same approach as for Lemma 3.10 (d), we obtain the following
list of lower bounds for θ(x):
Lemma 3.12 [145, theorem 10] We have θ(x) ≥ βx for x ≥ α when (α, β) are
in the list:
{(89 387, 0.995), (32 057, 0.990), (11 927, 0.985), (7481, 0.980),
(5381, 0.975), (3457, 0.970), (2657, 0.965), (1481, 0.960), (1433, 0.955),
(1427, 0.950), (853, 0.945), (809, 0.94), (599, 0.93), (557, 0.92),
(349, 0.91), (227, 0.89), (149, 0.86)}.

3.5 More Estimates

We also need the estimates for ψ(x) − θ(x) which are given in Lemma 3.13.
Compare these with [145, theorems 13, 14, 19 and 24].
Lemma 3.13
√
(a) For 1423 ≤ x ≤ 108 we have √x − 2 x < θ (x).
(b) For 121 < x ≤ 1016 we have x < ψ(x) − θ(x).
√
(c) For x ≥ 1013 we have ψ(x)√− θ(x) < 1.021 x.
(d) For 121 ≤ x we have 0.98 x < ψ(x) − θ(x).
√
(e) For 0 < x we have ψ(x) − θ(x) < 1.42620 x.
Proof (a) This may be verified by computation – see Section 3.3.
(b) For x√≤ 105 we can verify the equation numerically. For 105 ≤ x ≤ 108
use x − 2.1 x < θ(x), which is a consequence of√the proof of Lemma 3.10 –
see (3.8). Finally when 108 ≤ x ≤ 1016 we have x ≤ 108 so we can write
 √ √ √ √ √ √ √
ψ(x) − θ(x) ≥ θ( n x ) ≥ ( x − 2.1 4 x ) + ( 3 x − 2.1 6 x ) + ( 4 x − 2.1 8 x )
n≥2  
√ √3 1.1 2.1 2.1
≥ x + x 1 − 12 √ − √6 − 5/24
√ x x x
> x
since x ≥ 108 . √ √
1013 ≤ x ≤ 1022 use θ( x ) < x and (3.6), namely ψ(x) − θ(x) <
(c) For √
√
θ( x ) + 3 3 x. For x > 1022 use θ(x) < 1.02x, from Lemma 3.10(b), so that in
that case we have
 
√ 3 √
ψ(x) − θ(x) < x 1.02 + √6 < 1.021 x.
x
66 Estimates

(d) Up to 1016 the result follows by (b). √ For x > 1016 use the pair
(0.98, 7481) from Lemma 3.12 and write, since x ≥ 7481 in this range,
√ √
ψ(x) − θ(x) ≥ θ( x ) ≥ 0.98 x.

(e) Up to 106 the result follows

√ by √
a computation – see Section 3.3. For
x ≥ 106 use ψ(x) − θ(x) < 1.02 x + 3 3 x. All five parts of the lemma have
now been verified. 
We now give the final lemma, Lemma 3.14. Part (a) is the most useful.
The others are included with particular constants chosen to meet the needs of
particular applications.
Lemma 3.14 [145, theorems 4 and 29–31]
(a) For 19 421 ≤ x we have |θ(x) − x| < 0.125x/log x.
(b) For 1451 ≤ x ≤ e375 we have |θ(x) − x| < 0.31x/log x.
(c) For 809 ≤ x ≤ e575 we have |θ(x) − x| < 0.40x/log x.
(d) For 569 ≤ x we have |θ(x) − x| < 0.47x/log x.
(e) For 563 ≤ x we have |θ(x) − x| < 0.5x/log x with, in addition, for 0 < x we
get θ(x) < x(1 + 0.5/log x).
Proof (1) If 1 < x ≤ 1011 : each of (a)–(e) is verified by computation – see
Section 3.3.
(2) Next let 1011 ≤ x ≤ e5000 . We use Lemma 3.10(a), (b) and (c) to get for
x>0
√ √
ψ(x) − θ(x) < 1.0009 x + 3 3 x, x > 0,

and then use for b in Table 3.2 or 3.3

|ψ(x) − x| < b x, x ≥ eb ,

so therefore
√ √ √ √
(1 − b )x − 1.0009 x − 3 3 x < ψ(x) − 1.0009 x − 3 3 x < θ(x).

If b j and b j+1 are in the tables with b j < b j+1 and  j corresponds to a j = eb j ,
then for a j ≤ x ≤ a j+1 we have
    ⎛ ⎞
1 1 ⎜⎜⎜ 1.02 3 ⎟⎟⎟⎟
x 1− ≤ x 1− ⎜
≤ x ⎜⎝1 −  j − √ − 2/3 ⎟⎠ < θ(x)
8 log x 8 log a j+1 aj aj

if
1.02 3 1
 j + √ + 2/3 ≤ ,
aj aj 8 log a j+1
 3.6 Unsolved Problems 67

for x ≥ 1011 . Now e25 < 1011 and we were able to verify the inequality for  j
using four intervals for b, namely
[25, 30], [30, 40], [40, 100] and [100, 5000].
(3) For x ≥ e5000 use Theorem 3.8. In this range, with the notation in that
theorem and with R ≥ 8 we have
⎛  ⎞
 ⎜⎜⎜ log x ⎟⎟⎟ 1
(x) = 2 log x exp ⎜⎜⎝− ⎟⎟⎠ < .
R 8 log x

This completes the proof of (a) and therefore of (b)–(e) for x ≥ 1011 . 

3.6 Unsolved Problems

(1) Write a program to reproduce the tables given in Section 3.2, then
heuristically determine how  varies with b, with or without fixed m.
Derive analytical estimates to replace the tables.
(2) Examine the analytical estimates to derive the bound given in
Theorem 3.8 and derive a better bound (which is for example asymp-
totically smaller).
(3) Extend the finite ranges for any of the results given here which
were obtained through computation, using improved computation or
otherwise. See Section 3.3.
(4) Replace the approach using tables with something analytic and easier to
apply.
(5) Write a survey article or book giving an account of work on explicit
estimates and their applications in (a part of) number theory.
 4
Classical Equivalences

4.1 Introduction
In this chapter a number of classical equivalences to the Riemann hypothesis
are presented, several with complete proofs. In Section 4.2 the explicit
result of Schoenfeld is proved in detail, namely that RH is equivalent to the
inequality
√
x log2 x
|ψ(x) − x| ≤ ,
8π

being true for all x ≥ 74. We call this the Schoenfeld criterion. It is directly
related to the size of the error in different forms of the prime number theorem.
Note that RH is equivalent to the weaker form
√
|ψ(x) − x| x log2 x or even |ψ(x) − x|  x1/2+ .

Then in Section 4.3 material under the heading “Oscillation theorems” is

given. This material is presented without proof, except for two results which
are used explicitly in later chapters. A fundamental theorem of Landau,
Theorem 4.12, is used in many places in the volume to derive results when
the Riemann hypothesis is assumed to be false. In addition, a famous result of
Littlewood, Theorem 4.13, shows in particular that both positive and negative
gaps between ψ(x) and x occur
√ infinitely often as x → ∞, with gap sizes being
at least a constant times x logloglog x, even assuming RH.
In Section 4.4 a number of results are given wherein the hypothesis is
equivalent to the order of the error in arithmetic sums. These have been
known for a long time and the list is not at all comprehensive. Some failed
attempts to show RH is true have been based on them. It is expected that
a range of equivalences along this same line, some of them new, might be
derived.

68
 4.2 The Prime Number Theorem and Its RH Equivalences 69

A proof of the Littlewood criterion is given in Section 4.4, viz. that RH

is equivalent to the estimate that, for all  > 0, as x → ∞ we have

μ(n) x1/2+ .
n≤x

The following RHpack (see Appendix B) functions relate to the material

in this chapter: ComputeTheta, ComputeThetaValues, M and Psi.

4.2 The Prime Number Theorem and Its RH Equivalences

The original prime number theorem, as proved by de la Vallée-Poussin in
1889, is the statement that there exists an absolute constant a > 0 such that
we have
 √ 
π(x) = Li(x) + O x e−a log x , x → ∞,
x
where for x ≥ 2 we have Li(x) := 2 dt/log t, called the logarithmic integral.
This was improved by Walfisz [177, chapter 5] without using any special
hypotheses. He showed that there is a constant a > 0 such that as x → ∞ we
have
 3/5 1/5

π(x) = Li(x) + O x e−a log x/(loglog x) , x → ∞.
In 1901–1902 von Koch [93, 94] showed RH implied that a much better
error term in the prime number theorem could be assumed. Indeed von Koch
proved that the Riemann hypothesis is equivalent to the estimate
√
π(x) = Li(x) + O( x log x), x → ∞,
where the implied constant is absolute. The implied constant in the error
term was made explicit by Schoenfeld [153, 154] in 1976, who showed the
RH implies that for all x ≥ 2657 we have
√
x log x
|π(x) − li(x)| < ,
8π
where the alternative form of the logarithmic integral li(x) is defined by the
sum of improper integrals for x > 1,
 1−  x
dt dt
li(x) := lim + lim ,
→0 0 log t →0 1+ log t

and is such that Li(x) = li(x) − li(2) for x ≥ 2 with 1 < li(2) < 1.046.
Schoenfeld [153, 154] also found an explicit bound for the error  in the
estimate ψ(x) ∼ x in the case of the function ψ(x), where ψ(x) := pn ≤x log p
counts primes with a weighted sum. This function, and its close companion
θ(x) := p≤x log p, are fundamental components of this study. They are named
70 Classical Equivalences

after the famous Russian mathematician Chebyshev (1821–1894), who made

many contributions to mathematics, including orthogonal polynomials and
the theory of probability. He is well known for his proof of a weaker form of
the prime number theorem.
Here is the statement of Schoenfeld’s theorem: the Riemann hypothesis
implies that for all x ≥ 73.2 we have
√
x log2 x
|ψ(x) − x| < .
8π
It also implies that for all x ≥ 599 we have
√
x log2 x
|θ(x) − x| < .
8π
These two classical results have a famous corollary which gives the best
possible form of the prime number theorem error. For x ≥ 2657 we have
√
x log x
|π(x) − li(x)| < .
8π
The proof requires quite a few lemmas which are set out below. We use
a minimum of generality to make the proofs as simple as possible. The
corollary is proved following the theorem.
First, using the same method as in Abel’s identity, [6, theorem 4.2] we
derive the following well-known stronger result:
Lemma 4.1 Let a < γ1 < g2 < · · · < γn ≤ b be real numbers with n ≥ 2, and
a j ∈ C, 1 ≤ j ≤ n. For a ≤ x ≤ b let

A(x) = ai ,
γi ≤x

and let f (x) be a real- or complex-valued function on [a, b] with an integrable

derivative. Then
n  b
a j f (γ j ) = A(b) f (b) − A(a) f (a) − A(t) f  (t) dt.
j=1 a

Proof First note that

 γ j+1  γ j+1
A(t) f  (t) dt = A(γ j ) f  (t) dt = A(γ j ) f (γ j+1 ) − A(γ j ) f (γ j ),
γj γj

so therefore, noting that A(γn ) = A(b) and shifting the index in the first sum:
 b
A(t) f  (t) dt
a
 4.2 The Prime Number Theorem and Its RH Equivalences 71
 γ1  γn  b
= A(t) f  (t) dt + A(t) f  (t) dt + A(t) f  (t) dt
a γ1 γn
= A(a)( f (γ1 ) − f (a)) + A(γn )( f (b) − f (γn ))
n−1
+ (A(γ j ) f (γ j+1 ) − A(γ j ) f (γ j ))
j=1
= A(b) f (b) − A(a) f (a) + A(a) f (γ1 ) − f (γn )A(γn )

n−1 
n
− A(γ j ) f (γ j ) + A(γ j−1 ) f (γ j )
j=1 j=2
= A(b) f (b) − A(a) f (a) + A(a) f (γ1 ) − f (γn )A(γn )
n−1
+ (A(γ j−1 ) − A(γ j )) f (γ j ) + A(γn−1 ) f (γn ) − A(γ1 ) f (γ1 )
j=2

n
= A(b) f (b) − A(a) f (a) + A(a) f (γ1 ) − f (γn )A(γn ) − a j f (γ j )
j=1
+ A(γn−1 ) f (γn ) − A(γ1 ) f (γ1 )
 n
= A(b) f (b) − A(a) f (a) − a j f (γ j ).
j=1

This completes the proof of the lemma 

Remark By suitable indexing and setting a j = 1 with (γ j ) a distinct
increasing sequence of real numbers meeting the real interval (a, b] in at least
two points, so A(x) counts the number of the γ j not greater than x, we obtain
the useful corollary:
  b
f (γ j ) = A(b) f (b) − A(a) f (a) − A(t) f  (t) dt.
a<γ j ≤b a

By specializing to the case of the positive imaginary parts of zeta zeros, using
Lemma 4.1 we obtain a simplified upper bound for a functional sum. First
recall some definitions from Section 2.2 (11):
N(T ) := #{ρ : ζ(ρ) = 0, 0 < ρ ≤ T },
T  T  7
F(T ) := log + ,
2π 2πe 8
B(T ) := a1 log T + a2 loglog T + a3 .
Lemma 4.2 [146, lemma 7, p. 255] Let 1 < a < b and the real function f (y)
be such that f  (y) is continuous with f  (y) ≤ 0 and f (y) > 0 for a ≤ y ≤ b. Then
  b   b
1 y a2 f (y)
f (γ j ) ≤ f (y) log dy + a1 + dy + 2B(a) f (a).
a<γ ≤b
2π a 2π log a a y
j
72 Classical Equivalences

Proof By the remark following the proof of Lemma 4.1 we have

  b
f (γ j ) = N(b) f (b) − N(a) f (a) − N(y) f  (y) dy.
a<γ j ≤b a

Now Backlund’s explicit estimate for N(T ) (see Section 2.2) can be written,
for all T ≥ 2,

|N(T ) − F(T )| < B(T ) =⇒ N(T ) < F(T )

+ B(T ) and − N(T ) < B(T ) − F(T ).

Therefore, since f  (y) is negative, we can upper bound the sum,

  b
f (γ j ) < − (F(y) + B(y)) f  (y) dy
a<γ j ≤b a

+ (B(b) + F(b)) f (b) + (B(a) − F(a)) f (a).

Integrating by parts and using

1 y
F  (y) = log
2π 2π
and
   
 a2 1 a2 1
B (y) = a1 + ≤ a1 +
log y y log a y
for a ≤ y ≤ b, we get
  b   b
1 y a2 f (y)
f (γ j ) < f (y) log dy + a1 + dy + 2B(a) f (a).
a<γ ≤b
2π a 2π log a a y
j

which completes the proof. 

Now we define two sums depending on two real positive parameters for δ
and T :
 1
S 1 (δ) := (2 + δ) ,
0<γ≤T
|ρ|
(2 + 2δ + δ2 )  1
S 2 (δ) := 2 ,
δ T <γ
|ρ(ρ + 1)|

where ρ = β + iγ represents a zero of ζ(s). These sums are used as an

intermediate step to obtain the bounds we need for |ψ(x)− x| and |θ(x)− x|. The
derivation in Lemma 4.3 is a simplified version of that used in Lemma 3.2.
 4.2 The Prime Number Theorem and Its RH Equivalences 73

Lemma 4.3 [146, lemma 8, p. 256] Assume the Riemann hypothesis is true.
If x > 1 and 0 < δ < 1 − 1/x we have
  
 1 1  √ δx
ψ(x) − x + log(2π) + log 1 − 2  ≤ x (S 1 (δ) + S 2 (δ)) + .
2 x 2
Proof (1) First we need to define some functions:
 
1 1
η(x) := ψ(x) − x + log(2π) + log 1 − 2 ,
2 x
 h
E(x, h) := η(x + y) dy,
0
E(x, h) nhα
fn,α (x, h, z) := + − zhα−1 ,
hn 2
E(x, h) h
f (x, h, z) := f1,1 (x, h, z) = + − z,
h 2
E(x, −xδ) δ
1 := + ,
x2 δ 2
E(x, xδ) δ
2 := + .
x2 δ 2
Also assume x > 1 and x + h > 1. The variable h could be negative, indeed
we will choose later to have h = ±δx. In the properties below, (2) and (3) are
 and (4) can be deduced by induction on n ≥ 1.
immediate
h
(2) f1,α (x, h, z) dz = E(x, h).
0 h
(3) fn,α (x, h, y1 + · · · + yn ) dyn = fn−1,α+1 (x, h, y1 + · · · + yn−1 ).
0  h  h
(4) E(x, h) = ··· fn,α (x, h, y1 + · · · + yn ) dyn · · · dy1 .
0 0
(5) If 0 < h there is a z with 0 ≤ z ≤ h such that η(x + z) ≤ f (x, h, z): if not,
we would have by (2)
 h  h
E(x, h) = η(x + y) dy > f (x, h, y) dy = E(x, h),
0 0

which is false.
(6) If 0 > h there is a z with 0 ≥ z ≥ h such that η(x + z) ≥ f (x, h, z): if not,
we would have
 0  0
−E(x, h) = η(x + y) dy < f (x, h, y) dy = −E(x, h).
h h

(7) If < δ < 1 − 1/x then −1 x ≤ η(x) ≤ 2 x: let h = δx with δ > 0. By (5)
there is a z ∈ [0, h] such that
E(x, δx) δx
η(x + z) ≤ + − z ≤ x(1 + 2 ).
δx 2
74 Classical Equivalences

Therefore since z ≥ 0 and 12 log(1 − 1/(x + z)2 ) ≤ − 12 log(1 − 1/x2 ) we get

 
1 1
ψ(x) ≤ ψ(x + z) ≤ x(1 + 2 ) − log(2π) − log 1 − 2 .
2 x
In other words, η(x) ≤ 2 x. The proof of the lower bound is similar.
(8) Next we claim with ρ = 12 + iγ being a zeta zero that
 h  xρ+1 − (x + h)ρ+1
η(x + z) dz = .
0 ρ
ρ(ρ + 1)

To see this, use Theorem 2.2, which for x > 1 is

 x
x2  xρ+1 ∞
x−2n+1 ζ  (0) ζ  (−1)
ψ(t) dt = − − − x+ .
0 2 ρ
ρ(ρ + 1) n=1 2n(2n − 1) ζ(0) ζ(−1)

First integrate the definition of η(x):

 h
E(x, h) = η(x + z) dz
0
 h  h
= ψ(x + z) dz − (x + z) dz
0

0
 
1 h 1
+ log 1 − dz + h log(2π)
2 0 (x + z)2
 x+h   
1 2 1 x+h 1
= ψ(u) du − hx − h + log 1 − 2 du
x 2 2 x u
 xρ+1 − (x + h)ρ+1
= .
ρ
ρ(ρ + 1)

(9) Now we split the sum from (8) so that

   
  xρ+1 − (x + h)ρ+1    xρ+1 − (x + h)ρ+1 
|E(x, h)| ≤ 2   + 2   .
 ρ(ρ + 1)   ρ(ρ + 1) 
0<γ≤T T <γ

Let h = ±δx and recall we are assuming RH is true. In the first sum use the
bound
 ρ+1   δ
 x − (x + h)ρ+1  δ(2 + δ)
 ≤ x (1 + y) = x3/2 .
3/2

(ρ + 1) 0 2
In the second sum use the upper bound |1 − (1 ± δ)1+ρ | ≤ (1 + (1 + δ)2 ) so that
 ρ+1 
 x − (x + h)ρ+1  1
 ≤ x (2 + 2δ + δ ) .
3/2 2

ρ(ρ + 1) |ρ(ρ + 1)|
 4.2 The Prime Number Theorem and Its RH Equivalences 75

Therefore
 1  1
|E(x, ±xδ)| ≤ x3/2 (2 + δ)δ + 2x3/2 (2 + 2δ + δ2 ) .
0<γ≤T
|ρ| T <γ
|ρ(ρ + 1)|

Finally using this bound, the definitions of E(x, h), i , S i (δ) and Step (7), we
get
√ δx
|η(x)| ≤ x(S 1 (δ) + S 2 (δ)) + ,
2
and the proof of the lemma is complete. 
Now let T = 2(2 + 2δ + δ2 )/(δ(2 + δ)) so T → ∞ as δ → 0+. Set D :=
158.84998, just less than the y-coordinate of the 58th zeta zero. We also
in Lemma 4.4 and in Theorem 4.5 use the choices a1 = 0.137 and a2 =
0.443, which are sufficient for our purposes, but not the best available. See
Section 2.2 and the unsolved problems at the end of this chapter.
Lemma 4.4 [146, lemma 9, p. 262] If T 1 ≥ D and δ > 0 then
 2 
2+δ  T1
S 1 (δ) + S 2 (δ) < log +1 +1 .
4π 2π
Proof A computation gives
 1
 < 0.8113925.
γ≤D 1
4
+ γ2

Next take f (y) = 1/y in Lemma 4.2 to get

 1 1  2 T    
2 D 0.443 1 1 2B(D)
≤ log − log + 0.137 + − + .
D<γ≤T
γ 4π 2π 2π log D D T D

Therefore
S 1 (δ) 1 T 0.224212 0.224212 2B(D)
< 0.8113925 + log2 − + +
2+δ 4π 2π T D D
1 D
− log2
4π 2π
1 T 0.224212
≤ log2 − − 0.791068.
4π 2π T
Again using Lemma 4.2, this time with f (y) = 1/y2 , we get
 1  1
<
T <γ
|ρ(ρ + 1)| T <γ γ2
 ∞   ∞
1 1 y 0.443 dy 2B(T )
≤ 2
log dy + 0.137 + +
2π T y 2π log T T y3 T2
76 Classical Equivalences
 
1  T  1 0.443 2B(T )
= log + 1 + 2 0.137 + + .
2πT 2π 2T log T T2
Then, making the choice T = T 1 where (2 + δ)T 1 = 2(2 + 2δ + δ2 )/δ we get
 2 
2+δ  T1
S 1 (δ) + S 2 (δ) < log +1 +1 .
4π 2π

We are now able to present Schoenfeld’s proof of one side of the
equivalence, wherein we assume the Riemann hypothesis is true.
Theorem 4.5 [153, theorem 10, p. 337] The Riemann hypothesis implies
that for all x ≥ 23 × 108 we have
√
x log x (log x − 2)
|ψ(x) − x| <
8π
and
√
x log x (log x − 2)
|θ(x) − x| < .
8π
Proof Let x ≥ ξ ≥ 828 000. In the proof, constants δ and αi with 1 ≤ i ≤ 6
will be defined progressively, with each of them depending on ξ in the given
range, which finally will be chosen explicitly to enable the proof to succeed.
Let
log2 ξ log2 ξ
α1 := √ and α2 := √ < 0.0126. (4.1)
π ξ π ξ
Then
log x α1
δ := √ ≤ ≤ α2 .
π x log x
Now let T 1 := 2(2 + 2δ + δ2 )/(δ(2 + δ)) and D := 158.84998 so there are
precisely 57 zeros ρ = 12 + iγ of ζ(s) with 0 < γ ≤ D and T 1 > D. Then if
we set
α2 (1 + α2 )
δ1 := < 0.00634 and α3 := 2 loglog ξ − 2(1 + δ1 ) > 2.841,
2 + α2
we get
 √   √ 
T1 x 2 + 2δ + δ2 x
log + 1 = log + 1 ≤ log (1 + δ1 ) + 1
2π log x 2 + δ log x
1 1
≤ (log x − 2 loglog x + 2(1 + δ1 )) ≤ (log x − α3 ).
2 2
 4.2 The Prime Number Theorem and Its RH Equivalences 77

Now define
α23 + 4.152828
α4 := 2α3 − .
log ξ
By Lemmas 4.3 and 4.4, since
 
2+δ 1 α1
< 1+ ,
4π 2π 2 log x
we can write
   2 
|ψ(x) − x| 1 α1 T1
< √ 1+ log + 1 + 1.038207
x 2π x 2 log x 2π
  
log x 1 1 1
+ √ + log(2π) + log 1 − 2
2π x x 2 x
 
1 α1
< √ 1+ (log2 x − 2α3 log x + α23 + 4.152828)
8π x 2 log x
log x log(2π)
+ √ + .
2π x x
Thus
 
|ψ(x) − x| 1 α1
< √ 1 + (log2 x − α4 log x)
x 8π x 2 log x
 
log x 8π log(2π)
+ √ 4+ √ .
8π x x log x
Next, note that α3 < 2 loglog ξ < log ξ, so α4 increases with α3 , and therefore,
for this range,
2.8412 + 4.152828
α4 > 2 × 2.841 − > 0.
log ξ
Therefore, using the upper bound which was derived for |ψ(x) − x|/x and
defining
α1 8π log(2π)
α5 := α4 − −4− √ ,
2 ξ log ξ
we get
|ψ(x) − x| log x (log x − α5 )
< √ .
x 8π x
Finally define
8.16π 24π
α6 := α5 − − 1/6 .
log ξ ξ log ξ
78 Classical Equivalences

To get the analogous inequality for θ(x), use parts (a), (b) and
√ (d)√of
Lemma 3.10, which gives for 0 < x, the bound ψ(x) − θ(x) < 1.02 x + 3 3 x.
Then
|θ(x) − x| |ψ(x) − x| ψ(x) − θ(x) log x
< + ≤ √ (log x − α6 ).
x x x 8π x
Let ξ = 108 to compute α5 > α6 > 2, which competes the proof of the
theorem. 
√
Remark The inequality |θ(x) − x| ≤ x log2 x/(8π)√succeeds for 231 integers
n in the range [2, 598]. The inequality |ψ(x) − x| ≤ x log2 x/(8π) succeeds at
integers in the list
{13, 19, 20, 23, 24, 25, 26, 27, 29, 30, 31, 32, 33, 34, 35, 38, 39,
41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57,
58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73}
and fails at all other integers in [2, 73].
Theorem 4.6 [153, theorem 1] The Riemann hypothesis implies that for all
x ≥ 74 we have
√
x log2 x
|ψ(x) − x| <
8π
and for all x ≥ 599 we have
√
x log2 x
|θ(x) − x| < .
8π
Proof In the range x ≥ 23 × 108 the result follows from Theorem 4.5.
For 90 000 ≤ x < 23 × 108 we checked the inequalities through making a
computation at integers
1 √
n+1− n + 1 log(n + 1)(log(n + 1) − 2) < θ(n)
8π
1 √
< n+ n log n (log n − 2).
8π
Below 90 000 we checked the inequality
1 √ 1 √
n+1− n + 1 log2 (n + 1) < θ(n) < n + n log2 n.
8π 8π
The derivation in the case of ψ(x) is similar. 
More information about this process of verifying estimates over finite
ranges numerically is given in Chapter 3.
 4.2 The Prime Number Theorem and Its RH Equivalences 79

Corollary 4.7 [153, corollary 1, p. 339] Assume the Riemann hypothesis is

true. Then for all x ≥ 2657 we have
√
x log x
|π(x) − li(x)| ≤ .
8π
Proof First let x and ξ be real numbers such that x ≥ 23 × 108 ≥ ξ > 1. Then
by Abel’s identity [6, theorem 4.2] we have
 x  x
θ(x) θ(ξ) θ(t) − t dt
π(x) − π(ξ) = − + dt + .
log x log ξ 2
ξ t log t
2
ξ log t

Integrating 1/log t by parts we have also

 x  x
dt ξ x dt ξ x
= − + = − + li(x) − li(ξ).
2
ξ log t log ξ log x ξ log t log ξ log x
Therefore we get

θ(x) − x x
θ(t) − t
π(x) − li(x) = + dt − η(ξ),
log x ξ t log2 t
where η(x) := li(x) − π(x) − (x − θ(x))/log x. Therefore, using Theorem 4.5,
   
 θ(x) − x   x θ(t) − t 
|π(x) − li(x)| ≤  + dt + |η(ξ)|
 log x   ξ t log2 t 
√  x
x (log x − 2) 1 dt
≤ + √ + |η(ξ)|
8π 8π ξ t
√ √
x log x ξ
≤ − + |η(ξ)|.
8π 4π
√
At ξ = 108 , 0 < η(ξ) < 88.5 < 104 /(4π) so |π(x) − li(x)| < x log x/(8π) when
x ≥ 23 × 108 .
√ In the range n ∈ [2657, 23 × 10 ] we checked the inequality π(n) < li(n) +
8

n√log n/(8π) numerically at integers and then the inequality π(n) > li(n + 1)
− n + 1 log(n + 1)/(8π), also at integers, demonstrating that the theorem is
true for all x ≥ 2657. 
Remark
√ A computation shows that the inequality |π(n) − li(n)| <
n log n/(8π) is also true for a substantial number of integers n with 2 ≤ n ≤
2656. Indeed 1434 successes were found, splitting into about 40 subranges.
We now give an example of a well-known technique which is used a few
other times in this volume in case the Riemann hypothesis is false. Define
θ := sup{σ : ζ(σ + iγ) = 0} so RH is equivalent to θ = 12 , and always 12 ≤ θ ≤ 1.
Theorem 4.8 Let  > 0 be given. Then ψ(x) − x = Ω± (xθ− ), where the implied
constant is arbitrary and when Ωt is defined in Section 4.3.
80 Classical Equivalences

Proof First note that, by Lemma 5.28, for s > 1 we have

 ∞
ζ  (s) ψ(t)
− =s dt.
ζ(s) 2 t s+1
Now let 0 < α < θ ≤ 1 and let C  0 be a given constant. For 1 < x define
ψ(x) − x − Cxα
g(x) :=
x
so that
 ∞  ∞
g(x) ψ(x) − x − Cxα ζ  (s) 1 C
f (s) := dx = dx = − − − .
1 x s
1 x s+1 sζ(s) s − 1 s − α
Now let σ0 be the abscissa of convergence of the integral for f (s). The
integral represents a holomorphic function on the half plane σ > σ0 . In
addition, since ζ  (s)/ζ(s) has a pole at an imaginary zeta zero with real part
in (α, 1) which is not cancelled by any other term in the derived expression
for f (s), the half plane σ > σ0 can contain no zero of ζ(s). This implies
σ0 ≥ θ > α.
On the other hand, because ζ(s) is non-zero on the positive real axis (0, ∞),
and there is a function h(s) which is holomorphic on the positive real axis
with
ζ  (s) 1 1 + h(s) C
=− + h(s) =⇒ f (s) = − − ,
ζ(s) s−1 s s−α
f (s) (or its continuation) has no singularity on the real axis for σ > α. Because
α < σ0 , σ0 is not a singularity of f (s). By Theorem 4.12 we must have that
g(x) changes sign infinitely often as x → ∞ so ψ(x) − x > Cxα and ψ(x) − x <
−Cxα , each for an infinite number of values of x with limit value infinity. To
complete the proof, replace α by θ − . 
Now we can prove a principal result of this chapter, giving an explicit
equivalence to RH.
Theorem 4.9 (Schoenfeld criterion) The Riemann hypothesis is equivalent
to the inequality
√
x log2 x
|ψ(x) − x| ≤ , 74 ≤ x.
8π
Proof If RH is true the result follows from Theorem 4.6. If RH is false then
θ > 12 . By Theorem 4.8 we can choose  > 0 so that θ −  > 12 and, letting the
constant C = 1 we have
|ψ(x) − x| ≥ x1/2 xθ−−1/2
for an infinite sequence of values x with limit plus infinity. But for x
sufficiently large we have xθ−−1/2 > log2 x/(8π), so the inequality fails (an
infinite number of times). Therefore the inequality is equivalent to RH. 
 4.3 Oscillation Theorems 81

4.3 Oscillation Theorems

It is useful to note for what follows that, when the Riemann hypothesis fails,
there exist sequences of real numbers or integers with limit infinity which
contradict the estimates given by the Riemann hypothesis in different ways.
To describe this, the notation f (x) = Ω± (g(x)) is used. Here f (x) and g(x)
are real-valued and g(x) > 0 for all x > 0. This notation means that for some
positive constant C and an infinite increasing real sequence (xn )n∈N with limit
value infinity, we have f (xn ) ≥ Cg(xn ), and for some other such sequence
(yn )n∈N we have f (yn ) ≤ −Cg(yn ).
The material in this section is based mainly on the account of Grosswald
[70]. As given above, in 1901 √von Koch proved, using the Riemann
hypothesis, that π(x) = Li(x) + O( x log x). This was used by E. Schmidt in
1903 [152] to prove the following:
Theorem 4.10 (Schmidt) Let θ := sup{s : ζ(s) = 0} and
∞
π(x1/m )
f (x) := .
m=1
m

Then for all λ < θ we have f (x) = Ω± (xλ ) where the implied constant C may
be chosen arbitrarily large. In addition there are real sequences xn , yn → ∞
such that
√ √
xn yn
f (xn ) − Li(xn ) > and f (yn ) − Li(yn ) < − .
29 log xn 29 log yn
Then in 1905 Landau developed a new method to attack these “oscillation”
problems. As usual, we define θ(x) := p≤x log p.
Theorem 4.11 (Landau) As x → ∞ we have θ(x) = Ω± (x).
The principal tool in this work is the theorem of Landau given in his 1905
paper [104], but also in [80, theorem H] and [166, chapter II.1, theorems 6
and 7]. The proof is included because the result is fundamental to this work,
and used in many places.
Theorem 4.12 (Landau) Let s ∈ C and let f (x) : [1, ∞) → R be measurable
and bounded on all bounded intervals. Suppose that
 ∞
dx
F(s) := f (x) s
1 x
has a finite abscissa of convergence σc so F(s) ∈ C if s > σc .
(a) If there exists an a ∈ R such that f (x) is non-negative or non-positive for
x ≥ a, then the integral F(s) for σ = s > σc has a singularity at s = σc
and F(s) converges in a half plane such that it is holomorphic for σ > σc
but not in any half plane σ > σc −  for any  > 0.
82 Classical Equivalences

(b) If F(s) is holomorphic at s = σc , then f (x) changes sign at all points in

an infinite set xn with xn → ∞. We also have for every  > 0
f (x) = Ω± (xσc − ).
Proof (a) Suppose, to derive a contradiction, there exists an open connected
set Ω containing the half plane s > σc and the point s = σc so the
continuation of F(s) is holomorphic in Ω. Let α > σc and let δ > α − σc
be such that the open ball {s : |s − α| < δ} ⊂ Ω.
Since F(s) is holomorphic in Ω, for |s − α| < δ we can express it as a
convergent Taylor series,

∞
(s − α)n
F(s) = F (n) (α).
n=0
n!

Suppose that f (x) ≥ 0 for x ≥ a. (If f (x) ≤ 0 replace f (x) by − f (x).) Therefore
if b > a and n ≥ 0 we can write
 ∞  b
f (x) logn x f (x) logn x
(−1) F (α) =
n (n)
dx ≥ dx.
1 xα 1 xα
Therefore if s ∈ R satisfies α − δ < s ≤ α, F(s) is real also and we can write
 ∞   b ∞
(α − s)n b f (x) logn x (α − s)n f (x) logn x
F(s) ≥ dx = dx.
n=0
n! 1 xα 1 n=0 n! xα

To justify the interchange of the sum and the integral note that for 1 ≤ x ≤ b
we have
 n 
 (s − α) f (x) log x  ≤ M δ log b
n n n

 n! xα  n!
where M := max{| f (x)x−α | : 1 ≤ x ≤ b}, so the series is uniformly convergent.
Therefore
 b  b
f (x) (α−s) log x f (x)
F(s) ≥ α
e dx = dx.
1 x 1 xs
Because for real s ≥ a the integrand is positive, this integral is increasing with
b for b ≥ a, and bounded above by F(s). Hence the integral is convergent at
all points of the interval s ∈ (α − δ, α]. But this is impossible since the interval
properly contains s = σc . Therefore the function defined by the integral has a
singularity at s = σc .
(b) Assume, to derive a contradiction, that there exists a constant M > 0
such that f (x) ≤ Mxσc − for all x ≥ 1. Then if we define
g(u) := Me(σc −)u − f (eu )
 4.3 Oscillation Theorems 83

we get g(u) ≥ 0 for u ≥ 0, and setting x = eu

 ∞  ∞
M σc − dx
F(s) − = ( f (x) − Mx ) s+1 = − g(u)e−su du.
s − σc +  1 x 0
Because the integral
 ∞
xσc −−s−1 dx
1

is absolutely convergent when s = σ > σc − , the abscissa of convergence

of the integral of g(u)e−su is still σc . By part (a), the point s = σc is a singular-
ity of F(s)− M/(s−σc +) so F(s) cannot be holomorphic at s = σc . This con-
tradiction completes the proof. The proof for f (x) ≤ −Mxσc − is similar. 
Then in 1914 Littlewood [109] improved Theorem 4.11 by showing the
following:
Theorem 4.13 As x → ∞ we have
√
ψ(x) = x + Ω± ( x logloglog x), (4.2)
where the same formula is true if we replace ψ(x) by θ(x).
Proof (1) The proof is in eight steps. First we note that if RH is false then
the theorem follows from Theorem 4.8. Therefore in the rest of this proof we
can assume RH is true.
(2) To begin the proof proper, it follows from the proof of Dirichlet’s
lemma, Lemma 5.32, that if we define x to be the distance from x to the
nearest integer, given fixed real numbers α1 , . . . , αK and a positive integer N,
there exists an integer n with 1 ≤ n ≤ N K such that for 1 ≤ j ≤ K we have
α j n ≤ 1/N.
(3) Let ρ j = 12 + iγ j be the jth positive imaginary zeta zero with increasing
γ j . Let N be a sufficiently large positive integer and define α j = γ j log N/(2π),
where 0 < γ j ≤ T := N log N (see Section 2.2), so the number of the γ j under
consideration is K  T log T . By Step (2) there is an integer n with 1 ≤ n ≤ N K
such that for each j with 0 < γ j ≤ T we have
 
 nγ j log N  ≤ 1 .
 2π  N
Because for all real x we have |sin(πx)| ≤ πx we can write
|sin(2πα) − sin(2πβ)| = |2 sin(π(α − β)) cos(π(α + β))| ≤ 2πα − β.
Using these two bounds, if x = N n exp(1/N) we have
 
 γ j (log x − 1/N) 
|sin(γ j log x) − sin(γ j /N)| ≤ 2π  
2π
 
= 2π 
nγ j log N  2π
≤ .
2π  N
84 Classical Equivalences

Similarly, if x = N n exp(−1/N) we have

 
 γ j (log x + 1/N) 
|sin(γ j log x) + sin(γ j /N)| ≤ 2π  
2π
 
= 2π 
nγ j log N  2π
≤ .
2π  N
(4) Next we claim that for 1/(2x) ≤ δ ≤ 12 and δ = 1/N we have an
expression for the average value of ψ(x) − x over the interval [x/eδ , xeδ ],
namely, uniformly for x ≥ 4,
 x exp(δ)
1
A(δ) := (ψ(t) − t) dt
x exp(δ) − x exp(−δ) x exp(−δ)
√  ∞
sin(γ j /N) sin(γ j log x) √
= −2 x + O( x). (4.3)
j=1
γ j /N γj

To see this we first apply von Mangoldt’s theorem, Theorem 2.2, to get, for
some constant c ∈ R,
 x  xρ+1
(ψ(t) − t) dt = − − cx + O(1),
0 ρ
ρ(ρ + 1)

where the sum is over all of the non-trivial zeta zeros. This implies
δ  eδ(ρ+1) − e−δ(ρ+1) ρ
A(δ) = − x + O(1). (4.4)
sinh(δ) ρ 2δρ(ρ + 1)

We have for ρ = 12 + iγ as δ → 0+,

eδ(ρ+1) = eδiγ + O(δ) and e−δ(ρ+1) = e−δiγ + O(δ),
and (see Table 3.1)
 1  1
< < ∞.
ρ
ρ(ρ + 1) ρ
γ2
√
In addition we have |xρ | = x and, as δ → 0,
δ
= 1 + O(δ2 ) 1.
sinh(δ)
Using these estimates in the summation of (4.4) gives
⎛ ⎞
 eδ(ρ+1) − e−δ(ρ+1)  sin(γδ) xiγ ⎜⎜⎜  1 ⎟⎟⎟⎟
ρ
x = ix 1/2 ⎜
+ O ⎜⎜⎝ x 1/2
⎟⎟
ρ
2δρ(ρ + 1) ρ
δ ρ(ρ + 1) ρ
ρ(ρ + 1) ⎠
 sin(γδ) xiγ
= ix1/2 + O(x1/2 ).
ρ
δ ρ(ρ + 1)
 4.3 Oscillation Theorems 85

Next using
 ∞
d(t log t) 2 + log T log T
2
= ,
T t T T
we have
 
 sin(γδ) xiγ   sin(γδ) xiγ 

 
δ ρ(ρ + 1) δ ρ(ρ + 1) 
ρ γ≥1
 sin(γδ) γ 1  |sin(γδ)|
+
1≤γ≤1/δ
γδ γ2 δ 1/δ<γ γ2
 1 1 1
+
1≤γ≤1/δ
γ δ 1/δ<γ γ2
log2 (1/δ) + log(1/δ) log2 (1/δ).

If we substitute 1 + O(δ2 ) for δ/sinh(δ) in the right-hand side of (4.4) then

we get
 sin(γδ) xiγ
A(δ) = −ix1/2 + O(x1/2 ).
ρ
δ ρ(ρ + 1)

We have sin(γδ)/δ ≤ |γ| and by RH, 1/ρ = 1/(iγ) + O(1/γ2 ) and 1/(ρ + 1) =
1/(iγ) + O(1/γ2 ). Therefore
√  sin(γδ) xiγ √
A(δ) = − x + O( x)
ρ
γδ iγ
√  sin(γδ) xiγ − x−iγ √
= −2 x + O( x)
ρ
γδ iγ
√  sin(γδ) sin(γ log x) √
= −2 x + O( x),
γ>0
γδ γ

which completes the proof of the claim.

(5) Next, using the trivial bound for sin(x), Lemma 4.1 and the estimate
again
 ∞
d(t log t) log T
2
,
T t T
we have
 sin(γ j /N) sin(γ j log x)  1 N log(N log N)
N 1.
N log N<γ j
γ j /N γj N log N<γ
γj
2
N log N
j
86 Classical Equivalences

Therefore for x = N n exp(1/N), δ = 1/N and using the result of Step (3)
(this is the key to the proof) and Step (5), we see A(δ) can be expressed as
1
 ∞
sin(γ j /N) sin(γ j log x) √
A(δ) = −2x 2 · + O( x) (4.5)
j=1
γ j /N γj
1
 sin(γ j /N) sin(γ j log x)
= −2x 2 ·
γ j ≤N log N
γ j /N γj
⎛ ⎞
⎜⎜⎜ √  sin(γ /N) sin(γ log x) ⎟⎟⎟ √
+ O ⎜⎜⎜⎜⎝ x ⎟⎟⎟⎟ + O( x)
j j
· ⎠
γ j >N log N
γ j /N γj
 2
x 2  sin(γ j /N) √
1

= −2 + ( x)
N γ ≤N log N γ j /N
j
√  N log N  2
x sin(t/N) √
−2 log t dt + O( x)
N 1 t/N
√
1
x2
−2 N log N + O( x)
N
1 √
= −2x 2 log N + O( x).
(4.6)
Similarly for x = N n exp(−1/N) we have
∞
sin(γ j /N) sin(γ j log x) √
−2x 1/2
 +2x1/2 log N + O( x). (4.7)
j=1
γ j /N γj
K
(6) Because x ≤ N N e1/N , T = N log N and K  T log T  N log2 N we get
loglog x N log3 N.
Taking logarithms for a constant C > 0 we get from this
log N + logC + 3 loglog N = (1 + o(1)) log(N) ≥ logloglog x,
which implies
log N ≥ (1 + o(1)) logloglog x.
(7) Finally, by (4.5) and (4.7), and the bound established in Step (6), letting
δ → 0+ at a point of continuity of ψ(x) we can write
√
ψ(x) = x + Ω± ( x logloglog x).
The same is true at points of discontinuity, since the jump is O(log x).
(8) Considering the expansion of ψ(x) in terms of θ(x1/ j ), with 1 ≤ j ≤
log x/log 2, and the bound θ(x) x, we can write
√ √
ψ(x) = θ(x) + θ( x) + O(x1/3 log x) =⇒ 0 ≤ ψ(x) − θ(x) x.
 4.3 Oscillation Theorems 87
√
Thus ψ(x) = θ(x) + O( x), and substituting this in (4.2) completes the
proof. 
This theorem can be shown to be equivalent to
√ 
x logloglog x
π(x) = Li(x) + Ω± .
log x
The reader might compare this with theorem of Walfisz, giving the best-
known error in the prime number theorem.
Littlewood (Figure 4.1) gives an explicit form for the estimates of
Theorem 4.13. There are sequences of positive real numbers xn → ∞ and
yn → ∞ such that for all n ∈ N we have
√
xn logloglog xn
π(xn ) > Li(xn ) + π(xn )
3 log xn
and
√
yn logloglog yn
π(yn ) < Li(yn ) + .
3 log yn
It follows from these inequalities that there is an x > 0 such that π(x) >
li(x). The Skews number is the smallest natural number S for which this
inequality is true. This number is very celebrated, its proved upper bound
being at one time the largest number appearing in a mathematical proof.
Indeed Skews proved in 1955 [157] that
10964
S < 1010 .

Figure 4.1 J. E. Littlewood (1885–1977).

88 Classical Equivalences

There has been a great deal of work reducing the size of the known upper
bound and finding lower bounds for S . For example Zegowitz [183] showed
in 2010 that S < exp(727.951347) and Kotnik [97] in 2008 that 1014 < S .
For this section, and indeed all of this chapter, the reader would do well to
consider the article by Don Zagier [182].

4.4 Errors in Arithmetic Sums

Recall the definition of μ(n): μ(1) = 1 and, if n is not squarefree,
μ(n) = 0; otherwise μ(n) is (−1)ω(n) where ω(n) is the number of distinct
primes dividing n. Then the prime number theorem is equivalent to the
statement [6]
μ(1) + μ(2) + · · · + μ(N) 
lim = 0 or μ(n) = o(x).
N→∞ N n≤x

Now we will see below one of the most closely considered equivalences
to RH, due originally to Littlewood (see Littlewood [109] or Edwards [51,
section 12.1]), namely that RH is equivalent to the estimate that for all  > 0

μ(n) x1/2+ .
n≤x

First though two estimates are needed:

Lemma 4.14 [51, p. 54] Let a > 0, T > 0 and 0 < x < 1. Then
  a+iT s 
 1 x  xa
 ds ≤ .
2πi a−iT s  πT |log x|
Proof Let a < A and apply Cauchy’s residue theorem to the function x s /s on
the rectangle [a, A] × [−T, T ], on which the function is holomorphic. Thus the
contour integral of the function around the boundary of the rectangle is zero
so
 a+iT s  A+iT s  A−iT s  a+iT s
1 x 1 x 1 x 1 x
ds = − ds − ds − ds.
2πi a−iT s 2πi A−iT s 2πi a−iT s 2πi A+iT s
Therefore, since the modulus of the integrand is bounded by xA /A on the
right-hand vertical section of the contour and, if σ = s, by xσ /T on the
horizontal sections, we have
  a+iT s   A σ
 1 x  1 2T xA 1 x T xA (xa − xA )
 ds ≤ +2 dσ ≤ + .
2πi a−iT s  2π A 2π a T πA πT |log x|
Letting A → ∞ the lemma follows. 
 4.4 Errors in Arithmetic Sums 89

Lemma 4.15 [51, p. 55] Let a > 0, T > 0 and x > 1. Then
  a+iT s 
 1 x  xa
 ds − 1  ≤ .
2πi a−iT s πT |log x|
Proof The proof is similar to that of Lemma 4.14, except this time integrate
around [−A, a] × [−T, T ] and evaluate the residue of the function x s /s at
s = 0. 

Theorem 4.16 (Littlewood criterion) [108] The Riemann hypothesis is

equivalent to the estimate that for all  > 0 we have

μ(n) x1/2+ , x → ∞,
n≤x

where the implied constant depends on .

Proof (1) First assume the given condition, i.e. for all  > 0 we have

μ(n) x1/2+ , x → ∞.
n≤x

This part of the proof will show the condition implies the Riemann hypothesis
is true. Now
1 ∞
μ(n)
= , s > 1.
ζ(s) n=1 n s

Using Abel summation [6, theorem 4.2] with M(x) := n≤x μ(n),
 x  x
μ(n) M(x) M(t)
s
= s
+ s dt.
n=1
n x 1 t s+1

Therefore, letting x → ∞, for s > 1 we have

 ∞
1 M(x)
=s dx.
ζ(s) 1 x s+1
If M(x) grew less rapidly than x1/2+ for every  > 0 then the integral
representation for 1/ζ(s) would converge for s > 12 +  to an analytic
function. Therefore 1/ζ(s) would not have a pole in this half plane, so ζ(s)
would fail to have a zero. Since this is so for every  > 0, ζ(s) would have no
zero to the right of the critical line, so the Riemann hypothesis would be true.

(2) Now assume RH. Let T > 0 and x > 0, considered large, x  N and
  2+iT s 
 1 x ds 
Δ(x, T ) :=  M(x) − 
 2πi 2−iT ζ(s) s 
90 Classical Equivalences
  2+iT  
 ∞  x  s ds 
=  M(x) −
1
μ(n) 
 2πi n s 
 2−iT n=1
 2+iT  
 ∞  x  s ds 
=  μ(n) −
1
μ(n) 
 n<x 2πi 2−iT n=1 n s 
  2+iT 
  x  s ds
=  μ(n) −
1
μ(n)
 n<x 2πi 2−iT n<x n s
 2+iT  
 x  s ds 
−
1
μ(n)  .
2πi 2−iT n>x n s 
Therefore, using Lemma 4.14, Lemma 4.15, |μ(n)| ≤ 1, for values of x
restricted to being half an integer so that |log(x/n)| ≥ 2/n for x/2 ≤ n < 2x
and |log(x/n)| ≥ log 2 for x/2 > n or n > 2x, this gives
 (x/n)2  (x/n)2
Δ(x, T ) ≤ +
n<x
πT |log(x/n)| n>x πT |log(x/n)|
 (x/n)2  (x/n)2
≤ +
πT |log(x/n)| x/2≤n<2x πT |log(x/n)|
⎛
x/2>n or n>2x
⎞
x2 ⎜⎜⎜⎜ 1  1  1 ⎟⎟⎟
∞
≤ ⎜⎜ + ⎟⎟⎟
πT ⎝ log 2 n=1 n2 x/2<n≤2x n ⎠
 2 
x2 π
≤ + (log(2x) − log(x/2))
πT 6 log 2
x2
.
T
Now let  > 0 be given and integrate x s /(ζ(s)s) around the rectangle
 
[ 12 +  − iT, 2 − iT ] × 12 +  − iT, 12 +  + iT
to derive an alternative expression for the integral approximation to M(x)
using the approximation to 1/|ζ(s)| δ tδ given, since we are assuming RH,
in item (8) of Section 2.2:
 2+iT s  1/2+−iT s  1/2++iT s
1 x ds 1 x ds 1 x ds
= +
2πi 2−iT ζ(s) s 2πi 2−iT ζ(s) s 2πi 1/2+−iT ζ(s) s
 2+iT
1 x s ds
+
2πi 1/2++iT ζ(s) s
 1/2++iT0 s 
δ−1 2 
 1 x ds 
≤ 2Kδ T x + 
 2πi 1/2+−iT0 ζ(s) s 
 T
2 dt
+ x1/2+ Kδ tδ
2π T0 t
1/2+ δ
x K δ T
x1/2++2δ + .
πδ
 4.4 Errors in Arithmetic Sums 91

If T = x2 then M(x) will be less than some constant (dependent on δ) times

1/2++2δ
x for all large x of the form of an integer plus one half. But since
|μ(n)| ≤ 1, M(x) can change by at most one between these values of x, so
replacing  by /3 and δ by 2/3 we get M(x)  x1/2+ , which completes the
proof of Liouville’s theorem. 

There is an interesting story concerning the bounds for M(x) := n≤x μ(n)
[6, p. 91]. What became known as Mertens’ conjecture was the statement
√
|M(x)| ≤ x, x ≥ 1.
That this might be so came from computational evidence (see for example
Figure 4.2). However it was shown in 1985, by Odlyzko and te Riele [129],
that this conjecture was false, and failed for an infinite number of values
x with limit value infinity. The best-known unconditional result is that of
Walfisz [177] who showed in 1963 that
 
log3/5 x
M(x) x exp A .
loglog1/5 x
It follows from Theorem 4.16 directly that if the squarefree integers less
than or equal to N are considered, and D(N) denotes the absolute value of
the difference between the number of those divisible by an even number of
primes and the number of those divisible by an odd number of primes, then
we have the following:

M(x)

0.5

x
20 000 40 000 60 000 80 000 100 000

-0.5

√
Figure 4.2 The function M(x)/ x for 1 ≤ x ≤ 105 .
92 Classical Equivalences

Corollary 4.17 (Parity criterion) The Riemann hypothesis is equivalent to

the statement, as N → ∞,

D(N) N 1/2+ .

Another equivalence to RH, sometimes attributed to Landau, but an

easy consequence of the Littlewood criterion, is an expression in terms
of Liouville’s function. Let n ∈ N and as usual Ω(n) is the number of
primes dividing n including multiplicity, and let λ(n) = (−1)Ω(n) be Liouville’s
completely multiplicative function. Then the prime number theorem is
equivalent to the statement
λ(1) + λ(2) + · · · + λ(N)
lim = 0,
N→∞ N
or in other words

L(x) := λ(n) = o(x), x → ∞.
n≤x

Theorem 4.18 (Liouville criterion) The Riemann hypothesis is equivalent

to the estimate that for all  > 0 we have
λ(1) + λ(2) + · · · + λ(N)
lim = 0.
N→∞ N 1/2+
Proof Checking at prime powers and using the multiplicativity of λ and μ
we have for n ∈ N
 n
λ(n) = μ 2 .
d
2 d |n

Then
 n    x
L(x) = μ 2 = μ(e) = M 2 .
d √ √ d
d |n
1≤n≤x 2 2 d≤ x e≤x/d d≤ x

Therefore if M(x)  x1/2+ then it follows that L(x)  x1/2+ . By Möbius

inversion we also have
√
x x
M(x) = μ(d)L 2 .
d=1
d

From this it follows that if L(x)  x1/2+ then M(x)  x1/2+ . The criterion
then follows from the Littlewood criterion, Theorem 4.16. 
 4.5 Unsolved Problems 93

4.5 Unsolved Problems

(1) Find an  > 0 for which as x → ∞ we have M(x) x1− .
(2) Find explicit forms of the equivalence for the extended or generalized
Riemann hypothesis for Dirichlet L-functions, first for a single non-
trivial character and then for a class of characters [13, p. 56].
(3) Use Trudgian’s form for B(T ) in Section 2.2, and determine whether
Theorem 4.5 could be significantly improved.
(4) Examine the proof of Littlewood’s theorem, Theorem 4.13, and deter-
mine whether it could be improved. For example maybe using
√
ψ(x) = x + Ω± ( x loglog x).
 5
Euler’s Totient Function

5.1 Introduction
This chapter is based on the work of Jean-Louis Nicolas, J. Barkley Rosser
and Lowell Schoenfeld, as published in a variety of articles, but most
especially in [126, 127, 145].
J. Barkley Rosser (1907–1989) was a logician and student of Alonzo
Church. Together they proved the “Church–Rosser theorem” of the lambda
calculus. He also undertook research in prime number theory where
“Rosser’s rule” is named after him. Less is known about Lowell Schoenfeld,
who worked in the main in number theory and complex analysis. His joint
1962 paper with Rosser, giving explicit estimates for many arithmetical
functions, has 358 citations registered by MathSciNet. Parts of that work are
detailed in Chapter 3.
Jean-Louis Nicolas received his Ph.D. in 1968 with a thesis entitled “The
order of the maximal element of the group S n and permutations of highly
composite numbers” supervised by Charles Pisot. His work on the symmetric
group is outlined in Chapter 10. His first student was Guy Robin, who
graduated in 1983, some of whose work is described in Chapter 7.
The equivalences to RH, which are the subject of this chapter, are
expressed in terms of the well-known arithmetical function ϕ(n). This was
defined by Euler in 1763 as “the multitude of numbers less than n which
have no common divisor with it”. He proved that
 
ϕ(n)  1
= 1− ,
n p|n
p

where the product is over primes which divide n, and proved his theorem
aϕ(n) ≡ 1 mod n when (a, n) = 1. It was Gauss who introduced the now
standard notation ϕ(n) in his 1801 “Disquisitiones Arithmeticae”, and
J. J. Sylvester in 1879 who called the function a “totient”.

94
 5.1 Introduction 95

Figure 5.1 Jean-Louis Nicolas.

Some idea of the behaviour of ϕ(n) for large n can be obtained from the
average order
3n2
ϕ(1) + ϕ(2) + · · · + ϕ(n) ∼
π2
and the spread of values. These are such that {n/ϕ(n) : n ∈ N} is dense
in (1, ∞),
ϕ(n) ϕ(n)
lim inf =0 and lim sup = 1.
n→∞ n n→∞ n
In addition we have for all  > 0
ϕ(n)
lim = ∞.
n→∞ n1−
Finally, in this brief survey of some of the properties of ϕ(n), we have
ϕ(n) loglog n
lim inf = e−γ
n→∞ n
but for n ≥ 2,
n 3
< eγ loglog n + ,
ϕ(n) loglog n
96 Euler’s Totient Function

whereas eγ loglog n < n/ϕ(n) for an infinite number of integers n. These

properties are treated in this chapter.
Background material on ϕ(n), including many of these results, can be
found in the standard texts by Hardy and Wright [75] and De Koninck and
Luca [95, chapter 8], and in the “Handbook of Number Theory”, volume I
[150, chapter I]; see also Erdős [54]. Historical material is in Dickson [46,
volume I, chapter V].
Here is a summary of the content of the chapter. We begin with estimates
relating to ϕ(n) developed by Rosser and Schoenfeld. For example these
include the proof (Lemma 5.12) that for all x ≥ 286 we have
    
1 p 1
eγ log x 1 − < < e γ
log x 1 + ,
2 log2 x p≤x
p−1 2 log2 x

where the lower bound also holds for 1 < x < 286. We also call on explicit
estimates from Chapters 3 and 4. These explicit estimates provide an essential
and important underpinning for the RH equivalences.
The first main result, of Rosser and Schoenfeld, Theorem 5.25, comes after
quite a large number of lemmas.
Theorem 5.25 Without any hypotheses, for all n ≥ 3 except for the ninth
primorial (see below) N9 = 2 · 3 · · · 23, we have
n 5
< eγ loglog n + . (5.1)
ϕ(n) 2 loglog n
The second main result, that of Nicolas, Theorem 5.29, is an essential tool,
used in this chapter and in Chapter 7, when we assume RH is false. Here we
use the function
 1

f (x) := eγ log θ(x) 1− ,
p≤x
p

where as usual θ(x) := p≤x log p.

Theorem 5.29 Assuming RH is false, there exists a real number b with

0 < b < 12 such that as x → ∞
 
1
f (x) = Ω± b . (5.2)
x
We then build to proving what we call Nicolas’ first criterion, which is
Theorem 5.31, from 1983. It is very similar in structure to that of Robin’s
theorem, Theorem 7.11, but is expressed in terms of primorials. Let Nk be
the kth primorial, i.e. the product of the first k prime numbers in increasing
order of size.
 5.1 Introduction 97

Theorem 5.31 (Nicolas’ first criterion) If RH is true then for all k ≥ 1 we

have
Nk
eγ loglog Nk < . (5.3)
ϕ(Nk )
If however RH is false then this inequality is true for an infinite number of k
and false for an infinite number of k.
There has been, apparently, little discussion of expanding the set of
solutions to the inequality (5.3) beyond the primorials, even with RH. The
author suspects that much more is true than Theorem 5.31. For example, up
to 109 , every positive integer which is a multiple of a primorial and is less
than the following primorial satisfies the given inequality. There has also
been little discussion on which primorials qualify in the different situations
when RH fails.
Nicolas’ second theorem, Theorem 5.34, is much more recent and includes
four equivalences to RH in terms of ϕ(n). We will state just the first in this
introduction.
Theorem 5.34 Let
 
n 
c(n) := − eγ loglog n log n.
ϕ(n)
Then RH is equivalent to
lim sup c(n) = eγ (2 + β),
n→∞

where β = 2 + γ − log 4π.

The equivalence, Corollary 5.35, is easy to prove, but shows that
Theorem 5.34 is consistent with other results, such as those of Chapter 7.
It says that RH is equivalent to the inequality
n eγ (2 + β)
< eγ loglog n + 
ϕ(n) log n
for n ≥ N120569 . We call this form Nicolas’ second criterion.
Proving these theorems requires a set of lemmas which we prove in several
sections below. Estimates for prime products and sums are in Section 5.2.
Preliminary results with RH true, mostly dating from 1983, are in Section 5.3.
The further results with RH true, given in Section 5.4, correspond roughly to
Nicolas’ more recent work.
Nicolas’ first theorem follows in Section 5.6. In this section we include a
note regarding what would result if RH fails. In that case an infinite number
of primorials for which the inequality (5.3) were true would each be followed
by a primorial for which it was false. Finally we have in Section 5.7 the proof
of Nicolas’ second theorem.
98 Euler’s Totient Function

Thm 2.4 Lem 5.7

Lem 5.1 Lem 5.8

Lem 4.2 Lem 5.3

Lem 5.2 Lem 5.5

Lem 5.6
Lem 5.9

Lem 3.14

Lem 5.10

Lem 5.4

Thm 2.1

Figure 5.2 Some relationships between results in Chapter 5.

Navigating through the results in this chapter may at times be challenging.

To aid the reader, the relationships between some of the lemmas, theorems
and corollaries are set out in the directed graphs in Figures 5.2, 5.3
and 5.7.
The following RHpack (see Appendix B) functions relate to the material
in this chapter: NOverPhi, NicolasInequalityQ, Computef, NicolasTwoFail-
ures, ComputeH, ComputeDeltas, FHalf, PlotEulerPhiRatio, PlotInequality,
CheckCNk, ComputeC and PlotCNk.

5.2 Estimates for Euler’s Function ϕ(n)

We commence with a definition of the function V(x), which forms an
essential part of the statements and proofs of results in this part of the
development of ideas. Note that in the definition of V(x), the sum is over
 5.2 Estimates for Euler’s Function ϕ(n) 99

Thm 5.31 Lem 5.28

Lem 5.30

Thm 5.29 Thm 4.12

Lem 3.13

Lem 3.10 Lem 5.15 Lem 5.21

Lem 5.16

Lem 5.22

Lem 5.23

Lem 5.20

Figure 5.3 Some relationships between results in Chapter 5.

all imaginary zeros ρ = β + iγ of ζ(s):

 xβ−1
V(x) := .
ρ
γ2

Considering zeta zeros ρ = β + iγ, a critical parameter is denoted H, where for

all 0 < γ ≤ H, β = ρ = 12 , i.e. H is the height below which all zeta zeros with
positive imaginary part have been shown to be on the critical line. A variety
of values for H were used in the papers of Rosser and Schoenfeld. Even in the
1962 paper both γ57 and γ25000 were used. We will assume that ρ = 12 up to
H := 5.4 × 108 < γ15×108 (see Table 2.2), rather than the best-known available
value of H, which is larger. A variety of values for the parameter R were also
used in their papers. First we recall the description of a zero-free region in
the critical strip implied by Theorem 2.4: if ζ(ρ) = 0 with ρ = β + iγ and H < γ
then β < 1 − 1/(R log γ).
100 Euler’s Totient Function

This leads to the definition of the principal function in this part of the
development, φ(x, γ, R), which we write as φ(γ):
 
log x
exp −
R log γ
φ(γ) := .
γ2
We begin by deriving a preliminary upper bound for V(x) in terms of the
φ(γ) values. This will be made more explicit later.
Lemma 5.1 For x > 1 we have
1  1 
V(x) ≤ √ + φ(γ).
x ρ γ2 H<γ

Proof Splitting the sum in the definition of V(x) we get

 xβ−1  xβ−1
V(x) = + .
β≤1/2
γ2 β>1/2 γ2

For the first sum

 xβ−1 1  1 1  1
≤ √ ≤ √ ,
β≤1/2
γ2 x β≤1/2 γ2 x ρ γ2

and for the second

 xβ−1  xβ−1
= 2 .
β>1/2
γ2 β>1/2, γ>0
γ2

In addition, by Theorem 2.4, for γ > H we have

 
log x
exp −
xβ−1 R log γ
< = φ(γ),
γ 2 γ2
and so
 xβ−1  
< 2 φ(γ) ≤ φ(γ).
β>1/2
γ2 β>1/2, γ>0 γ>H

Putting these bounds together completes the proof. 

To see how this and the next few lemmas work together, we have the
network flow diagram given in Figure 5.2.
Recall from Section 2.2 Backlund’s result, and its improvements, for the
difference between the number N(T ) of zeta zeros with positive imaginary
part up to height T and its approximation F(T ), namely B(T ) := a1 log T +
a2 loglog T + a3 . Also recall the definition φ(γ) = exp(−log x/(R log γ))/γ2 .
 5.2 Estimates for Euler’s Function ϕ(n) 101

We are now able to bound sums of the φ(γ) in terms of two supplementary
functions which depend on a1 , a2 and a3 using the generalized form of Abel
summation.
Lemma 5.2 [145, theorem 27] If 2 ≤ h then
  ∞
t
φ(γ) < 2P(h)e−log x/(R log h) + Q(h) φ(t) log dt
h<γ h 2π

where
a1 log h + a2 loglog h + a3
P(h) =
h2
and
1 a1 log h + a2
Q(h) = + .
2π h log h log(h/(2π))
Proof By Lemma 4.2 and Backlund’s bound B(T ) we can write
  ∞  t    
1 a2 1
φ(γ) < log + a1 + φ(t) dt + 2B(h)φ(h).
h<γ h 2π 2π log h t

But for h ≤ t we have

h log(t/(2π)) 1 1
≤ and ≤ .
t log(h/(2π)) t log(t/(2π)) h log(h/(2π))
Therefore
   ∞  t 
1 a1 log h + a2
φ(γ) < + φ(t) log dt + 2B(h)φ(h)
h<γ
2π h log h log(h/(2π)) h 2π
 ∞  t 
−log x/(R log h)
= 2P(h)e + Q(h) φ(t) log dt.
h 2π
This completes the proof. 

For m ≥ 1 recall the definition km (h) := ρ,γ>h 1/γm+1 . These functions
received quite a lot of attention in Section 3.2 – see for example Table 3.1.
We next derive an upper bound for V(x) in terms of k1 (0) and k1 (H).
Lemma 5.3 [145, lemma 6] For x ≥ 1 and for H ≤ h we have
k1 (0) 2B(h) 1 + log(h/(2π)) k1 (H)
V(x) < √ + 2 + Q(H) + 1/(R log h) .
x h h x
Proof First note that
 ∞ −1/(R log t)  ∞
x log(t/(2π)) log(t/(2π)) 1 + log(h/(2π))
2
dt ≤ 2
dt = .
h t h t h
102 Euler’s Totient Function

Using this integral together with Lemmas 5.1 and 5.2 we get:
 xβ−1 1  1 
V(x) = ≤ √ + φ(γ)
ρ
ρ2 x ρ γ2 H<γ
k1 (0)   k1 (0)  x−1/(R log h)  1
< √ + φ(γ) + φ(γ) ≤ √ + +
x H<γ≤h h<γ
x H<γ≤h γ2 h<γ
γ2
k1 (0) 2B(h) 1 + log(h/(2π)) k1 (H)
< √ + 2 + Q(H) + 1/(R log h) .
x h h x
This completes the derivation. 
Next we link ψ(x) with V(x). This takes the form of upper bounds for two
infinite integrals. Von Mangoldt’s first theorem plays a crucial role here.
Lemma 5.4 [145, lemmas 7 and 8] For all x > 1 we have the upper bounds
 ∞ 
 (ψ(t) − t)  log 2π 1 − log 2
(1) I1 :=  2
dt < V(x) + +
x t x x3
and
 ∞ 
 (ψ(t) − t)(1 + log t) 
(2) I2 :=  dt
 x t2 log2 t 
 
2 + log x log 2π 1 − log 2
< V(x) + + ,
log2 x x x3
where log 2π < 1.84 and 1 − log 2 < 0.31.
Proof (1) Let ρ be a complex zeta zero and note that
 ∞
xρ−1
tρ−2 dt = − .
x ρ−1
Using von Mangoldt’s theorem, Theorem 2.1, and observing that |γ| ≤ |ρ| and
|γ| ≤ |1 − ρ| gives
 ∞   ∞
ψ(t) − t   xβ−1 log 2π − 12 log(1 − 1/t2 )
I1 =  dt 
 ≤ + dt
x t2 ρ
|ρ(ρ − 1)| x t 2

log 2π  x−2n−1
∞
≤ V(x) + +
x n=1
2n(2n + 1)
log 2π 
∞
x−3
≤ V(x) + +
x n=1
2n(2n + 1)
log 2π 1 − log 2
< V(x) + + ,
x x3
which gives the bound for the integral.
 5.2 Estimates for Euler’s Function ϕ(n) 103

(2) For the second integral, integrating twice by parts we get

 ∞ ρ−2  ρ−1  ∞ ρ−2 
t (1 + log t) xρ−1 ρ x t
dt = − − −2 dt .
x
2
log t (ρ − 1) log x (ρ − 1) log x
2 2 3
x log t

Also
 ρ−1  ∞ ρ−2   ∞ β−2
 x t  xβ−1 t
 2 − 2 3
dt ≤
 2
+2 3
dt
log x x log t log x x log t
 ∞
xβ−1 β−1 1 2xβ−1
< + 2x dt = .
log2 x 3
x t log t log2 x
Therefore
  ∞ ρ−2 
 1 t (1 + log t)  1 xβ−1 1 2xβ−1
 dt  < +
ρ x log2 t  |ρ(ρ − 1)| log x |ρ − 1|2 log2 x
2 + log x xβ−1
< .
log2 x γ2
Summing we get
 
  1  ∞ tρ−2 (1 + log t)  2 + log x
 dt < V(x).
ρ x log2
t  log2 x
ρ

Now we apply von Mangoldt’s theorem, Theorem 2.1, to get

 ∞ 
 (ψ(t) − t)(1 + log t)  2 + log x
I2 =  dt < V(x) + I3 (x),
 x t2 log2 t  log2 x
where
 ∞
(2 log 2π − log(1 − 1/t2 ))(1 + log t)
I3 (x) = dt
x 2t2 log2 t
   
1 + log x ∞ 1 dt
< 2 log 2π − log 1 − 2
2 log2 x x t t2
 
2 + log x log 2π 1 − log 2
< + .
log2 x x x3
This completes the proof. 
Next we have an easy-to-derive upper bound for an infinite integral which
appears in the proof of the previous lemma.
Lemma 5.5 [145, lemma 9] For 0 ≤ α < n and x > 1 we have
 ∞ α
t (1 + n log t) n 1
dt ≤ .
x
n+1
t log t 2
n − α x log x
n−α
104 Euler’s Totient Function

Proof Rewrite the form of the integral and then use integration by parts:
 ∞ α  ∞  
t (1 + n log t) α d 1
dt = − t dt
x tn+1 log2 t x dt tn log t
 ∞
1 tα
= n−α +α n+1 log t
dt
x log x x t
 ∞
1 α
≤ n−α + tα−n−1 dt
x log x log x x
n 1
= ,
n − α x log x
n−α

which completes the proof. 

The next step uses three functions to estimate θ(x):
 xβ−1
V(x) := ,
ρ
γ2
 
2 + log x 1.84 0.31 2.04 4.5 1.02
L(x) := V(x) + + 3 + 1/2 + 2/3 +
log x x x x x x−1
and
 
1 α
M(x, α) := log x log 1 + 2
− .
2 log x log x
We have already employed and found bounds for the function V(x). The
function L(x) arises below as part of a term in an upper bound, but is too
difficult to work with. We will find ranges of variables and values of a
parameter α for which L(x) < M(x, α), so we can replace L(x) by M(x, α)
in those ranges. The reader is invited to investigate whether L(x) and M(x, α)
are of comparable orders. If not, this step could be improved.
Lemma 5.6 [145, lemmas 11 and 12, theorems 29 and 31] We have the
following bounds:
(1) 108 ≤ x ≤ e600 =⇒ L(x) < M(x, 0.31),
(2) e375 ≤ x =⇒ L(x) < M(x, 0.47),   
0.31 0.31
(3) 1451 ≤ x ≤ e 375
=⇒ x 1 − < θ(x) < x 1 + ,
  log x   log x
0.47 0.47
(4) 569 ≤ x =⇒ x 1 − < θ(x) < x 1 + .
log x log x
Proof First we will derive a useful sufficient criterion. Let H ≤ u ≤ v and
0 < α < 12 be real numbers with
1
+α
log u ≥ κ(α) := 2
and L(u) < M(v, α),
1 − 2α
 5.2 Estimates for Euler’s Function ϕ(n) 105

and let u ≤ x ≤ v. Then, since L(x) and M(x, α) are decreasing as x increases,
we have

L(x) ≤ L(u) < M(v, α) ≤ M(x, α) =⇒ L(x) < M(x, α).

(1) In this case u = 108 , v = e600 . Let h → ∞ in Lemma 5.3 and set α = 0.31
to get κ(0.31) < 1.460 to obtain

L(u) = L(108 ) < 3.0 × 10−4 < M(e600 , 0.31) = M(v, 0.31)
=⇒ L(x) < M(x, 0.31) for 108 ≤ x ≤ e600 ,

proving part (1).

(2) Now set h = e25 in Lemma 5.3 to get
k1 (0) k1 (H) 1
V(x) < √ + 1/(25R) + 10 .
x e 10
Note also that κ(0.47) < 4.021 so
3
L(e375 ) < 2.9 × 10−5 < M(e10 , 0.47),
3 3
L(e10 ) < 7.0 × 10−6 < M(e4×10 , 0.47)

and
3 6
L(e4×10 ) < 9.0 × 10−9 < M(e3×10 , 0.47).

Then let in turn

[u, v] = [exp(375), exp(103 )], [exp(103 ), exp(4 × 103 )] and

[exp(4 × 103 ), exp(3 × 106 )],

to show L(x) < M(x, 0.47) for exp(375) ≤ x ≤ exp(3 × 106 ).

Finally, for x ≥ exp(3 × 106 ) set
log x
log h =
R(loglog x − log 400)
and again use the definitions of L(x) and M(x, α), the bound k1 (0) < 0.0463
and Lemma 5.3 to show
0.028
L(x) < < M(x, 0.47).
log x
(3) and (4) Parts (3) and (4) are proved in Lemma 3.14(b) and (d). 
Next we have a simple application of Abel summation to obtain an estimate
for an infinite sum of primes to negative powers.
106 Euler’s Totient Function

Lemma 5.7 Let α > 1. Then as x → ∞ we have

 1  
1 1
= +O ,
p≥x
pα (α − 1)xα−1 log x xα−1 log2 x
where the sum is over primes.
Proof Use Abel’s identity [6, theorem 4.2], π(x) = x/(log x) + O(x/(log x)2 )
and integration by parts to derive
 1  ∞
π(x) π(t)
α
= − α +α dt
p x x tα+1
p≥x
    ∞  ∞ 
1 x x dt dt
=− α +O + α α
+ O
x log x log!2 x α 2
x t log t x t log t
"∞  ∞ 
1 α 1 dt
= − α−1 + − α−1 +O
x  log x α −  1 t log t x
α 2
x t log t
1
+O .
α−1
x log2 x
Therefore
 1     ∞ 
1 1 1 dt
= + O + O
p α (α − 1)x α−1 log x α−1
x log x 2
log x x tα
2
p≥x
 
1 1
= +O .
(α − 1)x log x
α−1
xα−1 log2 x
This completes the proof. 
Next we derive a useful upper bound for another type of prime sum. The
constant 1.02 could be improved, but we do not need this.
Lemma 5.8 [145, p. 87] We have the upper bound for a sum over primes
valid for 1 < x:
1 1 1.02
< ,
n≥2
n p>x
p n (x − 1) log x

so if
 
1

1

S := − log 1 − + ,
p>x
p p

we have 0 ≥ S > −1.02/((x − 1) log x).

Proof A re-examination of the proof of Lemma 5.7, using f (x) := 1/xn , gives
the upper bound for n ≥ 2:
 1  ∞
θ(x) θ(t) 1 + n log t
<− n + dt.
p>x
p n x log x x t tn log2 t
 5.2 Estimates for Euler’s Function ϕ(n) 107

By Lemma 3.10 we have θ(t) < 1.02t for all t > 0 so therefore using
Lemma 5.5 we get
 1  ∞
1 + n log t 1.02n 1.02n
< 1.02 dt ≤ ≤ n−1 ,
p>x
p n
x
n 2
t log t (n − 1)x log x x log x
n−1

and so
1 1 1.02  1 1.02
< =
n≥2
n p>x
p n x log x n≥2
x n−2 (x − 1) log x

and the proof is complete. 

With these preliminaries in place, we now finesse the error in an explicit
form of Mertens’ estimate for the sum of prime reciprocals.
Lemma 5.9 [145, lemma 13] There is an absolute constant B such that for
all x ≥ 108 we have
   
 1  1 1.02
 
− loglog x − B < log 1 + − .
 p≤x p  2
2 log x (x − 1) log x

Proof (1) By Abel’s identity [6, theorem 4.2], with f (t) = 1/(t log t) and
an = log n for prime n and 0 otherwise, we have
1  x  
θ(x) d 1
= − θ(t) dt
p≤x
p x log x 2 dt t log t
 x  ∞  
θ(x) − x dt d 1
= + + B+ (θ(t) − t) dt,
x log x 2 t log t x dt t log t
where
 ∞  
1 d 1
B := − loglog 2 − (θ(t) − t) dt ∈ R.
log 2 2 dt t log t
It follows that
1  ∞
θ(x) − x (θ(t) − t)(1 + log t)
− loglog x − B = + dt.
p≤x
p x log x x t2 log2 t

(2) Next by Lemma 5.5 with n = 1 we get

 ∞
1 1 + log t 2
α= =⇒ 2
dt ≤ √
2 x t
3/2 log t x log x
and
 ∞
1 1 + log t 3
α= =⇒ 2
dt ≤ 2/3 .
3 x
5/3
t log t 2x log x
108 Euler’s Totient Function

Using these bounds and Lemmas 5.4 and 3.10 we get

   
 1  |θ(x) − x|  ∞ (ψ(t) − t)(1 + log t) 
 − loglog x − B ≤ + dt
 p≤x p  x log x  x t2 log2 t 
 ∞
(ψ(t) − θ(t))(1 + log t)
+ dt
x t2 log2 t
  
|θ(x) − x| 2 + log x 1.84 0.31
< + V(x) + + 3
x log x log2 x x x
 ∞
(1.02t + 3t )(1 + log t)
1/2 1/3
+ dt.
x t2 log2 t
Therefore, using the definition of L(x), namely,
 
2 + log x 1.84 0.31 2.04 4.5 1.02
L(x) := V(x) + + + 1/2 + 2/3 + ,
log x x x3 x x x−1
we get:
 
 1  |θ(x) − x| L(x)
− loglog x − B <
1.02
 + − .
 p≤x p  x log x log x (x − 1) log x

(3) To complete the proof we split the range. Firstly let 108 ≤ x ≤ e375 . By
Lemma 5.6 parts (1) and (3) we have in this range
0.31x
|θ(x) − x| < and L(x) < M(x, 0.31).
log x
Using the formula derived in Step (2), and the definition of M(x, α), namely,
 
1 α
M(x, α) := log x log 1 + 2
− ,
2 log x log x
we therefore have
 
 1  0.31 M(x, 0.31)
− loglog x − B <
1.02
 + −
 p≤x p  log x2
log x (x − 1) log x
 
1 1.02
= log 1 + − .
2
2 log x (x − 1) log x
(4) Now let e375 ≤ x. Using Steps (2) and (4) of Lemma 5.6, we have in
this range
0.47x
|θ(x) − x| < and L(x) < M(x, 0.47),
log x
and the result follows as in Step (3). 
 5.2 Estimates for Euler’s Function ϕ(n) 109

We now have a useful estimate which can easily be derived from

Lemma 5.9.
Lemma 5.10 [145, theorem 5] For all x ≥ 286 we have
1 1 1
loglog x + B − 2
< < loglog x + B + ,
2 log x p≤x p 2 log2 x
where the lower bound also holds for 1 < x < 286.
Proof Use the upper bound log(1 + x) < x for 0 < x < 1 and then remove the
absolute value to derive the result directly from Lemma 5.9 in the range x ≥
108 . For 2 ≤ x ≤ 108 a computation verifies the result – see Section 3.3. 
We now give two estimates for partial Euler products which apply in an
explicit finite range. The reader might compare [145, theorem 23, p. 73].
Lemma 5.11 For 2 ≤ x ≤ 108 we have
 1
−1 
2

γ
1− ≤ e log x 1 + √
p≤x
p x log x
and
 
 1 
− loglog x − B <
1
 .
 p≤x p  2 log2 x
Proof These follow from a computation at integers – see Section 3.3. 
The following explicit form of Mertens’ theorem [115] plays an essential
role in the development. See for example [75, chapter 22, theorems 427 and
428] or [6]. Recall that the constant B is normally defined by [95, p. 56]
  1

1

B := γ + log 1 − + ,
p
p p
where as usual γ is Euler’s constant.
Lemma 5.12 [145, theorem 8] For all x ≥ 286 we have
    
γ 1 p γ 1
e log x 1 − < < e log x 1 + ,
2 log2 x p≤x
p−1 2 log2 x
where the lower bound also holds for 1 < x < 286.
Proof First
     
p 1
log = − log 1 −
p≤x
p−1 p≤x
p
1 1 1
= + .
p≤x
p n≥2 n p≤x pn
110 Euler’s Totient Function

Now we can write, from the definition of B,

  1

1
 1 1
B=γ+ log 1 − + =γ− .
p
p p n≥2
n p>x pn

From Lemma 5.9 we have

 
1
B + loglog x − log 1 +
2 log2 x
1  
1 1.02
< < B + loglog x + log 1 + − .
p≤x
p 2
2 log x (x − 1) log x

Using these bounds and replacing B with the expression we have derived
gives
    
1 1
γ + loglog x − log 1 + < − log 1 −
2 log2 x p≤x
p
 
1
< γ + loglog x + log 1 + .
2 log2 x
Then because log(1 − x) < −log(1 + x) for 0 < x < 1 we get
    
1 1
γ + loglog x + log 1 − < − log 1 −
2 log2 x p≤x
p
 
1
< γ + loglog x + log 1 + .
2 log2 x
Finally, take exponentials to complete the proof. 

5.3 Preliminary Results With RH True

Since we are deriving equivalences of RH, one main part of any proof must
assume RH is true, and the other main part considers properties which apply
if it is assumed to be false. In order that these two vastly different situations
not be confused, every lemma or theorem which uses one or other of these
will include a reference to the particular assumption in the statement. If no
such reference appears, then the reader can safely assume no such assumption
is needed.
We use the standard definitions for θ(x) and ψ(x). For x > 0,
 
θ(x) := log p, ψ(x) := log p, (5.4)
p≤x n≥1, pn ≤x

where p is a prime number.

 5.3 Preliminary Results With RH True 111
√
Let S (x) := θ(x) − x and T (x) := x log2 x/(8π). Recall the result of
Theorem 4.6 – an estimate for S (x), assuming RH, which is used many times
in what follows: for x ≥ 599 we have |S (x)| ≤ T (x).
The function f (x), which we will now define, is fundamental in this work.
It comes from slightly varying Mertens’ theorem, an explicit form of which
has just been proved in Lemma 5.12:
⎛
⎜⎜⎜  ⎞⎟−1
1 1 ⎟⎟ 1
1− 2
< ⎜⎜⎜⎝eγ log x 1 − ⎟⎟⎟⎠ < 1 + , x > 1.
2 log x p≤x
p 2 log2 x
Now for x ≥ 2 let
 1

γ
f (x) := e log θ(x) 1− . (5.5)
p≤x
p
Estimating f (x) is a primary goal. We begin with an upper bound on a finite
domain.
Lemma 5.13 Assume RH is true. For 2 ≤ x ≤ 108 we have f (x) < 1.
Proof By Lemma 3.10, for 1 < x ≤ 108 we have θ(x) < x and by a computation
in this range (see Section 3.3)
 p
eγ log x < .
p≤x
p−1

Therefore for 2 ≤ x ≤ 108 we get, using Lemma 5.11,

 p−1  p−1
f (x) = eγ log θ(x) < eγ log x < 1.
p≤x
p p≤x
p

Treating the infinite range x > 108 is much more difficult and will take us
on quite a journey.
To begin, for x > 1 define R(x) := ψ(x) − x and define also an associated
integral expression which is also fundamental to this work:
 ∞  
R(t) 1 1
J(x) := + dt. (5.6)
x t2 log t log2 t
We need a simple lemma:
Lemma 5.14 Let a, b > 0 be real and let g(t) := loga t/tb . Then g(t) is positive
and decreasing for t > ea/b and max{g(t) : t ≥ 1} = (a/(eb))a .
Proof Since
g (t) a − b log t
= ,
g(t) t log t
the proof follows with some simple calculus. 
112 Euler’s Totient Function

To show how Lemma 5.15, and some of the results which follow, fit into
the overall scheme of the proof, we have another network flow diagram
(Figure 5.3).
We have defined S (x) := θ(x) − x. Analogous to J(x) let
 ∞  
S (t) 1 1
K(x) := + dt. (5.7)
x t2 log t log2 t
Note that the prime number theorem gives S (t) = O(t/log t) and ensures the
convergence of the integral. Then we have an estimate for f (x) for x > 1:
Lemma 5.15 [126, proposition 1] For x > 1
S (x)2 1
K(x) − ≤ log f (x) ≤ K(x) + . (5.8)
2
x log x 2(x − 1)
Proof First use Abel summation [6, theorem 4.2] and θ(x) = x + S (x) to
derive an expression, in terms of S (x), for the partial sum over primes p
of prime reciprocals:
1  x
θ(x) θ(t)(1 + log t)
= + dt
p≤x
p x log x 2 t2 log2 t
 x
S (x) 1 S (t)(1 + log t)
= + loglog x − loglog 2 + + dt,
x log x log 2 2 t2 log2 t
so
1  ∞
S (x) S (t)(1 + log t)
= + loglog x + B − dt
p≤x
p x log x x t2 log2 t
S (x)
= + loglog x + B − K(x),
x log x
with
 ∞
1 S (t)(1 + log t)
B := − loglog 2 + + dt.
log 2 2 t2 log2 t
Thus
  1  S (x)
K(x) = − + + loglog x + B. (5.9)
p≤x
p x log x
Note also the given expression for B can be compared with the classical
formula given by Lemma 5.9:
1  
1
= loglog x + B + O ,
p≤x
p log x
 5.3 Preliminary Results With RH True 113

where in this case

 
1

1

B=γ+ log 1 − + .
p
p p

Now we are able to derive an expression  for log f (x) where the reader will
γ
recall the definition f (x) := e log θ(x) p≤x (1 − 1/p) (see Figure 5.4). Then
  1

log f (x) = loglog θ(x) + γ + log 1 −
p≤x
p
⎛   ⎞ ⎛     ⎞
⎜⎜⎜  1 ⎟⎟⎟ ⎜⎜⎜ 1 1 ⎟⎟⎟
⎜
= ⎜⎝⎜loglog θ(x) + ⎟ ⎜
− + B⎟⎠⎟ + ⎜⎝⎜ log 1 − + + γ − B⎟⎟⎠⎟ .
p≤x
p p≤x
p p

Thus
log f (x) = U(x) + u(x),
where, using the classical expression for B, we have
 
1

1

u(x) = − log 1 − − .
p>x
p p

Therefore
 1 
U(x) := loglog θ(x) + − +B
p≤x
p

f(x)
0.99

0.98

0.97

0.96

0.95

0.94

x
0.93 200 400 600 800 1000

Figure 5.4 The function f (x).

114 Euler’s Totient Function

and
 1  1 1
 
1 1
0 < u(x) < 1 + + 2 + ··· ≤ ≤ .
p>x
2p 2 p p n>x
2n(n − 1) 2(x − 1)

Note that by Lemma 3.10 (d), for x ≥ 121 we have θ(x) ≥ 4x/5. Note also that
for t > 1 the function
d2 1 + log t
(loglog t) = −
dt 2
t2 log2 t
is increasing. A direct computation gives, for x ≥ 121 and t := 4x/5,
1 + log t 25 log(4x/5) + 1 2
− 2
=− 2
≥− 2 .
2
t log t 16 x log (4x/5)
2 x log x
Now, using this bound, apply Taylor’s theorem to the function loglog x for
θ(x) = x + S (x), expanded about x, to get for x ≥ 121
S (x) S (x)2 S (x)
loglog x + − 2 ≤ loglog θ(x) ≤ loglog x + .
x log x x log x x log x
Then, to derive the upper bound of the lemma, use (5.9) to get
1
log f (x) = U(x) + u(x) < U(x) +
2(x − 1)
 1  1
≤ loglog θ(x) + − + B+
p≤x
p 2(x − 1)
 
S (x)  1 1
≤ loglog x + + − + B++
x log x p≤x p 2(x − 1)
1
= K(x) + .
2(x − 1)
The derivation of the lower bound is similar using the lower bound from the
application of Taylor’s theorem. 
Next define for complex z with z < 1 and real x > 1 the integral
 ∞ z 
t 1 1
Fz (x) := + dt. (5.10)
x t
2 log t
log2 t
This very useful function has some easy-to-derive properties set out in
the next lemma. Again, integration by parts is the workhorse! There is an
illustrative plot of F1/2 (x) later in Figure 5.6.

Lemma 5.16 [127, lemma 2.2] For x > 1 and z < 1,

 ∞
xz−1 ztz−2
Fz (x) = + rz (x) where rz (x) := − dt. (5.11)
(1 − z) log x x (1 − z) log t
2
 5.3 Preliminary Results With RH True 115

If z = 1/2 then
 
1 4
|rz (x)| ≤ √ 1+ . (5.12)
|1 − z| x log2 x log x
If z = 1/2 then
2 2 2 2 8
√ − √ 2
≤ F1/2 (x) ≤ √ − √ 2
+ √ . (5.13)
x log x x log x x log x x log x x log3 x
Finally, if z = 1/3 we have
3
0 ≤ F1/3 (x) ≤ . (5.14)
2x2/3 log x
Proof Split the integral and integrate the first term by parts to derive
 ∞  
1 1
Fz (x) = t z−2
+ dt
x log t log2 t
 ∞ z−2  ∞ z−2
t t
= dt + 2
dt
x log t x log t
 ∞  ∞ z−2
xz−1 tz−2 t
= + dt + dt
(1 − z) log x x (z − 1) log t
2 2
x log t
 ∞
xz−1 −ztz−2
= + dt
(1 − z) log x x (1 − z) log t
2

xz−1
= + rz (x).
(1 − z) log x
Integrating by parts again we have
  ∞ 
z xz−1 2tz−2
rz (x) = − + dt dt. (5.15)
1 − z (1 − z) log2 x x (z − 1) log t
3

If z = 1/2 then 1 − z = z̄ so (5.12) follows from

 ∞
1 2 1
|rz (x)| ≤ √ + dt
|1 − z| x log x |1 − z| log3 x
2
x t3/2
 
1 4
= √ 1+ .
|1 − z| x log2 x log x
Let z = 1/2 in (5.11) and (5.15) and use the upper bound for |rz (x)| to get
(5.13). Finally, to get (5.14) use (5.11), noting that r1/3 (x) < 0. 
The following lemma involves rather intricate sums and estimates. It
is
√ a crucial ingredient for Lemma 5.18. Recall the definition T (x) :=
x log2 x/(8π).
116 Euler’s Totient Function

Lemma 5.17 [127, lemma 2.3] Let κ = κ(x) := log x/log 2. For x ≥ 16 define

κ
H(x) := 1 + x1/n−1/3
n=4

(see Figure 5.5). For x ≥ 4 and n ≥ 2 let

T (x1/n ) log2 x
ln (x) := = .
x1/3 8πn2 x1/3−1/(2n)
Finally, let

κ
L(x) := ln (x).
n=2

Then
(a) for j ≥ 9 and x ≥ 2 j we have H(x) ≤ H(2 j ), and
(b) for j ≥ 35 and x ≥ 2 j we have L(x) ≤ L(2 j ).

Proof (a) The function H(x) is continuous and decreasing on the interval
[2 j , 2 j+1 ) for all j ≥ 1. Therefore to prove (a) we need only verify H(2 j ) ≥
H(2 j+1 ) for all j ≥ 9. If j ≥ 20 then

j
H(2 j ) − H(2 j+1 ) = 2 j(1/k−1/3) (1 − 21/k−1/3 ) − 2( j+1)(1/( j+1)−1/3)
k=4

≥2 j(1/4−1/3)
(1 − 21/4−1/3 ) − 2( j+1)(1/( j+1)−1/3)

H(x)

2.6

2.4

2.2

2.0

1.8
x
200 000 400 000 600 000 800 000 1×106

Figure 5.5 The function H(x).

 5.3 Preliminary Results With RH True 117

= 2− j/3 ((1 − 2−1/12 )2 j/4 − 22/3 )

≥ 2− j/3 ((1 − 2−1/12 )220/4 − 22/3 ) > 0.
For 9 ≤ j ≤ 19, the inequality H(2 j ) ≥ H(2 j+1 ) can be verified numerically.
(b) If j ≥ 35 then 2 j ≥ e24 . By Lemma 5.14, for each k the function lk (x) is
decreasing for x ≥ 2 j so L(x) is decreasing on [2 j , 2 j+1 ). Then

j
L(2 ) − L(2
j j+1
)= (lk (2 j ) − lk (2 j+1 )) − l j+1 (2 j+1 )
k=2
≥ l2 (2 j ) − l2 (2 j+1 ) − l j+1 (2 j+1 )
log2 2 − j/3 j/4 2 −1/12
= 2 [2 ( j − 2 ( j + 1)2 ) − 4 × 21/6 ]
32π
>0
for j ≥ 35 since the term inside the outer square brackets is positive at j = 35
and increasing for j ≥ 35. 
The next lemma gives upper and lower bounds for ψ(x) − θ(x) assuming
RH. The reader might compare these with Lemma 3.13 where RH is not
assumed.
Lemma 5.18 [127, lemma 2.4] Assume RH is true. Then (a) for x ≥ 121 we
have √
x ≤ ψ(x) − θ(x), (5.16)
(b) for x ≥ 1 we have
√
4
x )x−1/3 ≤ .
(ψ(x) − θ(x) − (5.17)
3
Proof (a) We prove this part in two ranges of x values.
(1) If 112 ≤ x < 5993 the result may be verified numerically as follows.
Let qi be the increasing sequence of squarefull powers of primes p which
are less than 5993 , so q1 = 4, q2 = 8, q3 = 9, q4 = 16, q5 = 25, . . . . On each
interval [qi , qi+1
√ ) the function ψ(x) − θ(x) is non-negative and constant and
(ψ(x) − θ(x))/ x is decreasing.
Then for each i with 52 ≤ qi < qi+1 ≤ 5993 (so 5 ≤ i ≤ 1921), compute
√
recursively δi := (ψ(qi ) − θ(qi )/ qi+1 using
Δ1 = 2 log 2,
Δ1
δ1 = √ ,
q2
and, for i ≥ 1, qi = pni for some n ≥ 2 defines pi ,
Δi+1 = Δi + log(pi+1 ),
Δi
δi = √ ,
qi+1
118 Euler’s Totient Function

to get
min δi = δ1886 > 1.03795 > 1,
11≤i≤1921

while q5 = 112 and δ4 < 1. (See the RHpack function ComputeDeltas in

Appendix B.) Therefore (a) is true when 112 ≤ x < 5993 .
(2) Now assume x ≥ 5993 . By Theorem 4.6, with S (x) = θ(x) − x, since we
are assuming RH, we have
√
|S (x)| ≤ T (x) := ( x log2 x)/(8π), (5.18)
so
ψ(x) − θ(x) ≥ θ(x1/2 ) + θ(x1/3 ) ≥ x1/2 + x1/3 − T (x1/2 ) − T (x1/3 ).
By Lemma 5.14 for x ≥ 5992 we get
 
T (x1/2 ) T (x1/3 ) 1 log2 x log2 x 20
+ = + ≤ 2 < 1.
x1/3 x1/3 8π 4x1/12 9x1/6 πe
Therefore
 
√ 20 √
ψ(x) − θ(x) ≥ x + 1 − 2 x1/3 ≥ x.
πe
(b) To derive the stated inequality, consider three ranges for values of x.
(1) First let 1 ≤ x < 232 . The largest qi smaller than 232 is q6947 = 65 5212 . On
each of the intervals [qi , qi+1 ) in this range we find explicitly that the function
√
ψ(x) − θ(x) − x
G(x) :=
x1/3
is decreasing and that G(x) ≤ G(q103 ) < 1.3330 < 4/3.
(2) Now let 232 ≤ x < 64 × 1022 . Here we use [48, table 6.6] which gives
θ(x) < x for x ≤ 8 × 1011 (compare Lemma 3.10). Then, if as before we set
κ := log x/log 2, we get

κ 
κ
ψ(x) − θ(x) = θ(x1/n ) ≤ x1/n
n=2 n=2

so by Lemma 5.17
4
G(x) ≤ H(x) ≤ H(232 ) < 1.32 < .
3
(3) Finally let x ≥ 64 × 1022 ≥ 279 . Use θ(x) ≤ x + T (x) to write

κ 
κ
ψ(x) − θ(x) = θ(x1/n ) ≤ (x1/n + T (x1/n )),
n=2 n=2
 5.3 Preliminary Results With RH True 119

so using the notation and result of Lemma 5.17 we get

√
ψ(x) − θ(x) − x
G(x) =
x1/3
 κ  
T (x1/n )
≤ x 1/n−1/3
+ 1/3 − x1/2−1/3
n=2
x
κ κ
T (x1/n )
= 1+ x1/n−1/3 +
n=4 n=2
x1/3
= H(x) + L(x)
4
≤ H(279 ) + L(279 ) < 1.33 < .
3
This completes the proof of the lemma. 
The reader might wish to compare the lower bound of Lemma 5.18 with
the upper bound from Theorem 4.6, both of which assume RH:
√
√ x log2 x
x ≤ ψ(x) − θ(x) ≤ , x ≥ 599,
4π
or even, without assuming RH,
√
ψ(x) − θ(x) < 1.021 x, x > 0,
which is Lemma 3.13 (c).
Next recall the definitions R(x) := ψ(x) − x and S (x) = θ(x) − x and the
related integrals:
 ∞  
R(t) 1 1
J(x) := + dt,
x t2 log t log2 t
 ∞  
S (t) 1 1
K(x) := + dt, (5.19)
x t 2 log t log2 t
 ∞ z 
t 1 1
Fz (x) := + dt.
x t
2 log t
log2 t
Substituting the bounds for ψ(x) − θ(x) from Lemma 5.18 in the definitions of
J(x) and K(x), the following bounds for J(x) − K(x) can be derived directly:
Lemma 5.19 [127, corollary 2.1] Assume RH is true. Then for x ≥ 121 we
get
F1/2 (x) ≤ J(x) − K(x) ≤ F1/2 (x) + 43 F1/3 (x).
This lemma enables us to estimate K(x) by first estimating J(x), a task
which begins with Lemma 5.20 below, and which employs zeta zeros.
120 Euler’s Totient Function

To plot the graph of F1/2 (x) in Figure 5.6, we used the representation
 
2 log x
F1/2 (x) = √ − Ei − .
x log x 2

See the RHpack function FHalf in Appendix B.

We now express J(x) in terms of the values of Fρ (x) with an error which is
small for large x.

Lemma 5.20 [126, proposition 2] Assume the Riemann hypothesis is true.

There is a function J1 (x) such that
1
J(x) = − Fρ (x) − J1 (x),
ρ
ρ

where J1 (x) satisfies 0 ≤ J1 (x) ≤ log(2π)/(x log x), and where the sum over ρ
is over all of the imaginary zeros of ζ(s) with increasing absolute value of the
imaginary part.

Proof Firstly, there is a real positive number A such that

 1
A := .
ρ
|ρ(ρ + 1)|

(The reader can explore properties of this and similar sums in [51].)

F(1/2)(x)
0.10

0.08

0.06

0.04

0.02
x
200 400 600 800 1000

Figure 5.6 The function F1/2 (x).

 5.3 Preliminary Results With RH True 121

Next define g(t) := (1/log t + 1/log2 t)/t2 , t > 1, so

 
 1 2 3 2
g (t) = − 3 + + .
t log t log2 t log3 t

The series ρ g (t)tρ+1 /(ρ(ρ + 1)) is absolutely convergent for all t > 1, and by
the Riemann hypothesis the partial sums are all bounded above by A|g (t)|t3/2 ,
for some constant A > 0. For x > 1 this bound is integrable on [x, ∞). By the
Weierstrass M-test we can integrate term by term and write
 ∞ ⎛⎜ ⎞
 ∞
⎜⎜⎜ tρ+1 ⎟⎟⎟⎟ tρ+1
⎜⎜⎝ g (t)

⎟⎟⎠ dt = g (t) dt. (5.20)
x ρ
ρ(ρ + 1) ρ x ρ(ρ + 1)

Next, in the explicit formula of von Mangoldt’s theorem, Theorem 2.2,

namely,
 x
x2  xρ+1 
∞
x−2n+1 ζ  (0) ζ  (−1)
ψ1 (x) := ψ(t) dt = − − − x+ ,
0 2 ρ
ρ(ρ + 1) n=1 2n(2n − 1) ζ(0) ζ(−1)

which is valid for all x > 1, set

ζ  (0) ζ  (−1)  x1−2n
∞
g1 (x) := x − + .
ζ(0) ζ(−1) n=1 2n(2n − 1)
For x > 1 using the value for ζ  (0)/ζ(0) derived in Lemma 2.5, i.e.
ζ  (0)/ζ(0) = log(2π), we get
 
 1 1
g1 (x) = log(2π) + log 1 − 2 ,
2 x
and so for x ≥ 2 we have
0 ≤ g1 (x) < log(2π). (5.21)
Now consider the left-hand side of (5.20). First use integration by parts,
noting that ψ1 (t) = O(t2 ) and setting
 ∞
J1 (x) := g(t)g1 (t) dt, (5.22)
x

to obtain
 ∞   ∞
 tρ+1
g (t) dt = g (t)( 12 t2 − ψ1 (t) − g1 (t)) dt
x ρ
ρ(ρ + 1) x

= g(x)(ψ1 (x) − 12 x2 + g1 (x)) + J(x) + J1 (x),

where J(x) has been defined as
 ∞  ∞  
ψ(t) − t 1 1
(ψ(t) − t)g(t) dt = + dt.
x x t2 log t log2 t
122 Euler’s Totient Function

The right-hand side of (5.20), using integration by parts again, is

 xρ+1
 ∞
tρ
  Fρ (x)
−g(x) − g(t) dt = g(x)(ψ1 (x) − 12 x2 + g1 (x)) − .
ρ
ρ(ρ + 1) x ρ ρ
ρ

These derivations show that (5.20) can be written

 Fρ (x)
J(x) = − − J1 (x).
ρ
ρ

By (5.21) and (5.22) for x ≥ 2 we have

 ∞  
−1 log(2π)
0 ≤ J1 (x) ≤ log(2π) d = .
x t log t x log x
This completes the proof of the lemma. 
Lemma 5.21 [126, proposition 2] For x ≥ 121 we have
1
log f (x) ≤ J(x) − 0.98F1/2 (x) + (5.23)
2(x − 1)
and for x ≥ 3000 we have
0.9 1
√ ≤ 0.98F1/2 (x) − . (5.24)
x log x 2(x − 1)
√
Proof By Lemma 3.13 (d), √ for x ≥ 121 we have θ(x) ≤ ψ(x) − 0.98 x. This
shows S (x) ≤ R(x) − 0.98 x so by Lemma 5.15, in the range x ≥ 121, we
have
1
log f (x) ≤ J(x) − 0.98F1/2 (x) + .
2(x − 1)
Next by Lemma 5.16 we have
2
F1/2 (x) = √ + r1/2 (x),
x log x
√
and√for x ≥ 3000 > e8 we have |r1/2 (x)| ≤ 1/( √x log x). Therefore F1/2 (x) ≥
1/( x log x) and thus, since 1/(2x − 2) < 0.08/( x log x),
0.9 1
√ ≤ 0.98F1/2 (x) − ,
x log x 2(x − 1)
which completes the proof. 
Now we are able to bound the sum in the statement of Lemma 5.20 which
depends on non-trivial zeta zeros:
 5.4 Further Results With RH True 123

Lemma 5.22 [126, proposition 2] Assume RH so ρ = 12 . For x ≥ e4 we have

 
 Fρ (x)  0.08
  ≤ √ .
 ρ ρ  x log x
Proof By Lemma 5.16 we have
 Fρ (x)  xρ−1  rρ (x)
− = −
ρ
ρ ρ
ρ(ρ − 1) log x ρ
ρ
1  xiρ  rρ (x)
= √ − .
x log x ρ ρ(ρ − 1) ρ
ρ

The first series is convergent. Note that ρ(1 − ρ) > 0 and recall that by
Lemma 2.10,
 1
β := = 2 + γ − log(4π) ≤ 0.047.
ρ
ρ(1 − ρ)

Then using the estimate for rρ (x) given by inequality (5.12) of Lemma 5.16
we get for x ≥ e4
 
 Fρ (x)  0.047 × 1.5 0.08
  ≤ √ ≤ √ ,
 ρ ρ  x log x x log x
which completes the proof of the lemma. 

5.4 Further Results With RH True

Now we bound J(x) in terms of a function W(x), which will play a central
role in the proof of Nicolas’ second theorem. Define
 xiρ
W(x) := ,
ρ
ρ(1 − ρ)

and recall the definitions R(t) := ψ(t) − t and

 ∞  
R(t) 1 1
J(x) := + dt.
x t2 log t log2 t
Lemma 5.23 [127, lemma 2.5] Assume RH is true. For x > 1 we have
W(x)
J(x) = − √ − J1 (x) − J2 (x),
x log x
where J1 (x) and J2 (x) are functions which satisfy
 
log(2π) β 4
0 < J1 (x) ≤ and |J2 (x)| ≤ √ 1+ .
x log x x log2 x log x
124 Euler’s Totient Function

Proof First, by Lemma 5.20, for x ≥ 2 we have

1
J(x) = − Fρ (x) − J1 (x),
ρ
ρ

where J1 (x) is a function which satisfies 0 ≤ J1 (x) ≤ log(2π)/(x log x) – so is

positive and very small for large x.
ρ−1
Then, by (5.11) of Lemma 5.16  we can write Fρ (x) = x /((1 − ρ) log x) +
rρ (x), so if we define J2 (x) := ρ rρ (x)/ρ the result follows using the bound
for |rρ (x)| which is given by the estimate (5.12). 
In the next step, which provides the fundamental lemma for this chapter,
we bound f (x) in terms of W(x):
Lemma 5.24 Assume RH is true and let x ≥ 109 . Then
2 + W(x) 0.055 2 + W(x) 2.062
− √ + √ 2
≤ log f (x) ≤ − √ + √ (5.25)
x log x x log x x log x x log2 x
and
2 + W(x) 2.062 1 2 + W(x) 0.055
√ − √ 2
≤ −1 ≤ √ − √ . (5.26)
x log x x log x f (x) x log x x log2 x
Proof Let x ≥ 599. Use (5.8), (5.13), (5.14) and (5.18) and Lemmas 5.19
and 5.23 to derive the lower bound
4 log3 x
log f (x) ≥ J(x) − F1/2 (x) − F1/3 (x) −
3 64π2 x
W(x) 4 log3 x
=− √ − J1 (x) − J2 (x) − F1/2 (x) − F1/3 (x) −
x log x 3 64π2 x
W(x) + 2 2−β 8 + 4β log(2π)
≥− √ + √ 2
− √ 3
−
x log x x log x x log x x log x
2 log3 x
− −
x2/3 log x 64π2 x
and similarly the upper bound
W(x) + 2 2+β 4β 1
log f (x) ≤ − √ + √ + √ + . (5.27)
x log x 2
x log x x log x 2(x − 1)
3

Now let x0 = 109 . Because x ≥ x0 these bounds imply


W(x) + 2 1 8 + 4β
log f (x) ≥ − √ + √ 2
2−β−
x log x x log x log x0
⎞
log(2π) log x0 2 log x0 log5 x0 ⎟⎟⎟
− √ − 1/6 − √ ⎟⎠
x0 x0 64π2 x0
 5.4 Further Results With RH True 125

and
 √ 
W(x) + 2 1 4β x0 log2 x0
log f (x) ≤ − √ + √ 2 + β + + ,
x log x x log2 x log x0 2(x0 − 1)
from which (5.25) follows on substituting for √ x0 .
Now let v := −log f (x) and v0 := (2 + β)/( x0 log x0 ). For all real v we have
ev − 1 ≥ v so, because
1
− 1 = ev − 1 ≥ v = −log f (x),
f (x)
the upper bound of (5.25) implies the lower bound of (5.26). Then by (5.5)
and (5.25) we can write
W(x) + 2 2+β
v≤ √ ≤ √ ≤ v0 < 3.13 × 10−6 .
x log x x log x
Then, to obtain the upper bound notice that
ev0 v2 ev0 (2 + β)2 ev0 (2 + β)2 6.63 × 10−5
ev − 1 − v ≤ ≤ ≤ √ √ < √ .
2 2x log2 x 2 x0 x log2 x x log2 x
Then
1 6.63 × 10−5 6.63 × 10−5
− 1 = ev − 1 ≤ v + √ ≤ −log f (x) + √ .
f (x) x log2 x x log2 x
This completes the proof. 
Now we need three estimates, first given by Rosser and Schoenfeld and
proved here in Lemmas 3.14, 5.12 and 3.13, respectively:
(1) For 563 ≤ x we have x(1 − 1/(2 log x)) < θ(x) and for 1 < x we have
θ(x) < x(1 + 1/(2 log x)).
(2) For 286 ≤ x we have
 p  
γ 1
< e log x 1 + .
p≤x
p−1 2 log2 x
√
(3) For 1423 ≤ x ≤ 108 we have x − 2 x < θ(x).
The next theorem should be compared with an analogous upper bound for
the ratio σ(n)/n given in Chapter 7, which is also unconditional. It might
also be compared with the unconditional lower bound given in Chapter 6,
Lemma 6.8, where n > 1 is restricted to being superabundant, i.e. for all 1 ≤
m < n we have
σ(m) σ(n)
< .
m n
To show how it leads on to other results, we have a network flow diagram,
Figure 5.7. The proof is in six steps.
126 Euler’s Totient Function

Lem 5.9
Thm 5.29

Lem 5.27

Lem 5.12
Thm 5.34 Cor 5.35

Lem 3.14 Lem 5.26

Thm 5.25

Lem 5.33

Lem 3.13

Lem 5.32

Figure 5.7 Some relationships between results in Chapter 5.

Theorem 5.25 [145, theorem 15] For all n ≥ 3, except n = N9 = 2 · 3 · 5 · 7 ·

11 · 13 · 17 · 19 · 23 we have (with or without RH)
n 5
< eγ loglog n + =: g(n).
ϕ(n) 2 loglog n
Proof (1) As usual let p j be the jth prime with p1 = 2 and suppose that the
integers n and m are such that n < p1 · · · pm+1 . Let q j for 1 ≤ j ≤ r be all of the
distinct primes dividing n ordered so that q1 < q2 < · · · < qr . Thus for all j we
have p j ≤ q j . Then
p1 · · · pr ≤ q1 · · · qr ≤ n < p1 · · · pm+1 ,
so we must have r ≤ m. Therefore
n  r
qj  r
pj 
m
pj Nm
= ≤ ≤ = .
ϕ(n) j=1 q j − 1 j=1 p j − 1 j=1 p j − 1 ϕ(Nm )

(2) Next we verify the theorem for log n ≤ 294: indeed, if Nm is the mth
primorial, and g(n) the right-hand side of the inequality of this theorem,
which is increasing for n in this range, then by Step (1), if Nm ≤ n < Nm+1 ,
and we have verified the result of the theorem at each primorial in the range
9 < m ≤ M, then for these values of m we would have
n  p
≤ < g(Nm ) ≤ g(n).
ϕ(n) p≤p p − 1
m

Now checking the theorem at primorials is easily accomplished up to M = 106

giving at least log n ≤ 1.5 × 107 .
 5.4 Further Results With RH True 127

(3) Now let x ≥ 5 and suppose log n < θ(x). Choose m so pm+1 ≤ x < pm+2 .
If pm+1 was 2 or 3 then pm+2 would be 3 or 5 so we would have the bound
x < 5, which is false. Hence 5 ≤ pm+1 , and thus pm ≤ pm+1 − 2 ≤ x − 2. By Step
(1) we therefore have
n  p
≤ .
ϕ(n) p≤x−2 p − 1
(4) The next step is quite clever. Let n > 1 be an integer and the real number
y ≥ 2 be chosen to satisfy three conditions:
288 ≤ log n + y,
log n < θ(log n + y),
0.9 log n
y−2 ≤ .
loglog n
Then, by Lemma 5.12, replacing x by log n + y and using Step (3), since
log n < θ(log n + y) and (y − 2)/log n ≤ 0.9/loglog n we get
n 0.5
< log(log n + y − 2) +
eγ ϕ(n) log(log n + y − 2)
 
y−2 0.5
≤ loglog n + log 1 + +
log n loglog n
1.4
≤ loglog n + .
loglog n
Therefore
n 5
< eγ loglog n + .
ϕ(n) 2 loglog n

(5) Now assume 255 ≤ log n ≤ 1340 and set y = 2 + 2 1 + log n. This
implies 288 ≤ log n + y ≤ 1420.
Therefore using [145, theorem 19] (or compare with Lemma √ 3.13(a)),
which in particular shows that for 0 < x ≤ 1420.9 we have x − 2 x < θ(x), we
get 
log n = log n + y − 2 log n + y < θ(log n + y).
The other two conditions in Step (4) are readily found to hold, so in the given
range the theorem follows by the result of Step (4).
(6) Finally assume 1340 ≤ log n. Choose y as 0.9 log n/loglog n and use
Lemma 3.14(e). With these choices for log n and y each of the hypotheses
listed in (4) are easily checked. This completes the proof. 
Recall Nk := p1 · p2 · · · pk is the kth primorial, with pi the ith prime and
p1 = 2. Also recall the definition given in the introduction to this chapter:
 
n γ

c(n) := − e loglog n log n. (5.28)
ϕ(n)
128 Euler’s Totient Function

The following lemma shows that c(n) is very well behaved. In particular it
has a local maximum on the right-hand side at every primorial, and that if
ω(n) is small so is c(n).
Lemma 5.26 [127, lemma 3.1] Let n ≥ 2 and k ≥ 1 with n, k ∈ N. Fix k and
let n be such that either k = ω(n) or Nk ≤ n < Nk+1 . Then in each of these cases
we have c(n) ≤ c(Nk ).
Proof If k = ω(n) then Nk ≤ n and ω(n) ≤ k. If ω(n) > k then n ≥ Nk+1 , so
Nk ≤ n < Nk+1 implies Nk ≤ n and ω(n) ≤ k also.
α
Write n = Pα1 1 · · · P j j where P1 < P2 < · · · < P j are primes and αi ≥ 1 integers.
Then each Pi ≥ pi , where pi is the ith prime, so

n  j
1  1
j  1
k
Nk
= ≤ ≤ =
ϕ(n) i=1 1 − 1/Pi i=1 1 − 1/pi i=1 1 − 1/pi ϕ(Nk )

and therefore
 
Nk γ

c(n) ≤ − e loglog n log n =: h(n), (5.29)
ϕ(Nk )
where for fixed k the expression on the right-hand side of (5.29) defines a
function h(n). Interpolate h on [Nk , Nk+1 ) with a differentiable real function
h(x) so
 
 1 Nk γ γ
h (n) =  − e loglog n − 2e
2n log n ϕ(Nk )
 
1 Nk γ γ
≤  − e loglog Nk − 2e .
2n log n ϕ(Nk )
We claim the term in parentheses in the upper bound is negative. For k = 1, 2
this can be tested numerically. For k ≥ 3 it is also negative because N3 = 30,
so loglog Nk ≥ loglog 30 and we can use the bound given in Theorem 5.25,
namely for all n ≥ 3, without any special assumptions,
n 2.50637
≤ eγ loglog n + ,
ϕ(n) loglog n
where the slightly increased numerator 2.50637 > 2.5 has been introduced to
accommodate the exception N9 . Therefore h(n) ≤ h(Nk ) = c(Nk ) so by (5.29),
c(n) ≤ c(Nk ) as claimed. 
In Figure 5.8 there is plotted c(N3 ) and c(n) for 30 = N3 ≤ n < N4 = 210.
For x ≥ 1 recall the definition
 xiρ
W(x) = ,
ρ
ρ(1 − ρ)
 5.4 Further Results With RH True 129
c(n)

n
0 50 100 150 200

–2

–4

Figure 5.8 The relation c(n) ≤ c(Nk ) for k = 3.

which for real x is real, since the zeros of ζ(s) occur in complex conjugate
pairs.
Lemma 5.27 [127, proposition 3.1] Assume RH is true and that 109 ≤ pk ≤
x < pk+1 where pi is the ith prime. Then for c(Nk ) we have the upper bound
0.07 0.07
c(Nk ) ≤ eγ (2 + W(x)) − ≤ eγ (2 + β) −
log x log x
and the lower bound
3.7 3.7
eγ (2 − β) − ≤ eγ (2 + W(x)) − ≤ c(Nk ).
log x log x
Proof Firstly, the definitions of f (x) and c(n) ((5.5) and (5.28) respectively)
and the identity log Nk = θ(pk ) = θ(x) imply
 
 1
c(Nk ) = eγ θ(x) log θ(x) −1 . (5.30)
f (x)
Equation (5.18) and the bound θ(x) ≥ 4x/5, x ≥ 121, from Lemma 3.10(d),
give
 
  √   θ(x) 2 + log t 

 θ(x) log θ(x) − x log x =  √ dt
 x 2 t 
 
log(4x/5) + 2
≤ |θ(x) − x| √
2 4x/5
130 Euler’s Totient Function
√  
5 2 + log x
≤ T (x) √
4 x
√
5
= log2 x(log x + 2).
32π
Therefore, using x0 = 109 again:
 √  √
 θ(x) log θ(x)  5 log2 x (2 + log x)
 √ − 1  ≤ √
x log x 32π x log x
√
5 log2 x0 (2 + log x0 ) 6.9 × 10−3
≤ √ ≤ .
32π x0 log x log x
Now use Lemma 5.24 and (5.30), (5.26) and (5.5), recall that by
Lemma 2.10
 1
β := = 2 + γ − log(4π),
ρ
ρ(1 − ρ)

and observe |W(x)| ≤ β, to derive

  
0.055 0.0069
c(Nk ) ≤ eγ 2 + W(x) − 1+
log x log x
γ
e
≤ eγ (2 + W(x)) − (0.055 − 0.0069(2 + β))
log x
0.07
≤ eγ (2 + W(x)) − ,
log x
to complete the proof of the upper bound. To derive the lower bound, use the
lower bound from (5.26) of Lemma 5.24. 

5.5 Preliminary Results With RH False

In this section we assume RH is false and derive, in Theorem 5.29, a conse-
quence for the asymptotic values of log f (x). Logarithmic differentiation of
the Euler product representation for ζ(s) leads to the following lemma:
Lemma 5.28 For s > 1,
 ∞
ζ  (s) ψ(t)
− =s dt.
ζ(s)2 t s+1

Proof Since for s > 1 we have ζ(s) = p (1 − 1/p)−1 ,
   
1 1
log ζ(s) = − log 1 − s = sm
.
p
p p m≥1
mp
 5.5 Preliminary Results With RH False 131

Therefore, using integration by parts and uniform convergence for 1 <

σ0 ≤ s:
ζ  (s)   log p  Λ(n)
− = =
ζ(s) p m≥1
p sm n≥2
ns
 x  x  ∞
1 ψ(t) ψ(t)
= lim dψ(t) = lim s dt = s dt,
x→∞ 1 t s x→∞ 1 t
s+1
2 t s+1
which completes the derivation. 
Recall the definition: if g(x) is a real-valued function defined on some
domain (x0 , ∞) and h(x) is a positive real-valued function on the same
domain, we say g(x) = Ω+ (h(x)) as x → ∞ if there is a constant M > 0
and a real sequence (xi )i∈N with xi → ∞ such that g(xi ) ≥ Mh(xi ). We say
g(x) = Ω− (h(x)) if −g(x) = Ω+ (h(x)). Finally g(x) = Ω± (h(x)) means both of
these properties are true for g(x) with respect to h(x).
To prove the following theorem of Nicolas, the method developed by
Landau to obtain “Ω-theorems” is employed; see Theorem 4.12. (The reader
might compare [80, chapter V].) Also recall the definitions:
 1

γ
f (x) := e log θ(x) 1− ,
p≤x
p

ψ(x) := log p,
pn ≤x
R(x) := ψ(x) − x,
 ∞  
R(t) 1 1
J(x) := + dt.
x t2 log t log2 t
The following theorem provides the key element for Nicolas’ theorems,
Theorems 5.31 and 5.34, when RH is false, and, looking ahead, also for the
proof of Robin’s theorem, Theorem 7.11, in Chapter 7.

Theorem 5.29 If the Riemann hypothesis is false then there exists a real b
with 0 < b < 1/2 such that, as x → ∞,
 
1
log f (x) = Ω± b .
x
Proof The proof is in seven steps.
(1) First apply Theorem 4.12 with h(x) = J(x) for x ≥ 2 and h(x) = 0 for
1 ≤ x ≤ 2. For s > 1 set
 ∞  ∞  ∞   
J(x) 1 R(t) 1 1
G(s) := dx = + dt dx.
2 xs 2 x s x t2 log t log2 t
132 Euler’s Totient Function

The prime number theorem gives |R(t)| = O(t/log t). Hence, the double
integral is absolutely convergent. Therefore we can use Fubini’s theorem to
change the order of integration and write
 ∞   t 
R(t) 1 1 dx
G(s) = 2
+ 2
dt,
2 t log t log t 2 x s

which gives
  ∞   
1 R(t) 1 1
G(s) = 2 J(2) −
1−s
+ dt . (5.31)
s−1 2 t1+s log t log2 t

Note that the term on the right-hand side inside the outer parentheses is
holomorphic for s > 1 and vanishes at s = 1.
(2) Next, for s > 1 we can write
 ∞  ∞  ∞  2
R(t) ψ(t) dt dt
s+1
dt = s+1
dt − s
+ s
.
2 t 2 t 1 t 1 t

The third integral on the right-hand side is an entire function (1 − 21−s )/(s − 1)
which we call E1 (s).
Therefore for s > 1 we have, by Lemma 5.28,
 ∞
R(t) 1 ζ  (s) 1
dt = − − + E1 (s). (5.32)
2 t s+1 s ζ(s) s − 1
We know from the relationship [6, theorem 13.11], applied with 0 < s < 1,
∞    
(−1)n+1 1 1 1
(1 − 2 )ζ(s) =
1−s
= 1 − s + s − s + · · · > 0,
n=1
ns 2 3 4

that, since 1 − 21−s < 0 and ζ(0) = − 12 , ζ(s) has no zero in [0, ∞), so therefore
there is a real number δ > 0 such that there are no zeros of ζ(s) in a simply
connected set

Θ := {s : s > 1} ∪ {s : 0 < s ≤ 1, |s| < δ}.

Calculations of zeta zeros show we can take δ = 14.

Equation (5.32) also shows that if σ0 is the abscissa of convergence of the
integral on the left, then σ0 ≤ 1.
(3) Now note that ζ(s) = 1/(s − 1) + E0 (s), where E0 is entire, implies that
the combination of the first and second terms on the right-hand side of (5.32)
represents a function which is holomorphic in a neighbourhood of s = 1 in
Θ, and therefore has, on Θ, a holomorphic antiderivative G1 (s) and a second
antiderivative G2 (s). That is to say, there are complex numbers λ and μ such
 5.5 Preliminary Results With RH False 133

that
 ∞
R(t) 1
dt = −G1 (s) + λ,
t s+1 log t
 ∞2
R(t) 1
dt = G2 (s) + λs + μ.
2 t s+1 log2 t
By (5.31) we have
1
G(s) = (G1 (s) − G2 (s) + E2 (s)), (5.33)
s−1
where E2 (s) is entire. The expression in parentheses in (5.33) is a function
holomorphic on Θ which by Step (1) vanishes at s = 1. Therefore the right-
hand side of (5.33) is holomorphic on Θ.
In the neighbourhood of a zero ρ of ζ(s) of multiplicity m we have
 
 d 1 ζ  (s) 1 m
G1 (s) ∼ − ∼ ,
ds s ζ(s) ρ (s − ρ)2
1 ζ  (s) 1 m
G2 (s) ∼ − ∼− ,
s ζ(s) ρ s−ρ
so G1 (s) − G2 (s) would have a pole of order 2 at s = ρ and ρ ≥ 12 . This
shows that σ0 ≥ 12 so σ0 ∈ Θ and G(s) is holomorphic at σ0 .
If R(x) had the same sign for x sufficiently large, by Theorem 4.12, G(s)
would have a singularity at σ0 , but is holomorphic at that point. Therefore
R(x) changes sign infinitely often as x → ∞.
(4) Now let θ be a fixed real number satisfying 0 < θ < 1/2. Suppose there
exists a zero ρ of ζ(s) with ρ = β > 1 − b where b is a real number which
satisfies 1 − θ < b < 1/2. We will show that J(x) − x−b and J(x) + x−b do not
keep a constant sign as x → ∞. To see this first calculate the Mellin transform
 ∞
J(x) − x−b 1
dx = G(s) − + E3 (s), s > 1,
2 x s s − 1 +b
2
where E3 (s) := 1 x−b−s dx is an entire function. The second term extends to a
holomorphic function on Θb := Θ ∩ {s : s > 1 − b}.
If J(x) − x−b had a constant sign as x → ∞, by Theorem 4.12 the abscissa of
convergence of the Mellin transform would be less than or equal to 1 − b ≤ β.
This is impossible since, by Step (4), G(s) has a singularity ρ with real part
β. Then adopt the same procedure for J(x) + x−b to show that it too does not
ultimately have constant sign.
(5) Recall the definition
 ∞  
S (t) 1 1
K(x) = + t dt
x t2 log t log2
134 Euler’s Totient Function

where S (t) = θ(t) − t. By Lemma 3.13 (e), for x > 0 we have

√ √
ψ(x) − θ(x) < 1.42620 x < 3 x/2,
which gives directly, using θ(x) ≤ ψ(x) for the upper bound:
J(x) − 32 F1/2 (x) ≤ K(x) ≤ J(x).
The estimates we have for F1/2 (x) given in Lemma 5.16, together with the
sign changes exhibited by J(x) ± x−b , then show that K(x) = Ω± (x−b ) for all b
with 1 − θ < b < 1/2.
(6) We next study the function y(x) := K(x) − x−b . It is differentiable for all
x not equal to a prime number, and at such a point
 
 S (x) 1 1 b
y (x) = − 2 + + b+1 .
x log x log2 x x
This derivative can change sign at x = p, p a prime, and considering the
equation
(x − a)(log x + 1) + bx1−b log2 x = 0,
with a = S (x), which has only one root for a > 1 in [1, ∞), shows that y (x)
vanishes at most once in any interval (p, p ), where p and p are consecutive
primes with p ≥ 3. To see this note that the derivative of
bx1−b log2 x
x−a+
1 + log x
is
bx−b log x ((1 − b) log2 x + (2 − b) log x + 2)
1+ >0
(log x + 1)2
and use Rolle’s theorem.
Therefore the sign changes of y (x) form a discrete set in [1, ∞). Suppose
they constitute the increasing sequence (xi ). This sequence is infinite, since
we are assuming the Riemann hypothesis is false, and this implies S (x) =
Ω± (x(θ+1−b)/2 ); cf. [80, chapter V] or Theorem 4.12.
At a point xi where y (x) changes sign, so does
bx1−b log2 x
S (x) − .
1 + log x
This means that this quantity is either 0 or, in case xi is prime, lies between 0
and log xi . In all cases we have
 
S (x)2 log xi
S (xi ) = O(xi1−b log xi ) and thus = O .
xi2 log xi xi2b
 5.6 Nicolas’ First Theorem 135

(7) This is the final step. We claim there are an infinite number of values of
i with y(xi ) > 0. Indeed, if the function y(x) was negative or zero at all local
extrema, it would be negative everywhere, and this would mean, for all x,

K(x) ≤ x−b . But, as has been shown, K(x) = Ω+ (x−b ) where b := (1 + b − θ)/2.
At such a point xi we have
S (xi )2
K(xi ) − > xi−b + O(xi−2b log xi ),
xi2 log xi
giving, by Lemma 5.15, log f (x) = Ω+ (x−b ). To show that log f (x) = Ω− (x−b ),
also consider K(x)+ x−b and use the same approach. This completes the proof.


5.6 Nicolas’ First Theorem

Now that we have the key result of Section 5.5 we are able to prove Nicolas’
first theorem. First we derive a good upper bound for the logarithm of f (x)
provided x is sufficiently large.
Lemma 5.30 Assume RH is true. For x ≥ 3000 we have
0.8
log f (x) ≤ − √ .
x log x
Proof By Lemma 5.21 we have
1
log f (x) ≤ J(x) − 0.98F1/2 (x) + , x ≥ 121.
2(x − 1)
By the second part of Lemma 5.21, for x ≥ 3000 we have
1 0.9
−0.98F1/2 (x) + ≤− √ .
2(x − 1) x log x
By Lemmas 5.13, 5.20 and 5.22 we have
 
 Fρ  0.1
J(x) ≤   ≤ √ , x ≥ 55,
 ρ ρ  x log x
so using these inequalities it follows that
1
log f (x) ≤ J(x) − 0.98F1/2 (x) +
2(x − 1)
0.1 0.9
≤ √ − √
x log x x log x
0.8
=− √
x log x
which completes the proof. 
136 Euler’s Totient Function

Now Nicolas’ theorem is an easy consequence:

Theorem 5.31 (Nicolas first criterion) If the Riemann hypothesis is true
then for all x ≥ 2 we have f (x) < 1 and
Nk
eγ loglog Nk < , k ≥ 1.
ϕ(Nk )
If the Riemann hypothesis however is false then the given inequality is
true for an infinite number of k and false for an infinite number of k.
Proof If RH is true, then by Lemmas 5.13 and 5.30 for all x ≥ 2, f (x) < 1 so
we can substitute pk , the kth prime, for x and use log Nk = θ(pk ) to derive the
given inequality.
If RH is false then by Theorem 5.29 there is a real sequence (xk )k∈N with
xk → ∞ with log f (xk ) < 0 for all k ∈ N. Let p jk be the largest prime less than
or equal to xk . By taking a subsequence of the (xk ) we can ensure the p jk are
distinct. Then
ϕ(N jk )
1 > f (xk ) = f (p jk ) = eγ loglog N jk ,
N jk
so the given inequality is true for this infinite set of indices k. Similarly by
using xk such that log f (xk ) > 0 the same approach can be used to find an
infinite set of k such that the inequality is false. 
As pointed out in the introduction to this chapter, little appears to be known
about the solutions in general to the inequality
n
eγ loglog n <
ϕ(n)
where n is not necessarily a primorial. A limited amount of numerical
exploration appears to indicate that the proportion of solutions to this
Nicolas’ reversed inequality slowly declines, so the natural conjecture is that
they have density zero. See Figure 5.9 and Section A.6 in the Appendix.
We now briefly introduce an implication for when RH fails, so in this
last part of the section let RH be false. Then by Nicolas’ first theorem,
Theorem 5.31, there exists an infinite sequence of integers k → ∞ such that
Nk Nk+1
≤ eγ loglog Nk < eγ loglog Nk+1 < .
ϕ(Nk ) ϕ(Nk+1 )
These inequalities have some delicate consequences. We have the numeri-
cal example as shown in Figure 5.10, which is a plot of
   
Nk+1 Nk γ log Nk+1
Δk := − − e log .
ϕ(Nk ) ϕ(Nk ) log Nk
 5.7 Nicolas’ Second Theorem 137
count
0.20

0.15

0.10

0.05

n
0 20 000 40 000 60 000 80 000 100 000

Figure 5.9 Proportion of solutions to eγ loglog n < n/ϕ(n) for 1 ≤ n ≤ 105 .

Δk
k
1000 2000 3000 4000

–0.00002

–0.00004

–0.00006

–0.00008

–0.0001

Figure 5.10 The function Δk for 1 ≤ k ≤ 4000.

5.7 Nicolas’ Second Theorem

Primorials also play an important role in Nicolas’ second theorem. All the
hard work has been done, especially in Lemma 5.24. Corollary 5.35 shows
that the theorem is close in content to Robin’s theorem and Nicolas’ first
theorem. Let, as before, Nk = 2 · 3 · 5 · · · pk be the kth primorial. Recall the
138 Euler’s Totient Function

definitions:
 
n γ

c(n) := − e loglog(n) log n,
ϕ(n)
 xiρ
W(x) := ,
ρ
ρ(1 − ρ)
 1
β := = 2 + γ − log(4π) = 0.046191 . . . ,
ρ
ρ(1 − ρ)
 1

γ
f (x) := e log θ(x) 1− .
p≤x
p

For a plot of values of the sequence c(n) for n ∈ [1, 1000], see Figure 5.11.
Next we give a form of Dirichlet’s approximation theorem [5, section 7.2],
which we need to show the supremum of W(x) is β:
Lemma 5.32 Given  > 0, fixed real numbers α1 , . . . , αN and a real t0 > 0,
then there exist integers x1 , . . . , xN and real t ≥ t0 such that
|tα j − x j | < , 1 ≤ j ≤ N.
Proof Let q ∈ N satisfy 1/q <  and set a = (α1 , . . . , aN ) ∈ RN . Let the real
variable u take the 1 + qN values
0, t0 , 2t0 , . . . , qN t0 .

c(n)

n
200 400 600 800 1000

–2

–4

–6

Figure 5.11 The sequence c(n).

 5.7 Nicolas’ Second Theorem 139

Then, if as usual we denote the fractional part of the real variable x by {x}, the
1 + qN points ({uα1 }, . . . , {uαN }) ∈ [0, 1]N . This cube can be decomposed into
exactly qN regular subcubes each having all sides of length 1/q, so at least
two of these points are in a particular subcube. If the corresponding values
of u are u1 and u2 with u1 < u2 , then for 1 ≤ j ≤ N we have integers n j and m j
such that
1 1
|{u2 α j } − {u1 α j }| ≤ =⇒ |u2 α j − n j − u1 α j + m j | ≤
q q
1
=⇒ |(u2 − u1 )α j − (n j − m j )| ≤
q
=⇒ |tα j − x j | ≤ .
Finally observe that each x j is an integer and t = u2 − u1 ≥ t0 to complete the
proof. 
Lemma 5.33 Assume RH is true. Then
lim sup W(x) = β.
x→∞

Proof Let  > 0 and t0 > 0 be given. Choose N sufficiently large so that
N
2
> β − ,
j=1
|ρ j |2

and note that we can write

 2 cos(γ log x)
W(x) = .
γ>0
|ρ|2

Let α j := γ j /(2π) where ρ1 = 12 +iγ1 is the first complex zeta zero with positive
imaginary part. By Lemma 5.32 there is a t ≥ t0 such that for 1 ≤ j ≤ N we
have |tα j − x j | < /(2π). This enables us to write |tγ j − 2πx j | <  which implies
cos(tγ j ) ≥ cos() ≥ 1 − . Therefore letting x = et we have

N
2 cos(γ j log x)  1
W(x) ≥ −
j=1
|ρ j |2 |ρ|>|ρ |
|ρ|2
N


N
2(1 − )
≥ −
j=1
|ρ j |2
≥ (1 − )(β − ) − .
Since x may be chosen arbitrarily large and W(x) ≤ β this shows that
lim supx→∞ W(x) = β. 
140 Euler’s Totient Function

Theorem 5.34 The hypothesis RH is equivalent to each of the following:

(1) lim supn→∞ c(n) = eγ (2 + β) = 3.644415 . . . .
(2) For all n ≥ N120569 = 2 · 3 · · · 1591 883, we have c(n) < eγ (2 + β).
(3) For all n ≥ 2, c(n) ≤ c(N66 ) = c(2 · 3 · · · 317) = 4.0628356921 . . . .
(4) For all k ≥ 1, we have c(Nk ) ≥ c(N1 ) = c(2) = 2.208589 . . . .
Proof First assume RH is true.
(1) By Lemmas 5.33, 5.26 and 5.27 it follows that
eγ (2 + β) = lim sup eγ (2 + W(x)) ≤ lim sup c(Nk )
x→∞ k→∞
≤ lim sup c(n) ≤ lim sup c(Nk ) ≤ eγ (2 + β),
n→∞ k→∞

so (1) follows.
(2) Let, as before, x0 = 109 and k0 = π(x0 ) = 50 847 534. Setting k1 :=
120 568 and testing the equation numerically shows that for k1 < k ≤ k0 we
have c(Nk ) < eγ (2 + β). To do this we used a Mathematica procedure based on
the following pseudocode where γ is Euler’s constant and pk is the kth prime:
Set k0 , k1 and the precision d
Evaluate eγ numerically
Set the number of failures to 0
Set c to eγ (4 + γ − log(4π)) numerically
Set rk to the numerical product of pk /(pk − 1) from k = 1 to k = k1 − 1
Set LogNk to the numerical sum of log(pk ) from k = 1 to k = k1 − 1
For k from k1 to k0 update rk and LogNk and then evaluate
√
cNk which is (rk − eγ log(LogNk)) LogNk
Test to see if cNk > c, report any failure, update the number of failures
Return the number of failures
See the RHpack function NicolasTwoFailures. Also see Figure 5.12 for a plot
of c(Nk ) values.
For k > k0 Lemma 5.27 gives c(Nk ) < eγ (2+β) so, by Lemma 5.26, choosing
k appropriately, c(n) ≤ c(Nk ) < eγ (2 + β) so (2) follows.
(3) A Mathematica computation similar to that in (2) shows that for 1 ≤
k ≤ k0 we have c(N1 ) ≤ c(Nk ) ≤ c(N66 ) and then use Lemma 5.27 to complete
the proof that c(n) ≤ c(N66 ) for all n ≥ 2.
(4) By Lemma 5.27, for all k ≥ k0 we get
3.7
c(Nk ) ≥ eγ (2 − β) − = 3.30 · · · > c(2),
log x0
so, using the lower bound from the computation in (3), (4) is true also.
 5.7 Nicolas’ Second Theorem 141

3.64

3.62

3.60

3.58

200 000 400 000 600 000 800 000

Figure 5.12 The sequence c(Nk ) for 120 568 ≤ k ≤ 106 .

Now assume RH is false. By the prime number theorem in the form θ(x) ∼ x
and using Mertens’ theorem (see Lemma 5.12) we have lim x→∞ f (x) = 1.
Observe that f (x) is locally constant, which goes to the essence of
its definition. If pk is the kth prime, when pk ≤ x < pk+1 , f (x) =
eγ loglog(Nk )ϕ(Nk )/Nk . Also, when k → ∞, because log(1+y)/y → 1 as y → 0,
if we set 1 + 1/y = f (pk ) we get:
 
1 1
log 1 + ∼ log f (pk ) ∼ = f (pk ) − 1,
y y
so log f (pk ) ∼ f (pk ) − 1. Therefore, since log Nk ∼ pk and by Theorem 5.25
and Lemma 5.12 we have
Nk
∼ eγ loglog Nk ,
ϕ(Nk )
we can derive

log f (pk ) ∼ eγ loglog(Nk )ϕ(Nk )/Nk − 1

ϕ(Nk )c(Nk )
=− 
Nk log Nk
c(Nk )
∼− 
γ
e (loglog Nk ) log Nk
e−γ c(Nk )
∼− √ .
(log pk ) pk
142 Euler’s Totient Function

Because RH fails, there is a real number b with 0 < b < 1/2 such that, by
Theorem 5.29, we have log f (x) = Ω± (1/xb ), so in this case, where f (x) is
constant between primes, we must have
e−γ c(Nk )
= Ω± (1),
(log pk )p1/2−b
k

so lim inf n→∞ c(n) = −∞ and lim supn→∞ c(n) = ∞. Therefore each of (1), (2),
(3) and (4) is false also. This completes the proof of the equivalences. 
Now consider the inequality derived in Theorem 5.25, namely for all n ≥ 3
except n = N9 we have
n 5
< eγ loglog n + .
ϕ(n) 2 loglog n
From Theorem 5.34 part (2), we see that RH is equivalent to a stronger
form of this inequality, at least for n sufficiently large. We call this Nicolas’
second criterion, something of a misnomer since his second theorem is more
comprehensive, but is given because of its simplicity and because it takes the
same shape as other criteria.
Corollary 5.35 (Nicolas’ second criterion) The Riemann hypothesis is
equivalent to the following inequality holding for all n ≥ N120569 :
n eγ (2 + β)
< eγ loglog n +  .
ϕ(n) log n
Note also from Chapter 6, Lemma 6.8, for n a sufficiently large
superabundant number with maximum prime divisor p, that, without RH,
n σ(n) n 0.6
(1 − (p)) < < < eγ loglog n + ,
ϕ(n) n ϕ(n) loglog n
where (p) := (1 + 3/(2 log p))/log p and provided n ≥ Nk with pk ≥ 20 000.
Remark A computation shows the inequality of Corollary 5.35 fails for
primorials Nk with k in the following ranges:
[9, 27 220], [28 168, 40 969], [41 368, 56 605],
[70 250, 82 904], [118 349, 120 568],
and succeeds for all other primorials with 1 ≤ k ≤ 120 568.

5.8 Unsolved Problems

(1) Let k1 = 120 568. Find the largest number M such that both inequalities
M < Nk1 +1 and c(M) ≥ eγ (2 + β) are true.
(2) Find all the values of n ∈ N satisfying n < Nk1 +1 and c(n) > eγ (2 + β).
 5.8 Unsolved Problems 143

(3) Explain the remark following the corollary to Theorem 5.34 in terms of
the distribution of primes.
(4) Find a relationship between gaps, or sets of gaps, between primes which
is equivalent to the Riemann hypothesis.
(5) Prove that RH implies eγ loglog n < n/(ϕ(n) + 1) for all n ≥ 2.
(6) Lehmer’s conjecture of 1932: if ϕ(n) | n − 1 then necessarily n is prime.
(7) Carmichael’s totient function conjecture: for all m, #{n : ϕ(n) = m} > 1.
(8) The Duffin–Schaeffer conjecture of 1941: if f : N → (0, ∞) then for
almost all α (with respect to Lebesgue measure), the inequality
 
α − p  < f (q)
 q q
has infinitely many solutions in coprime integers p, q with q > 0 if and
only if the sum

∞
ϕ(q)
f (q) = ∞.
q=1
q
 6
A Variety of Abundant Numbers

6.1 Introduction
In this chapter we present a study of superabundant and colossally abundant
numbers. The material is included because these are the numbers which
appear as possible counterexamples to the inequality of Robin, which we
will show in Chapter 7 to be equivalent to RH.
Paul Erdős (Figure 6.1) made fundamental contributions to our understand-
ing of these numbers, especially in two joint papers, one with Alaoglu in 1944
[3] and the other with Nicolas in 1975 [56]. Erdős was the most prolific and
connected mathematician of all time, in terms of both the number of articles
he published and the number of his joint authors. He never held a regular
position, but travelled the world, staying for a time with one or other of his
joint authors and discussing mathematical problems. These were in a wide
range: graph theory, combinatorics, analysis, probability and set theory, as
well as number theory.
All of the numbers of interest share the property that, apart from exceptions
like 1, they have many divisors, with many of the divisors being large. Here
now are some definitions. A positive integer n is said to be highly composite
if the number-of-divisors function d(n) satisfies, for 1 ≤ m < n, d(n) > d(m).
In other words, a highly composite number n has more divisors than any
smaller number. It is called abundant provided σ(n)/n > 2. We say it is
highly abundant if σ(n) > σ(m) for all m < n. It is superabundant if
σ(n)/n > σ(m)/m for all 1 ≤ m < n. Some authors denote these numbers as
“SA”. It is colossally abundant, sometimes denoted “CA”, if there exists an
 > 0, depending on n, such that for all m > 1 we have
σ(m) σ(n)
≤ 1+ .
m1+ n
There are relations between these types of numbers. For example all
superabundant numbers are highly abundant and all colossally abundant

144
 6.1 Introduction 145

Figure 6.1 Paul Erdős, 1913–1996.

numbers are superabundant. To see this, suppose that n > 1 were colossally
abundant but not superabundant. Then for some m < n we would have
σ(n)/n < σ(m)/m. But this would mean that for some  > 0 we would have

σ(n) σ(m) mσ(n)

≥ 1+ > ,
n1+ m nm1+
which gives 1 > (m/n) > 1, a contradiction. Hence CA ⊂ SA. Examples of
small highly composite, superabundant and colossally abundant numbers are
given in Appendix A.
There are a range of counting problems for these types of numbers, some
of which have been solved. For example, the set of abundant numbers has
an asymptotic density. This was proved independently by Chowla [39],
Erdős [52] and Davenport [42]; and see [124, p. 266]. For equivalences to
the Riemann hypothesis, it is the superabundant and colossally abundant
numbers which appear, as will be seen in Chapters 7, 8 and 9.
A note about the history of the sum-of-divisors function σ(n), or, more
generally, divisor sum functions σ s (n), where s is a complex number, is
in order: they have an ancient past, being known to the Greeks, who were
fascinated with perfect numbers. They appear in Euclid’s “Elements” VII.22.
In Dickson’s “History of the Theory of Numbers” volume I [46], there are
146 A Variety of Abundant Numbers

extensive references in chapters II and X, including references to works by

Cardano, Mersenne, Newton, Waring, Descartes and Fermat.
The definition of σ s (n) is as follows:
   p s(ν p (n)+1) − 1  
σ s (n) := d =
s
s−1
= (1 + p s + p2s + · · · + pν p (n)s ).
d|n p|n
p p|n

If s = 0 we write d(n) := σ0 (n) = p|n (1 + ν p (n)), and if s = 1, the principal
focus in this chapter, then
  p(ν p (n)+1) − 1 
σ(n) := σ1 (n) = .
p|n
p−1

Here we are interested in the ratios σ(n)/n and, for  > 0, σ(n)/n1+ , over
initial sequences of natural numbers. If Nk = 2 · 3 · · · pk is the kth primorial,
then

σ(Nk ) 
k
σ(n)
= (1 + p j ) =⇒ lim sup = ∞,
Nk j=1 n→∞ n

but since for each prime p, σ(p)/p = 1 + 1/p, we have lim inf n→∞ σ(n)/n = 1.
It can be shown that {σ(n)/n : n ∈ N} is, in the usual topology, dense in (1, ∞).
Even though the values of both σ(n) and σ(n)/n exhibit a high degree
of randomness as functions of n, they have very nice Dirichlet series
representations. For example

∞
σ(n)
= ζ(s)ζ(s − 1), s > 2,
n=1
ns

and

∞
σa (n)σb (n) ζ(s)ζ(s − a)ζ(s − b)ζ(s − a − b)
= , s > σ0 ,
n=1
ns ζ(2s − a − b)

where

σ0 = max{(1 + a), (1 + b), (1 + a + b)}.

For further background material and references for σ(n) the reader might
consult [6, 75, 150].
The following RHpack (see Appendix B) functions relate to the mate-
rial in this chapter: HighlyCompositeQ, HardyRamanujanQ, SuperAbun-
dantQ, ColossallyAbundantQ, ColossallyAbundantInteger, ColossallyAbun-
dantF and CriticalEpsilonValues.
 6.2 Superabundant Numbers 147

6.2 Superabundant Numbers

Recall that a positive integer n > 1 is said to be superabundant if for all
1 ≤ m < n we have
σ(m) σ(n)
< .
m n
In this section we use the following notation for unique factorizations of
integers greater than 1:

n = 2a2 3a3 · · · pa p ,

where, for each prime q, aq ≥ 0, and p is the maximum prime dividing n, so

all of the primes up to p are included in the expression, and “ · · · ” here means
all consecutive primes appear to the same power as the leading and trailing
term. In other words, in an explicit prime factorization, q j · · · p j means q < p
and all of the consecutive primes between q and p are to be included, each to
the power j.
The table in section A.3 gives the 31 smallest superabundant numbers.
The database of Kilminster and Noe includes thousands, with the largest
at this time being numbered 8436 in the Online Encyclopedia of Integer
Sequences [158]

https://oeis.org/A004394/b004394.txt

which is:

n8436 = 214 · 38 · 55 · 74 · 113 · · · 173 · 192 · · · 712 · 731 · · · 27191 .

When it comes to Riemann hypothesis equivalents, superabundant num-

bers are not as conspicuous as some other types of integer. However, this
may change when the numbers are better understood. To this end we have
included some interesting recent results of Nazardonyavi and Yakubovich
[122], as well as part of the classical material, now 70 years old, of Erdős
and Alaoglu.
First we show that in the prime factorization of any superabundant number
all primes up to p appear since their exponents are in non-increasing order.
Lemma 6.1 If n is superabundant then the exponents are non-increasing,
i.e. aq ≥ aq for all primes q, q with q ≤ q ≤ p.
Proof All primes up to p must appear, since if a prime was missing we could
replace a larger prime in n with the missing prime raised to the same power as
the larger prime, and thereby obtain an integer m < n with σ(n)/n < σ(m)/m.
This is because for primes p and q with p < q we have σ(pe )/pe > σ(qe )/qe .
If the exponents of the prime powers were not non-increasing, first note that
148 A Variety of Abundant Numbers

for given e > f > 0 and x ∈ (1, ∞) the function

1 − x−e−1
g(x) :=
1 − x− f −1
is strictly decreasing because the numerator of the derivative is ex f − f xe + f −
e and its derivative is 0 at x = 1 and is negative. Thus if p, q, f, e are positive
integers with p, q prime, p < q and e > f , then p f qe > pe q f , but because
g(p) > g(q) we have
  
σ(pe q f ) 1 − p−e−1 1 − q− f −1
=
pe q f 1 − p−1 1 − q−1
   
1 − p− f −1 1 − q−e−1 σ(p f qe )
> = ,
1 − p−1 1 − q−1 p f qe
contradicting n is superabundant. Therefore the prime powers are non-
increasing and the proof is complete. 
Lemma 6.2 [3, theorem 2] For n superabundant and q < r ≤ p primes we
have the inequality
  
 log q 
 ar − aq  ≤ 1.
log r 
Proof Let β := aq log q/log r and suppose first that ar ≤ β − 2. Define an
integer x by the inequalities q x−1 < r < q x . If aq < x then aq ≤ x − 1 so

qaq ≤ q x−1 < r ≤ rar ≤ rβ−2 ≤ qaq r−2 < qaq ,

which is false. Therefore we must have aq ≥ x. This implies nr/q x is an integer

which is less than n. Using the fact that n is superabundant, and cancelling
common factors, we can write
σ(qaq −x rar +1 ) σ(qaq rar )
< aq ar ,
qaq −x rar +1 q r
which implies

(r − 1)qaq +1 + q x < (q x − 1)rar +2 + r.

Note that q x−1 ≤ r − 1 so q x ≤ q(r − 1) giving q x − 1 < q(r − 1). Using this we
have

(q x − 1)rar +2 + r ≤ (q x − 1)rβ + q x < q(r − 1)rβ + q x

≤ (r − 1)qaq +1 + q x < (q x − 1)rar +2 + r

since rβ ≤ qaq . This contradiction shows ar ≥ β − 1.

 6.2 Superabundant Numbers 149

Now suppose that ar ≥ β + 2 and again define x by q x−1 < r < q x so x ≥ 2.

Then r > q x−1 which implies qaq rar > qaq +x−1 rar −1 . Using the fact that n is
superabundant, and cancelling common factors, gives
qaq +x (r − 1) + q x−1 > rar +1 (q x−1 − 1) + r > rar +1 (q x−1 − 1) + q x−1 .
This implies q x (r − 1) > r2 (q x−1 − 1). Since the function g(r) := r2 /(r − 1) is
increasing with r ≥ 2 , and h(x) := q x /(q x−1 − 1) is decreasing with x ≥ 2, it
follows that
r2 qx q2
< x−1 ≤ .
r−1 q −1 q−1
The outermost terms give r2 /(r − 1) < q2 /(q − 1), which is false since q < r
and g(r) is increasing. Therefore we must have also ar ≤ β + 1, and the proof
is complete. 
The example n8436 given above had a p = 1. With two exceptions, this is
always the case for superabundant numbers.
Lemma 6.3 The exponent of the maximum prime divisor p of a superabun-
dant number is a p = 1 except for n = 4 and n = 36 in which cases we have
a p = 2.
Proof Suppose, to get a contradiction, that a p > 1 and let q be the penultimate
prime less than p and r the next higher prime to p, so q < p < r. Then m :=
nr/(pq) < n and n is superabundant, so σ(n)/n > σ(m)/m. Because aq ≥ a p ≥ 2,
this implies
     
1 p−1 q−1 1 1
1 + < 1 + a p +1 1 + aq +1 ≤ 1+ 2 1+ 2
r p −p q −q p +p q +q
  
1 1
< 1+ 2 1+ 2 .
p q
But the order and rank of the three primes implies, by Bertrand’s hypothesis,
that
  
p 1 1 4
< q < p < r < 2p =⇒ 1 + < 1+ 2 1+ 2
2 2p p p
and therefore 2 ≤ p ≤ 7.
If (q, p, r) = (3, 5, 7) or (5, 7, 11) then
  
1 1 1
1+ > 1+ 2 1+ 2
r p +p q +q
so we must have p ∈ {2, 3}. If p = 2, since 22 is superabundant, we are left to
consider n = 2e with e ≥ 3 so n > 6. But
1 σ(2e ) σ(6)
2− e = e > = 2,
2 2 6
150 A Variety of Abundant Numbers

which is false. Hence 4 is the only superabundant number which is a

power of 2.
Finally if p = 3 first note that 2 · 3, 22 · 3 and 22 · 32 are superabundant. If
n = 2e · 33 with e ≥ 3 first note that 23 · 33 = 216 > 120 so
 
1 40 σ(2e · 33 ) σ(120)
2− e = e 3 <3= ,
2 27 2 ·3 120
so no number of the form 2e · 33 , e ≥ 3, can be superabundant. Therefore 4
and 36 are the only superabundant numbers with a p > 1. This completes the
proof. 
The prime power of 2 provides a convenient upper bound for prime power
divisors of superabundant numbers. Generalizing this result is a problem
included in the list at the end of this chapter.
Lemma 6.4 If n is superabundant then for all primes q > 2 with q | n we have
qaq < 4 · 2a2 .
Proof If aq ≤ β := a2 log 2/log q then qaq ≤ 2a2 < 4 · 2a2 , so we may assume
aq > β. With Lemma 6.2 this implies we must have aq = β + 1. To get a
contradiction, assume also that for some superabundant number n we have
2a2 +2 < qaq . Note that if we define an integer x ≥ 1 by 2 x < q < 2 x+1 then we
would have the unitary divisor of n, 2a2 qβ+1 = 2a2 qaq > 2a2 +x qβ . Using that n is
superabundant and cancelling gives
qβ+2 (2 x − 1) + q < 2a2 +x+1 (q − 1) + 2 x . (6.1)
Therefore
qaq +1 (2 x − 1) < 2a2 +x+1 (q − 1)
2a2 +2 q(2 x − 1) < q2a2 +x+1
2a2 +2 (2 x − 1) < 2a2 +x+1
2a2 +x+1 < 2a2 +2
2 x < 21 which is false.
This contradiction shows that qaq ≤ 2a2 +2 and completes the proof of the
lemma. 
It seems that superabundant numbers have many divisors. One way of
quantifying this is the result of Caveney:
Lemma 6.5 [34, proposition 9] Every given integer n greater than 1 divides
all but a finite number of superabundant numbers.
Proof Note that in this proof primes are indexed by their sequential rank so
2 = p1 , . . . . Let n > 1 be a given integer with factorization
n = pa11 · · · pamm
 6.2 Superabundant Numbers 151

with ai ≥ 0, am ≥ 1, with pi the ith prime. Let S := {1, 2, 4, 6, 24, . . .} be the set
of all superabundant numbers. Let A := max(ai : 1 ≤ i ≤ m) be the maximum
of the exponents of n so n | (p1 · · · pm )A . Let
F := {s ∈ S : pmA  s}.
Because of Lemma 6.1, if pmA | s we would have n | s. Hence the set of
superabundant numbers s for which n  s is a subset of those for which pmA
does not divide s, i.e. F.
We claim F is a finite set. To see this let s = pb11 · · · pl l be in F so bm < A.
b

By Lemma 6.2 we have

| bm − b1 log 2/log pm  | ≤ 1

so b1 is bounded above, b1 ≤ B say, and therefore, by Lemma 6.4, pbi i < 4 × 2B

is bounded also for 1 ≤ i ≤ l, so F is finite and the proof of the lemma is
complete. 
Lemma 6.6 [123, proposition 11] If n1 < n2 are two successive superabun-
dant numbers then n2 ≤ 2n1 .
Proof Let n1 = 2k · · · p so 2n1 = 2k+1 · · · p. Then
σ(2n1 )/(2n1 ) 2k+2 − 1
= k+2 >1
σ(n1 )/n1 2 −2
so n2 ≤ 2n1 . 
It follows that there is a superabundant number in each real interval [x, 2x]
and that the number of superabundant numbers is thus infinite.
Recall that there is a nice upper bound for the prime power terms in a
superabundant number, namely that given in Lemma 6.4, qaq < 4 × 2a2 . In [3]
we see the remark that for q > 11 the factor of 4 can be removed. The next
lemma gives a convenient lower bound for the exponent:

Lemma 6.7 [123, lemma 14] If n = 2a2 · · · qaq · · · p is a superabundant

number, then for all q with 2 ≤ q ≤ p we have
# $
log p/log q ≤ aq .
Proof To get a contradiction set m := nqaq +1 /p and suppose that
 
log p m nqaq +1
aq ≤ − 1 =⇒ qaq +1 < p =⇒ = < 1.
log q n p
Since n is superabundant we must have
 
σ(n)/n 1 1
1< = 1+ =⇒ p < qaq +1 ,
σ(m)/m 1 + 1/qaq +1 p
152 A Variety of Abundant Numbers

which is a contradiction. Therefore log p/log q ≤ aq . This completes the

proof. 
Note that for all integers n > 1 we have n/ϕ(n) > σ(n)/n. The two ratios in
this inequality are a principal focus of this volume. For example they occur
in Chapters 5, 7 and 9. It is interesting to see that for superabundant numbers
the “gap” between the ratios has a nice expression in terms of the maximum
prime divisor of n. The next result of Nazardonyavi and Yakubovich was in
a preprint [122], but omitted in the final version [123].
Lemma 6.8 Let n be a superabundant number with maximum prime divisor
p and define
 
1 3
(p) := 1+ .
log p 2 log p
Then
σ(n) n
> (1 − (p)) .
n ϕ(n)
Proof Consider n = 2a2 · · · qaq · · · p and note that
 
σ(n)  1
> 1 − aq +1 .
ϕ(n) q≤p q
Take logarithms and use the inequality log(1 − 1/x) > −1/(x − 1), which is
valid for x > 1, and Lemma 6.7 to get
 
σ(n)  1
log > log 1 − aq +1
ϕ(n) q≤p q
 1  1
>− = −
q≤p
qlog p/log q − 1 q≤p
p−1
 
π(p) p 1.2762
=− >− 1+
p−1 (p − 1) log p log p
 
1 3
>− 1+ ,
log p 2 log p
where for the final inequality we need p ≥ 23 and we have used the estimate
π(x) < (x/log x)(1 + 1.2762/log x) which is valid for x > 1 [49]. Taking
exponentials and using e−x > 1 − x, which is true for all x > 0, we get
 
σ(n) 1 3
> 1− 1+ = 1 − (p).
ϕ(n) log p 2 log p
When p < 23 we note that the set of superabundant numbers with maximum
prime in this range is finite, and we are thus able to complete the proof by
computation; see Section 3.3. 
 6.3 Colossally Abundant Numbers 153

6.3 Colossally Abundant Numbers

The structure of the set of colossally abundant numbers is informed by the
concept of critical epsilon values. This idea was introduced by Erdős and
Nicolas [56] in their ground-breaking paper of 1975 on this subject. However,
the essential facts were already present in the section on colossally abundant
numbers in [3] published in 1944. The definition of colossally abundant
number given in the earlier paper was a little different, but here we will give
only the more modern version.
We give the definition which is now regarded as standard: recall that
we say that N is a colossally abundant number if for some  > 0 the
function σ(n)/n1+ attains its maximum at N. Therefore, for such an N and
its associated  we have for all n ∈ N
σ(n) σ(N)
≤ 1+ .
n1+ N
For a given  > 0 there is at least one colossally abundant number for that
given . This follows from bounds on σ(n)/n which show σ(n)/n1+ → 0 as
n → ∞. We have already seen in the introduction to this chapter that every
colossally abundant number is superabundant, so any result proved for that
type of number can be used if needed.
Let p be a prime. Define
%   & '
1
E p := log p 1 +  i
: j ∈ N and E := E p.
1≤i≤ j p p

The set E can be written with elements in decreasing order:

⎧ ⎫
⎪
⎪ 3 4 7 6 15 13 31
⎨ log( 2 ) log( 3 ) log( 6 ) log( 5 ) log( 14 ) log( 12 ) log( 30 ) ⎪
⎪
⎬
E=⎪ ⎪ , , , , , , , . . . ⎪
⎪
⎩ log(2) log(3) log(2) log(5) log(2) log(3) log(2) ⎭
= {0.5849 . . . , 0.2618 . . . , 0.2223 . . . , 0.1132 . . . , 0.09953 . . . ,
0.07285 . . . , 0.04730 . . . , . . .}.

We enumerate the elements of E in this order (i )i∈N so

1 = log2 ( 32 ) > 2 = log3 ( 43 ) > 3 = log2 ( 76 ) > 4 = log5 ( 65 ) > · · ·

and set 0 := ∞. The sequence E is called the critical epsilon values.

Let α ≥ 0 be an integer and p a prime. Define F(p, 0) := ∞ and for
α > 0 set
 
log[(pα+1 − 1)/(p(pα − 1))] 1
F(p, α) := = log p 1 + α ,
log p p + ··· + p
so F(p, α) is strictly decreasing as p and α increase (see Figure 6.2).
154 A Variety of Abundant Numbers
F(x,α)

0.04

0.03

0.02

0.01

α=2
x α=3
5 10 15 20

Figure 6.2 The functions F(3, α) and F(2, α) showing F(3, α) < F(2, α).

Lemma 6.9 [56, proposition 3] If N is colossally abundant for  and p is

any prime with pα  N then

F(p, α + 1) ≤  ≤ F(p, α).

Conversely, if N is a positive integer such that for all primes p with pα  N we

have F(p, α + 1) ≤  ≤ F(p, α), then N is colossally abundant for .
Proof If α ≥ 0 let n = N p. Then
σ(N p)  N p 1+
≤ = p1+ .
σ(N) N
Also
 
σ(N p) σ(pα+1 ) pα+2 − 1 1
= = α+1 = p 1+ ,
σ(N) σ(pα ) p −1 p(pα + · · · + 1)
so 1 + 1/(p(pα + · · · + 1)) ≤ p which gives F(p, α + 1) ≤ . Similarly if α > 0
we can use n = N/p to obtain F(p, α) ≥ . This completes the proof. 
Note that we have E p = {F(p, α) : α ≥ 1}
/and that for each η > 0 there are at
most a finite number of elements of E = p E p larger than η.
Now define for given  > 0 a strictly decreasing sequence of positive real
numbers (xk )≥1 , depending on  through the equations F(xk , k) = , and set
x = x1 and y = x2 .
Lemma 6.10 [56, proposition 4(a)] If   E, i.e.  is not a critical epsilon
value, then σ(n)/n1+ has a unique maximum at the colossally abundant
 6.3 Colossally Abundant Numbers 155

integer denoted n which has exponents for each prime q

  1+ 
q −1
αq () = logq − 1.
q − 1
In other words, αq () = k when xk+1 < q < xk for k ≥ 1 and αq () = 0 if q > x1 .
Proof Let q be a fixed prime number. Since   Eq and the function F(q, k)
is strictly decreasing in k, there exists a unique integer α such that
F(q, α + 1) <  < F(q, α). (6.2)
Solving these inequalities:
1 1
1+ < q < 1 + α
qα+1 + · · · + q q + ··· + q
1 q(q − 1) 1
< < α
qα+1 − 1 q−1 q −1
q−1
qα − 1 < < qα+1 − 1
q(q − 1)
q+1 − 1
qα < < qα+1
q(q − 1)
q+1 − 1
qα+1 <  < qα+2 .
q −1
Therefore
    
q+1 − 1 q1+ − 1
α + 1 < logq < α+2 =⇒ αq () = logq − 1.
q − 1 q − 1
If xk+1 < q < xk we have using (6.2)
F(xk , α + 1) < F(q, α + 1) < F(xk , k) = 
= F(xk+1 , k + 1) < F(q, α) < F(xk+1 , α).
Thus F(xk , α + 1) < F(xk , k) and F(xk+1 , k + 1) < F(xk+1 , α). Therefore since
F(x, α) is strictly decreasing in α, we get k < α + 1 and α < k + 1, so we must
have α = k, showing the alternative expression αq () = k is true also.
To demonstrate the final part of the lemma, note that we have  = F(x1 , 1)
which implies x1 = 1 + 1/x1 so
 +1   
x1 − 1 x1 (1 + 1/x1 ) − 1
log x1 = log x1 = log x1 x12 = 2.
x1 − 1 1/x1
But we have q > x1 which gives logq ((q1+ − 1)/(q − 1)) < 2 and thus
  1+ 
q −1
logq ≤ 1,
q − 1
so therefore αq () = 0.
156 A Variety of Abundant Numbers

The converse is immediate, via contradiction, because (6.2) determines the

α and they determine N. This completes the proof. 
Using the expression of Lemma 6.10 it is easy to generate colossally
abundant numbers. For example
 = 0.1 =⇒ n = 22 · 3 · · · 5
 = 0.01 =⇒ n = 26 · 32 · 52 · 7 · · · 29
 = 0.001 =⇒ n = 29 · 35 · 53 · 73 · 112 · 132 · 172 · 19 · · · 181
 = 0.0001 =⇒ n = 212 · 37 · 55 · 74 · 113 · 172 · · · 272 · 53 · · · 541
and
 = 0.00001 =⇒
n = 2 · 3 · 5 · 75 · 114 · 134 · 173 · · · 293 · 312 · · · 1392 · 149 · · · 541,
16 10 6

where the symbol “ · · · ” means a product of a contiguous set of primes in

increasing order all to the same power as those on its left and right.
Lemma 6.11 [56, proposition 4(b)] For each i ≥ 1, n is constant on the
open interval (i , i−1 ). These constant values, for different values of i, are all
distinct.
Proof If i is a critical epsilon value we have i = F(q, α) for some prime q
and positive integer α. A calculation then shows that
 1+ 
q i −1
1 + α = logq ,
qi − 1
so g() := logq ((q1+i − 1)/(qi − 1)) is an integer. Similarly if for given q this
value is an integer then the corresponding  is a critical  value. Therefore
if i+1 < i are two successive critical epsilon values, for each q the value of
αq () does not change when  ∈ (i , i−1 ), so n is constant on each of these
intervals.
Because the function g() is strictly increasing as  decreases, if
i+1 <   < i <  < i−1
then αq (  ) > αq (). This completes the proof. 
We denote by ni the constant value of n on the interval (i , i−1 ) so
n1 = 1, n2 = 2, n3 = 6, n4 = 12, n5 = 60, . . .
giving, by Lemma 6.10, the initial segment of the colossally abundant
numbers.
It follows from Lemma 6.11 that ni+1 > ni , so in particular the values of the
(ni ) are all distinct. Therefore the number of the colossally abundant numbers
corresponding to the sequence (ni )i≥1 is infinite.
 6.3 Colossally Abundant Numbers 157

Lemma 6.12 [56, proposition 4(c)] If the sets (Eq )q∈P are pairwise disjoint,
the set of colossally abundant numbers is equal to the set of ni with i ≥ 1. The
function n → σ(n)/n1+i attains its maximum at the two points ni and ni+1 .
Proof Choose  = i = F(q, β). For p  q, since the Eq are pairwise disjoint,
  E p so α p () is given by the expression of Lemma 6.10. By Lemma 6.9 the
exponent of q can be chosen to be β or β − 1. In the first case we get ni+1 and
in the second ni with
qβ+1 − 1 σ(ni ) σ(ni+1 )
q1+i = ⇐⇒ = ,
qβ − 1 n1+
i
i
n1+
i+1
i

so the function attains its maximum at these two points. 

Figure 6.3 shows two maxima at n4 = 12 and n5 = 60. Figure 6.4, with
 = 0.1 such that
log( 15 ) log( 65 )
14
<< =⇒  ∈ (5 , 4 ),
log 2 log 5
shows one maximum at the colossally abundant number n5 = 60.
Lemma 6.13 [56, proposition 4(d)] If the sets (Eq )q∈P are not pairwise
disjoint, for each i ∈ Eq ∩ Er , the function n → σ(n)/n1+i attains its maximum
at the four points
ni , qni , rni , qrni = ni+1 .

σ(n)/n1+∈4
2.0

1.5

1.0

0.5

n
0 20 40 60 80 100 120

Figure 6.3 The function n → σ(n)/n1+4 .

158 A Variety of Abundant Numbers
Table 6.1 Indexing for the first eight colossally abundant
numbers.

i ni i Value of i σ(ni )/n1+

i
i

1 1 log2 (3/2) 0.5849 1.00000

2 2 log3 (4/3) 0.2618 1.25102
3 6 log2 (7/6) 0.2223 1.34269
4 12 log5 (6/5) 0.1132 1.76086
5 60 log2 (15/14) 0.09953 1.86281
6 120 log3 (13/12) 0.07285 2.11659
7 360 log2 (31/30) 0.04730 2.46012
8 2520 log2 (63/62) 0.02308 3.09998

σ(n)/n1+∈
2.0

1.5

1.0

0.5

n
0 10 20 30 40 50 60 70

Figure 6.4 The function n → σ(n)/n1+0.1 .

Proof Let  = i = F(p, α) ∈ E. First we claim  is irrational. If not we could

write  = a/b for integers a and b so
 b
1
p = 1+ α
a
,
p + ··· + p

so the left-hand side is an integer and the right-hand side not an integer. (The
theorem of Gel’fond and Schneider, that if α and β are algebraic numbers
with α  0 or 1 and if β is not a rational number, then any value of αβ is a
transcendental number, would show that in fact  is a transcendental number.)
 6.3 Colossally Abundant Numbers 159

The theorem of Lang, given in Lemma 6.14 using

(u, v, w) = (log p, log q, log r),
a =  and b = 1, implies that if p, q, r are distinct prime numbers such that
p , q and r are all algebraic then  would be rational, which is impossible.
But if  ∈ E p , p is rational, so we conclude that E p ∩ Eq ∩ Er = ∅.
If there exist two sets Eq and Er which are not disjoint and we choose
i = F(q, β) = F(r, γ) ∈ Eq ∩ Er
and a prime p satisfying p  q and p  r, the maximum power of p dividing
N is determined by Lemma 6.10. The power of q is β or β − 1 and that of r is
γ or γ − 1, giving by Lemma 6.9 the four given possibilities. This completes
the proof. 
Note unsolved problem (2) at the end of this chapter: no pair of distinct
primes p, q are known with E p ∩ Eq  ∅, so we are not able to illustrate
Lemma 6.13.
Lemma 6.14 [106, chapter 2] If u, v, w are three complex numbers, linearly
independent over Q, and a, b are two complex numbers also linearly
independent over Q, then at least one of the six numbers
eau , eav , eaw , ebu , ebv , ebw
is a transcendental number.

Lemma 6.15 The quotient of two consecutive colossally abundant numbers

is either a prime or the product of two distinct primes.
Proof (1) Let n and n with n < n be two consecutive colossally abundant
numbers with associated epsilon values  and   , so for some i ∈ N there
is a critical epsilon value i with   < i < . Then, with the notation of
Lemma 6.11 write
  
n= pα p () and n = pα p ( ) ,
p p

where the products are over all primes, and for all p, 0 ≤ α p () ≤ α p (  ).
(2) For each prime p define a function on (0, 1) by
 1+t 
p −1
f p (t) := log p ,
pt − 1
and note that f p (t) increases as t decreases, and is a continuous function of t.
(3) By Lemma 6.10 we have α p () =  f p () − 1 and α p (  ) =  f p (  ) −
1. For fixed p, as t decreases from  to   , since n and n are consecutive
colossally abundant numbers and f p (t) is continuous, α p (  ) must differ from
160 A Variety of Abundant Numbers

α p () by at most one, and in such a case where they differ, f p (i ) must be a
positive integer greater than 1, say f p (i ) = m. This means

pi +1 − 1 pm − 1
= pm =⇒ pi = ,
pi − 1 pm − p
which is a rational number.
(4) If this exponent jump occurs at three or more distinct primes p, q, r
then we would have a real i with pi , qi and ri each rational. But again by
Lemma 6.14, this is impossible. This completes the proof. 

A long-standing conjecture of Erdős is that if p and q are distinct odd

primes with real  > 0 being such that p and q are both rational, then  is an
integer. We use the result of this conjecture in RHpack to generate colossally
abundant numbers using the function GenerateCA.
Now recall the definition
σ(n)
G(n) := .
n loglog n
The following lemma, originally due to Robin, is presented with a different
proof. It was used in crucial ways by Briggs in [16] to extend the lower bound
for any counterexample to Robin’s inequality. It is also needed in the chapters
which follow.

Theorem 6.16 [142, proposition 1] If M < N are consecutive colossally

abundant numbers and M ≤ n ≤ N then

G(n) ≤ max(G(M),G(N)).

Proof By Lemmas 6.12 and 6.13 there exists an index i ≥ 1 such that M = ni
and N = ni+1 and such that these two numbers are critical points giving the
same maximum value for the function n → σ(n)/n1+i .
The function f (x) := i x − loglog x is concave upwards for x ≥ 2, so for
2 ≤ a ≤ ξ ≤ b we have f (ξ) ≤ max( f (a), f (b)). Because ni ≤ n ≤ ni+1 this
means

i log n − logloglog n
≤ max(i log ni − logloglog ni , i+1 log ni+1 − logloglog ni+1 ),

or taking exponentials
 
ni ni i ni+1
i

≤ max , . (6.3)
loglog n loglog ni loglog ni+1
 6.4 Estimates for x2 () 161

But ni satisfies σ(n)/n1+i ≤ σ(ni )/n1+

i
i
so
 
σ(n) σ(ni ) ni loglog ni n i
G(n) = ≤ = G(ni ) (6.4)
n loglog n n1+
i
i loglog n loglog n ni
and the same inequality holds with ni replaced by ni+1 .
If we had G(n) > max(G(ni ),G(ni+1 )), by (6.4) we would have
   
loglog ni n i loglog ni+1 n i
1< and 1 <
loglog n ni loglog n ni+1
and therefore
 
ni i ni+1
i
ni
max , < ,
loglog ni loglog ni+1 loglog n
contradicting (6.3). Hence G(n) ≤ max(G(ni ),G(ni+1 )). This completes the
proof. 

6.4 Estimates for x2 ()

For what follows in Chapter 7 some explicit estimates for the numbers xi ,
introduced in Section 6.3, are needed. Erdős and Nicolas [56, p. 74] prove
that, for k ≥ 1 a fixed integer, we have
  
√ log 2 1
xk ∼ kx and x2 = 2x 1 −
1/k
+O . (6.5)
2 log x log2 x
Recall the notation x = x1 = x1 (). Here are given explicit inequalities relating
to these asymptotic forms in the following lemma.
Lemma 6.17 [142] Let  > 0 not be a critical epsilon value with xi := xi ()
for i ∈ N. Then
x1/k < xk (x > 1, k ≥ 2), (6.6)
√
x2 < 2x (x > 1), (6.7)
 
√ log 2
2x 1 − < x2 (x ≥ 1530). (6.8)
2 log x
Proof Setting t = x1/k , we note that proving inequality (6.6) is equivalent to
proving F(x, 1) = F(xk , k) < F(t, k). In other words
   
1 1
log 1 + k < k log 1 + . (6.9)
t t + t2 + · · · + tk
For u > 0 we have 1 + ku < (1 + u)k . Hence
   k
1 1 1
1+ k < 1+ < 1+ . (6.10)
t (t + t2 + · · · + tk )/k t + t2 + · · · + tk
Thus inequality (6.10) holds, and hence inequality (6.6) is true.
162 A Variety of Abundant Numbers
√
To derive inequality (6.7), since it is equivalent to F( 2x, 2) < F(x, 1) =
F(x2 , 2), we must show that
   
1 1
2 log 1 + √ log 1 +
2x + 2x x
< . (6.11)
log x + log 2 log x
In fact, we claim that
   
1 1
2 log 1 + √ < log 1 + , (6.12)
2x + 2x x
from which inequality (6.11) follows directly. √
To prove inequality (6.12), substitute y = 2x and simplify:
 2
1 1
1+ √ ≤ 1+ (x > 0). (6.13)
2x + 2x x
√
To derive inequality (6.8) we first set z = 2x (1 − u/2) where u =
log 2/log x. Because the given inequality is equivalent to F(x1 , 1) =
F(x2 , 2) < F(z, 2), we need to show
   
1 1
log 1 + log 1 +
x z + z2
< . (6.14)
log x log z
Using the inequalities 1/(a + 1) < log(1 + 1/a) < 1/a, a > 1, we find that, to
prove inequality (6.14) it is sufficient to show that
1 √  u  1
log 2x 1 − < , (6.15)
x log x 2 1 + z2 + z
which will follow if
 
u2
(1 + z + z ) 1 + u −
2
< 2x. (6.16)
log 2
Consider
    
u2 u 2 u2
z 1+u−
2
= 2x 1 − 1+u− . (6.17)
log 2 2 log 2
For x ≥ 212 we have u ≤ 1/12, so the right-hand side, and hence the left-hand
side, of (6.17) is bounded above by 2x(1 − 2.05u2 ).
The other part of (6.16) can be bounded, since 0 < u ≤ 1/12 and x ≥ 212 , as
follows:
 
u2 √  u √
(1 + z) 1 + u − ≤ 2x 1 + + 1 + u ≤ 1.06 2x. (6.18)
log 2 2
 6.5 Unsolved Problems 163

Therefore
 
u2 √
(1 + z + z ) 1 + u −
2
≤ 2x(1 − 2.05u2 ) + 1.054 2x
log 2
 2
√ log 2
= 2x + 1.06 2x − 4.1x < 2x,
log x
where the final inequality follows since for x ≥ 212 we have
 2
√ log 2
1.06 2x − 4.1x < 0.
log x
This proves inequality (6.8) for all x > 212 . For 1530 ≤ x ≤ 212 , inequality
(6.14) can easily be verified numerically, so the proof of the lemma is
complete. 
We also need the following with k = 3, stated in [56] to be proved along
the same lines as lemma 2(b) in [3].
Lemma 6.18 As  → 0 we have xk ∼ (kx1 )1/k .
Proof We have for x = x1 and y = xk
   
1 1
log 1 + log 1 +
y + y2 + · · · + yk x
= = .
log y log x
Note that  → 0+ if and only if x → ∞ if and only if y → ∞. Hence, using
log(1 + 1/x) ∼ 1/x as x → ∞ we get
yk log y ∼ x log x. (6.19)
Taking logarithms gives k log y ∼ log x. Inserting this in (6.19) then gives yk ∼
kx, completing the proof of the lemma. 

6.5 Unsolved Problems

(1) Show that the ratio of any two consecutive colossally abundant numbers
is always prime – see Table A.5.
(2) Show that the sets (Eq )q∈P ) are disjoint (Robin [142]).
(3) If p and q are distinct primes with x real and p x and q x simultaneously
rational, then x is an integer (Erdős and Alaoglu [3]).
(4) Erdős and Alaoglu [3] state that q > 11 implies qaq < 2a2 when n is
superabundant. Generalize this result.
(5) If n is superabundant then we have qaq ∼ p log p/log q when q → ∞ and
p ∼ log n when n → ∞ [3].
164 A Variety of Abundant Numbers

(6) For fixed k, the function t → F(t, k) is strictly decreasing for all t > 0,
and tends to zero when t → ∞. Therefore, as k → ∞, if n is colossally
abundant with xk := xk (), k ∈ N,
xk ∼ (kx)1/k .
This result is in [3]. Prove it and then give an error estimate.
(7) It was proved in [3] that the number of superabundant numbers N(x) up
to x > 0 satisfies
log x loglog x
N(x) .
(logloglog x)2
Improve this bound and find a lower bound for N(x).
(8) Find “book proofs” for some of the more difficult theorems and lemmas
in [3].
(9) Find an RH equivalence expressible in terms of superabundant numbers.
 7
Robin’s Theorem

7.1 Introduction
This chapter has as its focus Robin’s theorem, an explicit inequality involving
the sum-of-divisors function, valid on an explicit range, its validity being
equivalent to RH. It forms part of what we call the Ramanujan–Robin
criterion. We include Ramanujan because this great Indian mathematician
first demonstrated the truth of the inequality assuming RH, albeit the
asymptotic form.
The life of Srinivasa Ramanujan (pronounced Ra-manu-jan) requires no
introduction, since it has been celebrated in books, plays and film. His vast
mathematical contributions have been treated in depth in the monumental
works of Bruce Berndt, including [10].
Guy Robin was supervised by Jean-Louis Nicolas (the Nicolas of
Chapter 5) at the Université de Limoges, and received his Ph.D. in 1983 with
a thesis entitled “Grandes valeurs de fonctions arithmétiques et problèmes
d’optimisation en nombres entiers”. One of his students, Pierre Dusart, is
making important contributions to the ongoing development of this approach
to RH. According to [128], Robin did not learn of Ramanujan’s unpublished
result, showing Robin’s inequality was necessary if RH was true, until several
years after he had discovered his theorem and proof.
We have already seen some introductory and historical information
regarding σ(n) in Chapter 6. It should come as no surprise that, because of the
close relationships between the sum-of-divisors function and Euler’s totient
function ϕ(n), and the equivalences to RH developed in Chapter 5, there are
equivalences to RH which can be expressed in terms of σ(n).
Paul Erdős was interested in common properties of σ(n) and ϕ(n) through-
out his career, and authored or co-authored many papers exploring this
connection. For example Alaoglu and Erdős [3] state that both σ(n) and ϕ(n),
for every  > 0, apart from a set of density zero, are divisible by every prime
less than (loglog n)1− and by “relatively few” primes larger than (loglog n)1+ .

165
166 Robin’s Theorem

Aside from being multinomials in the prime factors of n, and obeying

the inequality σ(n)/n ≤ n/ϕ(n), at first glance these functions have very
little in common. Their definitions and genesis are completely different.
Nevertheless, properties of one function are frequently mirrored by the
properties of the other one. As an illustration, their average orders satisfy
 
σ(n)  ϕ(n),
n≤x n≤x

where f (x)  g(x), for positive functions f (x) and g(x), means that there are
numbers α > 0, β > 0 with, as x → ∞, αg(x) ≤ f (x) ≤ βg(x). Moreover the
same is true [53, 58] for the size of the sets of values
#{σ(n) : n ≤ x}  #{ϕ(n) : n ≤ x}.
In addition, the normal order of ω(σ(n)) is the same as the normal order
of ω(ϕ(n)), namely (loglog n)2 /2. Recall that we say arithmetic function f (n)
has normal order ρ(n), for some non-negative non-decreasing function ρ(n),
if for every  > 0 we have
%   &
# n ≤ x :  f (n)
− 1 < 
ρ(n)
lim = 1.
x→∞ x
See [75] or [95] for the development of this topic.
Work has also been done on special values of these functions; for example
see [27].
Recently there has been much work on “joint properties” shared by these
two functions. For instance there exists an infinite number of squarefree
integers n such that σ(n) ϕ(n) is a square [29]. In a similar vein Ford, Luca
and Pomerance [61] proved that the two functions have an infinite number
of values in common, that is σ(a) = ϕ(a ) for an infinite set of integer pairs
(a, a ) (see also Ford and Pollack [62, 63]). The case of when ϕ(n) divides
σ(n) has also been explored [28, 30].
Now we focus on the material in this chapter. Consider the growth of
σ(n)/n. On the one hand, whenever n is a prime we have σ(n)/n = 1 + 1/n,
so the range is arbitrarily close to 1. On the other hand, σ(n) cannot grow too
quickly. Grönwall [69] showed that
σ(n)
lim sup = eγ . (7.1)
n→∞ n loglog n
We do not use this result in this chapter, but it is proved and used in Chapter 9
as Theorem 9.2. What is interesting is the manner in which this limit superior
is approached. Ramanujan [136], assuming the Riemann hypothesis, proved
that σ(n)/(n loglog n) is asymptotically bounded above by eγ . That is he
showed:
 7.1 Introduction 167

Theorem 7.2 [136, 137] Assume the Riemann hypothesis is true. Then there
exists an n0 ≥ 1 such that
σ(n)
< eγ , (7.2)
n loglog n
for all n ≥ n0 .
Robin [142], using explicit estimates for arithmetic functions given by
Rosser and Schoenfeld [145, 153, 154], set out in Chapter 3, gave an explicit
value of n0 in Theorem 7.2. In [142, proposition 3, p. 204], he proves:
Theorem 7.11 [142, theorem 1] Assume the Riemann hypothesis is true.
Then
σ(n) < eγ n loglog n (n ≥ 5041). (7.3)
Moreover the values of 1 < n ≤ 5040 for which the inequality in (7.3) fails are
listed in the set A, where
A := {1, 2, 3, 4, 5, 6, 8, 9, 10, 12, 16, 18, 20, 24, 30, 36, 48,
60, 72, 84, 120, 180, 240, 360, 720, 840, 2520, 5040}. (7.4)
The inequality in expression (7.3) is known as “Robin’s inequality”. Note
that in this volume the term does not include the lower bound n ≥ 5041 for
the range of validity, which can sometimes cause confusion. We attempt to
make the range explicit whenever the term is used.
Since Grönwall’s result (7.1) is unconditional, one may expect an explicit
inequality similar to inequality (7.2) in which limit of the right-hand side was
eγ . Robin proved such a result (with a different constant) in:

Theorem 7.13 [142, theorem 2]

σ(n) 2
≤ eγ loglog n + (n ≥ 3). (7.5)
n 3 loglog n
One can compare Ramanujan’s theorem, Theorem 7.2, and Robin’s
theorem, Theorem 7.11, to see that, were Robin’s inequality (7.2) to fail for
some n ≥ 5041, it could only fail slightly.
However, inequality in (7.3) fails infinitely often if the Riemann hypothesis
is false! Robin [142, section 4] uses an oscillation result first proved by
Nicolas [126] to prove the following:
Theorem 7.15 [142, theorem 1] Assume the Riemann hypothesis is false,
and let θ be the supremum over all real parts of the non-trivial zeros ρ = β+iγ,
so θ > 12 . For any number b ∈ (1 − θ, 12 ), there exists a positive constant c such
that  
σ(n) c
> eγ 1 + , (7.6)
n loglog n (log n)b
for infinitely many values of n.
168 Robin’s Theorem

One may therefore combine Theorems 7.11 and 7.15 to give the following
equivalence for RH, which is known also as “Robin’s theorem”:
Theorem 7.16 (Ramanujan–Robin criterion) [142, theorem 1] A necessary
and sufficient condition for the Riemann hypothesis is that
σ(n) < eγ n loglog n (n ≥ 5041). (7.7)
Because of the function σ(n), the ratio of the left side to the right side of
inequality (7.3) is far from smooth.
We see in Figure 7.1 that, even for small values of n > 5040, the ratio is
frequently close to 1.0.
Lagarias [100] found a modification of Theorem 7.16, namely:

Theorem 7.18 (Lagarias’ criterion) [100, theorem 1.1] Let Hn = nj=1 j−1
represent the nth harmonic number in Q. A necessary and sufficient condition
for the Riemann hypothesis is that
σ(n) ≤ Hn + exp(Hn ) log(Hn ) (7.8)
is true for all n ≥ 1.
To summarize this chapter: Section 7.2 gives the proof of Ramanujan’s

theorem, which assumes RH. In Section 7.3 a lower bound
 for p≤x (1 −
1/p) is derived. Section 7.4 gives an upper bound for p≤x (1 − 1/p2 ). In
Section 7.5 there is a lower bound for loglog N and in Section 7.8 an

σ(n)/neγ loglog(n)

0.8

0.6

0.4

0.2

n
5200 5400 5600 5800 6000

Figure 7.1 The values of σ(n)/(neγ loglog n) for 5041 ≤ n ≤ 6041.

 7.2 Ramanujan’s Theorem Assuming RH 169

Figure 7.2 Timothy Trudgian.

upper bound. Section 7.6 includes the proof of Robin’s theorem in the
case RH is true. Section 7.7 derives Robin’s unconditional upper bound for
σ(n)/n. Section 7.9 gives the proof of Robin’s theorem when RH is false.
In Section 7.10, we give the proof of Lagarias’ theorem, which depends on
Robin’s theorem. In Section 7.11 an unconditional upper bound related to
Lagarias’ formulation is presented, and in the final Section 7.12, some ideas
relating to the sum of unitary divisors of an integer are discussed.
The substantial contributions of Tim Trudgian (Figure 7.2) to this volume,
especially this essential chapter, are gratefully acknowledged.
The following RHpack (see Appendix B) functions relate to the material
in this chapter: RobinsInequalityQ, PlotRobinsInequality, SigmaOverN,
LagariasInequalityQ and UnitaryDivisorSigma.
To obtain an overview of the dependences between the results in this
chapter, the reader is invited to consult Figures 7.3, 7.4, 7.5 and 7.6.

7.2 Ramanujan’s Theorem Assuming RH

Before preceding to a proof of Theorem 7.11 we first prove Theorem 7.2.
Thereafter, even though a moderate amount of work is required, all that we
need do is to calculate certain error terms explicitly. Recall that we have
defined

G(n) := σ(n)/(n loglog n), n > e.

170 Robin’s Theorem

Lem 3.14

Lem 7.7
Lem 7.10

Lem 5.7
Thm 4.6
Thm 6.16
Thm 7.11
Lem 5.24
Thm 7.2
Lem 7.8

Lem 7.1

Lem 6.17
Lem 7.9 Lem 2.10

Lem 3.12

Figure 7.3 Relationships between some of the results in Chapter 7.

First a simple but essential lemma:

Lemma 7.1 If a counterexample to Robin’s inequality exists for some n >

5040, then so does a counterexample which is colossally abundant.

Proof Let N be a counterexample to Robin’s inequality with N > 5040. It

follows from the results of Chapter 6 that there are an infinite number of
colossally abundant numbers (ni ) with ni → ∞. If each one satisfied Robin’s
inequality then G(ni ) < eγ for all i. Thus for some i we would have ni ≤ N ≤
ni+1 . By Theorem 6.16, we would have

G(N) ≤ max(G(ni ),G(ni+1 )) < eγ ,

so N would also satisfy Robin’s inequality, which is false. Hence at least

one of the ni cannot satisfy Robin’s inequality, giving a colossally abundant
counterexample. This completes the proof. 

Looking ahead, Lemma 8.9 of Akbary and Friggstad will show that if a
counterexample exists to Robin’s inequality greater than 5040, then the first
such example must be a superabundant number.
 7.2 Ramanujan’s Theorem Assuming RH 171

Lem 3.13

Lem 3.10

Lem 7.5

Thm 2.1
Lem 7.4
Lem 7.6
Lem 7.8 Lem 7.7
Lem 3.14

Lem 7.3

Thm 4.6

Lem 3.11

Figure 7.4 Further relationships between some of the results in Chapter 7.

Lem 3.14 Thm 5.25

Thm 7.13

Lem 7.12

Lem 5.9 Lem 5.11

Figure 7.5 Additional relationships between some of the results in Chapter 7.

172 Robin’s Theorem

Thm 5.29 Thm 7.2

Lem 7.14 Thm 7.15 Thm 7.16 Thm 7.11

Lem 5.7 Lem 6.17

Figure 7.6 Further relationships between some of the results in Chapter 7.

Now we present the statement of Theorem 7.2. Note that the proof depends
on the result of Lemma 7.10, which is proved in Section 7.5.
Theorem 7.2 (Ramanujan) [136, 137] Assume the Riemann hypothesis is
true. Then there exists an n0 ≥ 1 such that
σ(n)
< eγ , (7.9)
n loglog n
for all n ≥ n0 .
Proof Assume there exists an arbitrarily large counterexample N to inequal-
ity (7.9), which by Lemma 7.1 is colossally abundant. Then, by Theorem 6.16
and the results of Chapter 6 we have
σ(N)   
1 

1 1

= 1+ 1 + + 2 × ··· , (7.10)
N x <p≤x
p x <p≤x p p
2 3 2
 7.3 Ramanujan’s Theorem Assuming RH 173
√
where the xi are defined in Chapter 6. By Lemma 6.17, we have x2 < 2x, so
that we can bound (7.10) by
 √
x2 <p≤x (1 − 1/p ) 2x<p≤x (1 − 1/p )
2
σ(N) 2
≤  ≤  . (7.11)
N p≤x (1 − 1/p) p≤x (1 − 1/p)

We now proceed to estimate the numerator in the right-hand side of (7.11).

By Lemma 5.7 with α = 2 we get
 1  
1 1
= +O .
p>x
p2 x log x x log2 x

Using the Maclaurin series for −log(1 − z) and neglecting all terms other than
the first two we see that
   ⎛
⎜⎜⎜  1  ⎞⎟
⎟⎟
1 − 2 = exp ⎜⎜⎜⎜⎝− + O 3/2 ⎟⎟⎟⎟⎠
1 1
p p 2 x
√ √
2x<p≤x 2x<p≤x
⎛ √  ⎞
⎜⎜⎜ − 2 1 ⎟⎟⎟
= exp ⎝ √ ⎜ +O √ ⎟⎠ , (7.12)
x log x x log2 x
which gives an upper bound for the numerator in (7.11). To bound the
denominator below, one may use the result of Nicolas, Lemma 5.24, namely
 1

e−γ

−(2 + β)

1

1− ≥ exp √ +O √ , (7.13)
p≤x
p log θ(x) x log x x log2 x

where, because we are assuming RH, by Lemma 2.10,

 1  1
β := = = γ + 2 − log 4π = 0.0461 . . . .
ρ
ρ(1 − ρ) ρ
|ρ|2

We now proceed to estimate loglog N. Since N is colossally abundant we

have, by Lemma 7.10, for x ≥ 20 000:
⎛ √ ⎞
⎜⎜⎜ 0.986 2 0.484 ⎟⎟⎟
loglog N > log θ(x) exp ⎝⎜ √ − √ ⎟⎠ . (7.14)
x log x x log2 x
Finally, using (7.11), (7.12), (7.13) and (7.14), and that N does not satisfy
Robin’s inequality, we see that
⎛ √  ⎞
σ(N) ⎜⎜⎜ 2 + β − 3 2 1 ⎟⎟⎟
γ
e ≤ ≤ e exp ⎜⎝ √
γ
+O √ 2
⎟⎠ . (7.15)
N loglog N x log x x log x
√
Since 2 + β − 3 2 < 0, it follows that, for sufficiently large x, and hence for
sufficiently large N, the right-hand side of estimate (7.15) is less than eγ . This
contradiction completes the proof of the theorem. 
174 Robin’s Theorem

7.3 Preliminary Lemmas With RH True

In this and the following two sections we prove preliminary lemmas which
are needed for the proof of Theorem 7.11, which is Robin’s inequality under
RH for n ≥ 5041. We seek explicit versions of theequations and estimates
(7.12), (7.13) and (7.14). The first task is to bound p≤x (1 − p−1 ) from below.
The approach taken is along the same lines as that in Chapter 5.
Let S (x) := θ(x) − x and R(x) := ψ(x) − x as in Chapter 5. Using Stieltjes
integrals, and integrating by parts, we have
 1  x dθ(t)
=
p≤x
p 2− t log t
 x
S (x) 1 S (t){log t + 1}
= loglog x − loglog 2 + + + dt. (7.16)
x log x log 2 2− (t log t)2
Note that by the prime number theorem we have S (x) = O(x/log x), so the
integral in (7.16) converges as x → ∞. Note also we need a lower limit of 2−
to correctly evaluate the Stieltjes integral at a point of discontinuity. Letting
 ∞
1 log t + 1
B := − loglog 2 + S (t) dt, (7.17)
log 2 2 (t log t)2
we can rewrite (7.16) as
1  ∞
S (x) S (t){log t + 1}
= loglog x + B + − dt,
p≤x
p x log x x (t log t)2

and so
1  
S (x)
= loglog θ(x) + B + loglog x − loglog θ(x) +
p≤x
p x log x
 ∞
S (t)(log t + 1)
− dt. (7.18)
x (t log t)2
The purpose of writing the expression this way is to combine it with one
which comes from the more commonly found definition for B, which can be
written:
  1
  1   1

1

log 1 − = B−γ− − log 1 − + . (7.19)
p≤x
p p≤x
p p>x p p

Then, by exponentiating (7.19), we shall obtain an explicit version of (7.13).

Before proceeding directly, we shall introduce some notation to stand for
various error terms analogous to those used in Chapter 5. Let
 ∞
S (t)(log t + 1)
L1 (x) = dt = J1 (x) − K1 (x), (7.20)
x (t log t)2
 7.3 Preliminary Lemmas With RH True 175

where
 ∞  ∞
R(t)(log t + 1) (ψ(t) − θ(t))(log t + 1)
J1 (x) = dt and K1 (x) = dt.
x (t log t)2 x (t log t)2
(7.21)
The following Lemma 7.3 has the same derivation using integration by
parts as that of Lemma 5.16, which has a more restricted formulation.
Lemma 7.3 [142, lemma 1] For n ≥ 1 and ρ a complex number with 0 <
ρ < 1, set  ∞ ρ
t n log t + 1
Fρ,n (x) := dt. (7.22)
x t n+1
log2 t
Then
n xρ−n
Fρ,n (x) = + rρ,n (x), (7.23)
n − ρ log x
where   ∞ ρ−n−1 
ρ xρ−n t
rρ,n (x) := − 2 +2 dt . (7.24)
(ρ − n)2
log x x log3 t
If ρ = 12 then for n ≥ 1 we have |ρ| ≤ |ρ − n| and thus
 
1 4
|rρ,n (x)| ≤ 1+ . (7.25)
|ρ|xn−1/2 log2 x (2n − 1) log x
This completes the statement.
We shall apply Lemma 7.3 where ρ is a non-trivial zero of the Riemann
zeta function. In particular, if we assume the Riemann hypothesis is true we
have ρ = 12 .
The next lemma bounds not only the first integral appearing in (7.21), but
also a generalization, which we shall need in Section 7.4.
Lemma 7.4 [142, lemma 2] Assume the Riemann hypothesis is true. For
n ≥ 1 and for x ≥ 2, let
 ∞
n log t + 1
Jn (x) = R(t) dt (7.26)
x tn+1 log2 t

and recall β := ρ 1/|ρ|2 , the sum over the imaginary zeros of ζ(s). Then
 
1 4 2 log 2π
β n+ + + √
log x (2n − 1) log2 x β x
−
xn−1/2 log x
 
1 4
β n+ +
log x (2n − 1) log2 x
≤ Jn (x) ≤ . (7.27)
xn−1/2 log x
176 Robin’s Theorem

Proof Consider first the explicit formula in Theorem 2.1 (or see for example
[80, theorem 29]):
 xρ  
1 1
ψ(x) = x − − log 2π − log 1 − 2 (7.28)
ρ
ρ 2 x

and write
 ∞ %  &   ∞ g(t)tρ
1 1
g(t) ψ(t) − t + log 2π + log 1 − 2 dt = − dt (7.29)
x 2 t ρ x ρ

where
n log t + 1
g(t) := . (7.30)
tn+1 log2 t
For n ≥ 1 the dominated convergence theorem justifies the interchange of
summation and integration on the right-hand side of (7.29). Recalling that
R(t) = ψ(t) − t we may write
Jn (x) = Jn(1) (x) + Jn(2) (x), (7.31)
where
 ∞ ρ
t n log t + 1
Jn(1) (x) =− dt,
ρ x ρ tn+1 log2 t
 ∞   (7.32)
1 1 n log t + 1
Jn(2) (x) = − log 2π + log 1 − 2 dt.
x 2 t tn+1 log2 t
For t ≥ x ≥ 2 we have 0 < − 12 log(1 − t−2 ) < log 2π, and it then follows that
2 log 2π
− < Jn(2) (x) < 0. (7.33)
xn log x
Next use Lemma 7.3 with ρ = 12 + iγ to get
n  xiγ  rρ (x)
Jn(1) (x) = − . (7.34)
xn−1/2 log x ρ ρ(ρ − n) ρ
ρ

Using the upper bound (7.25), the bounds which have been derived for Jn(2) (x),
and the estimate  
 xiγ   1
 ≤ = β, (7.35)
 ρ ρ(ρ − n)  ρ
|ρ|2
completes the proof of the lemma. 
We now proceed to bound the integral K1 (x) in (7.21). In order to do this
we require the following lemma to bound the difference ψ(x) − θ(x). This
gives the same result as Lemma 5.18 (b), but the proof is different. Note that
 7.3 Preliminary Lemmas With RH True 177

the result improves on [153, (5.3*)], which, instead of the estimate (7.37),
has √ √
0.998697 x < ψ(x) − θ(x) < 1.001093 x + 3x1/3 . (7.36)
Lemma 7.5 [142, lemma 3] Assume the Riemann hypothesis is true. Then
for all x ≥ 121
0 < ψ(x) − θ(x) − x1/2 ≤ 43 x1/3 . (7.37)
Proof Let g(x) = (ψ(x) − θ(x) − x1/2 )/x1/3 . We may first check that g(x) ≤
1.332 for all x ≤ 4.1 × 1011 . To do this we need to examine only those primes
with p ≤ x1/2 , that is to say, the first 52 104 primes for which the equation is
able to be verified
 numerically.
Since ψ(x) = n≥1 θ(x1/n ) we may write
 
ψ(x) − 2ψ(x1/2 ) = (θ(x1/n ) − 2θ(x1/(2n) )) = (−1)n+1 θ(x1/n ), (7.38)
n≥1 n≥1

so that
ψ(x) − θ(x) ≤ ψ(x1/2 ) + (ψ(x1/2 ) − θ(x1/2 )) + θ(x1/3 ). (7.39)
Next, by Lemma 3.10 (a), for 0 < x ≤ 4.1 × 1011 we have θ(x) < x. For 4.1 ×
1011 ≤ x ≤ 1.68 × 1023 we have x1/2 ≤ 4.1 × 1011 , and therefore
ψ(x) − θ(x) ≤ θ(x1/2 ) + 2(ψ(x1/2 ) − θ(x1/2 )) + θ(x1/3 )
 
2 2.666
≤ x + x 1 + 1/12 + 1/6 ≤ x1/2 + 1.24637x1/3 .
1/2 1/3
(7.40)
x x
Now consider x ≥ 1.68 × 1023 and make use of the following results due to
Rosser and Schoenfeld (recall we are assuming RH):
ψ(x) − θ(x) ≤ θ(x1/2 ) + 3x1/3 , x > 0,
θ(x) < 1.000081x, x > 0, (7.41)
1 √
|ψ(x) − x| ≤ x log2 x, x ≥ 74,
8π
which are, respectively Lemma 3.10, Lemma 3.11 and Theorem 4.6. It
follows using inequality (7.39) that
 
log2 x 1.000081 3
ψ(x) − θ(x) ≤ x + x 1.000081 +
1/2 1/3
+ + 1/6
32πx1/12 x1/12 x
≤ x + 1.33128x ,
1/2 1/3
(7.42)
which proves the first part of the lemma.
For the second part of the lemma when 121 ≤ x ≤ 1016 we use
Lemma 3.13 (b), √
ψ(x) − θ(x) > x, (7.43)
178 Robin’s Theorem

and for x ≥ 1016 we use Theorem 4.6,

1 √
|θ(x) − x| ≤ x log2 x (x ≥ 599), (7.44)
8π
and the lower bound ψ(x) − θ(x) > θ(x1/2 ) + θ(x1/3 ), to show that
 
log2 x log2 x
ψ(x) − θ(x) > x + x 1 −
1/2 1/3
− . (7.45)
32πx1/12 72πx1/6
Since the right-hand side of the inequality (7.45) is decreasing for all x ≥
e24 = 2.64 . . . × 1010 , we have ψ(x) − θ(x) > x1/2 + 0.33x1/3 , and this completes
the proof of the lemma. 
Recall some definitions:
S (x) = θ(x) − x,
 1
β := ,
ρ
|ρ|2
 ∞  
(ψ(t) − θ(t)) 1 1
K1 (x) := + dt,
x t2 log t log2 t
 ∞  
(ψ(t) − t) 1 1
J1 (x) := + dt,
x t2 log t log2 t
L1 (x) := J1 (x) − K1 (x).
Using Lemma 7.5 we are now able to bound K1 (x) in (7.21), which leads to:
Lemma 7.6 [142, lemma 4] Under the assumption of the Riemann
hypothesis, for x ≥ 2, we have
 ∞
log t + 1
−L1 (x) = − S (t) dt
x (t log t)2
β − 2 8 + 4β 2 2 log 2π
2+β+ + 2
+ 1/6 +
log x log x x x1/2
≤ √ . (7.46)
x log x
Proof Using the notation of Lemma 7.3 and the result of Lemma 7.5 we have
K1 (x) ≤ F 1 ,1 (x) + 43 F 1 ,1 (x). (7.47)
2 3

Using the result of Lemma 7.3 and the bound in the estimate (7.25), we bound
the right side of the estimate (7.47), noting from Lemma 5.16 that r 1 ,1 (x) ≤ 0,
3
to obtain  
2 4 1
K1 (x) ≤ √ 1+ 2
+ 1/6 . (7.48)
x log x log x x
Finally, the bounds (7.48) and (7.27) can be used to complete the proof of the
lemma. 
 7.3 Preliminary Lemmas With RH True 179

−1
We are finally in a position to bound p≤x (1 − p ) from below. In
Lemma 7.7 we find an upper bound for the multiplicative inverse of this
product:
Lemma 7.7 [142, lemma 5] For x ≥ 20 000,
 1
−1 
2+β α(x)

1− ≤ eγ log θ(x) exp √ + √ 2
, (7.49)
p≤x
p x log x x log x

where
S 2 (x)(log x + 1.31) 8 + 4β 2 log x 2 log 2π log x
α(x) := 3/2
+(β−2)+ + 1/6 + . (7.50)
2x log x x x1/2
Proof We use Taylor’s formula to expand loglog θ(x) about the point x, so
S (x) S 2 (x)
loglog θ(x) = loglog x + − g(ξ), (7.51)
x log x 2
for some ξ ∈ (x, θ(x)) or (θ(x), x), where g(t) = (log t + 1)/(t2 log2 t). For x ≥
20 000 we use Lemma 3.14 (a):
x
θ(x) > x − (x ≥ 19 421). (7.52)
8 log x
Note that the function g(t) is decreasing, so we may use inequality (7.52) to
write g(ξ) < g(x − x/(8 log x)). This implies, since log(1 − ) < − for  > 0,
1
log x + 1 −
8 log x
g(ξ) <  2  2 .
1 1
x2 1 − log x 1 −
2
8 log x 7.8 log2 x
Note that because −log(1 − ) < /(1 − ) for 0 <  < 1 we have used:
    2  2
1 1 1
log x 1 −
2
> log x − > log x 1 −
2
.
8 log x log x − 1 7.8 log2 x
If y := log x, consider for y ≥ log(20 000) > 9.90349 the expression
1
y+1−
0.44 8y
y + 1.26 + − 2  2 .
y 1 1
1− 1−
8y 7.8y2
This expression is strictly positive for y ≥ log(20 000): to see this convert it
to a polynomial in y with leading term 0.01y7 and then divide by y7 to get the
180 Robin’s Theorem

form k(y) + 0.01 where the rational function k(y) satisfies |k(y)| < 0.00007.
Therefore
log x + 1.26 + 0.44/log x log x + 1.31
g(ξ) < < . (7.53)
x2 log2 x x2 log2 x
This and (7.51) show that
S (x) S 2 (x) log x + 1.5626
loglog θ(x) ≥ loglog x + − . (7.54)
x log x 2 x2 log2 x
Insert this into (7.18) and (7.19) gives:
S (x)
loglog x − loglog θ(x) +
x log x
 
S (x)2 log x + 1.31  1
≤ − log 1 −
2 x2 log2 x p≤x
p
1     
1 1
= −B + γ + + log 1 − +
p≤x
p p>x p p
S (x)2 log x + 1.31
≤ −B + γ + loglog θ(x) + B + − L1 (x),
2 x2 log2 x
so exponentiating we get
 1
−1 
S (x)2 log x + 1.31

γ
1− ≤ e log θ(x) exp −L1 (x) + .
p≤x
p 2 x2 log2 x

Finally we are able to use Lemma 7.6 to complete the derivation of the upper
bound. 


7.4 Bounding p≤x (1 − p−2 ) From Above With RH True
In this section we make (7.12) explicit by giving a more precise form of the
sum of Lemma 5.7 in the case α = 2. The proof takes seven steps and relies
on many of the results which have been proved.
Lemma 7.8 [142, lemma 6] Assume RH is true. For x ≥ 20 000,
   ⎛ √ ⎞
1 ⎜⎜⎜ − 2 4 ⎟⎟⎟
⎜
1 − 2 ≤ exp ⎝ √ + √ 2 ⎠
⎟. (7.55)
√ p x log x x log x
2x<p≤x

Proof (1) We consider separately small and large values of x. For 2 × 104 ≤
x ≤ 4 × 109 we write
  1

6 

1
−1  
1
−1
1− 2 = 2 1− 2 1− 2 . (7.56)
√ p π √ p p>x
p
2x≤p≤x p≤ 2x
 
7.4 Bounding p≤x (1 − p−2 ) From Above With RH True 181

We may verify, computationally, that

 −1 ⎛ √ ⎞
6  1 ⎜⎜⎜ − 2 3.7 ⎟⎟⎟
1− 2 ≤ exp ⎜⎝ √ + √ 2 ⎠
⎟, (7.57)
π2 √ p x log x x log x
p≤ 2x

for x ∈ [2 × 104 , 4 × 109 ]. As for the second product in (7.56) we write

⎛ ⎞⎟
 1
−1 ⎜⎜⎜  
1 ⎟⎟
1− 2 = exp ⎜⎜⎜⎝− log 1 − 2 ⎟⎟⎟⎠
p>x
p p>x
p
⎛ ⎞ ⎛ ⎞ (7.58)
⎜⎜⎜ 1 ⎟⎟⎟ ⎜⎜⎜  1 ⎟⎟⎟
≤ exp ⎜⎜⎝⎜ ⎟⎟⎟ ≤ exp ⎜⎜⎝⎜(1 + 2.51 × 10 )
−9
⎟⎟⎟ .
p>x
p2 − 1 ⎠ p>x
p2 ⎠

We can estimate the final sum appearing in the upper bound of (7.58) by
using Abel summation and integrating by parts so that
 1  ∞
−θ(x) 2 log t + 1
= 2 + θ(t) dt. (7.59)
p>x
p 2 x log x x t3 log2 t

Using the middle inequality in the estimate (7.41), and the lower bound
(7.52), we can bound the right side of (7.59) thus
 1 1.000081 1 0.070912
≤ + 2
≤ √ , (7.60)
p>x
p 2 x log x 8x log x x log2 x

where we have used θ(x) < 1.000081x from Lemma 3.11. Substituting this
bound in estimate (7.58), then using estimates (7.57) and (7.60) we establish
the lemma for x ∈ [2 × 104 , 4 × 109 ].
(2) For larger x we make use of the results from Lemmas 7.3, 7.4 and 7.5.
We write
⎛ ⎞⎟
⎜⎜⎜   1 ⎟⎟⎟  1  x dθ(t)
− log ⎜⎜⎝ ⎜ 1 − 2 ⎟⎟⎠ ≥ = √ 2
x2 <p≤x
p √ p2 2x t log t
2x<p≤x
 x  x
dt dS (t)
= √ 2 + √ 2 =: I1 + I2 . (7.61)
2x t log t 2x t log t

Integrating by parts gives us lower bounds on I1 and I2 .

(3) First consider I1 :
 x ! "x  x
dt 1 −dt
√ 2 log t
= − √
− √
− 2
2x t t log t 2x
2
2x t log t
! "x ! "x  x
1 1 dt
= − + + 2 √
.
t log t √2x t log2 t √2x 2 3
2x t log t
182 Robin’s Theorem

Hence
 x
√ √
dt 1 2 1 2 2
√
=− + √ + − √
2
2x t log t x log x x log 2x x log2 x x log2 2x
 x
dt
+2 √ 3
2
2x t log t
√   √
2 log 2 2 2 1
≥ √ 1− − √ 2
− .
x log x log x x log 2x x log x
Therefore since for x > 2 we have
 2
log x log x
+ √ < 1,
log 2x 2 2x
we get
√ √
2 2 (2 + log 2)
I1 ≥ √ − √ . (7.62)
x log x x log2 x
(4) For the second integral:
√
S (x) S ( 2x ) √
I2 ≥ 2 − + L2 ( 2x ) − L2 (x), (7.63)
x log x x log 2x
where  ∞
2 log t + 1
L2 (y) = S (t) dt = J2 (y) − K2 (y). (7.64)
y t3 log2 t
By Lemma 7.5, we have

F 1 ,2 (y) ≤ K2 (y) ≤ F 1 ,2 (y) + 43 F 1 ,2 (y), (7.65)

2 2 3

and so we may bound K2 (y) using the bounds on Fρ,n (x) in Lemma 7.3. Using
Lemma 7.4 to bound J2 (y) we find that
⎛ ⎞
1 ⎜⎜⎜ β + 29 4β 4 ⎟⎟⎟
L2 (y) ≤ 3/2 ⎜
⎝2β + + − ⎟⎠ . (7.66)
y log y log y 3 log2 y 3
It is straightforward to show that the right
√ side of estimate (7.66) is negative
for y ≥ 2. We may similarly bound L2 ( 2x ) from below:
√ √ √ √
L2 ( 2x ) ≥ J2 ( 2x ) − F 1 ,2 ( 2x ) − 43 F 1 ,2 ( 2x ),
2 3

which gives with c1 = 0.0198734

√ c1
L2 ( 2x ) > − √ , x > 4 × 109 . (7.67)
3/4
(2x) log 2x
 
7.4 Bounding p≤x (1 − p−2 ) From Above With RH True 183
√
(5) For 4 × 109 <√ x < 5 × 1021 we have 2x < 1011 , so therefore by
Lemma 3.10 (a), S ( 2x ) < 0. Moreover, for x ≥ 109 , by Theorem 4.6, we
have for c2 = 0.012
c2 x
|S (x)| ≤ . (7.68)
log x
Therefore by (7.59)
c2 c1
I2 ≥ − 2
− √
x log x (2x)3/4 log( 2x )
 
1 c2 c1 × 21/4 log2 x
=− √ √ + 1/4 .
x log2 x x x log(2x)

This shows that

c3
I2 ≥ − √ , (7.69)
x log2 x
where we can take c3 = 0.002666.
(6) When x ≥ 5 × 1021 we use the estimate (7.44) (we are assuming RH),
the bound (7.63) and the equation (7.64) to derive
 3 
1 log x log(2x) log2 x 21/4 log3 x c4
I2 ≥ − √ + √4 + ≥− √ , (7.70)
2
x log x 8πx 162 π x
3/4 8πx 1/4
x log2 x

where we take c4 := 0.02782.

(7) Finally combine the estimates (7.62), (7.69) and (7.70):
⎛ ⎞⎟
⎜⎜⎜   ⎟⎟
−log ⎜⎜⎜⎜⎝ 1 − 2 ⎟⎟⎟⎟⎠ ≥ I1 + I2
1
√ p
2x<p≤x
√ √
2 2 (2 + log 2) max{c3 , c4 }
≥ √ − √ − √
x log x x log2 x x log2 x
√
2 4
≥ √ − √
x log x x log2

and so
   ⎛ √ ⎞
1 ⎜⎜⎜ 2 4 ⎟⎟⎟
1 − 2 ≤ exp ⎜⎝− √ + √ 2 ⎠
⎟,
√ p x log x x log x
2x<p≤x

which finally completes the proof of the lemma. 

184 Robin’s Theorem

7.5 Bounding loglog N From Below With RH True

In this section we bound loglog N from below using RH. Recall we used this
lower bound in the proof of Ramanujan’s theorem, Theorem 7.2. Recall from
Chapter 6: if the positive integer N is colossally abundant with associated
parameter  > 0 then for all m > 1 we have
σ(N) σ(m)
≥ 1+ .
N 1+ m
Define a real number x > 1 as the unique solution to the equation x = 1 + 1/x,
so x depends on N through . Set x1 := x and for k ≥ 1 define a decreasing
sequence (xk ) by
 
1
log xk 1 + k = .
xk + · · · + xk

Lemma 7.9 [142, lemma  7] With the notation defined in the paragraph
above, if x ≥ 20 000 then i≥2 θ(xi ) ≥ 0.998x2 .

Proof To prove the lemma we make use of Lemma 3.12 which gives explicit
bounds of the form
θ(x) > αx (x ≥ a). (7.71)

Let A(x2 ) := θ(x2 ) + i≥3 θ(xi ). By Lemma 6.17 we have (x22 /2)1/k < xk , and so
⎛ ⎞
 ⎜⎜ x2 1/i ⎟⎟
A(x2 ) ≥ θ(x2 ) + ⎜
θ ⎜⎜⎝ 2 ⎟⎟⎟ . (7.72)
i≥3
2 ⎠

Now, let a and b with a < b be two positive real numbers. For y ∈ [a, b] we
have a bound of the form (7.71), so that
⎛ ⎛ 1/i ⎞⎞
 ⎛⎜⎜ a2 1/i ⎞⎟⎟ ⎜⎜⎜ 1  ⎜⎜⎜ a2 ⎟⎟⎟⎟⎟⎟⎟
A(y) ≥ αy + ⎜
θ ⎝⎜ ⎟ ⎜
⎠⎟ ≥ y ⎜⎝α + θ ⎜⎝ ⎟⎠⎟⎠ = yη, (7.73)
3≤i
2 b 3≤i 2

where η = η(α, a, b). The sum over all i ≥ 3 is finite, since the summands in
the expression (7.73) are zero whenever i > log(a2 /2)/log 2. We sum over
all 3 ≤ i ≤ log(a2 /2)/log 2 and use explicit numerical values for the θ(x)
evaluations needed on the right-hand side of (7.72). The range to be covered
needs to include the minimum value of x2 , which is not less than 193.
By choosing α, a and b appropriately we can bound A(y) in regions a ≤ y ≤
b. We do this in Table 7.1 for which we computed the values of η, given a, b
and α. The first four values of a, b and α are from [146, p. 265, corollary]
and the final values are from [145, theorem 10].
The lemma then follows from expression (7.73) and Table 7.1. 
 7.5 Bounding loglog N From Below With RH True 185
Table 7.1 Evaluation of η(α, a, b), given α, a and b.

a b α η

487 381 ∞ 0.998 0.99800

89 387 487 381 0.995 0.99901
32 057 89 387 0.990 1.00131
11 927 32 057 0.985 1.0022
7481 11 927 0.980 1.01433
5381 7481 0.975 1.10206
3457 5381 0.970 1.01896
2657 3457 0.965 1.02935
1481 2657 0.960 1.01876
1433 1481 0.955 1.0537
1427 1433 0.950 1.05201
853 1427 0.945 1.02161
809 853 0.94 1.06316
599 809 0.93 1.03981
557 599 0.92 1.06247
349 557 0.91 1.01651
227 349 0.89 1.02732
149 227 0.86 1.00836

Lemma 7.10 [142, lemma 8] Assume RH is true. If N is colossally abundant,

then for x ≥ 20 000 we have
⎛ √ ⎞
⎜⎜⎜ 0.986 2 0.484 ⎟⎟⎟
loglog N ≥ log θ(x) exp ⎜⎝ √ − √ ⎟⎠ .
x log x x log2 x
Proof From Table 7.1 let c1 = 0.99800. First consider the prime factorization
 
N= pi .
i≥1 xi+1 <p≤xi

Take logarithms and use Lemmas 7.9 and 3.11, which gives θ(x) <
1.000081x, x > 0, to get
 
log N = θ(xi ) = θ(x) + θ(xi )
i≥1 i≥2
≥ θ(x) + c1 x2
 
c1 x2
= θ(x) 1 +
θ(x)
 c1 x2 
≥ θ(x) 1 +
 1.000081x
c2 x2 
≥ θ(x) 1 + ,
x
186 Robin’s Theorem

where we can take c2 = 0.997797. Therefore we get with c4 = 1.00796973

   
c3 x2 c4 x2 x2
loglog N ≥ log θ(x) + > log θ(x) 1 + > log θ(x) 1 + ,
x x log x x log x
where we have used θ(x) < 1.000081x again and log(1 + x) > x/(x + 1), which
is true for x > 0.
Note that for real m, m , b > 0 and t ∈ [0, b], 1+mt ≥ exp(m t) holds provided
m ≤ log(1 + mb)/b. Also since using Lemma 6.17 we get
 
√ log 2 √
2x 1 − < x2 < 2x,
2 log x
we can take
√
x2 2
≤ √ < 0.00101 =: b at x = 20 000.
x log x x log x
This gives, with m = 1, the choice m = 0.999 and so we can then write
  
m x2
loglog N > log θ(x) exp
x log x
⎛ √  ⎞
⎜⎜⎜ m 2 log 2 ⎟⎟⎟
≥ log θ(x) exp ⎜⎝ √ 1− ⎟⎠
x log x 2 log x
⎛ √  ⎞
⎜⎜⎜ 0.986 2 log 2 ⎟⎟⎟
> log θ(x) exp ⎝ √ ⎜ 1− ⎟⎠ ,
x log x 2 log x
which completes the proof of the lemma. 

7.6 Proof of Robin’s Theorem With RH True

Now we are ready to prove Robin’s theorem, Theorem 7.11. First the
statement:
Theorem 7.11 [142, theorem 1] Assume the Riemann hypothesis to be true.
Then
σ(n) < eγ n loglog n (n ≥ 5041). (7.74)
Moreover the values of 1 < n ≤ 5040 for which the inequality in (7.3) fails are
the elements of the set A, where
A := {1, 2, 3, 4, 5, 6, 8, 9, 10, 12, 16, 18, 20, 24, 30, 36, 48,
60, 72, 84, 120, 180, 240, 360, 720, 840, 2520, 5040}. (7.75)
Proof Assume N is colossally abundant. Then, inequality (7.11), which is
part of the proof of Theorem 7.2, is
√
2x<p≤x (1 − 1/p )
2
σ(N)
≤  .
N p≤x (1 − 1/p)
 7.6 Proof of Robin’s Theorem With RH True 187

We now use the results of three of the lemmas which have been proved above,
all valid for x ≥ 20 000 and assuming RH. By Lemma 7.7,
 1
−1 
2+β α(x)

γ
1− ≤ e log θ(x) exp √ + √ ,
p≤x
p x log x x log2 x

where
S 2 (x)(log x + 1.31) 8 + 4β 2 log x 2 log 2π log x
α(x) := 3/2
+ (β − 2) + + 1/6 + ,
2x log x x x1/2
by Lemma 7.8,
   ⎛ √ ⎞
1 ⎜⎜⎜ 2 4 ⎟⎟⎟
1 − 2 ≤ exp ⎜⎝− √ + √ 2 ⎠
⎟,
√ p x log x x log x
2x<p≤x

and by Lemma 7.9,

⎛ √ ⎞
⎜⎜⎜ 0.986 2 0.484 ⎟⎟⎟
loglog N > log θ(x) exp ⎜⎝ √ − √ ⎟⎠ .
x log x x log2 x
These inequalities enable us to write
⎛ ⎞
⎜⎜⎜ c1 + c2 ⎟⎟⎟
σ(N) ⎜
⎜ log x ⎟⎟⎟⎟
≤ eγ exp ⎜⎜⎜⎜⎜ √ ⎟, (7.76)
N loglog N ⎜⎝ x log x ⎟⎟⎟⎠

where
c1 ≤ −0.7642, c2 = α(x) + 4.484, (7.77)
and where α(x) is defined in (7.50) as part of the statement of Lemma 7.7.
We will show that c1 + c2 /log x ≤ 0 for√certain ranges of x set out below.
(1) For x ≥ 100 000 we use |S (x)| ≤ x log2 x/(8π) (Theorem 4.6), which is
valid since we are assuming RH, to bound S (x) and hence α(x) from above.
This shows that α(x) ≤ 2.27146, thus c1 + c2 /log x ≤ −0.175 < 0.
(2) For x ∈ [2 × 104 , 105 ] we use inequality (7.52), i.e. 0 ≤ x − θ(x) ≤
x/(8 log x) (Lemma 3.14 (a)) to bound S (x), so c1 + c2 /log x ≤ −0.055 < 0 in
this case also.
(3) Finally, for x ≤ 20 000, we calculate all of the colossally abundant
numbers Ck with largest prime factor at most x. We do this using the
fact that the first 107 colossally abundant numbers satisfy that the ratio of
any two consecutive numbers is a single prime which can be determined
algorithmically. (A Mathematica implementation of this calculation is on the
page for A004990 in the Online Encyclopedia of Integer Sequences [158].)
Once the next prime greater than 20 000, namely 20 011, occurs (at the
2348th term) we have the complete set of primes which appear in order and
188 Robin’s Theorem

can easily generate the corresponding 2347 colossally abundant numbers; see
Table A.5.
Recall the definition G(n) := σ(n)/(n loglog n). By (1), (2) and (3) above
we have G(Ck ) < eγ for all Ck ≥ 5041. By Theorem 6.16, this shows that the
theorem is true for all n ≥ 5041. Finally, we compute G(n) for all 2 ≤ n ≤ 5040
to derive the set S . This completes the proof. 

7.7 An Unconditional Bound for σ(n)/n

In this section we neither assume that RH is true, nor that RH is false. The
main result, Theorem 7.13, is due to Robin.
Lemma 7.12 For x ≥ 104 we have
 p  
1

γ
≤ e log x 1 + . (7.78)
p≤x
p−1 2 log2 x

Proof When 104 ≤ x ≤ 108 we apply Lemma 5.11 which is:

 p  
2

γ
≤ e log x 1 + √ , 0 < x ≤ 108 . (7.79)
p≤x
p − 1 x log x
√
Because for x ≥ 104 we have log x ≤ x/10, from inequality (7.79) it follows
that inequality (7.78) is true for all x ∈ [104 , 108 ].
When x ≥ 108 using (7.19) we get
1
− loglog x − B
p
      
1 
p≤x
1 1
= −γ − loglog x − log 1 − + − log 1 − − .
p≤x
p p>x
p p

Therefore by Lemmas 5.8 and 5.9 we get

 ⎛ −1 ⎞ ⎛ −1 ⎞    
⎜⎜⎜ 1 ⎟⎟⎟  ⎜⎜⎜ 1 ⎟⎟⎟  1 1
log ⎝⎜ 1 − ⎠⎟ ≤ log ⎝⎜ 1 − ⎟⎠ + − log 1 − −
p≤x
p p≤x
p p>x
p p
 
1
≤ log 1 + + γ + loglog x,
2 log2 x
so, exponentiating,
  
p γ 1
≤ e log x 1 + ,
p≤x
p−1 2 log2 x
which completes the proof. 
We now proceed to derive Robin’s unconditional bound. Note that the
constant c0 is larger than that of Robin.
 7.7 An Unconditional Bound for σ(n)/n 189

Theorem 7.13 [142, theorem 2] Without assuming the Riemann hypothesis

is true or false we have
σ(n) c0
≤ eγ + , n ≥ 3, (7.80)
n loglog n (loglog n)2
where c0 = 2/3.

Proof (1) Let c0 > 0 be a constant to be determined. Let Nk = ki=1 pi be the
kth primorial, and suppose that pk ≥ 20 000, so k ≥ 2263, and so
y := loglog Nk ≥ 9.8942.
By Lemma 3.14 (a) we have
 
1
log Nk = θ(pk ) > pk 1 − . (7.81)
8 log pk
Since pk ≥ 20 000 we have x := log pk ≥ 9.90348, so
 
1 1
loglog Nk ≥ log pk + log 1 − ≥ log pk −
8 log pk 7.899 log pk
1
=⇒ y ≥ x − .
7.899x
By Lemma 7.12 with x replaced by pk , in this step we want
 
Nk γ 1 c0
≤ e log pk 1 + ≤ eγ loglog Nk + .
ϕ(Nk ) 2
2 log pk loglog Nk
Thus we need the minimum positive value of λ = δeγ such that for y ≥ x −
1/(7.899) we have
eγ λ 1 δ
eγ log pk + ≤ eγ loglog Nk + =⇒ x + ≤ y+ .
2 log pk loglog Nk 2x y
Since y + δ/y is increasing we can use the minimum y = x − 1/(7.899x) to get
1 1 δ
x+ − (x − )≤ =⇒ λ ≥ 0.66591,
2x 7.899x x − 7.899x
1

so c0 = 2/3 will work.

(2) Note next that if Nk ≤ n < Nk+1 then at most k distinct primes divide n.
If not and q1 , . . . , ql are those primes with l ≥ k + 1, then since if p1 , . . . , pl are
the first l primes in increasing order we must have for each j, p j ≤ q j , so
Nk+1 = p1 · pk+1 ≤ q1 · · · ql ≤ n,
which is false. Therefore
k    l  
ϕ(Nk )  1 1 ϕ(n)
= 1− ≤ 1− = .
Nk j=1
pj j=1
ql n
190 Robin’s Theorem

It follows that for Nk ≤ n < Nk+1 with pk ≥ 20 000 we have

σ(n) n Nk c0
< ≤ < eγ loglog Nk +
n ϕ(n) ϕ(Nk ) loglog Nk
c0
< eγ loglog n + . (7.82)
loglog n
When pk < 20 000, that is, when n ≤ N2263 , we generated the first 2377
colossally abundant numbers, in the same manner as in Theorem 7.11, and in
this way checked Robin’s inequality, and therefore the unconditional bound,
for 5041 ≤ n ≤ N2263 . For 3 ≤ n ≤ 5040 we evaluated the terms of the inequality
explicitly to obtain 0.648214, so c0 is the minimum value of the constant
computed by this method. This completes the proof. 

7.8 Bounding loglog N From Above Without RH

Before proving Theorem 7.15 we must first bound loglog N from above
without using RH.
Lemma 7.14 When N is colossally abundant we have
⎛ √  ⎞
⎜⎜⎜ 2 1 ⎟⎟⎟
loglog N ≤ log θ(x) ⎜⎝1 + √ +O √ 2
⎟⎠ .
x log x x log x
Proof By the proof of Lemma 9.13 we have

∞
log N ≤ θ(xi )
i=1
≤ θ(x1 ) + θ(x2 ) + O(θ(x3 ) log x)
  
θ(x2 ) x3 log x
= θ(x) 1 + +O .
θ(x) x
√
By Lemma 6.17 we have x2 = 2x (1 + O(1/log x)) and by Lemma 6.18,
x3 ∼ (3x)1/3 . In addition the prime number theorem gives θ(x) = x+O(x/log x)
as x → ∞. Therefore we can derive
⎛ √    ⎞
⎜⎜⎜ 2 1 log x ⎟⎟⎟
log N ≤ θ(x) ⎜⎝1 + √ + O √ + O 2/3 ⎟⎠
x x log x x
⎛ √  ⎞
⎜⎜ 2 1 ⎟⎟⎟
= θ(x) ⎜⎝⎜1 + √ + O √ ⎠⎟ .
x x log x
Therefore, since by the mean value theorem we have log θ(x) = log x +
O(1/log x), we get
⎛ √  ⎞
⎜⎜⎜ 2 1 ⎟⎟⎟
⎜⎜⎜ √ +O √ ⎟⎟
⎜⎜⎜ x x log x ⎟⎟⎟⎟
loglog N ≤ log θ(x) ⎜⎜⎜1 +   ⎟⎟⎟
⎜⎜⎜ 1 ⎟⎟⎟
⎝⎜ log x + O ⎠⎟
log x
 7.9 A Lower Bound for σ(n)/n With RH False 191
⎛ √      ⎞
⎜⎜⎜ 2 1 1 ⎟⎟⎟
= log θ(x) ⎜⎝1 + √ 1+O 1+O 2
⎟⎠
x log x log x log x
⎛ √  ⎞
⎜⎜⎜ 2 1 ⎟⎟⎟
≤ log θ(x) ⎜⎝1 + √ +O √ 2
⎟⎠ ,
x log x x log x
and the proof is complete. 

7.9 A Lower Bound for σ(n)/n With RH False

The theorem of this section is a vital ingredient for proving Robin’s theorem.
It relies heavily on Nicolas’ theorem, Theorem 5.29, giving an omega
estimate for the logarithm of f (x).
Theorem 7.15 [142, theorem 1] Assume the Riemann hypothesis is false.
Let θ be the supremum over all real parts of the non-trivial zeros ρ = β + iγ,
and thus θ > 12 . For any number b ∈ (1 − θ, 12 ), there exists a positive constant
c such that  
σ(n) γ c
> e 1+ , (7.83)
n loglog n (log n)b
for infinitely many values of n.
Proof First write, for N colossally abundant, following Lemma 6.9,
 −1  
σ(N)  1   1
= 1− 1 − k+1 , (7.84)
N p≤x
p k≥1 x <p≤x
p
k+1 k

where, by Lemma 6.17, we have xk+1 > x1/(k+1) . We can rewrite the right-hand
side of (7.84) as
 1
−1  
1

1− 1 − 2 E(x), (7.85)
p≤x
p x <p≤x p
2

where E(x) satisfies

  1

1
π(x3 )
1− 3 1− < E(x) < 1. (7.86)
x3 <p≤x2
p x

By Lemma 5.7 with α = 3 we have

⎛  ⎞⎟
 1
−1   
1
 ⎜⎜⎜ 1 1 ⎟⎟
1− 3 = exp − log 1 − 3 = exp ⎜⎜⎝ ⎜ + O 5 ⎟⎟⎟⎠ ,
p p p 3 x
p>x p>x p>x

so for some constant η > 0 we have

  ⎛
⎜⎜⎜  1  ⎞⎟   
1 ⎜ 1 ⎟⎟⎟ 1
1 − 3 ≥ exp ⎜⎜⎝−⎜ ⎟
+ O 5/3 ⎟⎟⎠ = exp O 2/3 .
p>x
p 1/3
p3 x x log x
3 p>ηx
192 Robin’s Theorem

In addition we have using Chebyshev’s prime number theorem and also

log(1 − 1/x) > −2/x:
 π(x )   
1 3 1
1− = exp π(x) log 1 −
x x
 1/3
   
x 2 1
> exp −η = exp O 2/3 .
log x x x log x
Therefore
    
1 1
E(x) > exp O √ = 1+O √ . (7.87)
x log2 x x log2 x
The estimate (7.12) of Theorem 7.2 is
   ⎛ √  ⎞
1 ⎜⎜⎜ 2 1 ⎟⎟⎟
1 − 2 = exp ⎜⎝− √ +O √ 2
⎟⎠ .
√ p x log x x log x
2x<p≤x

Therefore using (7.84), the bounds (7.86) and the estimate (7.87) we get
 −1     
σ(N)  1 1 1
≥ 1− 1− 2 1+O √ .
N p≤x
p √
2x<p≤x
p x log2 x

Next recall the definition of f (x) originally introduced by Nicolas and

employed as the key element in Chapter 5: for x > 0
 1

γ
f (x) := e log θ(x) 1− .
p≤x
p

Nicolas, in Theorem 5.29, proved that if the Riemann hypothesis is false there
is a b with 0 < b < 12 such that
⎛ ⎞
⎜⎜⎜   1
 ⎟⎟⎟
log f (x) = log ⎜⎜⎜⎝eγ 1 − log θ(x)⎟⎟⎟⎠ = Ω± (x−b ). (7.88)
p≤x
p

Therefore, using Lemma 7.14, we have

⎛ √  ⎞
⎜⎜⎜ − 2 1 ⎟⎟⎟
⎜
exp ⎝ √ +O √ ⎟⎠
σ(N) x log x x log2 x
≥ ⎛ √ ⎞
eγ N loglog N  1

⎜⎜⎜ 2

1 ⎟⎟⎟
e γ 1 − log θ(x) ⎝⎜1 + √ +O √ 2
⎟⎠
p≤x
p x log x x log x
⎛ √  ⎞
⎜⎜⎜ − 2 1 ⎟⎟⎟
exp ⎝⎜ √ +O √ 2 ⎠⎟
x log x x log x
≥   
1
exp Ω− b
x
 7.10 Lagarias’ Formulation of Robin’s Criterion 193
    
1 1
= exp Ω+ ≥ 1 + Ω+ .
xb xb
We also know, from Chapter 6, that if p | N is the largest prime divisor we
have
 p x
log N > log q > > ,
q≤p
2 4
√
since 2x < p ≤ x by the prime number theorem and Bertrand’s postulate.
Therefore there is a constant c > 0 such that
σ(N) c
γ
≥ 1+ .
e N loglog N (log N)b
Hence Robin’s inequality is false for an infinite number of N and the proof
of the theorem is complete. 
Finally in this section, combining Theorems 7.11 and 7.15 we get:
Theorem 7.16 (Ramanujan–Robin criterion) A necessary and sufficient
condition for the Riemann hypothesis is that
σ(n) < eγ n loglog n (n ≥ 5041). (7.89)

7.10 Lagarias’ Formulation of Robin’s Criterion

Jeffrey Lagarias (Figure 7.7) graduated from MIT in 1974 with a thesis
entitled “The 4-part of the class group of a quadratic field”. His adviser was
Harold Starke, whose adviser was Derrick Lehmer. Lagarias has taught since
2002 at the University of Michigan. In 1987 he received a Lester R. Ford
award from the Mathematical Association of America and in 2012 he became
a fellow of the American Mathematical Society.
Lagarias [100] considered another inequality involving the sum-of-divisors
function. His inequality does not directly involve Euler’s constant γ or a
number such as 5040. Recall the definition of the nth harmonic number:
1 1 1
Hn := 1 + + + · · · + (n ≥ 1).
2 3 n
To begin, in lemmas 3.1 and 3.2 of [100] sharp bounds on the quantity on
the right-hand side of inequality (7.92), namely Hn + exp(Hn ) log(Hn ), are
produced. We summarize this in the following:
Lemma 7.17 For n ≥ 3 we have
n
eγ n loglog n+ Hn ≤ Hn +exp(Hn ) log(Hn ) ≤ eγ n loglog n+ h(n), (7.90)
log n
where h(n) is a positive function decreasing on [4, ∞) to eγ , and 3 < h(4) < 4.
194 Robin’s Theorem

Figure 7.7 Jeffrey Lagarias.

Proof First consider the left inequality

 in (7.90). We use Abel’s identity [6,
theorem 4.2] applied to the sum Hn = nj=1 j−1 to obtain
 n
{t}
Hn = 1 + log n − 2
dt,
1 t

where {t} denotes the fractional part of t. Hence Hn < 1 + log n.

 ∞ 0 ≤ {t} < 1, the integral converges and so, upon writing γ =
Since
1 − 1 ({t}/t2 ) dt we have
 ∞
{t}
Hn = log n + γ + dt, (7.91)
n t2
thus Hn > log n + γ, so log Hn > loglog n and exp(Hn ) > eγ n, and therefore the
left inequality in (7.90) follows.
Now let
 n+1  
1 1
Rn := Hn − log(n + 1) = − dt → γ,
1 t t
so for all n ∈ N we have Rn+1 > Rn > 0. Therefore Hn − log(n + 1) < γ so Hn <
γ + log(n + 1). Exponentiating we get exp(Hn ) < eγ (1 + n). Taking logarithms
of the equation Hn < 1 + log n we have
  
1 1
log Hn < log(1 + log n) = log log n 1 + < loglog n + .
log n log n
Therefore
 
γ 1
Hn + log(Hn ) exp(Hn ) < e (n + 1) loglog n + + 1 + log n
log n
 7.10 Lagarias’ Formulation of Robin’s Criterion 195
 
1 n
= neγ loglog n + eγ loglog n + + + log en
log n log n
n
= neγ loglog n + h(n),
log n
where we have set
log n log2 n γ log n loglog n eγ
h(n) := eγ ++ +e + .
n n n n
The maximum of h(n) over the positive integers is at n = 4, with 3 <
h(4) < 4. 
Note that lemma 3.2 in [100] has a 4 instead of h(n) in the right-hand side
of (7.90), but otherwise the results are equivalent.
Now that bounds involving eγ n loglog n have been converted to bounds
involving Hn , we are in a position to prove Theorem 7.18. Most of the heavy
lifting has been done by Theorems 7.11, 7.13 and 7.15. Lagarias proves:

Theorem 7.18 (Lagarias’ criterion) [100, theorem 1.1] Let Hn = nj=1 j−1 for
n ≥ 1 denote the nth harmonic number. The inequality
σ(n) ≤ Hn + exp(Hn ) log(Hn ), n ≥ 1, (7.92)
is equivalent to the Riemann hypothesis.
Proof First, assume the Riemann hypothesis. Then by Theorem 7.11 and
Lemma 7.17,
σ(n) ≤ eγ n loglog n ≤ exp(Hn ) log(Hn ), n ≥ 5041, (7.93)
so that inequality (7.92) is satisfied for n ≥ 5041. One can check easily on a
computer that it is also true for 1 ≤ n ≤ 5040.
Now assume RH is false and that inequality (7.92) is true for all n ∈ N in
order that we might obtain a contradiction. By Theorem 7.15 there is a b with
0 < b < 12 such that for some c > 0 and an infinite set of n ∈ N, we have
c σ(n)
1+ b
< γ .
log n ne loglog n
Thus by the right-hand side of the inequality of Lemma 7.17 we get
 
γ neγ loglog n γ c
ne loglog n + c = ne loglog n 1 +
logb n logb n
< σ(n) ≤ Hn + exp(Hn ) log(Hn )
4n
≤ neγ loglog n + .
log n
Thus ceγ loglog n ≤ 4/log1−b n for an infinite set of n, which is false, and the
proof of the theorem is complete. 
196 Robin’s Theorem
σ(n)
Hn+) Hn log(Hn)

1.0

0.8

0.6

0.4

0.2

n
200 400 600 800 1000

Figure 7.8 The values of σ(n)/(Hn + exp(Hn ) log(Hn )) for 2 ≤ n ≤ 1000.

Note that, in the expression (7.93), we did not require the addition of the
term Hn . It was pointed out by Kaneko [100, p. 542] that one could instead
prove that σ(n) ≤ exp(Hn ) log(Hn ) for n > 60 is equivalent to the Riemann
hypothesis.
In Figure 7.8 we see how close the left side gets to the right side of the
inequality (7.92), even for small values of n.

7.11 Unconditional Results for Lagarias’ Formulation

In [101] Lagarias asked whether one could show that, unconditionally,
σ(n) ≤ Hn + 2 exp(Hn ) log(Hn ), n ≥ 1, (7.94)
with equality only for n = 1.
In [98] this question is answered in the affirmative. A result of Ivić [81]
(see also the improved result of Trudgian [170]) is that
σ(n)
< 2.59 loglog n, n ≥ 3. (7.95)
n
Using this inequality and the expression (7.90) we see that
2.59
σ(n) < exp(Hn ) log(Hn ) < 1.455 exp(Hn ) log(Hn ), n ≥ 3.
eγ
Checking the cases 1 ≤ n ≤ 2 by hand establishes
σ(n) ≤ Hn + 1.455 exp(Hn ) log(Hn ), n ≥ 1.
 7.12 Unitary Divisor Sums 197

One could also use a result of Robin [142, section 6, proposition 2 (1)], which
improves the right-hand side of inequality (7.95) to 2.57n loglog n for n ≥ 7.
There is an editorial comment in [102] that includes an observation by
the GCHQ Problem Solving Group. They contest that one could replace the
number 2 in inequality (7.94) by any constant K > 1 at the cost of having to
check more cases by hand when K is close to one. An example is cited: when
K = 1.2 one needs to verify inequality (7.94) for 1 ≤ n ≤ 106 .
Further advances have since obviated the need for such a large check.
Akbary, Friggstad and Juricevic [1] showed that
σ(n) σ(180)
< loglog n ≤ 1.0339eγ loglog n (n ≥ 121).
n 180 loglog 180
This was improved by Akbary and Friggstad [2, p. 3] by considering
superabundant numbers – see also Lemma 8.9:

σ(n)
≤ 1.013617eγ loglog n (n ≥ 5041).
n
This, and a computer check for 1 ≤ n ≤ 5040, establishes the unconditional
inequality given in Theorem 7.19.
Theorem 7.19 Without any hypotheses
σ(n) ≤ Hn + 1.013617 exp(Hn ) log(Hn ),
for all n ≥ 1.
One could reduce the size of the coefficient of exp(Hn ) log(Hn ) by
testing Robin’s inequality against more superabundant numbers. Given the
unconditional nature of Theorem 7.19, we note that if Theorem 7.18 is false,
it cannot be false by much. In this sense Theorems 7.19 and 7.18 complement
Theorems 7.11 and 7.13.

7.12 Unitary Divisor Sums

We say that d is a unitary divisor of n if d | n and (d, n/d) = 1. Let

σ∗ (n) = d
d|n
(d,n/d)=1

be the sum of all unitary divisors of n.

Robin [142, p. 210] notes that the proof of Grönwall’s result, equation
(7.1), can be adapted to show that
σ∗ (n) 6eγ
lim sup = 2 = 1.08 . . . .
n→∞ n loglog n π
198 Robin’s Theorem

Ivić [81] showed that

28
σ∗ (n) <
n loglog n (n ≥ 31).
15
This was improved by Robin, who showed that
σ∗ (n) < 1.63601n loglog n (n ≥ 31), (7.96)
except for n = 42 when σ∗ (n) = 1.7366 . . . n loglog n. Trudgian [170, theorem
1.1] improved on inequality (7.96), at least for large n, showing
σ∗ (n) ≤ 1.3007n loglog n (n ≥ 570 571). (7.97)
The method in [170] is incapable of reducing the right-hand side of inequality
(7.97) to anything less than 1.298n loglog n, even if n is taken to be
sufficiently large. Robin conjectured [142, proposition 1 (i), p. 210] that there
are infinitely many n for which
6eγ
σ∗ (n) > n loglog n. (7.98)
π2
A related conjecture was given in proposition 1 (ii) in [142], viz. that
σ(n)
< eγ ,
σ∗ (n) loglog n
for all n sufficiently large. This has been proved by Derbal [45].

7.13 Unsolved Problems

(1) Improve the power of loglog n on the right-hand side of inequality (7.80).
The right-hand side of (7.80) in Theorem 7.13 cannot be improved for all
n. However, for n sufficiently large, yet still explicit, one may certainly
improve both on the constant C0 = 2/3 and, more importantly, on the
power of loglog n in the denominator. Results of the form |θ(x) − x| ≤
ck x/logk x, for a positive constant ck and for k ≥ 2, could be helpful in
this regard. Such results are known – see e.g. [153, theorem 8*] and [173,
theorem 2].
(2) Can equation (7.97) be improved?
(3) Does equation (7.98) hold for infinitely many values of n?
(4) Can one determine an equivalence condition for the Riemann hypothesis
involving σ∗ (n)?
(5) Find an inequality or other equivalence to RH in terms of the number-of-
divisors function d(n). You may need to use the Dirichlet series

∞
d(n)
= ζ(s)2 (s > 1),
n=1
ns
 7.13 Unsolved Problems 199

or Mellin transform/Perron’s formula

  c+i∞
1 xz
d(n) = ζ(z)2 dz (c > 1).
n≤x
2πi c−i∞ z

(6) Attempt to show that RH is equivalent to the error term in Dirichlet’s

divisor problem

d(n) = x log x + x(2γ − 1) + O(xα+ ),
n≤x

taking the value α = 14 + ; but see [71] and [73].

(7) For N ≥ 2, let g(N) denote the largest value of
σ(n) − Hn
J(n) :=
exp(Hn ) log(Hn )
for 2 ≤ n ≤ N. According to Theorem 7.19 we have g(N) ≤ 1.0137 always;
according to Theorem 7.18, which assumes RH, we have g(N) ≤ 1 for all
N. When do we have large values of g(N)? We have g(10 000) = J(4) =
0.9873. What is the next n > 10 000 such that J(n) > 0.9873?
 8
Numbers That Do Not Satisfy Robin’s Inequality

8.1 Introduction
Recall the theorem of Robin in Chapter 7, Theorem 7.11, that the Riemann
hypothesis is equivalent to Robin’s inequality holding for all n > 5040:

σ(n)
< eγ loglog n.
n

This is the Ramanujan–Robin criterion. Here methods are developed to

derive properties of numbers which do or do not satisfy the inequality. For
example, all 11-free numbers larger than 5040 satisfy it. We begin with
the joint work of four authors from 2007 [38], YoungJu Choie from Korea
(Figure 8.1), Nicolas Lichiardopol from France, Pieter Moree from Germany
and Patrick Solé, also from France. We also include related work of Solé
and Michael Planat [159] from 2011, and conclude with the work of Tim
Trudgian and the present author.
The following notations for sets of integers will be used in this chapter. Let
the set of natural numbers satisfying Robin’s inequality be denoted R. Let

A := {1, 2, 3, 4, 5, 6, 8, 9, 10, 12, 16, 18, 20, 24, 30, 36, 48,
60, 72, 84, 120, 180, 240, 360, 720, 840, 2520, 5040},
B := {2, 3, 5, 6, 10, 30},
C := {1, 4, 8, 9, 16, 36},
D := {4, 8, 9, 16, 36, 72, 108, 144, 216, 900, 1800, 2700, 3600,
44 100, 88 200}.

Thus, by Theorem 7.11, A ∩ R = ∅. In addition B ∪ C ⊂ A and B ∩ C = ∅.

After some preliminaries it is shown that all squarefree integers which are
not in the set B satisfy Robin’s inequality. Then that all odd integers other

200
 8.1 Introduction 201

Figure 8.1 YoungJu Choie.

than {1, 3, 5, 9, 15} satisfy

n
< eγ loglog n,
ϕ(n)
which is called Nicolas’ inequality. Unfortunately it is the negation of the
Nicolas’ inequality that appears in Chapter 5 in equivalences to RH involving
ϕ(n). Observe that Nicolas’ inequality is stronger than that of Robin, since
σ(n)/n ≤ n/ϕ(n). The next result uses Hardy–Ramanujan numbers, which
are defined below. If Robin’s inequality is satisfied by all Hardy–Ramanujan
integers in a range, it is satisfied by all integers in that range. These results are
summarized in Theorem 8.10, namely any integer greater than 5040 and not
satisfying Robin’s inequality must be even, not squarefree and not squarefull.
The smallest such integer, if it exists, must also be a Hardy–Ramanujan
number and a superabundant number.
Following this work, successive improvements are given which show that
counterexamples to Robin’s inequality which are greater than 5040 must be
divisible by a high power of at least one prime: first the theorem from [38]
that all 5-free integers greater than 5040 satisfy Robin’s inequality, then from
[159] extending this to 7-free integers, and finally a work of Tim Trudgian
and the author extending this to 11-free integers.
To summarize this chapter: Section 8.2 discusses Hardy–Ramanujan
numbers and finishes with Theorem 8.10 discussed above. Then in each of
202 Numbers That Do Not Satisfy Robin’s Inequality

the three successive sections, Sections 8.3, 8.4 and 8.5, the given cases of
r-free integers are proved.
It is somewhat confusing that references to “Robin’s inequality” sometimes
implicitly include the lower bound n > 5040. In this volume by Robin’s
inequality we always mean
σ(n)
< eγ loglog n,
n
and any lower bound is given explicitly.
The following RHpack (see Appendix B) variables and functions relate
to the material in this chapter: SetA, SetB, SetC, SetD, HardyRamanujanQ,
NthPrimeBounds, LogKthPrimorialBounds, KthApproximatePrimorial, Pri-
morialFormToInteger and HardyRamanujanToPrimorialForm.

8.2 Hardy–Ramanujan Numbers

We say as usual that an integer n is squarefree if, for all primes p, p2  n.
We say it is squarefull if p | n implies p2 | n. Finally we say it is a Hardy–
Ramanujan number if it has the form for its standard factorization into
primes:
n = pα1 1 · · · pαmm ,
where p1 = 2, p2 = 3, etc. and, for all i with 1 ≤ i ≤ m, pi is the ith prime, and
the exponents are non-increasing, that is to say, αi ≥ αi+1 ≥ 1 for 1 ≤ i ≤ m − 1.
Lemma 8.1 [38, lemma 2.1(2)] For all x ≥ 4
1
≤ loglog x + γ,
p≤x
p
where the sum is over primes p and γ is Euler’s constant.
Proof By Lemma 5.11 we can write for 1 < x
1 1
≤ loglog x + B + ,
p≤x
p 2 log2 x
where B < 0.261498 is Mertens’ constant. Finally note that at x = 4 we have
1
B+ <γ
2 log2 x
to complete the proof. 
Lemma 8.2 [38, lemma 2.2 and corollary 2.1] If a ∈ A and p is a prime
number with p ≥ 7, then other than the case p = 7 and a ∈ {12, 120, 360}, we
have that ap satisfies Robin’s inequality. Thus if a ∈ B and p ≥ 7 is prime, we
have ap ∈ R.
 8.2 Hardy–Ramanujan Numbers 203

Proof Recall the definitions

A := {1, 2, 3, 4, 5, 6, 8, 9, 10, 12, 16, 18, 20, 24, 30, 36, 48,
60, 72, 84, 120, 180, 240, 360, 720, 840, 2520, 5040},
B := {2, 3, 5, 6, 10, 30}.
If p = 7 then checking numerically, for a ∈ {12, 120, 360}, ap does not satisfy
Robin’s inequality. Also numerically, for each a ∈ A, 11a satisfies Robin’s
inequality. Hence, if p ≥ 11 and a ∈ A we get
 
σ(ap) 1 σ(a) 12σ(a) σ(11a)
= 1+ ≤ =
ap p a 11a 11a
< eγ loglog(11a) ≤ eγ loglog(ap),
so ap satisfies Robin’s inequality. Since B ⊂ A \ {12, 120, 360}, ap satisfies
the inequality for all a ∈ B and p ≥ 7. 
We say a positive integer n satisfies Nicolas’ inequality if
n
< eγ loglog n.
ϕ(n)
Considering what we saw in Chapter 5, this definition is somewhat unnatural,
since in Nicolas’ first theorem the inequality is reversed to obtain an RH
equivalence. However, we will use it since it is traditional.
Now let S := {3a ·5b · pc : p ≥ 7 is prime, a, b, c are non-negative integers}, so
S = {1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, . . .}.

Lemma 8.3 [38, lemma 2.3] All elements of S other than {1, 3, 5, 9} satisfy
Robin’s inequality and all elements other than {1, 3, 5, 9, 15} satisfy Nicolas’
inequality.
Proof For p ≥ 7 and n ∈ S ∩ [31, ∞) we can write
σ(n) n 3 5 p 3 5 7
≤ ≤ · · ≤ · · = 2.1875 < eγ loglog 31 ≤ eγ loglog n.
n ϕ(n) 2 4 p − 1 2 4 6
The remaining elements of S can be checked numerically. 
Lemma 8.4 [38, theorem 1.1] All squarefree integers other than the set B,
satisfy Robin’s inequality.
Proof (1) Begin by noting that no element of B satisfies Robin’s inequality.
If p ≥ 7 is prime then p + 1 < 8p/7 < peγ loglog p, so all such primes satisfy
Robin’s inequality.
(2) We now proceed using induction. If m := ω(n) we have just shown the
lemma is true for m = 1. Assume it is true for all squarefree integers n with
ω(n) = m − 1 for some m ≥ 2 and let n = q1 · · · qm be an integer not in B and
with primes qi satisfying q1 < q2 < · · · < qm .
204 Numbers That Do Not Satisfy Robin’s Inequality

(3) Suppose first that qm ≥ log n. If, on the one hand, q1 · · · qm−1 ∈ B and
qm  B, then by Lemma 8.2, n = (q1 · · · qm−1 )qm satisfies Robin’s inequality.
If, however, qm ∈ B and n  B we must have n = 15, which can be shown to
satisfy Robin’s inequality directly.
If, on the other hand, q1 · · · qm−1  B, we must have, by the inductive
assumption,
(q1 + 1)(· · ·)(qm−1 + 1) < eγ q1 · · · qm−1 loglog(q1 · · · qm−1 ),
and so
(q1 + 1)(· · ·)(qm + 1) < eγ q1 · · · qm−1 (qm + 1) loglog(q1 · · · qm−1 ). (8.1)
Note that if 0 < a < b we have

log b − log a 1 b
dt 1
= > .
b−a b−a a t b
Therefore setting a = log(q1 · · · qm−1 ) and b = log(q1 · · · qm ) and using the
assumption qm ≥ log(n), we can write
loglog n − loglog(n/qm ) loglog(q1 · · · qm ) − loglog(q1 · · · qm−1 )
=
log qm log(q1 · · · qm ) − log(q1 · · · qm−1 )
1
>
log(q1 · · · qm )
loglog(q1 · · · qm−1 )
≥ ,
qm log qm
so that
(qm + 1) loglog(q1 · · · qm−1 ) ≤ qm loglog(q1 · · · qm ).
Therefore by (8.1)
(q1 + 1)(· · ·)(qm + 1) < eγ q1 · · · qm loglog(q1 · · · qm ),
so n ∈ R and the result of the lemma follows by induction.
(4) Suppose now it is the case that qm < log n. Recall we can assume m ≥ 2.
If qm = 3 then n = 6 but log 6 < 3. If qm = 5 then n ∈ {10, 15}, but log 10 < 5
and log 15 < 5. Hence we can assume qm ≥ 7. Note that for a > 0
 a+1
dt 1
log(a + 1) − log(a) = < ,
a t a
so, by Lemma 8.1, we can write
m m
1 1
log(qi + 1) − log(qi ) < ≤ ≤ γ + loglog qm
i=1
q p≤q p
i=1 i m

< γ + logloglog(q1 · · · qm ).
Taking exponentials we get σ(n)/n < eγ loglog n, which completes the
induction step in this case also. 
 8.2 Hardy–Ramanujan Numbers 205

Define the squarefree core γ(n) of n to be the largest squarefree integer

which divides n, i.e.

γ(n) := p.
p|n

Lemma 8.5 [38, theorem 2.1] All odd integers other than the set
{1, 3, 5, 9, 15}, satisfy Nicolas’ inequality.
Proof Let m := ω(n) ≥ 1. If m = 1 then, by Lemma 8.3, n satisfies Nicolas’
inequality. If m ≥ 2 then n/ϕ(n) = γ(n)/ϕ(γ(n)). Therefore if r is a squarefree
number satisfying Nicolas’ inequality so does every integer n with γ(n) = r.
Now let n = q1 · · · qm be squarefree with the qi prime and 3 ≤ q1 < q2 < · · · <
qm . In this case n satisfies Nicolas’ inequality if and only if

m
1  qi
m
= < eγ loglog(q1 · · · qm ).
i=1
1 − 1/qi i=1 qi − 1
Because q1 /(q1 −1) ≤ 3/2 and for 2 ≤ i ≤ m, qi /(qi −1) < (qi−1 +1)/qi−1 , setting
n1 := 2n/q1 gives
3  qi + 1 σ(n1 )
m−1
n
< = .
ϕ(n) 2 i=1 qi n1
It follows that since n1 is squarefree, by Lemma 8.4, other than elements of
B, n1 satisfies Robin’s inequality, so we have
n σ(n1 )
< < eγ loglog n1 < eγ loglog n.
ϕ(n) n1
Therefore n would satisfy Nicolas’ inequality. Elements of B with m ≥ 2 can
be checked numerically.
If however m = 1 then n would be a prime power so would be in S . The
result then follows by Lemma 8.3. 

Recall the definition θ(x) := p≤x log p for x ≥ 1. Recall that by
Lemma 3.10 (d), for x ≥ 121 we have θ(x) ≥ 0.8x. Recall also that by
Lemma 5.12 for x ≥ 286 we have
 p  
γ 1
≤ e log x 1 + .
p≤x
p−1 2 log2 x
These estimates enable us to derive the following upper bound for the left-
hand side of that expression in terms of the product of successive primes:
Lemma 8.6 [38, lemma 3.2] For all m > 4 we have
m
pi
≤ eγ log(2 log(p1 · · · pm )),
i=1
p i − 1
206 Numbers That Do Not Satisfy Robin’s Inequality

where as usual pi is the ith prime.

Proof If 4 < m < 62 a numerical check shows that the inequality is true, so
we can assume m ≥ 62. Note that p62 = 293. By Lemma 3.10 (d) we can write
log(p1 · · · pm ) = θ(pm ) > 0.8pm .
Therefore
1
log(2 log(p1 · · · pm )) > log pm + log 1.6 ≥ log pm + .
log pm
Hence, using Lemma 5.12 with x = pm , we get
m
pi
≤ eγ log(2 log(p1 · · · pm )),
i=1
p i − 1
and the proof is complete. 
Recall the definition
D = {4, 8, 9, 16, 36, 72, 108, 144, 216, 900, 1800, 2700, 3600, 44 100, 88 200},
so all elements of D are squarefull.
Lemma 8.7 [38, theorem 3.1] No element of D satisfies Nicolas’ inequality.
Every squarefull integer not in D satisfies Nicolas’ inequality, and thus
Robin’s inequality.
Proof Let n = qα1 1 · · · qαmm be a squarefull integer, with q1 < q2 < · · · < qm and
each q j prime, and not satisfy Nicolas’ inequality. Then if pi is the ith prime

m
pi  qi m
n
≥ = ≥ eγ loglog n ≥ eγ log(2 log(p1 · · · pm )).
i=1
pi − 1 i=1 qi − 1 ϕ(n)
By Lemma 8.6 we must therefore have m ≤ 4. Hence
35 2 3 5 7 n
= · · · ≥ ≥ eγ loglog n.
8 1 2 4 6 ϕ(n)
This implies n ≤ 116 144. There are 45 535 squarefull integers in this range.
Each can be checked numerically to complete the proof. 
Let n = qe11 · · · qkk be the standard prime factorization of the positive integer
e

n, where in this case the primes qi are ordered so that e1 ≥ e2 ≥ · · · ≥ ek .

Let ē = (e1 , . . . , ek ) and call ē the exponent pattern of n. For example since
50 = 52 × 21 , ē = (2, 1). Define a positive integer

k
m(ē) := pei i ,
i=1

where as usual pi is the ith prime, so m(ē) has exponent pattern ē.
 8.2 Hardy–Ramanujan Numbers 207

Lemma 8.8 [38, proposition 5.1] If Robin’s inequality holds for all Hardy–
Ramanujan integers n with 5041 ≤ n ≤ x it holds for all integers in this same
range.
Proof (1) First note that for given e > f > 0 and x ∈ (1, ∞) the function
1 − x−e
g(x) :=
1 − x− f
is strictly decreasing. To see this, observe that the numerator of the derivative
is ex f − f xe + f − e and its derivative is 0 at x = 1 and is negative for all x > 1.
(2) It follows from (1) that if p, q, f and e are positive integers with p, q
prime, p < q and e < f then
  
σ(p f qe ) 1 − p− f −1 1 − q−e−1
=
p f qe 1 − p−1 1 − q−1
 −e−1
 
1− p 1 − q− f −1 σ(pe q f )
> = .
1 − p−1 1 − q−1 pe q f
(3) For primes p and q with p < q we have σ(pe )/pe > σ(qe )/qe . Therefore
if the exponent pattern is fixed, the maximum value of the ratio σ(n)/n will
be assumed when the primes are as small as possible, i.e. for n = m(ē) and if
n1 = pβ11 · · · pβmm is any other number for which the maximum is assumed, by
(2), we must have β1 ≥ · · · ≥ βm . Thus
 
σ(m(ē)) σ(n)
= max : n has factorization pattern ē .
m(ē) n
(4) This shows that if σ(n)/n ≥ eγ loglog n, then the same inequality holds
with n replaced by m(ē).
(5) Now we claim that if Robin’s inequality is true for the integer m(ē),
then it is true for every positive integer n having exponent pattern ē. To see
this, let n be such that it has exponent pattern ē and m(ē) ≤ 5040. Assume
also that n ≥ 5041 and that it does not obey Robin’s inequality. Since the
maximum number of prime factors of any number less than 5041 is 5, we
can write using (3)
σ(n) σ(m(ē))  1 − p−6
eγ loglog n ≤ ≤ ≤ < 5,
n m(ē) p≤11
1 − p−1
so log(5041) ≤ log n < 14. Checking n values in this range numerically shows
that they satisfy Robin’s inequality, a contradiction, completing the proof of
the claim. Thus if n > 5040 and m(ē) ≤ 5040 then n obeys Robin’s inequality.
If however n > 5040 and m(ē) > 5040, since m(ē) is a Hardy–Ramanujan
number it must, by the hypothesis of the lemma, satisfy Robin’s inequality.
Hence, because of (3) and n ≥ m(ē),
σ(n) σ(m(ē))
≤ < eγ loglog(m(ē)) ≤ eγ loglog n.
n m(ē)
208 Numbers That Do Not Satisfy Robin’s Inequality

Therefore if Robin’s inequality holds for all Hardy–Ramanujan integers with

5041 ≤ n ≤ x it holds for all integers in this same range, which completes the
proof. 
Recall that a natural number n is superabundant if for all 1 ≤ m < n we
have σ(m)/m < σ(n)/n. For example, a simple computation shows that
{1, 2, 4, 6, 12, 24, 36, 48, 60, 120, 180, 240}
is the set of the first 12 superabundant numbers; see Table A.3.
Lemma 8.9 [2, theorem 3] The smallest integer n > 5040 which does not
satisfy Robin’s inequality must be a superabundant number.
Proof A computation shows all integers n in the range 5040 < n ≤ 10 080
satisfy Robin’s inequality, and that 10 080 is a superabundant number. In
order that we might get a contradiction, assume there is an integer n > 10 080
for which Robin’s inequality fails, and call n the smallest such integer.
Assume that n is not superabundant. Then there is an integer m > 0 such
that m < n and
σ(m) σ(n)
≥ ≥ eγ loglog n > eγ loglog m. (8.2)
m n
If m < 10 080, since 10 080 is superabundant we would have, by (8.2),
σ(n) σ(m) σ(10 080)
≤ ≤ ,
n m 10 080
and we could replace m by 10 080. Thus we may assume also 5040 < 10 080 ≤
m. Again by (8.2), m does not satisfy Robin’s inequality, a contradiction,
since n is the smallest such integer. This completes the proof. 
Summarizing Lemmas 8.5, 8.7, 8.8 and 8.9 we have:
Theorem 8.10 [38, theorems 1.3 and 1.4] Any integer greater than 5040 not
satisfying Robin’s inequality is even, not squarefree and not squarefull. The
smallest such integer, if it exists, must also be a Hardy–Ramanujan number
and a superabundant number.

8.3 Integers Not Divisible by the Fifth Power of Any Prime

Recall some standard definitions. If n is an integer, P(n) denotes the largest
prime dividing n. If r ≥ 2 is an integer we say n is r-free if, for all primes p,
pk | n implies k < r.
Lemma 8.11 [38, theorem 1.5] All 5-free integers greater than 5040 satisfy
Robin’s inequality.
 8.3 Integers Not Divisible by the Fifth Power of Any Prime 209

Proof (1) Suppose there exists an integer larger than 5040 which is 5-free
and does not satisfy Robin’s inequality. We claim that if n is the smallest
integer with this property then P(n) < log n. To see this first let q := P(n) with
n = mq. Since n is minimal, either m satisfies Robin’s inequality or it does
not and m ≤ 5040 so m ∈ A. If this latter is true, because
max {n ∈ A : P(n) ≤ 5} = 720,
and 5 × 720 = 5040, we can assume q ≥ 7. By Lemma 8.2, since n > 5040 and
does not satisfy Robin’s inequality, we obtain a contradiction in this case.
Hence m satisfies Robin’s inequality so m ≥ 7.
(2) Assume now, to get a contradiction, that we also have q ≥ log n. This
implies
q log q ≥ log n loglog n > log n loglog m,
which, since

loglog n − loglog m 1 log n
dt 1
= > ,
log q log n − log m log m t log n
implies first
q loglog m q(loglog n − loglog m) loglog m
> =⇒ >
log n log q log q log q
 
1
⇐⇒ 1 + loglog m < loglog n.
q
Then, since for all integers a, b ≥ 1 we have σ(ab) ≤ σ(a)σ(b), we get
   
σ(n) σ(qm) 1 σ(m) 1 γ
= ≤ 1+ < 1 + e loglog m < eγ loglog n,
n qm q m q
so n satisfies Robin’s inequality, a contradiction, completing the proof of the
claim.
(3) Note next that the set of 5-free Hardy–Ramanujan integers n > 5040
with P(n) ≤ 73 is finite. An explicit computation shows each member satisfies
Robin’s inequality.
(4) Now let t ≥ 2 and define for x ≥ 3
 1    1 − 1/pt 
Rt (x) := and S t (x) := .
p>x
1 − 1/pt p≤x
1 − 1/p

Then

log Rt (x) = − log(1 − p−t )
p>x
210 Numbers That Do Not Satisfy Robin’s Inequality
 1
=
p>x m≥1
mptm
 1
≤
p>x m≥1
(pm )t
1 t 1
≤ ≤ .
n>x
n t − 1 xt−1
t

(The reader might wish to experiment with the better upper bound given by
 log Rt (x)−t≤ tx /(t − 1).
1/((t − 1)(x − 1)t−1 ). Therefore 1−t

(5) Because Rt (x)/ζ(t) = p≤x (1 − p ), using the result of Step (3) we can
write
 t
log(1 − p−t ) = −log ζ(t) + log Rt (x) ≤ −log ζ(t) +
p≤x
(t − 1)xt−1

and so, by Lemma 5.12, we get

  1 − 1/pt 
S t (x) = log
p≤x
1 − 1/p
 
tx1−t 1
≤ −log ζ(t) + + γ + loglog x + log 1 + . (8.3)
t−1 2 log2 x
(6) Next we claim that if m is a 5-free integer such that P(m) < log m and
such that m does not satisfy Robin’s inequality, then we must have P(m) ≤ 73.
To see this we use the estimate derived in Step (4) with t = 5. First we set
x = log m. Since m is 5-free,
σ(m)  1 − p−ν p (m)−1  1 − p−5  1 − p−5
= ≤ ≤ = S 5 (log m),
m p|m
1 − p−1 p|m
1 − p−1 p≤log m 1 − p−1

so σ(m)/m ≤ S t (x). By the estimate (8.3),

   
σ(m) 5 1
log ≤ −log ζ(5) + 4 + γ + loglog x + log 1 + .
m 4x log2 x
If x0 satisfies
 
5 1
+ log 1 + < log ζ(5),
4x0t−1 log2 x0
then Robin’s inequality would be satisfied whenever x ≥ x0 , which is false.
For example we can choose x0 = 196 so here x < 196 and
σ(m)
eγ loglog m ≤ < S 5 (193) < 9.2.
m
 8.4 Integers Not Divisible by the Seventh Power of Any Prime 211

Hence we must have log m ≤ exp(S 5 (193)e−γ ) < 175 and the largest prime
less than 175 is 173.
We then repeat this process, replacing 193 with 173, and checking
exp(S 5 (173)e−γ ) ≤ 173, which is true. This continues until we reach the prime
73 for which exp(S 5 (73)) > 73: so the process necessarily stops:
193 → 173 → 151 → 137 → 109 → 103 → 97 → 89 → 83 → 79 → 73
and we conclude P(m) < log m ≤ 73.
(7) We are now able to give the final step in the proof. To obtain a
contradiction, let n be the smallest 5-free integer which is greater than
5040 and which does not satisfy Robin’s inequality. By Step (1) we have
P(n) < log n and by Step (6) P(n) ≤ 73.
Now let ē be the factorization pattern of n. Then necessarily m(ē) is 5-free
and m(ē) ≤ n. Since n is minimal, by Lemma 8.8 we cannot have 5041 ≤
m(ē) < n so either m(ē) = n, and n is Hardy–Ramanujan, or m(ē) ≤ 5040, so
by Lemma 8.8 Step (5) n satisfies Robin’s inequality, which is false. Hence n
is a Hardy–Ramanujan number, which by Step (3) implies n satisfies Robin’s
inequality. This contradiction completes the proof. 
Thus we have shown:
Theorem 8.12 [38, theorem 1.5] Any integer greater than 5040 which does
not satisfy Robin’s inequality must be divisible by at least the fifth power of
at least one odd prime.

8.4 Integers Not Divisible by the Seventh Power of Any Prime

First we set out some definitions. Let n0 := 2263. Recall Nn = p1 · · · pn for
n ≥ 1 is the nth primorial with N1 = 2. Let
7
f (n) := 1 +
10 log pn loglog Nn
and set n1 (t) to be the smallest integer not less than n0 and such that for
n = n1 (t) we have
e2/pn f (n) < ζ(t),
where t ≥ 2 is an integer.
Define
 1 − p−t
Ψt (n) := n ,
p|n
1 − p−1
and set Qt (n) := Ψt (n)/(n loglog n). Recall the definitions
σ(n) 
G(n) = and Rt (x) = (1 − p−t )−1 .
n loglog n p>x
212 Numbers That Do Not Satisfy Robin’s Inequality

In addition recall the definition of a colossally abundant integer from

Chapter 6: n is colossally abundant if for some  > 0 we have for all m > 1
σ(n) σ(m)
≥ 1+ .
n1+ m
The theorem of Solé and Planat, described below, depends critically
on computations of Briggs [16], which produced all colossally abundant
10
numbers up to 1010 , and verified that all of them larger than 5041 satisfied
Robin’s inequality. This computational fact is used also in Section 8.5 below.
Recall the result of Lemma 7.1: if there exists an n > 5040 for which Robin’s
inequality is false, then there exists such an n which is colossally abundant.
Lemma 8.13 [159, proposition 4] Let n ≥ 2263 and t ≥ 2. Then
 
Ψt (Nn ) exp (γ + 2/pn ) 7
≤ loglog Nn + .
Nn ζ(t) 10 log pn
Proof (1) Let Nn ≤ m < Nn+1 . Since m < Nn+1 the number of distinct prime
divisors of m, ω(γ(m)), must be less than or equal to n. If written in ascending
order, the ith of these prime divisors must be not less than pi so
Ψt (m) Ψt (Nn )
≤ .
m Nn
Also for n ≥ 2, we have 0 < loglog Nn ≤ loglog m which implies
Ψt (m) Ψt (Nn )
Qt (m) = ≤ = Qt (Nn ).
m loglog m Nn loglog Nn
(2) By Lemma 8.11 Step (4) we have
t 2
log Rt (pn ) ≤ ≤ ,
(t − 1)pn
t−1 pn
which implies
 

−t −1 2
(1 − p ) ≤ exp .
p>pn
pn

(3) By Lemma 3.14 we have for pn ≥ 2 × 104 ,

  
1
log Nn = θ(pn ) = log p > pn 1 − .
p≤pn
8 log pn

Thus, because for x < 12 , log(1 − x) > −x,

 
1 1
loglog Nn > log pn + log 1 − > log pn − ,
8 log pn 8 log pn
 8.4 Integers Not Divisible by the Seventh Power of Any Prime 213

and therefore
1
log pn < loglog Nn + .
8 log pn
(4) Now write
  −1
Ψt (Nn )  (1 − p−t p>pn (1 − p−t ) 
n
i ) 1
= = 1− .
Nn i=1
−1
1 − pi ζ(t) p≤pn
p

Therefore by Steps (1), (2) and (3) and Lemma 5.12, for n > π(20 000) = 2262
we get
 
Ψt (Nn ) exp(2/pn ) γ 1
≤ e log pn +
Nn ζ(t) 2 log pn
 
exp(γ + 2/pn ) 5
≤ loglog Nn +
ζ(t) 8 log pn
 
exp(γ + 2/pn ) 7
≤ loglog Nn + .
ζ(t) 10 log pn
This completes the proof of the lemma. 
Recall the definition of n1 (t): it is the smallest integer n with n ≥ n0 and
such that exp(2/pn ) f (n) < ζ(t) where f (n) := 1 + 7/(10 log pn loglog Nn ).
Lemma 8.14 [159, corollary 9] For n ≥ n1 (t) and M ≥ Nn we have
Qt (M) < eγ .
Proof Since n ≥ n1 (t) we have
 
7
e 2/pn
1+ < ζ(t),
10 log pn
which implies by Lemma 8.13
 
Ψt (Nn ) exp(γ + 2/pn ) 7
Qt (Nn ) = ≤ 1+ < eγ .
Nn loglog Nn ζ(t) 10 log pn loglog Nn
Because M ≥ Nn , for some n2 ≥ n ≥ n1 (t) we have Nn2 ≤ M < Nn2 +1 . Therefore,
by Step (1) of Lemma 8.13 we have
Qt (M) ≤ Qt (Nn2 ) < eγ ,
and the proof of the lemma is complete. 
Theorem 8.15 [159, theorem 10] If an integer N is 7-free and N > 5040 then
it satisfies Robin’s inequality.
Proof By Step (1) of Lemma 8.13 and Lemma 8.14, if n ≥ n1 (7) = 1292 and
N ≥ Nn then, since n is 7-free,
σ(N) ≤ Ψ7 (N) < Neγ loglog N,
214 Numbers That Do Not Satisfy Robin’s Inequality

so all such integers N ≥ Nn1 (7) < 3.51 × 104550 satisfy Robin’s inequality. By
the computations of Briggs [16, p. 253] Robin’s inequality is true for
10
5040 < N ≤ 1010 ,
and therefore true for all 7-free integers N > 5040. 

8.5 Integers Not Divisible by the 11th Power of Any Prime

This section differs from our usual practice of proving all, or at least most,
of the preliminary lemmas. It also relies on some unpublished material
of Dusart and some computational results of Briggs, which could prove
difficult to reproduce. It is proved that if there is an n ≥ 5041 for which
σ(n) ≥ eγ n loglog n, then n must be divisible by the 11th power of some prime.
Recall that Robin, in Theorem 7.16, proved that the Riemann hypothesis is
equivalent to the Ramanujan–Robin criterion, namely the inequality
σ(n) < eγ n loglog n (8.4)
being true for all n > 5040. Recall also that we say that a number is t-free if
it is not divisible by the tth power of any prime. Choie, Lichiardopol Moree
and Solé [38] showed that (8.4) is true for all 5-free integers; Solé and Planat
[159] showed that (8.4) is true for all 7-free integers. We have presented
these results in Theorems 8.12 and 8.15, respectively. Therefore if there is
some n ≥ 5041 for which σ(n) ≥ eγ n loglog n, then n must be divisible by the
seventh power of some prime.
The purpose of this section is to derive the following result.
Theorem 8.16 [31] If there is some n ≥ 5041 for which σ(n) ≥ eγ n loglog n,
then n must be divisible by the 11th power of some prime.
It is easy to check that the only two positive integers n ≤ 5040 that are
divisible by an 11th power of a prime, namely 211 and 212 , both satisfy
Robin’s inequality. In other words:
Corollary 8.17 The Riemann hypothesis is equivalent to Robin’s inequality
being satisfied by all integers greater than 2 that are divisible by the 11th
power of at least one prime.
Solé and Planat prove theirresults using primorials. Recall that the nth
primorial is defined as Nn = ni=1 pi , where pn denotes the nth prime with
p1 = 2. For an integer t ≥ 2 define as before
 1 1

Ψt (n)
Ψt (n) := n 1 + + · · · + t−1 , Qt (n) := .
p|n
p p n loglog n

Solé and Planat note that for a t-free integer n one has σ(n) ≤ Ψt (n). Using
their method, for a given t, first find an integer n1 (t) such that Rt (Nn1 (t) ) < eγ
 8.5 Integers Not Divisible by the 11th Power of Any Prime 215
10
and Nn1 (t) < 1010 . Solé and Planat in Lemma 8.14 showed that for all
N > Nn1 (t) we have Qt (N) < eγ . Since Briggs [16] has proved that Robin’s
10
inequality is true for all 5041 ≤ n ≤ 1010 , this then shows that Robin’s
inequality is true for all such t-free integers n ≥ 5041.
The main idea of this section is to use explicit estimates on sums over
primes to bound the primorials. This makes for an easy computation, and
one which is used to verify Theorem 8.16.
First the function Qt (n) is estimated using bounds on primorials. For t ≥ 2
we have by Step (4) of Lemma 8.13
  −t −1 
1 − p−t p>pn (1 − p )
n
1
Qt (Nn ) = k
= (1 − p−1 )−1 . (8.5)
loglog Nn k=1 1 − p−1k
ζ(t) loglog N n p≤p n

The first product on the right-hand side of the expression (8.5) is estimated
using Lemma 6 of Solé and Planat, namely,

(1 − p−t )−1 ≤ exp(2/pn ), (8.6)
p≥pn

for all n ≥ 2. Although this could be improved, such an improvement would

have negligible influence on the final result.
To estimate the second product on the right-hand side of (8.5), the
following result
  −1
−1 −1 γ 1
(1 − p ) ≤ e log x 1 − , x ≥ 2973, (8.7)
p≤x 5 log2 x

given in the unpublished result of Dusart [50, theorem 6.12] is used. This
improves on Lemma 5.12.
Now to apply the estimates (8.6) and (8.7) with x = pn , an explicit bound on
the nth prime is needed. It is possible to proceed without an explicit bound,
though this increases greatly the computation time when n is large. In fact, if
one considers the function
exp(2/x) log x
f (x) =  , (8.8)
1
1−
5 log2 x
which is increasing for all x ≥ 5, one sees that it is sufficient to consider only
an explicit upper bound on pn . Consider
pn ≤ b1 (n) := n(log n + loglog n − 12 ), n ≥ 20, (8.9)
and
pn ≤ b2 (n) := n(log n + loglog n − 0.9484), n ≥ 39 017, (8.10)
216 Numbers That Do Not Satisfy Robin’s Inequality

due respectively to Rosser and Schoenfeld [145, (3.11)] and Dusart [49,
section 4].
Now we bound the factor loglog Nn in expression
 (8.5) and make use of the
function θ(x) = p≤x log p. Since log Nn = ni=1 log pi = θ(pn ), bounds on Nn
can be derived using bounds on θ(pn ). These results are given in the following
lemma.
Lemma 8.18 For k ≥ 198,
 
loglog k − 2.1454
k log k + loglog k − 1 +
log k
 
loglog k − 2
≤ log Nk ≤ k log k + loglog k − 1 + . (8.11)
log k
Proof The left inequality follows from Robin [141, theorem 7] and is valid
for all k ≥ 3. The right inequality follows from Massias and Robin [114,
theorem B(v)] and is valid for all k ≥ 198. 
Note that the constant “2” in the right inequality in (8.11) cannot be
improved. One could reduce the “2.1454” that appears in the left inequality
at the expense of taking a much larger k. As shown below at the end of this
section, this has very little influence on our problem.
Now we use the estimates (8.6)–(8.10) and Lemma 8.18 to estimate the
right-hand side of expression (8.5). We have
⎧
⎪
⎪
⎨eγ g1 (t, n), 430 ≤ n ≤ 39 016,
Qt (Nn ) ≤ ⎪
⎪ (8.12)
⎩eγ g2 (t, n), n ≥ 39 017,

where
f (bi (n))
gi (t, n) = %  & ,
loglog n − 2.1454
ζ(t) log n log n + loglog n − 1 +
log n
for i = 1, 2. We require n ≥ 430 in the bounds (8.12) to ensure that the
conditions in the estimate (8.7) and Lemma 8.18 are met since p429 = 2971
and p430 = 2999. Following Solé and Planat, define n1 (t) to be the least value
of n ≥ 430 for which gi (t, n) < 1. Since gi (t, n) is decreasing in n, we will
have gi (t, n) < 1 for all n ≥ n1 (t). For such an n1 (t) consider the size of the
associated Nn1 (t) . The results are summarized in Table 8.1. Exact values of
Nn1 (t) for t = 11 and 12 are not given owing to the computational complexity
of calculating the nth primorial exactly.
As has been previously noted, Briggs [16] has proved that Robin’s
10
inequality is true for 5041 ≤ n ≤ 1010 , by demonstrating numerically it
holds for all colossally abundant numbers in this range, and then invoking
Theorem 6.16. To prove a statement such as “Robin’s inequality holds for all
 8.6 Unsolved Problems 217
Table 8.1 Values of n1 (t) and Nn1 (t) .

t n1 (t) Nn1 (t) Upper bound on Nn1 (t)

using Lemma 8.18

6 430 3.3 × 101273 1.4 × 101276

7 1847 3.3 × 106836 2.7 × 106851
8 39 017 4.9 × 10202520 2.3 × 10202725
9 39 017 4.9 × 10202520 2.3 × 10202725
10 234 372 1.2 × 101416098 1.8 × 101416984
11 48 304 724 — 2.8 × 10411504586
12
12 162 914 433 505 — >1010

10
t-free integers”, one must show that Nn1 (t) ≤ 1010 . The last entry in Table 8.1
shows that, at present, it is impossible to consider t = 12 without a new idea.
Conclusion. Here is a brief discussion of the possibility of proving
that Robin’s inequality is satisfied by all 12-free integers. Dusart [50]
(unpublished) has considered some improved versions of the estimates (8.7)
and (8.11), namely
  
−1 −1 γ 1
(1 − p ) < e log x 1 + , x ≥ 2973,
p≤x 5 log2 x
and
 
loglog k − 2.04
k log k + loglog k − 1 + ≤ θ(pk ), pk ≥ 1015 .
log k
These results appear respectively as Theorem 6.12 and Proposition 6.2 in
12
[50]. Even with these improvements one still has Nn1 (12) > 1010 . Without
injecting new ideas into the argument, one would have to increase the range
10
of Briggs’ computations beyond 1010 .
To extend the computation one need only check those numbers that are
colossally abundant and are divisible by the 11th power of some prime.
Presumably, this is a very thin set of numbers. Checking only these numbers
may motivate an extension of Briggs’ computations and hence the possibility
of extending the results in this section to t = 12.

8.6 Unsolved Problems

(1) Extend the method of Solé and Planat to t = 12 and beyond.
(2) Being divisible by the 11th power of at least one prime is perhaps only
one type of constraint that can be derived for numbers greater than 5040
which violate Robin’s inequality. This problem is to find new constraints
for such numbers.
(3) Show that any counterexample to Robin’s inequality (greater than 5040)
must be colossally abundant.
 9
Left, Right and Extremely Abundant Numbers

9.1 Introduction
This chapter explores further properties of numbers for which Robin’s
inequality fails. It is based on two papers [34, 35], by the same three authors
published in 2011 and 2012: Geoffrey Caveney, Jean-Louis Nicolas and
Jonathan Sondow. Two new equivalences to RH are derived. It also includes
an equivalence from 2013 of Nazardonyavi and Yakubovich.
Recall the definition of Grönwall’s function:
σ(n)
G(n) := , n > 1,
n loglog n
and note the important property which is used in this chapter and proved in
Section 9.2:

Theorem 9.2 (Grönwall’s theorem) [69; 75, theorem 323] We have that

lim sup G(n) = eγ .

n→∞

We say a composite positive integer n is left abundant if G(n/p) ≤ G(n)

for all prime divisors p | n. For example, for all p ∈ P, p is not left abundant
since G(1) is not defined. We say it is right abundant if G(n) ≥ G(an) for all
integers a > 1. It is said to be extraordinary if it is both left abundant and
right abundant. Finally, a natural number n is said to be extremely abundant
if either n = 10 080 or n > 10 080 and G(m) < G(n) for all m with 10 080 ≤
m < n. Note that 10 080 is the smallest superabundant number which is greater
than 5040.
Recall n is superabundant if for all 1 ≤ m < n we have
σ(m) σ(n)
< ,
m n

218
 9.1 Introduction 219

and denote by S the set of all superabundant numbers. It follows from

Theorem 9.2 that, since σ(n)/n = G(n) loglog n, we have lim supn→∞ σ(n)/n =
∞, so the number of superabundant numbers, the size of S, is infinite.
The first main result is that RH is equivalent to 4 being the only extra-
ordinary number, which is Theorem 9.8. We call this the Caveney–Nicolas–
Sondow criterion. The theorem has the corollary that if a counterexample
n to Robin’s inequality exists with n > 5040, then so does one which is an
extraordinary number.
The next main result, Lemma 9.16, is that Robin’s inequality is satisfied
for a very large range of integers, namely if 5040 < n < e19747 then G(n) <
eγ . This means any counterexample must be greater than 1010000 . Then in
Theorem 9.17 the proof is given that the set of right abundant numbers not
greater than 5040 is

A \ {1, 2, 9, 16, 20, 30, 84, 720, 840},

where A as defined at the start of Chapter 8 is the set of integers not greater
than 5040 for which Robin’s inequality is false. This theorem includes an
equivalence for the Riemann hypothesis, namely if the Riemann hypothesis
is true then there are no other right abundant numbers, and if the hypothesis
is false then infinitely many right abundant numbers exist and in that case

∞ > μ := max{G(n) : n > 5040} > eγ .

In addition, any integer M for which G(M) = μ is both colossally abundant

and right abundant.
Theorem 9.19 shows that infinitely many colossally abundant numbers are
also left abundant, and Theorem 9.21 that infinitely many colossally abundant
numbers are not left abundant.
In a final section, Section 9.6, we include the theorem of Nazardonyavi
and Yakubovich that the Riemann hypothesis is equivalent to the number
of extremely abundant numbers being infinite. This is the Nazardonyavi–
Yakubovich criterion.
This chapter has a complex network of dependences, and call-outs to
results in other chapters. To assist the reader to find a way through,
some dependences between results are presented as network flows. In
Figures 9.1, 9.2 and 9.3, an arrow from vertex A to B, say, means derivation
B requires the result of A.
The following RHpack (see Appendix B) functions relate to the material in
this chapter: GronwallG, ExtraordinaryQ, RightAbundantQ, LeftAbundantQ
and ExtremelyAbundantQ.
220 Left, Right and Extremely Abundant Numbers

Lem 5.13

Lem 9.1 Cor 9.9

Thm 9.2

Lem 9.7

Thm 9.8
Lem 9.6
Lem 9.3
Thm 9.2
Lem 9.5 Thm 7.16

Lem 9.4

Lem 6.1

Lem 6.2

Figure 9.1 Dependences between some results in this chapter.

9.2 Grönwall’s Theorem

The first preliminary result was proved by Grönwall in 1913 [69]. It
underpins all of the work in this volume since Chapter 6, but is not needed
explicitly until this chapter. First a standard lemma.
Lemma 9.1 If pn is the nth prime then as n → ∞ we have pn ∼ n log n.
Proof By the prime number theorem we have n log pn ∼ pn . Taking
logarithms gives
log n + loglog pn ∼ log pn .
Therefore
n log n n log n log n log pn − loglog pn
∼ = ∼ ∼ 1,
pn n log pn log pn log pn
 9.2 Grönwall’s Theorem 221

Lem

Thm 7.13
Lem 3.14

Thm 7.16
Lem 9.16 Thm 9.17
Lem 9.12

Lem 6.10

Lem 9.15
Lem 9.5

Figure 9.2 Further dependences between some results in this chapter.

Thm 4.16 Lem 9.20

Lem 9.14 Thm 9.21

Thm 9.19 Lem 9.13

Lem 9.18 Lem 9.11

Figure 9.3 More dependences between some results in this chapter.

which completes the proof. 

Theorem 9.2 (Grönwall’s theorem) [69; 75, theorem 323] We have that

lim sup G(n) = eγ .

n→∞
222 Left, Right and Extremely Abundant Numbers

Proof (1) Let m ∈ N and for each j let p j be the jth prime with p1 = 2. Define
an integer nm by

nm n
m +1

pj ≤ m < pj =⇒ θ(pnm ) ≤ log m < θ(pnm +1 ). (9.1)

j=1 j=1

By Lemma 9.1 we have

pn+1 (n + 1) log(n + 1)
lim = lim = 1. (9.2)
n→∞ pn n→∞ n log n
Using this limit and the prime number theorem in the form limx→∞ θ(x)/x = 1,
we can write
θ(pnm ) log m θ(pnm +1 ) pn +1 θ(pnm +1 )
1 = lim ≤ lim ≤ lim = lim m = 1.
m→∞ pn m→∞ pn m→∞ pnm m→∞ pnm pnm +1
m m

Thus
log m
lim = 1.
m→∞ pn
m

It follows that as m → ∞
pnm = log m (1 + o(1)). (9.3)
(2) By Mertens’ formula (Lemma 5.12) we can write, as x → ∞,
 1

1
1− = γ (1 + o(1)).
p≤x
p e log x

Let x = pm in this formula and use (9.3) to derive, using log(1 + o(1)) = o(1),
nm 
 
1 1 1
1− = γ (1 + o(1)) = γ (1 + o(1)). (9.4)
j=1
p j e log p nm e loglog m

(3) Next write the prime factorization of m as

m = pαj11 · · · pαjnn ,
where as before p j is the jth prime, j1 < j2 < · · · < jn and the exponents α j > 0.
Then

n
1 − 1/pαj i +1 
n
1
σ(m) = m i
<m . (9.5)
i=1
1 − 1/p ji i=1
1 − 1/p ji
Because n ≤ nm and pi ≤ p ji for all i it follows using Step (2) that

n
1  n
1 m
1
n
≤ ≤ = eγ loglog m (1 + o(1)).
i=1
1 − 1/p ji i=1 1 − 1/pi i=1 1 − 1/pi
 9.3 Further Preliminary Results 223

Using inequality (9.5) then gives

σ(m)
lim sup ≤ eγ . (9.6)
m→∞ m loglog m
(4) Next we construct a sequence of integers which shows the right-hand
side of inequality (9.6) is attained. (Note that this sequence has found little
use elsewhere in the work reported in this volume – which is somewhat
surprising.)
Define a sequence of natural numbers (an )n∈N by raising the nth primorial
to
# a power
$ which is determined by the logarithm of the largest prime, βn :=
log pn :
an = (p1 p2 · · · pn )βn . (9.7)
Then, by (9.3),

n
1 − 1/pβj n +1
σ(an ) = an
j=1
1 − 1/p j
 1

n
1
> an 1−
p
pβn +1 j=1
1 − 1/p j
an 
n
1
=
ζ(βn + 1) j=1 1 − 1/p j
an
= eγ log pn (1 + o(1)). (9.8)
ζ(βn + 1)
(5) The definition of the an and the prime number theorem give
log an = βn θ(pn ) = βn pn (1 + o(1)) = pn log pn (1 + o(1)).
This implies
log pn
loglog an = log pn + loglog pn + o(1) =⇒ lim = 1.
n→∞ loglog an
In addition, since ζ(σ) → 1 when σ → ∞, by the expression (9.8) we get
σ(an )
lim sup ≥ eγ .
n→∞ an loglog an
This, together with the result of Step (3), completes the proof. 

9.3 Further Preliminary Results

Lemma 9.3 [34, proposition 1, SA1] If S is the set of all superabundant
numbers, then
lim sup G(s) = eγ .
n∈S
224 Left, Right and Extremely Abundant Numbers

Proof By Theorem 9.2 there is an increasing sequence of positive integers

(n j ) j≥1 such that lim j→∞ G(n j ) = eγ . If for any j we happen to have n j ∈ S, the
set of all superabundant numbers, let s j := n j . Otherwise define
s j := max{s ∈ S : s < n j }.
Then if we set T := {s j + 1, . . . , n j } we would have T ∩ S = ∅. If we had
σ(s j )/s j < σ(n j )/n j , then the value of x ∈ T such that σ(x)/x is a maximum
would be in S, which is impossible. Therefore σ(s j )/s j ≥ σ(n j )/n j and so
G(s j ) > G(n j ). It follows that
lim sup G(s j ) ≥ lim sup G(n j ) = eγ
j→∞ j→∞

and thus lim sup s∈S G(s) = eγ . 

Lemma 9.4 [34, proposition 1, SA2] For every fixed positive integer a, every
sufficiently large n ∈ S is a multiple of a.
Proof Let a > 1, let k be the largest exponent in the standard prime
factorization of a, and let p be the maximum prime divisor of a. Then
a | (2 · 3 · · · p)k . Let
F := {s ∈ S : pk  s} = {s ∈ S : 0 ≤ ν p (s) < k},
where, as usual, ν p (s) is the highest power of p which divides s. By
Lemma 6.2 with q = 2 and r = p we see that ν2 (s) < B for some bound B
and all s ∈ F. If q | s ∈ F is any prime factor then by Lemma 6.1 we have
νq (s) < B, and then, using Lemma 6.4, qνq (s) < 2B+2 . Using this with q the
maximum prime dividing s (q = P(s)) gives P(s) < 2B+2 so for all s ∈ F we
have
s < (2B+2 !)B ,
and therefore F is a finite set. Hence, for all s sufficiently large we have pk | s.
The result of the lemma now follows using Lemma 6.1. 
Lemma 9.5 [34, lemma 1] If a ∈ N then lim supn→∞ G(an) = eγ . If n is right
abundant then G(n) ≥ eγ .
Proof The first part follows directly from Lemmas 9.3 and 9.4. For the
second part, since n is right abundant we have G(an) ≤ G(n), and so eγ =
lim supa→∞ G(an) ≤ G(n). 
Recall from the start of Chapter 8 that A denotes the set of positive integers
not greater than 5040 for which Robin’s inequality fails, i.e. A := {n ≤ 5040 :
G(n) ≥ eγ }. Then
A := {1, 2, 3, 4, 5, 6, 8, 9, 10, 12, 16, 18, 20, 24, 30, 36, 48,
60, 72, 84, 120, 180, 240, 360, 720, 840, 2520, 5040},
 9.4 Riemann Hypothesis Equivalences 225

Lemma 9.6 [34, lemma 2] If n ∈ A is greater than 5, then G(n) < G(n/p) for
some prime p | n.
Proof The smallest prime factor of n ∈ A \ {1, 2, 3, 4, 5} for which G(n) <
G(n/p) is given by
(n, p) ∈{(6, 2), (8, 2), (9, 3), (10, 2), (12, 2), (16, 2), (18, 3), (20, 5), (24, 3),
(30, 3), (36, 2), (48, 3), (60, 5), (72, 3), (84, 7), (120, 2), (180, 5),
(240, 5), (360, 5), (720, 3), (840, 7), (2520, 7), (5040, 2)},
so this result is a simple computation. 
Lemma 9.7 [34, lemma 3] If r ∈ A then there exists an  > 0 such that for all
primes p ≥ 11 we have G(pr) < eγ − .
Proof Let n be a positive integer, and let p and q be odd primes not dividing
n with q < p. Then
σ(pn) p + 1 σ(n) q + 1 σ(n)
G(pn) = = < = G(qn).
pn loglog pn p n loglog pn q n loglog qn
No prime p ≥ 11 divides any element of A. Also G(11r) < eγ for all r ∈ A.
Therefore G(pr) ≤ G(11r) < eγ . 

9.4 Riemann Hypothesis Equivalences

Recall that an extraordinary number is a positive integer N for which
G(N/p) ≤ G(N) for all primes p | N and G(aN) ≤ G(N) for all a ≥ 1.
Theorem 9.8 (Caveney–Nicolas–Sondow first criterion) [34, theorem 5]
The Riemann hypothesis is true if and only if 4 is the only extraordinary
number.
Proof Let N  4 be an extraordinary number. Then for all n ∈ N, since N is
right abundant, we have G(N) ≥ G(nN). Therefore, by Lemma 9.5 we have
G(N) ≥ lim sup G(nN) = eγ .
n→∞

If N ≤ 5040 then N ∈ A. But N being extraordinary, and not equal to 4, is

composite (recall the comment regarding G(p) in Section 9.1) so N > 5, and
thus by Lemma 9.6 G(N) < G(N/p) for some prime factor p | N. Since N is
right abundant this is impossible, so N > 5040 and G(N) ≥ eγ , and therefore
by Theorem 7.16 the Riemann hypothesis is false.
Now suppose that the Riemann hypothesis is false. Then since
lim supn→∞ G(n) = eγ (Theorem 9.2) and, by Theorem 7.16, there is an N
with G(N) > eγ , the maximum μ and minimum N exist where
μ := max{G(n) : n > 5040} and N := min{n > 5040 : G(n) = μ}.
226 Left, Right and Extremely Abundant Numbers

Then μ = G(N) ≥ eγ . We claim that such an N is extraordinary. To see this,

note first that G(N) ≥ G(n) for all n ≥ N, and thus N is right abundant.
Next, if N were prime then, since x → (x + 1)/(x loglog x) is decreasing for
x ≥ 5040, we would have
N +1 5041
1.7 < eγ ≤ G(N) = < < 0.5.
N loglog N 5040 loglog 5050
Therefore N is composite.
Now let p | N and set r := N/p. If r > 5040 since r < N and N is the minimal
integer with maximum value, we must have G(r) < G(N).
If however r ≤ 5040, when 5041 ≤ n ≤ 35 280 a computation shows that
G(n) < eγ , so N > 35 280 = 7 × 5040. Since in addition N = rp and r ≤ 5040
we must have p ≥ 11. If G(r) ≥ eγ we would have r ∈ A. Then Lemma 9.7
would give
eγ > G(pr) = G(N) ≥ eγ
so G(r) < eγ ≤ G(N). Therefore in both cases we have G(N) > G(r) = G(N/p)
so N is left abundant also and therefore N is an extraordinary number. 
The proof of Theorem 9.8 has the following direct corollary:

Corollary 9.9 If there is a counterexample to Robin’s inequality then the

maximum μ := max{G(n) : n > 5040} exists and the least N > 5040 with
G(N) = μ is an extraordinary number.
Lemma 9.10 [34, propositions 16 and 17] All left abundant numbers N with
exactly two prime factors are of the form N = 4 or N = 2p with p ≥ 7 a prime,
and all such numbers are left abundant.
Proof First we claim numbers of the form 2p are left abundant if and only
if p = 2 or p > 5. To see this, since G(2) < 0, numbers of the form 2p are left
abundant if and only if G(p) < G(2p), which holds certainly for p = 2 but not
for p = 3 or p = 5 using numerical checks. If p > 5 write
G(2p) 3(p + 1) p loglog p 3 loglog p
= = > 1,
G(p) 2p loglog 2p p + 1 2 loglog 2p
where the final inequality follows since 3 loglog x > 2 loglog 2x for x ≥ 7.
Hence 2p is left abundant for all primes p ≥ 7.
Now we show products of two distinct odd primes, say p and q, are not left
abundant. Let p > q ≥ 3 and write
G(pq) (p + 1) loglog q
= < 1,
G(q) p loglog pq
where we use the inequality (x + 1) loglog y < x loglog xy, which is true
whenever x ≥ y ≥ 3.
 9.4 Riemann Hypothesis Equivalences 227

Finally if p ≥ 3 then
G(p2 ) (p2 + p + 1) loglog p
= < 1.
G(p) (p2 + p) loglog p2
This completes the proof. 
Recall that in Chapter 7, Theorem 7.13, Robin’s unconditional upper
bound gives for all n > 1 the bound G(n) ≤ eγ + 1.1161/(loglog n)2 . We also
saw in Theorem 7.15 that if RH was false there exists a real number b with
1 − θ < b < 12 , where θ := sup{β : ζ(β + iγ) = 0}, such that when n → ∞ through
colossally abundant numbers
  
γ 1
G(n) = e 1 + Ω± .
logb n
Also recall from Chapter 6 we defined and used the function
 
1 1
F(z, k) := log 1 + .
z + · · · + z log z
k

Now we need an additional construction which was made in Chapter 6. Recall

we showed in Lemma 6.18 that xk ∼ (kx)1/k for all k ≥ 2, where x := x1 ().
Lemma 9.11 [35, lemma 1] In the notation of Chapter 6, let  > 0 and x =
x1 (). For all k ≥ 2 we have xk < (kx)1/k .
Proof Let z := (kx)1/k . If z > 1 and k ≥ 2, using (t + 1/2)−1 < log(1 + 1/t),
which is true for t > 0, we have
 
1 1
F(z, k) = log 1 +
z + · · · + z log z
k

k
<
(z + · · · + zk ) log zk
k
<
(k − 1 + zk ) log kx
k
≤
(k/2 + kx)log x
1 1
< log 1 + = F(x, 1),
x log x
so F(z, k) < F(x, 1). But this implies F(xk , k) =  = F(x, 1) > F(z, k), and the
lemma follows since t → F(t, k) is strictly decreasing on (1, ∞). 
Lemma 9.12 [35, lemma 2] Let N0 be a colossally abundant number of
parameter 0 , and N > N0 , for all n ≥ N0 , satisfies
σ(n) σ(N)
≤ 1+
n1+ N
for some fixed  > 0. Then N is also colossally abundant of parameter .
228 Left, Right and Extremely Abundant Numbers

Proof The given hypothesis applied to N > N0 and N0 being colossally

abundant of parameter 0 give
 1+  1+0
N σ(N) N
≤ ≤ =⇒  ≤ 0 .
N0 σ(N0 ) N0
If n < N0 , again using the hypothesis and that N0 is colossally abundant, we
have
σ(n) σ(n)n0 −
=
n1+ n1+0
σ(N0 )n0 −
≤ 1+
N0 0
σ(N0 )N00 −
≤ 1+
N0 0
σ(N)
≤ 1+ .
N
Therefore σ(n)/n 1+
≤ σ(N)/N , which completes the proof of the
1+

lemma. 
Lemma 9.13 [35, lemma 3] Let N be a colossally abundant number of
parameter  where  < F(2, 1) and let x = x be the unique solution to
F(x, 1) = . Then √
(i) for some constant c > 0 we have log N ≤ θ(x) + c x, and
(ii) if N is the largest colossally abundant number of parameter  then
√
θ(x) ≤ log N ≤ θ(x) + c x.
Proof Recall that x = x1 where for k ≥ 1 we have F(xk , k) = , and that the
sequence (xk ) is decreasing. Let K be the largest integer such that xK ≥ 2 so
that for all primes q with q | N we have 2 ≤ q ≤ xk and k = νq (N) ≤ K. Note
also that ν2 (N) = K or K − 1 and that if  < F(2, 1) then x > 2 and K ≥ 1. Then

K 
N= q. (9.9)
k=1 q≤xk

Therefore
log N ≤ θ(x1 ) + θ(x2 ) + · · · + θ(xK ).
Because t → F(t, k) is decreasing we have
F(2, K) ≥ F(xK , K) =  = F(xK+1 , K + 1) > F(2, K + 1).
But
 
1 1 1 1 2
F(2, K) = log 1 + < K+1 ≤ K < K,
2K+1 − 2 log 2 (2 − 2) log 2 2 log 2 2
 9.4 Riemann Hypothesis Equivalences 229

so therefore using F(x, 1) =  we get

log(1 + 1/x) 1 1 1
= > ≥ > 2.
log x (x + 1) log x (x + 1)(x − 1) x
Therefore
2 1 2 log x
> F(2, K) ≥  > 2 =⇒ K < 1+ . (9.10)
2K x log 2
Because k → xk is decreasing, we also get

log N ≤ θ(x1 ) + θ(x2 ) + Kθ(x3 ).

By Lemma 9.11 we have x2 ≤ (2x)1/2 and √x3 ≤ (3x)1/3 so using (9.10) and
θ(t) ∼ t as t → ∞, we get log N ≤ θ(x) + c x, which proves part (i) of the
lemma.
Equation (9.9) gives the largest colossally abundant number of parameter
 as

K 
N= q.
k=1 q≤xk

It follows that θ(x) ≤ log N, completing the proof of part (ii). 

Lemma 9.14 [35, lemma 4] For some c > 0 there are infinitely many primes
√
p such that θ(p) < p − c p logloglog p, and infinitely many other primes p
√
such that θ(p) > p + c p logloglog p.

Proof The proof employs Littlewood’s theorem [109], Theorem 4.13, in

an essential manner. In particular, that theorem gives two increasing and
unbounded sequences of real values (xn ) and (yn ) and a positive constant c
such that for all n
√ √
θ(xn ) < xn − 2c xn logloglog xn and θ(yn ) > yn + 2c yn logloglog yn .
(9.11)
In fact, the theorem refers to ψ(x) rather than θ(x), but
√ the θ(x) version is an
easy corollary since, assuming RH, ψ(x) = θ(x) + O( x log x), which follows
from

∞
ψ(x) = θ(x1/n ), θ(x) = 0 for x < 2.
n=1

If an infinite subsequence of the xn is prime we can replace (xn ) by that

subsequence and the first part of the lemma follows. So assume no xn is
prime and let pn be the smallest prime larger than xn . Then since t →
230 Left, Right and Extremely Abundant Numbers
√
t − c t logloglog t is increasing for t sufficiently large we have
√
θ(pn ) = θ(xn ) + log pn < xn − 2c xn logloglog xn + log pn
√
< pn − 2c pn logloglog pn + log pn ,
√
so for n sufficiently large we get θ(pn ) < pn − c pn logloglog pn .
Similarly let pn be the largest prime less than or equal to yn . By (9.11) again
we have
√ √
θ(pn ) = θ(yn ) > yn + 2c yn logloglog yn > pn + c pn logloglog pn ,
which completes the proof. 
Lemma 9.15 [35, lemma 6] Let  > 0 and for t > e define
g (t) :=  log t − logloglog t.
Then there is a unique t0 > e, dependent on , such that
 = 1/(log t0 loglog t0 ).
The function g (t) is decreasing for e < t < t0 and increasing for t > t0 .
Proof Note that
 
1 1
g (t) = − ,
log t loglog t t
so the result follows directly provided  = 1/(log t0 loglog t0 ) has a unique
solution t0 . But this is true because t → 1/(log t loglog t) is strictly decreasing
on (e, ∞) with image (0, ∞). 
Recall again from Chapter 8 that
A := {1, 2, 3, 4, 5, 6, 8, 9, 10, 12, 16, 18, 20, 24, 30, 36, 48,
60, 72, 84, 120, 180, 240, 360, 720, 840, 2520, 5040}
is the set of positive integers n, not greater than 5040, for which G(n) ≥ eγ .
Lemma 9.16 [35, p. 371] If 5040 < n < e19747 then G(n) < eγ .
Proof Recall from Step (3) of the proof of Theorem 7.11 in Chapter 7 that if
C is the largest colossally abundant number with P(C) < 20 000 then there is
no n with 5040 < n ≤ C such that G(n) ≥ eγ . Using Lemma 6.10 we see that
each colossally abundant number includes an initial segment of primes with
non-increasing exponents, and for every initial sequence of primes there is a
colossally abundant number with corresponding maximum prime divisor, so
P(C) = 19 993. Therefore by Lemma 9.13 logC ≥ θ(20 000). Hence can write
using Lemma 3.14 (a), for x ≥ 19 421,
x
θ(x) > x − =⇒ logC ≥ θ(20 000) > 19 747.
8 log x
 9.4 Riemann Hypothesis Equivalences 231

This completes the proof of the lemma. 

Recall that N is right abundant if for all a ≥ 1 we have G(N) ≥ G(aN).
Theorem 9.17 (Caveney–Nicolas–Sondow second criterion) [35, theorem
5] The set of right abundant numbers not greater than 5040 is A \
{1, 2, 9, 16, 20, 30, 84, 720, 840}. If the Riemann hypothesis is true then there
are no other right abundant numbers. If the hypothesis is false then infinitely
many right abundant numbers exist and in that case
∞ > μ := max{G(n) : n > 5040} > eγ ,
and, in addition, any integer M for which G(M) = μ is both colossally
abundant and right abundant.
Proof (1) Let
A2 := {3, 4, 5, 6, 8, 10, 12, 18, 24, 36, 48, 60,
72, 120, 180, 240, 360, 2520, 5040} ⊂ A
= A \ {9, 16, 29, 30, 84, 720, 840},
A3 := {n ≤ 5420 : n is right abundant}.
First we show that A3 ⊂ A2 . If N is in A3 then, by Lemma 9.5, G(N) ≥ eγ so N
is in A. For r ∈ A let Br := {a ∈ N : ar ∈ A, G(ar) > G(r)} and set ar := min Br
when Br is non-empty. Otherwise set ar := 0. A computation shows that the
r ∈ A for which Br  ∅ are {9, 16, 20, 30, 84, 720, 840}. But N is right abundant
so necessarily BN = ∅ giving N ∈ A \ {9, 16, 20, 30, 84, 720, 840} = A2 .
(2) Now let N ∈ A2 . We claim G(N) ≥ G(aN) for any multiple aN of N.
First consider multiples aN ≤ 5040. If G(aN) < eγ , because N ∈ A we have
G(aN) < eγ ≤ G(N). Otherwise aN  A which again gives G(aN) < eγ ≤ G(N)
so in both cases G(aN) < G(N) when aN ≤ 5040. (Alternatively a finite set of
possibilities can be checked numerically.)
(3) Now consider multiples aN > 5040 with N ∈ A2 as before. If log(aN) <
19 747 then Lemma 9.16 gives G(aN) < eγ ≤ G(N) and we are done. If
however log(aN) ≥ 19 747 then using Theorem 7.13 we have
0.667
G(aN) ≤ eγ + < 1.78849 < 1.790 < min{G(n) : n ∈ A2 } ≤ G(N).
log2 19 747
Therefore G(N) ≥ G(aN) when aN > 5040, so all elements of A3 are right
abundant. This shows that A2 ⊂ A3 , and therefore A3 = A2 , completing the
proof of the first part of the theorem.
(4) Now assume that the Riemann hypothesis holds. Then by Robin’s
theorem, Theorem 7.16, for all n > 5040 we have G(n) < eγ , and therefore,
by Lemma 9.5, n cannot be right abundant, so A3 is precisely the set of right
abundant numbers.
232 Left, Right and Extremely Abundant Numbers

(5) Finally assume that the Riemann hypothesis is false. Let the real
number η with 12 < η ≤ 1 be defined by

η := sup ρ.
ζ(ρ)=0

Using Theorem 7.15 we have for some b with 1−η < b < 12 , as N → ∞ through
colossally abundant numbers,
 
γ 1
G(N) = e + Ω+ .
logb N
Therefore there are infinitely many positive integers N which are colossally
abundant and satisfy G(N) > eγ . Thus, for all real t we have max{G(n) :
n ≥ t} > eγ .
Let μ1 := μ = max{G(n) : n > 5040} and let a1 be the smallest integer and b1
the largest integer both greater than 5040 with G(a1 ) = G(b1 ) = μ1 . If an and
bn have been defined for 1 ≤ n < i, we set

μi := max{G(n) : n > bi−1 },

and then let ai be the smallest integer and bi the largest integer, both greater
than bi−1 , such that G(ai ) = G(bi ) = μi . Because μi > eγ = lim supn→∞ G(n),
there is an infinite number of distinct integers ai , and they are such that if
n > ai then G(n) ≤ G(ai ). Therefore each ai is right abundant, so infinitely
many right abundant numbers exist.
(6) Now let an integer M > 5040 satisfy G(M) = μ. The method of the
previous paragraph can be used to show that M is necessarily right abundant.
Using Lemma 9.12 with N0 = 5040, 0 = 3/100 (in the notation of Section 6.3,
7 < 0.03 < 8 ), N = M and  = 1/(log M loglog M), and employing also
Lemmas 9.16 and 9.5, we get N = M > e19747 > N0 . For n ≥ N0 , using the
definition of M we get G(n) ≤ G(M). By Lemma 9.15, because  < N0 < M,
on the interval [N0 , ∞) the function g (t) attains its minimum at t = M, so for
n ≥ N0 we get
σ(n) loglog n G(n) G(M) G(M) σ(M)
= G(n) ≤ g (n) ≤ g (n) ≤ g (M) = 1+ .
n 1+ n e e e M
Therefore, by Lemma 9.12, M is also colossally abundant, completing the
proof of the theorem. 

9.5 Comparing Colossally and Left Abundant Numbers

Lemma 9.18 [35, lemma 7] Let N be colossally abundant of parameter 
and assume p := P(N) with p ≥ 5. If  > 1/(log(N/p) loglog(N/p)) then N is
also left abundant.
 9.5 Comparing Colossally and Left Abundant Numbers 233

Proof Let q | N be any prime divisor. Because p ≥ 5 and N is colossally

abundant we have 6p | N and so
 
N N N
≥ ≥ 6 > e =⇒ loglog > loglog e = 0.
q p q
Again because N is colossally abundant we get
σ(N/q) σ(N) σ(N/q) 1
≤ 1+ =⇒ ≤ 1+ . (9.12)
(N/q)1+ N σ(N) q
Because loglog N and loglog(N/q) are both positive we get
G(N/q) loglog N (N/q) loglog N
≤  =  = exp(g (N/q) − g (N)).
G(N) q loglog(N/q) N loglog(N/q)
(9.13)
By the result and notation of Lemma 9.15 we see g (t) is increasing for t > t0 .
Also, from the lower bound for  given in the hypothesis, we have
e < t0 < N/p ≤ N/q < N =⇒ exp(g (N/q) − g (N)) < 1,
giving G(N/q) < G(N), so N is left abundant. 
Theorem 9.19 [35, theorem 6] Infinitely many colossally abundant numbers
are also left abundant.
Proof Use Lemma 9.14 to find a prime p sufficiently large that it satisfies
√
θ(p) > p + c p logloglog p,
where c is defined in Lemma 9.13, and set  := F(p, 1). Let N be the largest
colossally abundant number of parameter , so F(p, 1) =  and P(N) = p so
p | N. By Lemma 9.13 we get
√
log N ≥ θ(p) > p + c p logloglog p,
which implies
√
log(N/p) > p + c p logloglog p − log p > p + 1.
Because log(1 + t) ≥ t/(1 + t) for t > 0 we have
log(1 + 1/p) 1 1
 = F(p, 1) = ≥ >
log p (1 + p) log p (p + 1) log(p + 1)
1
> .
log(N/p) loglog(N/p)
By Lemma 9.18, N is left abundant. Since there are an infinite number of
primes p given by Lemma 9.14, there are an infinite number of corresponding
left abundant N. This completes the proof of the theorem. 
234 Left, Right and Extremely Abundant Numbers

Lemma 9.20 [35, lemma 8] Let p ≥ 5 be an odd prime and N the largest
colossally abundant number of parameter  with  < 1/(log N loglog N). Then
N is not left abundant.
Proof Firstly we have
p + 1 σ(p)
p = = .
p p
Therefore, since F(p, 1) = , P(N) = p and p divides N at most to the first
power, inequality (9.12) with q = p is an equality
σ(N/p) σ(N)
= 1+ ,
(N/p)1+ N
and similarly for inequality (9.13). Defining g (t) and t0 as in Lemma 9.18
we get g (t) decreasing when t < t0 and
G(N/p)
= exp(g (N/p) − g (N)).
G(N)
Therefore the upper bound for  of the hypothesis gives N/p < N < t0 , so
G(N) < G(N/p) and N is not left abundant. This completes the proof of the
lemma. 
Theorem 9.21 [35, theorem 7] There are infinitely many colossally
abundant numbers which are not left abundant.
Proof Use Lemma 9.14 to find a prime p which satisfies
√
θ(p) < p − c p logloglog p,
where c is defined in Lemma 9.13, and then let  := F(p, 1). Let N be the
largest colossally abundant number of parameter . Using Lemma 9.13 for
the upper bound for θ(p) we get, for p sufficiently large,
√ √ √
log N ≤ θ(p) + c p < p − c p logloglog p + c p < p.
Therefore
log(1 + 1/p) 1 1
= < < .
log p p log p log N loglog N
By Lemma 9.20, N is not left abundant. The result of the lemma then follows
√
since an infinite number of primes satisfy θ(p) < p − c p logloglog p. 
A final note is now given here regarding a counting result which uses many
of the ideas in this chapter. A proper left abundant number is defined to
be a left abundant number with more than two prime factors. In [35, theorem
14] the authors showed that there exists a positive constant c such that
 
log x
Q(x) := #{N ≤ x : N is a proper left abundant number} ≤ exp c .
loglog x
 9.7 Unsolved Problems 235

9.6 Extremely Abundant Numbers

A simple observation is that the first counterexample n to Robin’s inequality
larger than 5040 must satisfy G(n) ≥ eγ > G(m) for all m with 5040 < m < n,
and by Lemma 8.9 be superabundant. So if we note also that 10 080 is the
smallest superabundant number which is larger than 5040, we see n must be
extremely abundant, that is, a “champion” for the function G(n) restricted
to [10 080, ∞). Using only Grönwall’s and Robin’s theorems, Theorems 9.2
and 7.16 respectively, Sadegh Nazardonyavi and Semyon Yakubovich found
an equivalence to the Riemann hypothesis with a very straightforward proof.
Theorem 9.22 (Nazardonyavi–Yakubovich criterion) [123, theorem 7] The
Riemann hypothesis is true if and only if the number of extremely abundant
numbers is infinite.
Proof If RH is true and the number of extremely abundant numbers is finite,
let m be the largest extremely abundant number. Then for all n > m we have
G(n) ≤ G(m). But this implies lim supn→∞ G(n) ≤ G(m) < eγ , where the last
inequality follows from Theorem 7.16 since we are assuming RH. But this
contradicts Theorem 9.2, so therefore the set of extremely abundant numbers
is infinite.
If RH is false however, then by Theorem 7.16 again, there exists an m ≥
10 080 with G(m) ≥ eγ . By Theorem 9.2, μ := supn≥10 080 G(n) is finite, and
there exists a maximal m0 ≥ 10 080 such that
G(m0 ) = μ ≥ eγ .
Any integer n > m0 satisfies G(n) < μ = G(m0 ), so n fails to be extremely
abundant. Therefore the set of extremely abundant numbers is finite, and the
proof is complete. 

9.7 Unsolved Problems

(1) Show that the integer 4 is the only left abundant number which is the
power of a prime [35].
(2) The smallest left abundant number that is not the power of a prime or a
product of at most two primes (called proper) is [35]
ν := 24 · 33 · 52 · 7 · 11 · 13 · 17 = 183 783 600.
(3) Show that the number ν defined in problem (2) is not right abundant.
(4) Show that any right abundant number n > 5 is even [35].
(5) Improve the upper bound for Q(x) given in [35, theorem 14].
(6) Find a non-trivial lower bound for Q(x).
(7) Find an upper bound for the number of extremely abundant numbers less
than or equal to real positive x.
 10
Other Equivalents to the Riemann Hypothesis

10.1 Introduction
In this chapter a selection of additional equivalences to RH are presented.
This is to show that the techniques used to derive those involving ψ(x), θ(x),
σ(n) and ϕ(n) are not unique, but that other methods and equivalences have
been derived. The equivalences are often surprising. Some are intricate, but
others are easy consequences of other equivalences. As before, in the main,
proofs are included to show depth or shallowness, and to provide the reader
with techniques which could be further developed. There is one notable
exception, namely the symmetric group application of Section 10.10 wherein
only the introductory proofs and ideas are presented. The selection is by no
means exhaustive, and tends to stray away from the arithmetic into other
realms, including the analytic, which is more property the domain of Volume
Two [32].
To begin, in Section 10.2 there is the interesting inequality, the Shapiro
criterion from 1976, which is Theorem 10.1. It is based on an arithmetic
function δ(n) which satisfies
 n
ψ(t) dt = log δ(n)
0

and has the form

 2
  1 n2 
 −  < 36n3 , n ∈ N,
1≤ j≤δ(n) k 2 

which is equivalent to the Riemann hypothesis.

In Section 10.3 there is the famous Farey fraction equivalence. This is
“doubly deep”, since it depends in anon-trivial way on the equivalence of
Littlewood, Theorem 4.16. Let m := 1≤ j≤n ϕ( j) and Fn := ( f jn : 0 ≤ j ≤ m) be
the nth Farey series, i.e. the set of rational numbers in [0, 1] in lowest terms,

236
 10.1 Introduction 237

with denominators not greater than n, in increasing order. Then we have the
theorem of Franel and Landau of 1924:
Theorem 10.6 (Franel–Landau criterion) The Riemann hypothesis is equiva-
lent to the statement that for all  > 0, as n → ∞,
m 
  n j 
 f j −  n1/2+ .
j=1
m

The equivalence in Section 10.4 is also a consequence of Littlewood’s

theorem, Theorem 4.16. It relies on the direct relationship between the
determinant of the matrix Rn describing divisibility and a sum over μ( j),
values of the Möbius function. The matrix Rn = (ri, j ) is such that the element
ri, j is zero except where j = 1 or i | j, in which case it is 1. The equivalence
was published by Redheffer in 1976. The relationship to the sum and the
following equivalence are easy to derive.
Theorem 10.9 (Redheffer criterion) The Riemann hypothesis is equivalent to
the statement that for all  > 0, as n → ∞,
det Rn n1/2+ .
The next equivalence, described in Section 10.5, is an easy consequence of
Theorem 10.9. It describes a property of particular finite directed graphs, the
so-called “divisibility graphs”. These are defined for each natural number n
by labelling vertices 1 through n and placing a directed edge between vertex
i and vertex j when i divides j. An additional set of edges goes from each j to
vertex 1. A closed cycle in one of these graphs is called “even” if the number
of edges is even, and “odd” otherwise. The number of even and odd cycles is
represented by e(Gn ) and o(Gn ) respectively. We have:
Theorem 10.10 (Divisibility graph criterion) The Riemann hypothesis is true
if and only if for all  > 0 we have
|o(Gn ) − e(Gn )| n1/2+
as n → ∞.

The function defined by ∞ n=1 (−1)
n+1 −s
n , which converges conditionally
for s > 0, is sometimes called η(s). It provides a very easy-to-derive
equivalence which is given in Section 10.6:
Theorem 10.12 (Dirichlet eta criterion) The Riemann hypothesis is equiva-
lent to the zeros of η(s) with 0 < σ < 1 all having real part 12 .
Now a different set of equivalences are described, each directly or indi-
rectly using inequalities concerning ζ(s). In Section 10.7 a famous statement
of an equivalence due originally to Speiser, who gave in 1934 a geometric/
238 Other Equivalents to the Riemann Hypothesis

topological but not rigorous proof [161], that RH is equivalent to all of the
zeros of the derivative of ζ(s) in the critical strip being on or to the right
of the critical line. Part of this statement was given rigorous justification by
Levinson and Montgomery in 1974:
Theorem 10.16 (Levinson–Montgomery criterion) If the Riemann hypothe-
sis is true, all non-trivial zeros of the derivative of the Riemann zeta function
ζ  (s) in the critical strip have real part greater than or equal to 12 . If the
Riemann hypothesis is false, the number of zeros of ζ  (s) in the critical strip,
with positive imaginary part less than T and real part less than 12 , is O(log T ).
A simple inequality equivalence was found by Spira in 1965:
Theorem 10.18 (Spira criterion) The Riemann hypothesis is equivalent to the
statement that if σ > 12 and t ≥ 6.29073 then |ζ(1 − s)| > |ζ(s)|.
As part of a re-examination of a lemma of Lagarias, the present author was
able to improve his Riemann zeta imaginary “zero gap” explicit estimate:
Theorem 10.32 If t ≥ 168π then ζ(s) has a zero with imaginary part satisfying
|t − γ| ≤ 1.5.
Later, in Section 10.9 the details are given of work of Lagarias and
Garunkstis on an inequality equivalence to RH given in terms of the
logarithmic derivative of Riemann’s function ξ(s).
Theorem 10.35 (Garunkstis–Lagarias criterion) Let a satisfy 12 ≤ a ≤ 1. If ζ(s)
has no zeros for σ > a then for s > a we have
%  &
ξ (σ) ξ (σ + it)
= inf  :t∈R .
ξ(σ) ξ(σ + it)
In Section 10.10 a summary and proofs of foundational results are
presented of a completely different type of equivalence. Nonetheless it was
considered that this should be included because the methods used related
strongly to those used in Chapters 5 and 7. Let g(n) be the maximum
multiplicative order of an element of the symmetric group S n . Then the
theorem of Massias, Nicolas and Robin of 1988 is:
Theorem 10.42 (Symmetric group criterion) The Riemann hypothesis is
equivalent to the statement that for all n sufficiently large we have

log g(n) < li−1 (n) ,
where li(n) is the Cauchy principal value of the logarithmic integral,
 n
dt
li(n) := ,
0 log t
and li−1 (n) represents its functional inverse.
 10.2 Shapiro’s Criterion 239

Finally in Section 10.11 we touch on the large body of work which goes
under the heading the “Hilbert–Pólya conjecture”. Broadly speaking, this is
the idea that there is a physical system which underlies the zeta function
so that the critical zeros are associated with eigenvalues of a self-adjoint
transformation on a (dense subset of) a Hilbert space.
The following RHpack (see Appendix B) functions relate to the material
in this chapter: FareyFractions, RedhefferMatrix, DivisibilityGraph, Plot-
DivisibilityGraph, Xi, Lambda, S, IntegerHeight and ZetaZeroCount.

10.2 Shapiro’s Criterion

This criterion was quite a minor part of a far-reaching summary of work on
Hilbert’s 10th problem by Davis, Matijasevic̆ and Robinson [44]. Published
in 1976, it contains an arithmetic inequality equivalent to RH which is
superficially simpler than Robin’s, and certainly much simpler to derive.
Recall the definitions. For x ≥ 1,
 
ψ(x) := Λ(n) = log p ≤ x log x,
1≤n≤x pα ≤x
 x
ψ1 (x) := ψ(t) dt.
1

If n = pα is a prime power define κ(n) := p. Otherwise set κ(n) := 1. For

x > 0 define

δ(x) := κ( j),
n<x j≤n

for example
δ(11) = 219 · 310 · 56 · 74
and
δ(102) = 2486 · 3288 · 5174 · 7148 · 1191 · 1389 · 1785 · 1983
· 2379 · 2973 · 3171 · 3765 · 4161 · 4359 · 4755 · 5349
· 5943 · 6141 · 6735 · 7131 · 7329 · 7923 · 8319 · 8913 · 975 · 101.
Note that with this definition we have δ(1) = 1, and for n ∈ N, ψ1 (n) = log δ(n).
Theorem 10.1 (Shapiro criterion) The inequality
 2
  1 n2 
 −  < 36n3 , n ∈ N, (10.1)
1≤ j≤δ(n) k 2

is equivalent to the Riemann hypothesis.

240 Other Equivalents to the Riemann Hypothesis

Proof (1) First assume RH is true and let S be the non-trivial zeros of ζ(s).
Then for ρ ∈ S we have 1 − ρ = ρ so by Lemma 2.10 we get
 1  1  1
≤ = = γ + 2 − log(4π).
ρ∈S
|ρ| · |1 + ρ| ρ∈S
|ρ| 2
ρ∈S
ρ(1 − ρ)

Using the explicit formula from Theorem 2.2, by RH we have |xρ+1 | = x3/2 .
We also have the sum
 ∞
1
= log 2,
n=1
2n(2n − 1)
so we get for x ≥ 1
  ⎛       ⎞
 x2  ⎜⎜⎜ ζ ζ ⎟
 + log 2⎟⎟⎟⎟ .
+   + 
1 (0) (−1)
ψ1 (x) −  ≤ x ⎜⎜⎜⎝
3/2
⎟⎠
2 |ρ| · |1 + ρ|
ρ∈S
ζ(0)   ζ(−1) 
Thus, by Lemma 2.5 and the evaluation |ζ  (−1)/ζ(−1)| = 1.985 . . . < 2 we can
write
 
 x2 
ψ1 (x) −  ≤ x (γ + 2 − log(4π) + log(2π) + 2 + log 2) < 5x .
3/2 3/2
2
But for all n ∈ N we have

n
1 
n
1
−1 + < log n < .
j=1
j j=1
j
Since for x ≥ 1 we have ψ1 (x) = log δ(x), we get
 
  1 
 − ψ1 (x) < 1.
1≤ j≤δ(x) j 
Therefore
 
  1 x2 
 −  < 1 + 5x3/2 ≤ 6x3/2 ,
 j≤δ(x) j 2
which gives the inequality (10.1).
(2) Now assume inequality (10.1) is true for all n ∈ N. Then for x a positive
integer we have ψ1 (x) = log δ(x). Hence |ψ1 (x) − x2 /2| < 1 + 6x3/2 . Since

ψ(x) ≤ log x ≤ x log x ≤ x3/2 ,
j≤x

this implies for real x ≥ 1

     x
 x2   x2  1
ψ1 (x) −  ≤ ψ1 (x) − + ψ(t) dt + |x2 − x2 |
2 2  x 2
< ψ(x) + 1 + 6x3/2 + x ≤ 9x3/2 .
 10.3 Farey Fractions 241

Therefore
 
 ψ1 (x) − x2 /2  9
  ≤ s+1/2 .
x s+2 |x |
(3) Next, taking the logarithmic derivative of the Euler product for ζ(s),
and using partial summation, the definition of ψ1 (x) and integration by parts,
we can derive for s > 1
 ∞
ζ  (s) ψ(x)
− =s dx
ζ(s) x s+1
 ∞
1
dψ1 (x)
=s
x s+1
1
 ∞
ψ1 (x)
= s(s + 1) dx.
1 x s+2
Therefore
 ∞
ζ  (s) s(s + 1) ψ1 (x) − x2 /2
− − = s(s + 1) dx. (10.2)
ζ(s) 2(s − 1) 1 x s+2
The inequality derived in Step (2) implies that the integral of (10.2) converges
to a function of s which is holomorphic on s > 12 . But this means ζ(s) has
no zero in this region, so RH is true. This completes the proof. 
Note, as far as numerical checking of inequality (10.1) is concerned, that
the expression  
 2 log δ(n) − 1 < 12
 n2  √n (10.3)

is better.
A subset T ⊂ N is computable if there is an algorithm to determine in
a finite number of steps whether or not an arbitrary given natural number
is a member of T [44]. From the theory of algorithms it follows that RH is
decidable, i.e. its truth or negation are able to be proved. In addition, using the
inequality (10.1) it would be possible to write down an explicit diophantine
equation with integer coefficients which has no solutions in integers if and
only if RH is true. See Davis et al. [44].

10.3 Farey Fractions

For n ∈ N let Fn be the finite sequence of rationals in the interval [0, 1] with
denominators in {1, 2, . . . , n}, in lowest terms, sorted in increasing order. For
example F3 = ( 01 , 13 , 12 , 23 , 11 ). These are the so-called Farey series. We write
f jn , 0 ≤ j ≤ m, for the jth term in the finite sequence Fn , where the number of
terms is given by Lemma 10.2.
242 Other Equivalents to the Riemann Hypothesis

This equivalence is not at all obvious, even though it relies on the classical
Littlewood criterion, Theorem 4.16. Before giving the proof we need several
lemmas. The proof is taken in the main from Edwards [51] which was taken
from the work of Franel and Landau of 1924, based on, apparently, the work
of the former [64].
Lemma 10.2 We have for all n ≥ 1

n
#Fn = 1 + ϕ( j).
j=1

Proof For 1 ≤ j ≤ n, the number of distinct rationals in (0, 1] with lowest-

terms denominator j is ϕ( j). The leading 1 in the sum accounts for the
rational 0 = 0/1. 

Below in this section we set m = nj=1 ϕ( j), the number of non-zero
elements in Fn . Also recall the definition for x > 0,

M(x) := μ(n),
n≤x

where μ(n) is the Möbius function.

The statement equivalent to the Riemann hypothesis says, roughly, that
when n is large the Farey fractions of order n are reasonably uniformly
distributed.
Lemma
 10.3 Let x > 1 be real and n = x. If m = #Fn − 1 and M(x) =
1≤n≤x μ(n), then for any function g : R → R, with g(x + 1) = g(x) for all x, we
have
m ∞  k  j  x
g( f jn ) = g M . (10.4)
j=1 k=1 j=1
k k

Proof Let χ(x) be the characteristic function

 of [1, ∞), i.e. χ(x) = 1 for x ≥
1 and χ(x) = 0 for x <1. Then M(x) = n∈N μ(n)χ(x/n), which by Möbius
inversion gives χ(x) = n∈N M(x/n). Given any rational number u/v with 0 ≤
u ≤ v and (u, v) = 1 then
u    
2u 3u
g =g =g = ···
v 2v 3v
occurs on the right-hand side of (10.4) with coefficient
 x x x  x
M +M +M + ··· = χ ,
v 2v 3v v
which is 1 if v ≤ x and 0 if v > x. This is exactly the coefficient of g(u/v) on
the left-hand side of (10.4). 
 10.3 Farey Fractions 243

For n ∈ N let Bn (x) be the nth Bernoulli polynomial. One way of defining
Bn (x) is that it is the unique polynomial which satisfies
 x+1
Bn (t) dt = xn .
x

These polynomials have many properties which are useful. The first few are
B0 (x) = 1,
B1 (x) = x − 12 ,
B2 (x) = x2 − x + 16 ,
B3 (x) = x3 − 32 x2 + 12 x.
Here we use also the related period-1 function bn (x) := Bn (x − x).
Lemma 10.4 For all n, k ∈ N and u ∈ R we have
    
1 k−1
Bn (ku) = kn−1 Bn (u) + Bn u + + · · · + Bn u + ,
k k
with the same identity holding with Bn replaced by bn .
Proof First we derive the identity for Bn (u). We have for any real x, changing
variables and then splitting the integral domain into k subintervals of equal
length,
 x+1/k 
1 kx+1
Bn (kt) dt = Bn (u) du = kn−1 xn
x k kx
 x+1
= kn−1 Bn (t) dt
    
x
 x+1/k
1 k−1
= k n−1
Bn (t) + Bn t + + · · · + Bn t + dt.
x k k
Because the integrands on both sides of this equation are polynomials, they
must be equal for every value of t. This gives the first identity.
Now let 0 ≤ u < 1/k ≤ 1 so 0 ≤ ku < 1 and we can write, since u + j/k < 1
for all j ≤ k − 1,
  
k−1
bn (ku) = Bn (ku) = kn−1 Bn (u) + · · · + Bn u +
k
  
k−1
= kn−1 bn (u) + · · · + bn u + ,
k
so the identity in bn (u) is true for these values of u. If 0 ≤ u < 1 there is an l < k
so u = l/k + (u − l/k) with 0 ≤ u − l/k < 1/k. Then, since bn (u) has period 1,
       
l l k−l−1
bn (ku) = bn k u − =k n−1
bn u − + · · · + b n u +
k k k
244 Other Equivalents to the Riemann Hypothesis
    
k−l k−l+1
=k n−1
bn u + + · · · + bn u +
k k
    
1 k−1
=k n−1
bn (u) + bn u + + · · · + bn u + ,
k k
where, for example, bn (u) occurs as the (l + 1)th term in the sum in the second
line of the derivation. This completes the proof of the lemma. 
Now define the “sawtooth function” b(u) := b1 (u) = u − u − 12 .
Lemma 10.5 For all j, k ∈ N we have
 1
( j, k)2
I( j, k) := b( ju)b(ku) du = ,
0 12 jk
where ( j, k) on the right-hand side represents the greatest common denomi-
nator (GCD) of j and k.
Proof First let k = 1. Then, since b(u) has period 1, and using Lemma 10.4
with n = 1,
   j−1   
1 j u 1 1 k u
I( j, 1) = b(u) b du = b(k + u) b + du
j 1 j j k=0 0 j j
 1    1
1 u 1 1
= b(u) b j · du = (u − 12 )2 du = .
j 0 j j 0 12 j
Next in case ( j, k) = 1 note that kv/ j modulo 1 when 0 ≤ v ≤ j − 1, since k has
a modulo j inverse, gives each rational v/ j exactly once, so that
  
1 1 ku
I( j, k) = b(u) b j · du
j 0 j

1 1 1 1
= b(u) b(ku) du = I(k, 1) = .
j 0 j 12 jk
In the general case let c = ( j, k), the GCD, and set j = cα and k = cβ so
(α, β) = 1. Then
 1
I( j, k) = b(cαu) b(cβu) du
0
 c
1
= b(αu) b(βu) du
c 0
 1
= b(αu) b(βu) du
0
1 c2
= I(α, β) = = .
12αβ 12 jk
This completes the proof. 
 10.3 Farey Fractions 245

We are now able to give the proof of the Franel–Landau criterion

for RH. Note the proof’s explicit dependence on Littlewood’s theorem,
Theorem 4.16. Recall that

n
m := ϕ( j) = #Fn − 1.
j=1

Theorem 10.6 (Franel–Landau criterion) [51, section 12.2] The Riemann

hypothesis is equivalent to the statement that for all  > 0, as n → ∞,
m 
  n j 
 f j −  n1/2+ .
mj=1

Proof (1) The proof is in five steps. First assume the statement is true. Fix
n ∈ N and, for 0 ≤ j ≤ m, write f j instead of f jn . Also use the same notation as
in Lemmas 10.2, 10.3 and 10.4. Let g(u) := exp(2πiu) in identity (10.4) to get
m ∞  k  x
e2πi f j = e2πi j/k M .
j=1 k=1 j=1
k
For 1≤ j ≤ m set δ j := f j − j/m, the so-called “discrepancy” at j. Now for
k > 1, kj=1 exp(2πi j/k) = 0 and for k = 1 the sum is 1 = k, so the right-hand
side of this expression simplifies directly to M(x). Thus
   
 m
 m   j 
2πi f j 
|M(x)| =  e  =  exp 2πi + δ j 
 j=1   j=1 m 
 
 m m

=  e 2πi j/m 2πiδ j
(e − 1) + e2πi j/m 
 j=1 j=1


m 
m 
m
≤ |e 2πiδ j
− 1| = 2 |sin(πδ j )| ≤ 2π |δ j | x1/2+ .
j=1 j=1 j=1

So by Theorem 4.16, the Riemann hypothesis is true.

(2) Now assume the Riemann hypothesis is true. Recall
b(u) := b1 (u) = u − u − 12 ,
and set g(u) to b(u) in Lemma 10.4. We get
    
1 k−1
b(ku) = k b(u) + b u + + · · · + b u +
0
.
k k
Therefore, using (10.4),
 m ∞  k  j  x
G(u) := b(u + f j ) = b u+ M
j=1 k=1 j=1
k k
∞  x
= b(ku)M .
k=1
k
246 Other Equivalents to the Riemann Hypothesis

Next in Steps (3) and (4) we use the definition  1 of G(u) and the expression
we have derived to evaluate the integral I2 := 0 G(u)2 du.
(3) Since Fn = {1 − f j : 0 ≤ j ≤ m} we have
m m
G(u) = b(u + 1 − f j ) = b(u − f j )
j=1 j=1

and b(0) = − 12 , b(0−) = 12 . This shows that G(u) decreases by 1 at each f j and
increases like m · u as u increases from f j to f j+1 . Also because

m−1 
m−1 
m−1 
m−1 
m−1
( f j − 12 ) = b( f j ) = b(1 − f j ) = (− f j + 12 ) = − b( f j ),
j=1 j=1 j=1 j=1 j=1

we have m−1 j=1 b( f j ) = 0 so limu→0+ G(u) = − 2 . All this goes to show that we
1

can write, for f j < u < f j+1 ,

G(u) = m · u − j − 12 .
We now use this formula to evaluate the integral I2 explicitly.
Recall f0 = 0 and fm = 1. Performing the integration (recall δ j = f j − j/m),
using δ0 = δm = 0 and shifting the second sum in the third line,
m  fj  2
1
I2 = mu − j + 1 − du
j=1 f j−1 2
f
m
(mu − j + 12 )3  j
= 
3m f
j=1 j−1
⎛  3  3 ⎞
1  ⎜⎜⎜ 1 ⎟⎟⎟
m
1
= ⎜⎝ m f j − j + − m f j−1 − ( j − 1) ⎟⎠
3m j=1 2 2
m ⎛ 3  3 ⎞
1  ⎜⎜⎜ 1 1 ⎟⎟⎟
= ⎜
⎝ mδ j + − mδ j − ⎟⎠
3m j=1 2 2

m
1
=m δ2j + .
j=1
12

(4) Now consider the representation G(u) = ∞ k=1 b(ku)M(x/k) from Step
(2). Since this sum is finite we can interchange sums and integrals to write
∞  ∞  1     
x x
I2 = b( ju) b(ku) du M M
j=1 k=1 0 j k
∞  ∞    x
( j, k)2 x
= M M ,
j=1 k=1
12 jk j k
where we have used Lemma 10.5 to evaluate the integrals.
 10.4 Redheffer Matrix 247

(5) For given j, k ≥ 1 let c = ( j, k) and set j = αc, k = βc. By the Riemann
hypothesis and Theorem 4.16, we can assume for every  > 0 there is an
implied constant, depending only on , such that M(x) x1/2+ for all x > 0.
Using this to bound our expression for I2 gives
 ∞  ∞  1/2+
( j, k)2 x2
I2
j=1 k=1
12 jk jk

∞ 
∞ 
∞
c2
x 1+2

α=1 β=1 c=1

c3+2 α3/2+ β3/2+
x1+2 .
We now combine the expression for the integral I2 , we derived in Step (3)
with this estimate to get

m
m δ2j x1+2 .
j=1

Finally the Cauchy–Schwarz inequality gives

⎛ m ⎞1/2 ⎛ m ⎞1/2 ⎛ m ⎞1/2
 m 
m ⎜⎜⎜ ⎟⎟⎟ ⎜⎜⎜ ⎟⎟⎟ ⎜⎜⎜  ⎟⎟⎟
|δ j | = |1 · δ j | ≤ ⎜⎜⎜⎝ 12 ⎟⎟⎟⎠ ⎜⎜⎜⎝ δ2j ⎟⎟⎟⎠ = ⎜⎜⎜⎝m δ2j ⎟⎟⎟⎠ x1/2+ ,
j=1 j=1 j=1 j=1 j=1

and we have verified the condition of the theorem. 


We explore the structure of the sums 1≤ j≤m | f j − j/m|, called Farey sums,
in Figures 10.1 and 10.2.√ Each plotted point in Figure 10.1 represents the
sum for Fn divided by n. Each plotted point in Figure 10.2 represents the
size of an individual term in the sum.
Remark The equivalence between Farey series and RH has been one of
the more popular equivalences. It has stimulated a considerable amount of
additional work. Some of this involves tight estimates for functions mapped
over Farey series, where the functions satisfy given conditions. Others
include relations between the prime number theorem and Farey series. See
for example [8, 85, 86, 87, 116, 117, 181].
There is another mysterious property connecting Farey series with zeta
functions. It was observed in [18] that the sums of the values of the Hurwitz
zeta function over the Farey fractions of a given order, other than zero, at a
zero of the zeta function, are all zero.

10.4 Redheffer Matrix

Fix n ∈ N and define an n × n matrix with 0, 1 entries by ri, j = 1 if j = 1 or
i | j. Otherwise ri, j is set to 0. We denote by Rn the matrix with entries ri, j ,
248 Other Equivalents to the Riemann Hypothesis
FareySum

0.4

0.3

0.2

0.1

n
0 20 40 60 80 100
√
Figure 10.1 The Farey sums over n with 1 ≤ n ≤ 100.

FareySumj

0.008

0.006

0.004

0.002

j
0 500 1000 1500 2000 2500 3000

Figure 10.2 Individual terms in the Farey sum for n = 101.

and call it the Redheffer matrix of order n. The determinant of this matrix
was first evaluated by Ray Redheffer in 1976 [138]. It was also evaluated,
using a combinatorial argument, by Herbert Wilf in 2006 [180]. Here we
present the method of Bryan Gillespie from 2011 [67], which better reveals
the underlying structure and gives a vast generalization.
 10.4 Redheffer Matrix 249

Let f : N → C be an arithmetic function with f (1)  0. Then [6, theorem

2.8] there is a unique arithmetic function f −1 , the so-called Dirichlet inverse
of f , such that
f ∗ f −1 = f −1 ∗ f = I,
where I(1) = 1 and I(n) = 0 for n > 1 is the identity for Dirichlet
multiplication, which is as usual denoted “∗”. Indeed, f −1 can be defined
recursively by f −1 (1) := 1/ f (1) and for n > 1
1   n  −1
f −1 (n) := − f f (d).
f (1) d|n, d<n d

If f is completely multiplicative this looks much simpler. In this case, for all
n ≥ 1 we have [6, theorem 2.17] f −1 (n) = μ(n) f (n).
Given any arithmetic function f : N → C with f (1)  0, and a natural
number n, define an n × n matrix by Rn ( f ) := (ri, j ) where, for 1 ≤ i, j ≤ n,
ri, j = f ( j/i) if i | j, ri, j = 1 if i > 1 and j = 1 and ri, j = 0 otherwise.
Lemma 10.7 [67, theorem 2.7] Let f be an arithmetic function with f (1)  0.
Then
⎛ ⎞
⎜⎜  n ⎟⎟⎟
n⎜
det Rn ( f ) = f (1) ⎜⎜⎜⎝1 + f ( j)⎟⎟⎟⎠ .
−1

j=2

Proof Define an upper triangular n × n matrix with complex coefficients

S n ( f ) = (si, j ) by si, j := f ( j/i) if i | j and si, j = 0 otherwise. Then S n ( f ) has
determinant f (1)n so is invertible in C. Note also that if f and g are two such
arithmetic functions then S n ( f ) × S n (g) = (ti, j ) where “×” is normal matrix
multiplication and ti, j = 0 if i  j and ti, j = ( f ∗ g)( j/i) if i | j. In other words
S n ( f ) × S n (g) = S n ( f ∗ g).
Therefore, if A is the n × n matrix with 0 in all positions except the leading
column, which has 1 in every position except the first, then
S n ( f −1 ) × Rn ( f ) = S n ( f −1 ) (S n ( f ) + A) = In×n + S n ( f −1 )A, (10.5)
where In×n is the n × n identity matrix. Then because of their
n shape S n ( f −1 )A
−1
is the zero matrix except for the (1, 1) element, which is j=2 f ( j), and the
(i, 1) elements with i > 1. Taking determinants of each side of (10.5) gives

n
f (1)−n det Rn ( f ) = 1 + f −1 ( j)
j=2

and the proof is complete. 

Now let f (n) = 1 for all n ∈ N so f −1 (n) = μ(n) in Lemma 10.7. We get:
250 Other Equivalents to the Riemann Hypothesis

Corollary 10.8 For all n ∈ N,


n
det Rn = μ( j).
j=1

Theorem 10.9 then follows directly from Theorem 4.16.

Theorem 10.9 (Redheffer criterion) [138] The Riemann hypothesis is
equivalent to the statement that for all  > 0, as n → ∞,
det Rn n1/2+ .

10.5 Divisibility Graph

The divisibility matrix Rn gives rise also to a graph-theory-based equivalence
to RH. These graphs are rather special, and designed to make the equivalence
an easy consequence of Section 10.4.
Form a directed graph Gn on the vertices labelled {1, . . . , n} by inserting a
directed edge from i to j if i | j and i  j, and a directed edge from j to 1
for 2 ≤ j ≤ n. We call Gn the divisibility graph of order n (see for example
Figure 10.3, which is G6 ).
A cycle of a directed graph is a sequence of edges such that the final point
of each edge is followed by an edge with the same initial point, other than the
last edge, whose final point corresponds to the initial point of the first edge.

5 1

Figure 10.3 The divisibility graph for n = 6.

 10.5 Divisibility Graph 251

In this context, other than the first and last, the vertices in a cycle are distinct.
A cycle is called odd if it consists of an odd number of edges. Similarly even
cycles are defined. Let o(Gn ) be the number of odd cycles of Gn and e(Gn )
the number of even cycles. Then we have the following parity equivalence to
RH:
Theorem 10.10 (Divisibility graph criterion) [9] The Riemann hypothesis
is true if and only if for all  > 0, as n → ∞, we have
|o(Gn ) − e(Gn )| n1/2+ .
Proof Let In be the n × n identity matrix and set Bn := Rn − In . Let Gn be
the directed graph with adjacency matrix Bn = (bi, j ) so the vertices of Gn are
{1, . . . , n} and an edge in Gn connects i to j if and only if bi, j = 1. Then Gn is a
divisibility graph. Define a polynomial in Z[x] by
qn (x) := det(Bn + xIn ).
Each cycle
1 → i1 → i2 → · · · → ik−1 → 1
in Gn , starting and ending at the vertex 1, having length k and distinct i j  1,
contributes to the determinant of Bn + xIn a term with value (−1)k−1 times the
determinant of the (n − k) × (n − k) submatrix which is obtained by deleting all
rows and columns indexed 1, i1 , i2 , . . . , ik−1 from that matrix. The submatrix is
upper triangular, so its determinant is simply xn−k , the product of terms on the
diagonal. Every cycle includes 1. Therefore each k-cycle in Gn contributes
(−1)k−1 xn−k to det(Bn + xIn ). In addition to cycles, the determinant includes
the diagonal xn and no other non-zero terms.
Now let c(n, k) be the number of distinct cycles of length 2 ≤ k ≤ n in Gn , so

n
qn (x) = xn + (−1)k−1 c(n, k)xn−k .
k=2

Then

n
det Rn = qn (1) = 1 + (−1)k−1 c(n, k) = 1 + o(Gn ) − e(Gn ).
k=2

Therefore
|o(Gn ) − e(Gn )| = |det(Rn ) − 1|.
By Theorem 10.9, it follows that RH is true if and only if for all  > 0, |o(Gn )−
e(Gn )| n1/2+ , which completes the proof. 
252 Other Equivalents to the Riemann Hypothesis

10.6 Dirichlet Eta Function

Perhaps the easiest equivalence to RH to derive is in terms of the so-called
Dirichlet eta function. An advantage of using this function is that its Dirichlet
series representation converges conditionally in the half plane σ > 0, in
particular, in all of the open critical strip.
Let s ∈ C and define for σ = s > 0

η(s) := (1 − 21−s )ζ(s).

Then we have:
∞
Lemma 10.11 If σ > 0 the series n=1 (−1)
n+1
/n s converges and has sum
η(s).

Proof If s = σ + it with σ, t real, then the series converges absolutely for

σ > 1 and therefore represents a holomorphic function in that half plane.
Indeed,

∞
(−1)n+1  ∞
1 ∞
1
=− +
n=1
ns n=1
(2n) s
n=1
(2n − 1) s
∞
1 1  1
∞ ∞
= −2−s + −
n=1
n s n=1 n s n=1 (2n) s
∞
1 1
∞
= −21−s +
n=1
n s n=1 n s
= (1 − 21−s )ζ(s) = η(s).

If σ > 0 then, by the alternating real series test, the series


∞
(−1)n+1
n=1
nσ

converges. Therefore, by [6, theorem 11.8], it converges for all s ∈ C with

σ = s > 0. It diverges at σ = 0 so the abscissa of convergence of the series
is σc = 0. Therefore we can continue η(s) with this series representation into
the half plane s > 0. 

Theorem 10.12 (Dirichlet eta criterion) The Riemann hypothesis is equiva-

lent to the zeros of η(s) with 0 < σ < 1 all having real part 12 .

Proof A simple calculation shows that none of the zeros of the factor 1 − 21−s
have real part in (0, 1). 
 10.7 The Derivative of ζ(s) 253

10.7 The Derivative of ζ(s)

In this section an “almost equivalence”, due originally to Speiser, is
described. In 1934 [161] he gave a geometric/topological but not rigorous
proof that RH is equivalent to all of the zeros of the derivative of ζ(s) in the
critical strip being on or to the right of the critical line. Part of this statement
was given rigorous justification by Spira in 1973, who proved that RH implies
the derivative ζ  (s) has no zero in the open left half of the critical strip [163];
see also Trudgian [171]. Levinson and Montgomery went part way to proving
the converse in their 1974 paper [107], by showing that there were at most
O(log T ) zeros of the derivative up to height T in the left half.
Let us start with some definitions:
N(T ) := #{ρ ∈ C : ζ(ρ) = 0, 0 < ρ ≤ T },
N − (T ) := #{ρ ∈ C : ζ(ρ) = 0, 0 < ρ < 12 , 0 < ρ ≤ T },
N1− (T ) := #{ρ ∈ C : ζ  (ρ) = 0, 0 < ρ < 12 , 0 < ρ ≤ T }.
In what follows we also need a well-known approximation to the
logarithmic derivative of the gamma function. This is defined for all
s ∈ C \ {0, −1, −2, . . .} by
 ∞
Γ(s) = t s−1 e−t dt.
0

Lemma 10.13 [51, section 6.3] Let w = u + iv ∈ C. Then

Γ (w) 1
= log w − + R,
Γ(w) 2w
where |R| ≤ 1/(10|w|2 ), |w| ≥ 2 and u ≥ 0.
This follows for example from the representation for log Γ(s) given in the
proof of Lemma 10.17.
The proof of the following very useful lemma, Jensen’s formula, is given
in Theorem B.5 in Volume Two [32] or [51, section 2.2].
Lemma 10.14 (Jensen’s formula) Let R > 0 and let f (z) be a function which
is holomorphic inside and on the boundary of the disc Ω = {z ∈ C : |z| ≤ R}
with no zeros on the boundary or at the centre. Let the zeros inside the disc
be labelled z1 , z2 , . . . , zn , where a zero of multiplicity m is included m times.
Then, if we let n(r) represent the number of zeros inside a circle with centre
0 and radius r with no zero on the boundary:
 R  n  2π
n(r) 1
dr = n log R − log |zi | = log | f (Reiθ )| dθ − log | f (0)|.
0 r i=1
2π 0
254 Other Equivalents to the Riemann Hypothesis

Recall Table 2.1 in Section 2.2. Let T ≥ e and denote by N(T ) the number
of zeros, including multiplicity, of ζ(s) in the rectangle [0, 1] × [0, T ] ⊂ C.
Then
  
 T T T 7 
N(T ) − log − +  < 0.122 log T + 0.278 loglog T + 2.510.
2π 2π 2π 8 
(10.6)

Lemma 10.15 For all T > 0 we have N1− (T ) = N − (T ) + O(log T ). If it is false

to say N − (T ) ≥ T + O(1) as T → ∞ then there exists a sequence T j → ∞ such
that for all j we have N1− (T j ) = N − (T j ).
Proof Denote the zeros of ζ(s) in the critical strip by ρ. Then because
s   s

−s/2
(s − 1)π Γ + 1 ζ(s) = ξ(s) = ξ(0) 1− ,
2 ρ
ρ

if s ≥ 0 and s is not a zero of zeta, taking logs, differentiating and then

taking the real part, we get
ζ  (s) 1 1 1 Γ (s/2 + 1)  1
 = − + log π −  + . (10.7)
ζ(s) s−1 2 2 Γ(s/2 + 1) ρ
s−ρ

Let s = σ + it. If ρ = β + iγ is a zero of ζ(s) and β < 12 so is 1 − ρ̄ = 1 − β + iγ

and
   
1 1 1 (t − γ)2 + (σ − 12 )2 − ( 12 − β)2
 + = −2 − σ .
s − ρ s − 1 + ρ̄ 2 |s − ρ|2 |s − 1 + ρ̄|2
If we set
 (t − γ)2 + (σ − 1 )2 − ( 1 − β)2  1
I1 := 2 2 2
+ , (10.8)
β<1/2
|s − ρ| 2 |s − 1 + ρ̄|2
β=1/2
|s − ρ|2

then we have
 1  
1
I :=  = − − σ I1 . (10.9)
ρ
s−ρ 2
By Lemma 10.13 we have an estimate for the logarithmic derivative of Γ(w)
where w = u + iv:
Γ (w) 1 1
= log w − + R, where |R| ≤ , |w| ≥ 2, u ≥ 0.
Γ(w) 2w 10|w|2
Therefore for |s| ≥ 3 and σ ≥ 0 we can write
Γ (s/2 + 1) σ+2 2
 = log |1 + s/2| − + R1 , where |R1 | ≤ .
Γ(s/2 + 1) |s + 2|2 5|s + 2|2
(10.10)
 10.7 The Derivative of ζ(s) 255

By (10.6), (10.7)–(10.10) and using β = 12 for |γ| < 103 , we see that when t =
10 and σ ∈ [0, 12 ] we have (ζ  (s)/ζ(s)) < 0. This was the method of Levinson
and Montgomery. Alternatively it is much easier to check this condition
numerically, where we find that in the given range (ζ  (s)/ζ(s)) < −0.2.
Consider now the line σ = 0. By (10.8), using 0 < β ≤ 12 all terms in I1 are
positive, so I < 0. Then using (10.7) and (10.10), when t ≥ 10, evaluating the
bounds for each term in (10.7), we have (ζ  (s)/ζ(s)) < 0 also.
On the line σ = 12 , unless we are at a zero of ζ(s), we have I = 0. If ρ0 =
1
2
+ iγ0 is a zero on this line then the first sum for I1 is positive and, in the
second sum, the term 1/|s−ρ0 |2 can be made arbitrarily large for s sufficiently
close to ρ0 . Thus on a sufficiently small semicircle with centre ρ0 and with
σ < 12 we get I1 > 0, and thus I < 0. Hence on these semicircles we also have
(ζ(s)/ζ(s)) < 0. Replacing the segment σ = 12 by a suitably indented contour
then gives (ζ  (s)/ζ(s)) < 0 for t ≥ 10 on that contour.
Now assume there exists a real sequence (T j ) j≥0 with T j → ∞ such that
(ζ  (s)/ζ(s)) < 0 on the segment (0, 12 )×{T j }. On the indented contour around
[10i, 1
2
+ 10i] × [iT j , 1
2
+ iT j ]
we must have (ζ  (s)/ζ(s)) < 0, so the change in argument of ζ  (s)/ζ(s) is
zero on the contour. By the principle of the argument, this implies the number
of zeros of ζ  (s) and ζ(s) are the same inside each of these contours, and the
proof is complete in this case.
Now suppose no such sequence (T j ) exists. This requires that there exists
a t0 > 0 such that, for all t ≥ t0 , (ζ  (s)/ζ(s)) ≥ 0 for s = σ + it and some
σ ∈ (0, 12 ). Because in (10.7) for t sufficiently large the first term on the right
can be made arbitrarily small and the third term arbitrarily large and negative,
this implies I > 0, so at least one term in the sum expression for I1 must be
negative. Therefore there exists a β < 12 such that
( 12 − β)2 > (t − γ)2 + (σ − 12 )2
so |t − γ| < 12 . If t = n is a sufficiently large integer, there must be at least one
zero ρ = β+iγ of zeta with β < 12 and |γ −n| < 12 . This shows N − (T ) ≥ T +O(1).
Now one can use Lemma 10.14 to show that the change in argument of
ζ(σ + it) and of ζ  (σ + it) for fixed t ≥ t1 and σ ∈ [0, 12 ] are both O(log t).
Alternatively, this is Step (4) of Theorem I.1 of Volume Two [32]. These
derivations Also use the estimates |ζ(s)| t A and |ζ  (s)| t A for −3 ≤ σ ≤ 1
and some A > 0 which is property (9) in Section 2.2.
Using this and the negativity (ζ  (s)/ζ(s)) on the other three components
of the indented rectangular contour around [0, 12 ] × [10, t] completes the
proof. 
Theorem 10.16 [107, theorem 1; 162] If the Riemann hypothesis is true, all
non-trivial zeros of the derivative of the Riemann zeta function ζ  (s) in the
256 Other Equivalents to the Riemann Hypothesis

critical strip have real part greater than or equal to 12 . Even if the Riemann
hypothesis is false, the number of zeros of ζ  (s) in this strip with imaginary
part less than T > 0 and real part less than 1/2 is O(log T ).
Proof If the Riemann hypothesis holds, then for all T > 0, N − (T ) = 0 so
N − (T ) > T + O(1) is false for all T ≥ T 0 . Hence, by Lemma 10.15, N1− (T ) =
N − (T ) = 0 for all T > 0, so ζ  (s) has no zero with 0 < σ < 12 .
Conversely if N1− (T ) = 0 for all T > 0 then again, by Lemma 10.15, N − (T ) =
O(log T ). (Note that this is not sufficient to give the Riemann hypothesis.) 
Recently Suriajaya derived an asymptotic formula for the number of zeros
of the kth derivative of ζ(s), assuming RH [164].

10.8 A Zeta-Related Inequality

The next equivalence is included because of the simplicity of its statement.
The proof was published by Dixon and Spira in 1965. It comes almost
directly from the functional equation for ζ(s), which makes it special when
compared with the other equivalences to RH described here. Also, in this
section, we include the derivation of explicit bounds for the function |Λ(s)|,
which is defined in Lemma 10.17. These could be useful, especially as a basis
for further work relating to the Dixon–Spira criterion.
Lemma 10.17 [47] If |t| ≥ 2π+0.1 and 1 > σ > 12 then |ζ(1− s)| > |ζ(s)| except
when ζ(s) = 0.
Proof Write the functional equation for ζ(s) as ζ(1 − s) = Λ(s)ζ(s) where as
usual
 πs 
Λ(s) := 2(2π)−s Γ(s) cos .
2
Because |Λ(s)Λ(1 − s)| = 1 we have |Λ(s)| > 0 for all s ∈ C and |Λ( 12 + it)| = 1.
Setting s0 = 12 + it define
 
|Λ(s)|
h(s) := log = log |Λ(s)|.
|Λ(s0 )|
Then we have h(s) > 0 if and only if |Λ(s)| > 1, so to prove the lemma, because
|ζ(1 − s)/ζ(s)| = |Λ(s)|, we need only show h(s) > 0 when 1 > σ > 12 .
Let 1 > σ > 12 and |t| ≥ 2π + 0.1. Then
   
1 cosh πt + cos πσ 1
h(s) = log |Γ(s)| − log |Γ(s0 )| + log − σ − log 2π
2 cosh πt 2
   ⎛ 1 ⎞
1 ∂ 1 ⎜⎜ | sin π(σ − 2 )| ⎟⎟⎟
≥ σ− log |Γ(σ + it)| + log ⎜⎝⎜1 − ⎠⎟
2 ∂σ σ=η 2 cosh πt
 
1
− σ − log 2π,
2
 10.8 A Zeta-Related Inequality 257

where we have used the mean value theorem to obtain η with 12 < η < σ.
If 0 ≤ x ≤ 12 we have 12 log(1 − x) ≥ −x. Therefore for |t| ≥ 12 we can derive
the lower bound
⎛ ⎞  
1 ⎜⎜⎜⎜ | sin π(σ − 12 )| ⎟⎟⎟ | sin π(σ − 12 )| 1 −π|t|
log ⎝1 − ⎟⎠ ≥ − > −2π σ − e .
2 cosh πt cosh πt 2
Therefore, dividing by σ − 12 we get
 
h(s) ∂
> log |Γ(σ + it)| − 2πe−π|t| − log 2π.
σ − 12 ∂σ σ=η

Now if we recall B3 (x) := x(2x2 − 3x + 1)/2, the Bernoulli polynomial of

order 3, we can write Stirling’s formula (see Volume Two [32, Appendix H])
in the form
  
1 1 1 1 ∞ B3 ({x})
log Γ(s) = s − log s − s + log 2π + − dx,
2 2 12s 3 0 (s + x)3
where {x} is the fractional part. This formula is true when s is real and
positive. Define the complex logarithm such that log 1 = 0, with the plane
cut along the negative x-axis. With this choice the formula remains true for
complex s in the open subset of C which is √ the complement of the cut.
Note that if x ∈ [0, 1] we have |B3 (x)| ≤ 3/36. Therefore
∂ d
log |Γ(σ + it)| =  log Γ(s)
∂σ ds
  ∞ 
1 1 B3 ({x})
=  log s − − + dx
2s 12s2 0 (s + x)
4

1 σ σ2 − t 2
> log(σ2 + t2 ) − −
2 2(σ2 + t2 ) 12(σ2 + t2 )2
√  ∞
3 1
− dx.
36 0 ((σ + x)2 + t2 )2
To bound the integral let y := σ + x and then tan θ := y/|t| to get
 ∞  ∞
1 1
dx ≤ dy
0 ((σ + x) + t ) 0 (y + t )
2 2 2 2 2 2
 π/2
1 π
= 3 cos2 (θ) dθ = .
|t| 0 4|t|3
Using this bound and 12 < σ < 1 we get
√
h(s) 1 3π
> log |t| − 2 − − 2πe−π|t| − log 2π.
σ − 12 2t 72|t|3
When |t| ≥ 2π + 0.1 the right-hand side is strictly positive, and therefore so is
h(s). This completes the proof of the lemma. 
258 Other Equivalents to the Riemann Hypothesis

Note that, for s = 12 + it, h(s) = 1, and for σ > 12 , ∂h(σ + it)/∂σ can easily
be shown, using the method of Lemma 10.17, to be quite positive.
Theorem 10.18 (Dixon–Spira criterion) [47, 162] The Riemann hypothesis
is equivalent to the statement that, if 12 < σ < 1 and t ≥ 6.29073, then
|ζ(1 − s| > |ζ(s)|.
Proof If the Riemann hypothesis is false and s is a zero not on the critical
line, then both sides of the inequality are zero, so it is false. If the Riemann
hypothesis is true, then Lemma 10.17 shows the inequality is true. 
Corollary 10.19 For |t| > 10 and 1 ≥ σ > 1
2
we have tσ−1/2 > |Λ(σ + it)| >
t(σ−1/2)/5 .
Proof First we derive an upper bound for h(s)/(σ − 12 ), where h(s) is defined
in Lemma 10.17. We have
   
cos(πσ) |sin(π(σ − 12 ))| 1 −π|t|
log 1 + ≤ < π σ− e
cosh(πt) cosh(πt) 2
and
∂ 1 σ σ2 − t 2
log |Γ(σ + it)| ≤ log(σ2 + t2 ) − −
∂σ 2 2(σ2 + t2 ) 12(σ2 + t2 )2
√  ∞
3 1
+ dx
36 0 ((σ + x)2 + t2 )2
√
1 1 3π
≤ log(σ + t ) +
2 2
2
+ .
2 12t 72|t|3
Therefore since σ ≤ 1
√
h(s) 1 1 3π
< log(1 + t ) +
2
+ − log(2π)
σ− 2 2
1
12t 2 72|t|3
√
1 3π √
≤ log |t| + + − log( 2π) < log |t|
12t2 72|t|3
for |t| > 1, giving the upper bound.
Next we derive a lower bound. Using the notation and argument of
Lemma 10.17 we first evaluate the integral:
 ∞
dx tan−1 (t/σ) σ
I(σ, t) := = − 2 2 2 .
0 ((σ + x)2 + t2 )2 2|t| 3 2t (σ +t )
Therefore, if we define as before
σ σ2 − t 2 1
F(σ, t) := − − + log(t2 + σ2 ),
2(t + σ ) 12(t + σ )
2 2 2 2 2 2
 10.9 The Real Part of the Logarithmic Derivative of ξ(s) 259

we can write
√
h(s) 3
> F(σ, t) − I(σ, t) − 2πe−π|t| − log 2π.
σ− 21
36
Since the integral is decreasing in σ, and F(σ, t) is increasing, we can
substitute σ = 12 and obtain a lower bound
√
h(s) 3 1
> F( 2 , t) −
1
I( , t) − 2πe−π|t| − log 2π.
σ − 12 6 2
Then for all σ > 12 we get h(s)/(σ − 12 ) > G(t) where
1 4t2 − 1 1
G(t) := log(4t2 + 1) + + √
2 3(4t + 1)
2 2
12 3t2 (4t2 + 1)
−1
1 tan (2t)
− 2 − √ − 2πe−πt − log(4π).
4t + 1 24 3 |t|3
Finally this expression is greater than 0.201675 log |t| > log |t|/5 for all |t| ≥ 10.
This completes the proof. 

10.9 The Real Part of the Logarithmic Derivative of ξ(s)

Lagarias [98] pointed out that RH was equivalent to a positivity criterion
expressed in terms of the real part of the logarithmic derivative of the
Riemann xi function ξ(s), namely
1 ξ (s)
s > =⇒  > 0.
2 ξ(s)
This is Lagarias’ positivity criterion.
Lagarias showed that the logarithmic derivative of ξ(s) was positive for
σ ≥ 9 and suggested the region could be extended to σ ≥ 1. This was shown
by the present author [24] for all |t| sufficiently large or small. This restriction
did not apply to the theorem of Garunkstis [66] who showed that
ξ (s)
s > 1 =⇒  > 0,
ξ(s)
and it is this work, following an account of the work of Lagarias, which we
have included here.
Lagarias’ work was set in the context of number fields and particular sets of
zeros of meromorphic functions. Let Ω ⊂ C be a discrete subset of points with
multiplicity. We say Ω is an admissible zero set if Ω = Ω̄, if corresponding
points ρ and ρ̄ have the same multiplicity denoted m(ρ), and if
 1 + |ρ|
< ∞. (10.11)
ρ∈Ω
1 + |ρ|2
260 Other Equivalents to the Riemann Hypothesis

This concept, considered along with Theorem 10.23, sheds new light on the
Riemann hypothesis and its generalizations.
Lemma 10.20 The zeros of the Riemann function ξ(s) form an admissible
zero set.

Proof By Lemma 2.8, Step (2), ρ 1/|ρ|2 converges. 
In what follows we need the Weierstrass product, which is developed and
proved in Volume Two [32, Theorem B.13]. See also Alfors [4, chapter 5].
Lemma 10.21 (Weierstrass product) The Weierstrass product
  s
m(ρ)  
s

s
m(ρ)
fΩ (s) := sm(0)
1− 1− 1−
ρ∈Ω∩R
ρ ρ∈Ω, ρ>0
ρ ρ̄

converges uniformly on compact subsets of C to an entire function which is

real on the real axis, where Ω is admissible.
Finally, in this summary of preliminary results, we need an expression for
the logarithmic derivative of fΩ (s). Note that the proof uses Cauchy’s well-
known bound for the first derivative of a function f (s) which is holomorphic
in an open neighbourhood of a closed disc B(s0 , r] with | f (s)| ≤ M on this
disc, namely
M
| f  (s0 )| ≤
.
r
This follows directly from Cauchy’s integral formula for the first derivative.
Lemma 10.22 The Mittag–Leffler expansion
 m(ρ)   
fΩ (s) 1 1
= + m(ρ) +
fΩ (s) ρ∈Ω∩R s − ρ ρ∈Ω, ρ>0 s − ρ s − ρ̄

converges uniformly on compact subsets K ⊂ V := C \ Ω.

Proof (1) First we show there exists r > 0 such that |s − w| > r for all s
in K and w in C \ V. If not there is a convergent sequence (sn ) in K and
corresponding points wn in C \ V = Ω with |sn − wn | < 1/n. But this means (sn )
and (wn ) have the same limit point, s0 say, in K. Since C \ V is closed, this
gives a contradiction since K ∩ Ω = ∅, showing ρ must exist.
(2) Next we show that for any r > 0 if K is compact so is the union of
closed discs
'
L := B(s, r].
s∈K

To see this let sn = xn + yn be a sequence in L with yn ∈ B(xn , r]. Then since

K is compact there is a convergent subsequence xnk → x0 in K. For each
 10.9 The Real Part of the Logarithmic Derivative of ξ(s) 261

j ∈ N we have ynk j ∈ B(x0 , r + 1/ j] such that this subsequence converges, to

y0 say, which necessarily is in B(x0 , r] ⊂ L. Hence (sn ) has the convergent
subsequence
snk j = ynk j + ynk j ,
so L is compact also.
(3) Now for n ∈ N let
  s
m(ρ)  
s

s
m(ρ)
m(0)
fΩ,n (s) := s 1− 1− 1− .
ρ∈Ω∩R
ρ ρ∈Ω, ρ>0
ρ ρ̄
|ρ|<n |ρ|<n


We claim fΩ,n → fΩ uniformly on K. Let  > 0 be given. By Steps (1) and
(2) there is a r > 0 such that the corresponding L is a compact subset of V.
Therefore fΩ,n → fΩ uniformly on L and thus there is an N such that for all
s ∈ L and n ≥ N we have
| fΩ,n (s) − fΩ (s)| < r.
In particular this inequality holds on each B(s, r] for every s ∈ K, so by
Cauchy’s bound applied to fΩ,n (s) − fΩ (s) we have for all s ∈ K and n ≥ N

| fΩ,n (s) − fΩ (s)| < .
(4) Since the convergence is uniform on compact subsets K for both fΩ,n

and fΩ,n , and fΩ,n (s)  0 on K, convergence is uniform for the logarithmic
derivative also. This completes the proof. 
Theorem 10.23 [98, theorem 1.1] If Ω ⊂ C is an admissible zero set with
corresponding entire function fΩ (s) then the following two conditions are
equivalent for a given real θ:
(a) All elements ρ ∈ Ω have ρ ≤ θ.
(b) For s > θ we have
  
f (s)
 Ω > 0.
fΩ (s)
Proof If s > θ and (b) is true, then fΩ (s)/ fΩ (s) is well defined at s ∈ C,
hence does not have a pole, so s  Ω. Thus if ρ ∈ Ω we have ρ ≤ θ, which is
condition (a).
Since Ω is admissible, by Lemma 10.22 there is a Mittag–Leffler expansion
for fΩ (s)/ fΩ (s). If (a) is true and s > θ then s > ρ for all ρ ∈ Ω. We
claim that if c > 0 is real and ρ = σ + iγ, z = x + iy, then if x > σ we have
(c/(z − ρ)) > 0. This follows directly from the identity
 
c c(x − σ)
 = .
z−ρ (x − σ)2 + (y − γ)2
262 Other Equivalents to the Riemann Hypothesis

It follows that every term in the expansion of ( fΩ (s)/ fΩ (s)) is positive and
therefore so is the sum. Therefore condition (b) is true also. 
In Garunkstis [66] from 2002, there is a better result than that given in [24],
and we give the details of this below. Let 12 ≤ a ≤ 1 and suppose that ζ(s) has
no zero when σ > a. Then Garunkstis [66] showed that for all such σ we have
for all t ∈ R
ξ (σ + it) ξ (σ)
 ≥ .
ξ(σ + it) ξ(σ)
There are a number of lemmas leading to this result; some are due to
Lagarias [98].
Note that (see Volume Two [32, Theorem 4.6] or Edwards [51, chapter 2])
we can write the Weierstrass product form of ξ(s) valid for all s ∈ C
 s

ξ(s) = ξ(0) 1− , (10.12)
ρ
ρ
where the product is over all of the complex zeros ρ of zeta taken by
combining complex conjugate pairs.
Lemma 10.24 [98; 99, lemma 3.1] For non-zero t, σ and γ we have
1 1 2
+ 2 ≥ 2 ⇐⇒ 3γ2 ≥ σ2 + t2 . (10.13)
σ + (t + γ)
2 2 σ + (t − γ)2 σ + γ2
Proof No denominator is zero, so clearing denominators of the first
inequality, dividing by 2 and simplifying gives the second inequality. These
steps can be reversed so the second implies the first. 
Note that if t = 0 the left-hand inequality is an equality.
Lemma 10.25 [66, lemma 2] Assume all zeros ρ = β +iγ of ζ(s) satisfy β < a.
For a < σ ≤ 10 and 0 < |t| ≤ 21 we have
  
ξ (σ + it) ξ (σ)
 > .
ξ(σ + it) ξ(σ)
Proof First note that 1 − a < β < a and |γ| ≥ 14.134 . . . . If we set σ0 := σ − β
and let t satisfy |t| ≤ 21 we get 3γ2 > 3(14)2 > 102 + 212 ≥ σ20 + t2 so the second
inequality of Lemma 10.24 holds strictly, hence so does the first. Since σ > a
we can write using (10.12)
    
ξ (σ + it) σ−β σ−β
 = +
ξ(σ + it) γ>0
(σ − β)2 + (t − γ)2 (σ − β)2 + (t + γ)2
 2(σ − β) ξ (σ)
> =
γ>0
(σ − β)2 + γ2 ξ(σ)
and the proof of the lemma is complete. 
 10.9 The Real Part of the Logarithmic Derivative of ξ(s) 263
√
Lemma 10.26 [66, lemma 3] Let 1
2
≤ β ≤ a ≤ 1, σ > a and γ ≥ (σ + 12 )/ 3.
Then we have two inequalities:
σ−a σ+a−1 σ−β σ+β−1
+ ≤ + ,
(σ − a) + γ
2 2 (σ + a − 1) + γ
2 2 (σ − β) + γ
2 2 (σ + β − 1)2 + γ2
σ−β σ+β−1 2(σ − 12 )
+ ≤ .
(σ − β)2 + γ2 (σ + β − 1)2 + γ2 (σ − 12 )2 + γ2
Proof Define
σ− x
f (x) := and g(x) := f (x) + f (1 − x)
(σ − x)2 + γ2
with x ∈ [a − 1, a]. Then
2(σ − x)3 − 6(σ − x)γ2
f  (x) = .
((σ − x)2 + γ2 )3
√
Because γ ≥ (σ + 12 )/ 3 we have f  (x) ≤ 0 so f (x) is concave down, giving
2 f ((x+y)/2) ≥ f (x)+ f (y). Setting x = β, y = 1−β gives the second inequality.
Since 12 < β ≤ a, g ( 12 ) = 0 and g (x) < 0 for 12 < x ≤ a the first inequality follows
also. 
Lemma 10.27 [66, lemma 4] For all a and σ with a < σ ≤ 9/2 we have
ξ (σ)
0< ≤ 0.047(σ − 12 ).
ξ(σ)
Proof Taking the logarithmic derivative of the definition of ξ(s) we get for
ζ(s)  0
ξ (s) 1 1 1 1 Γ (s/2) ζ  (s)
= + − log π + + .
ξ(s) s s − 1 2 2 Γ(s/2) ζ(s)
From the Laurent expansion of ζ(s) about s = 1 (Volume Two [32, Theorem
I.2] or [82]) we get, with C being Euler’s constant,
ζ  (s) 1
=− + C + O(|s − 1|).
ζ(s) s−1
We also know [Γ ( 12 )/Γ( 12 )]/2 = −C/2 − log 2 (using for example the
representation for Γ (z)/Γ(z) given in Theorem I.1 in Volume Two [32]
– and see Lemma 2.5 in this volume) so we can compute ξ (1)/ξ(1) =
0.0230957 . . . .
Now let A := {ρ : ζ(ρ) = 0, β = 12 , γ > 0} and B := {ρ : ζ(ρ) = 0, β > 12 ,
γ > 0}. Then by Lemma 10.26 we have for σ ≥ a
 
ξ (σ)  1 1
= +
ξ(σ) γ>0 σ − ρ σ − ρ̄
264 Other Equivalents to the Riemann Hypothesis
 2(σ − 12 )  σ−β σ−1+β

= +2 +
ρ∈A
(σ − 12 )2 + γ2 ρ∈B
(σ − β) + γ
2 2 (σ − 1 + β)2 + γ2
 2(σ − 1
)  4(σ − 12 )
≤ 2
+
ρ∈A
(σ − 12 )2 + γ2 ρ∈B
(σ − 12 )2 + γ2
 2(σ − 12 )
=
(σ − 12 )2 + γ2
γ>0
 1
≤2 (σ − 12 ).
γ>0
γ 2

Now we find an upper bound for the coefficient of σ − 12 . First note by

Lemma 10.26 and (10.12) that, if γ0 = 14.1 . . . is the imaginary part of the
lowest zero of ζ(s) in the critical line,
ξ (1)  2   4(1 − β) 4β

0.04619 > 2 = + +
ξ(1) ρ∈A (1/2)2 + γ2 ρ∈B (1 − β)2 + γ2 β2 + γ2
 2  4
≥ +
ρ∈A
1/4 + γ 2
ρ∈B
1/4 + γ2
 2 γ02  1
≥ > 2 ,
γ>0
1 + γ2 1 + γ02 γ>0 γ2

so γ>0 2/γ
2
< 0.0465 . . . and the proof of the lemma is complete. 
Lemma 10.28 [66, lemma 5] Suppose there are no zeta zeros in the half
plane σ > a. Let 12 ≤ a < σ ≤ 92 and t ∈ R be given. Suppose further that at
least one of the following conditions is satisfied:

(i) there are two non-trivial zeros or a double zero of ζ(s) with |t − γ| ≤ 5, or
(ii) there is a complex zero ρ = β + iγ of ζ(s) with |t − γ| ≤ 2. Then
  
ξ (σ + it)) ξ (σ)
 ≥ .
ξ(σ + it) ξ(σ)
Proof Case (i): In case both zeros, not necessarily distinct, lie on the critical
line, ρ1 = 12 + iγ1 and ρ2 = 12 + iγ2 say, then by Lemma 10.27
  
ξ (σ + it) σ − 12 σ − 12
 ≥ +
ξ(σ + it) (σ − 12 )2 + (t − γ1 )2 (σ − 12 )2 + (t − γ2 )2
2(σ − 12 ) ξ (σ)
≥ > .
4 2 + 52 ξ(σ)
In case one of the zeros is off the critical line, consider the two zeros ρ1 =
β + iγ and ρ = 1 − β + iγ with 12 < β and |t − γ| ≤ 5. Then since β < a using
 10.9 The Real Part of the Logarithmic Derivative of ξ(s) 265

Lemma 10.26
  
ξ (σ + it) σ−β σ−1+β
 ≥ +
ξ(σ + it) (σ − β) + (t − γ)
2 2 (σ − 1 + β)2 + (t − γ)2
σ−β σ−1+β
≥ +
(σ − β) + 25 (σ − 1 + β)2 + 25
2

σ−a σ−1+a
≥ +
(σ − a) + 25 (σ − 1 + a)2 + 25
2

= h(σ, a)(σ − 12 ),
where we define
 ⎛ ⎞
σ−a σ−1+a ⎜⎜⎜⎜ 1 ⎟⎟⎟⎟
h(σ, a) := + ⎝ ⎠.
(σ − a)2 + 25 (σ − 1 + a)2 + 25 σ − 12
Using Mathematica to evaluate the derivatives of h(σ, a) and simplify the
resulting expressions we find
∂h(σ, a) (2σ − 1)(a2 − 11a − 20 + σ − σ2 )(a2 + 9a − 30 + σ − σ2 )
=−
∂σ ((σ − a)2 + 25)2 ((σ − 1 + a)2 + 25)2
and
∂h(9/2, a) 32(2a − 1)(16a4 − 32a3 − 1288a2 + 1340a − 29 031)
= .
∂a (4a2 − 36a + 181)2 (4a2 + 28a + 149)2
Now we have assumed a ≤ σ ≤ 9/2 and 12 < a ≤ 1 giving ∂h(9/2, a)/∂σ ≤ 0
and thus h(σ, a) ≥ h(9/2, a). In the range a ∈ [ 12 , 1], ∂h(9/2, a)/∂a ≤ 0 giving
h(9/2, a) ≥ h(9/2, 1) = 0.0483518 . . . .
Feeding this inequality back into the lower bound for the logarithmic
derivative of ξ(s) and then using Lemma 10.27 completes the proof in this
case.
Case (ii): If the zero does not lie on the critical line then we have case (i)
which is already covered. Hence we can write ρ = 12 +iγ and assume |t −γ| ≤ 2.
Then by (10.12) we have
  
ξ (σ + it) σ − 12
 ≥
ξ(σ + it) (σ − 12 )2 + (t − γ)2
   
1 1 1
≥ 2 2 σ − = 0.05 σ −
4 +2 2 2

ξ (σ)
≥
ξ(σ)
by Lemma 10.27 and the proof is complete. 
266 Other Equivalents to the Riemann Hypothesis
Table 10.1 The first 20 Riemann zeros and gaps.

Rank n Zero ρ = 12 + iγn Gap γn+1 − γn

1 14.134725141734693790 6.8873144970368612022
2 21.022039638771554993 3.9888179413741337706
3 25.010857580145688763 5.4140185457138244471
4 30.424876125859513210 2.5101854618796764804
5 32.935061587739189691 4.6511165710864815666
6 37.586178158825671257 3.3325408533218239302
7 40.918719012147495187 2.4083542687675043321
8 43.327073280914999519 4.6780776002521602084
9 48.005150881167159728 1.7686815965051424540
10 49.773832477672302182 3.1964890000421584622
11 52.970321477714460644 3.4759262193489341602
12 56.446247697063394804 2.9007963055389582753
13 59.347044002602353080 1.4847345220074567646
14 60.831778524609809844 4.2807655234717968166
15 65.112544048081606661 1.9672664814125670536
16 67.079810529494173714 2.4665911816798055384
17 69.546401711173979253 2.5207559633079283296
18 72.067157674481907583 3.6375330246020255858
19 75.704690699083933168 1.4401493697908722044
20 77.144840068874805373 2.1925349513745625501

Note that ρ300 = 0.5 + i541.847 . . . . Table 10.1 gives the first 20 imaginary
parts of the imaginary zeros of zeta. The third column is the size of the gap
to the next zero. Figure 10.4 is a plot of the first 300 coordinate gaps.
We see from the data in Table 10.1 that when 21 ≤ t ≤ 79 there is either a
zero with |t − γ| ≤ 2 or two zeros with |t − γ| ≤ 5.
There was some discussion of the error in finding the number of zeros N(T )
of ζ(s) in the critical strip up to a given height in Chapter 2, Section 2.3. It is
a well-studied problem and well known that we can write
 T  7  
T 1
N(T ) = log + + S (T ) + O ,
2π 2πe 8 T
where S (T ) is the argument of ζ(s) on the critical line at the point 12 + iT
which is not the ordinate of a zero. The argument is fixed by taking the
value found by continuous variation from an argument value of 0 along the
horizontal line from (∞, iT ) to ( 12 , iT ). At a zero the value is defined to be
limt→0+ S (t). This function and its properties are discussed fully by Edwards
[51, chapters 8 and 9] for example. In this section we use the explicit estimate
of Trudgian:
 10.9 The Real Part of the Logarithmic Derivative of ξ(s) 267
γn+1–γn
6

n
0 50 100 150 200 250 300

Figure 10.4 The first 300 imaginary coordinate gaps between zeros of ζ(s).

Lemma 10.29 [168, theorem 2.2] If t + h > t > 168π then

 t+h 
 
 S (u) du  ≤ 2.067 + 0.059 log(t + h).
t

Another way of writing the formula for the number of zeros of ζ(s) up to
height T is
ϑ(T )
N(T ) = + S (T ) + 1, (10.14)
π
where (see [51, section 6.5])
    
1 1 + 2it 1 − 2it log π
ϑ(t) := log Γ − log Γ − t
2i 4 4 2
 
t t 1 1 1
= log − + +O 3 .
2 2πe 8 48t t
Note that if γn and γn+1 are the positive imaginary parts of the ordinates of
sorted successive zeta zeros and γn < t < γn+1 , then N(t) = n for all such t so
S (t) decreases on (γn , γn+1 ). At t = γn , S (t) has a jump discontinuity of size
a positive integer. See Figure 10.5, where the discontinuities are represented
by vertical line segments.
Lemma 10.30 For t ≥ e we have
 
 1 2t 21
ϑ (t) = log + R, where |R| ≤ .
2 π 160t2
268 Other Equivalents to the Riemann Hypothesis
S(T )

1.0

0.5

T
410 420 430 440 450

–0.5

–1.0

Figure 10.5 The function S (T ) on the domain [400, 450].

Proof Firstly
 
 1 1 Γ (1 + 2it) Γ (1 − 2it)
ϑ (t) = − log π + + .
2 4 Γ(1 + 2it) Γ(1 − 2it)
Now use the estimate given in Lemma 10.13, namely
Γ (s) 1 1
= log s − + R1 , where |R1 | ≤ , |s| > 2, s > 0,
Γ(s) 2s 10|s|2
to get
 
 1 2t 21
ϑ (t) = log + R2 , where |R2 | ≤ , t > 2,
2 π 160t2
completing the proof. 
Lemma 10.31 For 168π ≤ t < t + h, if ζ(s) has no zero with imaginary part
in [t, t + h] and S (t) has no zero in (t, t + h) then h < 1.5 uniformly in t.
Proof Assume, to get a contradiction, that the given conditions hold with
h = 1.5. If the sign of S (t) is uniformly negative on [t, t + 1.5], because
ϑ(t) is concave up, considering the area of a triangle we can write, using
Lemma 10.30 and (10.14),
 t+h 
  h2
 S (u) du  ≥ ϑ (t).
t
 2π
 10.9 The Real Part of the Logarithmic Derivative of ξ(s) 269

If the sign of S (t) is positive then, since ϑ (t) is increasing,

 t+h 
  1
 S (u) du  ≥ h2 (S (t) − S (t + h))
t
 2
h
= (ϑ(t + h) − ϑ(t))
2π
h2
≥ ϑ (t).
2π
Hence in both cases, by Lemma 10.30 we get
 t+h   
  h2 2t 21h2
 S (u) du  ≥ log − .
t
 2π π 320πt2
Therefore, by Lemma 10.29
 2 
h 21h2 h2 π h
2.067 + 0.059 + + log ≥ − 0.059 log t.
t 320πt2 2π 2 2π
Substituting h = 1.5, this inequality implies 168π < t < 1724. In the range
t ∈ [168π, 1724] the maximum separation of the imaginary parts of zeta zeros
is less than 2.981 (with the maximum being γ380 − γ379 ). 
It follows from Lemma 10.31 that for t ≥ 168π, if ζ(s) has no zero with
imaginary part in [t, t + 1.5] then S (t) cuts (t, t + 1.5) at a point of continuity.
Theorem 10.32 If t ≥ 168π then ζ(s) has a zero with imaginary part
satisfying |t − γ| ≤ 1.5.
Proof Assume this is false for some t so ζ(s) has no zero with imaginary
part in [t − 1.5, t] or in [t, t + 1.5]. By Lemma 10.31, S (t) changes sign
on (t − 1.5, t) and on (t, t + 1.5). These points of sign change must be
separated by at least one point of discontinuity of S (t), which are in one-
to-one correspondence with the imaginary parts γ of the zeros of ζ(s), so
there is at least one such γ with t − 1.5 < γ < t + 1.5. 
Corollary 10.33 [98, lemma 3.5] For each t with |t| ≥ 21 at least one of the
conditions (i) or (ii) of Lemma 10.28 is true.
Remark The study of distances (“gaps”) between zeros of the Riemann
zeta function has received a great deal of attention: for the initial papers
see [119] and [129], but also the very large body of work which is random
matrix theory – see for example [89]. A theorem of Littlewood has been made
explicit, at least in part, by Hall and Hayman [72, theorem 1]: there exists a
T 0 such that for T > T 0 there exists a zeta zero ρ = β + iγ such that
π/4
|γ − T | ≤ .
logloglog T − loglogloglog T − 32
270 Other Equivalents to the Riemann Hypothesis

One could find the explicit value of T 0 and relate this result to that of
Theorem 10.32 for example.

Lemma 10.34 [98, lemma 3.6] If σ > 9/2 and |t| ≥ 21 then
  
ξ (σ + it) ξ (σ)
 ≥ .
ξ(σ + it) ξ(σ)
Proof Let (σ, t) be in the given ranges, so in particular σ > 1. From
the definition of ξ(s), taking the logarithmic derivative and using the
representation of the logarithmic derivative of Γ(z) (see Volume Two [32,
Theorem H.1])
∞  
1 Γ (s/2) γ 1  1 1
=− − + − ,
2 Γ(s/2) 2 s n=1 2n 2n + s

where γ is Euler’s constant, gives

   
ξ (s) 1 1 1 Γ (s/2) ζ  (s) 1
 = + +  + − log π
ξ(s) s s−1 2 Γ(s/2) ζ(s) 2
σ σ−1 γ σ
= 2 2+ − − 2 2
σ +t (σ − 1) + t
2 2 2 σ +t
∞  
1 σ + 2n ζ  (s) 1
+ − +  − log π
n=1
2n (σ + 2n)2 + t2 ζ(s) 2

and
ζ  (s) ∞
Λ(n)
 =− cos(t log n).
ζ(s) n=1
nσ

From this we can conclude the minimum of this latter sum is attained at t = 0,
since each cosine term is bounded above by 1 and each coefficient is positive.
Now let
ξ (s) ζ  (s) 1 1 1 Γ (s/2) 1
g(s) := − = + + − log π.
ξ(s) ζ(s) s s − 1 2 Γ(s/2) 2
Then

(g(σ + it) − g(σ))

   ∞  
σ−1 1 1 σ + 2n
= − + −
(σ − 1)2 + t2 σ − 1 n=1
σ + 2n (σ + 2n)2 + t2
⎛∞ ⎞
2⎜
⎜⎜ 1 ⎟⎟⎟ t2
≥ t ⎜⎜⎝ ⎟⎟⎠ − .
n=1
(σ + 2n)((σ + 2n)2 + t2 ) (σ − 1)((σ − 1)2 + t2 )
 10.10 The Order of Elements of the Symmetric Group 271

Therefore all we need to show is

∞
1 1
> . (10.15)
n=1
(σ + 2n)((σ + 2n)2 + t2 ) (σ − 1)((σ − 1)2 + t2 )

To this end let k := 1 + 2n and σ1 = σ − 1. If we consider the ratio of each

fixed term on the left with the term on the right
σ1 t2 + σ31
R(σ1 , t) := ,
(σ1 + k)t2 + (σ1 + k)3
then for t > 0 we have
∂R 2t((σ1 + k)3 − σ31 (σ1 + k))
= > 0,
∂t ((σ1 + k)t2 + (σ1 + k)3 )2
so for fixed σ1 the minimum of R(σ1 , t) is at t = 21. In addition
∂R kt4 + k3 t2 + 3kσ1 (σ1 + k)2
= > 0,
∂σ1 ((σ1 + k)t2 + (σ1 + k)3 t2 )2
so for fixed t > 0 the minimum occurs at σ1 = 7/2. Therefore we need only
check the result of the theorem at σ = 9/2, t = 21. At this point we find

3
1 1
> .
n=1
(σ + 2n)((σ + 2n)2 + t2 ) (σ − 1)((σ − 1)2 + t2 )

This completes the proof since all the terms in the sum are positive. 
Theorem 10.35 [66, p. 2] Let a satisfy 12 ≤ a ≤ 1. If ζ(s) has no zeros for
σ > a then for s > a we have
%  &
ξ (σ) ξ (σ + it)
= inf  :t∈R .
ξ(σ) ξ(σ + it)
Proof This result follows directly from Lemmas 10.25, 10.27, 10.28
and 10.34, together with Corollary 10.33. 

10.10 The Order of Elements of the Symmetric Group

In this section we give a summary of the equivalence to RH arising out of the
work of Landau and Shah, and developed by Massias, Nicolas and Robin,
namely the symmetric group criterion. We give some of the introductory
proofs, but leave the interested reader to consult the references for details of
the main derivations, especially those references given in [112].
Let n ∈ N. Denote by S n the group of all permutations of n unlike symbols,
so the size |S n | = n!. The group product is composition of functions, where a
permutation which sends i → σ(i) is regarded as a function σ. These groups
272 Other Equivalents to the Riemann Hypothesis

are central to any study of finite groups. This is because of Caley’s theorem,
which states that any group of order n is isomorphic to a subgroup of S n .
Indeed S n has many subgroups; for example S 4 has 30 and S 5 over 100.
These groups appear in many parts of mathematics and its applications,
including Galois theory, the definition of the determinant, symmetries of
objects including molecules, cryptography, etc. For low values of n these
groups have concrete isomorphic realizations: for example S 1 consists only
of the identity, S 2 is the largest Abelian S n , S 3 is the group of rotational and
reflection symmetries of an equilateral triangle, S 4 is the group of rotations
about opposite faces, diagonals and edges of a cube, and S 5 is the Galois
group of the general polynomial of the 5th degree.
In this section we need only a few simple properties of S n . A cycle is
a permutation which can be represented as an m-tuple of distinct elements
(n1 , n2 , . . . , nm ), with m ≤ n, and such that the corresponding permutation is

n1 → n2 → · · · → nm → n1 .

We say the length of such a cycle is m. The order of a permutation σ is

the least k ≥ 1 such that σk = e, the identity. Two cycles are equal if the
corresponding permutations are equal as functions. The main properties of
cycle representations are given by the following:

• Every σ ∈ S n may be written as a cycle or a product of disjoint cycles.

• If two cycles are disjoint, the corresponding permutations commute.
• The representation of a permutation in terms of disjoint cycles is unique up
to commutation.
• The order of a permutation σ ∈ S n is the least common multiple (LCM) of
the lengths of the cycles in the cycle representation of σ.

For an easy-to-read introduction to S n , see Gallian [65, chapter 5]. More

advanced material is in Sagan [149].
Now let g(n) be the maximum order of any permutation in S n . In 1903
Landau [103] proved that as n → ∞ we have
 √
log g(n) ∼ n log n or g(n) ∼ e n log n .

Values of g(n) are given in Table A.6 and in part A000793 of the
Online Encyclopedia of Integer Sequences [158]. Landau’s theorem [103]
is included here as Theorem 10.39. We based our presentation of this result
on the proof of William Miller [118].
In 1939 Shah [156] improved Landau’s asymptotic formula by proving
  
 loglog n 1
log g(n) = n log n 1 + +O .
2 log n log n
 10.10 The Order of Elements of the Symmetric Group 273

In 1980 Szalay [165] made further improvements to Landau’s theorem,

arriving at the very precise expression
 
loglog n − 1 loglog n − 2 + o(1)
log g(n) = n log n 1 + + .
log n log2 n
In this section we summarize the properties of g(n) and prove some pre-
liminary lemmas, including a simpler proof, as Theorem 10.40, of a result of
Shah which is essential to the development of the equivalence. Then we state
two theorems. The second spells out the equivalence to RH, but it is really a
corollary of the first more detailed result. This is based on the work of Mas-
sias, Nicolas and Robin [112]. Here is the statement of the second theorem:
Theorem 10.42 The Riemann hypothesis is equivalent to the statement that
for all n sufficiently large we have

log g(n) < li−1 (n)
where li(x) for x > 0 is the Cauchy principal value of the logarithmic integral
 x
dt
li(x) := ,
0 log t

and where li−1 (x) is the functional inverse.

Finally in this section, we summarize the elements of a statistical study of
S n made by Erdős and Turan [55]. This is in order that the reader might see
that elements having the maximum order in S n are exceptional, and that the
order of an “average” element is significantly less.
To begin, we define the arithmetic function
 
h(n) := pai i , where n = pai i
1≤i≤k 1≤i≤k

is the standard prime factorization. Also set h(1) = 0 so h(n) is an additive

function. A plot of h(n) for 1 ≤ n ≤ 103 is given in Figure 10.6.
Lemma 10.36 [118] For all n ∈ N we have
g(n) = max m.
h(m)≤n

Proof If n = 1 the result is immediate since h(1) = 0 and h(2) = 2, so we can

assume n > 1.
(1) Let a1 , a2 , . . . , ak be a set 
of positive integers and m = [a1 , . . . , ak ] their
LCM. Then we claim h(m) ≤ 1≤i≤k ai . To see this, suppose first m = 1 so
m = [1, 1, . . . , 1] and the inequality is immediate, and we can assume m > 1.
Suppose we have a set of positive integers with m = [a1 , . . . , ak ] but h(m) >
274 Other Equivalents to the Riemann Hypothesis
h[n]
500

400

300

200

100

n
200 400 600 800 1000

Figure 10.6 Values of the function h(n) excluding prime powers.


1≤i≤k ai . Let {ai } be a set with the smallest such sum. Since removing a term
with ai = 1 reduces the sum, but preserves the value of the LCM, we can
assume each ai ≥ 2. If we could write one of the ai as ai = xy with (x, y) = 1 and
both x, y ≥ 2 then since x+y < xy = ai we could replace ai with the two integers
x and y, and again reduce the sum. Hence we can assume each of the ai is
a power of a prime. Finally if two of the ai were powers of the same prime
we could delete the one with the smaller or equalpower, and get a smaller
sum, but the same LCM and so satisfying h(m)  > 1≤i≤k ai . But in case the ai
are powers of distinct primes we have h(m) = ki ai , a contradiction, which
proves the claim.
(2) We will now show that there exists a permutation on n elements of order
m if and only if h(m) ≤ n. To see this, first suppose h(m) ≤ n. Let m = 1≤i≤k pαi i
and define a permutation σ ∈ S n with cycle lengths (pαi i ) and n − h(m) fixed
points. Then the order of this permutation is m. If such a permutation of order
m exists let (ai ) be its cycle lengths so their sum is n and LCM is m. Hence,
by Step (1) we have h(m) ≤ n.
(3) Now we claim that among all of the permutations on n symbols with
maximum order there exists at least one with non-trivial cycle orders being
powers of distinct primes: let m be the maximum order of a permutation
so we can write g(n) = m. Then by Step (2) we have h(g(n)) ≤ n. Use the
construction of Step (2), with m having this value g(n), to construct the
required permutation.
 10.10 The Order of Elements of the Symmetric Group 275

(4) We can now complete the proof. Let θ := maxh(m)≤n m. Now by Step (3)
using a permutation of maximum order, h(g(n)) ≤ n so g(n) ≤ θ. If on the
other hand m is such that h(m) ≤ n, by Step (2) there is a permutation of order
m so m ≤ g(n). Hence θ ≤ g(n) so θ = g(n) and the proof of the lemma is
complete. 

This section includes a version of Landau’s theorem log g(n) ∼ n log n
[105, pp. 222–229] based on the proof in Miller’s paper [118], which on the
face of it looks very similar to Landau’s. In Table A.6 the first 160 values
of g(n) are given. Extensive evaluations were carried out by Morain up to
n = 32 000 with some details
 in [120].
For x > 0 let A(x) := p≤x p be the sum of all primes up to x.
Lemma 10.37 As x → ∞ we have
 2 
x2 x
A(x) = +O .
2 log x log2 x
Proof This is a simple application√ of Abel’s identity [6, theorem 4.2],
splitting the integral for the error at x. 
The reader may wish to compare the use of Lemma 10.37 with the method
of Shah [156] who takes A(x) to the fourth order. This is not needed in the
proof, but deriving it could be good practice for the energetic reader.
Lemma 10.38 As x → ∞ we have
 2 
x2 1 x2 1 x2 x
A(x) = + 2
+ 3
+O .
2 log x 4 log x 4 log x log4 x
Proof First use induction to show that
 x
p=− π(n) + π(x)(x + 1).
p≤x n=1

In addition, integrating the definition of Li(x) successively by parts leads to

the asymptotic expansion, for x → ∞,
x  n!
∞
Li(x) = .
log x n=0 logn x

Using these expressions and the prime number theorem in the form

π(n) = Li(n) + O(x exp(−c log x ))
enables us to derive
x  x   
n n 2n n
π(n) = + + +O
n=1 n=2
log n log2 n log3 n log4 n
276 Other Equivalents to the Riemann Hypothesis
 x    x 
1 1 2 u du
= u + + du + O
2 log u log2 u log3 u 4
2 log u
 
x2 3 x2 7 x2 x2
= + + + O .
2 log x 4 log2 x 4 log3 x log4 x
Therefore
  2 
x2 1 x2 1 x2 x
A(x) = p= + + + O .
p≤x
2 log x 4 log2 x 4 log3 x log4 x
This completes the proof. 

Theorem 10.39 [105, 118] As n → ∞ we have log g(n) ∼ n log n.
Proof (1) Let n be a positive integer and let P = P(n) be the largest prime
such thatthe sum of the primes less than P does not exceed n and define
F(n) := p<P p. Then because log F(n) = θ(P − 1) and P ∼ P − 1 we have, by
 ∼ P.
the prime number theorem, log F(n)
(2) Now for x > 0 let A(x) := p≤x p. By the definitions of A(x) and P we
have A(P − 1) ≤ n < A(P). But for all x > 0 we have A(x − 1) ∼ A(x) and
therefore, by Lemma 10.37, we have P2 /(2 log P) ∼ n, which implies that
P2 /(2n log P) → 1. 
If P is not asymptotic to n log n then there is an  > 0 such that for an
infinite number of values of n one of the inequalities
 
P ≤ (1 − ) n log n or P ≥ (1 + ) n log n
√
holds. Because x → x2 /log x is increasing for x > e, if the first inequality is
true then
P2 (1 − )2 log n
≤ .
2n log P log n + loglog n + 2 log(1 − )
However, as n → ∞ the right-hand side approaches (1 − )2 but the left-hand
side tends to 1. Similarly  the second inequality does not hold for infinitely

many n. Hence P ∼ n log n and so, by Step (1) we have log F(n) ∼ n log n.
(3) Because, by Step (2) of Lemma 10.36, there is a permutation on n
letters of order F(n) we have F(n) ≤ g(n).
(4) Let q1 , . . . , q s be all of the distinct primes dividing g(n) in increasing
order. Then we claim that
 s
log q j < 2 + log F(n) + log P.
j=1

Note that we can assume n ≥ 2 so P ≥ 3. Let p1 , . . . , pr be the odd primes

with p j ≤ P and such that p j  g(n) for 1 ≤ j ≤ r. Hence the set of primes
not exceeding P, other than possibly 2, is the list p1 , . . . , pr , q1 , . . . , qt−1 where
 10.10 The Order of Elements of the Symmetric Group 277

these q are the qi ≤ P which divide g(n). Now because h(m) ≥ q1 + · · · + q s , by

Lemma 10.36 we get

s 
q j ≤ h(g(n)) ≤ n < p.
j=1 p≤P

Cancelling like terms we get


s 
r
qj ≤ 2 + p j.
j=t j=1

But for t ≤ j ≤ s we have 3 ≤ P ≤ q j and for 1 ≤ i ≤ r we have 3 ≤ pi ≤ P.

Therefore since for x ≥ 3 the function x → x/log x is increasing, we can write
P P
log q j ≤ q j and pi ≤ log pi .
log P log P
Therefore

s
2 log P 
r  r
log q j ≤ + log p j ≤ 2 + p j.
j=t
P j=1 j=1

Finally add the sum 1≤ j≤t−1 log q j to each side, and use the fact that
p1 , . . . , pr , q1 , . . . , qt−1 is a list of all of the primes not greater than P to
complete the proof of the claim.
(5) If q is a prime and e > 1 an integer being the maximum power for
which qe | g(n), we claim qe ≤ 2P. To see this let Q be the smallest prime not
dividing g(n) so each of the primes less than Q divides g(n). Thus their sum
is at most h(g(n)) ≤ n. Also the sum of the primes p ≤ P is greater than n.
Hence Q ≤ P. To complete the proof we will show that qe ≤ 2Q. Assume, to
get a contradiction, that qe > 2Q.
Now let t ∈ N be the unique positive integer such that Qt−1 < q < Qt and let
m := (Qt /q)g(n). Then m > g(n) and
h(m) = h(g(n)) + (Qt − qe + qe−1 ). (10.16)
If q < Q, so t = 1, then since qe > 2Q we have
qe
< −Q
−qe + qe−1 ≤ −
2
so the second term on the right-hand side of (10.16) is negative. If q > Q
we get
Qt − qe + qe−1 < qQ − q(q − 1) ≤ qQ − qQ = 0.
Therefore, in each case we have h(m) < h(g(n)) ≤ n. But m > g(n), which is a
contradiction. Therefore qe ≤ 2Q, from which the claim follows directly.
278 Other Equivalents to the Riemann Hypothesis
 e
(6) Now let g(n) = 1≤ j≤s q j j be the prime factorization. Write log g(n) as
the sum of the terms log q j where e j = 1 and e j log q j where e j > 1. By Step
(2) the first sum is at most 2 + log F(n) + log P. √By Step (5) each term in the
second sum is at most log 2P and there at most 2P terms. Therefore
√
log F(n) ≤ log g(n) ≤ 2 + log F(n) + log P + 2P log 2P
log g(n)
=⇒ lim = 1,
n→∞ P
and so by Steps (1) and (2) we have

log g(n) ∼ P ∼ log F(n) ∼ n log n,

which completes the proof. 

The present author found parts of the proof of the theorem of Shah [156]
mysterious, so here we present a different derivation of the same result, based
on the proof of Landau’s theorem given by William Miller, Theorem 10.39.
Theorem 10.40 As n → ∞,
  
 loglog n 1
log g(n) = n log n 1 + +O .
2 log n log n

of the proof of Theorem 10.39 we have P ∼ n log n,
Proof (1) By part (2) 
and we can write P = n log n (1 + α), so that α → 0 as n → ∞. Also we have
A(P − 1) ≤ n < A(P), so by Lemma 10.37 we have
 2 
P2 P
n= +O
2 log P log2 P
 
n log n (1 + α)2 n
= +O
log n + loglog n + 2 log(1 + α) log n
 
1 + 2α + α 2
1
=⇒ 1 =   +O .
loglog n α log n
1+ +O
log n log n
Therefore
 
loglog n 1
α= + O(α ) + O
2
2 log n log n
 2  
loglog n loglog n
(α + O(α2 ))2 = +O
2 log n log2 n
⎛ 2 ⎞  
⎜⎜⎜⎜ loglog n ⎟⎟⎟⎟ 1
α (1 + O(α)) = O ⎝
2
⎠=O .
log n log n
 10.10 The Order of Elements of the Symmetric Group 279

Therefore
 
loglog n 1
α= +O .
2 log n log n
(2) By Steps (1) and (6) of the proof of Theorem 10.39 we have
√
θ(P − 1) ≤ log g(n) ≤ 2 + θ(P − 1) + log P + 2P log(2P).
Therefore, using the result of Step (1),
⎛ ⎛√ ⎞⎞
⎜⎜⎜ ⎜⎜⎜ P log P ⎟⎟⎟⎟⎟⎟
log g(n) = θ(P − 1) ⎜⎝1 + O ⎜⎝ ⎟⎠⎟⎠
θ(P)
     
1 log P
= (P − 1) 1 + O 1+O √
log P P
        
1 1 log P
= P 1+O 1+O 1+O √
P log P P
  
1
= P 1+O
log P
     
 loglog n 1 1
= n log n 1 + +O 1+O
2 log n log n log n
  
 loglog n 1
= n log n 1 + +O
2 log n log n
and this completes the proof. 

Figure 10.7 is a plot of log g(n)/ n log n.
Nicolas studied the properties of g(n) in detail in the 1960s, and Massias
continued this study, publishing related work in his Limoges doctoral thesis
in 1985. Several papers were also published on the topic by these two authors
together with Robin. See the references in [112] and [113]. The first set of
results stated below are for the main part proved in detail in Nicolas’ paper
[125], where the methods are similar to those used in Chapter 6. There is
also a summary with sketched proofs in [113]. The RH equivalence is proved
in [112]. The introductory results we have given above should be sufficient
background for the reader to read these papers. 
Recall the definition: h(1) = 0 and, if n = p|n pv p (n) is the standard prime
factorization set,

h(n) := pv p (n) .
p|n

We have seen in Lemma 10.36 that

g(n) = max m.
h(m)≤n
280 Other Equivalents to the Riemann Hypothesis
log g(n)/ n logn
1.0

0.9

0.8

0.7

0.6

n
0 50 100 150 200

Figure 10.7 Values of the function log g(n)/ n log n for 3 ≤ n ≤ 1000.

It follows from this expression that we have m = g(n) for some n if and only
if m > m implies h(m ) > h(m).
Definition of the set G: we say that n ∈ G if there exists ρ > 0 such that

∀ m ≥ 1, g(m) − ρ log m ≥ g(n) − ρ log n. (10.17)

It follows, using a similar method to that used in the introduction to Chapter 6

to show that colossally abundant numbers are superabundant, that G ⊂ g(N).
To each number ρ > 2/log 2 there corresponds an nρ ∈ G defined in the
following manner:

nρ = pα p (10.18)
p≤x1

with x1 /log x1 = ρ so x1 > 4. We also introduce the real numbers xi , i ≥ 2, by

xii − xii−1
= ρ,
log xi
which enables us to write

∞ 
nρ = p. (10.19)
i=1 p≤xi

We have relationships such as

   
x1 1
x2 = 1+O . (10.20)
2 log x1
 10.10 The Order of Elements of the Symmetric Group 281

Recall that θ := sup{ζ(ρ) : ζ(ρ) = 0}, and that f (n) = Ω± (g(n)) as n → ∞

provided there are two unbounded positive integer sequences (ni ) and (mi )
such that for all i ∈ N we have f (ni ) ≥ g(ni ) and f (mi ) ≤ −g(mi ) .
The symmetric group criterion is a consequence of a more refined result
than Theorem 10.40, and is Theorem 10.41 stated below:
Theorem 10.41 [112, theorem 1]
(i) If θ = 1 there is an a > 0 such that
 √ √ 
−1 −a log n
log g(n) = li (n) + O ne .

If θ < 1 then
  
log g(n) = li−1 (n) + O (n log n)θ/2 .

(ii) If θ > 12 then for all ξ < θ

  
log g(n) = li−1 (n) + Ω± (n log n)ξ/2 .

(iii) If there exists a zero of ζ(s) with real part θ > 12 then
  
log g(n) = li−1 (n) + Ω± (n log n)θ/2 .
(iv) If the Riemann hypothesis is true then for all n sufficiently large

log g(n) < li−1 (n).
(v) If the Riemann hypothesis is true then as n → ∞
 ⎛ √ ⎞
−1
⎜⎜⎜ 2 − 2 ⎟⎟⎟  
log g(n) = li (n) − ⎝ ⎜ ⎟⎠ (n log n)1/4 + Ω± (n log n)1/4 .
3
Theorem 10.42 (Symmetric group criterion) The Riemann hypothesis is
equivalent to the statement that for all n sufficiently large we have

log g(n) < li−1 (n),
where li(n) is the Cauchy principal value of the logarithmic integral
 n
dt
li(n) := .
0 log t

Of course, knowing the maximal order of any element of S n , or any group,

does not give insight into the way the orders of elements are distributed.
In a series of seven papers published in the 1960s and 1970s, commencing
with [55], Erdős and Turan revealed many facts regarding the distribution of
282 Other Equivalents to the Riemann Hypothesis

orders, including that few elements σ 

∈ S n have orders comparable with the
maximal asymptotic order, g(n) ∼ exp( n log n ). They noted for example that
there were (n − 1)! elements σ, consisting of a single cycle, having order n.
They proved that the order of a “generic” element is much smaller than the
maximum. Here we simply state two of their main results.
Theorem 10.43 Given , δ > 0 and n sufficiently large, depending on  and
δ, we have, other than for at most δn! permutations σ:
2 2
e(1/2−) log n ≤ ord(σ) ≤ e(1/2+) log n .
They indicated they could use the method they developed to show also
that:

Theorem 10.44 Provided n → ∞ through integers such that ω(n) → ∞ we

have for σ ∈ S n
 
log ord(σ) − 1 log2 n ≤ ω(n) log3/2 n
 2 
with at most o(n!) exceptions.

10.11 Hilbert–Pólya Conjecture

To describe this conjecture and equivalence we first need some definitions.
Let H be a complex Hilbert space and let Δ be a dense linear subspace and
T, Δ → H a linear transformation. Let
Δ∗ := {x ∈ H : y → x, T y is a continuous linear functional on all of H}.
By the Riesz representation theorem (Volume Two [32, Theorem J.1]), for
each x ∈ Δ∗ there is a unique z ∈ H such that x, T y = z, y for all y ∈ Δ. We
define T ∗ (x) = z, and call the linear transformation T ∗ the adjoint of T . The
map T is called self-adjoint if the domain of its adjoint transformation T ∗ is
also Δ, so Δ∗ = Δ, and on this domain we have T = T ∗ , so for all f, g ∈ Δ we
have T f, g =  f, T g. Thus T is symmetric, a weaker condition than being
self-adjoint.
The Hilbert–Pólya conjecture, as formulated by Hugh Montgomery in
1973 [119] possibly for the first time, is that if we write each non-trivial
zero of ζ(s) with positive imaginary part as 12 + iγn , then the numbers γn
correspond to the eigenvalues of an unbounded self-adjoint operator T . This
of course would force the γn to be real and solve RH.
Since (see Lorch [110, theorem 4-1]) a self-adjoint transformation defined
on all of H is necessarily bounded, we must have Δ  H.
It appears that the conjecture goes back to the very conception of RH.
In considering what was left of Riemann’s handwritten working paper
collection in the Göttingen University library, John Keating found a note
 10.11 Hilbert–Pólya Conjecture 283

relating to the stability of a rotating fluid on the same page as notes on the
zeros of ζ(s) (see du Sautoy [151, p. 286]). The condition for stability of
the fluid subject to perturbation was that a set of eigenvalues should be on a
straight line! An implicit connection between RH and something physical!
Andrew Odlyzko attempted, in the early 1980s while Pólya was still alive,
to trace the formulation of the conjecture back to Pólya and Hilbert by
corresponding with Pólya (see Volume Two [32, Chapter 5]). Pólya recalled
answering a question of Landau as to whether or not there was a physical
reason why RH should be true. Pólya’s answer was that it would be the case
if the nontrivial zeros of the function ξ(s) were so connected with the physical
problem that RH would be equivalent to all of the eigenvalues of that problem
being real. Odlyzko was not able to find any reported statement by Hilbert,
but the name of the conjecture is now well established by use.
While up to the time of writing Hilbert–Pólya has failed to provide a proof
of RH, the conjecture has inspired a great deal of work and progress in
the intersection of mathematics and physics. The main focus has been to
design the Hilbert space H, domain Δ and transformation T . Macroscopic
classical mechanical systems, quantum mechanical and quite a few others
have been investigated, but no one seems to have come close to providing
what is needed. Some believe asking for the transformation to be self-adjoint
is too much, or that the structures should be constructed out of mathematics
alone, i.e. starting with the integers. Sir Michael Berry and his co-workers
have given quite precise specifications on what a successful structure might
need to satisfy [11].
As part of this story of successful synergies between mathematics and
physics, there is the well-known and often retold interaction between Hugh
Montgomery and Freeman Dyson over tea in the Institute for Advanced
Study, Princeton [151, p. 262]. This became related to the great body of
work on random matrices, which was originally developed as a way to study,
statistically, the energy levels of heavy nuclei, i.e. those with large numbers
of neutrons and protons. The relationships with the results of computations
of zeta zero imaginary parts, especially those carried out by Andrew Odlyzko
[79, 130], have been startling.
Background reading might include Conrey [41], Katz and Sarnak [90] and
Rudnick and Sarnak [147]. On the physics side, there is the survey preprint
of Schumayer and Hutchinson [155].
One of the remarkable particular discoveries of random matrix theory
has been very precise expressions for moments of distribution functions,
providing corresponding conjectures for the moments of ζ(s). Here are the
conjectured moments of John Keating and Nina Snaith [91] for ζ(s) for all
k ∈ N and s on the critical line:
  k 2
1 T 1
|ζ( 2 + it)|2k dt ∼ gk a(k) log 2π
T
,
T 0
284 Other Equivalents to the Riemann Hypothesis

where
 1
k 2 
∞ 
Γ(m + k)
2
1
a(k) := 1−
p
p m=0
m!Γ(k) pm

and

g1 = 1, g2 = 1/12, g3 = 42/9!, g4 = 2024/16!,

and if

n−2
G(k + 1)2
G(n) := i! =⇒ gk = .
i=0
G(2k + 1)

Estimates for moments of zeta have been very significant, and have been
so for a long time. For example Landau in his treatise of 1908 included the
following theorem for the case k = 1 [51, section 9.7]: Given  > 0 and σ0 > 12
there is a T 0 such that for all σ ≥ σ0 and T ≥ T 0 we have
  T 
 1 
 |ζ(σ + it)| dt − ζ(2σ) < .
2
T −1 1 

This estimate was used by Bohr and Landau [51, section 9.6], together with
Jensen’s formula (Volume Two [32, Theorem B.5]), to show that the number
of zeta zeros in any rectangle R := [δ, 1] × [0, T ] for any δ > 12 is bounded
above by Kδ T . Since the total number of zeta zeros N(T ) in [0, 1] × [0, T ]
satisfies N(T ) ∼ T/(2π) log(T/(2πe)), the proportion of roots in R divided by
the total number up to T tends to zero as T → ∞. This result is still one of
the best pieces of analytic evidence for the truth of RH.
For more recent significant developments on the topic of estimates for
moments and references, there is the article of Soundararajan [160].
Searching for a single self-adjoint transformation which would resolve
RH could be asking too much. For example associating each zero with an
individual transformation depending on the zero would give the same result.
The work of Ross Barnett and the author leading to [21], consistent with
random matrix theory, seems to indicate a relationship between rotations and
zeta zeros, which has yet to be fully developed.

Theorem 10.45 (Hilbert–Pólya criterion) The Riemann hypothesis is equiv-

alent to the following property: for each critical zero ρ of ζ(s), when written
in the form ρ = 12 + iη, there is a self-adjoint transformation T ρ of a Hilbert
space Hρ such that η is an eigenvalue of T ρ .

Proof If RH is true let H = 2 (N, C) with Euclidean inner product, and for n ∈
N let en := (0, 0, . . . , 1, 0, . . . ) be the nth standard basis element. If ρn = 12 + iηn
 10.12 Epilogue 285

is the nth complex zero with positive imaginary part, define T : Δ → H by

setting T (en ) = ηn en , so

T (x) = ηn x n e n .
n∈N

Let Δ := {x ∈ H : T (x) < ∞}. Since the linear subspace generated by any
finite subset of the (en ) is in Δ, Δ is dense in H. Since by RH each ηn is real,
T is self-adjoint. Setting T ρ = T completes the derivation. 

10.12 Epilogue
Looking back over the equivalences to RH described in Chapters 4
through 10, the reader might be surprised at their variety, but also at the
few properties of ζ(s) which have been used to derive them: von Mangoldt’s
theorem, a simple zero-free region, both known for over 100 years, and a
better critical zero line height H – but even deriving improvements to H often
uses methods based on those dating back to Turing. The functional equation
barely appears, neither does the lower bound for the proportion of zeros on
the critical line, nor upper bounds for zeros possibly off the line.
In addition, the properties of primes that are used are slight and based on
not much more than the prime number theorem – no quadratic reciprocity,
no modular restrictions. It is some explicit estimates for arithmetic quantities
which are used to derive the simple inequalities which have been shown to
be equivalent to RH. Work on these estimates has thus been shown to be
invaluable.
Another observation is that when RH fails, the failure has dramatic
consequences. Just one off-critical-line zero will result in an infinite number
of failures for an equivalent inequality.
One approach to resolving RH, begun as reported in Chapter 8, would
be to further explore properties of potential counterexamples, or sets of
counterexamples, i.e. integers which do not satisfy an equivalent inequality.
Finding a set of restrictions which are inconsistent would be the goal.
Another approach would be to work at weakening a given inequality and
strengthening a related unconditional form, until they merge.
Bringing to bear on equivalences recent better understanding of prime
properties, such as primes in progressions and infinite subsequences of
primes with pn+1 − pn ≤ K for some explicit (and small) value of K, as well
as more classical properties, could also be fruitful. In this way it is properties
of primes which could resolve RH, rather than the reverse!
In addition to the equivalences described here, there are a host of other
equivalences which have been demonstrated. These are often fascinating and,
on the face of it, it is difficult to detect any connection to RH. Many of these
are described in Volume Two [32].
286 Other Equivalents to the Riemann Hypothesis

One thing is for certain – when it comes to equivalences to RH, the last
word has not been spoken.

10.13 Unsolved Problems

(1) From Section 10.3, the trivial bound is
m 
  n j 
 f j −  n2 .
j=1
m

Improve this bound by finding a lower power of n.

(2) Again from Section 10.3, show for all j with 1 ≤ j ≤ m that
 
 f n − j  ≤ 1 .
 j m n
To this end, using telescoping sums from 0 to 1 and the bound f j+1n
− f jn ≥
2 n
1/n we should be able to get upper and lower bounds for f j in terms
of j.
(3) Prove that if the Riemann hypothesis is false there exists a zero ρ = β + iγ
of ζ  (s) with 0 < β < 12 .
(4) Find an explicit lower bound for n in the RH equivalence of
Section 10.10.
(5) In Section 10.9 find the relationship between the result of Hall and
Hayman and Theorem 10.32.
 Appendix A
Tables

The following RHpack (see Appendix B) variables and functions relate

to the material in this appendix: GenerateHC, GenerateHR, GenerateSA,
GenerateCA, GenerateLandau, FindNextXA, CAPrimes, InitialNR,
InitialHC, InitialSA, InitialSA1000, InitialCA, InitialCA150, InitialXA,
InitialLandau, InitialCAPrimes, HardyRamanujanToPrimorialForm and
PrimorialFormToInteger.

A.1 Extremely Abundant Numbers

See Chapter 9 and the item A217867 in the Online Encyclopedia of Integer
Sequences [158]. Recall that

σ(n)
G(n) := , n ≥ 2.
n loglog n

Table A.1 The first seven extremely abundant numbers.

Rank Factors eγ − G(n)

1 25 · 32 · 5 · 7 0.0252580790
2 2 · 3 · 5 · 7 · 112 · 132 · 171 · · · 1131
8 5 3 2
0.0238889777
3 28 · 35 · 53 · 72 · 112 · 132 · 171 · · · 1271 0.0237040357
4 28 · 35 · 53 · 72 · 112 · 132 · 171 · · · 1311 0.02342819838
5 28 · 35 · 53 · 72 · 112 · 132 · 171 · · · 1371 0.0232960582
6 28 · 35 · 53 · 72 · 112 · 132 · 171 · · · 1391 0.0228625950
7 29 · 35 · 53 · 72 · 112 · 132 · 171 · · · 1391 0.0228130145

287
288 Tables

A.2 Small Numbers Not Satisfying Robin’s Inequality

See Chapter 8. Numbers not satisfying Robin’s inequality satisfy

σ(n)
> 1.
neγ loglog n

Table A.2 Numbers n ≤ 5041 with σ(n)/n ≥ eγ loglog n.

Rank n Factors σ(n)/(neγ loglog n)

1 1 1 —
2 2 2 −2.29784
3 3 3 7.95991
4 4 2 3.00812
5 5 5 1.41579
6 6 2·3 1.92545
7 8 23 1.43797
8 9 32 1.03024
9 10 2·5 1.21174
10 12 22 · 3 1.43927
11 16 24 1.06673
12 18 2 · 32 1.14614
13 20 22 · 5 1.07462
14 24 23 · 3 1.21395
15 30 2·3·5 1.10079
16 36 22 · 32 1.1196
17 48 24 · 3 1.07517
18 60 2 ·3·5
2
1.11527
19 72 23 · 32 1.04641
20 84 2 ·3·7
2
1.00581
21 120 23 · 3 · 5 1.07559
22 180 22 · 32 · 5 1.03387
23 240 24 · 3 · 5 1.0231
24 360 23 · 32 · 5 1.02942
25 720 24 · 32 · 5 1.00087
26 840 23 · 3 · 5 · 7 1.0094
27 2 520 23 · 32 · 5 · 7 1.01322
28 5 040 24 · 32 · 5 · 7 1.00556
 A.3 Superabundant Numbers 289

A.3 Superabundant Numbers

Recall that a superabundant number n is a positive integer for which
σ(m)/m < σ(n)/n for all 1 ≤ m < n. See the item A004394 in the Online
Encyclopedia of Integer Sequences [158].

Table A.3 The first 31 superabundant numbers.

Rank n Factors σ(n)/n

1 1 1 —
2 2 2 −2.29784
3 4 22 1.750
4 6 2·3 2.000
5 12 22 · 3 2.333
6 24 23 · 3 2.500
7 36 22 · 32 2.528
8 48 24 · 3 2.583
9 60 22 · 3 · 5 2.800
10 120 23 · 3 · 5 3.000
11 180 22 · 32 · 5 3.033
12 240 24 · 3 · 5 3.100
13 360 23 · 32 · 5 3.250
14 720 24 · 32 · 5 3.385
15 840 23 · 3 · 5 · 7 3.429
16 1 260 22 · 32 · 5 · 7 3.467
17 1 680 24 · 3 · 5 · 7 3.543
18 2 520 23 · 32 · 5 · 7 3.714
19 5 040 24 · 32 · 5 · 7 3.838
20 10 080 25 · 32 · 5 · 7 3.900
21 15 120 24 · 33 · 5 · 7 3.937
22 25 200 24 · 32 · 52 · 7 3.966
23 27 720 23 · 32 · 5 · 7 · 11 4.052
24 55 440 24 · 32 · 5 · 7 · 11 4.187
25 110 880 25 · 32 · 5 · 7 · 11 4.255
26 166 320 24 · 33 · 5 · 7 · 11 4.294
27 277 200 24 · 32 · 52 · 7 · 11 4.327
28 332 640 25 · 33 · 5 · 7 · 11 4.364
29 554 400 25 · 32 · 52 · 7 · 11 4.396
30 665 280 26 · 33 · 5 · 7 · 11 4.398
31 720 720 2 · 32 · 5 · 7 · 11 · 13
4
4.509
290 Tables

A.4 Colossally Abundant Numbers

See the item A004490 in the Online Encyclopedia of Integer Sequences
[158]. Recall that n is colossally abundant if there is an  > 0 with, for all
m > 1, σ(n)/n1+ ≥ σ(m)/m1+ . See Chapter 6 for some larger examples.

Table A.4 The first 30 colossally abundant numbers.

Rank n Factors Prime

1 2 2 2
2 6 2·3 3
3 12 22 · 3 2
4 60 22 · 3 · 5 5
5 120 23 · 3 · 5 2
6 360 23 · 32 · 5 3
7 2 520 23 · 32 · 5 · 7 7
8 5 040 24 · 32 · 5 · 7 2
9 55 440 24 · 32 · 5 · 7 · 11 11
10 720 720 24 · 32 · 5 · 7 · 11 · 13 13
11 1 441 440 25 · 32 · 5 · 7 · 11 · 13 2
12 4 324 320 25 · 33 · 5 · 7 · 11 · 13 3
13 21 621 600 25 · 33 · 52 · 7 · 11 · 13 5
14 367 567 200 25 · 33 · 52 · 7 · 11 · 13 · 17 17
15 6 983 776 800 25 · 33 · 52 · 7 · 11 · 13 · 17 · 19 19
16 160 626 866 400 25 · 33 · 52 · 7 · 11 · 13 · 17 · 19 · 23 23
17 321 253 732 800 26 · 33 · 52 · 7 · 11 · 13 · 17 · 19 · 23 2
18 9 316 358 251 200 26 · 33 · 52 · 7 · 11 · 13 · 17 · 19 · 23
· 29 29
19 288 807 105 787 200 26 · 33 · 52 · 7 · 11 · 13 · 17 · 19 · 23
· 29 · 31 31
20 2 021 649 740 510 400 26 · 33 · 52 · 72 · 11 · 13 · 17 · 19 · 23
· 29 · 31 7
21 6 064 949 221 531 200 2 · 3 · 5 · 7 · 11 · 13 · 17 · 19 · 23
6 4 2 2

· 29 · 31 3
22 224 403 121 196 654 400 26 · 34 · 52 · 72 · 11 · 13 · 17 · 19 · 23
· 29 · 31 · 37 37
23 9 200 527 969 062 830 400 26 · 34 · 52 · 72 · 11 · 13 · 17 · 19 · 23
· 29 · 31 · 37 · 41 41
24 c24 26 · 34 · 52 · 72 · p5 · · · p14 43
25 c25 27 · 34 · 52 · 72 · p5 · · · p14 2
26 c26 27 · 34 · 52 · 72 · p5 · · · p15 47
27 c27 27 · 34 · 52 · 72 · p5 · · · p16 53
28 c28 27 · 34 · 52 · 72 · p5 · · · p17 59
29 c29 27 · 34 · 53 · 72 · p5 · · · p17 5
30 c30 27 · 34 · 53 · 72 · p5 · · · p18 61
 A.5 Primes to Make Colossally Abundant Numbers 291

A.5 Primes to Make Colossally Abundant Numbers

See Chapter 6 and the item A073751 in the Online Encyclopedia of Integer
Sequences [158].

{2, 3, 2, 5, 2, 3, 7, 2, 11, 13, 2, 3, 5, 17, 19, 23, 2, 29, 31, 7,

3, 37, 41, 43, 2, 47, 53, 59, 5, 61, 67, 71, 73, 11, 79, 2, 83, 3,
89, 97, 13, 101, 103, 107, 109, 113, 127, 131, 137, 139, 2, 149, 151,
7, 157, 163, 167, 17, 173, 179, 181, 191, 193, 197, 199, 19, 211, 3,
223, 227, 229, 5, 233, 239, 241, 251, 2, 257, 263, 269, 271, 277,
281, 283, 293, 23, 307, 311, 313, 317, 331, 337, 347, 349, 353, 359,
367, 373, 379, 383, 389, 397, 401, 409, 419, 421, 431, 433, 439, 443,
449, 457, 461, 2, 463, 467, 29, 479, 487, 491, 499, 503, 509, 521,
523, 541, 31, 547, 11, 557, 563, 3, 569, 571, 577, 587, 593, 599,
601, 607, 613, 617, 619, 631, 641, 643, 647, 653, 659, 661, 673, 677,
683, 691, 701, 709, 719, 727, 733, 739, 743, 751, 757, 761, 37, 769,
773, 787, 797, 809, 811, 7, 821, 823, 827, 829, 839, 2, 853, 857,
859, 863, 877, 881, 883, 887, 13, 907, 911, 919, 5, 929, 41, 937,
941, 947, 953, 967, 971, 977, 983, 991, 997, 1009, 1013, 1019, 1021,
43, 1031, 1033, 1039, 1049, 1051, 1061, 1063, 1069, 1087, 1091, 1093,
1097, 1103, 1109, 1117, 1123, 1129, 1151, 1153, 1163, 1171, 1181,
1187, 1193, 1201, 1213, 1217, 47, 1223, 1229, 1231, 1237, 1249, 1259,
1277, 1279, 1283, 1289, 1291, 1297, 1301, 1303, 1307, 1319, 1321,
1327, 1361, 1367, 1373, 1381, 1399, 1409, 1423, 1427, 1429, 1433,
1439, 1447, 1451, 1453, 1459, 1471, 3, 1481, 1483, 1487, 1489, 1493,
1499, 1511, 1523, 1531, 1543, 2, 53, 1549, 1553, 1559, 1567, 1571,
1579, 1583, 1597, 1601, 1607, 1609, 1613, 1619, 1621, 1627, 1637,
1657, 1663, 1667, 1669, 1693, 1697, 1699, 1709, 1721, 1723, 1733,
1741, 1747, 1753, 1759, 1777, 1783, 1787, 1789, 1801, 1811, 1823,
1831, 1847, 1861, 1867, 1871, 1873, 1877, 1879, 1889, 1901, 1907, 59,
1913, 1931, 1933, 1949, 1951, 17, 1973, 1979, 1987, 1993, 1997, 1999,
2003, 2011, 2017, 2027, 2029, 2039, 61, 2053, 2063, 2069, 2081, 2083,
2087, 2089, 2099, 2111, 2113, 2129, 2131, 2137, 2141, 2143, 2153,
2161, 2179, 2203, 2207, 2213, 2221, 2237, 2239, 2243, 2251, 2267,
2269, 2273, 2281, 2287, 2293, 2297, 2309, 2311, 2333, 2339, 2341}
292 Tables

A.6 Small Numbers Satisfying Nicolas’ Reversed Inequality

The numbers in bold are primorials and those marked with an asterisk (∗ )
are multiples of a primorial which are less than the next primorial.

{1, 2, 3, 4∗ , 5, 6∗ , 8∗ , 9, 10∗ , 12∗ , 14∗ , 15, 16∗ , 18∗ , 20∗ , 22∗ , 24∗ , 26∗ ,
28∗ , 30, 36, 40, 42, 48, 50, 54, 60∗ , 66, 70, 72, 78, 84, 90∗ , 96, 102, 108,
114, 120∗ , 126, 132, 138, 140, 144, 150∗ , 156, 162, 168, 174, 180∗ , 186,
192, 198, 204, 210, 216, 222, 228, 234, 240, 246, 252, 258, 264, 270,
276, 294, 300, 306, 312, 330, 336, 342, 360, 378, 390, 396, 420∗ , 450,
462, 468, 480, 504, 510, 528, 540, 546, 570, 588, 600, 630∗ , 660, 672,
690, 714, 720, 750, 756, 780, 798, 810, 840∗ , 858, 870, 882, 900, 924,
930, 960, 966, 990, 1008, 1020, 1050∗ , 1080, 1092, 1110, 1122, 1134,
1140, 1170, 1176, 1200, 1218, 1230, 1260∗ , 1290, 1302, 1320, 1350,
1380, 1386, 1410, 1428, 1440, 1470∗ , 1500, 1530, 1554, 1560, 1590,
1596, 1620, 1638, 1650, 1680∗ , 1710, 1722, 1740, 1770, 1800, 1830,
1848, 1860, 1890∗ , 1920, 1932, 1950, 1980, 2010, 2040, 2070, 2100∗ ,
2130, 2142, 2160, 2184, 2190, 2220, 2250, 2280, 2310, 2340, 2370,
2394, 2400, 2430, 2460, 2490, 2520, 2550, 2580, 2610, 2640, 2670,
2700, 2730, 2760, 2772, 2790, 2820, 2850, 2856, 2880, 2910, 2940,
2970, 3000, 3030, 3060, 3090, 3120, 3150, 3180, 3210, 3234, 3240,
3270, 3276, 3300, 3330, 3360, 3390, 3420, 3450, 3480, 3510, 3540,
3570, 3600, 3630, 3660, 3690, 3696, 3720, 3780, 3810, 3822, 3870,
3900, 3930, 3960, 3990, 4020, 4080, 4110, 4140, 4158, 4170, 4200,
4230, 4260, 4290, 4350, 4368, 4380, 4410, 4440, 4560, 4590, 4620∗ ,
4650, 4680, 4770, 4830, 4920, 4950, 5040, 5070, 5082, 5100, 5130,
5160, 5220, 5250, 5280, 5460, 5520, 5544, 5550, 5580, 5610, 5670,
5700, 5850, 5880, 5940, 6006, 6090, 6120, 6210, 6240, 6270, 6300,
6510, 6600, 6630, 6720, 6840, 6900, 6930∗ , 6960, 7020, 7140, 7260,
7350, 7410, 7560, 7590, 7650, 7770, 7800, 7854, 7920, 7980, 8160,
8190, 8250, 8280, 8400, 8550, 8580, 8610, 8670, 8778, 8820, 8910,
8970, 9030, 9120, 9180, 9240∗ , 9282, 9360, 9450, 9570, 9660, 9690,
9750, 9870, 9900}
 A.8 Maximum Order of an Element of the Symmetric Group 293

A.7 Heights of Integers


The height h(n) of an integer n = pα1 1 · · · pαmm is ni=1 pαi i .

Table A.5 Initial values of h(n); compare with item A008475 in OEIS [158].

n h(n) n h(n) n h(n) n h(n)

1 0 13 13 25 25 37 37
2 2 14 9 26 15 38 21
3 3 15 8 27 27 39 16
4 4 16 16 28 11 40 13
5 5 17 17 29 29 41 41
6 5 18 11 30 10 42 12
7 7 19 19 31 31 43 43
8 8 20 9 32 32 44 15
9 9 21 10 33 14 45 14
10 7 22 13 34 19 46 25
11 11 23 23 35 12 47 47
12 7 24 11 36 13 48 19

A.8 Maximum Order of an Element of the Symmetric Group

Table A.6 Initial values of g(n); compare with item A000793 in OEIS [158].

n g(n) n g(n) n g(n) n g(n)

1 1 13 60 25 1 260 37 13 860
2 2 14 84 26 1 260 38 16 380
3 3 15 105 27 1 540 39 27 720
4 4 16 140 28 2 310 40 30 030
5 6 17 210 29 2 520 41 32 760
6 6 18 210 30 4 620 42 60 060
7 12 19 420 31 4 620 43 60 060
8 15 20 420 32 5 460 44 60 060
9 20 21 420 33 5 460 45 60 060
10 30 22 420 34 9 240 46 60 060
11 30 23 840 35 9 240 47 120 120
12 60 24 840 36 13 860 48 120 120
 Appendix B
RHpack Mini-Manual

B.1 Introduction
This appendix is the manual for a set of functions written to assist the reader
to reproduce, and possibly extend, the calculations mentioned in the main
part of the book. The software for the package is provided over the world
wide web at the webpage for the book linked to the author’s homepage:
www.math.waikato.ac.nz/∼kab
and is in the form of a standard Mathematica add-on package. To use the
functions in the package you will need to have a version of Mathematica at
level 7.0 or higher.

B.1.1 Installation
First connect to the website given in the paragraph above and click on the
link for RHpack listed under the heading “Software” to get to the RHpack
homepage. Instructions on downloading the files for the package will be
given on the homepage. If you have an earlier version of RHpack, first delete
the file RHpack.m. The name of the package file is RHpack.m. To install
the package, if you have access to the file system for programs on your
computer, place a copy of the file in the standard repository for Mathematica
packages or any other directory, which will be listed by evaluating $Path in
Mathematica, to which you have access. You can then type <<RHpack.m
and then press the Shift and Enter keys to load the package. You may need
administrator or super-user status to complete this installation. Alternatively,
place the package file RHpack.m anywhere in your own file system where it
is safe and accessible.
Instructions for Windows systems The package can be loaded by typing
SetDirectory["c:\\your\\directory\\path"]<Shift/Enter>
<<RHpack.m;ResetDirectory[]<Shift/Enter>

294
 B.1 Introduction 295

or
Get["RHpack.m",Path->{"c:\\your\\directory\\path"}]
<Shift/Enter>
where the path-name in quotes should be replaced by the actual path-name
of directories and subdirectories which specify where the package has been
placed on the computer you are using. If the double backslash does not work,
try a single backslash.
Instructions for Unix/Linux systems These are the same as for
Windows, but the path-name syntax style should be like
/usr/home/your/subdirectory
Instructions for Macintosh systems These are the same as for Windows,
but the path-name syntax style should be like
HD:Users:ham:Documents:
All systems The package should load printing a message. The functions
of RHpack are then available to any Mathematica notebook you subsequently
open.

B.1.2 About This Mini-Manual

This appendix contains a list of all of the functions available in the package
RHpack followed by a manual entry for each function in alphabetical order.
To obtain information about bug fixes and updates to RHpack, consult the
website for the text linked to the author’s homepage (see Section B.1).
There are many issues to do with computer algebra and mathematical
software that will arise in any serious evaluation or use of Mathematica and
RHpack. Some notes follow:
RHpack function arguments are first evaluated and then checked for correct
data type. If the user calls a function with an incorrect number of arguments
or an argument of incorrect type, rather than issue a warning and proceeding
to compute (default Mathematica style), RHpack prints an error message,
aborts the evaluation and returns the user to the top level, no matter how
deeply nested the function which makes the erroneous call happens to be
placed. This is to assist users to debug programs which include calls to this
package.
Every function in RHpack has a short name which can be used in place
of the Mathematica style long name. Functions which include numerical
floating-point evaluations frequently have an integer argument “d”. This
represents the precision which is used for internal computations, and not the
accuracy of the resulting output.
The time that it takes for functions to evaluate is variable. For some
argument ranges, stored values have been used to speed up the computations.
This applies for example to M(x) and θ(x).
296 RHpack Mini-Manual

Appendix A gives tables of numbers such as superabundant and colossally

abundant. RHpack includes in some instances larger sets of numbers: for
example 1000 superabundant and 20 extremely abundant numbers are stored
explicitly.

B.2 RHpack Functions

 AllPrimesFactor (apf)
This function returns the prime factors of a natural number with all
primes up to the maximum prime divisor appearing.
AllPrimesFactor[n] −→ L
n is a natural number,
L is a list of lists of the form {{2, a2 }, {3, a3 }, . . . , {p, a p }} where p is the
maximum prime dividing n and, for each sublist {q, aq }, aq is the
maximum power to which q divides n.

 BacklundB (blb)
This function computes an upper bound for the error in the approx-
imation to the number of critical zeta zeros up to height T , given
by
|N(T ) − F(T )| ≤ B(T ).
We have B(T ) := a1 log T + a2 loglog T + a3 where the values ai are stored
in the global variables $a1 , $a2 , $a3 , and are best possible at the time of
writing. See Section 2.2.
BacklundB[T, d] −→ r
T is a positive real number representing the height up to which zeta zeros
are being counted,
d is a precision to which evaluations to compute B(T ) will be made
internally,
r is a floating-point number being the value of B(T ).

 CAPrimes (cap)
This function computes a list of primes for generating colossally
abundant numbers – see Table A.5. This has been shown to work up
to n = 1011 .
CAPrimes[n] −→ L
n is a natural number being the number of primes which are to be generated,
L is an ordered list of primes such that the product of the first j primes is the
jth colossally abundant number.
 B.2 RHpack Functions 297

 CheckCNk (cnk)
This function computes the k ≤ k0 such that c(N1 ) ≤ c(Nk ) ≤ c(N66 ) fails,
to precision d – see Theorem 5.34.
CheckCNk[k0, d] −→ L
k0 is an integer being the upper bound for which the values of c(Nk ) are
checked,
d is a natural number being the precision to which internal calculations are
made,
L is a list of values of k for which either c(Nk ) < c(N1 ) or c(Nk ) > c(N66 ).

 ColossallyAbundantF (caf)
This function evaluates the function F(x, α) defined in Section 6.3, to
precision d.
ColossallyAbundantF[x, a, d] −→ r
a is a natural number,
x is a real number greater than 1,
d is a natural number being the precision for internal computations,
r is a floating-point number being the value of
 
1
log x 1 + a a−1 .
x + x + ··· + x

 ColossallyAbundantInteger (cai)
This function returns the colossally abundant integer corresponding to a
given e > 0 – see Section 6.3.
ColossallyAbundantInteger[e] −→ L
e is a positive real number,
L is a list, the first element being a colossally abundant number, the second
being a list of prime powers for the prime divisors of n, in ascending
order.

 ColossallyAbundantQ (caq)
This function determines whether n is a colossally abundant integer –
see Table A.4.
ColossallyAbundantQ[n] −→ P
n is a natural number,
P is True if n is colossally abundant and False otherwise.
298 RHpack Mini-Manual

 CompositeQ (coq)
This function determines whether or not n is a composite integer.
CompositeQ[n] −→ P
n is a natural number,
P is True if n is composite, False otherwise.

 ComputeC (coc) 
This function evaluates c(n) = (n/ϕ(n) − eγ loglog n) log n, to precision
d – see Section 5.4.
ComputeC[n, d] −→ r
n is a natural number 2 or more,
d is the precision to which evaluations to compute c(n)will be made
internally,
r is a floating-point number.

 ComputeDeltas (cod)
This function computes the values of
ψ(qi ) − θ(qi )
δi := √
qi+1
which are used in the proof of Lemma 5.18. The qi are the ordered
squarefull powers of primes up to 5993 . It also prints a table of the first
11 δi .
ComputeDeltas[d] −→ min
d is a natural number being the precision for internal computations,
min is the minimum value of the δi .

 Computef (cof) 
This function evaluates f (x) = eγ log θ(x) p≤x (1 − 1/p), to precision d –
see Chapters 5 and 7.
Computef[x, d] −→ r
x is a positive real number,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the value of f (x).

 ComputeH (coh)
This function computes the value of
log
x/log 2
H(x) := 1 + x1/n−1/3 .
n=4
 B.2 RHpack Functions 299

See the proof of Lemma 5.17.

ComputeH[x, d] −→ r
x is a positive real number not less than 16,
d is a natural number being the precision for internal computations,
r is a floating-point number being the value of H(x).

 ComputeKm (ckm)
Note the definition
 1
km (h) = ,
γ>h
γm+1

where ρ = β+iγ are complex zeta zeros. This function finds upper bounds
for the values km (0) recursively, to precision d, starting with the bound
k1 (0) ≤ 0.0463 – see Table 3.1.
ComputeKm[m, d] −→ L
m is a natural number,
d is a natural number being the precision for internal computations,
L is a list of floating-point numbers representing computed upper bound for
{k1 (0), k2 (0), . . . , km (0)}.

 ComputeTheta (cth) 
This function evaluates θ(x) = p≤x log p.
ComputeTheta[x, d] −→ r
x is a real number,
d is a natural number being the precision for internal computations,
r is a floating-point number which is an approximation to θ(x).

 ComputeThetaValues (ctv)
This function computes b values, at integer points commencing at 1, of
a list of θ(x) values, to precision d.
ComputeThetaValues[b, d] −→ L
b is a natural number being the number of θ(x) values to be computed,
d is a natural number being the precision for internal computations,
L is a list of b θ(x) values.

 CriticalEpsilonValues (cev)
This function computes a list of critical values at primes up to b and
powers up to jmax, in decreasing order – see Section 6.3.
CriticalEpsilonValues[b, jmax] −→ L
300 RHpack Mini-Manual

b is a natural number,
jmax is a natural number,
L is a list of critical epsilon values in decreasing order in symbolic form for
primes up pb and exponents j up to jmax.

 DivisibilityGraph (dvg)
This function returns the n × n divisibility graph of Section 10.5.
DivisibilityGraph[n] −→ L
n is a natural number,
L is a list of terms i → j where 1 ≤ i, j ≤ n is a subset of pairs, representing
the divisibility graph.

 ExtraordinaryQ (exq)
This function determines whether n is an extraordinary number up to a
finite number of multiples of n – see Chapter 9.
ExtraordinaryQ[n, b, d] −→ P
n is a natural number,
b is a natural number,
d is a natural number being the precision for internal computations,
P is True if n is left abundant and right abundant up to multiples not greater
than b, both to precision d, otherwise False.

 ExtremelyAbundantQ (eaq)
This function determines whether n is an extremely abundant number –
see Table A.1
ExtremelyAbundantQ[n, d] −→ P
n is a natural number with n ≥ 10 080,
d is a natural number being the precision for internal computations,
P is True if G(m) < G(n), to precision d, for 10 080 ≤ m < n, otherwise False.

 FareyFractions (ffr)
This function computes the list of Farey fractions for denominator n, in
increasing order – see Section 10.3.
FareyFractions[n] −→ L
n is a natural number ,
L is a list of the Farey fractions {0, 1/n, . . . , 1} of order n.

 FHalf (fha)
This function computes F1/2 (z) – see Lemma 5.16.
FHalf[x, d] −→ r
 B.2 RHpack Functions 301

x is a positive real number,

d is a natural number being the precision for internal computations,
r is a floating-point number representing the value of F1/2 (x).

 FindNextXA (fnx)
This function finds an extremely abundant number larger than a given
extremely abundant number – see Chapter 9. This function does not
check that its argument is extremely abundant, and simply checks the
size of Grönwall’s function G(n) against its values at multiples by primes
which preserve its Hardy–Ramanujan status. Because of this, some
extremely abundant numbers will be missed.
FindNextXA[n] −→ m
n is a natural number which should be a Hardy–Ramanujan number,
m is a natural number with G(m) > G(n) or False if it fails to find such an m.

 GenerateCA (gca)
This function generates the b smallest colossally abundant numbers, with
some restriction on the size of b – see Table A.4.
GenerateCA[b] −→ L
b is a natural number,
L is a list of b colossally abundant numbers in increasing order.

 GenerateHC (ghc)
This function generates all highly composite numbers up to b – see
Section 8.2.
GenerateHC[b] −→ L
b is a natural number,
L is a list of the highly composite numbers up to b.

 GenerateHR (ghr)
This function generates a list of all Hardy–Ramanujan numbers up to b
– see Section 8.2.
GenerateHR[b] −→ L
b is a natural number,
L is a sorted list of all Hardy–Ramanujan numbers less than or equal to b.

 GenerateLandau (gla)
This function generates values of Landau’s function g(n) up to n = b –
see Section 10.10. It uses the algorithm of Heinz and Alcover.
GenerateLandau[b] −→ LL
302 RHpack Mini-Manual

b is a natural number,
LL is a list of lists, each sublist being a pair {n, g(n)}.

 GenerateSA (gsa)
This function generates all superabundant numbers up to b – see
Section 6.2.
GenerateSA[b] −→ L
b is a natural number,
L is a list of all superabundant numbers up to b in increasing order.

 GronwallG (grg)
This function evaluates Grönwall’s function G(n) = σ(n)/(n loglog n), to
precision d.
GronwallG[n, d] −→ r
n is a natural number,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the value of G(n).

 HardyRamanujanQ (hrq)
This function determines whether n is a Hardy–Ramanujan integer – see
Section 8.2.
HardyRamanujanQ[n] −→ P
n is a natural number,
P is True if n is a Hardy–Ramanujan number, otherwise False.

 HardyRamanujanToPrimorialForm (hr2pf)
This function converts a HardyRamanujan integer to primorial form –
see Section 6.1.
HardyRamanujanToPrimorialForm[n] −→ L
n is a natural number which is Hardy–Ramanujan,
L is a list of primes in non-increasing order of prime value such that n is the
product of the primorials corresponding to those primes to the given
powers such that each prime is the largest in an associated primorial.

 HighlyCompositeQ (hcq)
This function determines whether n is a highly composite integer – see
Section 6.1.
HighlyCompositeQ[n] −→ P
 B.2 RHpack Functions 303

n is a natural number,
P is True if n is highly composite, otherwise False.

 Initial Set Elements

The following symbols evaluate to sorted lists of the smallest types of
numbers given by
InitialCA (ica) colossally abundant numbers,
InitialCAPrimes (icap) primes to make colossally abundant numbers,
InitialCA150 (ca1) the first 150 colossally abundant numbers in primorial
form,
InitialHC (ihc) the first 41 highly composite numbers ,
InitialLandau (ila) values of Landau’s function g(n),
InitialSA (isa) superabundant numbers,
InitialSA1000 (sa1) the first 1000 superabundant numbers in primorial
form,
InitialNR (inr) natural numbers which do not satisfy Nicolas’ inequality,
InitialXA (ixa) extremely abundant numbers.
Primorial form for a Hardy–Ramanujan integer is the encryption n →
{p1 , p2 , . . . , pm } where the pi are primes, which are not necessarily distinct,
in increasing order, and which are such that n is the product of the primorials
pi #, which in each case is the product of all of the primes up to and
including pi . See the RHpack functions HardyRamanujanToPrimorialForm,
PrimorialFormToInteger and PrimorialFormQ.
 IntegerHeight (iht)
This function computes the function h(n) of Section 10.10.
IntegerHeight[n] −→ m
n is a natural number,
m is a natural number being h(n).

 KFreeQ (kfq)
This function determines whether the integer n is k-free.
KFreeQ[n, k] −→ P
n is a natural number,
k is a natural number,
P is True if n is k-free, otherwise False.

 KthApproximatePrimorial (app)
This function returns the kth primorial, to precision d – see Chapter 5.
KthApproximatePrimorial[k, d] −→ r
k is a natural number,
304 RHpack Mini-Manual

d is a natural number being the precision for internal computations,

r is a floating-point number representing Nk = p1 · · · pk .

 KthPrimorial (kpp)
This function returns the kth primorial Nk as an exact integer.
KthPrimorial[k] −→ m
k is a natural number,
m is a natural number being the value of Nk , the kth primorial.

 LagariasInequalityQ (liq)
This function checks σ(n) < Hn + exp(Hn ) log(Hn ), to precision d – see
Section 7.10.
LagariasInequalityQ[n ,d] −→ P
n is a natural number,
d is a natural number being the precision for internal computations,
P is True if the inequality is satisfied at precision d, False otherwise.

 Lambda (lam)
This function evaluates the function Λ(s) of Section 10.8 for complex s.
Lambda[s, d] −→ r
s is a real or complex number,
d is a natural number being the precision for internal computations,
r is a floating-point complex number.

 LargestZetaZero (lzz)
This function finds the zeta zero ρ = 1
2
+ iγ with maximum γ ≤ h, to
precision d for h ≤ 106 .
LargestZetaZero[h, d] −→ r
h is a positive real number with h ≤ 106 ,
d is a natural number being the precision for internal computations,
r is the value of the y-coordinate of the largest zeta zero with imaginary part
less than or equal to h.

 LeftAbundantQ (laq)
This function determines whether n is a left abundant number – see
Chapter 9.
LeftAbundantQ[n, d] −→ P
n is a natural number,
 B.2 RHpack Functions 305

d is a natural number being the precision for internal computations,

P is True if G(n/p) ≤ G(n), to precision d, for all prime divisors p | n, False
otherwise.

 LogKthPrimorialBounds (lppb)
This function gives upper and lower bounds of Lemma 8.18 for log Nk .
LogKthPrimorialBounds[k, d] −→ L
k is a natural number with k ≥ 198,
d is a natural number being the precision for internal computations,
L is a list of two floating-point numbers {l, u} such that l ≤ log Nk ≤ u.

M
This function evaluates the sum of the Möbious μ(n) function up to x > 0
– see Chapter 4 and Section 10.3.
M[x] −→ s
x is a positive real number,
x
s is an integer representing the sum n=1 μ(n).

 MaxPrime (mxp)
This function finds the maximum prime less than or equal to x.
MaxPrime[x] −→ p
x is a real number,
p is a prime number which is maximal such that p ≤ x.

 NextZetaZero (nzz)
This function finds the next zeta zero with height larger than h, to
precision d. It is suitable only for modest values of h.
NextZetaZero[h, d] −→ r
h is a positive real number,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the y-coordinate of the smallest
zeta zero ρ with γ > h, where ρ = 12 + i 12 .

 NiceFactorInteger (nfi)
This function factors a natural number n with primes to the power 1
appearing as singletons.
NiceFactorInteger[n] −→ L
n is a natural number,
306 RHpack Mini-Manual

L is list of list pairs {p, e} where p is prime and pe  n and e ≥ 2 and singletons
p where p1  n.

 NicolasInequalityQ (niq)
This function checks n/ϕ(n) < eγ loglog n to precision d – see Chapter 5.
NicolasInequalityQ[n, d] −→ P
n is a natural number,
d is a natural number being the precision for internal computations,
P is True if the inequality is satisfied to precision d and False otherwise.

 NicolasTwoFailures (n2f)
This function computes the number of failures to Nicolas inequality at
primorials
 
n 
− e loglog n log n < eγ (4 + γ − log π − 2 log 2)
γ
ϕ(n)
in the range k1 ≤ k ≤ k0 with n = Nk – see the proof of Theorem 5.34 and
Corollary 5.35. This function is suitable only for small values of k.
NicolasTwoFailures[k1, k0, d] −→ n
k1 is a natural number,
k0 is a natural number with k1 ≤ k0,
d is a natural number being the precision for internal computations,
n is a non-negative integer representing the number of integer points k in
[k1, k0] at which the inequality fails at n = Nk .

 NOverPhi (nop)
This function evaluates n/ϕ(n), to precision d – see Chapter 5.
NOverPhi[n, d] −→ r
n is a natural number,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the value n/ϕ(n).

 NthHarmonicNumber (nha)
This function evaluates the n harmonic number, Hn , to precision d – see
Section 7.10.
NthHarmonicNumber[n, d] −→ r
n is a natural number,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the value of Hn .
 B.2 RHpack Functions 307

 NthPrime (npr)
This function returns the nth prime pn where p1 = 2 – see Section 8.5.
NthPrime[n] −→ c
n is a natural number,
d is a natural number being the precision for internal computations,
p is a rational prime being the nth largest prime number.

 NthPrimeBounds (prb)
This function gives the upper and lower bounds of Rosser and
Schoenfeld [145] for the nth prime – see Section 8.5.
NthPrimeBounds[n, d] −→ L
n is a natural number being the index for the nth prime pn with n > 20,
d is a natural number being the precision for internal computations,
L is a list of two floating-point numbers, being {l, u} where l ≤ pn ≤ u.

 PlotCNk (pcnk)
This function prints a plot of the values of c(Nk ) for k1 ≤ k ≤ k0 – see
Chapter 5. It returns the value of c(Nk ) at k = k0.
PlotCNk[k1, k0, d] −→ r
k1 is a natural number,
k0 is a natural number with k1 ≤ k0,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the value of c(Nk0 ).

 PlotDivisibilityGraph (pdg)
This function returns a plot of the divisibility graph of Section 10.5. The
edges are directed and vertices labelled 1 through n.
PlotDivisibilityGraph[n] −→ G
n is a natural number,
G is a labelled, directed graph.

 PlotEulerPhiRatio (pep)
This function prints a plot of the graph of the proportion of positive
integers n for which loglog n < n/ϕ(n) up to b – see Chapter 5. It returns
the value of this proportion at b.
PlotEulerPhiRatio[b] −→ r
b is a positive real number,
308 RHpack Mini-Manual

r is a floating-point number representing the number of integers which

satisfy the inequality divided by b.

 PlotRobinsInequality (pri)
This function prints a plot of n vs σ(n)/n and eγ loglog n on the same
axes, for a ≤ n ≤ b – see Chapter 7.
PlotRobinsInequality[a, b, d] −→ L
a is a natural number,
b is a natural number with a < b,
d is a natural number being the precision for internal computations,
L is a two-element list being {σ(b)/b, eγ loglog b}.

 PlotVonMangoldtPsi (pvm)
This function prints a plot of the explicit formula for ψ0 (x) – see
Section 2.2. It returns the value of this function at the end point of its
domain.
PlotVonMangoldtPsi[a, b, nx, d] −→ r
a is a positive real number being the start of the domain to be plotted,
b is a positive real number with a < b being the end point of the domain to
be plotted,
r is a floating-point number being the value of ψ0 (b),
nx is a natural number being the number of terms of the form xρ /ρ used to
approximate ψ0 (x),
d is a natural number being the precision for internal computations.

 PowerfulQ (pfq)
This function determines whether a natural number n is powerful.
PowerfulQ[n] −→ P
n is a natural number,
P is True if for each prime p | n we have p2 | n, otherwise False.

 PrimePower (prp)
This function returns the maximum power to which p divides n. The
number p need not be prime.
PrimePower[p, n] −→ e
p is a natural number with p > 1,
n is a natural number
e is a non-negative integer being ν p (n).
 B.2 RHpack Functions 309

 PrimorialFormQ (pfq)
This function determines whether a list of integers is in primorial form, i.e. a
list of primes in non-decreasing order
PrimorialFormQ[L] −→ P
L is a list of integers,
P is True if each element of L is prime, else False.

 PrimorialFormToInteger (pf2i)
This function computes the integer corresponding to the primorial form.
PrimorialFormToInteger[P] −→ n
P is a list of primes in non-decreasing order,
n n is a natural number being the product of the primorials corresponding to
the primes in P.

 Psi
This function computes ψ(x) using the expression
∞  
ψ(x) = θ x1/n .
n=1

See Section 2.2.

Psi[x, d] −→ r
x is a positive real number being the point at which ψ is to be evaluated,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the value of ψ(x).

 Psi0
This function computes ψ0 (x) using nz terms – see Section 2.2.
Psi0[x, nz, d] −→ r
x is a positive real number being the point at which ψ0 is to be evaluated,
nz is a natural number being the number of terms in the von Mangoldt sum
which will be used,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the value of ψ0 (x).

 PsiLargeXEpsilon (ple)
This function computes
 
(x) = log x exp(− log x/R ),
to precision d, for R ≥ 16 – see Section 3.4.
PsiLargeXEpsilon[x, R, d] −→ r
310 RHpack Mini-Manual

x is a positive real number,

d is a natural number being the precision for internal computations,
R is a real number with R ≥ 16,
r is a floating-point number representing the value of (x).

 PsiLargeXEpsilon2 (ple2)
This function computes
 
(x) = 2 log x exp(− log x/R ),
to precision d for 8 ≤ R ≤ 18 – see Section 3.4.
PsiLargeXEpsilon2[x, R, d] −→ r
x is a positive real number,
R is a real number in the range 8 ≤ R ≤ 18,
d is a natural number being the precision for internal computations,
r is a floating-point number representing (x).

 PsiSmallXEpsilon (pse)
This function computes b based on Rosser and Schoenfeld’s 1962
approach, to precision d – see Section 3.2.
PsiSmallXEpsilon[b, m, d] −→ r
b is a positive real number such that x = eb is the point at which ψ(x) is
estimated,
m is a natural number being the parameter of the estimation,
d is a natural number representing the precision for internal computations,
r is a floating-point number representing the value of b .

 RedhefferMatrix (rem)
This function gives the divisibility matrix of Redheffer – see
Section 10.4.
RedhefferMatrix[n] −→ LL
n is a natural number,
LL is a list of n lists, being the square divisibility matrix. It can be printed in
standard matrix form using the function MatrixForm.

 RightAbundantQ (raq)
This function determines whether n is a right abundant number up to
multiples a · n with 1 ≤ a ≤ b – see Chapter 9.
RightAbundantQ[n, b, d] −→ P
n is a natural number,
b is a natural number,
 B.2 RHpack Functions 311

b is a natural number,
P is True if G(n) ≥ G(an) for 1 < a ≤ b, False otherwise.

 RobinsInequalityQ (riq)
This function checks the inequality σ(n)/n < eγ loglog n, to precision d –
see Chapter 5.
RobinsInequalityQ[n,d] −→ P
n is a natural number,
d is a natural number being the precision for internal computations,
P is True if the inequality is satisfied, otherwise False.

S
This function computes the argument of ζ( 12 + it) as in Section 10.9.
S[t] −→ r
t is a positive real number,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the value of S (t) = N(t) −
(ϑ(t)/π + 1).

 SetA, SetB, SetC, SetD

These symbols evaluate to the sets defined in Chapter 8. For example
SetB −→ {2, 3, 5, 6, 10, 30}
 SigmaOverN (son)
This function evaluates σ(n)/n as an exact rational.
SigmaOverN[n] −→ q
n is a natural number,
q is a rational number being the value of σ(n)/n.

 SuperAbundantQ (saq)
This function determines whether n is a superabundant integer – see
Chapter 6.
SuperAbundantQ[n] −→ c
n is a natural number,
P is True if n is superabundant, False otherwise.

 UnitaryDivisorSigma (uds)
This function evaluates the sum the unitary divisors of n – see
Section 7.12.
UnitaryDivisorSigma[n] −→ m
312 RHpack Mini-Manual

n is a natural number,
m is a natural number being the sum of the unitary divisors of n.

 VerifyThetaX (vth)
This function verifies the equation θ(x) < x for a ≤ x ≤ b to precision d –
see Section 3.3
VerifyThetaX[a, b, d] −→ bad
a is a real number being the initial point of the x values to be tested,
b is a real number being the end point of the domain of x values to be tested,
d is a natural number being the precision for internal computations,
bad is a list of integers representing points in the domain where the
inequality θ(x) < x fails at precision d/2.

 Xi
This function evaluates the Riemann ξ(s) function of Section 10.9 at
complex numbers s.
Xi[s, d] −→ c
s is a real or complex number,
d is a natural number being the precision for internal computations,
c is a floating-point complex number.

 ZetaZeroCount (zzc)
This function returns a count of the zeta zeros up to height T up to height
106 – see Section 2.2.
ZetaZeroCount[T, d] −→ n
T is a positive real number not greater than 106 ,
d is a natural number being the precision for internal computations,
n is a non-negative integer, representing the number of zeros of ζ(s) on the
critical line, above the x-axis, with imaginary part satisfying 0 ≤ γ ≤ T ,
to precision d.

 ZetaZeroHeight (zzh)
This function computes the ordinate of the nth highest positive
imaginary part of a zeta zero, to precision d – see Section 2.2.
ZetaZeroHeight[n, d] −→ r
n is a positive integer being the rank of the zero with n ≤ 107 ,
d is a natural number being the precision for internal computations,
r is a floating-point number representing the y-coordinate of the nth zero.
 References


[1] A. Akbary, Z. Friggstad and R. Jurecevic, Explicit upper bounds for p≤pω(n) p/(p − 1),
Cont. Discrete Math. 2 (2007), 153–160.
[2] A. Akbary and Z. Friggstad, Superabundant numbers and the Riemann hypothesis,
Amer. Math. Monthly 116 (2009), 273–275.
[3] L. Alaoglu and P. Erdős, On highly composite and similar numbers, Trans. Amer.
Math. Soc. 56 (1944), 448–469.
[4] L. Alfors, Complex Analysis, 2nd edn, McGraw-Hill, 1966.
[5] T. M. Apostol, Modular Functions and Dirichlet Series in Number Theory, Springer,
1976.
[6] T. M. Apostol, Introduction to Analytic Number Theory, 2nd edn, Springer, 1990.
[7] R. J. Backlund, Über die nullstellen der Riemannschen Zetafunktion, Acta Math. 41
(1918), 345–375.
[8] R. Balasubramanian, S. Kanemitsu and M. Yoshimoto, Euler products, Farey series
and the Riemann hypothesis II, Publ. Math. Debrecen 69 (2006), 1–16.
[9] W. W. Barrett, R. W. Forcade and A. D. Pollington, On the spectral radius of a (0,1)
matrix related to Mertens’ function, Linear Algebra Appl. 107 (1988), 151–159.
[10] B. C. Berndt, Ramanujan’s Notebooks, Parts I–V, Springer, 1985–1998.
[11] M. V. Berry and J. P. Keating, The Riemann zeros and eigenvalue asymptotics, SIAM
Rev. 41 (1999), 236–266.
[12] A. R. Booker, Turing and the Riemann hypothesis, Notices Amer. Math. Soc. 53
(2006), 1208–1211.
[13] P. Borwein, S. Choi, B. Rooney and A. Weirathmueller, The Riemann Hypothesis: A
Resource for the Afficionado and Virtuoso Alike, CMS Books in Mathematics, Springer,
2008.
[14] R. P. Brent, On the zeros of the Riemann zeta function in the critical strip, Math. Comp.
33 (1979), 1361–1372.
[15] R. P. Brent, J. van de Lune, H. J. J. te Riele and D. T. Winter, On the zeros of the
Riemann zeta function in the critical strip II, Math. Comp. 39 (1982), 681–688.
[16] K. Briggs, Abundant numbers and the Riemann hypothesis, J. Exp. Math. 15 (2006),
251–256.
[17] K. A. Broughan, Structure of sectors of zeros of entire flows, Topology Proc. 27 (2002),
1–16.
[18] K. A. Broughan, Vanishing of the integral of the Hurwitz zeta function, Bull. Austral.
Math. Soc. 65 (2002), 121–127.

313
314 References

[19] K. A. Broughan, Holomorphic flows on simply connected subsets have no limit cycles,
Meccanica 38 (2003), 699–709.
[20] K. A. Broughan and A. R. Barnett, The holomorphic flow of the Riemann zeta function,
Math. Comp. 73 (2004), 987–1004.
[21] K. A. Broughan, Holomorphic flow of the Riemann xi function, Nonlinearity 18 (2005),
1269–1294.
[22] K. A. Broughan and A. R. Barnett, Linear law for the logarithms of the Riemann
periods at simple critical zeros, Math. Comp. 75 (2006), 891–902.
[23] K. A. Broughan and A. R. Barnett, Corrigendum to “The holomorphic flow of the
Riemann zeta function”, Math. Comp. 76 (2007), 2249–2250.
[24] K. A. Broughan, Extension of the Riemann ξ-function’s logarithmic derivative
positivity region to near the critical strip, Can. Math. Bull. 52 (2009), 186–194.
[25] K. A. Broughan and A. R. Barnett, Gram lines and the average of the real part of the
Riemann zeta function, Math. Comp. 81 (2012) 1669–1679.
[26] K. A. Broughan, Website for phase portraits of the zeros of ζ(s) and ξ(s). www.math.
waikato.ac.nz/∼kab.
[27] K. A. Broughan, J.-M. De Koninck, I. Katai and F. Luca, On integers for which the sum
of divisors is the square of the squarefree core, J. Integer Seq. 15 (2012), art. 12.7.5.
[28] K. A. Broughan and D. Delbourgo, On the ratio of the sum of divisors and Euler’s
totient function I, J. Integer Seq. 16 (2013), art. 13.8.8 (17pp.).
[29] K. A. Broughan, K. Ford and F. Luca, On square values of the product of the Euler
totient function and the sum of divisors function, Colloq. Math. 130 (2013), 127–137.
[30] K. A. Broughan and Q. Zhou, On the ratio of the sum of divisors and Euler’s totient
function II, J. Integer Seq. 17 (2014), art. 14.9.2 (22pp.).
[31] K. A. Broughan and T. S. Trudgian, Robin’s inequality for 11-free integers, Integers
15 (2015), #A12 (5pp.).
[32] K. A. Broughan, Equivalents of the Riemann Hypothesis II: Analytic Equivalences,
Cambridge University Press, 2017.
[33] J. Büthe, J. Franke, A. Jost and T. Kleinjung, A practical analytic method for
calculating π(x), Preprint.
[34] G. Caveney, J.-L. Nicolas and J. Sondow, Robin’s theorem, primes, and a new
elementary reformulation of the Riemann hypothesis, Integers 11 (2011), #A33 (10pp.).
[35] G. Caveney, J.-L. Nicolas and J. Sondow, On SA, CA and GA numbers, Ramanujan J.
29 (2012), 359–384.
[36] Y. Cheng, An explicit upper bound for the Riemann zeta-function near the line σ = 1,
Rocky Mountain J. Math. 29 (1999), 115–140.
[37] Y. Cheng and S. W. Graham, Explicit estimates for the Riemann zeta function, Rocky
Mountain J. Math. 34 (2004), 1261–1280.
[38] Y. Choie, N. Lichiardopol, P. Moree and P. Solé, On Robin’s criterion for the Riemann
hypothesis, J. Théor. Nombres Bordeaux 19 (2007), 357–372.
[39] S. Chowla, On abundant numbers, J. Indian Math. Soc. (2) 1 (1934), 41–44.
[40] J. B. Conrey, More than two fifths of the zeros of the Riemann zeta function are on the
critical line, J. reine angew. Math. 399 (1989), 1–16.
[41] J. B. Conrey, L-functions and random matrices, in Mathematics Unlimited – 2001 and
Beyond, pp. 331–352, Springer, 2001.
[42] H. Davenport, Über numeri abundantes, Sitzungsber. Preuss. Akad. Wiss. Berlin 27
(1933), 830–837.
[43] H. Davenport, Multiplicative Number Theory, 3rd edn, Springer, 2000.
[44] M. Davis, Y. Matijasevic̆ and J. Robinson, Hilbert’s tenth problem: diophantine
equations: positive aspects of a negative solution, in Mathematical Developments
 References 315

Arising from Hilbert Problems, Proc. Symp. Pure Math., XXVIII, De Kalb, IL, 1974,
pp. 323–378. American Mathematical Society, 1976.
[45] A. Derbal, Grandes valeurs de la fonction σ(n)/σ∗ (n), C. R. Math. Acad. Sci. Paris
364 (2008), 125–128.
[46] L. E. Dickson, History of the Theory of Numbers, vol. I, Dover, 2005.
[47] R. D. Dixon and L. Schoenfeld, The size of the Riemann zeta-function at places
symmetric with respect to the point 1/2, Duke Math. J. 33 (1966), 291–292.
[48] P. Dusart, Inégalités explicites pour ψ(X), θ(X), π(X) et les nombres premiers, C. R.
Math. Acad. Sci. Soc. R. Can. 21 (1999), 53–59.
[49] P. Dusart, The kth prime is greater than k(log k + loglog k − 1) for k ≥ 2, Math. Comp.
68 (1999), 411–415.
[50] P. Dusart, Estimates of some functions over primes without RH, Preprint,
arXiv:1002.0442v1, 2010.
[51] H. M. Edwards, Riemann’s Zeta Function, Academic Press, 1974; reprinted by Dover,
2001.
[52] P. Erdős, On the density of abundant numbers, J. London Math. Soc. 9 (1934), 278–
282.
[53] P. Erdős, On the normal number of prime factors of p − 1 and some related problems
concerning Euler’s φ-function, Quart. J. Math. 6 (1935), 205–213.
[54] P. Erdős, Some remarks on Euler’s ϕ function, Acta Arith. 4 (1958), 10–19.
[55] P. Erdős and P. Turan, On some problems of a statistical group theory I,
Z. Wahrscheinlichkeitstheorie verw. Geb. 4 (1965), 175–186.
[56] P. Erdős and J.-L. Nicolas, Repartition des nombres superabondants, Bull. Soc. Math.
France 103 (1975), 65–90.
[57] L. Faber and H. Kadiri, New bounds for ψ(x), Math. Comp. 84 (2015), 1339–1357.
[58] K. Ford, The distribution of totients, Ramanujan J. 2 (1998), 67–151 (Paul Erdős
memorial issue).
[59] K. Ford, Zero-free regions for the Riemann zeta function, in Number Theory for the
Millennium II, Urbana, IL, 2000, pp. 25–56, A. K. Peters, 2002.
[60] K. Ford, Vinogradov’s integral and bounds for the Riemann zeta function, Proc.
London Math. Soc. (3) 85 (2002), 565–633.
[61] K. Ford, F. Luca and C. Pomerance, Common values of the arithmetic functions φ and
σ, Bull. London Math. Soc. 42 (2010), 478–488.
[62] K. Ford and P. Pollack, On common values of φ(n) and σ(n) I, Acta Math. Hungar.
133 (2011), 251–271.
[63] K. Ford and P. Pollack, On common values of φ(n) and σ(n) II, Algebra Number Theory
6 (2012), 1669–1696.
[64] J. Franel and E. Landau, Les suites de Farey et le probléme des nombres premiers,
Göttinger Nachr. (1924), 198–206.
[65] J. A. Gallian, Contemporary Abstract Algebra, 5th edn, Houghton Mifflin, 2002.
[66] R. Garunkstis, On a positivity property of the Riemann ξ-function, Lithuanian Math. J.
43 (2002), 140–145.
[67] B. R. Gillespie, Extending Redheffer’s matrix to arbitary arithmetic functions,
Bachelor with Honors Thesis, Pennsylvania State University, 2011.
[68] X. Gourdon, The 1013 first zeros of the Riemann zeta function, and zeros computation
at very large height, Preprint, 2004.
[69] T. H. Grönwall, Some asymptotic expressions in the theory of numbers, Trans. Amer.
Math. Soc. 14 (1913), 113–122.
[70] E. Grosswald, Oscillation theorems of arithmetical functions, Trans. Amer. Math. Soc.
126 (1967), 1–28.
316 References

[71] J. L. Hafner, New omega theorems for two classical lattice point problems, Invent.
Math. 63 (1981), 181–186.
[72] R. R. Hall and W. K. Hayman, Hyperbolic distance and distinct zeros of the Riemann
zeta-function in small regions, J. reine angew. Math. 526 (2000), 35–59.
[73] G. H. Hardy, The average order of the arithmetical functions P(x) and Δ(x), Proc.
London Math. Soc. 15 (1916), 192–213.
[74] G. H. Hardy and J. E. Littlewood The zeros of Riemann’s zeta function on the real line,
Math. Z. 10 (1921), 283–317.
[75] G. H. Hardy and J. M. Wright, An Introduction to the Theory of Numbers, 6th edn,
Oxford University Press, 2008.
[76] D. R. Heath-Brown, Simple zeros of the Riemann zeta function on the critical line, Bull.
London Math. Soc. 11 (1979), 17–18.
[77] D. R. Heath-Brown, Zero-free regions for Dirichlet L-functions, and the least prime in
an arithmetic progression, Proc. London Math. Soc. 64 (1992), 265–338.
[78] D. A. Hejhal and A. M. Odlyzko, Alan Turing and the Riemann zeta function, in Alan
Turing – His Work and Impact, eds S. B. Cooper and J. van Leeuwen, Elsevier, 2012.
[79] G. A. Hiary and A. M. Odlyzko, The zeta function on the critical line: numerical
evidence for moments and random matrix theory models, Math. Comp. 81 (2012),
1723–1752.
[80] A. E. Ingham, The Distribution of Prime Numbers, Cambridge University Press, 1932.
[81] A. Ivić, Two inequalities for the sum of divisors functions, Univ. Novom Sadu, Zb.
Rad. Prirod.-Mat. Fak. 7 (1977), 17–22.
[82] A. Ivić, Riemann Zeta Function: Theory and Applications, Dover, 2003.
[83] W.-J. Jang and S-H. Kwon, A note on Kadiri’s explicit zero free region for the Riemann
zeta function, J. Korean Math. Soc. 51 (2014), 1291–1304.
[84] H. Kadiri, Une région explicite sans zéros pour la fonction ζ de Riemann, Acta Arith.
117 (2005), 303–339.
[85] S. Kanemitsu and M. Yoshimoto, Farey series and the Riemann hypothesis, Acta Arith.
75 (1996), 351–374.
[86] S. Kanemitsu and M. Yoshimoto, Farey series and the Riemann hypothesis III,
Ramanujan J. 1 (1997), 363–378.
[87] S. Kanemitsu and M. Yoshimoto, Euler products, Farey series and the Riemann
hypothesis, Pub. Math. Debrecen 56 (2000), 431–449.
[88] A. A. Karatsuba and S. M. Voronin, Riemann zeta function (transl. from Russian by N.
Koblitz), De Gruyter, 1994.
[89] N. M. Katz and P. Sarnak, Zeroes of zeta functions and symmetry, Bull. Amer. Math.
Soc. 36 (1999), 1–26.
[90] N. M. Katz and P. Sarnak, Random matrices, Frobenius eigenvalues, and monodromy,
AMS Colloquium Publications, 45. American Mathematical Society, 1999.
[91] J. P. Keating and N. C. Snaith, Random matrix theory and ζ( 21 + it), Comm. Math.
Phys. 214 (2000), 57–89.
[92] D. E. Knuth, The Art of Computer Programming I: Fundamental Algorithms, 3rd edn,
Addison-Wesley, 1997.
[93] H. von Koch, Sur la distribution des nombres premiers, Acta Math. 24 (1901),
159–182.
[94] H. von Koch, Über die Riemannsche Primzahlfunction, Math. Ann. 55 (1902),
440–464.
[95] J.-M. De Koninck and F. Luca, Analytic Number Theory: Exploring the Anatomy of
Integers, American Mathematical Society, 2012.
 References 317

[96] N. M. Korobov, Estimates of trigonometric sums and their applications, Uspehi Mat.
Nauk 13 (1958), 185–192.
[97] T. Kotnik, The prime counting function and its analytic approximations, Adv. Comp.
Math. 29 (2008), 55–70.
[98] J. C. Lagarias, On a positivity property of the Riemann ξ-function, Acta Arith. 99
(1999) 217–213.
[99] J. C. Lagarias, Errata: On a positivity property of the Riemann ξ-function, Acta Arith.
99 (1999) 765.
[100] J. C. Lagarias, An elementary problem equivalent to the Riemann hypothesis, Amer.
Math. Monthly 109 (2002), 534–543.
[101] J. C. Lagarias, Problem 10949, Amer. Math. Monthly 109 (2002), 569.
[102] J. C. Lagarias and W. Janous, A generous bound for divisor sums: problem 10949,
Amer. Math. Monthly 111 (2004), 264–265.
[103] E. Landau, Über die Maximalordnung der Permutationen gegebenen Grades, Arch.
Math. Phys. Ser. 3 5 (1903).
[104] E. Landau, Über einen satz von Tschebyschef, Math. Ann. 61 (1905), 527–550.
[105] E. Landau, Handbuch der lehre von der Verteilung der Primzahlen, 2nd edn, vols 1
and 2, Chelsea, 1953.
[106] S. Lang, Introduction to Transcendental Numbers, Addison-Wesley, 1966.
[107] N. Levinson and H. L. Montgomery, Zeros of the derivatives of the Riemann zeta-
function, Acta Math. 133 (1974), 49–65.
[108] J. E. Littlewood, Quelques conséquences de l’hypothése que la fonction ζ(s) n’a pas
de zéros dans le demi-plan s > 12 , C. R. Acad. Sci. Paris Sér. I Math. 158 (1912),
263–266.
[109] J. E. Littlewood, Sur la distribution des nombres premiers, C. R. Acad. Sci. Paris Sér.
I Math. 158 (1914), 1869–1872.
[110] E. R. Lorch, Spectral Theory, Oxford University Press, 1962.
[111] J. van de Lune, H. J. J. te Riele and D. T. Winter, On the zeros of the Riemann zeta
function in the critical strip IV, Math. Comp. 46 (1986), 667–681.
[112] J.-P. Massias, J.-L. Nicolas and G. Robin, Evaluation asymptotique de l’ordre
maximum d’un element du groupe symétrique, Acta Arith. 50 (1988), 221–242.
[113] J.-P. Massias, J.-L. Nicolas and G. Robin, Effective bounds for the maximal order of
an element in the symmetric group, Math. Comp. 53 (1989), 665–678.
[114] J.-P. Massias and G. Robin, Bornes effectives pour certaines fonctions concernant les
nombres premiers, J. Théor. Nombres Bordeaux 8 (1996), 215–242.
[115] F. Mertens, Über einize asymptotische Gesetse der Zahlentheorie, J. reine angew.
Math. 77 (1874), 46–62.
[116] M. Mikolás, Farey series and their connection with the prime number problem I, Acta
Univ. Szeged. Sect. Sci. Math. 13 (1949), 93–117.
[117] M. Mikolás, Farey series and their connection with the prime number problem II, Acta
Sci. Math. Szeged 14 (1951), 5–21.
[118] W. Miller, The maximum order of an element of a finite symmetric group, Amer. Math.
Monthly 94 (1987), 497–506.
[119] H. L. Montgomery, The pair correlation of zeros of the zeta function, in Analytic
Number Theory, Proc. Symp. Pure Math. XXIV, pp. 181–193, American Mathematical
Society, 1973.
[120] F. Morain, Tables sur la fonction g(n), Département de Mathématiques, Université de
Limoges, 1988.
[121] M. J. Mossinghoff and T. S. Trudgian, Non-negative trigonometric polynomials and a
zero-free region for the Riemann zeta-function, Preprint, arXiv:1410.3926v1, 2014.
318 References

[122] S. Nazardonyavi and S. Yakubovich, Superabundant numbers, their subsequences and

the Riemann hypothesis, Preprint, arXiv:1211.2147v3, 2013.
[123] S. Nazardonyavi and S. Yakubovich, Extremely abundant numbers and the Riemann
hypothesis, J. Integer Seq. 17 (2014), art. 14.2.8 (23pp.).
[124] M. B. Nathanson, Elementary Methods in Number Theory, Springer, 2000.
[125] J.-L. Nicolas, Ordre maximal d’un element du groupe des permutations et ‘highly
composite numbers’, Bull. Soc. Math. France 97 (1969), 129–191.
[126] J.-L. Nicolas, Petites valeurs de la fonction d’Euler, J. Number Theory 17 (1983),
375–388.
[127] J.-L. Nicolas, Small values of the Euler function and the Riemann hypothesis, Acta
Arith. 155 (2012), 311–321.
[128] J.-L. Nicolas, Ramanujan, Robin, highly composite numbers, and the Riemann
hypothesis, Preprint, arXiv:1211.6944v4, 2013.
[129] A. M. Odlyzko and H. J. J. te Riele, Disproof of the Mertens’ conjecture, J. reine
angew. Math. 357 (1985), 138–160.
[130] A. M. Odlyzko, On the distribution of spacings between zeros of the zeta function,
Math. Comp. 48 (1987), 273–308.
[131] S. J. Patterson, An Introduction to the Theory of the Riemann Zeta-Function,
Cambridge Studies in Advanced Mathematics 14, Cambridge University Press, 1988.
[132] D. J. Platt, Computing degree 1 L-functions rigorously, Ph.D. Thesis, University of
Bristol, 2011.
[133] D. J. Platt, Computing π(x) analytically, Math. Comp. 84 (2015), 1521–1535.
[134] D. J. Platt and S. A. Trudgian, An improved explicit bound on |ζ( 12 + it)|, J. Number
Theory 147 (2015), 842–851.
[135] D. J. Platt and S. A. Trudgian, On the first sign change of θ(x) − x, Math. Comp. 85
(2016), 1539–1547.
[136] S. Ramanujan, Highly composite numbers (Part 1), Proc. London Math. Soc. 14
(1915), 347–409.
[137] S. Ramanujan, annotated by J.-L. Nicolas and G. Robin, Highly composite numbers
(Part 2), Ramanujan J. 1 (1997), 119–153.
[138] R. Redheffer, Eine explizit lösbare Optimierungsaufgabe, in Numerische Methoden bei
Optimierungsaufgaben, Band 3 (Tagung, Math. Forschungsinst., Oberwolfach, 1976),
Int. Ser. Numer. Math., Vol. 36, pp. 213–216, Birkhäuser, 1977.
[139] H.-E. Richert, Zur Abschätzung der Riemannschen Zetafunktion in der Nähe der
Vertikalen σ = 1, Math. Ann. 169 (1967), 97–101.
[140] B. Riemann, Gesammelte Werke, Teubner, 1893; reprinted by Dover, 1953.
[141] G. Robin, Estimation de la fonction de Tchebychef θ sur le k-ième nombre premier et
grandes valeurs de la fonction ω(n) nombre de diviseurs premiers de n, Acta Arith. 42
(1983), 367–389.
[142] G. Robin, Grandes valeurs de la fonction sommes des diviseurs et hypotheses de
Riemann, J. Math. Pures Appl. 63 (1984), 187–213.
[143] J. B. Rosser, The n-th prime is greater than n log n, Proc. London Math. Soc. 45 (1938),
21–44.
[144] J. B. Rosser, Explicit bounds for some functions of prime numbers, Amer. J. Math. 63
(1941), 211–232.
[145] J. B. Rosser and L. Schoenfeld, Approximate formulas for some functions of prime
numbers, Illinois J. Math. 6 (1962), 64–94.
[146] J. B. Rosser and L. Schoenfeld, Sharper bounds for the Chebyshev functions θ(x) and
ψ(x), Math. Comp. 29 (1975), 243–269.
 References 319

[147] Z. Rudnick and P. Sarnak, Zeros of principal L-functions and random matrix theory. A
celebration of John F. Nash, Jr., Duke Math. J. 81 (1996), 269–322.
[148] K. Sabbagh, The Riemann Hypothesis, Farrar, Straus and Giroux, 2003.
[149] B. Sagan, The Symmetric Group: Representations, Combinatorial Algorithms, and
Symmetric Functions, Springer, 2001.
[150] J. Sándor, D. S. Mitrinović and B. Crstici, Handbook of Number Theory I, Springer,
2006.
[151] M. du Sautoy, The Music of the Primes, HarperCollins, 2003.
[152] E. Schmidt Über die Anzahl der Primzahlen unter gegebener Grenze, Math. Ann. 57
(1903), 195–204.
[153] L. Schoenfeld, Sharper bounds for the Chebyshev functions ϑ(x) and ψ(x) II, Math.
Comput. 30 (1976), 337–360.
[154] L. Schoenfeld, Corrigendum: Sharper bounds for the Chebyshev functions ϑ(x) and
ψ(x) II, Math. Comput. 30 (1976), 900.
[155] D. Schumayer and D. A. W. Hutchinson, Physics of the Riemann hypothesis, Rev. Mod.
Phys. 83 (2011), 307–330.
[156] S. M. Shah, An inequality for the arithmetic function g(x), J. Indian Math. Soc. 3
(1939), 316–318.
[157] S. Skews, On the difference π(x) − li(x), J. London Math. Soc. 8 (1933), 277–283.
[158] N. J. A. Sloan, Online Encyclopedia of Integer Sequences (OEIS), https://oeis.org.
[159] P. Solé and M. Planat, The Robin inequality for 7-free integers, Integers 11 (2011),
#A65 (8pp.).
[160] K. Soundararajan, Moments of the Riemann zeta function, Annals Math. 170 (2009),
981–993.
[161] A. Speiser, Geometrisches zur Riemannschen Zetafunktion, Math. Ann. 110 (1934),
514–521.
[162] R. Spira, An inequality for the Riemann zeta function, Duke Math. J. 32 (1965), 247–
250.
[163] R. Spira, Zeros of ζ  (s) and the Riemann hypothesis, Illinois J. Math. 17 (1973),
147–152.
[164] A. I. Suriajaya, On the zeros of the k-th derivative of the Riemann zeta function under
the Riemann hypothesis, Funct. Approx. Comment. Math. 53 (2015), 69–95.
[165] S. M. Szalay, On the maximal order in S n and S n∗ , Acta Arith. 37 (1980), 321–331.
[166] G. Tenenbaum, Introduction to Analytic and Probabilistic Number Theory, Cambridge
University Press, 1995.
[167] E. C. Titchmarsh and D. R. Heath-Brown, The Theory of the Riemann Zeta-Function,
2nd edn, Oxford University Press, 1986.
[168] T. S. Trudgian, Improvements to Turing’s method, Math. Comp. 80 (2011), 2259–2279.
[169] T. S. Trudgian, An improved upper bound for the argument of the Riemann zeta-
function on the critical line II, J. Number Theory 134 (2014), 280–292.
[170] T. S. Trudgian, The sum of the unitary divisor function, Publ. Inst. Math. (Belgrad)
(N.S.) 97 (2015), 175–180.
[171] T. S. Trudgian, A short extension of two of Spira’s results, J. Math. Inequalities 9
(2015), 795–798.
[172] T. S. Trudgian, Improvements to Turing’s method II, Rocky Mountain J. Math. 46
(2016), 325–332.
[173] T. S. Trudgian, Updating the error term in the prime number theorem, Ramanujan J.
39 (2016), 225–234.
[174] A. M. Turing, A method for the calculation of the zeta-function, Proc. London Math.
Soc. 48 (1943), 180–197.
320 References

[175] A. M. Turing, Some calculations of the Riemann zeta-function, Proc. London Math.
Soc. 3 (1953), 99–117.
[176] I. M. Vinogradov, A new estimate for ζ(1 + it), Izv. Akad. Nauk SSSR, Ser. Mat. 22
(1958), 161–164.
[177] A. Walfisz, Weylsche Exponentialsummen in der neueren Zahlentheorie, Mathematis-
che Forschungsberichte, XV, VEB Deutscher Verlag der Wissenschaften, 1963.
[178] S. Wedeniwski, Section 1.1 History, ZetaGrad website, www.zetagrid.net/zeta/
math/zeta.result.100billion.zeros.html.
[179] A. Weil, Sur les “formules explicites” de la théorie des nombres premiers, Comm.
Sém. Math. Univ. Lund (1952), 252–265.
[180] H. S. Wilf, The Redheffer matrix of a partially ordered set, Electron. J. Combin. 11
(2004/2006), Research Paper 10 (5pp.).
[181] M. Yoshimoto, Farey series and the Riemann hypothesis II, Acta Math. Hungar. 78
(1998), 287–304.
[182] D. Zagier, The first 50 million prime numbers, Math. Intelligencer (1977), 7–19.
[183] S. Zegowitz, On the positive region of π(x) − li(x). Master’s Thesis, University of
Manchester, 2010.
 Index

Abel’s identity, 70, 79, 106, 107, 194, 275 CAPrimes (RHpack), 287, 296
abundant number, 144–146 Cauchy, A. L., 5
adjoint of a transformation, 282 Caveney, G., 11, 218, 314
admissible zero set, 259, 261 CheckCNk (RHpack), 98, 297
Akbary, A., 170, 197, 313 Cheng, Y., 19, 314
Alaoglu, L., 10, 144, 313 Choi, S., 14, 21, 313
Alfors, L., 260, 313 Choie, Y., 10, 200, 201, 314
AllPrimesFactor (RHpack), 296 Chowla, S., 145, 314
Apostol, T. M., 16, 26, 31–33, 70, 79, 88, 89, 91, colossally abundant number, 12, 144, 153–161,
106, 107, 109, 112, 132, 146, 194, 249, 252, 170, 212, 219, 280, 290
275, 313 ColossallyAbundantF (RHpack), 146, 297
applications of S n , 272 ColossallyAbundantInteger (RHpack), 146, 297
author homepage, xviii ColossallyAbundantQ (RHpack), 146, 297
CompositeQ (RHpack), 298
computable, 241
background reading, 13 ComputeC (RHpack), 98, 298
Backlund, R. J., 20, 21, 39, 47, 48, 54, 72, 101, ComputeDeltas (RHpack), 98, 118, 298
254, 313 Computef (RHpack), 98, 298
BacklundB (RHpack), 16, 296 ComputeH (RHpack), 98, 298
Balasubramanian, R., 247 ComputeKm (RHpack), 41, 49, 299
Barnett, A. R., 17, 284, 314 ComputeTheta (RHpack), 69, 299
Barrett, W. W., 251, 313 ComputeThetaValues (RHpack), 69, 299
Berndt, B. C., 7, 165, 313 Comrie, L. J., 21
Berry, M. V., xx, 283, 313 Conrey, J. B., xviii, 283, 314
Bohr, H., xvii critical epsilon value, 153, 161
Bombieri, E., xvii, xx critical line, 6, 17
Borwein, P., 14, 21, 93, 313 CriticalEpsilonValues (RHpack), 146, 299
Brent, R. P., 21, 313 Crstici, B., 96, 319
Briggs, K., 216, 313 cycle in Sn , 272
Broughan, K. A., 11, 17, 97, 166, 200, 201,
238, 259, 260, 262, 278, 284, 313,
314 Davenport, H., 145, 314
Büthe, J., 22, 314 Davis, M., 239, 315
De Koninck, J.-M., 96, 109, 166, 314, 316
de la Vallée-Poussin, 69
CA number, 144 definition of F(p, α), 153
Caley’s theorem, 272 definition of G(n), 160

321
322 Index

definition of ni , 156 functional equation, 16

definition of P(n), 208
definition of xk , 154
GA1 number, 11
Delbourgo, D., xx, 166, 314
GA2 number, 11
Derbal, A., 198, 315
Gallian, J. A., 272, 315
Dickson, L. E., 146, 315
Garunkstis, R., 238, 259, 315
Dirichlet eta criterion, 252
Gauss, J. C. F., 3
Dirichlet eta function, 11, 252
GenerateCA (RHpack), 287, 301
Dirichlet, J. L., 1, 3, 6, 138, 146, 198, 199, 249,
GenerateHC (RHpack), 287, 301
252
GenerateHR (RHpack), 287, 301
divisibility graph, 11, 250–251
GenerateLandau (RHpack), 287, 301
divisibility graph criterion, 251
GenerateSA (RHpack), 287, 302
DivisibilityGraph (RHpack), 239, 300
Gillespie, B. R., 248, 315
divisor sum, 12
Gourdon, X., 21, 315
Dixon, R. D., 256, 315
Grönwall’s function, 218
Dixon–Spira criterion, 256–259
Grönwall, T. H., 10, 166, 197, 315
du Sautoy, M., 14, 319
Graham, S. W., 19, 314
Dusart, P., 41, 165, 215–217, 315
Gram, J. P., 21, 314
graph cycle, 250
Edwards, H. M., 14, 16, 23, 88, 89, 242, 245, 253, GronwallG (RHpack), 219, 302
262, 266, 267, 284, 315 Grosswald, E., 81, 315
epilogue, 285
Erdős, P., 7, 10, 11, 144, 145, 147–161, 163, 313, Hadamard factorization, 33
315 Hadamard, J., 29
estimates for θ(x), 54–67 Hafner, J. L., 199, 316
estimates for x2 , 161 Hall, R. R., 270, 286, 316
Euclid, 1 Hardy, G. H., xvii, 7, 11, 96, 109, 146, 166, 199,
Euler’s constant, 7, 140, 193, 202 316
Euler’s phi function, 13, 94, 98–110, 165 Hardy–Ramanujan number, 10, 201–208
Euler, L., 1, 2, 10, 16, 94 HardyRamanujanQ (RHpack), 146, 202, 302
even cycle, 251 HardyRamanujanToPrimorialForm (RHpack),
exact computational methods, 51–54 202, 287, 302
examples of Sn , 272 harmonic number, 13
exponent pattern, 206 Hayman, W. K., 270, 286, 316
extraordinary number, 11, 218, 225 Heath-Brown, D. R., 30, 316, 319
ExtraordinaryQ (RHpack), 219, 300 height H, 21, 29, 46, 49, 99
extremely abundant number, 12, 218, 235 Hejhal, D. A., 22, 316
ExtremelyAbundantQ (RHpack), 219, 300 Hiary, G. A., 283, 316
highly abundant number, 144
highly composite number, 6, 144
Faber, L., 51, 315
HighlyCompositeQ (RHpack), 146, 302
Farey fractions, 11
Hilbert, D., xvii, 6
Farey rationals, 241
Hilbert–Pólya conjecture, 11, 282–285
Farey series, 241
Hilbert–Pólya criterion, 285
Farey sum, 247
Holmes, G., xx
FareyFractions (RHpack), 239, 300
Hutchinson, D. A. W., 283, 319
FHalf (RHpack), 98, 120, 300
Hutchinson, J. I., 21
FindNextXA (RHpack), 287, 301
Forcade, R. W., 251, 313
Ford, K., xviii, 19, 21, 29, 41, 166, 315 Ingham, A. E., 81, 131, 134, 176
four exponentials theorem, 159 InitialCA (RHpack), 287, 303
Franel, J., 242, 315 InitialCA150 (RHpack), 287, 303
Franel–Landau criterion, 241–247 InitialCAPrimes (RHpack), 287, 303
Franke, J., 22, 314 InitialEX (RHpack), 303
Friggstad, Z., 170, 197, 313 InitialHC (RHpack), 287, 303
 Index 323

InitialLandau (RHpack), 287, 303 logarithmic integral, 11, 13

InitialNR (RHpack), 287, 303 LogKthPrimorialBounds (RHpack), 202, 305
InitialSA (RHpack), 287, 303 Lorch, E. R., 282, 317
InitialSA1000 (RHpack), 287, 303 Luca, F., 96, 109, 166, 314–316
InitialXA (RHpack), 287, 303
IntegerHeight (RHpack), 239, 303
Ivić, A., 14, 16, 196, 198, 316 M (RHpack), 69, 305
Massias, J.-P., 11, 273, 317
Mathematica, xix
Jacobi, G., 3 Matijasevic̆, Y., 239, 315
Jang, W.-J., 21, 29, 316 maximum order, 12
Janous, W., 317 MaxPrime (RHpack), 305
Joe, S., xx Meller, N. A., 21
Jost, A., 22, 314 Mertens’ conjecture, 91
Jurecevic, R., 313 Mertens, F., 141, 317
Mikolás, M., 247, 317
Miller, W., 272, 275, 317
Kadiri, H., 21, 29, 51, 315, 316
Mitrinović, D. S., 96, 319
Kanemitsu, S., 247, 316
Montgomery, H. L., 238, 253, 270, 282, 317
Karatsuba, A. A., 14, 316
Morain, F., 275, 317
Katai, I., 166, 314
Moree, P., 10, 200, 314
Katz, N. M., 269, 283, 316
Mossinghoff, M. J., 30, 317
Keating, J. P., 283, 313, 316
multiplicative order, 13
KFreeQ (RHpack), 303
Klein, F., 5
Kleinjung, T., 22, 314 Nathanson, M. B., 145, 318
Knuth, D. E., 31, 316 Nazardonyavi, S., 11, 147, 218, 219, 318
Korobov, N. M., 20, 317 NextZetaZero (RHpack), 16, 305
Kotnik, T., 88, 317 NiceFactorInteger (RHpack), 305
KthApproximatePrimorial (RHpack), 202, 303 Nicolas’ first theorem, 135–136
KthPrimorial (RHpack), 304 Nicolas’ inequality, 201, 203, 205, 206
Kwon, S.-H., 21, 29, 316 Nicolas’ second theorem, 137–142
Nicolas, J.-L., 7, 10, 11, 94, 96, 110, 142, 153,
Lagarias’ positivity criterion, 259–271 173, 192, 218, 273, 314, 315, 317, 318
Lagarias, J. C., 10, 11, 168, 169, 193–197, 238, NicolasInequalityQ (RHpack), 98, 306
259, 262, 317 NicolasTwoFailures (RHpack), 98, 140, 306
LagariasInequalityQ (RHpack), 169, 304 notations, 12–13
Lambda (RHpack), 239, 304 NOverPhi (RHpack), 98, 306
Landau, E., 9, 11, 21, 29, 68, 81, 134, 275, 284, NthHarmonicNumber (RHpack), 306
315, 317 NthPrime (RHpack), 307
Landau–Vinogradov symbols, 13 NthPrimeBounds (RHpack), 202, 307
Lang, S., 159, 317
LargestZetaZero (RHpack), 16, 304 odd cycle, 251
LCM, 272 Odlyzko, A. M., 22, 91, 270, 283, 316, 318
left abundant number, 11, 218, 219, 226, 235 OEIS, 147, 187, 272, 287, 289–291, 293
LeftAbundantQ (RHpack), 219, 304 order of a permutation, 272
Lehman, R. S., 21 oscillation theorems, 68, 81–88
Lehmer, D. H., 21
length of a cycle, 272
Levinson, N., 238, 253, 256, 317 Patterson, S. J., 14, 318
Lichiardopol, N., 10, 200, 314 Planat, M., 212, 319
Liouville criterion, 92 Platt, D. J., 19, 21, 29, 30, 64, 318
Liouville’s function, 13, 92 PlotCNk (RHpack), 98, 307
Littlewood criterion, 69, 88–92 PlotDivisibilityGraph (RHpack), 239, 307
Littlewood, J. E., 69, 83, 88, 89, 229, 236, 245, PlotEulerPhiRatio (RHpack), 98, 307
247, 250, 269, 316, 317 PlotInequality (RHpack), 98
324 Index

PlotRobinsInequality (RHpack), 169, 308 Rosser, J. B., 9, 20, 21, 29, 39–41, 49, 54, 94, 96,
PlotVonMangoldtPsi (RHpack), 16, 308 98–110, 125, 167, 177, 216, 318
Pollack, P., 166, 315 Rudnick, Z., 283, 319
Pollington, A. D., 251, 313
Pomerance, C., 166, 315
PowerfulQ (RHpack), 308 S n , 238, 271, 272
prime number theorem, 15, 69–80 S (RHpack), 239, 311
PrimePower (RHpack), 308 Sándor, J., 96, 319
primorial, 9, 96 SA number, 144
PrimorialFormQ (RHpack), 309 Sabbagh, K., 13, 319
PrimorialFormToInteger (RHpack), 202, 287, 309 Sagan, B., 272, 319
proper left abundant number, 235 Sarnak, P., 269, 283, 316, 319
properties of Sn , 272 Schmidt, E., 81, 319
Psi (RHpack), 69, 309 Schoenfeld criterion, 68
Psi0 (RHpack), 16, 309 Schoenfeld, L., 9, 20, 21, 29, 39–41, 54, 64, 69,
PsiLargeXEpsilon (RHpack), 41, 309 94, 96, 98–110, 125, 167, 177, 198, 216, 315,
PsiLargeXEpsilon2 (RHpack), 41, 310 318, 319
PsiSmallEpsilon (RHpack), 51 Schumayer, D., 283, 319
PsiSmallXEpsilon (RHpack), 41, 310 self-adjoint transformation, 282
set A, 167, 200, 203, 219, 224
set B, 200, 203
Ramanujan, S., 6, 10, 11, 165, 318 set C, 200
Ramanujan–Robin criterion, 214 set D, 200, 206
Rankin, R., 7 set G, 280
Redheffer criterion, 247–250 set R, 200
Redheffer matrix, 11 set S , 203
Redheffer, R., 248, 318 SetA (RHpack), 202, 311
RedhefferMatrix (RHpack), 239, 310 SetB (RHpack), 202, 311
RH, xvii, xviii, 89, 94, 96, 97, 111, 117, 119, 123, SetC (RHpack), 202, 311
144, 164, 165, 168, 169, 173, 174, 180, SetD (RHpack), 202, 311
184–188, 191, 198, 201, 203, 218, 219, 227, Shah, S. M., 11, 319
235, 236, 238, 251, 253, 256, 259, 271, 273, Shapiro criterion, 236, 239–241
279, 283, 285 Siegel, C. L., 6, 24
RH classical equivalences, 68–92 SigmaOverN (RHpack), 169, 311
RHpack, xix, 12, 41, 69, 98, 120, 140, 146, 169, Skews number, 88
202, 219, 239, 287, 294, 295 Skews, S., 88, 319
Richert, H.-E., 19, 318 Sloan, N. J. A., 147, 319
Riemann hypothesis, xvii, 6–8, 15, 17, 21, 68, 89, Snaith, N. C., 283, 316
145, 167, 214, 219, 225, 231, 235, 273, 283 Solé, P., 10, 200, 212, 314, 319
Riemann hypothesis history, 1–7 Sondow, J., 11, 218, 314
Riemann xi function, 11, 16 Soundararajan, K., 284, 319
Riemann zeta function, 2, 8, 11, 15–39 Speiser, A., 237, 253, 319
Riemann, B., 5, 8, 21, 24, 283, 318 Spira, R., 238, 253, 255, 256, 319
Riemann–Siegel formula, 24 squarefree core, 205
right abundant number, 11, 218, 219, 231, 235 squarefree number, 201–203
RightAbundantQ (RHpack), 219, 310 squarefull number, 201, 202, 206
Robin’s inequality, 12, 201, 203, 206, 212, 214, standard notation, 12
218 sum-of-divisors function, 7
Robin’s inequality failures, 200–217 summary of Volume One, 8–12
Robin’s theorem, 165–193 superabundant number, 10, 12, 144, 147–161,
Robin, G., 7, 10, 11, 64, 144, 160, 163, 165–193, 170, 201, 208, 218, 280, 289
197, 273, 317, 318 SuperAbundantQ (RHpack), 146, 311
RobinsInequalityQ (RHpack), 169, 311 Suriajaya, A. I., 256, 319
Robinson, J., 239, 315 symmetric group criterion, 271–282
Rooney, B., 14, 21, 313 Szalay, S. M., 273, 319
 Index 325

table for Nicolas’ reversed inequality, 292 Walfisz, A., 69, 91, 320
table for Robin’s inequality, 288 Wedeniwski, S., 21, 30, 320
table of ni and i , 157 Weierstrass product, 26, 260, 262
table of colossally abundant numbers, 290 Weil, A., 30, 320
table of extremely abundant numbers, 287 Weirathmueller, A., 14, 21, 313
table of maximum symmetric group orders, 293 Wilf, H. S., 248, 320
table of superabundant numbers, 289 Wilson, M., xx
tables of bounds for ψ(x), 41–51 Winter, D. T., 21, 30, 313, 317
te Riele, H. J. J., 21, 30, 91, 313, 317, 318 Wright, J. M., 96, 109, 146, 166, 316
Tenenbaum, G., 81, 319
Titchmarsh, E. C., 14, 16, 21, 319
trivial zeros, 16 Xi (RHpack), 239, 312
Trudgian, T. S., xx, 19–21, 29, 30, 64, 93, 196,
198, 200, 201, 266, 314, 317–319
Turan, P., 11, 315 Yakubovich, S., 11, 147, 218, 219, 318
Turing, A. M., 21, 285, 320 Yohe, J. M., 21
Turner, J. C., xx Yoshimoto, M., 247, 316, 320

unitary divisor, 197 Zagier, D., 1, 320

UnitaryDivisorSigma (RHpack), 169, 311 Zegowitz, S., 88, 320
unsolved problems, 14, 39, 67, 93, 142, 163, 198, zero-free region, 6, 9, 21, 30, 99
217, 235, 286
zeta derivative, 253–256
zeta function properties, 16–21
van de Lune, J., 21, 30, 313, 317 zeta product representation, 30–39
VerifyThetaX (RHpack), 41, 312 zeta sums numerical estimates, 40–41
Vinogradov, I. M., 20, 320 zeta zero-free regions, 21–30
von Koch, H., 69, 316 zeta zeros, 6
von Mangoldt, H. C. F., 15, 21, 40, 102, 103, 121, ZetaZeroCount (RHpack), 239, 312
176, 285 ZetaZeroHeight (RHpack), 16, 312
Voronin, S. M., 14, 316 Zhou, Q., 166, 314